gaas/alas aspat diodes for millimetre and sub millimetre

235
GaAs/AlAs ASPAT Diodes for Millimetre and Sub-Millimetre Wave Applications A thesis submitted to The University of Manchester for the degree of Doctor of Philosophy In the Faculty of Sciences and Engineering 2017 MOHD RASHID REDZA BIN ABDULLAH SCHOOL OF ELECTRICAL AND ELECTRONIC ENGINEERING

Upload: others

Post on 16-Nov-2021

6 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre

GaAsAlAs ASPAT Diodes for Millimetre

and Sub-Millimetre Wave Applications

A thesis submitted to The University of Manchester for the degree of

Doctor of Philosophy

In the Faculty of Sciences and Engineering

2017

MOHD RASHID REDZA BIN ABDULLAH

SCHOOL OF ELECTRICAL AND ELECTRONIC ENGINEERING

2

LIST OF CONTENTS

LIST OF CONTENTS 2

LIST OF TABLES 6

LIST OF FIGURES 8

LIST OF PUBLICATIONS CONFERENCE PRESENTATIONS 13

PUBLICATIONS 13

CONFERENCE PRESENTATIONS 14

ABSTRACT 15

DECLARATION AND COPYRIGHT STATEMENT 16

ACKNOWLEDGEMENTS 17

DEDICATION 18

1 INTRODUCTION 19

11 Background 19

12 Aims and objectives 26

13 Outline of this Thesis 27

2 LITERATURE REVIEW 29

21 Introduction 29

22 Historical review of III-V Compound Semiconductor for RF applications 30

23 The Concept of Heterostructures 33

231 Homojunctions Heterojunctions and Band Discontinuities 34

232 Lattice-Matched and Pseudomorphic Materials 36

233 Quantum well and 2DEG 41

24 Metal-Semiconductor Contact 42

241 Schottky Contact 43

242 Ohmic Contact 46

25 Asymmetrical Spacer-layer Tunnel (ASPAT) diode Background work 48

26 Asymmetrical Spacer-layer Tunnel (ASPAT) diode The Basics 50

261 Principle of Quantum Tunneling 51

262 ASPAT Structural Parameters of GaAsAlAs materials System 55

Figure 214 Epitaxial layer structure of device XMBE 304 (lateral structure)

used in this study 55

3

263 ASPAT Electrical Parameters 61

27 Characterization of Ohmic Contacts 68

271 Transmission Line Measurement (TLM) 69

28 Basic Characterization Techniques and procedures 72

281 Measuring tools and apparatuses 72

282 Measurement steps using a VNA 75

283 Measurement Practice and Flowchart 76

3 RESULTS OF FABRICATION PROCESS OF ASPAT DIODES 79

31 Introduction 79

32 Epitaxial Layer Growth Techniques 80

321 Molecular Beam Epitaxy (MBE) 80

33 Basic Principles of Common Fabrication techniques 81

331 Sample cleaning 81

332 Photolithography 82

333 Etching Process 86

334 Sputtering (dielectric deposition) 88

335 Metallization Process Lift-off and Annealing 88

34 GaAsAlAs ASPAT Process Optimization 92

341 ASPAT Devices used in Fabrication 93

342 Fabrication process of GaAsAlAs ASPAT diode toward repeatability

reproducibility and manufacturability 95

343 Fabrication process of GaAsAlAs ASPAT diode toward High

frequency Applications 117

35 Conclusions 125

4 PHYSICAL MODELLING OF THE GAASALAS ASPAT DIODE USING

SILVACO 126

41 Introduction 126

42 SILVACO modelling Tools 127

43 SILVACO Implementation GaAs AlAs ASPAT Modelling 130

44 Simulation Result and Analysis 132

45 Structure Analysis of ASPAT Diode 135

451 Dependencies of current on AlAs Barrier thickness 135

452 Dependence of current on Spacer I length l1 137

4

453 Dependence of current on Spacer II length l2 138

46 Temperature Dependent Simulation 140

47 Temperature Dependence characteristics of GaAsAlAs ASPAT Diodes 142

48 Conclusions 147

5 DC amp RF CHARACTERISTIC OF ASPAT DIODES 148

51 Introduction 148

52 General Analysis of Current-Voltage (I-V) characteristics of GaAsAlAs

ASPAT diodes 149

53 RF Test Fixture Theory and Experiment 153

531 On-Wafer Measurement and Small Signal One-Port Characterizations

154

54 Device Calibration 155

541 Open and Short De-Embedding Technique 155

55 S-Parameter Measurement Result and Analysis 157

551 Diode to diode uniformity 158

552 Wafer to wafer uniformity 160

553 Small devices RF measurements 161

56 Extracting RF models of ASPAT at Zero Bias Voltage 164

561 Extraction of ASPAT parasitic element 165

562 Extraction of ASPAT intrinsic elements 168

563 Capacitances -Voltage (C-V) Extraction 172

57 Conclusions 173

6 MILLIMETRE WAVE GAASALAS ASPAT DETECTOR DESIGN USING

ADS 175

61 Introduction 175

62 Detection Theory 176

63 Definition of Characteristics of Interest Detection Parameters in small signal

analysis 179

64 Noise Consideration in a Detector diode 182

65 Modelling of a 100GHz Zero-biased ASPAT Detector 184

66 Conclusions 199

7 FREQUENCY MULTIPLIERS (DOUBLER) DESIGN USING GAASALAS

ASPAT DIODES 201

71 Introduction 201

5

72 Motivation and Background 202

73 Frequency Multiplier Architecture the Basics 203

731 Types of frequency multipliers 205

74 Parameters of interest for Frequency Multipliers 206

75 20GHz40GHz Varistor Mode ASPAT Diode Frequency Doubler 208

76 Conclusions 214

8 CONCLUSION AND FUTURE WORK 215

81 Conclusion 215

82 Future Work 217

REFERENCES 218

APPENDICES 228

Word count including footnotes and endnotes 61500(approximately)

6

LIST OF TABLES

Table 11 Sample XMBE304 XMBE307 and XMBE368 GaAsAlAs materials

structure grown on GaAs Substrates by MBE 23

Table 12 Quantum wells sandwiching the quantum barrier for sample XMBE314

grown on a GaAs substrate by MBE 24

Table 13 Novel InGaAsAlAs ASPAT diode grown on InP substrate by MBE on InP

substrate 24

Table 21 Lattice constant (a) and band gap for group III-V binary and ternary

compound semiconductors a room temperature [41 42] 38

Table 22 epitaxial layer of sample XMBE304 which is the main focus of this work 56

Table 31 Chemical wet etch characteristics using Orthophosphoric-based and

Ammonia on GaAs and InGaAs materials 87

Table 32 Epitaxial layer of Doped substrate samples 93

Table 33 The epitaxial layers of sample XMBE304 with barrier 10ML~28nm 94

Table 34 Generic fabrication steps established by Dr Md Adzhar [101] 97

Table 35 Standard process flow for Air-Bridge design fabrication 101

Table 36 Standard fabrication process flow for Dielectric-Bridge design 107

Table 37 New arrangement of the mask number and step in Second Run 110

Table 38 New arrangement for the Third run using Dielectric-Bridge mask 113

Table 39 The outcome of the spreading resistance before and after using LOR

technique 115

Table 310 DC and RF characteristics for XMBE304 118

Table 311 3rd

Gen Mask process step 119

Table 312 Standard deviation at two different voltages 124

Table 41 The parameter values used in this simulation 134

Table 42 The calculated values of bandgap at different temperatures 140

Table 43 The calculated effective masses for each temperature used in this simulation

141

Table 44 Epitaxial layer profile for Schottky Barrier Diode (XMBE104) 145

Table 51 Performance of the ASPAT diode obtained from measured IV characteristics

in this work 153

Table 52 Device to device uniformity check for large ASPAT diode 159

Table 53 The variation of reflection coefficient (S11) for GaAsAlAs ASPAT diode at

four different frequencies[117] 159

7

Table 54 Wafer to wafer uniformity check on XMBE304A and XMBE304B 161

Table 55 Comparison between calculated (fully Depleted) and extracted (different

biases) values from equivalent circuit parameters for different ASPAT mesa sizes at zero

bias voltage 170

Table 61 A summary of all the important parameters of the 4x4 microm2 diode 185

Table 62 Noise Equivalent Power (NEP) for zero bias detector diode 190

Table 63 Summary of characteristic of interest for GaAsAlAs ASPAT detector 198

Table 64 Comparison 4times4microm2 GaAsAlAs ASPAT diode to the-state-of-the-art zero

bias detector at W-band (75GHz-110GHz) 199

Table 71 Comparison parameter obtained from this simulation to the state-of-the-art

multiplier diode 213

8

LIST OF FIGURES

Figure 21 III-V compound semiconductors mobility and band gap[24] 31 Figure 22 illustration of Homojunctions band structure material before (left) and after

(right) equilibrium 34 Figure 23 Energy band diagrams before (left) and after (after) interface combination at

equilibrium 35

Figure 24 Lattice Matching for both materials when aL=aS 37 Figure 25 Energy gap and lattice constant for direct and indirect band gap of compound

semiconductor at ambient temperature[40] 37 Figure 26 Lattice mismatched material 39

Figure 27 The formation of Pseudomorphic-layer in two situations (a) compressive and

(b) tensile strain [1] 40 Figure 28 The band diagram of an ideal quantum well formed by a narrow band gap

semiconductor eg InGaAs sandwiched between two large band gap semiconductors eg

GaAs (a) Structure (b) energy band diagram and (c) Conduction band diagram when

AlGaAs is n-doped[43] 41 Figure 29 Metal and semiconductor in two conditions (a) separation (b) in contact 43

Figure 210 Energy band diagram of Schottky contact on n-type material under (a)

reverse and (b) forward bias 45 Figure 211 Figure 10 N-type semiconductor Ohmic contact with different doping

concentration ND (a) Low (b) Intermediate and (c) high 47 Figure 212 Classical view of whether an electron is can surmount a barrier or not

Quantum mechanical view allows an electron to tunnel through a barrier The probability

(blue) is related to the barrier thickness 51

Figure 213 Rectangular tunnelling with incident reflected and transmitted wave

function[70] 52

Figure 214 Epitaxial layer structure of device XMBE 304 (lateral structure) used in

this study 55 Figure 215 The layer profile of ASPAT with 28nm barrier thickness [27] 57

Figure 216 Conduction band diagram showing band bending and 2DEG formation at

the L1 spacer 60 Figure 217 I-V characteristics of a fabricated ASPAT diode 63 Figure 218 Intrinsic Elements of the Asymmetric Spacer Tunnel Layer (ASPAT) Diode

63 Figure 219 Cross-sectional view of ASPAT vertical structure the dimensions of d b

and h are not drawn to scale 65 Figure 220 Cross-section view of the complete process ASPAT diode (XMBE304)

The dimensions are not drawn to scale 67 Figure 221 A simple TLM structure with effective length and sheet resistance

underneath 69 Figure 222 Top view of TLM ladder structure use in this work 71 Figure 223 Typical plot of resistance versus TLM spacing 71

Figure 224 The set of DC measurement apparatus arrangement to measure the TLM 73

9

Figure 225 The set of RF measurement for on-wafer DC and RF one-port measurement

75 Figure 226 Actual VNA system that was used for RF characterization 75 Figure 227 Block diagram of the ASPAT measurement step 77

Figure 31 3D illustration of Optical lithography process used in this research 85 Figure 32 Actual picture of thermal evaporator used in this study 89 Figure 33 Single layer lift-off process using negative photoresist 91 Figure 34 Current-Voltage characteristic of sample XMBE368 used in this study at

two different locations on the wafer tile 96

Figure 35 ASPAT mask design for (a) 100times100microm2 (b) 30times30microm

2 and (c) 15times15 microm

2

diode dimensions designed in the 1st Gen Mask 96

Figure 36 Air Bridge design (red circle) for large mesa area (100times100microm2) device with

option for doped substrates 100 Figure 37 The layout of 1

st design of Dielectric Bridge (green circle) mask design for

100 times 100microm2 emitter size with option for doped substrate processing 100

Figure 38 Dry Etching for the first run in this study 102 Figure 39 Severe undercut of 2times2 microm

2 and 6times6 microm

2 devices 103

Figure 310 Device with emitter area of 2times2 microm2 and 6times6 microm

2 after dry etch and wet

etched 104 Figure 311 SEM Images of the GaAs sample 104

Figure 312 The cross-sectional view of InGaAs (left) and GaAs (right) samples used in

this study 105

Figure 313 SEM images taken from the XMBE304 samples and a GaAs test sample

105

Figure 314 SEM images for InP and InGaAs taken from [56] 105 Figure 315 Short circuit behaviour on one of the fabricated device in this run 106

Figure 316 The surface of the sample after final processing 108 Figure 317 IV characteristics of sample XMBE304 for emitter sizes of 1000 microm2

2500 microm2 900 microm2 400 microm2 225 microm2 100 microm2 and 36 microm2 109

Figure 318 Optical images after MESA etching (a) 1 microm Tolerance (b) 05 microm

Tolerance 110

Figure 319 After lift-off processing 111 Figure 320 The current density of XMBE304 processed using Si3N4 as dielectric layer

112

Figure 321 side view of lateral ASPAT structure 113

Figure 322 The measured size of the emitter area and the length D (blue color marked)

114 Figure 323 Summary of LOR technique steps 115 Figure 324 IV characteristic in 3rd run (reduced d-Gap) for each diode size on the mask

116

Figure 325 3rd

Gen Mask layout showing actual devices de-embedding TLM and

alignment mark structures used in this study 118 Figure 326 Step by step device fabrication (in 3D and cross-sectional view) for semi-

insulating substrate device type used in this study 121 Figure 327 Example finished process device with bond pad using 3

rd Gen mask 121

Figure 328 XMBE304 TLM measurement for the top contact after annealing 122

10

Figure 329 XMBE304 TLM measurement for the bottom contact after annealing 122 Figure 330 XMBE304 GaAsAlAs ASPAT measured IV characteristics at room

temperature for 4times4microm2 mesa size 123

Figure 331 XMBE304 GaAsAlAs ASPAT measured IV characteristics at room

temperature for 6times6microm2 mesa size 124

Figure 332 XMBE304 GaAsAlAs ASPAT measured IV characteristics at room

temperature for 10times10microm2 mesa size 124

Figure 41 SILVACO Atlas simulation process flow 128 Figure 42 The GaAsAlAs ASPAT structures produced from simulation based on the

diode multilayer heterostructures on the right 131 Figure 43 The Energy-momentum diagram for AlAs heterojunction semiconductor 131 Figure 44 (a) The conduction band and the valence band of the ASPAT diode structure

(b) the energy band diagram of the ASPAT diode structure when under three different

biases 132 Figure 45 The current-voltage characteristic of an ASPAT diode (100times100microm

2) and

(4times4microm2) using SILVACO Atlas simulator for structure device XMBE304 showing

excellent agreement between simulated and experimental data 134

Figure 46 IV characteristics of the dependencies of current on AlAs barrier 136 Figure 47 Example of analysis at -1 and 1V to the current 136 Figure 48 I-V characteristic of the dependencies current to Spacer I layer 137

Figure 49 Current changes with layer thickness l1 138 Figure 410 IV characteristic of the dependencies current to Spacer 1 layer 139

Figure 411Current change with layer thickness l2 139

Figure 412 Measurement and simulation comparison result as a function of temperature

range from 100K to 398K 142 Figure 413 Log Current vs voltage as a function of temperature for ASPAT sample

XMBE304 143

Figure 414 The effective barrier high of the Schottky barrier diode and the ASPAT

Diode [3] 144

Figure 415 Log Current vs voltage as a function of temperature for SBD sample

XMBE104 145

Figure 416 Influence of temperature on IV characteristic for both ASPAT diode and

SBD 146

Figure 51 IV Characteristics of measured ASPAT diodes for emitter sizes of 4x4um2

6x6um2 and 10x10um

2 Note the good scalability 149

Figure 52 Junction resistance versus voltage 151

Figure 53 The rectifying action of the diode at zero voltage which gives the ASPAT

high sensitivity near zero bias detection 152

Figure 54 One port S-parameter measurements 155

Figure 55 Fabricated open short de-embedding structure and real device (zoomed) use

for RF calibration and measurements (Note Images are not to scale) 156

11

Figure 56 Real S11 RF measurements for (XMBE304A) were done for four devices

from 15times15microm2 to 100times100microm2 device sizes for within-wafer uniformity check 158

Figure 57 Imaginary S11 measurements (XMBE304A) to qualify the repeatability and

reproducibility were done on four devices from 15times15 microm2 to 100times100 microm2 at zero

bias 158

Figure 58 Wafer to wafer uniformity test on 1st and 2

nd RF measurement(XMBE304B)

to qualify the repeatability and reproducibility were done for 30times30microm2 to 15times15 microm

2

device sizes (Real and Imaginary) Note blue colour is XMBE304A and red colour is

XMBE304B 160

Figure 59 RF measurement performed on device sizes of 10times10microm2 6times6mmicro

2 and

4times4microm2

(Real and Imaginary) Note that green red and blue colour represents 4times4microm2

6times6mmicro2 and 10times10microm

2 diodes respectively 162

Figure 510 RF measurement performed on device sizes of 10times10 microm2 6times6 microm2 and

4times4 microm2 (Smith Chart) Note that green red and blue colour represents 4times4microm2

6times6mmicro2 and 10times10microm2 diodes respectively 162

Figure 511 Cross-sectional of fabricated ASPAT diode and the corresponding

equivalent circuit model 164

Figure 512 The S-parameter Touchstone file is used to read the measured files 165

Figure 513 Equivalent Circuit Model of Open CPW or de-embedded structure 166

Figure 514 Equivalent circuit model for short de-embedded structure 166

Figure 515 Smith chart representative S-parameter measurement for short (left) and

open (right) CPW The blue lines represent simulated data and the red is measured data

167

Figure 516 Equivalent circuit of the ASPAT diode 169

Figure 517 Excellent fitting between measured (red colour) and simulated (blue colour)

for S11 (Real and Imaginary) results for various small device designs 169

Figure 518 Excellent fitting between measured (red colour) and simulated (blue colour)

for S11 results (Smith Chart) for various small device designs 170

Figure 519 Capacitance extraction showing full depletion at -025 volts (modelled

capacitance vs Voltage) 172

Figure 61 Block diagram represent a complete direct receiver system 177

Figure 62 The detection process of a single wave through a non-linear IV characteristic

of a diode 177

Figure 63 Lumped element illustration of microwave detector circuit 178

Figure 64 The mixing process where the signals are processed by the non-linear I-V

characteristic to produce an intermediate frequency (fIF=fLO-fRF) when signals at fLO and

fRF are applied to the diode 179

Figure 65 Measurement of Tangential Sensitivity[108 129] 181

12

Figure 66 The configuration circuit to verify 10th order polynomial equation extracted

from MATLAB to realize a virtual GaAsAlAs ASPAT diode 186

Figure 67 Verification of actual (blue measured) and virtual (red_10th order

polynomial) I-V characteristic of the 4times4 microm2 diode used in this study 186

Figure 68 Direct detector circuit topology using an ASPAT diode 187

Figure 69 Output voltage and detector sensitivity over wide range of input power 188

Figure 610 Simulated dependence of Voltage Sensitivity on incident power and load

resistance of the ASPAT detector 189

Figure 611 Junction resistance as a function of forward voltage 189

Figure 612 Curvature coefficient for the measured I-V characteristic with a device size

of 4times4μm2 191

Figure 613 Voltage Sensitivity against frequency at -20dBm fixed input power 192

Figure 614 Reflection Coefficient versus operating frequency without matching

circuitry 193

Figure 615 Detector circuit with impedance matching circuit placed in between diode

and source 194

Figure 616 Reflection Coefficient over wide frequency band with matching 195

Figure 617 Simulated voltage sensitivity of GaAsAlAs ASPAT detector at W-band

frequency 195

Figure 618 Lowest detectable signal at 100GHz operating frequency 196

Figure 619 The dynamic range of the ASPAT diode obtained from linear region of

diode operation 197

Figure 620 Simulation result of diode sensitivity for three devices geometries obtained

from the fabricated ASPAT in this work 198

Figure 71 performance of state-of the-art millimetre wave source [166] 202

Figure 72 Frequency multiplier has the property that foutput is an integer multiple of fin

204

Figure 73 Principle of operation for frequency multiplier utilising a non-linear

resistance [10] 204

Figure 74 A standard system for two port frequency multiplier circuit 207

Figure 75 Block diagram of frequency doubler with series ASPAT mounted diode 209

Figure 76 Empirical modelling of ASPAT doubler utilising Keysight ADS tool 210

Figure 77 Conversion loss and conversion efficiency as a function of input power 211

Figure 78 Conversion loss and Conversion efficiency as a function of output frequency

212

13

LIST OF PUBLICATIONS CONFERENCE PRESENTATIONS

PUBLICATIONS

1 MRR Abdullah Y K Wang J Sexton M Missous and M J Kelly ldquoGaAsAlAs

Tunnelling Structure Temperature Dependence of ASPAT Detectorsrdquo 8th UK-Europe-

China Workshop on mm-waves and THz Technologies 2015 Cardiff University IEEE

proceedings DOI 101109UCMMT20157460591

2 Yuekun Wang Mohd Rashid Redza Abdullah James Sexton and M Missous

ldquoInGaAs-AlAs asymmetric space layer tunnel (ASPAT) diodes for THz electronicsrdquo 8th

UK-Europe-China Workshop on mm-waves and THz Technologies 2015 Cardiff

University IEEE proceedings DOI 101109UCMMT20157460589

3 K N Zainul Ariffin S G Muttlak M Abdullah M R R Abdullah Y Wang and M

Missous ldquoAsymmetric Spacer Layer Tunnel In018Ga082AsAlAs (ASPAT) Diode using

Double Quantum Wells for Dual Functions Detection and Oscillationrdquo 8th UK-

Europe-China Workshop on mm-waves and THz Technologies 2015 Cardiff

University IEEE proceedings Doi 101109UCMMT20157460599

4 K N Zainul Ariffin M R R Abdullah Y K Wang S G Muttlak O S

Abdulwahid J Sexton MJ Kelly and M Missous ldquoAsymmetric Spacer Layer Tunnel

Diode (ASPAT) Quantum Structure Design Linked to Current-Voltage Characteristics

A Physical Simulation Studyrdquo UK-China Millimetre Waves and Terahertz Technology

Workshop September 2017 Submitted 14 July 2017 Conference held on 11th -13th

September 2017 DOI 101109UCMMT20178068358

5 K N Zainul Ariffin Y Wang M R R Abdullah S G Muttlak Omar S

Abdulwahid J Sexton Ka Wa Ian Michael J Kelly and M Missous ldquoInvestigations of

Asymmetric Spacer Tunnel Layer (ASPAT) Diode for High-Frequency Applicationsrdquo

DOI 101109TED20172777803

6 Omar S Abdulwahid S G Muttlak M R R Abdullah K N Zainul Ariffin J

Sexton Ka Wa Ian Michael J Kelly and M Missous ldquoA 100GHz Zero-Biased

Quantum Tunnelling ASPAT Detectorrdquo Submitted to IEEE TED on DEC 2016 under

correctionamendment Pending fabrication data

14

CONFERENCE PRESENTATIONS

1 Mohd Rashid Redza Abdullah J Sexton Kawa Ian MJKelly and M Missousldquo

G2040GHz Frequency Doubler Varistor Mode using ASPAT diodesrdquo UK

Semiconductors 2017 2017 University of Sheffield Oral presentation

2 M R R Abdullah YueKun Wang J Sexton Kawa Ian and M Missousldquo

Microwave Performance of GaAsAlAs Asymmetric Spacer Layer Tunnel (ASPAT)

Diodesrdquo UK Semiconductors 2016 2016 University of Sheffield Oral presentation

3 M R R Abdullah J Sexton and M Missousldquo GaAsAlAs Tunnelling Structures

THz RTD oscillators and ASPAT detectorsrdquo UK Semiconductors 2015 2015

University of Sheffield Oral presentation

4 Yuekun Wang Mohd Rashid Redza Abdullah and M MissousldquoInGaAs-AlAs

asymmetric space layer tunnel (ASPAT) diodes for THz electronicsrdquo UK

Semiconductors 2015 2015 University of Sheffield Oral presentation

5 Mohd Rashid Redza Abdullah and M Missousldquo GaAsAlAs Tunnelling

Structure Temperature Dependence of ASPAT Detectorsrdquo PGR Conference2016

2016 University of Manchester Poster presentation

6 YueKun Wang KNZainul Ariffin Mohd Rashid Redza Abdullah J Sexton

Kawa Ian and M Missous ldquoPhysical Modelling and Experimental Studies of

InGaAsAlAs Asymmetric spacer Layer Tunnel Diodesrdquo UK Semiconductors 2016

2016 University of Sheffield Oral presentation

7 K N Zainul Ariffin S G Muttlak M R R Abdullah Y Wang Omar S

Abdulwahid M Missous ldquoExperimental and Physical Modelling of Temperature

Dependence of a Double Quantum Well In018Ga082AsAlAs ASPAT Dioderdquo UK

Semiconductor Conference 2016 Sheffield 6 ndash 7 July 2016 Oral Presentation

8 Omar S Abdulwahid Mohd Rashid Redza Abdullah S G Muttlak K N Zainul

Ariffin Mohamed Missous ldquoTunnelling Barrier Diode for Millimetre Wave

Mixingrdquo UK Semiconductor Conference 2016 Sheffield 6 ndash 7 July 2016 Oral

Presentation

9 M Abdullah K N Zainul Ariffin MRR Abdullah J Sexton M Missous and

MJ Kelly ldquoA Novel In18Ga82As-AlAs Asymmetric Spacer Layer Tunnel (ASPAT)

Diode with Double Quantum Wells for Microwave Detectionrdquo UK Semiconductor

Conference 2015 Sheffield 1 ndash 2 July 2015 Oral Presentation

15

ABSTRACT

Thesis Title GaAsAlAs ASPAT Diodes for Millimetre and Sub-Millimetre Wave

Applications

Institute School of Electrical and Electronic Engineering the University of Manchester

Candidate Mohd Rashid Redza bin Abdullah

Degree Doctor of Philosophy (PhD)

Date 3 October 2017

The Asymmetric Spacer layer Tunnel (ASPAT) diode is a new diode invented in

the early 90s as an alternative to the Schottky barrier diode (SBD) technology for

microwave detector applications due to its highly stable temperature characteristics The

ASPAT features a strong non-linear I-V characteristic as a result of tunnelling through a

thin barrier which enables RF detection at zero bias from microwaves up to

submillimetre wave frequencies In this work two heavily doped GaAs contact layer on

top and bottom layers adjacent to lightly doped GaAs intermediate layers enclose

undoped GaAs spacers with different lengths sandwiching an undoped AlAs layer that

acts as a tunnel barrier The ultimate ambition of this work was to develop a MMIC

detector as well as a frequency source based on optimized ASPAT diodes for millimetre

wave (100GHz) applications The effect of material parameter and dimensions on the

ASPAT source performances was described using an empirical model for the first time

Since this is a new device keys challenges in this work were to improve DC and

RF characteristic as well as to develop a repeatable reproducible and ultimately

manufacturable fabrication process flow This was investigated using two approaches

namely air-bridge and dielectric-bridge fabrication process flows Through this work it

was found that the GaAsAlAs heterostructures ASPAT diode are more amenable to the

dielectric-bridge technique as large-scale fabrication of mesa area up to 4times4microm2 with

device yields exceeding 80 routinely produced The fabrication of the ASPAT using i-

line optical lithography which has the capability to reduce emitter area to 4times4microm2 to

lower down the device capacitance for millimetre wave application has been made

feasible in this work The former challenge was extensively studied through materials

and structural characterisations by a SILVACO physical modelling and confirmed by

comparison with experimental data The I-V characteristic of the fabricated ASPAT

demonstrated outstanding scalability demonstrating robust processing A fair

comparison has been made between ASPAT and SBD fabricated in-house indicating

ASPAT is extremely stable to the temperature The RF characterisations were carried out

with the aid of Keysight ADS software

The DC characteristic from fabricated GaAsAlAs ASPAT diodes were absorbed

into an ADS simulation tool and utilized to demonstrate the performance of MMIC

100GHz detector as well as 20GHz40GHz signal generators Zero bias ASPAT with

mesa area of 4times4microm2 with video resistance of 90KΩ junction capacitance of 23fF and

curvature coefficient of 23V-1

has demonstrated detector voltage sensitivity above

2000VW while the signal source conversion loss and conversion efficiency are 28dB

and 03 respectively An estimate noise equivalent power (NEP) for this particular

device is 188pWHz12

16

DECLARATION AND COPYRIGHT STATEMENT

No portion of the work referred to in the dissertation has been submitted in support of an

application for another degree or qualification of this or any other university or other

institute of learning

COPYRIGHT STATEMENT

i The author of this thesis (including any appendices andor schedules to this thesis) owns

certain copyright or related rights in it (the ldquoCopyrightrdquo) and he has given The University of

Manchester certain rights to use such Copyright including for administrative purposes

ii Copies of this thesis either in full or in extracts and whether in hard or electronic copy

may be made only in accordance with the Copyright Designs and Patents Act 1988 (as

amended) and regulations issued under it or where appropriate in accordance with licensing

agreements which the University has from time to time This page must form part of any

such copies made

iii The ownership of certain Copyright patents designs trademarks and other intellectual

property (the ldquoIntellectual Propertyrdquo) and any reproductions of copyright works in the thesis

for example graphs and tables (ldquoReproductionsrdquo) which may be described in this thesis

may not be owned by the author and may be owned by third parties Such Intellectual

Property and Reproductions cannot and must not be made available for use without the prior

written permission of the owner(s) of the relevant Intellectual Property andor

Reproductions

iv Further information on the conditions under which disclosure publication and

commercialisation of this thesis the Copyright and any Intellectual Property andor

Reproductions described in it may take place is available in the University IP Policy

(httpdocumentsmanchesteracukDocuInfoaspxDocID=487) in any relevant Thesis

restriction declarations deposited in the University Library The University Libraryrsquos

regulations (httpwwwmanchesteracuklibraryaboutusregulations) and in The

Universityrsquos policy on Presentation of Theses

17

ACKNOWLEDGEMENTS

First and foremost all gratefulness and praise is to Allah swt for everything in my

life He is the one and the only one who granted me knowledge health patience and

ability to complete this thesis as well as colouring the whole journey of my PhD

I give my deepest and sincere gratitude to my PhD supervisor Professor Mohamed

Missous for his time support patience and guidance throughout the journey of this PhD

studies His encouragements valuable advice precious ideas and a wealth of knowledge

amp experiences have had a direct inspiration on this research Special thanks also to our

experimental officer Dr James Sexton for not only sharing his knowledge advice and

semiconductor fabrication skills but also his effort in maintaining our clean room

facilities to a great level My gratitude also extends to Mr Mallachi McGowan for his

help and assist in the lab-related issue

I am also obligated to Prof MJ Kelly from University of Cambridge and Dr Kawa

Ian from ICS limited for their measurement of the ASPAT samples on realizing the RF

characteristics This collaboration effort can hopefully last longer in designing and

implementing the ASPAT MMIC detectors

My deepest appreciation also goes to my PhD colleagues Khairul Nabilah Saad

GMuttlak Omar AbdulWahid and Yuekun Wang for their support as well as working

together with me to realize this exciting project directly and indirectly A sincere

thankfulness similarly to my seniors Dr Md Adzhar Zawawi and Dr Fauzi Packeer for

their support during the first and second year of my research For other friends and staff

members under Prof Missous and Dr M Migliorato I will always remember the strong

bond and friendship we made

I am really fortunate that I been blessed with my motherrsquos care who always make doarsquo

for my success every day during my studies As for my beloved wife Dr Nik Maryam

Anisah Nik Mursquotasim who had always encouraged me supported me and gave me

patience through all the hardship in this journey thank you very much

Finally I also would like to thank and acknowledge my sponsor Majlis Amanah

Rakyat (MARA) for financially supporting me during this studies I am greatly indebted

with your kind support which was vital to my study

18

DEDICATION

This thesis dedicated to

My respected and beloved parentshellip

My loving wife dearest siblings and in-lawshellip

19

1 INTRODUCTION

11 Background

It is an undeniable fact that semiconductors have changed the world much further

than anything people could have predicted in the last 60 or 70 years ie after the lsquocats

whisker and vacuum tube eras This field of research has been expanding from year to

year starting from the discovery of the first semiconductor (silver sulfide) in 1833 by

Michael Faraday [1 2] and it still remains very active to the present Semiconductors

have a large range of applications and are not just limited to use in communications they

can be found everywhere in other applications from Earth to space The widespread

usage and sheer number of applications have led to it growing very quickly and

contributing greatly to the growth of World Economics Over time the successful

development of semiconductor growth techniques such as Molecular Beam Epitaxy

(MBE) has enabled researchers to tailor and precisely control the semiconductor

material for new electronic devices with extra functionalities This has led to the

development of advanced devices such as high electron mobility transistors (HEMTs)

and Heterojunction bipolar transistors (HBTs) for use in wireless communication

technology Given this development today electronic devices such as computers

handheld smartphone tablets etc are no longer perceived as luxury and attractive items

but rather have become crucial in everyday life Such devices provide the means to allow

for people to remain connected to each other via the sending and receiving of

information electronically The huge demand for such types of devices has resulted in

competition in both the electronic market and technologies which only goes on to

advance the semiconductor industry

Nowadays the demand for electronic devices characterised by high speed high

efficiency ultra-low power and low manufacturing cost has increased exponentially To

fulfil this growth in demand high data rate systems are required in other words the

system must work at a higher frequency for both the transmitter and receiver The

frequency of interest for advanced wireless communication is in the Millimetre and sub-

20

millimetre wave region which is around 30-300GHz and 300 - 3000GHz respectively

The second frequency region is also sometime known as the terahertz (THz)

electromagnetic region This band lies between the microwave and infrared frequency

bands From the first time it was revealed in the late 80s[3-5] the THz region has gained

a lot of international attention due to its unique properties and since then the motivation

to develop these devices has increased significantly To date the THz frequencies region

has shown its ability to fulfil various applications such as high-resolution imaging in

medical security and surveillance field atmospheric monitoring and environment radio

astronomy as well as compact range radars[3] to name a few

However despite these developments not much effort has been made in exploring

alternative compact THz devices As a result electronic THz devices are still in the state

of immaturity as compared to microwave and photonics devices This is due to their high

cost and absence of compact amp solid-state THz sources (oscillators) and receivers

(detectors) that are capable of operating at both room and extreme temperatures[6 7] A

great deal of work still continues to fill up the lsquoTHz gaprsquo (between 300GHz and 3THz)

used for the most important part of a communication system namely the front-end

receiver or first stage Such a system is responsible for receiving detecting and

processing the received signal to be translated into useful information Furthermore THz

receivers systems still require the best-integrated components such as source mixer and

detector to reach their complete competencies[8] The detector which remains the

critical part of the receiver system requires devices or components that are able to fulfil

the THz gap requirement Studies conducted over a number of years have found out that

the key element in improving THz detection relies upon the use of passive devices ie

diodes Based on these findings many types of diode ie tunnelling diode point-contact

diode and Schottky barrier diode (SBD) have been proposed for detection applications

Amongst microwave and millimetre wave detector diode devices the Schottky

Barrier Diode (SBD) is the dominant detector that has been used since the 1940s[9] The

reason for this dominance is the ease of fabrication of a SBD (by either a point-contact

or evaporated semiconductor-metal structure) and its ability to produce a non-linear

current-voltage (I-V) characteristic which is necessary for rectifyingdetecting diodes [9

10] SBDs also have high cut-off frequency good dynamic range and are low cost To

21

date the SBD has been able to detect signals up to 100GHz [11] 1THz[12] and as high

as 10 THz[13] However the current transport mechanism in a SBD relies on thermionic

emission and therefore is strongly dependent on temperature and means that using them

in extreme conditions ie military and automotive applications is complex The SBD

also suffers from high noise figure[14] and is susceptible to burnout at a modest pulse

power level this will limit the use of ultra-high frequencies and low power signal

applications Other diodes that share the same characteristics are Planar Doped Barrier

(PDB) Germanium Backward Diode (GBD) ie a type of Esaki tunnel diode These

diodes are well known and are reliably used as millimetre wave detectors However it

still proves inefficient to substitute the SDB with any of the previously mentioned

diodes This is due to some drawbacks such as strong temperature dependence limited

dynamic range fabrication complications and hence poor reproducibility (ie GBD) and

other circuit complexities

Hence there is strong compulsion to study examine and produce new detector

diode structures that are able to solve the mentioned diodes limitations and which have

high sensitivity larger dynamic range low noise strong independence to temperature as

well as being able to work efficiently in the high-frequency band and at zero bias The

advantages of working at zero-bias relates very much to the need for a system with less

power consumption so that the device (ie mobile communication) is able to run off

small batteries for a reasonable length of time eliminating extra biasing circuit as well as

noise Therefore a new tunnelling device namely the Asymmetrical Spacer Layer

Tunnel diode (ASPAT) developed by RT Syme [15 16] and refined by Missous et

al[17] has been examined in this work The ASPAT which is in essence a

Semiconductor-insulator-semiconductor structure relies on tunnelling through a barrier

to provide current compared to conventional thermionic emission in SBDs The ASPAT

diode has many advantages a zero bias turn-on voltage very weak sensitivity to changes

in temperature (due to tunnelling) very low noise large dynamic range high resistance

to pulse burn-out [18] and as demonstrated recently can be reproducibly

manufactured[17] The growing interest in THz frequencies nowadays makes the

ASPAT an excellent choice to fulfil all requirements for ultra-high speed applications

22

ie communication (mobile computer networking) radar (military equipment) scalar

analyser and built-in test equipment

In this work an ASPAT diode based on group III-V elements of the periodic table

comprising compound semiconductors of large band gap material Aluminium Arsenide

(AlAs) sandwiched between two lower bandgap Gallium Arsenide (GaAs) are used and

intensively examined The AlAs semiconductor which is ten-monolayer thick has

almost the same lattice constant as GaAs but has a larger bandgap Consequently in the

conduction band a thin barrier of the AlAs is formed from the arrangement of such

structure The structure is made up of GaAs and AlAs both materials are grown on

GaAs substrate using Solid Source Molecular Beam Epitaxy (SSMBE) Therefore in

this study the ASPAT diode will be referred to as ldquoGaAsAlAs ASPATrdquo diode The

conventional GaAsAlAs ASPAT diode has been developed and successfully fabricated

in two different stages This work was the first carried out using facilities provided by

the University of Manchester The first stage of the work was to qualify the

reproducibility and repeatability of growth and fabrication technique which is mostly

performed on larger emitteranode size The second was to develop conventional

ASPAT diodes that can perform at Millimetre and sub-millimetre wave frequencies and

which are comprised of small emitter area

Prior to this work full physical modelling using SILVACO design software was

undertaken to generate models and to fully characterise and identify the fundamental

physical phenomenon of multi-junction ASPAT diode Therefore insight into and

performance based on diode structure and electron movement can be understood and

predicted which lead to the crucial idea in helping and advising iterations to epitaxial

growth as well as diode fabrication The verification of the physical models must be set

as a priority goal by comparing the results of statistically fabricated measured data The

advantage of physical modelling is that it can help reduce materials resources cost and

fabrication time

Further research into the field has led to the development of two other types of

ASPAT diodes that are used to compare with the conventional GaAsAlAs ASPAT

diode Their configuration involved the use of a more advanced semiconductor

technology which comprises InGaAsAlAs materials and GaAsAlAs with InGaAs

23

quantum wells The latter was a novel ASPAT diode and the former is identified as

advanced ASPAT diode However these two advanced ASPAT diodes have not been

extensively studied in this thesis as they will be covered by other co-workers at

Manchester Hence due to these some important parameters are compared to the

conventional one as it is the main focus of this work In the case of temperature

dependent studies the DC characteristic of conventional ASPAT is compared to in-

house fabrication AuGaAs SBD All the ASPATs epitaxial layer materials structures are

shown in the following tables

Table 11 Sample XMBE304 XMBE307 and XMBE368 GaAsAlAs materials structure grown

on GaAs Substrates by MBE

Thickness (Aring)

Layer Material Doping (cm-3

) XMBE368 XMBE304 XMBE307

Top Ohmic1 GaAs (Si) 4times1018

~3000 ~3000 ~3000

Buffer 1 GaAs(Si) 4times1017

~50 ~400 ~50

Spacer1 GaAs NID 50 50 50

Barrier AlAs NID 28 28 28

Spacer 2 GaAs NID 1000 2000 1000

Buffer GaAs(Si) 4times1017

50 400 50

Bottom Ohmic GaAs(Si) 5times1018

~3500 ~4500 ~3500

Substrate GaAs (Si) 50000 50000 50000

Note that sample XMBE368 and XMBE304 are grown on doped GaAs

substrates Sample XMBE368 was grown un-rotated to study the effect of barrier

thickness variation

24

Table 12 Quantum wells sandwiching the quantum barrier for sample XMBE314 grown on a

GaAs substrate by MBE

XMBE314

Layer Material Doping (119836119846minus120785) Thickness (Aring) Bandgap (eV)

Ohmic Layer GaAs (Si) 4times1018

3000 142

Emitter GaAs (Si) 1times1017

400 142

Spacer GaAs Undoped 50 142

Quantum Well In18Ga82As Undoped 60 116

Barrier AlAs Undoped 28 283

Quantum Well In18Ga82As Undoped 60 116

Spacer GaAs Undoped 2000 142

Collector GaAs (Si) 1times1017

400 142

Ohmic Layer GaAs (Si) 4times1018

4500 142

Substrate GaAs

Table 13 Novel InGaAsAlAs ASPAT diode grown on InP substrate by MBE on InP substrate

XMBE326

Layer Material Doping (cm-3

) Thickness (Aring) Bandgap (eV)

Top Ohmic1 In053Ga047As(Si) 5times1019

3000 075

Buffer 1 In053Ga047As(Si) 1times1017

350 075

Spacer1 In053Ga047As NID 50 073

Barrier AlAs NID 283 283

Spacer 2 In053Ga047As NID 2000 075

Buffer In053Ga047As(Si) 1times1017

350 075

Bottom Ohmic In053Ga047As(Si) 15times1019

4200 075

Substrate InP (Si) NID 620000

From the above tables it can be noted that XMBE307 is the first batch that was grown

in-house using a Riber V100HU MBE machine followed by XMBE368 XMBE 304

XMBE314 and finally XMBE326 The two earlier batches were grown on n+

substrate hence their fabrication process flow is simpler On the other hand the three

other batches were grown on semi insulating substrate thus requiring the development

of new repeatable reproducible and robust process flow which will be covered in this

thesis

25

In general the fabrication of the ASPAT diode is based on top-down processes this

is because the ASPAT is a vertical structure device and the junction capacitance of the

ASPAT is directly associated with the size of the anodeemitter area Further to these

the capacitance directly influences the diode cut-off frequency Therefore the simplest

way to reduce the capacitance is by reducing the lateral area of the device of the ASPAT

structure since it can be represented by a parallel plate capacitor where the capacitance

is inversely proportional to the area of the device In order to achieve high cut-off

frequencies minimising capacitance via small dimensions ie sub-micrometre level is

essential However this will also increase the series resistance of the diode As a result

the cut-off frequency will be degraded Thus there is a trade-off between small

dimension of device and high cut-off frequency to be achieved

Finally successful growth and fabrication for small area GaAsAlAs ASPAT diode

in this work has led to carefully extracted RF characteristics This becomes a stepping

stone to designing a millimetre wave integrated circuit (MMIC) detector using empirical

modelling in Keysight ADS tools Therefore a predicted performance for a 4times4 um2

fabricated ASPAT is that can operate at 100GHz ASPAT as a zero-bias detector with a

voltage sensitivity of over 2000VW Additionally the design of a millimetre wave

source using similar ASPAT diodes was also carried out The performance of a 2040

GHz doubler using GaAsAlAs ASPAT in varistor mode is demonstrated for the first

time with a conversion loss of 33dB and conversion efficiency of ~ 02

26

12 Aims and objectives

The aim of this study is to further improve the performance of microwave and

millimetre wave technology by incorporating the Asymmetrical Spacer Layer Tunnel

Diode (ASPAT) for ultimate operation near THz frequencies by designing a range of

low power high-speed devices enhancing the methods of Simulation layout and

materials amp structural characterisations with fabrication process optimization using the

facilities available at the University of Manchester

There are three main objectives in this research firstly to streamline the physical

device design and modelling using the GaAsAlAsGaAs materials systems in order to

produce a zero bias detector which is basically a rectifier of a microwave signal by

using the SILVACO Atlas simulation tools

Secondly to achieve reproducibility and manufacturability of the fabrication

process for new type of GaAsAlAs ASPAT structure (lateral structure) hence small size

ASPAT emitter by improving the device processing technique and maximising the

capability limit of the conventional i-line optical lithography that is available in Prof

Missousrsquos group laboratory

Thirdly to optimise DC parameters through electrical properties investigation as a

stepping stone to the next objective that is to characterise the RF performance of the

GaAsAlAs ASPAT detector circuit The detection properties of microwave and

millimetre wave diode will also be investigated with different ASPAT diode size at

100GHz Further to these the properties of microwave signal source will also be

developed by way of utilizing the non-linearity feature of the diode Therefore this new

type of tunnelling diode can be applied to both applications of signal detection and

signal source in the microwave and millimetre wave ranges

27

13 Outline of this Thesis

This thesis is organized into eight chapters The first chapter discusses the

contextual information that motivates the undertaking of the study An overview of the

work which includes the details of the studied samples the aim and objectives of the

whole research project are also outlined in this chapter

Chapter 2 deals with the literature review of the basic principles and concepts of

the group III-V compound semiconductors The historical background of such

semiconductors which is essential to the development of ternary structures etc and the

advancement of semiconductor materials engineering is presented The types of existing

tunneling diode as well as conventional microwave diodes are also discussed and

compared The fundamentals of ASPAT diode which includes structural parameters and

its operation are then explained in detail Finally discussions of the ASPAT key DC

characteristics which are important for detection purposes are presented

Chapter 3 focusses on the development of the experimental techniques which can

be divided into two stages In the first stage the development is towards repeatability

reproducibility and manufacturability of the ASPAT grown in-house by MBE and

fabricated by conventional i-line optical lithography The second stage involves

optimisation and fine tuning such fabrication method for GaAsAlAs ASPAT samples

that can operate at high frequency ie 100GHz detector For both stages of the

fabrication process all techniques including mask design generic and special process

flow are presented The chapter ends with discussions on issues related to sample

processing and improvements that are proposed to solve these issues

Chapter 4 dwells on the modelling of the GaAsAlAs ASPAT using the SILVACO

simulation package The discussions are expected to offer a better understanding or

insight into each layer that forms the ASPAT diode structure The chapter begins with

discussions of the operation of the SILVACO Atlas tool A validation of physical

modelling is essential and presented according to the fabricated mesa sizes of the diode

28

Thereafter towards the end of the chapter the analyses of the relationship between

device current-voltage (I-V) characteristics the structural parameter including various

temperatures dependent simulations with a comparison to an in-house fabricated SBD

are offered

Chapter 5 presents relevant DC results based on optimized fabrication process and

RF characterization which enable obtaining an intrinsic and extrinsic element of the

GaAsAlAs ASPAT diode The discussions also highlight the analysis of DC zero bias

equivalent circuit and de-embedding extraction using ADS The chapter ends with

discussions on the RF reproducibility performance which includes the performance as

well for millimeter-wave and sub-millimeter wave applications

Chapter 6 discusses the main applications of ASPAT diodes The chapter begins

with discussions on detection theory followed by the parameters of interest and ends

with circuit design as well as the performance of a 100GHz detector The circuit design

was conducted using Keysight ADS software via harmonics balance simulation tool The

performance in term of sensitivity depending on measured ASPAT emitter size is

demonstrated Finally a comparison with conventional Schottky diode is presented

towards the end of the chapter

Chapter 7 discusses a secondary application that can be applied to the ASPAT by

utilizing the nonlinearity feature of the diode to create a signal source namely a 20 to

40GHz frequency doubler in varistor mode The doubler performance of ASPAT will be

explored through circuit design constructed via Keysight ADS simulation software Each

key parameter is highlighted and discussed in detail

The final chapter of this thesis that is Chapter 8 discusses the conclusions of the

study with emphasis on the overall key research findings The chapter also highlights

suggestions for further research in this particular field of study

29

2 LITERATURE REVIEW

21 Introduction

Since 1940s the development in the technology of semiconductor electronics has

been expanding and now has led to the establishment one of the most astonishing

industries of the 3rd

-millennium era Leading this advancement is the integrated circuit

(IC) or chip which was driven mostly by silicon (Si) Overtime the IC has undergone

substantial revolution in term of power economics size and efficient energy

consumption Currently it covers every aspect of human life ie from desktop personal

computers in the office and house to the compact smartphone in the pocket and from a

gigantic satellite in space to small satellite navigation in cars In other words

semiconductor technology is crucial to human life Without developments in

semiconductor materials engineering and shrinking of device size such accomplishment

may not have been realised today Therefore this chapter presents a macro view of the

development in compound semiconductor technology especially in radio frequency (RF)

towards Millimetre and submillimetre wave applications with regard to the improvement

of material and device structures

The essentials of group III-V compound semiconductor will be emphasised for its

points of interest and application in this field (RF technology) This chapter comprises

five main sections The first section is an overview of the semiconductor history with

concentration on its advantages and applications in the RF field while the second and

third discuss the effects of III-V compounds when the interface occurs between

semiconductor-semiconductor and semiconductor-metal respectively which leads to a

basic understanding of hetero-structures device as well as contacts namely Schottky and

Ohmic The fourth section is predominantly concerned with high-speed devices ie

diodes and materials in this field which leads to the exploitation of the main researchrsquos

device Then the following section describes in detail the background works basic

principle and intrinsic amp extrinsic parameters of the Asymmetric spacer Tunnel Diode

(ASPAT) Finally the basic way of characterising the device will also include giving an

overview of how the device is measured and what parameters are needed

30

22 Historical review of III-V Compound Semiconductor for RF applications

The beginning of commercial electronic devices was marked with the first point-

contact semiconductor transistor developed in 1947 by William Shockley at Bell

Laboratories in New Jersey Shockley developed a device based on a Germanium

Bipolar Junction Transistor (Ge BJT) structure [19] with operating frequency above 1

GHz Since then and until early 1950s the development of Ge BJT was fast and it

became foremost in the market of semiconductor technology However the emergence

of Silicon (Si) challenged Ge in the market in the 1960s Si has the upper hand primarily

because it has better electron transport and low manufacturing costs compared to Ge

[19] By the 1970s almost all RF transistors were based on Si BJT Additionally the

development of Si which forms a new material from the formation of native oxide

namely Silicon Dioxide (SiO2) led to the invention of the Metal Oxide Semiconductor

Field Effect Transistor (MOSFET) [19] The future of digital electronic industries has

been ldquobrightrdquo ever since the MOSFET was ldquobornrdquo as it has become a fundamental

building block component in complex microprocessors and flash memories Despite this

development the exploitation of Si at RF frequencies did not last long since Si is not an

optimum semiconductor for RF electronic devices The emergence of GaAs has

improved RF applications for high-speed transistors

Ge and Si which are categorised as single element semiconductor are the earliest

materials used to build the first transistor devices These devices played a crucial role

towards the development of more advanced material such as GaAs of the compound

semiconductor type[20] A compound semiconductor is a semiconductor formed by the

ionic bond of different types of semiconductor material most widely known as the group

III-V compound semiconductors The main reason for the progression of the III-V

compound semiconductors is due to their better electron mobility compared to the single

element semiconductors The term ldquomobilityrdquo in the semiconductor industry refers to the

easiness of movement of charges in many directions inside a crystal In fact it is

determined by the access resistances values with saturated velocity under certain values

of electric fields (bias) the higher the electric field the faster is the carrier movement in

the crystal Figure 21 shows the electron mobility and band gap for the most common

31

group III-V compound semiconductors Besides higher mobility III-V compound

semiconductors also have light-emission capability and are suitable for bandgap-

engineering techniques

The work on III-V compound semiconductors mainly on GaAs FETs led to a new

change for the whole RF electronics industry For example in 1966 the first GaAs

MESFET was invented[21] and achieved a maximum operating frequency of operation

of 3GHz [22] Three years later the frequency increased to 30GHz [23]

Figure 21 III-V compound semiconductors mobility and band gap[24]

With better features in terms of having a higher electron mobility compared to Si

electronic devices based on III-V materials developed rapidly This attracted attention in

many aspects especially in military radar application electronic warfare system missile

guidance control electronic for smart warfare system and secure communication To be

specific those demands were fulfilled through the application of microwave mixer and

detectors [25] which were achieved based on Schottky barrier diodes and FETs

However these applications remained largely as niche markets for use only in military

and exotic scientific projects until 1980 In addition to the microwave industry two

important diodes that played a large role in very high-frequency power source namely

the Gunn diode and the Impact Avalanche and Transit Time (IMPATT) diode which

were discovered in the 1960s[26]

100

1000

10000

100000

0 05 1 15

Bu

lk M

ob

ility

(cm

2 V

-1 s

-1

BandGap (eV)

InS

b

InA

s

Ge

Ga

Sb

In

GaA

s

Si

GaA

s

InP

32

Furthermore the invention of Molecular Beam Epitaxy (MBE) growth technique at

the beginning of 1970s has enhanced the full potential of the III-V compound

semiconductors[27] This technique has led to the formation of a new class of materials

and heterojunction device with high-quality interfaces and accurate control of the

thickness during growth[28] The advancement of material engineering that tailored the

III-V compound semiconductor with MBE effect has been beneficial for both three-

terminal and two-terminal devices As a result of this more advanced devices in both

electronics and optics were developed such as quantum well (QW) laser Resonant

tunnelling diode (RTD) high electron mobility transistor (HEMT) and many more[29]

The aim was to achieve high-speed devices transporting data at high data rates and

robust devices These devices promised an excellent option to conventional transistor

(three terminal devices) in high-frequency systems especially in the terahertz (THz) or

Millimetre and sub-millimetre wave regions [30]

One of the promising diodes that received a lot of attention is the resonant

tunnelling diode (RTD) which was first described in 1974 by Chang [31] This device

which consists of a double barrier and one quantum well is the classical tunnelling diode

Due to its good symmetrical non-linearity in its current- voltage characteristic it can be

exploited for signal generation and detection However the main focus of RTD to date

has been in the generation of continuous wave (CW) ultra-high frequency and to a lesser

extent in detection Therefore other tunnelling based diodes were developed specifically

for detection purposes which are the main foci of this work The PDB and ASPAT

diodes are the workhorse candidates for detection purposes Most of these are built based

on group III-V compound semiconductors [32]

Unlike the Very Large Scale Integration (VLSI) market ie CMOS for personal

computer (PCs) the RF electronic device for civilian application reached the consumer

market only in the late 1980s through satellite television with operating frequencies

around 12GHz [19] Since then many RF application have been deployed on the mass

market depending on their operating frequency such as 09GHz ndash 25GHz for wireless

communication 20GHz to 30GHz for satellite communication 77GHz for car radar

systems and above 90GHz for different sensor applications Utilising GaAs as the main

material RF devices have become the key underpinning components for modern

33

communication systems As a result in 1998 the volume production of mobile phones

was greater than that of PCs for the first time in history Presently production is being

made for devices like smartphones cellular phones mobile internet access and new

communication services and tablets

The development of the RF field is never ending More and more improvements are

being made especially through the design and fabrication of oscillators and detectors

which are mainly built based on group III-V compound semiconductors When RF

devices were used by the military (in the 1970s to 1980s) cost was not a concern

However after getting into civil application market (ie in the 1990s) the most frequent

issues highlighted were performance and cost[19]

The ability to generate or receive high operating frequencies with high power large

bandwidth and high sensitivity is an indicator for a good performance of RF devices

(depending on specific applications) For example the highest room temperature based

oscillator of up to 186 THz was achieved in thin well AlAs-InGaAs RTD by Professor

Masahiro Asada from Tokyo Institute of Technology [33] An excellent review on THz

sources can be found in [34] For ultra-high frequency detector and mixer applications

the two terminals RF device that is mostly used is the SBD In 1996 the highest cut off

frequency achieved by a mixer utilizing the SBD was about 5THz[35] and this has kept

increasing ever since The factor that motivates the development of THz devices is the

requirement to have a compact coherent source in the THz range Undoubtedly in the

future there will be very exciting times for enthusiasts of terahertz sources and receiver

as new generations of compact broadband and tuneable solid source device based on

advanced compound semiconductor are developed

23 The Concept of Heterostructures

A III-V compound semiconductor is mostly grown on a single semiconductor

substrate forming a layer called epitaxial heterojunction layer It is a starting point and

the key feature that brings the idea of realising the most advanced semiconductor

devices currently being developed and manufactured by combining several epitaxial

semiconductors [36-38] Heterojunctions have the capability of manipulating carrier

transport ie electron and holes transport in crystal separately unlike homojunctions

34

This has resulted in the successful development of new devices for high-speed and high-

frequency applications as well as optical sources and detectors [37] This section will

discuss lattice matched material pseudomorphic material hetero-junction band

discontinuities and quantum wells

231 Homojunctions Heterojunctions and Band Discontinuities

The term homo-junction refers to the interface between identical semiconductor

materials that have different polarity ie p-type or n-type but similar in energy gap This

phenomenon is usually applied in forming p-n junction diodes and can be understood by

referring to Figure 22 below

Figure 22 illustration of Homojunctions band structure material before (left) and after (right)

equilibrium

The materials A and B which have similar bandgap (Eg) and different dopant

types ie p-type and n-type will have their Fermi levels (Ef) closer to the valence band

(EV) and conduction band (EC) respectively before ldquothermal equilibriumrdquo Once

equilibrium is achieved Ef of both p-type and n-type will be aligned causing band

bending of EC and EV As a result a built-in electric field is introduced (via diffusion of

carriers) for both holes and electrons and forcing them to move in one direction

On the contrary a heterojunction occurs when the interface between two

semiconductor materials with different bandgap energy are brought together (ie large

energy band gap material combined with a low band gap one eg wide band gap AlAs

and narrow bandgap GaAs) This results in a steep band bending which leads to the

formation of energy band discontinuities at the junction as shown in Figure 23 In a

semiconductor heterojunction the most important parameter is the band gap energy

EC

EV

Ef

Eg

p-type E

g

n-type

EC

EV

Ef

Material A

p-type

n-type

Material B Material A Material B

35

associated with each material in the structure where the degree of discontinuity can be

utilised in varying the carriers transport properties as well as the quality of the junction

depending on the interest of the designer This leads to flexibility in tailoring device

characteristics leading to vastly improved performance of the device

Figure 23 Energy band diagrams before (left) and after (after) interface combination at

equilibrium

Based on Figure 23 above Material A indicated with blue line is a large band

gap energy material and Material B highlighted with the red line is a low band gap

material EV represents the valence band EC the conduction band and Ef is the Fermi

level of the materials Alternatively the two materials band discontinuities are denoted

by ΔEC for the conduction band and ΔEV for the valence band χ and Eg represent the

electron affinity and band gap energy respectively

At some point where by the Fermi energy of both semiconductor materials are levelled

the structure would have reached its thermal equilibrium The band gap of materials A

and B have a discontinuity at the interface (ΔEg) of these two materials In general this

is given by

120549119864119892 = 119864119892119860 minus 119864119892

119861 (21)

Furthermore when thermal equilibrium is achieved ΔEg is then divided between

conduction band and valence band discontinuities (ΔEC and ΔEV respectively) at the

material A and B junction interfaces Their relationships can be expressed as

EC

A

EV

A

Ef

A

Eg

A

Material A Material B

χA

χ

B

ΔEC

ΔEV

Ef

B

E

g

B

EC

B

EV

B

EV

A

χA

ΔE

C

ΔEV

EC

A

Ef

EC

B

EV

B

χB

Material A Material B

Vacuum

Level Vacuum

Level

36

∆119864119862 = 120594119860 minus 120594119861 (22)

∆119864119881 = (119864119892119861 minus 119864119892

119860) minus (120594119860 minus 120594119861) (23)

∆119864119892 = 119864119892119860 minus 119864119892

119861 = ∆119864119862 + ∆119864119881 (24)

However these relationships which were introduced by Anderson can only offer an

approximation In practice the results are always different since dislocation and

interface strain occur at the junction Therefore precise control during epitaxial growth is

always required and growth technologies such as MBE are employed In due course the

band gap discontinuity can be further exploited by using different types of material

combination Examples are GaAsAl052Ga048As and In053Ga047As In052Al048As [39]

232 Lattice-Matched and Pseudomorphic Materials

As discussed earlier a heterojunction happens when any two different

semiconductor materials that have different bandgap are joined together At the atomic

level both materials often differ in lattice constant The easiest way to explain this is by

setting the formation of heterojunction which can be separated into two types lattice

matched and lattice mismatched (pseudomorphic)

2321 Lattice Matched Systems

To create discontinuities for use as a high-performance device the combination of

semiconductor materials is essential Selecting the appropriate materials that have

similar or very close lattice constants to combine is crucial to avoid disruption at the

atomic level heterojunction interface Figure 24 shows that the lattice constant of a

material A ie substrate (aS) and material B ie deposited over layer (aL) are identical

or very close and their surface atoms are perfectly matched This scenario is known as

lattice matching

37

Figure 24 Lattice Matching for both materials when aL=aS

As can be seen in Figure 25 while there are restricted binary materials available to

form good heterojunction interfaces it is possible to combine semiconductor materials in

binary ternary and quaternary forms to allow the formation of a variety of lattice-

matched heterojunction interfaces The examples of materials that have successfully

been alloyed are In053Ga047As In052Al048AsInP and GaAsAlxGa(1-x)As (x=0 to 1)

Even though the materials system hetero-junction of these materials has close lattice

constant value their band-gap will experience an abrupt variation

Figure 25 Energy gap and lattice constant for direct and indirect band gap of compound

semiconductor at ambient temperature[40]

Material A(layer)

Material B(substrate)

aL

aL

aS

aS

38

The existence of binary ternary and quaternary semiconductors formed by alloying

semiconductors has expanded the opportunity for heterojunction formation in devices

The alloy semiconductor which is produced by the combinations of two semiconductors

A and B has a lattice constant that obeys Vegardrsquos Law as follows

119886(119886119897119897119900119910) = 119909119886119860 + (1 minus 119909)119886(119861) (25)

For the alloy the band gap normally follows the virtual crystal approximation

119864119892(119886119897119897119900119910) = 119909119864119892(119860) + (1 minus 119909)119864119892(119861) (26)

Table 21 shows the list of the semiconductor alloy band gap and lattice constant for

common binary and ternary for group III-V compound semiconductors[41]

Table 21 Lattice constant (a) and band gap for group III-V binary and ternary compound

semiconductors a room temperature [41 42]

Alloy Lattice constant a(Aring) Band gap EgeV)

AlAs 5661 2239

AlSb 6136 1581

GaAs 5653 1424

GaN 3189 34

InAs 6058 0417

InP 5869 1344

Al052Ga048As 5657 2072

In053Ga047As 5868 0773

In052Al048As 5852 1543

39

2322 Pseudo-morphic Materials

The other scenario is when two different materials with different lattice constants

are brought into contact The observation can be made at the atomic level where the

atom will try to match each other as shown in Figure 26 below

Figure 26 Lattice mismatched material

In fact for both situations (ie lattice matched and lattice mismatched) the atom of

the material at the hetero-interface will change their position to maintain the geometry of

the lattice Due to distortion at this atomic level a strain is then induced at the hetero-

interface In order to form a good hetero-junction interface the strain must not exceed a

certain specific critical value which will cause crystal dislocations to occur The result of

crystal dislocation is generally bad as it will affect the carriers which will be

concentrated in the defect area thus degrading the carriersrsquo mobility This then makes

the overall function or performance of the device to become poor

Nowadays the Molecular Beam Epitaxial (MBE) technique is able to grow epitaxial

layers of mismatched semiconductor layers profile ie mismatched in their lattice

constant (aLneaS) The growth method works when the grown epitaxial layer assumes the

lattice parameters of the layer it is deposited on Nonetheless the layers must be kept

within a certain limit and the deposited layer must be thin enough to avoid defect or

dislocation formations This new layer known as a ldquopseudomorphicrdquo material will alter

its original crystal structure and physical properties

Material A(layer)

Material B(substrate)

aL

aL

aS

aS

40

Figure 27 The formation of Pseudomorphic-layer in two situations (a) compressive and (b) tensile

strain [1]

Figure 27 shows material A in which the pseudomorphic materials can be related to

two situations compressive and tensile strain The compressive strain occurs when the

deposited layer has a larger lattice constant than the substrate (aL gtaS) while tensile strain

happens when the deposited layer has a smaller lattice constant than the substrate

(aLltaS) These leads to aL either to compress or stretch to fit aS respectively Note that

the pseudomorphic layers can only be grown to a certain critical thickness hc From

Figure 27 the strain between the substrate and the deposited epitaxial layer is given by

휀 =119886119871 minus 119886119878

119886119878 (27)

Where Ɛ is strain between two layers aL is lattice constant of the deposited layer and

aS is lattice constant of the substrate layer The concern in deposition of the over layer is

to avoid dislocation occurring at the interface if there is too much strain at the junction

The strain is naturally influenced by the thickness of the deposited layer and thus the

thickness of growth must be controlled below the critical thickness hc which is

expresses as

(b)

41

ℎ119888 =119886119904

(28)

Moreover one needs to appreciate that even though the crystal structure and their

physical properties change the total energy within the unit cell is maintained This is

possible by distortion of the deposited layer in the direction perpendicular to the growth

direction while leading to lattice matching in the lateral plane Example of lattice

matched materials is GaAsAlAs and pseudomorphic material is In08Ga02AsInP

233 Quantum well and 2DEG

A typical application of heterojunction interface is one in which utilises ΔEC and ΔEV to

form barriers for electrons and holes One example of barriers that confines these

carriers is known as a Quantum Well (QW) A QW is a layered semiconductor usually

very thin ie about ~ 100 Aring thicknesses in which many quantum mechanical effects can

occur It is formed by a thin layer of a low bandgap energy semiconductor material eg

GaAs sandwiched between two similar large bandgap energy semiconductors eg AlAs

or AlGaAs The growth technique to achieve thin layers of QW is usually MBE The

benefit of this method is that it allows the formation of heterojunction with very thin

epitaxial layer

Figure 28 The band diagram of an ideal quantum well formed by a narrow band gap

semiconductor eg InGaAs sandwiched between two large band gap semiconductors eg GaAs (a)

Structure (b) energy band diagram and (c) Conduction band diagram when AlGaAs is n-doped[43]

42

The thickness of the layer that can be achieved can be as thin as the electron mean free

path (De Broglie wavelength) which is around 100 Aring to 300 Aring [44] The expression for

the De Broglie wavelength is given by

120582 = ℎ120588frasl (29)

Here h and ρ are Planckrsquos constant and momentum of the electron respectively Figure

28(b) illustrates a quantum well formation in abrupt semiconductor interfaces It can be

observed that the heterojunction boundary will experience discontinuities at the edges of

the conduction band and valence band with a quantum well generated for the carriers

(both electron and holes) The quantised energy sub-bands in the quantum well structure

in Figure 28(b) can be determined from [43]

119864 = 119864119899 + (

ℏ2

2119898lowast) (119896119909

2 + 1198961199102)

(210)

Where 119864119899 = (ℏ21205872

2119898lowast ) (119899

119871)2

and n is the energy level index that can be n=1 2 3hellip

The dopants in a semiconductor with large band gap layers may supply the

carrier to the quantum well and this occurs when the base or bottom of the quantum well

is lower than the Fermi Level and hence the high energy donors will go down to the

well therefore creating a Two-Dimensional Electron Gas (2DEG) In the 2DEG the

electrons and holes move freely in the quantum well in the plane perpendicular to the

growth direction however they are not capable of moving in the crystal growth

direction (confinement direction)[45 46] The 2DEG phenomena can be seen in Figure

28(c)

24 Metal-Semiconductor Contact

A semiconductor device is incomplete if there is no connection between the

semiconductors and the outside world A metal which is usually gold (Au) or gold

germanium (AuGe) is diffused into the semiconductor to allow for electrical connection

from the outside world to the semiconductor and vice versa The metal-semiconductor

contact can be either a Schottky contact or an Ohmic contact The Schottky contact is a

43

rectifying contact while the Ohmic contact is a contact that provides a low resistance

path between semiconductor and metal

241 Schottky Contact

The Schottky contact is basically a metal contact to the gate to enter a region or

channel in a transistor Figure 29 shows a schematic band diagram of a metal-

semiconductor contact before and after contact (Schottky-Mott concept)[47]

Figure 29 Metal and semiconductor in two conditions (a) separation (b) in contact

In Figure 29 the work function of the metal is represented by qm while the

semiconductor work function is qS The qχ is the energy difference of an electron

between the vacuum level and conduction band edge ie known as the electron affinity

and qVn is the difference between the conduction band and Fermi level in the

semiconductor EV EC Ef is the valence band energy conduction band energy and the

Fermi level respectively

The metal and semiconductor are brought together as showed in Figure 29(a)

both materials are at steady state However when the metal and semiconductors are in

contact as illustrated in Figure 29(b) the electrons that flow from the conduction band

in the semiconductor into the lower energy state of the metal will cause the Fermi level

to be aligned in thermal equilibrium Due to this process the positive charge donor is

trapped in the semiconductor interface hence forming a depletion region Xdep

Thereafter the upward bending of the energy in the semiconductor takes place On the

qχ(s)

Eg(s)

Vacuum Level

EV

EF(m)

E

F

EC

qϕ(m)

qϕ(s)

Metal Semiconductor

qVn

qϕB

qϕ(s)

qVbi

X

qχ(s)

Eg(s)

Vacuum Level

EV

EF

EC

qϕ(m)

Metal Semiconductor

Xdep

(a) (b)

44

other hand the negative charge (electron) will be accumulated within a narrow region in

the metal interface The existence of two different charges at the metal-semiconductor

boundary generates an electric field This leads to a potential barrier qB as seen by

electrons in the metal moving into the semiconductor and a built-in potential qVbi as

seen by electrons in the semiconductor trying to move into the metal

The built-in potential qVbi is defined as follows

119902119881119887119894 = 119902empty119861 minus 119902119881119899 (211)

The barrier height empty119861 in the ideal case is specified by the dissimilarity between a metal

work function empty119898 and electron affinity of the semiconductor

119902empty119861 = 119902empty119898 minus 119902120594 (212)

Referring to Eq (212) above the barrier height empty119861 rises linearly with the metalrsquos work

function empty119898 Nevertheless this is only in theory as the presence of localised surface

stated at the edges causes empty119861 to become unresponsive to the metal work function

Consequently Eq(211) is then reordered to match the difference in metal and

semiconductor work function Thus the new equation becomes

119902119881119887119894

= 119902(120601119898 minus 120601119904) (213)

A Schottky contact appears when a metal-semiconductor contact has a large

barrier height (B ge kT) and low doping concentration in the semiconductor (ND le NC) In

the case when the metal-semiconductor contact is under some bias eg reverse bias the

semiconductor will react to a positive bias according to the metal by a voltage V=-VR

This condition will affect the built-in potential and leads to increase from Vbi to

(Vbi+VR) thus increasing the barrier height empty119861 in the semiconductor as well

Consequently electrons are less able to flow from the semiconductor and cross into the

metal Therefore the current flow will be very small This phenomena is shown in

Figure 210(a)

45

Figure 210 Energy band diagram of Schottky contact on n-type material under (a) reverse and (b)

forward bias

As can be seen from Figure 210(b) when a forward bias is applied the semiconductor

is biased negatively with respect to the metal by a voltage V=Vf This will result in a

reduction in built-in potential from Vbi to Vbi-Vf The electrons in the semiconductor will

lower the barrier height and a lot of electrons will escape into the metal causing a large

current to flow Thus a large current flow in the forward direction compared to the

reverse direction Essentially this is the origin why the Schottky contact is named a

rectifying contact [48 49] For the metal side both forward and reverse biases applied

do not affect the barrier high empty119861 because there is no voltage drop there

In this system the electron and holes are transported by a phenomenon called

Thermionic Emission (TE) which happens when the semiconductor layer is lightly

doped Nd lt 1x1017119888119898minus3 The electron will only be thermionically emitted into the metal

when the energy is higher than the potential barrier[50] There is another phenomenon

called Thermionic Field Emission (TFE) which happens when the potential barrier

thickness is very thin (thin enough) to allow the electron to tunnel through the barrier

This will be discussed in the next section as this phenomenon leads to the formation of

an ohmic contact

(a) (b)

46

242 Ohmic Contact

Basically an ohmic contactrsquos purpose is to provide a low resistance path from

the semiconductor to the outside world It is different to a Schottky contact as it is a non-

rectifying contact and does not control the current flow the I-V characteristic of an

ohmic contact is linear in both forward and reverse directions (equality in current flow)

The ohmic contact also has a small voltage drop across it compared to the voltage drop

across the device

If a metal and semiconductor are bought together unavoidably a Schottky

contact will be formed Therefore to create an ohmic contact some techniques to reduce

barrier height and width of the depletion region must be used ie increase Nd In carrier

transport theory there are three mechanisms of carrier transport across the barrier

Firstly the Thermionic Emission (TE) which happens when the carries are excited to

overcome the barrier when the thermal energy is present Secondly the Thermionic

Field Emission (TFE) occurs when the electronholes have enough energy to tunnel

through an adequately thin barrier and some has overcome the low barrier at the top

Finally the Field Emission (FE) which results when carriers can tunnel through the

entire barrier The FE is the most favoured mechanism in the ohmic contact approach

[51]

From the three mechanisms above the current can be determined by the following

equations

1) Current in Thermionic emission (Figure 211(a))

exp (empty119861)

119896119879

(214)

2) Current in Thermionic field emission (Figure 211(b))

exp [

(empty119861)

11986400119888119900119905ℎ11986400

119896119879

] (215)

3) Current in Field Emission (Figure 211(c))

exp (empty119861)

11986400

(216)

47

Where k is the Boltzmann constant T is the temperature 11986400 is the tunnelling parameter

and is related to the doping concentration radic119873119863 The barrier height is denoted by empty119861

Figure 211 Figure 10 N-type semiconductor Ohmic contact with different doping concentration

ND (a) Low (b) Intermediate and (c) high

Figure 211 shows that the carrier transport mechanism is varied by the doping

concentration (ND) As can be seen from Figure 211(c) the doping concentration here is

the highest and influences the depletion region width Xdep to become smaller Therefore

Field Emission (FE) becomes dominant This FE method is the favourite method for

ohmic contact formation [51] and will be utilised in the fabrication carried out in this

work

In fabricating practical devices the ohmic contact is often split into two types

alloyed and Non-Alloyed The difference between the two is that the alloyed type is used

when the semiconductor is doped with a low doping concentration ie less than

1x1018119888119898minus3 while the Non-Alloyed is designed for heavily doped semiconductors with

more than1 times 1019119888119898minus3 doping

The alloyed ohmic contact requires thermal annealing to have a good performance

for electron transport In multi-layer metals one of the metals has the role of donor or

acceptor which is used to increase the doping concentration of the semiconductor If a

temperature anneals eg 420˚ Celsius is applied the metal will diffuse into the

semiconductor and carry the dopant into the semiconductor Therefore a heavily doped

region will be formed and the depletion width becomes narrow establishing the ohmic

contact The key example of this is the usage of the Gold-Germanium-Nickel (Au-Ge-

Ni) alloy where the Ge is the n-type dopant [52] which diffuses into the semiconductor

(a) (b) (c)

48

and perform atom replacement in the semiconductor ie in GaAs it replaces Ga On the

other hand the Non-Alloyed does not require any thermal annealing as it already has a

very high doping concentration and will automatically reduce the depletion region width

The Non-Alloyed ohmic contact has some advantages such as reproducible contact

reduced processing time and good uniformity [53]

25 Asymmetrical Spacer-layer Tunnel (ASPAT) diode Background work

In this section some historical background of the Asymmetrical Spacer-layer

Tunnel (ASPAT) diode is given Since this is the first thesis reporting about this new

device it is worth to mention some historical background about this tunnelling diode

The ASPAT was first proposed by a group of scientists from General Electrical

Company (GEC) in 1990[16] The works led by Richard T Syme and assisted by

Michael JKelly Angus Condie and Ian Dale initiated the idea of launching a new type

of tunnel diode The idea managed to attract the interest of many parties following the

development of resonant tunnelling diode (RTD) which earlier had shown a promising

weak temperature dependence [54] However the interest in RTD is mainly limited to

microwavesub-millimetre wave generation For THz detection the requirement is to

have a significantly asymmetric IV characteristic Given this the ASPAT which has

only a single energy barrier and most importantly weak temperature dependence and

large dynamic range would be a promising candidate for this application

The development of ASPAT is a kind of reverse engineering since it was built

purposely to replace the earliest receiver diode especially the Schottky Barrier Diode

(SBD) which has strong dependence on operating temperature [55] From the time when

it was first revealed a lot of works have been done to realise this most sophisticated

tunnel diode The first attempt which was reported in [56] was meant to gather some

insights into the device by using the well-known Schrodinger and Poissonrsquos equations

for simulation The second attempt on the other hand was directed to physically grow

and fabricate the device Here the real problem occurs At the first stage of qualifying

this device it was found not to be manufacturable Since then a new tunnel diode

structure based on GaAsAlAs materials system was built by both MBE and MOCVD

Its microwave performance was then tested at 94GHz [18] The same paper also

49

reported performance comparisons between ASPAT and another microwave diode ie

Germanium Backward Diode (GBD) PDB and SBD

Work on these devices stopped due to the inability to commercialise the ASPAT

and other tunnel based devices [57-60] The problems associated with low-cost

manufacture of tunnel diodes are due to firstly the thickness of the AlAs barrier layer

the dependence of tunnelling probability (electron) through a single barrier is

exponential and varies by a factor of more than 350 for one monolayer change in the

AlAs barrier thickness[61] The tunnelling of the electron through a barrier is

proportional to the current through a barrier as a function of a bias across the AlAs

barrier[62] Secondly the bandgap which is predominantly happens to be a ternary alloy

with relative composition ie AlxGa1-xAs Here the x can vary the bandgap in the

semiconductor layer For the ASPAT a 1 change in x results in a 30 change in the

current[62] To design an ASPAT for microwave and THz applications the designer

often allows at most plusmn10 variation of the absolute current through a specific diode at a

pre-identified bias This implies that within a wafer the uniformity that the ASPAT must

achieve is less than plusmn01 monolayers while between wafer to wafer the reproducibility

in barrier thickness in average must be identically controlled[63] This explains why at

the qualification stage of investigating ASPAT there was a need to focus on

GaAsAlAs-based material to diminish further errors because of the change in x This

type of work on ASPATs has been carried out by other co-workers at Manchester and

Cambridge

Thereafter the work then focused on repeatability and manufacturability tests

These result in many attempts being carried but failed with unacceptable between wafer

to wafer reproducibility [61 64] The development of reproducibility and repeatability of

the ASPAT was pursued for over 10 years until precise control of the growth of the

thickness AlAs layer was finally achieved using MBE[65] [66 67] This achievement is

confirmed by a current density produced which varied by less than plusmn30 indicating that

the reproducibility of AlAs barrier of the order of plusmn 02 monolayers A final step to

achieve the level control for the ASPATrsquos AlAs barrier thickness was carried out and

resulted in a 1 standard deviation of the IV characteristics for both within a wafer and

different wafer (2 inches wafer size)[17] In the early stage of this work some

50

repeatability and manufacturability test was also carried out Once this vital step is

accomplished further investigation on the ASPAT was made most recently and which

will be covered in this thesis ie temperature independence[68] and new ASPAT

types[69] characterisation to achieve smaller device RF measurement and development

of THz detectors The material systems that have been investigated so far are

GaAsAlAs and InGaAsAlAs both in the ManchesterCambridge group

Recently the ASPAT was commercialised by Linwave Technology as a wideband

zero-bias detector diode This was done in April 2016 Although it is now on the market

the ASPAT remains immature in term of research and development A lot of work is still

required to enhance the device ie working at the sub-millimetre wave using ternary

material etc

26 Asymmetrical Spacer-layer Tunnel (ASPAT) diode The Basics

The basic building block of the ASPAT diode is based on heterojunction of three

multilayer semiconductor structures which have two different band gaps The structure

comprises a thin layer of wider-gap semiconductor sandwiched between two

semiconductors with narrower-gap forming a tunnel barrier The basic principle of the

ASPAT device is based on the exploitation of quantum mechanics theory using

heterojunctions interface According to quantum mechanics theory moving particles

(electrons) with less energy than the barrier height have a probability of appearing on the

other side of the barrier by a tunneling through it This can be achieved in conditions

where the barrier must be very thin (~ 10 monolayer ) This is in contrast to classical

physics where a particle must have kinetic energy at least slightly greater than the

potential barrier height in order to overcome the barrier otherwise the probability of the

particle to appear on the side of the barrier is zero

Since the ASPAT diode operation is based on tunnelling through a barrier one

needs to know that the tunnelling mechanisms can be classified into two types[44]

intraband and interband The latter is described as tunnelling that occurs from

conduction band to valence band (electron) and valence band to conduction band (holes)

This normally happens in bipolar device ie p-n junction diode which has n-type and p-

type doped regions On the other hand intraband refers to tunnelling which occurs when

51

electron tunnel from the conduction band of a semiconductor to the conduction band of

its neighbouring semiconductor The same thing happens to the holes in the valence

band The device with this type of tunnelling is normally a unipolar device which is

either p-type or n-type doped The ASPAT diode can be considered as a device that is

based on intraband tunnelling mechanism Therefore the focus will be entirely based on

its principles

261 Principle of Quantum Tunneling

Generally all tunneling diodes obey the concept of quantum mechanical

tunneling Quantum mechanical tunneling is a phenomenon where a particle is able to go

through an energy barrier higher than the kinetic energy of the particle and if it is thin

enough compared to the de Broglie electron wavelength (λ) If the electron wave is

greater than the barrier the probability of the wave to occur at both side of the barrier is

higher

Figure 212 Classical view of whether an electron is can surmount a barrier or not Quantum

mechanical view allows an electron to tunnel through a barrier The probability (blue) is related to

the barrier thickness

For the case of classical physics (Figure 212(a)) the particles can be confined by

energy barriers of a semiconductor if their kinetic energy is less than the barrier energy

The particles thus require higher kinetic to escape to other states this phenomenon is

called thermal emission In quantum mechanics (Figure 212(b)) the particle is

described in two ways as a wave and as a particle If the particle moves like a wave it

will carry all the waversquos properties Therefore it will not brusquely end up at the

En

erg

y

(a) Classical view (b) Quantum mechanical view

En

erg

y

En

erg

y

52

boundary of the energy barrier Hence when the particles collide with the barrier

(incidence) there will be a probability of penetrating the barrier if the barrier is thin

enough and has a finite height For thicker barrier the probability of a wave that can be

found on the other side of the barrier is very small However the possibility of the

electron wave to appear on the other side of the barrier is increased by thinning the

barrier The potential barrier of semiconductor material technology can be determined by

using Homojunctions structures with different doping profile This will result in a

difference in band alignment and multilayer heterojunction structure (different

semiconductors have different band gap) which includes semiconductors insulators and

conductors (metals)

The easiest way to explain the phenomenon is by considering a potential barrier

Epot(x) with barrier height E0 energy bigger than the total energy E as shown in Figure

213 the potential energy occurs in a finite space between 0ltxlta and is 0 outside

Figure 213 Rectangular tunnelling with incident reflected and transmitted wave function[70]

The electron outside the region of the potential barrier (xlt0 and xgta) is free to

move The effective mass of the electron is different inside and outside of the barrier in

real tunneling devices when implemented using semiconductor heterostructures The

quantum mechanical equations predict the wave nature of matter which states that matter

unveils wavelike properties under some conditions and particle-like properties under

other conditions The wavelike properties as described by the Schrodinger Formula of

E0 Transmitted Ψ = 119862119890minus119894119896119909

Incident Ψ = 119860119890119894119896119909

Reflected Ψ = 119861119890minus119894119896119909 Ψ = 119863119890minus119894119896119909

0 a

E

x

V(x

)

53

quantum mechanics represent a particle penetrating through a potential barrier most

likely as an evanescent wave coupling of electromagnetic waves[44]

To start the calculation of the tunnelling probability the Schrodinger equation is given

by

119894ℏ

120597

120597119905Ψ(119903 119905) = ΨΗ(r t)

(217)

Where ℏ is Planckrsquos constant (662606957 times 10119890minus341198982119896119892119904

2120587frasl ) Ψ(119903 119905)is the wave

function at position r and time t Η is the Hamiltonian operator given by

Η = minus

ℏ2

2119898nabla2 + 119881(119903 119905)

(218)

Where 119881(119903 119905)is the potential energy which is dependent on space and time 119881(119903 119905) is

considered zero for a particle traveling in free space without any potentials The plane

wave with vector r and t is given by

Ψ(119903 119905) = 119890119894(119896119903minus120596119905) (219)

This equation satisfies Eq (217) above under the condition where the particle is

travelling in free space without potential k is the wave vector which is equivalent

to2120587120582frasl and the angular frequency 120596 is 2120587 multiplied by the frequency

In the case of tunneling through a potential barrier the method of separation of

variables is used to simplify the problem as in the equation below

Ψ(119903 119905) = 119877(119903)119879(119905) (220)

It is assumed that the problem above is divided into time-dependent and time-

independent parts 119877(119903) is the spatial component and 119879(119905) is the time-based component

of the wave function The time dependent problem as shown above in the Schroumldinger

Equation (1) can easily be solved by filling up all the finite parameters The solution of

54

the time-independent part gives the tunneling probability For the one-dimensional (1D)

time-independent Schroumldinger equation[44]

119864120595(119909) = minus

ℏ2

2119898

1198892

1198891199092120595(119909) + 119881(119909)120595(119909)

(221)

E is the total energy and 120595(119909)is the spatial component of the wave function along the x

axis The combination to the wave function is given by

120595(119903 119905) = 120595(119903)119890minus(

119894119864119905ℏ

)

(222)

The time-independent plane wave solution of 120595 = 119890119894119896119909 satisfies the equation (221) for

any constant potential V0 in space Plugging in the wave solution yields the condition

that

119896 = radic2119898lowast(119864 minus 1198810)

ℏ2

(223)

Referring to Figure 213 also the barrier with exact rectangular shape with height E0 and

width W the solution of the wave functions and tunneling probability can be extracted

by using the below equation [44]

119879119905 =

[119862119890minus119894119896119909]2

[119860119890119894119896119909]2= [1 +

11986402 sin ℎ2 (119896119882)

4119864(1198640 minus 119864)]minus1 asymp

16119864(1198640 minus 119864)

11986402 exp(minus2radic

2119898lowast(1198640 minus 119864)

ℎ2119882)

(224)

For more complex barrier shape Wentzel-Kramers-Brillouin has simplified the

Schrodinger equation for tunneling probability of carrier which becomes[44]

119879119905 =

[119862119890minus119894119896119909]2

[119860119890119894119896119909]2asymp 119890minus2int |119896(119886)|119889119886

1198860 asymp 119890

minus2int radic2119898lowast

ℎ2 [119880(119886)minus119864]1198891198861198860

(225)

55

This means that the incident electron has a finite probability T of tunneling

through the potential barrier and this leads to the concept of tunneling probability as well

as a tunnelling current Therefore this becomes the basis of tunneling phenomena and

thus all devices which are related to tunneling can be modelled and analysed based on

this basic example The tunneling phenomenon is a majority-carrier effect and the

tunneling time is set by the quantum transition probability per unit time (which is on the

order of picoseconds) rather than the transit time concept [44 71] This enables the

tunneling devices to work at a much higher switching speed They can also be used in

high-frequency applications such as microwave circuit and high-speed oscillators

262 ASPAT Structural Parameters of GaAsAlAs materials System

Figure 214 Epitaxial layer structure of device XMBE 304 (lateral structure) used in this study

Conduction band profile

56

The core materials that make up the ASPAT diode in this work is based on

heterostructure of group III-V compound semiconductors Such materials are chosen due

their mature excellent properties and their band gap which can be tailored to fit the

desired design as well as to improve the carrier mobility In this work the primary layers

that form an ASPAT diode are very thin pure Aluminium Arsenic (AlAs) of thickness

ten monolayers buried in between dissimilar thickness of pure Gallium Arsenic (GaAs)

layers as can be seen in the red circle in Figure 214 above These two GaAs layers are

known as spacer layer which normally have a ratio of 401 or 201 in thickness The

asymmetrical spacers layer and the thin barrier in such arrangement lead to an

asymmetric current-voltage characteristics as proposed firstly by Syme and Kelly[15]

To examine and investigate this GaAsAlAs ASPAT structure in term of electrical and

RF characteristics the device have been grown according to Table 22 below

Table 22 epitaxial layer of sample XMBE304 which is the main focus of this work

Material Doping (cmminus3) Thickness (Aring) Bandgap (eV)

Ohmic Layer GaAs (Si) 4times1018 3000 142

Emitter GaAs (Si) 1times1017 400 142

Spacer GaAs Undoped 50 142

Barrier AlAs Undoped 28 283

Spacer GaAs Undoped 2000 142

Collector GaAs (Si) 1times1017 400 142

Ohmic Layer GaAs (Si) 4times1018 4500 142

Substrate GaAs - 650 microm 142

The arrangement of the multi-layers that form a lattice matched GaAsAlAs ASPAT

diode can be transformed into band structure profile view for easy understanding The

conduction band profile at equilibrium is as sketched in Figure 214 As can be seen in

the Figure 214 the ASPAT diode is generally a heterojunction multilayer structure

tunnelling diode

57

Figure 215 The layer profile of ASPAT with 28nm barrier thickness [27]

The generic structure of ASPAT diode which is shown in Figure 215 above with a

schematic band structure comprises the following (starting from the top)

(1) A thick layer of heavily doped n++

about 4e+18cm-3

of GaAs with a thickness of

approximately 300nm

(2) An intermediate layer of lightly doped n-type about 1e+17cm-3

of GaAs with

thickness of approximately 40nm

(3) A spacer layer not intentionally doped (NID) GaAs with thickness of approximately

5nm

(4) An ultra-thin layer of NID AlAs with thickness of approximately 28nm

(5) A spacer layer of NID GaAs with thickness of approximately 200nm

(6) An intermediate layer of lightly doped n+ ~1e+17

GaAs with thickness of

approximately 40nm

(7) A thick layer of heavily doped n+ about 4e+18 of GaAs with a thickness of

approximately 750nm

Each layer has its own role For instance layer (1) and (7) are used as ohmic

contacts via connection to a AuGeNiAu metal stack This explains why they are

purposely heavily doped (gt 1018

cm-3

) for better low resistance ohmic contacts Two

intermediate layers (layers 2 and 6) are used to prevent the carrier in the contact layers

from diffusing into the undoped layers The two unequal length spacer layers with ratio

58

1198971 1198972 of about 401 are used as voltage arms to yield an asymmetric current-voltage

characteristic The asymmetry means that after a positive bias is applied from the long

spacer region an accumulation layer is formed and it is deeper than that formed by the

negative bias The thin layer positioned in the middle (Layer 4) is the tunneling barrier

The performance of a single barrier ASPAT diode can be optimised depending on

the applications by appropriate selection of the material system so that the band gap and

barrier height of such material can be modified Furthermore the mobility of electron

and doping concentration of the contacts region can be tuned The parameters to tune

during the growth for instant growth interrupt time and growth temperature will also

affect the performance of this diode The key layers that will affect the performance are

the barrier thickness and the two spacer layers enclosing it The study has shown that a

one monolayer change in thickness results in 300 change in in tunneling current for a

fixed voltage point[65]

The following discussions account for the effect of the main structure of the ASPAT

which is related to their performance

2621 Barrier Thickness and height

The probability of an electron tunnelling through a barrier depends exponentially on

the width and height of the barrier as well as the energy that is incident on the barrier[72

73] All these will affect the I-V characteristic of the ASPAT diode The tunnel current is

obtained by summing over all incidents electrons energies with tunnelling probabilities

through the barrier The tunnelling varies approximately as[74]

119879 prop 119890minus120581119889 (226)

Where d is the barrier thickness and κ is defined by the expression below

120581 =

radic2119898lowast(1198810 minus 119864)2

ℏfrasl

(227)

From textbook the tunnelling probability is given by

119879(119864) = 412058121198702

[(1205812 + 1198702)2119904119894119899ℎ2119870119897 + 412058121198702]frasl (228)

Where K is expressed as

59

119870 = radic2119898lowast119864

ℏfrasl (229)

Where ℏ represents the reduced Planckrsquos constant (h2π) E and m are the electron

energy and effective mass respectively Thus by inputting appropriate value into these

equations one finds that reducing the barrier thickness by one monolayer increases the

tunnelling probability by a factor of nearly three for every electron that tunnels through

the barrier As a result the current will also increase Further it indicates that the

tunnelling strongly depends upon barrier thickness and height By contrast the current

does not strongly depend on temperature

2622 Spacer Thickness

The reason for having two dissimilar undoped spacer lengths is mainly to avoid

diffusion of dopant to the barrier and subsequent layers during growth but in the case of

the ASPAT diode the spacer can also act as a voltage arm Varying the thick spacer

layer (1198971) results in the reverse current decreasing as the layer thickness increases and

varying the thin spacer layer (1198972) will affect the forward current which increases as the

thickness reduces To obtain appropriate asymmetrical I-V characteristics one needs to

maintain an adequate ratio between these two spacer thicknesses While a too thin

1198971 results in high leakage current at reverse bias a too thick 1198972 results in low forward

current One also needs to keep it thick enough to prevent carrier diffusion

These two spacers must be kept undoped or very low doped to allow the electron

moving in the electron mean free path region as it is clear from ionised donors Under

large forward bias an accumulation layer is formed between the spacer and barrier

segment and it is more noticeable compared to the accumulation layer that is formed if a

negative bias is applied

60

Figure 216 Conduction band diagram showing band bending and 2DEG formation at the L1

spacer

Consequently a triangular well is formed which creates an emitter 2D electron gas

(2DEG) population The electrons in this 2DEG occupy the quasi-bound states which

mean high excitation energy thus allowing the electron to tunnel through the barrier as

depicted in Figure 216

In term of RF performance it is important to highlight that a thicker spacer layer

will affect the depletion region which gets wider and thus will reduce the junction

capacitance of the device as per the following expression

119862 = 휀0휀119903

119860

119889

(230)

Where A is the area of the device d is the thickness of the main device structure which

consists of spacers barrier and well layers 휀0 is the permittivity of free space and the

relative permittivity of the spacer material is denoted by εr However the intrinsic delay

time will also increase and hence degrade the device high frequency performance

Therefore optimisation through spacer thickness requires trade-off between reducing

leakage current at reverse bias and degrading device junction capacitance

GaAs

GaAs

AlAs

2DEG

Γ

X

61

263 ASPAT Electrical Parameters

The classical approach in determining the current flow through an ASPAT diode is

by solving the Schrodinger and Poison equations Prior work had been done by Syme et

al in 1991 Due to the fact that AlAs barrier is very thin tunnelling is assumed to occur

at the Gamma valley ie AlAs bandgap= 283eV (rather than X valley)[18 75] and only

from accumulation layer (2DEG)[59] Here the DC characteristic of the ASPAT diode

can be calculated by solving Schroumldinger equation with the position vector represented

by z (in this case) Thus the equation is expressed as

minus

ћ2

2nabla

1

119898lowast(119911)nabla120569 + |119890|120593(119911)120569 = 119864120569(119911)

(231)

Supposing the current is uniform across x and y planes then this can be simplified to one

dimension Therefore the 1 D Schroumldinger equations becomes

minusћ2

2119898lowast

1198892

1198891199112120595 + |119890|(120595 minus ∆120595) = 119864119911120595

(232)

Where

120595 =

120569

exp (119894119896119911119911)

(233)

where ∆120595 is the correction term which reduces the effective barrier height The

Schroumldinger equation is solved using different values for Ez thus the quantum

mechanical current density in the z-direction is now expressed as[15]

119895119911 =

minus|119890|ћ

2119898lowast(120595lowast

119889120595

119889119911minus 120595

119889120595lowast

119889119911 )

(234)

It is different for a heavily doped contact which can describe as below the envelope

functions in the left and right contacts respectively can be described by plane wave

120595119897 = exp(119894119896119897119911) + 119877 exp(minus119894119896119897119911) (235)

120595119903 = 119879119890119909119901[119894119896119903(119911 minus 120580119873)] (236)

62

In this case the left contact covers the region zlt0 while the right contact covers the

regions zgt 120580119873 Equations (35 and 36) are then inserted into Eq (34) to form the

following expression

119895119911 =

|119890|ћ1198961

119898lowast(1 minus |119877|2) =

|119890|ћ1198961

119898lowast|119879|2

(237)

Where R (Ez) and T (Ez) are the complex reflection and transmission coefficients

respectively and they are solved by using the transfer matrix method This method has

been described in reference [56] The next step is to integrate the current in the z-

direction 119895119911 for all possible Ez values Thus the expression for the current density

becomes

119895119911 =|119890|119898lowast119896119861119879

2120587ћ2int(1 minus |119877|2)

infin

0

119897119899 |1 + exp (

119864119891 minus 119864119911

119896119861119879)

1 + exp (119864119891 minus 119881|119890| minus 119864119911

119896119861119879

|

(238)

The equations above are used to calculate the current density approximation from

the ASPAT main structure (two spacer layers and one barrier) only and based on

intraband tunnelling from the conduction band profile For real fabricated structure the

calculation must take into account both intrinsic and extrinsic elements of the diode

While the latter is mostly related to the pad and probe that is used to extract the I-V

characteristic the first element mostly comes from the epitaxial layer of the ASPAT

diode itself The ASPAT I-V characteristic is shown in Figure 217 which clearly

indicates nonlinear characteristics and thus can be used for detection applications

63

Figure 217 I-V characteristics of a fabricated ASPAT diode

2631 Intrinsic Elements of the ASPAT diode

In order to extract the intrinsic electrical characteristic of the ASPAT diode a

generic structure as shown in Figure 218 is essential Two main sources of contribution

to the electrical characteristics are the interfaces of each layer and properties of the

materials themselves

Figure 218 Intrinsic Elements of the Asymmetric Spacer Tunnel Layer (ASPAT) Diode

64

The electrical current flowing from the top contact to the bottom contact will go

through each epitaxial layer producing a close loop Each junction interface limits the

current flow and sum up the total resistance resulting in what is known as the diode

series resistance (Rs) In the ASPAT main structure there is a junction capacitance (Cj)

due to the undoped regions surrounded by heavily doped contacts thus acting as a

parallel plate capacitor The fully depleted capacitance (Cj) of the diode can be

expressed as in Eq (230)

2632 Series resistance of the ASPAT diode

The total series resistance (Rs) of an ASPAT diode can be calculated based on the

finished fabricated diode structure In general Rs depends on three contributors namely

the non-uniformities in the contact metallization the un-depleted epitaxial layer (total

thickness) on both side of the heterostructures and the resistance caused by the

spreading current from the Mesa into the much wider second contact layer ie doped

substrate or 2nd

ohmic layer[76] In fact the contribution toward building up the total Rs

solely depends on how the structure is designed In this work two types of structures

were deployed namely lateral structures and vertical structures These will be described

in the next subsection

For both types of structures the series resistance (Rs) of the ASPAT diode consists

of aspecific Ohmic contact resistance (ρcA) contact epitaxial layer resistance (Repi-

Layers) and spreading resistance (Rspr)[33 77] where Rspr is influenced by the type of

structure The specific contact resistance is obtained from Transmission Line

Measurements (TLM) of the sample The theory of the TLM will be discussed in detail

in the next section The expression for the specific contact resistance is

120588119888 = 119877119888119871119879119908sinh119889

119871119879frasl

cosh 119889119871119879

frasl

(239)

Here Rc is the contact resistance LT is the transfer length (effective length) w is the

contact pad width and d is the length of the contact pad

65

Repi-Layer is the sum of all doped layers that sandwich the main ASPAT device For

each doped layer the resistance is given by

119877119890119901119894minus119897119886119910119890119903 = 120588

119871

119860

(240)

Where ρ is the resistivity which is given by 120588 = 1

120583119899119902119873119863 L is the epitaxial layer thickness

in cm A is the device area (emitter size) micron denotes the mobility of the electron q is the

electron charge and ND is the donor concentration The spreading resistance depends on

the structure design of the device This will be elaborated in detail in the following

section

26321 Vertical structure (doped Substrate)

The XMBE307 structures were grown on n+ GaAs substrate to provide the

simplest fabrication process The cross section of the finished single diode can be seen in

Figure 219 below

Figure 219 Cross-sectional view of ASPAT vertical structure the dimensions of d b and h are not

drawn to scale

66

At low frequency and in a mesa that is etched into a doped substrate material the

spreading resistance can be approximated by[76]

119877119904119901119903 =120588119904

2119889

(241)

Where ρs is the substrate resistivity and d is the ASPAT diode mesa length

However the spreading resistance is increased at high enough frequencies as the skin

depth (δ) in the substrate is much lower than the effective mesa length (d) of the diode

A new spreading resistance is then calculated also based on the assumption that the skin

depth is much lower than the chip thickness (h) but much larger than the mesa length

Thus the spreading resistance at high frequency is given by

119877119904119901119903(119891) =

120588119904

120587120575[05 ln (

119887

119889) +

119887]

(242)

Where the skin depth (δ) is taken from standard planar formula and can be expressed as

120575 = [

2120588

(120583120596)]

12frasl

(243)

Where micro is the permeability and ω is the angular frequency During DC measurement

this type of structure requires having good suction on the stage for a good contact

However for small die (15mm times15mm) the suction sometimes is not strong enough to

provide a very good adhesion to the sample Therefore another type of ASPAT diode is

deployed which is based on the lateral structure by utilising semi-insulating substrate

and both contacts are connected to probes

67

26322 Lateral structure (Semi-insulating Substrate)

Figure 220 Cross-section view of the complete process ASPAT diode (XMBE304) The dimensions

are not drawn to scale

In order to obtain accurate measurement results so as to avoid contact errors to the

substrate between stage and Device Under Test (DUT) a lateral structure as shown in

Figure 220 above is deployed This type of design offers many advantages ie it

provides a path for on-wafer RF measurement However the proper design has to take

into account the increase in RS due to improper attention to the spreading resistance

This spreading resistance is different from the vertical structure that was discussed

above In this case it is mainly caused by a gap at the bottom contact The gap in the

horizontal direction between epitaxial layer and metal at bottom contact is denoted as D

gap in in Figure 220 Therefore Rspr for the lateral design can be expressed as [77]

119877119904119901119903 =

1

120587120590119889119866119886119860119904ln (

119886

119886119898119890119904119886)

(244)

Where σ is the conductivity between two coaxial half-cylindrical electrodes with inner

(amesa) and outer (a) rectangular length or bottom ohmic layer which is given by (σ =1

ρ) a d and amesa are the length and thickness indicated in Figure 220 above Noticeably

68

the D gap will have direct effect on the outer length (a) of the device which is also

proportional to Rspr For high-frequency operation where the skin depth is less than

d(GaAs) σ becomes

120590(120596) =

120590(0)

[1 + (120596120591119903119890119897)2]

(245)

Where τrel =microme and micro m as well as e are the mobility effective mass and electron

charge respectively It is recommended that for high-frequency applications the device

series resistance must be as low as possible

Hence for both type of structure the ASPAT series resistance is calculated based on

all the above-stated resistances and these are set by

119877119904 =120588119862

119860+ 119877119890119901119894minus119897119886119910119890119903119904 + 119877119904119901119903

(246)

The total RS can be decreased by increasing the emitter area of the diode However a

large device will not able to reach millimetre and submillimeter wave region (THz) as

the capacitance will also increase (Eq 230) Therefore both parameters will have a

trade-off between them to be able to work at ultra-high frequencies

27 Characterization of Ohmic Contacts

The semiconductor Ohmic contact can be characterised using techniques that will

be described in the following section First is the Cox-Strack technique which is

specially designed to characterise bulk type semiconductor (thick) contact resistance on

two opposite sides The detailed description of this technique can be found in [78]

Second is a technique called Four Point Probe This technique was developed in 1954 by

Valdes etel [79] to characterise semiconductor resistivity It can also be used to

characterise the contact resistance for planar type devices As this research does not

cover this method the details of the measurement can be referred to [80] Finally the

most common method which is also extensively used in this research is the standard

Transmission Line Measurement (TLM) The details of this method will be explained in

the next section There are simplified versions of the TLM method which require just

one lithography step but are nevertheless very powerful in characterising and optimising

the contact resistance known as Circular Transmission Line Measurement (CTLM)[81

69

82] However in this research this is not to be covered as the standard TLM is already

adequate for planar type devices

271 Transmission Line Measurement (TLM)

The formation of metal and semiconductor interfaces will create a contact that

becomes very important for the characterisations of any fabricated device Additionally

it enables the quality of certain process flow to be determined This interface must be

evaluated by a technique known as the Transmission Line Measurement (TLM) TLM

which was first introduced by Murrmann and Widmand [83] in 1969 underwent some

refinements by Berger [84] in 1971 The theory of the TLM can be described by

constructing a TLM structure which comprises a set of metals contact pads placed in

series on a highly doped semiconductor layer as depicted in Figure 221 The structure is

designed like a series of the islands to permit current flow in parallel to the contact pads

[80] which is a direction defined by etched patterns Each contact metal pad behaves

like a MESA which has a thickness (t) and width (W) The distance between each metal

contact pads is defined by d1 the gap between two neighbouring contact pads which are

beneath each contact pad is defined as effective length LT This will allow current flow

in and out of the subsequent neighbouring metal pad The resistance elements that will

be extracted are RA and RB which sit under the contact and in between two metal pads

These two elements represent the sheet resistance under the metal contact pad area and a

sheet resistance of the material between two metal pads

Figure 221 A simple TLM structure with effective length and sheet resistance underneath

t L

T

LT

RA R

A R

B

dn

Probe Probe

Metal Pad

GaAs

MESA

70

The basic relationship of resistance R with respect to the size of the metal contact

or in the standard transmission line can be expressed as [44]

119877 = 120588

119871

119860= 120588

119871

119905 times 119882=

120588

119905times

119871

119882= 119877119904ℎ

119871

119882

(247)

Where ρ is the materialrsquos resistivity L is the length t is the thickness W is the width

Rsh is the sheet resistance and A is the cross-sectional area of the transmission line The

unit for ρ and Rsh are Ωm and Ωm2 respectively

The total resistance RT of this structure can be taken from the sum of the two

neighbouring padrsquos resistance RA and RB In order to relate with Eq (247) above this

RT will be substituted into Eq(247) to become

119877119879 = 2119877119860 + 119877119861 = 2119877119904ℎ119860

119871119879

119882+ 119877119904ℎ119861

119889119899

119882

(248)

As suggested in [85] RshA and RshB are assumed to be identical Therefore Eq(248)

can be reorganised into specific contact resistance RC and semiconductor sheet

resistance Rsh above as 119877119862 = 119877119904ℎ119860119871119879

119882 the new equation is then

119877119879 = 2119877119862 + 119877119904ℎ119861

119889119899

119882

(249)

The common practice throughout this research is to design a TLM structure that

has a ladder structure consisting of 10 metal pads with each one measuring to a size of

100microm width and 50microm length and the space between the first and second metal pad

5microm The gap is increased after the second metal pad by a further 5microm until ten metal

pads are completed produce a separation between the ninth and tenth pad of 45microm The

TLM ladder structure as depicted in Figure 222 is supplied by a constant current of

1mA at the very left and right metal contacts by two probes This allows the extraction

of RC (Ωmm) and Rsh (Ω) instantly from such structure The potential difference

between the two adjacent metal pads is measured by another two probes and the reading

of a voltmeter is recorded The total resistance is obtained by using Ohm law where

voltage is divided by current (VI) Another voltage reading is taken for the next two

neighbouring metal pads until the largest gap is reached The readings of (conversion

71

VI) which result in the corresponding resistance RT are then plotted against spacing and

the result can be seen in the graph in Figure 223

Figure 222 Top view of TLM ladder structure use in this work

Additionally in the measurement the voltmeter used in work has a very high resistance

Otherwise there will be leakage of current occurring through the probes and cables

Therefore the parasitic resistance of the cable or connector and the probe contact can be

ignored The key parameters that can be extracted from the graph will be discussed in

the next paragraph

Figure 223 Typical plot of resistance versus TLM spacing

y = 02592x + 02375

0

2

4

6

8

10

12

14

0 10 20 30 40 50Res

ista

nce

Rn (

Oh

ms)

TLM Spacing dn (um)

LT

d1 d

2 d

3 d

n

I (1mA)

V

V

LT

ME

SA

Su

bst

rate

72

The straight-line graph plotted in Figure 223 can be referred to Eq (249) and

this must be done by assuming the sheet resistance Rsh of the material is constant If the

straight line of the graph is extended up to initial gap (d=0) the intercept on the y-axis

provides the 2RC value To extract the 2LT further extrapolation is made until the

interception at RT = 0 is reached Therefore an important parameter which is the specific

contact resistance ρC can be found from this expression[85 86]

120588119888 = 119877119862119871119879119882sinh

119897119871119879

cosh119897

119871119879

(250)

Where l represents the total conductive semiconductor thickness The final part that can

be extracted from the graph is the slope of the line This is obtained by dividing the sheet

resistance with the width of the metal pad ie represented by expression (119877119904ℎ

119882frasl )

28 Basic Characterization Techniques and procedures

281 Measuring tools and apparatuses

The success of every experiment is determined by the backend results that are

obtained from the measurements It is important to choose an appropriate instrument

which will provide the required data for a valid and detail analysis Thus this section

will give a brief explanation of the measuring instruments and methods that were

exploited in this research The measurement apparatus systems that are available and

have been utilised in completing this thesis are ldquoset of DCrdquo and ldquoSet of RFrdquo

measurements The DC set measurement consists of room temperature and variable

temperature system This system is built to perform process monitoring during device

fabrication and it comprises of five main components

The fundamental component in the ldquoDCrsquos set toolrdquo is the Agilent B1500A

Parameter Analyser [87] used to provide fixed currentbias during testing The other

component is a Karl Suss PM5 Cascade Prober [88] which is used to receive fixed

currentvoltage and to supply its to the semiconductor contact via probe tip The PM5

Prober has at least four probe arms and each of them is fitted with ldquoneedlerdquo called probe

73

tip The size of the tip that is normally used here is 2microm and in some cases the tip size

of 1microm is also utilised The currentbias supplied to the sample must go through the two

of Source Measurement Units (SMU) namely SMU1 and SMU2 to ensure no mismatch

occurs between parameter analyser and diode All the testing are controlled using a

software called Integrated Circuit Characterization and Analysis Program (IC-CAP

2009) brought from Keysight Technologies[89] The software is installed on a standard

Personal Computer (PC) and the PC is connected to a General Purpose Bus Interface

(GPIB) to link with the B1500A Parameter Analyser This system can be organised

based on the purpose of measurements ie IV characteristics TLM Schottky Diode

and transistor as it is very flexible to change the configuration For examples

Transmission Line Measurement (TLM) configuration requires the addition of a digital

multi-meter and four-point probe tip while diode measurement just needs two point

probe tip without a digital multi-meter Figure 224 illustrates the measurement system

for a set of DC to test the TLM structure

Figure 224 The set of DC measurement apparatus arrangement to measure the TLM

On the other hand the set of RF measurement consists of five block elements

configuration in the system namely the Vector Network Analyser (VNA) DC

sourcemonitor Cascade Microtech Prober SMU and Control PC via GPIB The RF set

Agilent B1500A

Parameter analyser

DUT

(TLM Diode

Capacitance)

Karl Suss

PM5 Cascade

Prober

PC

(MS windows 2000)

Digital Multi-meter

ICCAP 2014 Provide current

source

Measure and read the

voltage Stage and Probe

Signal

Current Source

SMU1 SMU2

General purpose bus interface

74

performs RF characterisation after the device fabrication is completed This system can

also perform DC characterisation as its basic instrument has this function too The VNA

machine used in this research is the Anritsu 37369A [90] which can perform the

Scattering Parameter (S-Parameter) measurement with a frequency range of 40MHz to

40GHz The DC sourceMonitor utilised in this experiment is the HP 4142B Modular

DC SourceMonitor [91] Both of these sub-systems are controlled by a standard PC

which exploits GPIB port to link them During operation the HP 4142B is connected to

the VNA by an internal bridge network and two SMUs to the Cascade Microtech Prober

which is then connected to the bond pads of the device It is identical to the set of DC

measurement where the SMU setting and data assembly are accomplished by IC-CAP

software package therefore the data that was obtained before and after completing the

fabrication can be compared This will enhance the validation of the data

However the stage and probes in both sets of measurement are different The

Cascade Microtech Prober has only two probe arms with each arm fitted with a 3-

fingers probe tip in the arrangement of Ground-Signal-Ground (GSG) as shown in

Figure 225 Each finger (pitch) is separated by 100microm thus to fit in with the pitch and

to reduce the mismatch in resistance the bond pad design must follow this separation

between each contact The GSG is configured by connecting the outer pads (Collector)

to the Ground probe tip while the inner pads (Emitter) are attached to the Signal probe

tip where the RF signal is sent and received through it Figure 226 shows the actual set-

up for RF measurement used in this research

75

Figure 225 The set of RF measurement for on-wafer DC and RF one-port measurement

Figure 226 Actual VNA system that was used for RF characterization

282 Measurement steps using a VNA

The RF measurement steps can be summarised in the following

- Step 1 Select or find the suitable VNA depending on applications

PC

(MS windows 2000)

ICCAP 2014

VNA

Anristru 32379A

40MHz to

40GHz

HP4142B

Modular DC

source

DC Bias Source

Cascade

Microtech

Probe Station

One 50microm pitch G-

S-G probe tips

Ground

SMU1

Cathode

Gen

eral

purp

ose

bus

inte

rfac

e

Lo

w P

ow

er H

igh P

ow

er

Cathode

Anode

SMU2

Ground

Ground

Signal

76

There are few factors that need to be well-thought-out before starting to use a VNA

especially for S-parameter measurements The factors are the availability of the VNA in

term of operational frequency and measurement port types air-filled metallic waveguide

or on-wafer Not all VNA can have an operational frequency for banded measurement

ie W-band Ku-band etc All these may require external signal sources to extend the

operational frequency

- Step 2 Properly setting up the VNA

The VNA can be set up depending on its application and the goal of measurement for an

instant number of point requires a desired frequency span IF bandwidth and the supplied

power level This very important to ensure the desired measurements are correct and

appropriate

- Step 3 Appropriate calibration system

In order to have a valid calibration appropriate calibration method has to be chosen

depending on the applications This will determine the accuracy and standards of

calibration As for on-wafer calibrations the de-embedding is normally used while for

the off-wafer the SOLR technique is more suitable to employ

- Step 4 Validation or verification of the calibration results

It is vital to validate the calibration results to ensure that the system has been properly

calibrated

- Step 5 Proper measurement

Proper alignment positioning and touching from probes tip to DUT is necessary to

guarantee a good repeatability and reproducibility of measurement results Normally

when positioning the probe an alignment marker is used as an aid By doing this similar

travel distance for the probes can be achieved The measuring plane will also be equally

well-defined

283 Measurement Practice and Flowchart

Essentially the device characterisation is performed in two stages ie during

fabrication as a process monitoring and after completion for data collection and

analysis Figure 227 shows the block diagram of a flow chart for testing a 15mm times

15mm wafer processing performance In the wafer processing after reaching the top and

77

bottom contactrsquos step the sample can be examined by measuring the TLM structure

according to the TLM procedure The measurement is conducted by exploiting a set of

DC measurement apparatus as mentioned above analogous to the TLM ladder structure

on the wafer surface Based on the TLM results the presence of any process issue during

the fabrication can be identified by examining the parameter such as contact resistance

and sheet resistance Thus a decision can be made either to proceed or to terminate the

fabrication should any issue is found early on As a result no materials will be wasted

further When a dielectric layer is involved Capacitance Dielectric measurement can be

tested This practise can be used to obtain the quality of the dielectric layer To connect

between a diode effective area and probe a bridge is requires It can be attached to the

diode emitter (to determine the diode size) to a bond pad for probe tip to touch This

bridge can either be left hanging in air or sticking to the isolated dielectric layer For the

GaAsAlAs material system the latter technique is preferable since it avoids issues with

the air-bridge which will be discussed in detail in Chapter 3 The opening area (via) for

metal connection can be checked by measuring the resistance on a special design pad

after a plasma dry etching step

Figure 227 Block diagram of the ASPAT measurement step

The next stage of characterising the device is the on-wafer diode measurement

which takes place after completion of all processing (including bond padsco-planar

pads) The work is carried out using the above set for RF measurement and employed

Ohmic Contact

Opening Via

TLM

Qualified

Device

Bad device

Dielectric

capacitance

Bonding pad DC and RF

Good

Good Good

Fail

Fail

Fail

78

purposely to access the DC and RF performance of each diode where the current-voltage

(I-V) characteristic and S-Parameter results are obtained In fact DC measurements are

first performed using a set of DC measurement and a rough IV characteristic can be

obtained to ensure the diode is working properly Usually the yield of any fabricated

device on 15mm times 15mm wafer in this research is between 70 to 90 In this research

the outcomes that will be discussed in the subsequent chapters are in term of the average

of measured values Thus it is very important to have a meaningful data to compare with

physical modelling and simulation in the future The diode IV characteristics are studied

by applying different DC bias at the emitter to collector terminals to extract its keys

parameters ie turn on voltage (supposedly zero bias) non-linear characteristic Rj RS

Cj etc

The device that has been measured by DC and having produced a valid result

will be marked for the next investigation ie the S-Parameter measurement This

measurement is executed using a similar system but with different probe types The

three fingers probe type is used and the device frequency response is measured via the

one-port network from the VNA The important parameters extracted from this

measurement are usually S11 (depending on how many ports are measured) Although

working devices are selected to measure accordingly there is a need to ensure the VNA

RF cable and probe tips are calibrated so that only valid data without errors will be

obtained A calibration technique called SHORT-OPEN-LOAD (SOL) is performed

prior to each daily measurement by exploiting a calibration sample with WinCal

software (Cascade Microtech)[92]

To avoid confusion it is worth mentioning that this technique used to calibrate

the device structure is different from what is used in SOL calibration to the equipment

Furthermore the de-embedding calibration is made on-wafer with the same substrate of

the actual device whereas the SOL is performed on a special calibration substrate

Normally the de-embedding results are not constantly automated with the VNA

equipment However the measurement is done separately starting with the special DUT

substrate then followed by the OPEN and SHORT de-embedding structures

79

3 RESULTS OF FABRICATION PROCESS OF ASPAT DIODES

31 Introduction

This chapter presents in detail the general fabrication techniques for a generic and

development work of micron scale ASPAT diodes The discussions will focus on the

semiconductor growth technique used in this work ie MBE and the fabrication process

steps which include sample cleaning photolithography etching and contact

metallization These techniques are ample to build and deliver commercially marketable

fabricated structure Hence all photolithography techniques used to complete this

project are based on conventional i-line optical lithography which is adaptable to

industry and commercial purposes

The fabrication process in this work can be fragmented into two major works

firstly the fabrication towards reproducibility repeatability and manufacturability in

term of device structure process flow and DC amp RF characteristics This work will

involve relatively larger emitter area which varies from 15times15microm2 to 100times100microm

2 The

larger area provides for ease and fast fabrication as well as DCamp RF measurements

Once repeatability and reproducibility of the process flow and performance is confirmed

the second part of this work which is concerned with applications in millimetre and sub-

millimetre-wave then took place In this work small emitter designs varying from

2times2microm2 up to 10times10microm

2 were considered with appropriate measurement pads The

successful fabrication of smaller diode geometries in the second part of the programme

naturally leads to a further investigations to understand its epitaxial layer structure and

extracting intrinsic components ie junction resistance junction capacitance and series

resistance which will determine the device performance in high-frequency applications

The I-V characteristic is obtained from DC measurement which is usually performed at

room temperature Its results are then compiled and compared with advanced simulation

It is worth mentioning that all samples that are investigated in this project are grown by

means of MBE and the activities related to the epitaxial layer growth using MBE in the

University of Manchester were done by the Materials Growth Team (Prof Missous) and

the authors has no responsibility for this particular task

80

32 Epitaxial Layer Growth Techniques

Before discussing the principle of the common lithography technique it is

important to discuss the wafer growth technique as it comes first before the fabrication

The growth technique that is extensively used in this study as well contributing a lot in

the electronic semiconductor industry is Molecular Beam Epitaxy (MBE) The following

section will discuss the basic operating principle of solid source MBE

321 Molecular Beam Epitaxy (MBE)

The MBE technique was developed in the early 1970s [27] and is purposely used for

growing high purity epitaxial layers of compound semiconductors Such sophisticated

growth technique provides significant functionality ie precise control of the thickness

(to one atomic layer) and contributes to the growth of various types of complex

semiconductor multilayers high quality and advanced materials This level of control is

vital for an assortment of heterostructures devices that are being utilized as part of the

development of the advanced electronics devices especially for the ASPAT diode which

require 01ML control over the AlAs barrier to attain acceptable variability in device

characteristics Additionally the accurate doping profile and excellent junction

abruptness also can be achieved by using this technique

Practically the MBE system used in this work is a solid source MBE which

utilises beams produced by heating up various sources The sources can be Si Al Ga

As and other group III-V compound semiconductors When the crucibles which contain

the sources are heated atoms or molecules of the various elements are evaporated and

travel in straight lines paths like beams directed toward a target (heated and rotating

substrate surface) The condition of the vacuum during evaporation is ultra-high vacuum

(UHV) ~10-11

torr in order to have high quality crystals The substrate is heated and

rotated to provide good growth uniformity over large areas ( up to 4times4rdquo wafers) [93]

The growth rate in typical MBE growth is ~1ML second and can be controlled

by the source temperature in the crucible The abruptness at the heterojunctions interface

and switching of the growth compositions can be obtained by precise control of shutters

that are placed in front of the crucibles Therefore an abrupt junction at GaAs and AlAs

81

interface can be formed to realize the barrier in the ASPAT diode In order to monitor

the quality of the growing crystal and measure the layer by layer growth mode

Reflection High Energy Electron diffraction (RHEED) technique is utilised This

technique works based on the diffraction of electrons from the crystal surface [93]

Additionally given that the ASPAT current density is very sensitive to the barrier

thickness a study has been made using different growth techniques namely MBE and

Metal-Organic Chemical Vapour Deposition (MOCVD)[59] From this investigation it

was concluded that the percentage local variability of current density produced by MBE

grown diodes is better than those grown by MOCVD Thus in this study to get benefit

from its performance all wafers are MBE grown

33 Basic Principles of Common Fabrication techniques

This section covers the generic fabrication process which underpins

reproducibility repeatability and process optimisations for high-frequency applications

331 Sample cleaning

Essentially semiconductor processing requires a ldquoclean environmentrdquo to produce

devices The clean environment is classified according to how many ldquounwantedrdquo

particles are present in a cubic meter There are four categories of clean room available

in the industry Class 10 Class 100 Class 1000 and Class 10000 [94]

The fabrication of all ASPAT diodes in this project was performed in a clean

room environment of Class 1000 equipped with laminar air flow and filter system to

give Class 100 or better during processing Although the sample was processed in a

highly controlled particle environment there is still a high chance for a sample to get

contaminated when handled by a human Besides this the source of particle which

contributes to the contamination can be from the apparatuses and processing equipment

used in the laboratory themselves Thus the process of cleaning the sample wafer

surface before the start of each step is vital

Generally in a clean room the standard solutions that are used to clean a sample are

N-Methyl Pyrrolidone (NMP) Acetone Propan-3-ol (Iso-Propane-ol) (IPA) and

82

deionized (DI) water The sample which is cut up into 15times15mm2 size tiles is cleaned

based on the following procedures

1 Hot NMP- The sample is dipped into the solution at 80˚C for 10 minutes This

solution acts as an organic type of nature pollutants removal

2 DI water- acts as NMP remover The sample is then washed by flowing DI water

throughout the samplersquos surface

3 Acetone- to ensure any remaining NMP is completely removed from the sample

The sample is dipped into Acetone for 5 minutes in a low power ultrasonic bath

at ambient temperature

4 IPA- is used to remove the Acetone from the samplersquos surface The sample is

then dipped into the IPA for 5 minutes in a low power ultrasonic bath at room

temperature The use of low power for the ultrasonic bath is to avoid the sample

cracking or breaking

5 Once done the sample is then blow-dried using nitrogen (N2) gun to remove any

moisture coming from the IPA Fortunately it is easy to remove the IPA

completely from the sample surface given its higher rate of evaporation

Once all these steps are accomplished a visual inspection using a high magnification

optical microscope is conducted to ensure the sample surface is clean The cleanliness of

the sample is determined by the (lack) of particles or other spots (liquid mark) that can

sometime be observed during the inspection Obviously the lower the number is the

better the sample is as it is impossible to totally remove dirt especially marks

Sometimes it is hard to remove the particles in one go There is always a need to repeat

each cleaning step for a few times However this will not affect the sample in term of

electrical performance as the cleaning solutions used are non-destructive

332 Photolithography

Lithography or sometimes called pattern transfer is the most important step in

realizing microelectronic devices The designed geometry and dimensions on a quartz

glass plate called a ldquophoto-maskrdquo must be done prior to the fabrication process While

the photo mask can be designed by using various software tools in this work the design

83

is done via the Advance Design System (ADS) by Keysight The details of the design

which includes three different mask designs will be covered in Section 34 The mask

consists of the desired patterns (master) that can be printed onto solid materialrsquos surface

by means of an electrochemically sensitive polymer (photoresist) using

photolithography In fact this type of optical lithography technique can be performed

with and without a mask due to its simplicity It is also easy and cheaper compared to

other techniques such as x-ray lithography or Electron-beam lithography (EBL)

Although the latter is expensive it is still worth to have since it provides a higher

resolution which is preferable when developing sub-micrometre technology

processing[95] The special feature about EBL is that it does not need a mask to pattern

samples but can be produced by the movement of an electron beam point source and

hence writing the patterns directly on the surface

The ultra-violet (UV) based light sources are more popular among researchers

and development workers because they are cheaper and have modest resolution Usually

photolithography operates at wavelengths (λ) from 193 nm to 436nm The source that

provides the UV light is a Mercury (Hg) arc lamp which uses narrowband filters to select

single emission lines First is the i-line at λ= 365nm then the h-line which is of lower

resolution and has λ of 405nm and thirdly the g-line with a λ= 436nm[96] The

conventional optical lithography used in this research uses the ldquoi-linerdquo at a wavelength of

365nm For shorter wavelengths than these excimer laser or krypton fluoride laser with

a λ= 248nm and argon-Fluoride with a λ = 193nm are also used in the industry The

higher power levels enable higher productivity (throughput) while narrower spectral

widths reduce chromatic aberration provide better resolution and larger depth-of-focus

In this research all the fabrication process are done by utilizing a conventional optical

lithography (i-line) using a Karl Suss MA4 mask aligner Before starting any UV

exposure it is important to check the UV light intensity as it will affect the resolution

and thus desired device dimensions Therefore every corner that is exposed to the UV

light is calibrated to be at 09mWatt power exposure

The complete set of photolithography components consists of photoresist

(photosensitive polymer) photo-mask (chromium) which is used to block the UV to

form a pattern mask-aligner and developer (chemical solutions) Standard fabrication

84

process usually practiced at the University of Manchester starts with sample cleaning as

mentioned earlier Then the sample needs to go through heat treatment to remove all the

moisture with a temperature set to be 150˚C and bake for 5-10 minutes After having

cooled down the samplersquos surface is coated with a thin layer of photoresist via a

technique called spin-coating using a Laurell CZ-650 series spinner The spinning speed

is set depending on the type of resist ie 3000rpm for negative photoresist and 5000rpm

for positive photoresist The rotating speed of the spinner does not have much effect on

the coated photoresist thickness but will have consequence on the uniformity distribution

over the sample surface The coated thickness photoresist however depends on the

concentration of the specific photoresist Once spin-coated is done another short heat

treatment (1 minute) is required to ensure the resist is hard enough to contact a

photomask and to remove any excess solvent The temperature is set on the hot plate to

about 115˚C and 110˚C for positive and negative photoresists respectively

The important segments contained in the photoresist are a polymer (base resin)

a sensitizer and a casting solvent [97] The polymer will react by changing its structure

when exposed to the radiation While sensitizers will govern the reaction of the

photochemical in the polymeric phase the casting solvent will permit the spin-coated

application on the wafer surface The photoresist consists of positive and negative

photoresist both of which were used in this research Their basic difference is with

respect to the area that is exposed by UV light ie whether it will remain on the

semiconductor surface or will be removed In the case of a positive resist the exposed

area will be removed by the developer and the covered area will remain In other words

whatever is displayed on the photomask goes onto the sample surface On the other

hand for the negative photoresist the area that is covered from UV exposure will be

dissolved by the developer Figure 31 shows a 3D picture of both processes used in this

research

85

Figure 31 3D illustration of Optical lithography process used in this research

To obtain the desired pattern on the surface after UV exposure for a certain

duration the sample is required to go through a development procedure by using a

developer The developer is used to expel the dissolvable part of the photoresist after

being exposed to UV The usefulness of the developer depends on the photoresist ie

MF319 developer is specifically for positive photoresist while MIF326 is suitable for

negative photoresist Both types of developers will not harm the devices as they are a

kind of metal free ion solution Thus no free ion will change the characteristic of the

device The common practice in the clean room at the University of Manchester is to use

positive photoresist namely Shipley Microposit S1800 supplied by The Dow Chemical

and negative resist AZnLOF2020 (AZ2microm) which is supplied by MicroChem For the

S1800 series the thickness of the photoresist is determined by the last two digits ie

13microm thick for S1813 and 05microm thick for S1805 Both positive photoresists are used

Resist

GaAs

Dielectric

Photo Mask

Negative Resist Positive Resist

After Etching

86

mostly as protective area during wet chemical etching and as sacrificial dielectric layer if

higher temperature is applied ie 190˚C On the contrary AZ2microm which normally has a

thickness of 2microm is useful for patterning small dimension which leaves small gaparea

for the metallization process to fill It has good aspect ratio and useful in single layer lift-

off (post metallization process) In fact this type of resist can be thinned by diluting into

an Edge Bead Removal (EBR) solution and smaller device feature size can be obtained

The final dimension of certain devices (mesa size) is governed by the exposure

time the distance between photomask and samplersquos surface and the development time

The appropriate UV exposure time is required to avoid over-exposure which will cause

the spreading of light into the purportedly dark-field area[98] However the effect of the

exposure time differs between the photoresist types smaller opening area than the mask

pattern for negative resist and larger opening area for positive resist[45] The gap

between the photomask and wafer surface must be reduced as much as possible to avoid

the UV light going through the unwanted area To obtain a good gap value an applied

pressure from stage to the mask is required The normal pressure used in this research is

between 04 to 1 Pascal depending on the type of photoresist (negative and positive) as

well as its thickness Lastly the development time also influences the dimension of the

device Appropriate development time is required because ie over-development will

cause the polymerized photoresist to etch laterally resulting in bad patterned geometries

On the other hand under-development will cause non-uniformity in the surface after wet

chemical etching (positive resist) as well as causing lift-off problem (negative resist)

333 Etching Process

The etching process is a process of removing unwanted semiconductor layers to

define device geometry and isolate each individual device in one sample There are two

types of etching technique used in this research ie wet chemical etching and plasma

dry etching The wet chemical etching is based on Orthophosphoric(H3PO4) solution

which is a selective etchant to materials like GaAs InGaAs and AlAs The selective

etchant is referred to a solution that can etch away a specific semiconductor with a

specific etch rate The etch rate depends on the mixture ratio and concentration of the

solutions ie the higher the concentration the higher the etching rate In practice the

87

temperature humidity and epitaxial layer doping level also have an impact on the etch

rate Hence to minimize variation in the etch rate both temperature and humidity in the

clean room are constantly monitored and regulated Ambient temperature between 18degC

to 19degC is usually suitable while humidity is kept within 30 to 40 The advantage of

the wet chemical etching is that it is inexpensive controllable and with high throughput

highly selective and simple The mixture solution that is used in this work is

Orthophosphoric (H3PO4) Hydrogen Peroxide (H2O2) and Di-ionised water (H2O) with

ratios of 3150 3110 and 212 The ratio of 3150 provides an etch rate of about

600Aminute and is good to define the area and opening the under-cut in air bridges for

InGaAs samples[99] The 3110 ratio results with highly anisotropic shape but is easy to

control as the etch rate is about 1500Aminute for GaAs material system However the

212 etchant solution will provide extremely anisotropic etch rate of about

1000Asecond and is quite difficult to control For GaAsAlAs ASPAT diode it can still

be controlled since the thickness that needs to be removed is about 7000A (refer to

section 342 for details on the fabrication of ASPAT) Table 31 summarizes the etch

rate with different ratios and different selective materials Common to all the ratios

mentioned above are the isotropic side walls with lateral and vertical etch rate of 11

Table 31 Chemical wet etch characteristics using Orthophosphoric-based and Ammonia on GaAs

and InGaAs materials

Material Etchant Ratio Etch Rate (Aringminute)

GaAs H3PO4H2O2H2O 3150 600

GaAs H3PO4H2O2H2O 3110 1500

GaAs H3PO4H2O2H2O 212 60000

GaAs NH4OHH2O2H2O 118 2000~3500

InGaAs H3PO4H2O2H2O 3150 850

The plasma dry etch is purposely run to obtain extremely high anisotropic etch

profile vertically and horizontally It can be done via mask with positive photoresist and

self-align mask which is metal contact as a mesa protector The precursors that are used

to etch away GaAs InGaAs and AlAs layer in this technique are Methane and

Hydrogen (CH4+H2) On the other hand Carbon Tetrafluoromethane (CF4) and H2 is

88

used in removing Si3N4 The etch rate is determined by how much power is applied to

the plasma to hit the sample surface the pressure inside the chamber and amount of the

precursor In this research plasma Technology is used for dry etching This machine

which is a conventional OXFORD INSTRUMENTS 1990 machine can produce an etch

rate on average of 100Aminute for an RF power of 100mWatt

334 Sputtering (dielectric deposition)

In this work a Kurt JLesker PVD 75 is used to deposit Si3N4 layer on the sample

surface The deposition rate depends on the RF power that is applied To avoid surface

damage on the sample surface a sacrificial layer formed by SiO is deposited using the

Bio-Rad Thermal evaporator before transferring to the PVD 75 to start deposition with a

low (75Watt) RF power Once the deposition time reaches 30 minutes the power is

increased up to 200Watt As a result good uniformity of the dielectric layer is obtained

335 Metallization Process Lift-off and Annealing

The metallization process is a process in which metal contacts on semiconductor

devices are created The purpose of this process is to make a proper interconnection

between the semiconductor devices to other parts of the circuit elements In other words

it is to connect the semiconductor device to the outside world This process will allow

the device to be examined electrically so that all electrical characteristics can be obtained

ie resistance I-V curve capacitance conductance etc The metal scheme used in this

process depends on the semiconductor material In the case of GaAsAlAs ASPAT

diode Gold-Germanium (AuGe) Nickel (Ni) and Gold (Au) are used However in the

case of InGaAsAlAs ASPAT diode the metal scheme used is Titanium (Ti) and Au

The technique used in this process is resistive thermal evaporation

The metallization process starts by cleaning all the metallic sources and boats

ie tungsten boat Au Ni and Ti metals by using Trichloroethylene Acetone and IPA

consecutively for 5 minutes each in a high power ultrasonic bath This step is very

important to reduce the risk of contamination during thermal evaporation Once done all

the metallic materials are dried using a high-pressure nitrogen gun and then dipped in

89

Hydrochloric acid (HCL) solution for 2 minutes to de-oxidize the metals so that it has

minimal effect on the series resistance of the ASPAT diode

Two types of thermal evaporators were used extensively in this study both

Edwards Auto 306 (one denoted as Junior Auto 306) The latter is used to deposit alloy

type of metals while the former is used for the non-alloyed type of metals In the case of

GaAsAlAs ASPAT alloy type metals are used while for InGaAsAlAs non-alloyed

metals are used The cleaned metals are then loaded in the thermal evaporators and

placed on a resistive tungsten boat Each metal is placed on its specific tungsten boat to

avoid unnecessary mixture during the evaporation

Figure 32 Actual picture of thermal evaporator used in this study

Prior to loading the sample into the evaporator a standard fabrication process for

ASPAT diodes takes place by patterning the samples with AZ2microm negative photoresist

At the end of this step an opening area is created for the metal contact to be filled and

connected to the ohmic contact of the semiconductor The sample is also deoxidised

using a mixture of HCL and water in the ratio of 11 prior to evaporation This has to be

done in a very short time in order to avoid re-development of the native oxide layer

Inside the chamber the sample is securely placed on a chuck upside down facing the

filled tungsten boat The distance between the sample and metallic source boat is about

90

40 cm Figure 32 illustrates the actual thermal evaporation system used in this study

The thermal evaporator chamber is pumped down to reach a minimum pressure below

1times10-5

mbar before vaporizing the metal It is important that the mean free path between

metal amp sample is created and each vaporized metal stick firmly on the samplersquos surface

The normal practice in this study is to keep the vacuum pressure under 1 times10-6

bar so

that better device performance can be obtained The amount of current required to melt

down the metal is between 4 Amps to 6 Amp This amount of current is forced through

the tungsten boat and generates very high heat melting down the metallic source and

vaporizing it towards the sample surface The deposition rate for each metal can be

monitored by using a built-in film thickness monitoring (FTM) on the thermal

evaporator which proportionally depends on the amount of materials deposited The

GaAsAlAs sample is started with deposit of 55nm AuGe 13nm Ni and 500nm of Au

The reason of depositing AuGe first is due to the fact that the ohmic layer of the

ASPAT only can only be doped with a maximum doping of 4 times1018

cm-3

which is not

high enough for good conductivity Thus here Ge atoms will diffuse into the GaAs and

replaces Ga atoms during annealing process leading to higher doping levels (gt1 times1019

cm-3

) and hence improved conductivity

3351 Lift-off

The use of AZ2microm allows for the exact patterning of metals without the need for etching

using a single layer lift-off technique The negative photoresist also provides an undercut

profile which will create disjointedness between the desired metal pattern (on the

semiconductor) and undesired metal (on photoresist) The process of getting rid of the

unwanted metal from a sample surface is called lift-off The process starts once the

evaporation process is accomplished The sample is soaked into 80˚C N-Methyl-2-

Pyrrolidone (NMP) solution for usually 20 minutes (fast lift-off process) In most

instances the sample is in NMP for more than 12 hours in ambient temperature (slow

lift-off process) as the NMP solution is not destructive to the sample In this solution

the negative resist will be softened and the metal part which sticks on it will also be

eliminated from the sample As depicted in Figure 33 the lift-off process for a single

device shows the usual photoresist undercut profile observed To ensure that NMP

91

residues on the sample surface are completely removed DI water is used to rinse the

sample which is then blown dry with a nitrogen gun

Figure 33 Single layer lift-off process using negative photoresist

3352 Annealing

The alloyed (AuGeNiAu) metal stack requires a thermal treatment called annealing to

improve the ohmic contact between metal and semiconductor The sample which has

deposited top and bottom contacts is loaded into an annealing furnace at a temperature of

420˚C for 2 minutes In the case of GaAs during thermal annealing Ge atoms penetrates

into the GaAs crystal for approximately 70nm-250nm depending on evaporated

thickness of the metal layers annealing temperature as well as time [100] In this work

the total metal thickness evaporated for each contact layer is around 500nm This

thermal annealing treatment will also melt down the Au if it is too thick and is subjected

to too long a heat treatment Therefore it is not advisable to do annealing after the

sample is coated with bond pad metals as it can result in short circuited devices

sometimes

Az2micro Negative

photoresist profile

after UV exposure

and development

Evaporation to

form metal layer

(AuGe Ni Au)

Desired metal

contact

Metallisation

process

Lift-off undesired metal

92

34 GaAsAlAs ASPAT Process Optimization

As mentioned earlier this section present details two major process flows of the

fabrication process for the ASPAT diodes which utilised three stages of development of

photomasks design namely a ldquoFirst generation mask designrdquo (1st Gen) a ldquoSecond

Generation mask (2nd

Gen)rdquo and a ldquoThird Generation mask (3rd

Gen)rdquo design The 1st

and 2nd

generations mask designs are the designs that were produced in the first stage of

this work to develop the fabrication process know-how and to get familiarized with the

actual fabrication techniques in the cleanroom The difference between the 1st Gen and

2nd

Gen masks is in term of the development towards realizing Air Bridges and

Dielectric Bridges which were mostly covered by the 2nd

Gen mask design The

analyses in term of reproducibility repeatability and manufacturability for process

control and current-voltage characteristic as well as ultimately RF measurement are

obtained on relatively large size devices via these two mask designs The large area

emitter dimensions range in size from 15times15 microm2 to 100times100microm

2

The 3rd

Gen mask design was designed based on the optimization of the 2nd

Gen

mask which was to realize ASPAT diodes that are able to work at very high operating

frequencies Therefore in such design the ASPAT devices have to have a minimal

amount of capacitance and low series resistance this is can be achieved by shrinking the

emitter size of the diode to the smallest area possible as well as optimising the

fabrication process ton reduce parasitics The smallest fabricated devices designed on the

new mask has an 2times2 microm2 MESA area which also includes Ground Signal Ground bond

pads for both device and de-embedding structures (open and Short) for RF

measurements

Before the commencement of any fabrications and designing any layouts on eg

Si GaAs InP etc the epitaxial layer must be grown first to a desired design In the

University of Manchester the epitaxial layer structures are grown using one of the two

Molecular Beam Epitaxy (MBE) machines which are either the RIBER V90H or the

V100HU system Both systems are managed by Professor Missous Epitaxial wafers or

sample grown by each system are identified by a prefix and numbers that are prefix

VMBE for the V90H system and XMBE for the V100 system The epitaxial layers are

93

grown on four-inch wafer diameter The maximum diameter that can be grown on using

the V100H is 8 inches but generally single 4rdquo or 4x4rdquo wafers are used The wafers are

then diced and cut using a diamond scriber into 15mm times 15mm tiles for easy handling

and fabrication in the D12 cleanroom lab in the University of Manchester

In this section the structures fabrication and performance of the various ASPAT

diodes for both repeatability and high frequency applications will be discussed further

The ASPAT is manufactured on wafer sample XMBE304 which is a GaAsAlAsGaAs

lattice matched to a GaAs semi insulating substrate is the main focus The ground works

on this ASPAT such as the initial design and fabrication process flow optimization had

been conducted by fellow PhD colleagues in the group led by Professor Missous at the

University of Manchester

341 ASPAT Devices used in Fabrication

3411 XMBE368 and XMBE307

Table 32 Epitaxial layer of Doped substrate samples

Thickness (Aring)

Layer Material Doping (cm-3

) XMBE368 XMBE307

Top Ohmic1 GaAs (Si) 4times1018

~3000 ~3000

Buffer 1 GaAs(Si) 4times1017

~50 ~50

Spacer1 GaAs NID 50 50

Barrier AlAs NID 28 28

Spacer 2 GaAs NID 1000 1000

Buffer GaAs(Si) 4times1017

50 50

Bottom Ohmic GaAs(Si) 5times1018

~3500 ~3500

Substrate GaAs (Si) Doped 50000 50000

These two samples were the first batch of diode structures used in this work and

were grown using the RIBER V100 MBE machine A great deal of work was expanded

to ensure that it is able to produce appropriate non-linear I-V characteristics The work

carried out including finding suitable fabrication process steps mask designs process

control limitations ie etching rates etc The results obtained from processing these

94

samples mostly on large area anode and cathode sizes and their analysis included both

growth profiles and fabrication process flows These samples were grown on doped

GaAs substrates As such the finished diodes were vertical structures and the fabrication

process has marked differences compared to undoped substrate samples While

XMBE368 had similar epitaxial layer profile to XMBE307 during growth the AlAs

layer was set to be stagnant (ie no-rotation of the substrate during growth) to

investigate the effects of slight variations in barrier thickness

3412 XMBE304

The next sets of samples were all grown on semi-insulating GaAs substrates

Table 33 details the epitaxial layer profile of sample XMBE304 the main work horse

of this research work The growth of this structure was performed on a multi-wafer

platen and consisted of 9 x 2rdquo wafers (from XMBE304A to XMBE304I)

Table 33 The epitaxial layers of sample XMBE304 with barrier 10ML~28nm

Epitaxial layer Material Doping(cm-3

) Thickness(Aring)

Emitter GaAs(Si) 4e+18

3000

Emitter 2 GaAs(Si) 1e+17

400

Spacer GaAs NID 50

Barrier AlAs NID 28

Spacer 2 GaAs NID 2000

Collector GaAs(Si) 1e+17

400

Collector 2 GaAs(Si) 4e+18

4500

Substrate GaAs(SI) Semi-Insulating

For a typical ASPAT structure the emitter is essentially highly doped to

4times10+18

cm-3

to provide accumulation of electron in the emitter contact region It was

purposely highly doped to also achieve low ohmic contact with the metal The spacers

are used to avoid diffusion of dopants to the subsequent layers The ASPAT structure as

mentioned earlier has a single AlAs barrier with a very small thickness sandwiched

between two different length GaAs spacer layers

Batch XMBE304 is the main focus in these studies All activities required for

repeatability reproducibility process flow and devices as well as qualifying new or

95

optimization fabrication technique for small devices and RF performance which were

then used for high frequency were based on the set of wafers grow in this batch

342 Fabrication process of GaAsAlAs ASPAT diode toward repeatability

reproducibility and manufacturability

As the new wafer structures have to be tested and evaluated to gauge the

performance of the ASPAT diodes their uniformity also needs to be tested so as to

ensure it exhibited fully functional diode with zero bias detection in minimal variation in

IV characteristics between diodes As described previously 4rdquo wafers are always diced

up into 15mm times15 mm size for ease of handling in the cleanroom and for masks cost

purposes

3421 Doped n+ Substrate Wafers

For wafers grown on n+ substrates (XMBE368 and XMBE307) a two-level mask

set is sufficient to complete the fabrication process In this case the devices are designed

with mesa structures and top metal contact on the upper surface of the wafer and a

bottom contact on the backside of the wafer One mask plate is used for defining the

mesa structure (wet etch) and the other mask plate is used for defining the metal pad

Since the mesa is relatively large (30times30microm2

to 100times100microm2) there is no requirement to

define bond pads for measurement purposes A prior RTD mask designed by fellow

colleague Dr Md Adzhar for his PhD work was used for processing the ASPAT diodes

as well in the first instance The detail fabrication process is summarized in appendix I

for the doped substrate wafers Figure 34 shows typical IV characteristics for 30x30microm2

emitter size diodes obtained from sample XMBE368

96

The I-V measurement was taken from two different tiles (15mm times15 mm) located

on top and bottom of a 4rdquo XMBE368 wafer The location 1 marked as a blue line in

figure 34 refers to a device on a tile taken from the top of the 4rdquo wafer while location

2 (red) refers to a device on a tile taken from the bottom of 4rdquo wafer At 07V the

separation difference in current is ~228 for both tiles This shows that the device in

tile 1 is less resistive than the sample in tile 2 implying that the AlAs layer for tile

located on the top corner of the wafer is thinner and thus allows more electrons to tunnel

through it at a given bias

3422 First Generation Mask Design (1st Gen)

-0015

0005

0025

0045

0065

0085

-15 -1 -05 0 05 1 15

I C

urr

en

t (A

mp

)

V Voltage (Volt)

Current vs Voltage (XMBE368_1) for 30x30microm2

MidLeft2

MidRight3

Figure 34 Current-Voltage characteristic of sample XMBE368 used

in this study at two different locations on the wafer tile

(a) (b) (c)

Figure 35 ASPAT mask design for (a) 100times100microm2 (b) 30times30microm

2 and (c) 15times15 microm

2 diode dimensions

designed in the 1st Gen Mask

Location 2

Location 1

97

The fabrication of the GaAsAlAs ASPAT conducted in this 1st Gen mask is

followed the established generic process flow for InGaAs material which was done by

former co-worker Dr Md Adzhar Md Zawawi[101] The generic process flow is fairly

simple consisting of four mask steps as shown in Table 34 below The fabrication

process starts with the sample being cleaned in a NMP solution at a temperature of 80˚C

with Acetone and Propan-2-ol (IPA) The purpose of cleaning using NMP and Acetone

is to remove any organic material The IPA is used to remove Acetone residues

Table 34 Generic fabrication steps established by Dr Md Adzhar [101]

Step number Process

1 Top Contact

2 Mesa Etch

3 Isolation

4 Bottom Contact

In the 1st Gen mask the first step in fabricating the GaAsAlAs ASPAT diodes is

to use the first mask to define the emitter contact area The emitter has three different

sizes 100times100microm2 30times30microm

2 and 15times15microm

2 The lithography technique uses the

negative photo-resists AZ2micron with 55 second UV-photolithography to pattern the

top contact Then it is developed using MC319 developer to clear and define the

exposed area for the metals to stick to The sample is then subjected to plasma etching to

remove all organic residues and contaminants Then it is dipped into a mixture of

diluted Hydrochloric acid and water HCL H2O with a concentration of 11 for de-

oxidation This must be done in a short time right before the evaporation in order to

ensure good contact between metal and semiconductor surface with very low oxide

formation in between The ASPAT device performance depends on this step as contact

to the channel is by means of current flowing through the anode to the cathode terminals

In our lab the approach taken to achieve the Ohmic contact is by evaporation of Gold

Germanium (AuGe) Nickel (Ni) and Gold (Au) metals layers on top of the cap layer

Subsequently the metal is defined via a lift-off process using NMP

98

The next critical step is to define the MESA or island The mesa etch mask is

designed with two options with 05 microm or 10 microm tolerance The different mesa

tolerances are introduced to act as a safeguard for the emitter from producing excessive

undercut caused by lateral etching This step is to isolate and eliminate the unwanted

GaAs which will electrically link the active layers as many devices will be fabricated at

the same time on the same wafer tile (15mm times 15mm size) The lithography process in

this step uses positive resist and is developed using MC326 developer This step is

achieved using a wet etch process where a non-selective etchants mixture of

H3PO4H2O2H2O etches down the epitaxial layers until it slightly exceeds the AlAs

barrier with an etch rate of about 600Aring to 900Aring per minute The outcome of the MESA

or island step is the formation MESA active layers which are surrounded by inactive

layers of semi-insulating substrate (when using semi-insulating substrates)

The isolation mask is a step that is basically the same as the MESA etch step

eg using similar etchants mixture same lithography process but a different mask This

step is used to etch down until the GaAs substrate is reached which means that the

etching time is longer than that in the MESA etching step The purpose of this step is to

fully isolate the device from other neighboring devices hence ensuring no electrical

connection exists between each device within one sample Since these ASPAT diodes

employ an air bridge structure of size 1times5 microm the two minutes mesa etches will

simultaneously provide an initial undercut through lateral etching for the air bridge

formation

Based on the results so far obtained using the first-generation mask which

provided large mesa areas the current voltage characteristics of the ASPAT as far as the

non-linear zero bias is concerned did work very well However there was a need to

reduce the ASPAT mesa area down to very small dimensions to achieve mm-wave or

THz detection frequencies It is certainly a general rule for semiconductor device which

operate at very high frequency to have extremely small lateral dimension which

minimises the capacitance within the ASPAT device Furthermore for wider adoption of

the technology it is also important to develop a simple reproducible and low-cost

fabrication method for ASPAT diodes Details of this fabrication process are attached in

appendix II

99

3423 Second Generation (2nd

Gen) mask (ASPAT-GSG)

This work intended to enhance the 1st Gen-Large Area ASPAT photomask by

adding many features including 2 times 2microm2 mesa areas ground-signal-ground (GSG) bond

pads to enable RF measurement and three options device processing eg Air Bridge

Dielectric Bridge for semi-insulating doped substrate and dry etch-mesa Other reasons

for designing this mask was also to qualify process steps when deploying thin (1microm

width) bridge to connect small mesa area ASPAT diodes to the co-planar GSG bond pad

for DC and ultimately RF measurements

The mask was designed to fulfil the basic rules of fabricating various types of

tunnelling diode for instance RTD and ASPAT The diodes layouts were designed using

the commercial software Advanced Design Software (ADS) from Keysight

Technologies Ltd Once the design was completed it was ready to be sent to Compu-

Graphics Company for printing and patterned on a special chrome coated glass plate

The new 2nd

Gen mask design was termed ASPAT with Ground-Signal-Ground

(ASPAT-GSG) and consisted of two main designs ldquoAir Bridge and Dielectric Bridgerdquo

where each contained eight diodes with different emitter sizes (100times100μm2 50times50μm

2

30times30μm2 20times20μm

2 15times15μm

2 10times10μm

2 6times6μm

2 and 2times2μm

2) The Air Bridge

design is comprised of Design 1 (doped substrate) and Design 2 (undoped substrate)

This can be selected by changing the order of each individual layer of the mask steps

The same options are applied for Dielectric Bridge design which included processing for

doped and undoped substrates The details of the masks will be explained in the

following paragraph

(1) Air Bridge

Design 1 Contains five steps or layers mask of size 15mm times 15mm

suitable for air-bridge for undoped semi-insulating substrate

Design 2 Consist of seven layers steps mask of 15mm times 15mm size

There are 413 die chips in about 6mm times 6mm sizes in this design Figure 36 shows both

type of design for Air Bridge mask processing

(i) Mask 1- Top Contact

(ii) Mask 2- MESA

100

(iii) Mask 3-Isolation

(iv) Mask 4- Bottom Contact

(v) Mask 5- Collector Bond pad with GSG

(vi) Mask 2A-Dielectric

(vii) Mask 6A-Collector Bond Pad with GSG

(2) Dielectric Bridge

Design 1 Consists of seven identical steplayers masks with lateral

lengths of 15mm times 15mm each suitable for fabrication on semi-insulating

substrate

Design 2 Also has seven steps layers masked with dimension of 15mm

times15mm each suitable for doped substrate processing

There are 357 die chips in this type of mask design with an estimated size of 6mm times

6mm separately The smallest emitter size that is connected with a dielectric as the

bridge is 6times6microm2 The smallest 2times2microm

2 diodes was designed with the air-bridge

connected to the emitter bond pad due to the difficulties of opening viascavity for less

than 2 microm2 devices Figure 37 show both options for Dielectric Bridge mask processing

(i) Mask 1- Top Contact

(ii)Mask 2- MESA

(iii)Mask 3-Isolation

(iv) Mask 4- Bottom Contact

(v) Mask 5-GSG bond pad

(vi) Mask 6- Via Dielectric

(vii) Mask 7- Dielectric Bridge

(vii)Mask 5A- Via Dielectric

(ix) Mask 6A-GSG bond Pad

(x) Dielectric Bridge

Figure 36 Air Bridge design (red circle) for large mesa area (100times100microm2) device with

option for doped substrates

Figure 37 The layout of 1st design of Dielectric Bridge (green circle) mask design for 100 times

100microm2 emitter size with option for doped substrate processing

101

34231 Fabrication Process of the Air-Bridge Design

The fabrication using 1st Gen mask as mentioned in Section 3422 is less

complex however the fabrication process for Air-Bridge mask design contains a few

additional steps which are to add bond pads for the Ground-Signal-Ground radio

frequency (RF) layouts If the sample is on a doped substrate another layer needs to be

added leading to a total of six steps all together

Table 35 Standard process flow for Air-Bridge design fabrication

Mask Number Step number Process

1 1 Top Contact

22A 2 Mesa EtchDielectric

3 3 Isolation

4

5

4

5

Bottom Contact

Bond Pad

In this work the samples that have been processed so far are XMBE304

XMBE314 (GaAsAlAs base material system) and XMBE326 (InGaAsAlAs base

material system) All of them are built up on semi-insulating substrate thus the normal

process flow was followed Step 1 to step 4 follow exactly the same route as the 1st Gen-

ASPAT Large Area mask formerly discussed in Section 3422 This process is then

continued by spin coating AZ2microm on the surface Mask 5 is used to pattern the bond

pads The metallisation scheme used for bond pads is TiAu and the thickness must be at

least 1 micrometre thick This is to minimise series resistance at the pads and ensure a

robust surface is created when used for DC and RF measurements

342311 Air-Bridge Process Optimization

Since the Air Bridge design approach is focused more on developing air-bridges

which have a width of 1 microm and as the smallest device is 2 times2 microm2 the negative resist

(Az2microm) which was used in the 1st Gen design type was changed to Az1microm After spin

coating an Edge Bead Remover (EBR) is required to remove the beads at the edge of the

sample this is to ensure no gap is created between samples and mask when ldquohard-

102

contactrdquo is applied during the photolithography step The use of Az1microm and EBR are

critical to enable the fabrication of small air-bridges and emitter sizes

The exposure time during the photolithography technique also needs to be

increased to achieve a 1microm size air bridge A longer time than normal is required thus

95 seconds is used for exposure under UV light Once finished appropriate

development time must be applied to ensure the line for the air bridge is perfectly

opened for the metal to fill up To get impeccable result in this step both combination of

exposure and development have to be tuned naturally leading to trade off in both Too

long exposure and developing time will break the bridge whereas for too short exposure

time and developing the air bridge will not open

There are two types of etching method used in this fabrication wet

(H3PO4H2O2H2O) and dry (NH4H2) etching Both have their own advantages and

disadvantages Wet etching is faster than dry etching However the etching profile is

highly isotropic and causes serious undercut The smaller devices which have

dimensional sizes of 6times6microm2 and below will ldquoshrinkrdquo the effective area under the top

contact metal Even though dry etching can have highly anisotropic etching profile

which can prevent excessive undercuts it requires a lot of time for etching the

semiconductor layers The NH4 and H2 plasma that are used in this technique do not only

etch the GaAs and AlAs layers but also etch the metal contact at the same rate as the

semiconductors Thus the metal area must be covered with a photoresist or thicker metal

(~1microm) must be used to counter this issue Figure 38 shows that the emitter bond pad

which is not covered by the resist will eventually be etched away by the dry etching

Figure 38 Dry Etching for the first run in this study

As mentioned in section 333 the advantage of H3PO4H2O2H2O wet etching is

that it can be made using two solutions fast and slow The fast solution is based on a

concentration ratio of 212 while the slow solution is based on a 3150 mixture The

103

212 fast solution will produce good anisotropic etch profile but is tricky to control due

to the rather high etch rate of 1000Aring per second Both fast and slow wet etching

solutions being carried out in this experiment are replicated from successful recipes that

were developed by other co-worker from Prof Missousrsquos group

342312 Issue of Over Etch under Top Metal (Wet Etching for Air-Bridge

Design)

The first run using the solutions mentioned above started with the deposited

metal as a top contact (Top contact must be defined first to make a bridge) then Mesa

etching to remove ~7000Aring GaAs using the fast etch solution The next process to be

followed was to employ the slow etch solution to etch down the epitaxial layer to the

substrate Doing this ensured that air-bridge was open and the devices isolated

individually Unfortunately unexpected results were observed where large undercuts

still appeared for both 2times2 microm2 and 6times6microm

2 devices Figure 39 shows severe undercut

under the big device that can clearly be seen in the digital scope Most probably this

problem occurs due to excessive time used for the wet etching (10 seconds for fast

solution and 10 minutes for the slow etch solution)

Figure 39 Severe undercut of 2times2 microm2 and 6times6 microm

2 devices

The second run was employed to reduce the wet etching time in total thus dry

etching was needed As discussed above CH4 and H2 are used to etch away the GaAs

and AlAs semiconductor material while O2 is used to remove polymer residues To

reduce redundant generated polymer during dry etching the process must be separated

into several goes (runs) In this run at the mesa etch step the metal area is covered by

S1805 resist and then the plasma etches down until the heavily doped Ohmic layer is

reached For the isolation step the slow wet etching solution is used to ensure the

semiconductor under the bridge is removed The etching time is still long ie over 5

Zoomed Zoomed

104

minutes Hence this will still generate appreciable undercuts Although this run

managed to reduce the wet etching time it still did not solve the undercut Figure 310

shows undercuts under the effective mesa area still occurring in this run

Figure 310 Device with emitter area of 2times2 microm2 and 6times6 microm

2 after dry etch and wet etched

Further investigation has been made to find the root cause of this issue Scanning

Electron Microscopy (SEM) picture were taken to see deep into the completed devices to

investigate what is actually happening Figure 311 shows clearly that the semiconductor

under the bridge and effective area under the 2times2 microm2 device went missing The same

goes to for the large device where about half the size of the designed effective area under

to metal was also unintentionally removed by the solution

Figure 311 SEM Images of the GaAs sample

Since this work was also run together with an InGaAs based sample

(XMBE326) new knowledge regarding etching profile was acquired The lesson learnt

from this run is that the wet etch profile of the Gallium Arsenic (GaAs) material is

totally different from that of InGaAs Figure 312 shows the cross section of GaAs and

InGaAs materials after wet etching The evidences of these issues are also indicated by

the SEM images in Figure 313 and Figure 314 This more or less explained the reason

why the semiconductor under the top metal contact is always missing when wet

chemical etching is applied

Zoomed Zoomed

105

Figure 312 The cross-sectional view of InGaAs (left) and GaAs (right) samples used in this study

Figure 313 SEM images taken from the XMBE304 samples and a GaAs test sample

Figure 314 SEM images for InP and InGaAs taken from [56]

Furthermore previous work done by co-worker Dr Md Adzhar Md Zawawi has

optimised the process for RTD samples based on InGaAs and InAlAs heterojuction

semiconductor materials In this work Dr Zawawi found out that achieving submicron

dimensions is possible when using the soft reflow technique on InGaAs[102] However

it does not apply to GaAs when using the same technique There are few reasons to

106

believe that the work that was carried out by the previous student is not repeatable to the

XMBE304 GaAsAlAs material The main reason is due to the much thicker Ohmic

layers used in the GaAs of ASPAT structures which requires longer etch time The

sample that was used in the previous submicron work namely XMBE277 (RTD) [101]

and corresponding epitaxial layer structure is attached in appendix III The thickness

needed to be removed for the MESA step is 1421Aring compared to XMBE304 (ASPAT)

which is ~6900Aring

The fabrication results shown from the Air-Bridge design does not seem

favourable to the GaAsAlAs heterojunction sample These include unreproducible IV

characteristics and excessive undercut under the metal Therefore the GaAsAlAs

ASPAT sample cannot be processed using this type of mask (Air-Bridge approach) The

fabrication efforts were then moved to the Dielectric Bridge design described below

342313 DC measurement

Fully functional ASPAT I-V characteristics were not obtainable in this run due to

severe damages caused by the undercuts that happened underneath the top metal contact

As can be seen from Figure 315 the behaviour of the current response suggests a very

leaky diode This confirms that the air bridge approach does not work with the

GaAsAlAs ASPAT diode

Figure 315 Short circuit behaviour on one of the fabricated device in this run

-0015

-001

-0005

0

0005

001

0015

002

0025

003

0035

-3 -2 -1 0 1 2 3

Cu

rren

t (A

mp

)

Voltage (Volt)

30x30BP

107

34232 Fabrication Process of the Dielectric-Bridge Design

Since the Air-Bridge design was not successful an alternative Dielectric-bridge

process flow was then designed to solve this issue The first run in this Dielectric-bridge

process was performed to ensure that when the fabrication was completed it was able to

produce the correct and reproducible ASPAT current-voltage characteristics The

process flow in this run is to follow the initial design which identifies the steps according

to the mask number Step 1 to step 4 in the process are exactly the same as in the 1st Gen

mask design Step 5 which is the bond pad process is then continued with spin coating

the AZ2microm After exposure and development a TiAu metal scheme with a minimum

thickness of approximately 1 microm is deposited

Table 36 Standard fabrication process flow for Dielectric-Bridge design

Mask Number Step number Process

1 1 Top Contact

22A 2 Mesa Etch

3 3 Isolation

4

5

6

7

4

5

6

7

Bottom Contact

Bond Pad

Via Etch

Bridge

Step 6 is quite complicated compared to the other steps where the introduction of

S1805 resist is used as a dielectric layer to prevent short circuit between top and bottom

contacts This includes opening the smallest via (holes within the dielectric layer) of 2times2

microm2 and 6times6 microm

2 emitter sizes This step started by spin-coating the S1805 resist and

was then followed by baking it at 150˚C for 30 minutes The longer baking time is to

ensure it is hard enough to be deployed as a dielectric layer Once the S1805 had

hardened as dielectric layer the sample was then spin-coated again with S1813 (thicker

resist) on the dielectric layer to act as a mask for vias opening (to cover the dielectric

layer from via etching damages) After the S813 was developed O2 plasma etches was

108

applied to remove the S1805 that covered the top metal and bond pads The hole on the

metal was then exposed This required only two minutersquos plasma etches time

As soon as the holes were created within the dielectric layer a quick clean to

remove S1813 was performed using Acetone and IPA This has to be fast enough to

remove the S1813 without removing the S1805 dielectric layer Mask 7 then defined the

area for the metal that will fill up into the open vias This metal connected the top

contact metal of the device to the bond pad (GSG pad) The same metal scheme as

mentioned in Section 3422 was used as the bond pad

The fabrication using the dielectric bridge approach in the 2nd

Gen mask design

appeared to work successfully when initial I-V characteristics were taken and it also

showed that the ASPAT diode was fully functional However using hardened S1805 as

a dielectric layer is not a good practise for manufacturing device since there were left

over residue on the sample surface as depicted in Figure 316 This residue comes from

the non-uniformity of S1805 resit that was formed during heat treatment Therefore the

next run was to avoid using S1805 but replacing it with a standard dielectric layer based

on Silicon nitrite (Si3N4)

Figure 316 The surface of the sample after final processing

342321 DC measurements

Given that the process of qualifying mask steps for GaAsAlAs ASPAT diodes

looked promising using the dielectric bridge approaches initial I-V characteristics

measurements were carried out for XMBE304A Figure 317 shows that the I-V

characteristics are comparable to those of the other runs ie in the 1st Gen mask

Noticeably the IV characteristic (Figure 317) between each device size does not scale

109

on a single line This is attributed to the unintentional variation in the size of the emitter

during wet etching within each sample

Figure 317 IV characteristics of sample XMBE304 for emitter sizes of 1000 microm2 2500 microm2 900

microm2 400 microm2 225 microm2 100 microm2 and 36 microm2

The current densities shown in Figure 317 are calculated based on a reduced

effective area of the devices by a factor of 08 This assumption is made as actual area

(without metal) is un-measurable Theoretically devices with small areas will have

higher resistances compared to larger devices However the opposite occurred in these

devices This problem is still under investigation It could be due to the spreading

resistance producing different values according to device sizes (smaller devices have

longer D while bigger devices have shorter D)

34233 Dielectric-Bridge Process Optimization (Si3N4)

Due to poor surface roughness of the sample when using S1805 as dielectric

layer this run was employed to improve the surface quality by using Si3N4 In order to

get the actual size of the emitter the processing needed to start by defining the

semiconductor area using either top contact mask or mesa etch mask and not to deposit

the metal first Smaller effective area will be obtained if the top contact mask is used to

define the area compared to the MESA mask since there are 05 microm and 1 microm tolerances

designed in the MESA mask

-000001

0

000001

000002

000003

000004

000005

-2 -1 0 1 2

Cu

rren

t D

en

sity

(A

mp

microm

2 )

Voltage (Volt)

Current Density vs Voltage

density36

density100

density225

density400

density900

density2500

density10000

110

Table 37 New arrangement of the mask number and step in Second Run

Mask Number Step number Process

3 1 Isolation

22A 2 Mesa Etch

4 3 Bottom Contact

6

1

7

5

4

5

6

7

Via Etch

Top Contact

Bridge

Bond Pad

The fabrication was initiated by isolating each device using Mask 3 to define

them individually Then MESA mask (Mask 2) was used to cover the area from wet

etching optical measurement were used to obtain the actual size of the emitter without

metal Figure 318 shows the process results after H3P04H2O2H2O etching for a ratio of

3150

Figure 318 Optical images after MESA etching (a) 1 microm Tolerance (b) 05 microm Tolerance

Although the mask was designed with smaller size of 6times6microm2 using the slow etchant

solution the effective emitter size can still shrink to ~3times3 microm2 and ~2times2 microm

2 as can be

seen in Figure 318 The next step was to define the bottom contact using a metal scheme

that is suitable for GaAs ohmic layer ie AuGeNiAu After lift-off the samples were

then cleaned and spin-coated with AZ2microm Then the area was defined by using Mask 6

allowing the exposed area to be filled by Si3N4 which is the dielectric layer used in the

second run However a tricky issue happened here as the Si3N4 was difficult to lift-off

especially on the smallest emitter area in this mask In this run the lift-off process was

successfully done after three days Step five which uses Mask 1 was to define the top

111

contact The same metal scheme will fill up the tiny holes within the dielectric layer to

attach on the emitter semiconductor Figure 319 below shows an optical image of a

6times6microm2

device after the top contact lift-off process Once lifted-off annealing took place

to ensure Ge diffuse into the GaAs contact layer to lower the resistance

Figure 319 After lift-off processing

Step six uses Mask 7 to define the area of metal connection between active areas

to the bond pad In this step and step seven a different metal scheme from the top and

bottom contact was used to connect metal to metal Here a TiAu scheme was used The

final step (step 7) was to define the bond pad area In this context mask 5 is used The

bond pad requires thicker metal thickness to reduce the series resistance and to increase

the robustness of the metal surface when probed with needles during measurements

This second run of 2nd

Gen mask dielectric approach went well with better samplersquos

surface when using Si3N4 as the dielectric layer compared to the previous run using

S1805 Once the bond pad were defined for each device as done in previous run the

preliminary current-voltage characteristics of the Si3N4 run were obtained This is

described in the next section

342331 DC measurements

As mentioned in the previous section the Si3N4 run started with a defined emitter

area This allows actual ASPAT dimension to be measured thus the current density

versus voltage that are plotted in the following figures are based on actual measured

sizes

112

Figure 320 The current density of XMBE304 processed using Si3N4 as dielectric layer

Although the actual area of each device is obtainable from measurement the current

density of this sample does not scale well either Figure 320 show the current density is

not scalable from 03V to 1V This probably happened due to the same issue related to

the spreading resistance for each diode varying To investigate this issue new runs were

required with intention of reducing Rsprd and were expected to lead to more scalable

devices Therefore a study of reducing the spreading resistance (Rsprd ) due to

contribution of the large D-gap was carried out using the 2nd

Gen mask with the Si3N4

dielectric bridge approach The following section discusses in detail the steps used in

reducing the gap between the top and bottom contact for the GaAsAlAs ASPAT diode

34234 Dielectric-Bridge (Si3N4) Process Optimization by varying the D-Gap

The series resistance as discussed in detail in Section 26322 is generally due to

the total contribution of specific contact resistance in particular the diode size the sum

of all doped layer resistances that sandwich the main ASPAT layer and the spreading

resistance which comes from the lateral structure diode design[33] Thus in order to

acquire better performance at high frequency these contributors must be controlled One

parameter that can be controlled through this fabrication process is to reduce the

spreading resistance by controlling the separation between top and bottom contact ie

the D-gap as indicated in Figure 321

-000001

0

000001

000002

000003

000004

000005

-1 -05 0 05 1

Cu

rren

t D

ensi

ty

(Am

pm

2)

Voltage (Volt)

Current Density vs Voltage

D36

D100

D225

D400

D900

D2500

D10000

113

Figure 321 side view of lateral ASPAT structure

Without designing a new mask at this stage the same 2nd

Gen Dielectric-Bridge

mask was used but the process flows did not follow the sequence order of mask

numbers The reason for deciding not to design a new mask to reduce the D-Gap was

because the fabrication process flow had not yet confirmed it repeatability and

reproducibility Thus the available photo mask at that moment was fully utilised

Furthermore the feature of the 2nd

Gen mask that included the mesa tolerance can be

exploited in this study Hence the initial idea was to reduce the length of D by using the

tolerance that was designed in with Mask 2 (MESA etch mask)

Table 38 New arrangement for the Third run using Dielectric-Bridge mask

Mask Number Step number Process

1 1 Mesa Etch

3 2 Isolation

3

2

3

4

Bottom Contact

Mesa Cover

6

1

7

5

5

6

7

8

Via Etch

Top Contact

Bridge

Bond Pad

The process therefore was initiated by cleaning the sample and was then

followed by spin-coating it with S1805 to cover the emitter contact and define the actual

D-gap

114

size of the diode This first step used Mask 1 with wet chemical etching using the slow

solution (H3P04H2O2H2O etch with a ratio of 3150) Once the resist was striped

optical measurements were performed to obtain the effective area of the emitter Figure

322 below shows the smallest emitter area obtained after etching

Figure 322 The measured size of the emitter area and the length D (blue color marked)

Step two was to isolate the devices individually by using Mask 3 This was

performed using the fast etch solution (H3P04H2O2H2O etch with ratio of 212) This

took about 10 seconds to remove about 10000Aring of material

Step three which involved two masks was the most complicated in this process

Firstly Mask 3 was used to define the bottom contact by covering the sample with

AZ2microm and then the sample was hard- baked at 190 ˚C for 4- 5 minutes to ensure it had

totally dried before applying Polydimethylglutarimide (PMGI) The PMGI type used in

this process is Lift-off Resist (LOR) SF11 After spin-coating the SF11 at 7000RPM for

45 seconds it was post-baked at 190 ˚C for 4 minutes This was done to ensure that the

LOR SF11 was hard enough for the S1805 to stick on it This was followed by spin-

coating S1805 at 4000RPM for 30 seconds and exposing it under i-line UV for 20

seconds and developed using Micro Dev mix with DI water (11 ratio) for one minute

When the correct shape of S1805 was formed the sample was exposed under flood UV

for 15 minutes The SF11 was developed with 101A developer for 1 minute Figure 323

below summarizes the whole process in step three The step three processes concluded

by depositing the bottom contact with alloyed metal scheme and once lift-off had taken

place the next step continues as usual

115

Figure 323 Summary of LOR technique steps

Step four as well as the subsequent steps were completed in this run by copying

from the previous run (ie second run) which was to deposit the Si3N4 as the dielectric

layer and so on All the devices that underwent fabrication using this technique were

measured under optical microscope MC60 assisted by the Walsall software tool in the

lab From the measured value obtained it was expected that reducing the length of D in

this technique would improve the spreading resistance by up to 70 from the original

2nd

Gen mask design Table 39 summarizes the calculation of original spreading

resistance and improvement using this technique The calculation method and equation

are taken from Section 26322

Table 39 The outcome of the spreading resistance before and after using LOR technique

Device Size(microm2) D-original(microm) RSprd(Ω) D-new(microm) RSprd(Ω)

100x100 4 056 163 023

50x50 14 327 194 054

30x30 19 602 163 097

20x20 21 832 212 135

15x15 22 1007 202 18

10x10 24 1293 205 252

6x6 25 1643 230 418

116

Technically from the experiment in this run it was observed that the development time

also controlled the final length of D longer development time produced shorter D

lengths while shorter time yielded larger D gap lengths Therefore in this run the

development time was kept constant because of the desired D length from the mesa

tolerance in general is only ~2microm

342341 DC measurement

As usual when wafer processing was completed early DC measurement took

place to check the diodes performance Each diode size was measured and compiled

into one graph as shown in Figure 324 It is clear that the performance was better than

that in the previous run in term of scalability and current conductivity

Figure 324 IV characteristic in 3rd run (reduced d-Gap) for each diode size on the mask

To compare the current conductivity from this run and the previous run current

density at 05V was measured for both 2nd

and 3rd

runs Assuming the contact resistances

(TLM) for both runs were constant for 50times50microm2 device size the improvement of

current in the 3rd

run is about 92 This showed that such approach to reduce the D-gap

-500E-06

000E+00

500E-06

100E-05

150E-05

200E-05

-2 -15 -1 -05 0 05 1 15 2Cu

rre

nt

De

nsi

ty (

Amicro

m2)

Voltage (Volt)

Den_100microm2

Den_225microm2

Den_400microm2

Den_900microm2

Den_2500microm2

Den_10000microm2

Den2_36microm2

117

was successful in this run A new mask design was ready to take the challenge for

processing toward millimetre and sub-millimetre wave application

343 Fabrication process of GaAsAlAs ASPAT diode toward High frequency

Applications

So far the information that can be gathered from previous processing is that the

optimum process flow is achieved through dielectric approach design The effort in

reducing the series resistance by lowering its biggest contribution was attained through

lowering the D-Gap in the structure Once everything had been optimized ie the

process flow series resistance junction capacitance etc it was time to develop a new

mask design which only focused on the development of small ASPAT devices for use in

the millimetre and sub-millimetre wave regions

3431 Third Generation (3rd

Gen) mask design

The 3rd

Gen mask design was developed by taking into the account every aspect

of parameters that can contribute to the device by means of robust devices that are able

to function properly at ultra-high frequency The device cut-off frequency is given by

119891119888119906119905minus119900119891119891 =

1

2120587119877119904119862119895

(31)

Here RS is the series resistance and Cj is the junction capacitance To obtain high cut-off

frequencies Rs and Cj must be kept as low as possible From the fabrication point of

view two parameters that can be directly and easily controlled are the device area (A)

and the D-gap (which contributes to Rs)

Therefore calculations were made to find the best option Table 310 shows the

calculated value of capacitance (eq230 in page 44) and cut-off frequencies (eq31

above) for the ASPAT diodes studied These two equations extract the cut-off frequency

of the ASPAT assuming no external effects and for fully depleted devices The

XMBE304 ASPAT sample is expected to be suitable for millimetre wave applications

if small devices are successfully made In real devices sub-millimetrewave operation

can be hard to achieve due to increased series resistance and other process related

parasitics

118

Table 310 DC and RF characteristics for XMBE304

Device Size (microm2) Fully depleted

Capacitance(fF)

Calculated Series

Resistance (Ω)

Fully depleted Cut-Off

Frequency(GHz)

10000 5490 04 72

900 490 15 216

36 198 7 1148

16 879 11 1646

4 22 29 2500

Therefore in the 3rd

Gen mask design the smallest device that can possibly be obtained

in the GaAs based material fabricated using i-line lithography which is available at the

University of Manchester is 2times2microm2 and the gap between top and bottom is 15microm at

least The connecting bridge technique applied only utilised dielectric bridge method for

GSG features Figure 325 shows the layout of actual 3rd

Gen mask used in this study

There are 344 die chips on this type of mask design It also includes the Ground-signal

ground pad with 50um pitch for each chip and six de-embedding test structures as well

as eight TLM structures

Figure 325 3rd

Gen Mask layout showing actual devices de-embedding TLM and alignment mark

structures used in this study

119

This mask is designed generally from optimizations from 2nd

Gen Mask which

deploys a dielectric bridge for connection between the devices to the bond pads

Consequently the processing steps are not being much different but mostly follow what

is shown in Table 311 below The difference only applies to the much smaller mesa

size Other features included in this mask are de-embedding structures for RF

measurements via-hole test structure TLM structures and parallel plate capacitance test

structures

Table 311 3rd

Gen Mask process step

Mask Number Step number Process

1 1 Mesa Etch

2 2 Isolation

3

4

3

4

Bottom Contact

Mesa Cover

5

6

7

5

6

7

Via Etch

Top Contact

Bond Pad

The high frequency fabrication process flow is summarized in the following section

which shows illustrations in three dimensional and cross sectional (Figure 326) views

for easy understanding

34311 Step by Step Processing

0 Wafer preparation

Cleaning using NMP and DI

water or

Acetone and IPA

120

1 Mesa Etch

2 Isolation

3 Bottom Contact

4 Mesa Cover (Dielectric Deposition)

5 Via Etch

Etching using etchant solution

H3P04H2O2H2O etch with ratio

of 3110

Etching using etchant solution

H3P04H2O2H2O etch with ratio

of 212

Metal deposition with metal

scheme AuGeNiAu for

~500nm thick

Dielectric deposition 500nm

thick Si3N4

Via etch to open holes for metal

contact using reactive ion

etching using CF4

121

6 Top Contact

7 Bridge and Bond Pad

Figure 326 Step by step device fabrication (in 3D and cross-sectional view) for semi-insulating

substrate device type used in this study

Figure 327 Example finished process device with bond pad using 3rd

Gen mask

3432 TLM measurements

Transmission Line model (TLM) measurements for this run were carried out

after annealing (ie at 420˚C for 2 minutes) for process control monitoring by extracting

contact resistance (Rc) values These values are a measure of the quality of ohmic metal

contacts for a given process As discussed in Chapter 2 the TLM technique used four-

Metal deposition with metal

scheme AuGeNiAu for ~500nm

thickness and thermal annealing

420˚C for 2 minutes

Metal deposition with metal

scheme TiAu (1microm thick)

122

point measurement on TLM test structures located around the 15times15mm2 tiles as shown

in Figure 222 (in page 55) Normally five TLM structures are measured across the tile

for both top and bottom contacts Figures 328 and 329 display graphical TLM results

for top and bottom contact respectively As can be seen both graphs exhibit excellent

uniformity

Figure 328 XMBE304 TLM measurement for the top contact after annealing

Figure 329 XMBE304 TLM measurement for the bottom contact after annealing

Based on the graphs above for the top metal contact the average contact

resistance (RC) value using the metal scheme of AuGeNiAu is found to be ~005Ωmicrom

and the sheet resistance (RSH) is 22Ω However for the bottom contact the average

value for contact resistance is 012Ωmicrom and sheet resistance obtained is 26Ω Both

values are in a very good agreement with the known doping of both ohmic contact

0

2

4

6

8

10

0 10 20 30 40 50

Res

ista

nce

(O

hm

s)

Spacing (microm)

After 420C for 2mins top contact

y = 02592x + 02375

0

2

4

6

8

10

12

14

0 10 20 30 40 50

Res

ista

nce

(O

hm

s)

Spacing (microm)

After 420C for 2mins bottom contact

y = 02157x + 01053

123

layers Therefore the specific contact resistance (Eq 250) that contributes to the total

series resistance can be calculated and the value obtained is 15Ωmicrom2 and 54Ωmicrom

2 for

top and bottom TLM structures respectively

3433 DC characteristic measurements

Again once the wafer processing is completed room temperature DC

characteristics are taken using an HP 414B or HP500B parameter analyser and its actual

setup is as described in Chapter 2 This initial measurement was to ensure the

functionality of the diodes For this run using the 3rd

Gen mask the I-V measurements

for mesa active area of 4times4microm2 6times6 microm

2 and 10times10 microm

2 were taken and are depicted in

Figure 330 Figure 331 and Figure 332 respectively Nine diodes were measured for

each three device sizes to check their uniformity Thus the average current and standard

deviation are taken at two voltage steps of 1V and 15V Table 312 summarizes the

standard deviation for each measured data obtained from this final run

Figure 330 XMBE304 GaAsAlAs ASPAT measured IV characteristics at room temperature for

4times4microm2 mesa size

-00005

0

00005

0001

00015

0002

00025

0003

00035

-3 -2 -1 0 1 2 3

Cu

rren

t (A

)

Voltage (V)

C01_4um_1P_SI

C22_4um_1P_SI

C40_4um_1P_SI

W01_4um_1P_SI

W22_4um_1P_SI

W40_4um_1P_SI

AI01_4um_1P_SI

AI22_4um_1P_SI

AI40_4um_1P_SI

124

Figure 331 XMBE304 GaAsAlAs ASPAT measured IV characteristics at room temperature for

6times6microm2 mesa size

Figure 332 XMBE304 GaAsAlAs ASPAT measured IV characteristics at room temperature

for 10times10microm2 mesa size

Table 312 Standard deviation at two different voltages

Device size (um2) Standard Deviation 1V () Standard Deviation 15V ()

4times4 14 14

6times6 65 72

10times10 44 39

Noticeably the standard deviation of the device increases for the smaller size

devices This trend happened probably because the active mesa area is not uniform

causing different series resistances Since the smallest mesa active area which achieved

-0001

0

0001

0002

0003

0004

0005

0006

0007

0008

-3 -2 -1 0 1 2 3

Cu

rren

t (A

)

Voltage (V)

C02_6um_1P_SI

C23_6um_1P_SI

C41_6um_1P_SI

W02_6um_1P_SI

W23_6um_1P_SI

W41_6um_1P_SI

AI02_6um_1P_SI

AI23_6um_1P_SI

AI41_6um_1P_SI

-0005

0

0005

001

0015

002

0025

-3 -2 -1 0 1 2 3

Cu

rren

t (A

)

Voltage (V)

C03_10um_1P_SI

C24_10um_1P_SI

C42_10um_1P_SI

W03_10um_1P_SI

W24_10um_1P_SI

W42_10um_1P_SI

AI03_10um_1P_SI

AI24_10um_1P_SI

AI42_10um_1P_SI

125

a good I-V characteristic is 4times4microm2

in this run these devices were used for the next step

of characterisation which is the S-parameter measurement to extract their behaviour at

different frequencies

35 Conclusions

In this chapter basic fabrication techniques of GaAsAlAs ASPAT on both doped

and semi insulating substrates using standard I-line lithography as well as step by step

descriptions to achieve reproducibility in the process fabrication flow has been

demonstrated with relevant initial measured results Two major outcomes have been

demonstrated firstly related to repeatability reproducibility and manufacturability

mostly done on large device areas (15times15microm2 to 100times100 microm

2) Secondly a successful

process flow for small emitter size devices (2times2 microm2 to 10times10 microm

2) has been

developed

Subsequently two types of designs during process optimization were developed

namely Air-Bridge and Dielectric-Bridge approaches The latter approach seems

favourable for GaAsAlAs materials but was only successful when reducing the series

resistance of the device This problem was addressed by optimising the D-gap between

top and bottom contact which resulted in good scalability of each ASPAT dimensions

improving current conductivity by 92 and achieving a reproducible process On the

other hand in the former approach issues were encountered with over etching underneath

the diode effective area and thus it was hard to achieve reproducible devices

Repeatability reproducibility and manufacturability fabrication processes based on

dielectric bridge method were successfully developed This new process provides a

highly efficient and economical solution for the fabrication of GaAsAlAs ASPAT

diode Emitter sizes down to 4times4 microm2 dimensions are routinely and reproducibly

achieved in this process Series resistance which is an important parameter in

determining high frequency application are greatly reduced by changing the gap between

top and bottom contact

With all the new improvements implemented the possibility of the proposed Dielectric

Bridge method fabrication process was successfully applied for the fabrication of

ASPAT diodes

126

4 PHYSICAL MODELLING OF THE GAASALAS ASPAT

DIODE USING SILVACO

41 Introduction

Fabricating any device or circuit requires a lot of time resources cost etc

especially elements related to the production of semiconductor devices Furthermore in

realising such a device a clean room is required Hence fabrication and processing in

wafer fab need full attention to get it done with a higher rate of success reproducibility

and manufacturability One solution that can be highlighted which will be able to reduce

all the resources that are mentioned above is by using computer simulation approaches

or to be precise physical modelling For epitaxial layer based devices physical

modelling is a good choice as it would give a better understanding and insight into each

layer and how electrical characteristics are derived A software that is most suitable for

epitaxial layer physically modelled is SILVACO This software is a very comprehensive

tool to simulate epitaxial based devices and to predict their behaviour Such software

covers many aspects starting from the first principles of physic epitaxial layer

definition as well as device layout thus making it the most powerful virtual wafer fab

tool in the market

In this chapter the SILVACO packages are discussed The discussions include the

method of defining a new material defining models and constructing the AC or DC

supply to obtain the output characteristics of the virtual device Apart from these the

focus will also be on the ASPAT diode modelling simulation and analyses of the results

which will include the ASPAT structure suitable models and DC current-voltage

characteristics The dependencies of individual structure on the I-V curves will also be

highlighted Finally the discussion of results and analyses involving a range of

operational temperatures dependency as well as from comparison made to conventional

SBD used in this study will be examined

127

42 SILVACO modelling Tools

SILVACO is a modelling software introduced in 1984 by Dr Ivan Pesic It is

purposely created for electronicsrsquo devices physical modelling and characterization This

software company has become the major supplier for most of the Electronic Design

Automation (EDA) for circuit simulation amp design of analogue Mixed-Signal and RF

circuit market This Technology Computer Aided Design (TCAD) software can predict

the simulated device performances starting from first principles It has a package which

can provide Virtual Wafer Fabrication (VWF) simulation to the device designer and

which has the capability to perform two or three dimensional physical device modelling

by using the ATLAS simulator [103] SILVACO allows all parameters such as

electrical thermal and optical characteristics of a device to be simulated under desired

bias conditions It offers cost effectiveness as well as quick prediction of results for

many semiconductor devices compared to real experiments Some hands-on experiment

may not always perform hence SILVACO can be used as an alternative

The core of SILVACO is Atlas itself which provides a platform to perform DC AC

and transient analysis for such dimensional device structure regardless of the

heterojunction material type ie binary ternary quaternary etc As the brain of

SILVACO Atlas which receives input command files containing instruction text for

execution from a runtime environment known as Decbuild will process the instruction

text and display progression error and warning via Runtime Output All the calculations

of the resultsoutcomes of the simulation are plotted via a tool called TonyPlot which is a

tool to visualise the output Figure 41 below shows how precisely the physical

modelling takes place the process of building the structure how its parameters and

variables are defined how an appropriate model statement is selected how performance

is analysed and lastly how the outcomes are displayed

128

Figure 41 SILVACO Atlas simulation process flow

As can be seen in Figure 41 the structure specification statement is used to define

any desired structure by setting the command in Deckbuid of the following parameters

a) Mesh where the structure can be defined either in 2D or 3D Cartesian grids The

unit of the coordinates used is in microns and the spacing parameters which

define the netting size can be used to improve the accuracy of the analysis at any

given position The density of netting size in this statement determines the

processing time

b) Region where the multi-layers in a structure are defined and this statement has

to outline each layer that represents a separated region independently The mesh

must be assigned to a region and the sequence of the region is arranged from low

to high

-Mesh -Region -Electrode

-Contact -Material

Structure Specification

-Model -Interface

Model Specification

Method

Numerical Method

-Log -Solve -Load -Save

Solution Specification

-Tony Plot

Display Results

129

c) Electrode are used to define the location of bias point for a designed structure

when performing the electrical analysis In this work or for the case of a diode

two electrodes are allocated as an anode and a cathode In a vertical device the

latter electrode is placed at the bottom of the device while the anode is at the top

d) Doping this statement refers to doping concentration injected into the desired

region and it normally depends on the material types

e) Materials since SILVACO was developed specifically for Silicon-based devices

default parameters are set up for Silicon properties However the use of

materials statement allows SILVACO to run for different material ie GaAs

InGaAs etc In order to make it work this parameter is defined first and then

followed by the material name and its properties such as bandgap permittivity

conductionamp valence band discontinuities mobility etc

The most crucial part is to determine whether the simulated structure of a particular

device is correct or incorrect This is done by properly choosing the specific model

statement The Specific model statement is employed to express the physic equations

that are used during the device analysis The models statement depends solely on the

structure definition Examples are device structure with double barrier use Non-

equilibrium Green function (NEGF) model and single barrier uses Semiconductor-

insulator-Semiconductor (SIS) model for effective and accurate analysis process For the

case of a single barrier in SILVACO there are many model statements that can be used

for such analysis Therefore it is recommended to check model by model in order to

ensure that all the needed parameters are defined in the material statements and results

produced are valid

SILVACO is able to calculate such models by using different numerical methods

which means semiconductor device problematic is computed to make successive

solutions by random discretisation There are three different numerical methods that are

regularly used by SILVACO to perform its calculations which are Newton Gummel and

Block Basically this is solved by using a non-linear iteration procedure which begins

from an initial guess and which then uses an iterative process to find the predicted

solution The detail of these can be found in reference [103]

130

In order to turn on the problem solution the solution specification is defined This

include log solve save and load statement All these work together to provide data for

analysis by other functions The log statement is a file type that saves in memory and can

be loaded by Atlas Any solved device will be stored in the log file Therefore it is

necessary to define the LOG before SOLVE statement and close it after the calculations

While the save statement is used to store all data point to a node in the output file the

load statement is utilised to recall all the saved data to be read by Atlas

Finally all these files can be displayed or plotted on TonyPlot it is recommended to

make SET files at plotting point for a better visualisation The plotted or displayed files

in TonyPlot can be manipulated for scaling graphs overlaying different curves and most

importantly to export the data to other files formats

43 SILVACO Implementation GaAs AlAs ASPAT Modelling

The structures play an important role in determining the terminal output

characteristics Therefore to start simulating the ASPAT device the first thing that must

be specified is its structure As mentioned earlier in Section 3412 the ASPAT diode is

a top down multilayer structure This structure which is adopted from that of Section

3412 will be used as a basis to perform the ASPAT simulation Figure 42 shows the

structure of the ASPAT in this real simulation which is exactly the same as been

discussed in Section 26322 The ASPAT diode consists of two heavily doped (up to

5x1018119888119898minus3) GaAs contact layers on top and bottom slices adjacent to lightly doped (up

to 3x1017119888119898minus3) GaAs intermediate layers In between these layers is a sandwiched

structure consisting of two different lengths of undoped GaAs spacer layers and a thin

layer of AlAs that act as a tunnel barrier In this work the simulation result will be

compared with the fabricated measurement result depending on the size of the diode

The actual device that will be used to validate the simulation is the main device used in

this study (XMBE304) which is based on a lattice matched GaAsAlAs grown on a

semi-insulating substrate Therefore the design structures that are proposed in the

fabrication are compatible with the fabricated devices and are based on lateral structures

as can be seen in Figure 42

131

Figure 42 The GaAsAlAs ASPAT structures produced from simulation based on the diode

multilayer heterostructures on the right

In the consequent simulations a key observation regarding the AlAsGaAs

heterojunction is that there are two types of tunnelling processes direct tunnelling and

indirect tunnelling Figure 43 shows the Energy-Momentum (Ek) diagram depicting the

three valleys for the AlAs conduction band namely L Γ and X points In normal

circumstances the transition of electrons happens to the X-point which is the lowest

energy in the conduction band In the case of very thin barrier the tunnelling process

occurs at the Γ point in both AlAs and GaAs materials which is the direct tunnelling

process[18] It has been reported that in the case of an ASPAT diode tunnelling which

occurs at the Γ point will be the dominant component in the tunnelling current

Therefore the actual band gap will be different from the one at the X-point which is

which 216eV[15] By contrast the energy band gap at the Γ-point is around

283eV[104] Thus this simulation uses this band gap value

Figure 43 The Energy-momentum diagram for AlAs heterojunction semiconductor

216eV

Γ

X

E

K

L

AlAs

CB

VB

283e

132

The simulation code as attached in Appendix IV and the output of the simulation with all

the input mentioned above are shown in Figure 44 (a) and Figure 44 (b) Figure 44 (a)

is the band diagram at equilibrium and Figure 44 (b) is the band diagram when a bias is

applied

Figure 44 (a) The conduction band and the valence band of the ASPAT diode structure (b) the

energy band diagram of the ASPAT diode structure when under three different biases

44 Simulation Result and Analysis

Basically the SILVACOrsquos Atlas simulation package is used to calculate the I-V

characteristics from multilayers structures In the case of GaAsAlAs heterojunctions all

details of the structure as shown in Figure 44 above are calculated based on solving the

Schrodinger time-independent equation in each layer taking into account the variation of

effective mass and conduction band offset between GaAs and AlAs

For thermionic emission and tunnelling mechanism across an abrupt heterojunction

interface the general method used in SILVACO is taken from K Y Yang work [105]

The tunnelling current of the ASPAT diode uses the equation below[74]

119869 = sum2119898lowast119864119894(1 minus 119877)119896119879

1205872ħ3

119873

119894=1ln [

1 + exp (119864119865 minus 119864119894

119896119879)

1 + exp (119864119865 minus 119881 minus 119864119894

119896119879)]

(41)

Where m is the electron effective mass E denotes the energy of the electron R is the

total resistance k represent Boltzmann constant ħ is the reduced Planck constant EF is

the fermi level V is applied voltage and Ei is the electron energy perpendicular to the

-2

-15

-1

-05

0

05

1

15

1

82

16

3

24

4

32

5

40

6

48

7

56

8

64

9

73

0

81

1

89

2

97

3

Ene

rgy(

eV

)

Thickness (microm)

VB

CB

133

barrier The important parameters that enable SILVACO Atlas to perform correct

calculations and analysis are the choice of appropriate models The suitable model that is

available for evaluating the GaAsAlAs ASPAT is based on the Semiconductor-

insulator-semiconductor (SIS) model

Thus in this run Non-local Quantum Barrier Tunnelling Model (SISEL and

SISHO) are utilized specifically semiconductor-insulator-semiconductor mode This

model enables the tunnelling current between two semiconducting regions separated by a

quantum barrier to be calculated [103] It is assumed that the charge tunnels across the

whole barrier with the source or sinks at the interface with the semiconductor regions

Under the Non-Local Quantum Barrier tunnelling model another model that can be used

is semiconductor-semiconductor-semiconductor (SS) tunnelling model if the materials

are specified By correctly inserting all parameters with the right model an excellent DC

IV characteristic match between simulation and measurement can be produced as shown

in Figure 45

441 DC Current-Voltage Characteristic

In this simulation the current at each bias step and each mesh point can be set up by the

user however the detailed calculation such as formula usage methodology and

approach that is adopted by SILVACO Atlas is unknown As mentioned above in order

to produce the energy band diagram of the ASPAT the DC characteristic of the structure

can be solved by using the Schroumldinger and Poisson equation self-consistently

To ensure that these simulations are valid one fabrication was performed on an

ASPAT diode sample XMBE304 For this sample the structure parameters are the

same as has been set in this simulation The result of the measurement and simulation

are then compared Figure 45 shows that the simulation result is in excellent agreement

with the measured data for this sample

134

Figure 45 The current-voltage characteristic of an ASPAT diode (100times100microm2) and (4times4microm

2)

using SILVACO Atlas simulator for structure device XMBE304 showing excellent agreement

between simulated and experimental data

These fitted results were performed on both a large 100times100microm2 device and the

smallest obtainable from fabrication (4times4microm2) which was to be used for the repeatability

amp reproducibility studies as well as for high-frequency applications study respectively at

room temperature In order to get a good fit a few parameters had been modified in the

SILVACO software via the Deck-built tool for example energy band effective masses

and bandgap discontinuity of GaAs spacer and AlAs barrier (The mentioned parameters

values are summarized in Table 41) The percentages of bandgap discontinuity in

SILVACO using the ALIGN parameter is given by[103]

119860119897119894119892119899 =

Δ119864119862

Δ119864119862 + Δ119864119881

(42)

Where ΔEc and ΔEv are the conduction band discontinuity and valence band

discontinuity respectively The m0 in the tables denotes the electron rest mass Once all

agreement between measurement and simulation has been met the simulation is then

carried out with structure analysis at room temperature and different temperatures

simulation

Table 41 The parameter values used in this simulation

Material Bandgap(eV) ΔEg(eV) Effective mass(kg)

GaAs 1424 03 0067m0

AlAs 2835 071 0126m0

-00002

0

00002

00004

00006

00008

0001

00012

-2 -1 0 1 2

Cu

rre

nt

I (A

mp

)

Voltage V (Volt)

Current vs Voltage

Measurment

Simulation

135

45 Structure Analysis of ASPAT Diode

Once the device structure was modelled and having successfully produced a

precise band diagram as well as validated the simulation results with experimental I-V

characteristics the next step is to further analyse the relationship between basic device

structure and its I-V characteristics This approach is used to predict what would happen

to the DC output if some of the parameters were varied especially with regards to the

AlAs barrier thickness In the subsequence simulations the thickness of each main

ASPAT (unequal spacers and barrier) layer will be studied independently as a variable in

order to determine how each parameter affects the I-V characteristic in both magnitude

and curvature The analysis will also include manufacturing tolerance where the

structure parameter which will result in a 10 difference in their I-V characteristics is

examined[59 75] This will provide an overview of how precisely to manufacture each

layer of the device The following simulation is based on the XMBE304 structure with

emitter size of 4x4microm2

451 Dependencies of current on AlAs Barrier thickness

Since the AlAs barrier is what limits the transportation of electron flow and

hence the current (which depends exponentially on the tunnelling barrier thickness)

therefore the first analysis to run on the simulation is the variation on barrier thickness

In the simulation the barrier thickness is measured in term of the monolayer Generally

one monolayer can be calculated by dividing the lattice constant of the material by two

In the case of AlAs one monolayer is calculated as follows

119900119899119890 119898119900119899119900119897119886119910119890119903(1119872119871) =

119860119897119860119904 119897119886119905119905119894119888119890 119898119886119905119888ℎ119890119889 (5666Å)

2

(43)

= 283Å

The nominal value of the AlAs barrier thickness for sample XMBE304 is 283nm

ie ten monolayers In the first simulation test the current change due to the barrier

thickness variation from 9ML to 11ML was examined first followed by the amount of

barrier thickness change that would produce a 10 change in current The simulation is

setup by fixing all other parameters and varying the AlAs barrier thickness as mentioned

136

above with a step of 02ML In order to determine what fraction of a ML would yield a

10 difference in the current the barrier thickness is slightly changed to fit the curve for

both 5 above and 5 below the original curve

Figure 46 IV characteristics of the dependencies of current on AlAs barrier

The current-voltage characteristics of the ASPAT diode do change dramatically with

barrier thickness in forward bias but not that much in reverse bias (Figure 46) The

current decreases as the barrier thickness increases For a 1ML change in layer thickness

(from 9ML to 11ML) the current changes by over ~300 at 05V

Figure 47 Example of analysis at -1 and 1V to the current

-00002

0

00002

00004

00006

00008

0001

00012

00014

-15 -1 -05 0 05 1 15

Cu

rre

nt

(A)

Voltage(V)

9ML

92ML

94ML

96ML

98ML

10ML

102ML

104ML

106ML

108ML

11ML

973E-06

0

0000005

000001

0000015

000002

0000025

000003

0000035

85 95 105 115

Cu

rren

t a

t 1V

B

ias

(Am

p)

Barrier Thickness (ML)

Current change with tunnel barrier thickness

Forward Current

5

-5

137

From the simulation result shown in Figure 46 a 9945ML barrier thickness will

give a 5 higher current and a 10056ML barrier will give a 5 lower current (Figure

47) Therefore in total 01 ML difference yields 10 current difference These indicate

that in order to control the current within 10 barrier thickness difference the growth

precision in the barrier must be precise to better than 01ML Extensive studies have

also shown that the I-V characteristic of a GaAsAlAsGaAs diode is very sensitive to

the thickness of AlAs barrier This work has been reported elsewhere[17]

452 Dependence of current on Spacer I length l1

For the longer spacer length (l1) five different values are chosen from 01microm to

03microm The lengths are changed in the order of 005microm Therefore the arrangement of

length is as follows l1=01microm l1=015microm l1=02microm l1=0 25microm and l1=03microm

respectively The results are plotted from -15V to 15V anode voltage in Figure 48 The

I-V characteristic of the ASPAT diode does not change much with the length in the

forward bias region but in the reverse bias region the current decreases as the layer

thickness increases Here the l1 layer acts as a voltage arm and a small size device

cannot sustain big changes in spacer length Changing the length at the forward region

will also change the energy states on the anode side as well changing the states

distribution on the cathode side in reverse bias

Figure 48 I-V characteristic of the dependencies current to Spacer I layer

-00002

0

00002

00004

00006

00008

0001

-2 -1 0 1 2

Cu

rren

t (A

)

Voltage (V)

L=01

L=015

L=02

L=025

L=03

138

By fixing the current at -1V and 1V the current and the layer thickness relationship is

illustrated in Figure 49

Figure 49 Current changes with layer thickness l1

It is noticeable that there is a dramatic change in reverse current from 01 microm to

05 microm layer thicknesses However the forward current only falls slightly from 01microm to

015microm and is stable afterwards Hence for a small size device a large change in spacer

layer at the cathode will allow more current to pass

453 Dependence of current on Spacer II length l2

Finally is the variation in the spacer II Similar to spacer I above five values are

chosen for the shorter undoped GaAs layer length l2 the thickness is varied from

00025microm to 00075 microm (steps are l2=00025microm l2=000375microm l2=0005microm

l2=000625microm and l2=00075microm respectively The results are plotted from -15V to

15V anode voltage as shown in Figure 410 In this case a slight change in I-V

characteristic in the forward bias can be seen clearly which means the I-V

characteristic depends on the length of the shorter undoped layer Therefore the l2 layer

also acts as another voltage arm due to the asymmetrical length The effect is quite

similar to the spacer l1 but this times the forward current only slight changes The reason

for the small change in current is that the length change is small and it linearly affect the

states distribution

-000002

-0000018

-0000016

-0000014

-0000012

-000001

-0000008

-0000006

-0000004

-0000002

0

0

000005

00001

000015

00002

000025

00003

000035

01 015 02 025 03

Rev

erse

Cu

rren

t (A

)

Forw

ard

Cu

rren

t (A

)

Spacer Lenght (microm)

Forward Current (1V)Reverse Current(-1V)

139

Figure 410 IV characteristic of the dependencies current to Spacer 1 layer

Fixing the current at -1V and 1V the current versus layer thickness relationship is

illustrated in Figure 411

Figure 411Current change with layer thickness l2

The I-V curve depends on the length of the shorter undoped spacer layer quite linearly

The forward current changes in increase to the layer compared to the backward current

The layer thickness l2 should be small as long as it prevents carrier diffusion Therefore

all these three layers must be kept within limit to ensure that the high performance of the

ASPAT diode can be fully utilised

-00002

0

00002

00004

00006

00008

0001

-2 -1 0 1 2

Cu

rren

t (A

)

Voltage (V)

L=75n

L=25n

L=625n

L=5n

L=375n

-2E-07

-18E-07

-16E-07

-14E-07

-12E-07

-1E-07

-8E-08

-6E-08

-4E-08

-2E-08

0

0

00002

00004

00006

00008

0001

00012

0002 0004 0006 0008

Rev

erse

Cu

rren

t (A

)

Forw

ard

Cu

rren

t (A

)

Spacer Lenght (microm)

Forward Current (1V)

Reverse Current(-1V)

140

46 Temperature Dependent Simulation

This section will discuss in detail how changes in operating temperatures influence

the IV characteristics of the ASPAT diode The same codes as in the previous

simulation with fitted results are used for this temperature dependence study but a few

parameters were changed for different temperatures

Theoretically the material parameters that are influenced by the change of

temperature are band gap electron effective mass the density of state (NC NV) light

hole mass heavy hole mass permittivity and electron amp hole mobilities However not

all mentioned parameters will have a large impact on the IV characteristic in the

SILVACO Atlas simulation The most significant factors that give appreciable impact on

the DC output current-voltage were the energy bandgap and the effective mass The

GaAs bandgap as a function of temperature is given by the equation below [106]

119864119892 = 1198641198920 minus

120572 1198791198712

120573 + 119879119871

(44)

Here Eg is the bandgap Eg0 denotes the bandgap at 0K TL is the Temperature α=

Constant (Varshni Parameter) AlAs6e-4 GaAs5405e-4 β= Constant (Varshni

Parameter) AlAs408 and GaAs204 The calculated parameters that are used in this

simulation are shown in Table 42

Table 42 The calculated values of bandgap at different temperatures

Temperature (K) GaAs Eg(eV) AlAs Eg(eV)

77 1506 2903

100 1500 2899

125 149 2893

150 1486 2887

175 1478 2879

200 1470 2871

225 1461 2863

250 1452 2854

275 1443 2844

300 1424 2835

325 1419 2824

350 1414 2814

375 1404 2803

398 1394 2793

141

The effective mass of the materials used in this simulation can be expressed by

119898119899 = 1198980119899 + 11989810 (

119879119871

300119870)

119898119901 = 1198980119901 + 1198981119901 (119879119871

300119870) + 1198982119901 (

119879119871

300119870)2

(45)

Where mn is the effective electron mass mp represents the effective hole mass m1n and

m1p are constant number for the basic GaAs material m0n

for 119898119883119898119871 119886119903119890 119892119894119907119890119899 119887119910 119905ℎ119890 119890119902119906119886119905119894119900119899 (1198981198991199052 lowast 119898119899119897)

13 while m0p is based on the

expression 1198980119901 = (11989811990111989732

+ 119898119901ℎ32

)23 The calculated parameters are shown in Table

43

Table 43 The calculated effective masses for each temperature used in this simulation

Temperatures (K) GaAs Effective Mass (kg) AlAs Effective Mass (kg)

77 00660m0 03790 m0

100 00658 m0 03788 m0

125 00655 m0 03785 m0

150 00652 m0 03782 m0

175 00649 m0 03779 m0

200 00646 m0 03776 m0

225 00643 m0 03773 m0

250 0064 m0 0377 m0

275 00637 m0 03767 m0

300 00634 m0 03764 m0

325 00631 m0 03761 m0

350 00628 m0 03758 m0

375 00625 m0 03755 m0

398 00622 m0 03752 m0

142

Figure 412 Measurement and simulation comparison result as a function of temperature range

from 100K to 398K

Figure 412 above shows excellent agreement between simulation and

measurement results at various temperatures The IV characteristics correspond to a

device of size 100times100microm2 as presented in Chapter 3

47 Temperature Dependence characteristics of GaAsAlAs ASPAT Diodes

As mentioned earlier in chapter one the tunnelling diode has many advantages

over conventional Schottky barrier diodes some of which are a large dynamic range

low power consumption and very weak temperature dependence This section will

discuss the effect of variable temperature applied to the GaAsAlAs ASPAT diode and a

similarly processed TiAu Schottky diode Two samples were fabricated together for

these studies (XMBE304 and XMBE104 representing an ASPAT and a SBD

respectively) The fabrication technique is exactly the same as has been discussed in

Chapter 3 In order to make it fair for direct comparisons as well as easy probing both

diodes were fabricated with the same emitter size (100times100microm2) The DC measurements

at different temperature were carried out using a Lakeshore Cryogenic probe station over

the range of 77K to 398K in 25K step interval

-002

-001

0

001

002

003

004

005

006

-2 -15 -1 -05 0 05 1 15 2

Cu

rren

t I

(Am

p)

Voltage V (Volt)

T=100K_Simu

T=100K_Meas

T=398K_Simu

T=398K_Meas

T=200K_Simu

T=200K_Meas

T=300K_Simu

T=300K_Meas

143

471 GaAsAlAs ASPAT diode vs TiAu SBD

Once fabrication and measurement were completed both DC outputs of the diodes

were characterised and analysed Figure 413 shows a semi-logarithmic plot for

measured current versus voltage as a function of temperature for ASPAT sample

XMBE304 In forwards bias the current changes for different temperatures from 77K

to 398K are less than 5 percent This confirms the very weak temperature dependence of

current transport as it is dominated by tunnelling through the barrier On the other hand

the backward bias shows the current changes at different temperature are slightly bigger

than in forward this due to band bending occurring faster (making the effective barrier

lower) and allowing thermionic emission to significantly contribute to transport of the

current The only other study of temperature dependence for the ASPAT was made by et

el RT Syme[15] but details were not stated in their report

Figure 413 Log Current vs voltage as a function of temperature for ASPAT sample XMBE304

The effective barrier height for the GaAsAlAs ASPAT diode is higher than that of

the SBD (See Figure 414) therefore there is an expectation of more limited thermionic

current flow in the ASPAT than the SBD As mentioned earlier the conventional

Schottky Barrier diode that is used in this study consists of a Gold Titanium and GaAs

(AuTiGaAs) interface which is the baseline for the temperature dependence study

0000001

000001

00001

0001

001

-15 -1 -05 0 05 1 15

Log

Cu

rren

t (A

)

Voltage (V)

T=77K

T=100K

T=125K

T=150K

T=175K

T=200K

T=225K

T=250K

T=275K

T=300K

T=325K

T=350K

T=375K

T=398K

144

Figure 414 The effective barrier high of the Schottky barrier diode and the ASPAT Diode [3]

The SBD epitaxial layers profile is as shown in Table 44 below Theoretically the SBD

obey thermionic emission transport[44] and its I-V characteristic is given by

119868 = 1198680[exp (

119902119881

119899119896119879) minus 1]

(46)

Where q is the electron charge V is the applied voltage across the diode n denotes the

diode ideality factor k is the Boltzmann Constant T is the absolute temperature in

Kelvin and I0 is the diode saturation current which is given by the expression

1198680 = 119860119860lowast1198792 exp (minus

119902empty1198870

119899119896119879) exp (minus120572120594119890

12120575)

(47)

here A is the area of the diode A denotes the effective Richardson constant Oslashb0 is the

barrier height at zero bias δ represents the thickness of interfacial insulator layer χ

denotes the mean tunnelling barrier and α = radic(4120587

ℎ)(2119898lowast) is a constant value The

ideality factor n is taken from the slope of the SBD current-voltage characteristic and in

this study its value varies from 1 to 2 (depending on temperature) In the case of the

ASPAT diode thermionic emission can also happen if a thicker barrier is used (~ 100Aring

or thicker) as shown by CS Kyono et el [104] who concluded that when a thicker

barrier of AlAs barrier is used the current transport is dominated by thermionic emission

145

Table 44 Epitaxial layer profile for Schottky Barrier Diode (XMBE104)

Layer Material Doping(cm-3

) Thickness(Aring) Bandgap (eV)

Schottky GaAs(Si) 500E+15 7500 14

Semiconductor GaAs(Si) 300E+16 7500 14

Semiconductor GaAs(Si) 100E+17 7500 14

Ohmic GaAs(Si) 500E+17 7500 14

Buffer GaAs(Si) 300E+18 7500 14

Substrate GaAs(Si) N+ 3000 14

The fabricated SBD was also measured and its I-V characteristic is plotted as a

function of temperature in Figure 415 Unlike the ASPAT diode the current at forward

bias for the SBD change enormously with temperature from 77K to 398K and at all

biases For the ASPAT diode the slight change in current only started after 08V bias as

the current starts to have some component of thermionic emission over barrier

Figure 415 Log Current vs voltage as a function of temperature for SBD sample XMBE104

In order to see clearly how much the current is changing in forward bias for both

ASPAT and SBDs diode a log current at different voltages versus 1000temperature is

plotted as shown in Figure 416

1E-08

00000001

0000001

000001

00001

0001

001

01

-2 -1 0 1 2

Log

Cu

rren

t (A

)

Voltage (V)

T=398K

T=375K

T=350K

T=325K

T=300K

T=275K

T=250K

T=225K

T=200K

T=175K

T=150K

T=125K

T=100K

T=77K

146

Figure 416 Influence of temperature on IV characteristic for both ASPAT diode and SBD

Semi-logarithmic plots of current (at V= 05 06 07 08V) versus inverse

temperature for both SBD and ASPAT are shown in Figure 416 When the temperature

is increased the current also increases in the SBD as a result of thermionic emission

over the barrier for sample XMBE104 This is in contrast to the temperature-

independent tunnelling through the thin AlAs barrier of sample XMBE304 ASPAT

diode where when the temperature is increased the current is almost constant

At low and high temperatures the ASPAT shows excellent temperature

independence with a constant current flow It exhibits a tunnelling current in excess of

values expected by the elastic tunnelling current calculation equation suggested by RT

Syme [16 18] above (Eq 1) using a Oslash value of 105eV (ΓGaAs to Γ AlAs tunnelling) By

contrast for the SBD at low temperature (77K-275K) the changes of currents were very

high and for every 02V there is an exponential change of more than 40 This

temperature dependent study was also reported in[68]

147

48 Conclusions

This chapter demonstrated the establishment of an excellent physical model and

comparison of room temperature I-V characteristics of GaAsAlAs ASPAT diodes for

different emitter sizes their scalability as well as an investigation of their characteristics

at different temperatures from 77K to 398K Simulation are validated on well-

characterized experimental data and excellent fitting which had been achieved in this

work permit the designer to extract all related parameters of heterojunctionmultilayer

ASPAT structures thus creating modification for future growth specification in order to

achieved precise designs

It is clear that the work which had been carried out in this chapter is able to

achieve with adequate accuracy a claim of reverse engineering capability The ability of

the GaAsAlAs ASPAT to act as a zero-bias detector has been analysed and compared

with the SBD It is clear that the temperature stability which is shown by the

GaAsAlAs ASPAT is much better than that of the SBD thus demonstrating that the

tunnelling current is dominant over the thermionic emission in ASPAT diodes

148

5 DC amp RF CHARACTERISTIC OF ASPAT DIODES

51 Introduction

To assist in circuit designs for any type of high-frequency circuits such as

millimetre wave detectors frequency multipliers and mixer circuits which are built

based on non-linear devices (diodes) an equivalent-circuit model for the diode is

required This is among the simplest and most effective method for analysing

semiconductor devices which work at high frequency where the electrical characteristics

measured obtained from the devices are extracted and presented in a circuit consisting of

lumped elements components (resistor inductor capacitor etc) However accurate DC

and RF measurement data is essential to extract the equivalent-circuit elements quickly

and correctly The extracted parameters values from the circuit that are taken into

account usually depends on bias and frequency associated with the device physically

which is also interrelated to the semiconductor material parameters device structure as

well as fabrication process flows

In this work the DC and RF data were derived from DC and S-parameter

measurements respectively These measurements were carried out both in-house and at

the University of Cambridge by a collaborator partner (Prof MJ Kelly) The I-V

characteristics of the diode obtained from DC measurements were measured from -2V to

2V while the S-parameters were carried out over a wide frequency range from 40MHz to

40GHz with eight different biases In this chapter the DC measurements for various

sizes of the diodes with analysis of their IV characteristics will be discussed The one-

port on-wafer ASPAT measurement setup as well as the de-embedding method will

also be explained Thereafter the equivalent circuit models with all lumped element

effect will be discussed This work is carried out with the help of the VNA which

principle has been described in Chapter 2 and Keysight ADS simulation tool All

technical details regarding the equivalent circuit models will be explained together with

the method used for the ASPAT diode evaluation The equivalent circuit model also will

cover the diode intrinsic elements such as Cj Rj and Rs and extrinsic elements ie CP

149

and RP Finally an equivalent circuit model with the small signal characterization of the

fabricated ASPAT diodes will be presented

52 General Analysis of Current-Voltage (I-V) characteristics of GaAsAlAs

ASPAT diodes

The recent development of state-of-art for DC measurement apparatus has led to

capabilities for high-level accuracy of measuring voltages to a few nano Volts and

current signals in the femto Amp range[107] This can easily be obtained by exploitation

of proper connections and high-quality cables connecting the equipment to the Device

under Test (DUT) In this work the DC measurements were carried out using an Agilent

B1500A Parameter analyser whose description was covered in Chapter 2

As was discussed in Chapter 3 the GaAsAlAs ASPAT diodes have been

fabricated with different mesa areas between 2times2microm2 to 100times100microm

2 but the smallest

size obtained with good I-Vs was 4times4microm2 In this section the focus will be on how the

extracted data can be expanded further for empirical modelling Figure 51 shows typical

results for measured ASPAT diodes with various dimensions to check for their

uniformity According to our standard procedure the DC measurement has to be

conducted prior to the RF to ensure the diode is in fully working order as this will later

save a lot of time during RF characterization

Figure 51 IV Characteristics of measured ASPAT diodes for emitter sizes of 4x4um2 6x6um

2 and

10x10um2 Note the good scalability

1E-10

1E-09

1E-08

00000001

0000001

000001

00001

0001-2 -1 0 1 2

Cu

rre

nt

De

nsi

ti (

Amicro

m2)

Voltage (V)

4x4microm^2

6x6microm^2

10x10microm^2

150

Figure 51 above demonstrates the IV Characteristics of measured GaAsAlAs

ASPAT diodes (XMBE304) for emitter sizes of 4times4um2 6times6um

2 and 10times10um

2 This

sample was processed using the dielectric bridge technique developed in this work It

can be observed that current per unit area for each dimension fits and scales to each

other The scalability of each diodes measurement is very important to ensure no process

related issues are hampering the devicersquos proper operations This also confirms that the

diodes are completely functional and can be used for the next stage of measurements

The advantage of having an excellent scalability of those diode sizes is that a prediction

of smaller emitter size can be made

This type of IV characteristic shows asymmetric behaviour which results from

the unequal spacer lengths of the device This behaviour is very useful for detection

application as it obeys a square law model The square law predicts that the current is

proportional to the square of the applied bias

119868 = 1198861198812 119908ℎ119890119899 119881 gt 0

119868 = 0 119908ℎ119890119899 119881 lt 0

(51)

To extract the first order effects of ASPAT diodes DC measurements which

result in asymmetric I-V characteristics are analysed The slope of the non-linear region

is used to determine the junction resistance (Rj) which is obtained from the first

derivative of voltage versus current (dVdI) The expression of Rj is given by

119877119895 =

120597119881

120597119868

(52)

In order to understand the relationship between Rj and diode sizes of the ASPAT the IV

characteristic for each diode displayed in Figure 51 is used to extract the Rj This has

been done by using the expression in equation (52) above and their response is plotted

against bias as displayed in Figure 52

151

Figure 52 Junction resistance versus voltage

As can be seen in the Figure 52 above the Rj for each device decreases strongly

when the voltage increases At zero bias the 4times4um2 devices show the highest Rj value

followed by 6times6um2 and 10times10um

2 devices The junction resistance at zero bias obtained

from the 4times4um2 diode is around 86KΩ while reducing by a third for the 6times6um

2 and

10times10um2 diodes with Rj of 27KΩ and 10KΩ respectively A diode with a smaller

forward current under the same applied voltage will exhibit a larger Rj For a good

millimetre wave detector a device with a large value of Rj is desirable since it will

provide high detection sensitivity

The slope at the IV characteristic contributes to an important parameter that is

commonly used by electronic manufacturers to describe diode specification namely the

video impedance (RV) which is also known as the non-linear resistance The RV which is

extracted from the real part of the diode small signal impedance is highly dependent on

the DC bias current and only weakly depends on the series resistance of the diode (RS)

Therefore the video impedance is given by

119877119881 = 119877119895 + 119877119878 (53)

Where RS is the series resistance of the diode whose value is normally very small and

does not contribute much to the whole slope and hence RV is dominated by Rj The RV

changes in behaviour if any DC current is flowing through the diode Practically small

DC current in the range of 1 to 10 microAmp or total zero bias is used to maintain the

appropriate RV value (1-2KΩ to several MΩ) RV will also determine the voltage

-10

10

30

50

70

90

110

130

150

-01 0 01 02 03

Rj(

)

Voltage (V)

6x6um^2

10x10um^2

4x4um^2

152

sensitivity of the whole detector circuit This will be explained further in the next

chapter In the case of a detector with an amplifier RV of the diode acts as the RF

impedance which needs to be matched with the video amplifier ( impedance looking into

the diode from the amplifier)[108 109]

The quotient of the second order derivative to the first derivative

((d2IdV

2)dIdV)) when calculated from the whole I-V characteristic translates directly

into a curvature coefficient (k) This is the most commonly used figure-of-merit to

quantify diode nonlinearity at zero bias Figure 53 below shows the variation of k with

bias and more importantly the zero bias rectifying action for device sizes of 4times4 um2

6times6 um2 10times10 um

2 This parameter which represents the small-signal rectifying

action of the diode will affect the performance of the detector (voltage sensitivity)

Detailed discussions on how this parameter effect the detector performance will also be

discussed in the next chapter

Figure 53 The rectifying action of the diode at zero voltage which gives the ASPAT high sensitivity

near zero bias detection

Figure 53 above shows calculated curvature coefficient of the measured I-V

characteristics from the same diodes shown in Figure 51 The highest k value is

obtained from the diodes with size of 4times4um2 followed by 6times6um

2 then 10times10um

2

The curvature coefficient decreases sharply as the bias increases for each diode This can

be attributed to a significantly increasing number of electrons that tunnel through the

thin barrier which were accumulated in the 2DEG formed in the intrinsic spacer region

-5

0

5

10

15

20

25

30

-001 004 009

Cu

rvat

ure

Co

effi

cien

t(V

-1)

Voltage (V)

k(10x10 um^2)

k(6x6 um^2)

k(4x4 um^2)

153

An ASPAT diode with a smaller size will have a larger Rj with a corresponding smaller

current under the same bias condition and hence will demonstrate a larger k value In

this calculation the curvature coefficient at zero bias obtained from 4times4um2 6times6um

2

and 10times10um2 diode is 23V

-1 17V

-1 and 16V

-1 respectively

A summary of the ASPAT diodes parameters obtained from measured I-V

characteristics that have been translated into first and second order differentials are

gathered in Table 51 below and compared to other diodes in the literature

Table 51 Performance of the ASPAT diode obtained from measured IV characteristics in this work

Sample Rj(Ω) k(1volt)

ASPAT 10times10 microm2 10K 16

ASPAT 6times6 microm2 27K 17

ASPAT 4times4 microm2 86K 23

Ge Backward diode 182K[110] 159[110]

InGaAs Backward diode 154[110] 23[110]

Sb Backward diode 5K[111] 47[111]

Si-Backward diode 135K[112] 31[112]

PDB 15K[8] -

AlGaAs SBD 20-100K[113] 34-38[113]

GaN HBD - 16[114]

From Table 51 above it is clearly that the ASPAT diode has a comparable value of Rj

and k to existing detector diode in the research community and in the commercial

market Based on literature of each diodes stated in the table the key to obtaining a high

value of k at zero bias is to minimize any forward tunnelling current Furthermore the

largest ASPAT diode used here (10times10microm2) has very close performances to that of a

commercial diode ie discrete Ge backward diode (ref[110]) where both Rj and k value

are close to each other

53 RF Test Fixture Theory and Experiment

RF measurements differ from DC measurement as they are more complicated

and it is necessary to comprehend the basic measurement principles to achieve

meaningful data This is obligatory especially for on-wafer RF characterization and

154

analysis to attain precise results Most of the electronics component measurements

which have input and output for instance antenna amplifier cables etc are based on a

two-port network configuration The characteristics which can be extracted from these

components are usually used to define their impact on a more complicated system

The performance of the two-port network can be described by a few parameters

ie scattering (S-Parameter) admittance (Y-Parameter) Impedance (Z-parameter) and

Hybrid (H-Parameter) However the S-parameter approach is favoured for high-

frequency measurements as it is relatively easier to characterize the microwave

performance and is able to convert to other parameters when necessary The advantage

of S-parameters is that they can straightforwardlydirectly convert into other two-port

parameters as mentioned above in term of currents and voltages[115] In fact to obtain

the device capacitance the appropriate S-parameters needs to be transformed into Y-

parameters using specific equations Furthermore the devicersquos cut-off frequency can

also be obtained when S-parameter measurements are performed over a wide frequency

range

531 On-Wafer Measurement and Small Signal One-Port Characterizations

In this work the arrangement of the RF measurement setup is assumed to be a

linear system as small voltage amplitude signals are used this means that the signals

have only a linear effect on the network without any gain compression or attenuation

The assumption is still acceptable even though the typical ASPAT is characterised as

non-linear in nature because it is a passive device which will act linearly at any input

power level

Generally the S-Parameter measurements on a diode can be adequately and

suitably performed using a one port measurement The technique used to characterise the

output is similar to the two ports technique but only incident and reflected waves are

used to characterise the input and output ports of the device Essentially this is because

the ASPAT has only two terminals and it is a passive device like other diodes

Therefore the analysis will revolve around the S11 parameter Figure 54 below show the

S11 is a ratio of reflected wave to the incident wave

155

11987811 =

119877119890119891119897119890119888119905119890119889

119868119899119888119894119889119890119899119905=

1198871

1198861 119908ℎ119890119899 1198862 = 0

(54)

A VNA as described in Chapter 2 is used to measure the ASPAT diodes This

powerful equipment is able to measure S-parameters up to 40 GHz To conduct accurate

S-parameter measurement at the diode the measurement setup must be calibrated prior

to the actual measurements taking place

54 Device Calibration

541 Open and Short De-Embedding Technique

Further calibration to be made involves anything related or attached to the

device The co-planar waveguide (CPW) bond pad and interconnect line that are

attached to the intrinsic diode are the main contributors of the errors also are required to

be calibrated In general the bonds pad could generate a capacitance (parasitic) in

parallel with the intrinsic diode and its contribution depends on the size of the bonds pad

as well as the operating frequency Meanwhile the CPW and interconnect line may

cause a parasitic inductance in series with the diode

The method that is used to get rid of this parasitic is called de-embedding and the

most common technique to realise it is by introducing OPEN-SHORT structure[116]

This method is based on a lumped-elements model Parasitic elements of the diode

De-Embedding Structure

Incident wave

Reflected wave

One-port device

a1

b1

Figure 54 One port S-parameter measurements

156

equivalent circuit correlate directly to the access section of the CPW hence can be

derived from de-embedding structures The aim of the de-embedding technique is to

represent these parasitic elements so that the one-port characteristic of the actual diode

can be determined

The two types of de-embedding structure OPEN and SHORT are conventional

techniques that are widely used in this study The design of all structure must be

identical (in size) to the device to avoid any discrepancy It is very simple to design all

these three structures for example open structures are obtained by eliminating the diode

layout and keeping the bond pad layer only The short structure just adds a bridge and

ensures ground and signal pad are connected to each other Through structures are

realised by disconnecting both ground pad and leaving the signal pad to connect to each

other

To gain more accuracy this external effect must be removed by the implementation of

de-embedding structures on the same tile as the actual device Figure 55 shows the

fabricated de-embedding structures used in this study

Figure 55 Fabricated open short de-embedding structure and real device (zoomed) use for RF

calibration and measurements (Note Images are not to scale)

In summary the de-embedding which is used to extract out the parasitic elements

from entire single diode measurement is a very important step as normally on-wafer

measurement requires coplanar waveguide (CPW) to access the diode structure (active

region) The CPW will have some effects which will disturb the accuracy of the device

characteristics

157

55 S-Parameter Measurement Result and Analysis

This section will only present RF measurement results after all VNA setup and

calibration were performed The S-Parameter measurements were carried out on ASPAT

diodes at five different DC biases from -2 to 05 volt with a sweep frequency from

40MHz to 40GHz using a calibrated VNA and the input power was fixed at -30dBm

The measurement procedure as described in Chapter 2 was performed on the device

(on-wafer) equipped with the appropriate bond pads This is important to ensure the

results obtained are valid The reason for using different biases is to find at what voltage

the device capacitance is fully depleted This is also very important in determining the

cut- off frequency of the devices

In this research two phases of the experiment on the S-parameter measurements

were carried out The first phase is to qualify the process flow ie for manufacturability

and repeatability which can be obtained from the consistency of the result The S-

parameter measurements taken on the same wafer dies are repeated several times on

different GaAsAlAs ASPAT diodes There are three different tiles taken from 3

different wafers namely XMBE304A XMBE304B and XMBE304C carried out in

this experiment The repeatability tests are done mostly on the large devices (15times15microm2

up to 100times100 microm2) and the results are analysed based on the reflection coefficient (S11)

on Real and Imaginary measurements

In the second phase the measurement is toward producing devices that can

perform at high-frequencies This can be realised by utilising small emitter size devices

(4times4microm2 6times6 microm

2 and 10times10 microm

2) The measurement results of these devices will be

used to build the equivalent circuit models while both the intrinsic as well as the

parasitics of the device will be evaluated Hence all the values obtained from these S-

parameter measurements will be used to design the device that can be used in

millimetres-wave applications As can be seen in Figure 56 below the extracted S-

parameter measurement results comprise of a reflection coefficient (S11) for real

imaginary and Smith chart for XMBE304A While these measurement results are

extracted at zero bias voltage the other bias voltages will be used to extract the

capacitance This will be described in the final section of this chapter

158

551 Diode to diode uniformity

In order to study within tile uniformity and reproducibility statistics of the RF

performance five devices of different mesa sizes in the same tile (XMBE304A) were

measured at zero bias and represented in term of Real and Imaginary reflection

coefficient (S11) the uniformity check is carried out at three different frequencies step

under 15GHz since the cut-off frequency for these big devices is relatively low at about

~20GHz on average The variation of the reflection coefficient is taken from the

percentage of the (standard deviationmean values) for all five diodes from this run The

following figures show Real and Imaginary S11 of large mesa area ASPAT diode from

15times15microm2 up to 100times100microm

2 which are represented by lines graph in a few different

colours

Figure 56 Real S11 RF measurements for (XMBE304A) were done for four devices from

15times15microm2 to 100times100microm2 device sizes for within-wafer uniformity check

Figure 57 Imaginary S11 measurements (XMBE304A) to qualify the repeatability and

reproducibility were done on four devices from 15times15 microm2 to 100times100 microm2 at zero bias

159

The variance data extracted from the graph (Real S11) above for each device within-

wafer (device to device) uniformity study is summarize in Table 52 below

Table 52 Device to device uniformity check for large ASPAT diode

Device Size 100times100 microm2 50times50 microm

2 30times30 microm

2 20times20 microm

2 15times15 microm

2

Variation 5GHz 181 115 405 151 145

Variation 10GHz 106 133 522 424 293

Variation 15GHz 119 198 281 76 509

The majority of diodes show that the variations of S11 measurements are below

3 and only a few are below 8 These finding still can be considered as good for

manufacturing control since absolute I-V characteristics reported in [63] is set by

designerrsquos specification to be not more than plusmn10 variation Further extensive RF

measurements were carried out by the research collaboration with the University of

Cambridge on the same GaAsAlAs ASPAT diodes wafer [117] In their study they

focused on 50times50 microm2

mesa size 17 of diodes were chosen to be measured The study of

uniformity of RF characteristic only focused on frequencies below 20GHz The same

approach was used to get the variation of the reflection coefficient for all 17 diodes but

this work was carried at four different frequencies Table 53 shows the zero bias S11

result for four different frequencies and standard deviation of the devices

Table 53 The variation of reflection coefficient (S11) for GaAsAlAs ASPAT diode at four different

frequencies[117]

Frequency (GHz) 5 10 15 20

Variation () 197 243 26 276

From the results the variations of 50times50 microm2 mesa sizes measured in-house and

at the University of Cambridge are comparable with all variations showing good

uniformity ie recording variations below 3 This indicates that the RF performance

of the GaAs AlAs diode is valid and reproducible and is thus considered as a good

achievement for manufacturing Once the reproducibility and repeatability of the large

devices showed stable results the fabrication process then continued to obtain smaller

emitter size for work at high-frequencies

160

552 Wafer to wafer uniformity

Other RF measurements were conducted on sample XMBE304B which was

fabricated in-house using the same process steps but the only difference from

XMBE304A was the use of SiN3 as dielectric In this run three different mesa sizes

were measured (15times15mmicro2 20times20microm

2 and 30times30 microm

2) and Real and Imaginary S11

plotted against frequency The RF performances of both samples are gathered in one

graph as shown in Figure 58 below

Figure 58 Wafer to wafer uniformity test on 1st and 2

nd RF measurement(XMBE304B) to qualify

the repeatability and reproducibility were done for 30times30microm2 to 15times15 microm

2 device sizes (Real and

Imaginary) Note blue colour is XMBE304A and red colour is XMBE304B

For this wafer to wafer uniformity study four diodes with three different sizes as

specified previously were measured from sample XMBE304B and four diodes from

previous measurements of XMBE304A The blue line in Figure 58 represent

measurement result of real and imaginary for sample XMBE304A while the red line

161

represents XMBE304B The uniformity data is compared at three different frequencies

and gathered in the table below

Table 54 Wafer to wafer uniformity check on XMBE304A and XMBE304B

Device Size 30times30 microm2 20times20 microm

2 15times15 microm

2

Wafer A vs wafer B variation 5GHz 305 31 1 314

Wafer A vs wafer B Variation 10GHz 352 344 329

Wafer A vs wafer B Variation 15GHz 321 376 359

As can be seen in the Table 54 above the wafer to wafer uniformity is rather

large (30) on average The main reasons being that sample XMBE304A was

processed by utilizing S1805 as a dielectric layer while sample XMBE304B used

Si4N3 Although the process steps are similar for both wafer processing the use of

different dielectric layer will influence the diode parameters especially resistance and

capacitance as the dielectric constant for each materials is different Secondly the wafer

processing is not run concurrently at the same time thus the moisture and temperature in

the clean room might differ for both processing Although the wafer to wafer uniformity

test for this run might not be favourable for manufacturing tolerance at least the use of

different dielectric layer shows some significant result in term of capacitance resistance

effect to the GaAsAlAs ASPAT diode

553 Small devices RF measurements

The first objective of this study was to make smaller size mesa devices ie

1times2microm2 1times3 microm

2 2times2 microm

2 and 3times3 microm

2 However for GaAsAlAs ASPAT type this is

difficult to achieve in practise These issues were discussed in detail in Chapter 3

Hence the smallest emitter size that yields repeatable and reproducible results was

4times4um2 The final measurement which was done on sample XMBE304C focused on

small devices The measurements were done on four devices two with the diode bond

pads sitting on substrate (GaAs SI) and the other two sitting on dielectric layer (Si4N3)

Figure 59 below shows three measured results obtained from sample XMBE304C

using the 3rd

Gen Mask

162

Figure 59 RF measurement performed on device sizes of 10times10microm2 6times6mmicro

2 and 4times4microm

2 (Real and

Imaginary) Note that green red and blue colour represents 4times4microm2 6times6mmicro

2 and 10times10microm

2 diodes

respectively

Figure 510 RF measurement performed on device sizes of 10times10 microm2 6times6 microm2 and 4times4 microm2

(Smith Chart) Note that green red and blue colour represents 4times4microm2 6times6mmicro2 and 10times10microm2

diodes respectively

30MHz

40GHz

163

As can be seen in Figure 59 (Real Imaginary) and Figure 510 (Smith Chart) are

obtained from measurement of four diodes in the same tile The diode to diode

uniformity that is sitting on the same platform obtained at 35GHz frequency in this run

on average is ~15 25 and 1 for 4times4microm2 6times6mmicro

2 and 10times10microm

2 diodes

respectively On the other hand the uniformity between diode to diode sitting on

dielectric and substrate is quite high due to different capacitance value of devices on

average ~7 are attained from three different sizes of diode

From the measurement results above the Real S11 measurement of four different

sizes show the same trend for each frequency At low frequency resistances for each

diode is high as the S11 value is large At intermediate frequency the values drop

tremendously for big devices (30times30 mmicro2 and 20times20 mmicro

2) ie in Figure 57 At high

frequency all diode reach saturation limit as the value are constant Small devices

(4times4mmicro2 6times6mmicro

2 and 10times10microm

2) ie Figure 59 show S11 values that are higher than

those of large devices as smaller emitter diode have larger resistance value

The imaginary S11 value also shows the same trend as for big devices However

for small devices in this run (4times4mmicro2 6times6mmicro

2 and 10times10microm

2) the S11 value keep

dropping toward negative values at increasing frequencies This indicates that bigger

devices with positive value at high frequency are more capacitive than the smaller

devices It is worth mentioned that the capacitance and inductance values for device

sizes of 4times4microm2and 6times6microm

2 come from the CPW layouts and these are dominant while

for device sizes of 15times15 mmicro2 and above the device capacitance itself is dominant

The Smith Chart shows the reflection coefficient (S11) as a function of the

applied frequency (30MHz to 40GHz) All measurements from each mesa size follow

unique impedance circle which is that most of the lines are at the lower right outer ring

This means that the diode capacitance value is frequency dependent For the case of

10times10microm2

devices these impedance circles are mostly toward the outer ring meaning a

higher capacitance than the other two mesa dimensions All the device constantly follow

the outer ring without crossing any real axis at any frequency point meaning that the Cj

is not shorted at the maximum 40GHz measurement frequency (not reached cut-off

frequency) Therefore the entire small GaAsAlAs diodes in this run have capability to

work in the millimetre wave frequencies range

164

56 Extracting RF models of ASPAT at Zero Bias Voltage

The methodology used in the S-Parameters measurement for high-frequency

analysis must ensure that the derivation of the equivalent circuit corresponds to their port

characteristics In other words the component representing the ASPAT in the equivalent

circuit model must have physical significance otherwise the circuit will be meaningless

The fabricated ASPAT diodes as discussed in Chapter 3 have the cross section shown in

Figure 511

There are two main components that can be extracted from the fabricated

ASPAT depicted above ie intrinsic and parasitic The intrinsic refers to the main

structure of the diode itself and are represented by three bias dependent elements

namely Junction Resistance (Rj) Junction Capacitance (Cj) and diode Series Resistance

(Rs) The parasitic is the elements related to the bond pad of the anode and cathode as

well as interconnects They are represented by parasitic inductance (LP) resistance (RP)

and capacitance (CP)

The diode parameter extraction is different from the three terminal devices

(FETs) in the sense that FETs are a kind of direct extraction in which all the elements in

the transistor have linear functions to the port characteristics ie S-parameter Y-

Parameter Z-Parameter and can easily be solved by the matrix calculation method for

those particular parameters[118] The same extraction method cannot be applied to the

diodes because its elements will embroil with each other Therefore only one method is

used to extract the diode element which is optimisation by tuning the initial value toward

the measured S-Parameter values

Figure 511 Cross-sectional of fabricated ASPAT diode and the corresponding equivalent circuit

model

165

The strategy used to model the ASPAT is based on an initial fitting value of the

lumped elements to the extracted value from measurement on three S-parameter graphs

(real imaginary and Smith chart) for the reflection coefficient (S11) The refinement is

accomplished by optimisation and fine tuning of the values which result in minimum

error between extracted and modelled values Figure 517 (on page 153) shows fitted S-

parameter result for extracted and model numbers with each one fitted in a single line as

an example However to achieve this excellent fitting key prior steps have to be used

de-embedding fitting the intrinsic value and optimisation

561 Extraction of ASPAT parasitic element

Once the S-parameter measurements achieve stability repeatability and

reproducibility for each measurement in term of S11 results as mentioned above the

results of the de-embedding structure which had been measured prior to the device

structure are then extracted to form a well-defined equivalent circuit In order to build

and analyse the equivalent circuit firstly the measured data is imported into the ADS

software prior to any fitting This can be realised via the ldquoStart The Data File Toolrdquo

features provided by this particular software When successfully imported the data is

read by the function S2PMDIF (These files are a natural extension of two-port S-

parameter Touchstone files) as depicted in Figure 512 below

Figure 512 The S-parameter Touchstone file is used to read the measured files

166

For the open and short techniques after de-embedding the equivalent circuit

model which is represented by mainly a capacitor and an inductor is built The open

structure requires resistance and the capacitance values of 20KΩ and 26fF respectively

connected in parallel to be well fitted to the real imaginary and Smith chart (S11) output

On the other hand for the short structure the Real imaginary and Smith chart (S11) have

to satisfy the values of resistance and inductance elements of 1Ω and ~47pH respectively

connected in series These values strongly rely on the bond pad or CPW dimension and

length The Equivalent circuit models and fitted data as well as measurement can be

seen in Figure 513 and Figure 514 below

Figure 513 Equivalent Circuit Model of Open CPW or de-embedded structure

Figure 514 Equivalent circuit model for short de-embedded structure

To satisfy the equivalent circuit a self-consistence method introduced by

Ren[119] is utilised This approach accurately extracts the CPW capacitance (Cpad) and

inductance (Lpad) as well as intrinsic Junction capacitor (Cj) which is attained from the

one-port S-parameter measurements Therefore the pad capacitance introduced by the

self-consistence method for the open structure can be expressed by

167

119862119875 =

119868119898(11988411119874119901119890119899)

120596

(55)

Lpad which represents the short structure is given by

119871119875 =

1

120596(119868119898 (11988411119874119901119890119899minus 11988411119878ℎ119900119903119905

))

(56)

Here Y is the admittance parameter (Y-parameter) converted from the S-parameter

measurement data and ω is the angular frequency The extracted measurement data

represented in the equivalent circuit fits with the simulated data in three S11 graphs as

can be seen in Figure 515 below From the Smith Chart it can be clearly seen that both

open and short S11 results are on the circumference which means the resistance of the

short structure is very small while in the open it is very large Additionally the

calculated Cpad and Lpad using Equations 55 and 56 above produce results similar to

those obtained in the equivalent circuit model for the open and short structure The

values are ~25fF and ~45pH respectively These data completely verify and validate

both results

Figure 515 Smith chart representative S-parameter measurement for short (left) and open (right)

CPW The blue lines represent simulated data and the red is measured data

Short

Open

168

562 Extraction of ASPAT intrinsic elements

Once the parasitic elements are determined it is easy to build a complete ASPAT

equivalent circuit The ASPAT is not like other tunnelling diode which their equivalent

circuit models widely studied ie RTD [120] IMPATT and PDB The only literature

which reports ASPAT equivalent circuits can be found in [15] and other RT Symersquos

journal paper[16] Fortunately its equivalent circuit model is not much different

compared to other diode video detectors Thus other literature which is based on

Schottky diode equivalent circuit model used for detector application can be referred to

The simplest form of ASPAT equivalent circuit and other video detectors intrinsically

consist of junction capacitance (Cj) series resistance (RS) and junction resistance (Rj)

First and foremost to extract the equivalent circuit one must know the theory behind

each parameter that is developedbuilt as a spine to become a complete element This is

vital to ensure the equivalent circuit is correct In the case of the ASPAT Cj is predicted

from a simple fully depleted parallel plate capacitor approximation which was discussed

previously in Eq (230)[15] Additionally for the S-parameter measurements the Cj can

also be validated by the self-consistence method mentioned earlier and thus can be

expressed by

119862119895 =

[

(1

120596)

1

1

119868119898 (11988411119905119900119905119886119897minus 11988411119874119901119890119899

)+

1

119868119898 (11988411119874119901119890119899minus 11988411119878ℎ119900119903119905

)]

(57)

This approach helps to verify both the fully depleted parallel plate capacitor in S-

parameter measurements The basic component which is responsible for the ASPAT

series resistance RS was discussed in detail in Chapter 2 RS and Cj are key contributors

to the high-frequency operation as expressed by the device cut-off frequency Equation

(58) below

119891119888119906119905minus119900119891119891 =

1

2120587119877119904119862119895

(58)

The R and C parameters must be kept as low as possible in order to obtain high cut-off

frequencies for millimetre wave applications From the fabrication point of view Cj can

169

be reduced by making as small a diode emitter size as possible while for RS reducing

the D gap is of paramount importance as it dominates the series resistance The ASPAT

contact resistance in the electrodes (contacts between metal and semiconductor) can be

reduced by using high doping in the ohmic layers

The junction resistance (Rj) of the ASPAT is taken from the 1st derivative or

slope of the current-voltage characteristics Normally the value of Rj is very large

(several kilo Ω) compared to Rs The small signals ASPAT equivalent circuit built with

intrinsic and extrinsic components is shown in Figure 516 below while the fitting

results is shown in Figure 517 and Figure 518

Figure 517 Excellent fitting between measured (red colour) and simulated (blue colour) for S11

(Real and Imaginary) results for various small device designs

Rj

Cj

Cpad Rpad

Rs Lpad

Figure 516 Equivalent circuit of the ASPAT diode

170

Figure 518 Excellent fitting between measured (red colour) and simulated (blue colour) for S11

results (Smith Chart) for various small device designs

The equivalent circuit that was built for the ASPAT is taken from sample

XMBE304C with emitter dimensions of 4times4 microm2 6times6 microm

2 and 10times10 microm

2 These

devices are expected to work in the millimetre-wave region and have cut-off frequencies

(intrinsic) of ~650GHz ~200GHz and ~100GHz respectively

Table 55 Comparison between calculated (fully Depleted) and extracted (different biases) values

from equivalent circuit parameters for different ASPAT mesa sizes at zero bias voltage

Parameters 4times4microm2 6times6 microm

2 10times10 microm

2

Calculated

(fully

depleted)

Extracted Calculated

(fully

depleted)

Extracted Calculated

(fully

depleted)

Extracted

0V -05V -1V 0V -05V -1V 0V -05V -1V

Cj(fF) 879 23 7 6 198 55 148 139 549 171 486 46

Cpad(fF) - 15 15 15 - 15 152 152 - 15 15 15

Lpad(pH) - 45 43 42 - 50 473 473 - 51 51 46

Rj(KΩ) - 90 833 522 - 35 392 392 - 12 125 13

Rs(Ω) 99 11 11 11 67 95 8 7 41 95 45 37

fcut-off

Cj(GHz)

1828 629 2066 241

1

1208 192 1344 163

5

710 107 728 935

fcut-off

Cj+Cp(GHz)

- 380 658 688 - 151 663 781 - 98 556 705

171

The focus in this study is purposely to build ASPATs as zero bias detectors that are able

to work in the millimetre and sub-millimetre frequency range therefore all the

parameters which are obtained from equivalent circuit were extracted at zero bias

voltage Theoretically the calculations which are derived from both self-consistence amp

theory can only be solved for fully depleted device capacitance (using a parallel plate

configuration) Hence both extracted and calculated results are compiled in Table 55

Noticeably the calculation can only produce the intrinsic parameters of the

ASPAT for fully depleted capacitance On the other hand both intrinsic and extrinsic

parameters of GaAsAlAs ASPAT are obtainable from extraction and thus help to

determine at what bias the diode is start to deplete The junction and series resistance (Rj

and RS) of each dimensions shown in Table 55 above were achieved by fitting the

elements of the equivalent circuit with the three measured S11 graphs whereas the Cpad

and Lpad were extracted by utilising the self-consistent method from the S-parameter

measurements which is fitting the de-embedding structure Additionally the Cj values

are obtained via fitting the measured S11 data and employing the self-consistence

approach Results obtained from both techniques are identical

At zero bias all extracted junction capacitance from each device sizes are very

different from the calculated one while the extracted series resistance are closer to the

calculation This means that Cj is a highly voltage dependent parameter and Rs is

voltage independent but solely dependent on device structure and material used to

fabricate it The extraction at -05V and -1V shows that the values of junction resistance

is changing for most of the devices which means this parameter also rely on bias voltage

as discussed earlier in Section 52

The cut-off frequency for each calculated devices are near the THz range even

for the 100microm2 emitter area However with the introduction of parasitics elements ie

pad capacitances fcut-off is degraded tremendously Therefore it is important to make sure

all the intrinsic elements have optimum values so that the target operating frequency of

the ASPAT diode can be met Due to this it is advisable to operate the devices at no

more than 13 of fcut-off when designing detector systems

The parameters extraction at -05 and -1V also show that Cj values are closer the

calculated ones which means the ASPAT diode is reaching full depletion

172

563 Capacitances -Voltage (C-V) Extraction

Theoretically the junction capacitance of the ASPAT is calculated from the fully

depleted formula 119862119895 = 휀119900휀119903119860119889 which was also discussed in Equation (230) in Chapter

2 in page 43 Its value depends on the change of voltages to depletion at the emitter

contact[15] and make it one of the voltage dependent parameter for the diode[121]

Therefore the C-V characteristic of the GaAsAlAs must be precisely extracted

In practice the capacitance is difficult to measure due to the very low resistance at zero-

bias However alternatively it can be measured and extracted by applying different

voltage and identifying the point at which there is change which essentially represents

full-depletion Apart from these it can also be extracted from S-parameters measurement

which is then converted to Y-parameters A C-V characteristic of the GaAsAlAs

ASPAT from XMBE304C is extracted and plotted as depicted in Figure 518 below

Figure 519 Capacitance extraction showing full depletion at -025 volts (modelled capacitance vs

Voltage)

From the graph shown in Figure 518 the capacitance is extracted at eight

different biases for 4times4 microm2 6times6 microm

2 and 10times10 microm

2 The devicersquos junction

capacitance for each dimension increases and reaches a maximum value at 025V There

are additional quantum capacitance effect which comes from an increase in the negative

charges in the 2DEG region (when band bending happens creating an accumulation

0

50

100

150

200

250

-2 -15 -1 -05 0 05

Cap

acit

ance

(fF

)

Voltage (V)

Cj(4x4microm^2)

Cj(6x6microm^2)

Cj(10x10microm^2)

173

region at the barrier) This charge is imaged by the positive charge in the whole

depletion region Increasing the voltage toward positive values leads to a lowering of

the AlAs barrier and thus allowing thermionic emission to take place after certain bias

values leaving only the depletion capacitance and making the quantum capacitance

negligibly small

In the reverse bias case the device junction capacitance reaches a saturation

(fully depleted capacitance) at a voltage of -025V and remain constant up to -2V If the

reverse bias voltage is increased further the ASPAT may reach breakdown Therefore it

is important to know how far the diode can withstand applied reverse bias to ensure it

can still give full performance

57 Conclusions

In this chapter scalable DC characteristics of GaAsAlAs ASPAT diode derived

from three different emitter sizes of 4times4 microm2 6times6 microm

2 and 10times10 microm

2 was

demonstrated The current density obtained at zero bias is several microAmicrom2 These allow

1st order differential effect to exhibit high value of Junction resistance (Rj) at zero bias

However Rj is highly bias dependent The 2nd

order differential effect on IV

characteristics display a high value of curvature coefficient leading to high voltage

sensitivity when applied in millimeter wave detector applications These two parameters

are vital in the design of millimeter wave detectors and especially those operating at zero

bias

Subsequently RF measurement up to 40GHz of uniformity study for both within

wafer and wafer to wafer variance were undertaken An average uniformity below 7

was obtained for within wafer study on large device area ( 15times15 microm2 to 100times100 microm

2)

while for small device area ( 4times4 microm2 to 10times10 microm

2) a smaller 3 uniformity variance

was achieved in average However for wafer to wafer study the variant uniformity was

quite high at around 30 on average for relatively large device (15x15 microm2 to 30x30

microm2) This was mainly attributed to different dielectric layers used in the process flows

of the sample rather than fundamental MBE control of the AlAs barrier thickness

174

It was demonstrated in this chapter that careful on-wafer RF measurements of small

size GaAsAlAs ASPAT diodes allow accurate device parameter extraction of both

extrinsic and intrinsic parameters The extrinsic parameters are namely pad capacitance

and inductance with obtained values of 26fF and 47pH respectively These values were

obtained from de-embedding structure fabricated on the same tile as the real devices

The intrinsic parameters such as junction capacitor junction resistor and series

resistance had different values according to device dimensions The smallest zero bias

value of Cj obtained from 4times4 microm2 diodes was 23fF ensuring a high cut-off frequency of

380GHz and hence in the next chapter a 100GHz detector will be presented working at

slightly less than 13 of this cut off frequency

The C-V data extraction confirmed that the fully depleted capacitance started to

happen at around -025V The maximum junction resistance occurs at +025V largely

caused by the depletion region and an additional quantum capacitance effect CQ This

effect is strongly related to the size of a 2DEG which occurs under forward bias (01V to

025V) and can be reduced by having a smaller thickness AlAs barrier

175

6 MILLIMETRE WAVE GAASALAS ASPAT DETECTOR

DESIGN USING ADS

61 Introduction

The ASPAT diode having features of non-linear IV characteristics at zero voltage

make it useful for signals rectification ie detector and mixer for millimetre-wave

applications Additionally ASPAT diodes have a range of advantages such as large

dynamic range strong temperature insensitivity etc[15]over other rectifier diode This

makes ASPAT an appropriate candidate in RF detection applications Since 1940 the

only two terminals device that has been the workhorse for RF applications is the

Schottky Barrier Diode (SBD)[9] In its earliest form the SBD was built based on a

point-contact device which could not perform at high frequencies It was then developed

to work at higher frequencies by exploitation of epitaxial structures [10] and to date the

SBD remains the mainstay of two terminal devices that are able to work in the

millimetre and submillimetre-wave regions However as discussed in Chapter 4 the

performance of SBD is degraded at extremes of temperatures and these circuits

employing SBDs require temperature compensating circuitry Thus there is additional

complexity associated with technologies and applications related to SBDs

Before this work was carried out no model for the ASPAT diode as detector had

been available or developed especially using the empirical modelling ADS software

When the ASPAT diode was first introduced its function was conceptually explained it

was then built and tested to compare with other microwave detectors at X-band

frequency (95GHz) The comparisons were made in terms of detector parameters ie

sensitivity dynamic range temperature dependence etc[15] These early works lead by

RT Syme et al supplied the basic knowledge to model the ASPAT diode as a zero-bias

detector for mm-wave frequency gt100GHz For the SBD many models and equivalent

circuit approaches have been reported [122 123] The modelling of conventional SBD is

mostly implemented through fitting the S-parameter curves of the model to the

experimental one This approach is also carried out in this research since it is accurate to

predict the performance of the device under test [124]

176

This Chapter aims to introduce low cost reliable and sophisticated detector design

based on ASPAT diodes which is believed to be able to improvereplace SBD in

millimetre-wave applications especially in imaging The focus was on developing and

establishing an appropriate circuit design that suit the new ASPAT diode for such

applications The detector sensitivity as its key parameter ultimately limits the quality

and acquisition time of the detector In the subsequence section the theory of detection

including both direct and heterodyne will be discussed This is followed by definition of

detector characteristics of interest as well as noise consideration Section 65 present the

main focus of this chapter which is the development of 100GHz ASPAT detectors and

their result will be explained in term of all detector characteristic of interest

62 Detection Theory

Any incoming signal such as RF microwave or mm-wave in the form of envelope

function or single wave can be detected by rectification of the signal using a nonlinear

device ie transistor or diode The input and output signal (RF) signals are normally in

the form of amplitude as a function of time Typically the detector output is a low-

frequency signal known as the video signal which has amplitudes that are proportional to

the square of the input RF signalrsquos voltage amplitude

A complete receiver system as shown in Figure 61 below consists of receiver

antenna and a circuit designed to extract the signal and then amplify it The function of

the receiver system is the converse of the function of the transmitter side At the receiver

side the antenna is used to receive the signal it then conveys the signal to the extraction

circuit for detection as the information-bearing part of the signal (using the nonlinear

device as its heart) The signal is finally amplified to avoid any information strength

decay Additionally in a digital system the output signal which had been processed by

the detector circuit has to maintain an optimum input signal conveyed to in-phase and

quadrature (IQ) demodulators The output signal will go through a low pass filter then

to an analogue-to-digital converter (ADC) and thus a digital baseband output signals

will be produced

177

Figure 61 Block diagram represent a complete direct receiver system

There are two types of millimetre wave integrated circuit (MMIC) used for

detection purposes namely direct detectors and heterodyne detectors The direct detector

MMIC is the simplest circuit used for detector applications and has the simplest way of

extracting the RF information Due to its simplicity the direct detection method is

inexpensive and most attractive method used for measuring power in RF Laboratories

and Industries This detection scheme is also sometimes known as video detection[125]

The simplest way of explaining the detection process is that the incoming RF or

microwave signals depicted in Figure 62 below with an appropriate input power (Pin) is

rectified by using a diode and results in a corresponding output voltage (Vout) A detector

IC designed based on diodes is normally able to rectify very low levels of RF power (lt-

40dBm) then produces an output DC voltage that is proportional to the RF power A

rectifier diode can function at zero bias (which is very good for reducing noise) at very

small DC bias (003mA) and relatively high RF impedance which will produce around

600Ω This will affect the capacitance value and a low capacitance is needed to realise a

high detection sensitivity

Figure 62 The detection process of a single wave through a non-linear IV characteristic of a diode

Detector

Speakerdisplay unit Amplifier

Antenna

Vout

Pin

Tuner Amplifier

178

However this type of detector has a drawback which is itrsquos relatively low signal to

noise ratio Thus it will also rectify any incoming electrical noise at all frequencies and

up to the cut-off frequency (fC) The basics lumped components circuit as shown in

Figure 63 is used to build such detector which consists of Source impedance (Zo)

Rectifier diode (ASPAT or Schottky) wire or pad inductance and capacitance and Load

impedance (RL)

Figure 63 Lumped element illustration of microwave detector circuit

Another type of detector is the heterodyne method which mixes incoming RF or

microwave signal (fRF) with another constant signal produced by a secondary circuit

called the Local Oscillator (LO) The LO frequency (fLO) must be slightly higher than fRF

to enhance the RF signal This mixing between fRF and fLO happens in the nonlinear

device as depicted in the Figure 64 This will produce a signal at a different frequency

called the intermediate frequency (fIF) which can then be amplified and detected as

explained in the previous paragraph Theoretically a basic requirement of the mixer is to

have fIF as efficient as possible while practically the minimum conversion efficiency

obtained is around 20 The main reason for using a mixer is due to the fact that

selective amplifiers at RF frequencies are costly and hard to achieve Hence a mixer is a

good technique as it only convert the signal to a lower frequency in which good

selectivity and high gain can be more effortlessly realised[14] A good mixer diode is the

one that can produce a high cut-off frequency and reduce conversion losses (Lc) A

mixer and detector diode with a low driven input power result in reducing overall noise

figure and thus in the ideal case the fIF amplifier also should have a low noise figure for

better performance The advantage of the heterodyne method is that it has a higher

179

sensitivity compared to the direct detection method this is achieved by producing an fIF

which has a lower frequency than the incoming RF signal[14] Obviously a zero-bias

voltage diode is more favourable to be used in mixer and detector applications

Figure 64 The mixing process where the signals are processed by the non-linear I-V characteristic

to produce an intermediate frequency (fIF=fLO-fRF) when signals at fLO and fRF are applied to the

diode

To have good detection efficiency for both types of detectors the operating

frequency (fO) must be several times smaller than fC In the case of an incoming

maximum modulated signal fM the frequency that can be acquired is in the range of fO

plusmnfM and will normally come with noise The standard method that is used to reduce the

noise is using a filter of bandwidth about 2fM at the centre of fO with the condition that fO

must be smaller than fM (f0lefM) Otherwise it would be difficult to attain However in

most cases fM is smaller than fO and fO is smaller than fC thus this will make the video

impedance (RV) (or nonlinear impedance) very close to the differential resistance of the

diode (at fO) in the equivalent circuit[74]

63 Definition of Characteristics of Interest Detection Parameters in small signal

analysis

Theoretically a transfer function measurement is preferable prior to any empirical

modelling since no assumption can be made due to the detector non-linearity

Furthermore measuring voltage output at high frequency can be very low while

measuring the power incident on the detector is hard to achieve where the linearity of

180

the typical power meter is normally less than 3 over its operating range[126]

Therefore the modelling of detector output voltage vs input power (ie transfer

function) can help to determine both nonlinearity correction and appropriate operating

range for the detector itself

The performance of the diode that is often taken into account is the transfer function

(output voltage Vout versus incident power Pin) and the main parameters that is used to

characterise and determine the quality of any detector diode are the voltage sensitivity

(βV) tangential sensitivity (TSS) dynamic range (P1dB) ie under 1 dB roll off power and

variation of output voltage when examined in extreme temperature situation (ΔV(T))

The voltage sensitivity in small signal analysis can use the approach introduced by

Torrey and Whitmer [9] then βV can be expressed as

120573119881(119894119889119890119886119897) =

119877119895119877119871120581

2(119877119881 + 119877119871) (1 +119877119904

119877119895)

2

[1 +1205962119862119895

2119877119878119877119895

1 + 119877119904 119877119895frasl]

(61)

Where ω is the angular frequency (2πf) Cj is the junction capacitance of the diode

active region RL is the load resistance RV is the video impedance taken from the

expression of Rj + Rs and κ is the curvature efficiency that give small signal rectifying

action of the ASPAT diode which is given by the second order term and itrsquos expression

is

κ =

11988921198681198891198812frasl

119889119868119889119881frasl

(62)

The curvature coefficient or responsivity (κ) is translated directly from the non-linearity

of the IV characteristics of the ASPAT diode for detector application Both RV and κ are

the parameters that can be extracted directly from diode DC measurement as discussed

in Chapter 5

The voltage sensitivity is actually a quantitative relationship between input

power and detector response Meaning that it is a change in signal output over change in

input power Normally output power is measured in Volts and input signal is measured

in Watts Therefore the unit of responsivity is VW[127 128]

181

TSS is referred to the lowest or minimum signal that the detector could detect it is

determined by the diodersquos βV and total noise available in the system (from the diode and

any amplifiers in the detector circuit) For any diode with fO=10 GHz and low noise of

1MHz bandwidth amplifier the TSS is typically less than -55 dBm[74] The TSS equation

is given by

119879119878119878 =

radic[4119884119870119861119879119861(119865119886 + 119905 minus 1)]

119872

(63)

Where M is a figure of merits and is derived based on the expression M= 120573119881 radic119877119881frasl t

denotes the diode noise temperature Fa is the noise figure B T and Y are the amplifierrsquos

bandwidth temperature and power for signal-to-noise ratio respectively For a low-level

video detector ie lt10GHz the sensitivity mainly depends upon three factors firstly on

RF matching structure secondly on the rectification efficiency output impedance and

noise properties of the diode and finally the input impedance bandwidth and noise

properties of the video amplifier at the detector output The RF matching structure

controls the quantity of overall energy at the active junction for rectification The second

factor controls the reaction of the diode to incident microwave radiation and the last

factor will influence the detector sensitivity in general[109]

In practical the Tss is a direct measure of the signal-to-noise ratio of a detector and

is achieved by varying the amplitude of the input pulse (RF signal) until a point in which

the top of the noise level with no signal applied is at the same level of noise at the

bottom level of RF signal It is commonly measured on an oscilloscope as depicted in

Figure 65 below It is defined as the input power at which a signal to noise ratio of 251

is produced[109]

Figure 65 Measurement of Tangential Sensitivity[108 129]

182

The transfer function in many detector diodes is often divided into three sections

Firstly at low incident power secondly at higher input power and finally at very high

power static (continuous) In the first region the detector diode performs as a square-law

detector in which Vout is proportional to Pin This region normally is used to extract the

dynamic range of the diode detector In the second region Vout is approximately

proportional to Vin and this region is known as the linear regime Finally at higher Pin

still the transfer function or response rolls off and thus Vout ultimately become saturated

This roll-off point where Vout has dropped by 1dB below an extrapolation of the

dependence at low Pin is termed the ldquo1dB roll-off pointrdquo and this value is usually in the

range of -11 to 12 dBm[15] Therefore a dynamic range of the detector diode can be

obtained by taking the interval between TSS and 1dB roll-off point (in dBm)

Finally the temperature dependence of Vout for a detector is normally taken from

two extreme points of the temperature (-40C˚ to +80C˚) and thus can be determined

from

Δ119881(119879) = 10 11989711990011989210 |

119881(1198791)

119881(1198792)|

(64)

This Vout variation between -40C˚ and +80C˚ normally expressed in dB

64 Noise Consideration in a Detector diode

The existence of noise in a system limits the accuracy of device performance and

the precision of measurements In a detector system specifically using a diode the noise

which can reduce the sensitivity of signal encryption is called the Noise Equivalent

Power (NEP) By definition the NEP is a noise power density over the detection

sensitivity and it can be exploited to determine the overall noise performance of a

detector[130 131] In other words NEP is defined as the power from the input source

(Pin) that is required to supply a voltage output (Vout) equal to the root means square

noise at Vout [132] For an ideal lossless match and assuming only Johnson-Nyquist is

present the NEP of a zero-bias detector can be expressed as

119873119864119875119900119901119905 = radic4119896119879119861119877119881120573119900119901119905 (65)

183

Where βopt is responsivity with an optimum match which is given by1 2frasl 119877119895120581 This type

of noise appears when changing voltages across a diode and a noise voltage (Vn)

normally will arise Theoretically the NEP has units of Watts (as it is actually a power)

but it often normalized to 1Hz as it is independent of bandwidth and thus the unit

becomes WHz12

Additionally there are also several noise sources that contribute to Vn in a

semiconductor diode which are Johnson-Nyquist noise Flicker Noise and shot noise

Johnson-Nyquist noise [133 134] appears across any conductor or semiconductor at

thermal equilibrium this is due to the thermal agitation of the carriers or charges It can

be expressed in root mean square voltage as below

119881119869minus119873 = radic4119896119879119861119877119895

(66)

Where Rj is the differential intrinsic resistance B denotes the post-detection bandwidth

T is the device temperature and k is the Boltzmann constant[134]

The second noise that is taken into consideration when dealing with semiconductor

devices is Flicker noise more commonly known as 1f noise It is a group of known and

unknown noise sources that can be observed in the frequency spectrum and normally

display an opposite to the frequency power density curve[135] It comes from a variety

of different causes ie recombination effects at a defect in semiconductor mobility

fluctuation and flow of direct current as well as interface phenomena [136-138] In term

of voltage source Flicker noise can be expressed as[139]

1198811119891 = 119870119891119881119909119891119910 (67)

Where Kf denotes a device-specific constant V is the voltage and f is the frequency The

value of x and y typically used are 2 and -1 respectively This type of noise (1f noise)

will be neglected at frequencies high enough due to the fact that the NEP of the diode is

proportional to the thermal noise and resistance of the diode

Finally the noise that causes time-dependent fluctuation in a flow of electrical

current because of the carrier or electron charge crossing a potential barrier is called shot

noise Shot noise is due to the randomness in the diffusion and recombination of both

majority and minority carriers[140] The equation of shot noise term at random time is

given by[141]

184

119881119878ℎ119900119905 = 2119902119868119861 (68)

Where q is the electron charge I is the current and B is the bandwidth This type of noise

is not affected by changes in temperature or device parameters Therefore the total noise

voltage appearing in the semiconductor is found to be [141]

1198811198992 = 119881119869minus119873

2 + 11988111198912 + 119881119904ℎ119900119905

2 (69)

However a zero-bias device will greatly eliminate both shot and flicker noise

compared to a biased device This has been explained by Equation (66) and Equation

(67) above where both noises are significantly related to the current and voltage Thus

if V and I =0 in the nonappearance of incident power then Vn will also become zero For

a detection process with bias the diode will be self-biased by ΔV which causes both

flicker and shot noise to appear But the shot noise in practical situation is much smaller

and thus normally ignored [142-144] Usually the noise in a zero-bias detector is

estimated by considering the low power limit as good first order estimation in which the

presence of only Johnson-Nyquist noise and ΔV is arbitrarily low [11 145 146]

Additionally it has been reported that the noise in tunnelling type diodes displays very

low or no excess noise in the bias region of the current-voltage characteristic[147 148]

Therefore in general most of the noise specifically in tunnelling type of diode will be

neglected this is a great advantage compared to SBD or transistors

65 Modelling of a 100GHz Zero-biased ASPAT Detector

Once the DC and RF characteristics of the ASPAT diode had been accurately

obtained the next step is to model and design a detector circuit based on S-parameter

measurement results as was explained in the previous chapter The aim is to realise a

detector circuit design which can be operated at millimetre and sub-millimetre wave

regions from an accurate diode model prior to the circuit design A diode detector model

puts experimental observations into context and offers insight into future experiment

results Consequently an electrical model based on lumped element component is vital

for a deeper understanding of how and to what extent a new device like the ASPAT

diode can affect all the key detector parameters that were previously discussed The

prediction of the detector parameters mostly depends upon the ASPAT diode

185

geometrical emitter size and material parameters However for millimetre wave

operating frequency the accuracy of the model is more sensitive not only to diode size

but also to the diode periphery ie substrate as well as coplanar amp transmission line

adopted in the circuit Therefore both extracted intrinsic and parasitic element of such a

device must be taken into account

In this work an ASPAT diode with an emitter size of 4times4microm2 which is the

smallest size that could be fabricated so far was chosen to be exploited for detector

designs The important parameters related to the 4times4microm2 GaAsAlAs ASPAT which

works at 0V is summarize in table 61 below

Table 61 A summary of all the important parameters of the 4x4 microm2 diode

Device Rj(Ω) Rs(Ω) Cj(fF) Cp(fF) κ(V-1) Intrinsic_fcut-off(GHz)

4times4microm2 90K 11 21 15 23 629

The actual measured I-V characteristic is used to model the diode since the library

in the ADS simulation tool does not have an ASPAT diode model or any tunnelling

diode for that matter The procedure of realizing the diode model is by taking the I-V

characteristic obtained from the 4times4microm2 emitter size measurement results and

converting it into a10th

orders polynomial equation via MATLAB software to create a

virtual I-V characteristic Thereafter this equation is then defined as a two terminals

device namely Symbolically-defined Device (SDD1P) ie a component of the non-

linear equation provided by ADS (Figure 66) to represent the ASPAT Figure 67

shows the measured data and 10th

orders polynomial equation fit very well to each other

Hence this new component used to represent the whole ASPAT diode will be used in

this research for MMIC detector and Frequency Multiplier designs The device chosen to

be modelled (4times4microm2) has measured junction capacitance of 21fF (at 0V at 40GHz) The

detector circuit is designed to operate at 100GHz for a safe side due to the extrinsic

calculated fcut-off is around 380GHz Since there are a lot of advantages in using unbiased

detectors compared to biased one this work will discuss the performance of millimetre-

wave detector at zero-bias and their result will be compared to the current performance

of other diodes reported in the literature

186

Figure 66 The configuration circuit to verify 10th order polynomial equation extracted from

MATLAB to realize a virtual GaAsAlAs ASPAT diode

Figure 67 Verification of actual (blue measured) and virtual (red_10th order polynomial) I-V

characteristic of the 4times4 microm2 diode used in this study

To realize the ASPAT detector circuit a simple detector circuit topology as

depicted in Figure 68 was constructed Initial simulation was run to perform a

functionality check of the detector circuit utilizing the Harmonic Balance (HB)

simulation tool embedded in that particular software Such simulation tools will analyse

the detector performances in the frequency-domain as it is mostly beneficial and fully

compatible with microwave and millimetre wave problems The frequency domain is

also suitable for single and multi-tone power excitation The importance of harmonic

balance are described in [149]

-00005

0

00005

0001

00015

0002

00025

0003

-3 -2 -1 0 1 2 3

Cu

rre

nt

(A)

Voltage (V)

ADS

4x4

187

Figure 68 Direct detector circuit topology using an ASPAT diode

Initially the circuit topology that consists of P1_Tone power supply ASPAT

bypass capacitor and load resistance is simulated by setting up a fixed input frequency at

100GHz The ASPAT diode provides a DC output voltage proportional to the input

power strength depending on the absolute values of the DC terms associated with the

nonlinearity of the I-V characteristics The capacitance in the output part is a bypass

capacitor used to prevent millimetre-waves from leaking to the output The load

resistance is large enough to ensure the voltage divider between load impedance and

device impedance gives maximum voltage sensitivity This large load resistance is

achieved by creating an open circuit at the end of detector circuit terminals

Noted that this simulation was run using diode parameters that were extracted from on-

wafer one port S-Parameter measurements as described in chapter 5 To apply them in a

two port application ie detector circuit may or may not provide a very accurate

outcome it however worked adequately in the particular circuit described in this chapter

but may not work in other circuits in general Thus the one port extractions in this work

still provide adequate parameters to build and design specific MMIC detector circuits

but not in general applications

The main reason for these simulations and their results to be used in high frequency

applications is due to the fact that actual RF measurement were done up to 40GHz

Additionally the 100GHz operating frequency was obtained from extrapolation of each

ASPATrsquos component Since the on-wafer measurement that were carried out were

limited to one port characterization applying them to two port network applications may

188

have extra consequences which are unknown Therefore actual MMIC detectors are

needed to be built and test to validate this work

To find out what power the 4times4microm2 ASPAT diode can withstand the input power

is varied from -40dBm to 10dBm via control by the P1_tone As can be seen in Figure

69 the diode starts to saturate when the received input power is about -8 dBm Above

this power limit both output voltage and sensitivity drop dramatically

Figure 69 Output voltage and detector sensitivity over wide range of input power

This diode detector circuit can thus operate adequately at given input powers from -30

dBm to -8 dBm with a sensitivity of 950VW However for the best possible sensitivity

over a range of input frequencies only one optimized input power needs to be chosen

The parameters that directly influence the voltage sensitivity are the curvature

coefficients load resistance and video resistance as can be seen from Equation (61)

Therefore in the following simulation the values RL will be optimized according to the

diodes optimum input power with regards to the highest possible voltage sensitivity

Consequently five values of load resistance were chosen from few ohms to infinity

and with the same applied input power as depicted in Figure 610 For most load

resistors the sensitivity is constant at low input power and drop at the diode saturation

region (-8dBm and greater) However for an RL value of 100KΩ and below the loaded

voltage sensitivity shows a peak near 0dBm input power which corresponds to the

maximum slope of the ASPAT detector transfer function The highest voltage sensitivity

is obtained by using an Open circuit load impedance as shown in the graph The Open

circuit load impedance gives the highest voltage sensitivity due to the voltage divider

189

between source and load impedance therefore RL must be at least 5-10 times larger than

Rj to give a better sensitivity

Figure 610 Simulated dependence of Voltage Sensitivity on incident power and load resistance of

the ASPAT detector

Furthermore the value of voltage sensitivity (βV) depends on the junction

resistance of the diode and thus the large value of Rj of the ASPAT diode yields the high

βV observed Rj which was taken from the non-linear measured IV characteristic is a

voltage dependent parameter [142]and is inversely proportional to the forward bias

voltage as depicted in Figure 611 below

Figure 611 Junction resistance as a function of forward voltage

1

10

100

1000

-40 -30 -20 -10 0 10

Vo

lta

ge

Sen

siti

vit

y (

VW

)

Incident Power Pin (dBm)

Infinity

1KΩ

10KΩ

1MΩ

100Ω

100KΩ

0

20

40

60

80

100

0 001 002 003 004 005 006 007 008 009 01

Ju

nct

ion

Res

ista

nce

(k

Ω)

Bias Voltage (V)

190

In fact the expression of (120597119881120597119868frasl ) contribute to the video impedance expression via the

expression RV=Rj+RS (nonlinear resistance) As RS is very small compared to Rj thus it

was ignored when calculating RV Although the large value of Rj will increase the

voltage sensitivity it will also make matching difficult to achieve Therefore there will

be a compromise between the size of the matching circuit and voltage sensitivity to

attain the correct value of Rj Additionally a very high Rj (around ~1MΩ) will also

increase the detectorrsquos noise equivalent power (NEP in Eq(65)) Thus an average value

of RJ typically around 100kΩ is satisfactory[150] By having a large value of RJ one can

benefit from a low input power to drive the diode into the non-linear region and thus the

detector can work at very low RF input power The NEP for GaAsAlAs ASPAT diode

is then calculated based on parameters obtained from this simulation at room

temperature The values are compared to other diode detector available in the literature

as shown in Table 62 below

Table 62 Noise Equivalent Power (NEP) for zero bias detector diode

Device NEP (pWHz12) RJ(KΩ) Frequency(GHz)

ASPAT (4times4 microm2) 188 92 100

Tunnel Diode (08times08 microm2)[150] 370 26 220-330

Zero bias SBD[11] 15 3 150

Sb-Heterojunction Backward Diode[145] 024 32 94

VDI Zero bias SBD[151] 2 18 110

From the Table 62 above the NEP of the ASPAT is calculated based on RF input power

which is required to obtain an output signal-to-noise ratio of unity in a 1Hz at detector

output[142] and also the assumption of only Johnson-Nyquist (thermal noise) is

dominant for small incident power (-25dBm)[152 153] The prediction of NEP for the

ASPAT is comparable to the VDI Zero bias detector since the value of their junction

resistance is much lower than that of the ASPAT Therefore it is very important to

obtain a reasonable value of junction resistance From this it is clear that a trade-off of

high voltage sensitivity and low junction resistance is best to obtain low noise

191

Finally the curvature coefficient at specific operating voltages also influences

the voltage sensitivity of the detector Figure 612 shows the calculated curvature

coefficient of the 4times4microm2 ASPAT diode used in this work

Figure 612 Curvature coefficient for the measured I-V characteristic with a device size of 4times4μm2

The high zero-bias curvature (23V) is reached from the mutual effect of the intra-band

tunnelling in the GaAs-AlAs-GaAs and the highly doped GaAs at the anode and cathode

(Rs amp Rj) This was shown in the numerical simulation in[154] where the combined

effects of the optimum anode AlAs composition increases the curvature coefficients by

thinning the energy tunnelling window (intraband tunnelling process)

Other approaches that can lead to a large curvature is using smaller device area

with minimum series resistance [155] as was also discussed in Section 52 Having a

better curvature coefficient leads to increased voltage sensitivity as seen from Equation

(61) above and βV is also proportional to κ [111] The curvature coefficient calculated

using the above formula is nearly 23V at 1mV peak and high voltage sensitivity can be

achieved by having large value of curvature coefficient But one needs to remember that

this will also decrease with increasing input power because RJ which will also decrease

Therefore for a safe operating region the incident power that can be applied through the

diode is in between -30dBm to -8dBm for a 100GHz operating frequency

In this simulation -25dBm is chosen to simulate the GaAsAlAs ASPTAT diode

detector working at 100GHz input frequency By fixing the P1_Tone to this input power

-5

0

5

10

15

20

25

30

-001 001 003 005 007 009

Cu

rva

ture

Co

effi

cien

t(V

-1)

Voltage (V)

k(4x4 um^2)

192

the frequency is varied from 90GHz to 110GHz and the results are depicted in Figure

613 below

Figure 613 Voltage Sensitivity against frequency at -20dBm fixed input power

Noticeably the voltage sensitivity of the diode detector decreases linearly with

increasing input frequency At 100GHz a sensitivity of around 540VW is obtained

However the sensitivity in this case is roughly estimated from the Equation (61) above

and will not be sufficiently accurate because of the effect of other key factors such as

reflective power which was not included This parameter must be taken into the account

due the P_1Tone power source which provides a 50 Ω impedance source (Zin) which

does not match the load impedance (ZL) which consists of JωL 1JωC and Z and which

mainly comes from the ASPAT diode itself In order to determine the load impedance at

the diode a typical ohmrsquos law (Z=VI) equation must be used at the input side of the

diode with regards to the applied frequency (100GHz)

As a result the total load impedance obtained from the simulation is 11055-

j69057Ω which is clearly not matched to the 50Ω impedance source The mismatch

between source and load leads the available power from the source to be not fully

delivered to the load and hence there is loss of power leading to a lower detector

sensitivity Therefore actual calculation must take into the account the reflection impact

as expressed in the equation below

193

120573119881(119886119888119905119906119886119897) =

119877119869119877119871120581(1 minus |Γ|2)

2(119877119881 + 119877119871) (1 +119877119904

119877119895)2

[1 +1205962119862119895

2119877119878119877119895

1 + 119877119904 119877119895frasl]

(610)

Where the term (1-|Г|2) refers to the normalized power absorption by the ASPAT diode

and Г is the reflection coefficient due to discrepancy between 50Ω input impedance (Zin)

of the input port and the diode Consequently the calculation of reflection coefficient

(eq611) is carried out by using this expression and the result is shown in Figure 614

Γ =

119885119871 minus 119885119894119899

119885119871 + 119885119894119899

(611)

Figure 614 Reflection Coefficient versus operating frequency without matching circuitry

As can be seen in Figure 614 without the matching circuit the S11 is decrease

linearly with frequency but only very slowly This means that most of the RF power

transmitted from the source is reflected back (by that ratio) when it went through the

diode Therefore in order to resolve the mismatch a matching circuit is introduced in

between the source and load of the detector circuit as shown in Figure 615 (red

rectangular) This matching circuit works by transforming the load impedance into an

impedance that is identical to the source or input impedance Note that for any

impedance matching circuit the main purpose is usually to obtain maximum power

transfer to the load however in some cases (ie oscillators) the matching circuit is to

achieve a lower noise figure Hence in a broad sense the introduction of the matching

194

circuit in a detector circuit can be defined as a circuit that convert available impedance

into wanted impedance by obeying the maximum power transfer theorem [156]

Figure 615 Detector circuit with impedance matching circuit placed in between diode and source

There are many type of matching circuit that can be used to achieve both

objectives above such as circuits using lumped element transmission-line-impedance

matching circuit single and double-stub tuners as well as a quarter-wavelength In this

design the technique used to match source and load impedance is the single open and

short stub (in red rectangular) as it is simple convenient and very efficient in ADS

simulation The important parameter that needs to be tuned in both stubs is the electrical

length (E) at any designed frequency ie 100GHz in this case The E tuning is realized

by using the Smith Chart features available in the ADS simulation tools Figure 616

shows the reflection coefficient with matching circuit modelled over a broad frequency

band and it is clearly shown that at the desired operating frequency the reflection is very

low Note that it is difficult to obtain wide frequency band matching

Matching circuit

195

Figure 616 Reflection Coefficient over wide frequency band with matching

The simulation is continued to find the effect of the matching circuit placed in

the detector circuit on the voltage sensitivity The same input power (-25dBm) is applied

to the diode and the voltage sensitivity is plotted against frequency as depicted in Figure

617 Obviously at the desired operating frequency (100GHz) the sensitivity rises up to

a maximum value of 2100VW with this value obtained without any reflection and the

input port being completely matched with the ASPAT diode model used in this work

Figure 617 Simulated voltage sensitivity of GaAsAlAs ASPAT detector at W-band frequency

Once the matching circuit to be used for 100GHz operating frequency was confirmed

further simulations were made by applying a series of low input power to find the

tangential sensitivity (Tss) of the detector diode When determining the Tss it is very

important to include the matching circuit as it will minimize any power losses through

the diode thus a very small input power can be detected

196

Figure 618 Lowest detectable signal at 100GHz operating frequency

The transfer function depicted in Figure 618 shows incident power of -80 dBm

to -50 dBm applied and the lowest detectable signal that can be obtained with 4times4microm2

mesa size ASPAT diode is around 138microV at -68dBm Although a typical value of Tss is

normally not more than ~-55dBm as in ref [74] the lower value obtained in the

simulation is because the device is operated at zero bias operation and does not use any

amplifier therefore the noise and values related to amplifier as in Equation 63 have been

neglected Even though the TSS appeared very low it is most likely very dependant to

the 10th

order polynomial equation embedded into SDD in ADS software Therefore in

near future the TSS value has to be determined in real fabricated MMIC ASPAT detector

As discusses in Section 63 other important parameter that can be extracted from

the diode transfer function is the Dynamic Range of the diode It can be obtained in a

region called square law region which is a region where the Vout of the ASPAT diode is

proportional to the square law of the input power signal From Figure 619 the square

law region is in between -68dBm and -12dBm and the linear region or in this case

saturated region is above -12dBm Taking to the account the roll-off point where Vout

has dropped 1dB below the extrapolation of the dependence at low input power and

therefore the dynamic range of the detector diode can be obtained by taking the interval

between TSS and the 1dB roll-off point (in dBm) which is ~55dBm

197

Figure 619 The dynamic range of the ASPAT diode obtained from linear region of diode operation

The figure of Merit (M) of the detector (ie equation 63) is 2100 (90K) 12

where 2100 is the sensitivity and 90K is the value of RJ at zero bias which is equivalent

to 652 W The M value should be large however in this case due to RJ being very

large it has dropped tremendously when compared to the voltage sensitivity obtained in

this simulation The results obtained indicate a reasonably successful design of the

MMIC detector using the 4times4microm2

emitter size GaAsAlAs ASPAT diode The results

obtained lead to the design of other MMIC detector using the other fabricated diode

sizes (6times6microm2 and 10times10microm

2) A Similar procedure to the one used for the 4times4microm

2

diode was followed The only difference was the use a slight higher input power (-

20dBm) than in the 4times4microm2 design Hence the simulation results obtained are then

compiled and compared as depicted in Figure 620

Saturated

Region

198

Figure 620 Simulation result of diode sensitivity for three devices geometries obtained from the

fabricated ASPAT in this work

The graph plotted for each dimension was taken after matching circuits were included

As can be seen in Figure 620 the highest sensitivity is achieved using the smallest

device size as this has the highest cut off frequency Table 63 below summarises the

performances of the 100GHz ASPAT detectors obtained from the simulation in this

work

Table 63 Summary of characteristic of interest for GaAsAlAs ASPAT detector

Device Tss(dBm) fcut-

off(GHz)

Bv

(VW)

Dynamic range dB(dBm) M(W-12

)

4times4microm2 -68 380 2100 55 65

6times6microm2 -50 151 1445 48 45

10times10microm2 -40 98 247 40 21

From table 63 above it is clear that a lower cut-off frequency will affect the

voltage sensitivity The dynamic range between each ASPAT is different because larger

size area will allow more power to go through the diode as a result of high current that

such a device can handle before reaching the saturation their lowest detectable is higher

Small diode size will detect lowest voltage but cannot handle high power On the other

hand a large diode size is able to receive high power however can only offer lower

voltage output Therefore there is a trade-off between small diode size and receiving

input power which will directly affect Tss and 1dB roll off Once again all the

100

1000

10000

90 95 100 105 110

Vo

lta

ge

Sen

siti

vit

y (

VW

)

Frequency (GHz)

10times10microm^2

6times6microm^2

4times4microm^2

199

parameters obtained in this simulation are just estimation from the 10th

order polynomial

equation thus real ASPAT detector has to be fabricated for verification

The best device performance among all GaAsAlAs ASPAT diodes was obtained

with the 4times4microm2 mesa area size diode which was compared to other exiting millimetre

wave detector diode available in the literature Since the 100GHz is located in the W-

band spectrum frequency therefore the comparison will be performed in this frequency

band but with low input power The parameters for the-state-of-the-art zero bias

detectors are gathered in Table 64 below

Table 64 Comparison 4times4microm2 GaAsAlAs ASPAT diode to the-state-of-the-art zero bias detector

at W-band (75GHz-110GHz)

Device Size(microm2) Tss(dBm) βv (VW) Pin (dBm) f (GHz)

GaAsAlAs ASPAT 4times4 -68 2100 -25 100

GaAs SBD HSCH-

9161[157]

- -49 2200 94

HBD[158] 15x15 2540 -20 95

Planar SBD[159] - -68 2100 -25 100

Note that the ASPAT diode retains its favourable temperature stable characteristics

which are not the case for all the diodes used for comparison in Table 64

66 Conclusions

In this chapter all theory regarding RF detection using diodes ie parameters of

interest noise consideration etc have been discussed The aim to design and develop a

low cost reliable and sophisticated zero bias 100GHz detector circuit was achieved

through exploitation of a 4times4microm2 GaAsAlAs ASPAT diode The design was performed

with the aid of Keysight ADS modelling software utilizing harmonic balance simulation

The effect of load resistance junction resistor to the detector voltage sensitivity was also

discussed in details The 90KΩ Rj value and open circuit load resistance was chosen for

high sensitivity

200

A step by step design of a W-band ASPAT detector was presented The effect of

matching circuit was discussed in detail and where an unmatched sensitivity of 843VW

is obtained which then increases to 2100VW after matching Through RF

characterization simulation a detection at 100GHz (W-band) was successfully achieved

with a relatively large device mesa area (4times4microm2) at an input power of -25dBm

(8microWatt) leading to a 2100VW voltage sensitivity a -68dBm TSS and 55dBm dynamic

range All these values are comparable to others fabricated diodes in the literature

The zero bias ASPAT detectors based on the GaAsAlAs material system in this

work are still at an early stage of development a lot of work is still required to realize

high yielding integrated millimeter and sub-millimeter wave (MMIC) detector circuits

However as this work is on-going at Manchester it is expected that fabricated ASPAT

MMICs with even higher voltage sensitivity will be fabricated in the near future through

collaborating bodies involved in this research especially the University of Cambridge

and ICS Limited

201

7 FREQUENCY MULTIPLIERS (DOUBLER) DESIGN USING

GAASALAS ASPAT DIODES

71 Introduction

Originally the key application for the ASPAT diode was for use as microwave and

millimetre wave detectors[18] This is due to the fact that such diode demonstrates

strong non-linearity low noise and high cut-off frequency features as described in the

previous chapters However these features are not only beneficial for detection purposes

but also allow them to be used and designed as microwave and millimetre wave sources

The only way to generateenhance continuous wave (CW) power using a non-linear

device is through frequency multiplication techniques It is known that the frequency

multiplier is the alternative approach (to 3 terminal transistors) using non-linear devices

that are used to generate high frequency low phase noise signals Any high quality low

frequency signal that goes through a frequency multiplier circuit can be generated to any

desired high output frequency[160] Therefore the main objective of this chapter is to

demonstrate the feasibility of the ASPAT diode as a compact source of microwave and

millimeter-wave receiver for imaging applications[161]

The study of the ASPAT diode as a power source begins with a brief explanation

of the importance of a frequency source and the lack of compact device and technologies

at high-frequency signals The state-of-the-art for frequency multiplier will also be

discussed In the next section (Section 74) the fundamentals of the frequency multiplier

architecture ie the principle of operation and appropriate devices will be presented

Since this is the first attempt at using GaAsAlAs ASPAT diodes a simple multiplier

circuit design and topology was built This will be discussed in detail in the subsequent

sections where simulation results are discussed The focus of the discussion will be to

demonstrate the possibility of a GaAsAlAs ASPAT diode functioning as a frequency

multiplier and comparison with other state-of-the-art varistor mode frequency

multipliers

202

72 Motivation and Background

Typically continuous wave (CW) sources generating below 100 GHz can be

obtained through oscillators amplifiers and pin diode comb generator Below 10THz

the sources can be made from RTD IMPATT diodes and Gunn oscillators and above 10

THz it is commonly done by photonic mixing quantum cascade laser (QCL) and gas

lasers [162] Both types of sources and their performance are plotted in Figure 71

However these conventional ways of generating millimetre and sub-millimetre waves

have their own limitations ie high cost complexity and sometimes requirement for

cryonic cooling The most effective way to tackle the limitations of conventional mm-

wave and THz sources is by implementing frequency multiplication technique using

solid state nonlinear diodes [163-165] such as SBD and ASPAT diodes

Figure 71 performance of state-of the-art millimetre wave source [166]

Twenty years ago there were only two types of diodes (SBD and P-N junction

diodes) often used for frequency multiplication To date besides the SBD there are

many types of diode that have been used as frequency multipliers These include the

high electron mobility varactor (HEMV) single barrier varactor (SBV) [167] and hetero-

structure barrier varactor (HBV) (270GHz with 90mW input power and Conversion

Efficiency of 72)[168 169] Other variants that have developed to enhance the

frequency multiplier performance of the classic SBD [170] include the Barrier-intrinsic-

203

n+ (BIN)diode and Barrier N-layer N+ (BNN) diode [171] Other diodes for use in such

applications are the planar doped barrier diode (PDB) Resonant tunnelling diode (RTD)

amp it families ie Quantum well diode (QWD) and step recovery diode which is a

modification of the P-N junction diode

Although three terminal devices ie FET GaAs MESFET and HEMT had

shown better performance and are capable of achieving greater efficiency and

bandwidth as well as having additional conversion gain features [160] two terminal

devices (ie varactor diodes) which are passive multiplier are still preferred This is due

to their simplicity and most importantly their ability to generate very little noise Among

these types of device technologies the SBD is preferable as it is mature and has been

shown to be very suitable for high-frequency applications [172 173]

The ASPAT diode is exploited to investigate the possibility and the feasibility of

generating microwave and millimetre-wave power through well-known frequency

multiplication methods The utilisation of the ASPAT in frequency multiplication will

also aid in generating local oscillator sources which are critical components in

heterodyne receivers The ASPAT diode will work in resistive I-V mode (varistor mode)

and has features to work also at zero bias condition thus offering low power handling

than traditional high-efficiency varactor diode since the varactor diode requires a large

reverse bias supply of several tens of volts

73 Frequency Multiplier Architecture the Basics

In principle a frequency multiplier is an electronic circuit that gives an output

frequency that is a multiple integer of its input frequency signal pumped from a local

oscillator as depicted in Figure 72 The ability to generate any desired multiple output

signals is realised by a nonlinear device ie diode or transistor Such devices though

also can give distortion or cause sudden change to the input frequency Additionally

these devices generate multiples of the input frequency (fout) The distortion of the

sinusoidal signal refers to an abruptsudden change versus amplitude or time which thus

generates higher frequency with lower amplitudes of the input signal Usually a

frequency multiplier circuit will include a bandpass filter to select the desired harmonic

204

frequencies and deselect undesired harmonic frequencies especially fundamental

harmonic at the output for further processing

Any non-linear device either in symmetricalantisymmetric current-voltage or

capacitance-voltage can be utilised to realise a frequency multiplier source [168 174]

Figure 73 describes the method where a nonlinear resistance is utilised to convert a

harmonic input signal into periodic output signal containing components at multiples of

the input frequency Both non-linear resistance and reactance characteristics can be

extended into power series methods

Figure 73 Principle of operation for frequency multiplier utilising a non-linear resistance [10]

The operating principle of the frequency multiplier is shown in Figure 73 where

the I-V curve converts a harmonic frequency input into a periodic frequency output

including components at multiples of the input frequency The non-linear I-V

characteristic can be explained in term of a power series at the operating fixed point of

bias voltage (VB) [174]

Frequency

Multiplier Circuit

finput

foutput

= nfinput

Figure 72 Frequency multiplier has the property that foutput is an integer multiple of fin

205

119868(119881119861 + ∆119881) = 1198860 + 1198861∆119881 + 1198862∆1198812 + 1198863∆1198813 helliphellip (71)

For a given input voltage as below

∆119881 = 119881119904 cos120596119878119905 (72)

The input signal harmonics will become

119868(119905) = 1198680 + 1198681 cos120596119904119905 + 1198682 cos 2120596119904119905 + 1198683 cos 3120596119904119905 (73)

Where t and ω are the time and angular frequency respectively based on equation (73)

the output contains both signal source and harmonics Therefore a complete frequency

multiplier circuit has to have non-linear device and filter to allow the selection of any

frequency components needed

731 Types of frequency multipliers

Frequency multipliers can be classified into passive and active multipliers This

classification is based on the ability of the frequency multiplier to yield any conversion

gainlosses The passive multiplier is the one that only produces conversion losses In

other words it can be described as a multiplier that generates an output power level

lower than the excitation input power and it is mostly dominated by passive nonlinear

devices ie Diodes On the other hand the active multipliers refer to a device that would

produce an output signal with a power level that is greater than the input signal power

This conversion of power is termed as conversion gain These types of multipliers attract

much attention as they do not only increase the frequency at the output but also the

signal power

Passive frequency multiplier can be formed by using diodes that are classified as

being of the varistor (non-linear I-V) or varactor (non-linear C-V) type [160 174] The

varistor type will influence the frequency multiplication with a non-linear resistance or

conductance (resistive diodes) and this results in a very large potential bandwidth at the

output but poor conversion efficiency The varactor diode type where the frequency

multiplications are affected by the non-linear capacitance (reactive diode) as their

reactive element typical result is high conversion efficiency A diode that is used in this

206

application must have strong nonlinearity stable electrical characteristic repeatable and

has fast enough response to an applied frequency Therefore multipliers are classified

into Doubler Tripler quintuple and so forth depending on the highest power of output

harmonic signal

In general all varactor type diodes with such characteristics will produce high

power at odd-order harmonic oscillation if any microwave signal is pumped into them

The benefit of having odd-order in multiplier design is that it reduces the complexity of

the overall circuit ie it eliminates even-order idler frequencies [175] The varactor type

diode had been shown by Manley-Rowe to a get maximum 100 conversion efficiency

for generating an ideal harmonic [176] compared to the varistor type where the

maximum efficiency achievable ideally is 1 1198992frasl where n is the multiplication factor

(output harmonics number) [174] In the case of power handling (input excitation power)

for multipliers varactor mode diode required greater power (several milli Watt) than the

varistor mode due to the fact that reverse applied voltages are very large (many tens of

volts) Therefore these types of frequency multipliers may not be suitable for the case of

high input power excitation There no report in the literature of varistor based

multipliers working with high power excitations

74 Parameters of interest for Frequency Multipliers

The simplest way to describe an equivalent circuit for a complete frequency

multiplier is by setting a Source impedance (ZS) at the input side and load impedance

(ZL) at the output side as depicted in Figure 74 below This circuit usually has the same

properties as described in the previous chapter and most of the others two ports

networks However in this case the purpose is different and is the conversion of a sine

wave signal source (Vs) with angular frequency ωs to an output signal with frequency

nωs where n is the multiplication integer or the order of multiplication

207

Figure 74 A standard system for two port frequency multiplier circuit

Referring to Figure 74 above there are few sets of parameters for the frequency

multiplier to be taken out and compared Examples are the conversion loss maximum

input signal power Impedance at source and load Bandwidth multiplication factor or

harmonic amp subharmonic content and noise conversion properties The conversion loss

(CL) is described by the ratio of available power at source (Ps) to the output harmonic

power delivered to load resistance (PL) and is normally expressed in dB It occurs due to

the nature of passive semiconductor diodes and the electronic circuit itself that are lossy

and dissipate energy On the other hand the conversion efficiency (ηn) is a ratio of the

output power at load (PL) to the available power at the input (Ps) This is often expressed

in percentage () The conversion efficiency can be determined as

120578 =

119875119878

119875119871

(74)

while the conversion loss is expressed as[174]

119871119899[119889119861] = 10 log

119875119904

119875119871= 10 log (

|1198811199042|

4 119877119890 |119885119904||1198681198712|119877119890|119885119871|

) (75)

Where Vs is the input voltage and IL is the output current amplitude Besides this the

conversion efficiency is often referred to as the inverted value of the Ln In designing a

frequency multiplier it is crucial to minimise the conversion loss and maximise the

conversion efficiency value

To achieve a perfect multiplier with minimum conversion loss the impedance of

source and load must be at an optimum level This implies that the source impedance

Frequency

Multiplier Z

L

Zout

Z

in

V

Z

SWR Г

208

(Zs) must be very close to the complex conjugate of the multiplier input impedance

(Zin) hence minimum reflection loss will occur at the input side This can be realised by

introducing an impedance matching circuit between the diode and source The power

transfer between the source and the multiplier is quantitatively described by the value of

the multiplier input reflection coefficient (Ѓ) with source Zs assumed to represent a

reference impedance This specification also can be explained in the standing wave ratio

(SWR) Both relationships are described below respectively [174]

Γ =

119885119894119899 minus 119885119904lowast

119885119894119899 + 119885119904

(76)

119878119882119877 =

1 + |Γ|

1 minus |Γ|

(77)

Where the asterisk () represents the complex conjugate of the Zs impedance

On the other hand the situation of the load impedance is different when a standard

or an optimum value is provided by the designer This will either increase the conversion

loss or decrease the output power Thus one has to keep in mind that frequency

multipliers are non-linear devices and power transfer condition both at the input and the

output depend on each other and the input signal level[174]

75 20GHz40GHz Varistor Mode ASPAT Diode Frequency Doubler

In this work a similar ASPAT diode (4times4microm2) to that in designing millimetre

wave detector in the previous chapter is used The main objective of designing the

frequency multiplier circuit was first to investigate the performance of the ASPAT

diode as a microwave or millimetre wave signal source A design is deemed successful

when the diode physical parameters are optimised and the suitable impedance matching

network is produced for each desired harmonic as well as maximising the output power

These goals however are hard to achieve when a higher frequency operation is targeted

for use

There are many types of multiplier circuit topologies that can be implemented

using GaAsAlAs ASPAT diode in varistor mode to achieve high order of multiplication

209

Examples are single diode multiplier series or parallel connected diode multiplier anti-

parallel amp anti-series connection diode pair multiplier anti-parallel-series connected

diode multiplier and bridge frequency multiplier as well as nonlinear transmission line

frequency multiplier [174] Before designing a circuit there is one most important

consideration to make Prior to choosing any mentioned circuits to be used for frequency

doubler the design considerations are made based on the capability of ASPAT diode to

receive an optimum amount of input excitation RF power From the discussions in

Chapter 6 the ASPAT diode will reach saturation level (linear regime) at power ~

-10dBm for a device size of 4x4microm2 Once the optimum input power was confirmed the

circuit topology was carefully chosen to balance between the requirements of the

ASPAT to work at high frequencies ie low Rs and Cj amp high diode cut-off frequency

as well as the desired output signal frequency that needs to be produced

To realise the first attempt of an ASPAT diode as a signal source a simple circuit

topology of a frequency doubler was deployed as depicted in Figure 75 below The

frequency doubler circuit consists of a voltage source (can be power source) input

filtering with matching network ASPAT diode output filtering with matching network

and load impedance (ZL)

Figure 75 Block diagram of frequency doubler with series ASPAT mounted diode

In order to investigate the doubler performance the Keysight ADS simulation

tool and similar procedure to obtain accurate ASPAT model using a 10th

order

polynomial equation as in Section 65 was used The circuit in Figure 75 is translated

into ADS format as illustrated in Figure 76 Once the circuit was constructed the

analysis was performed using the Harmonics Balance (HB) simulator The circuit

requirements are matched terminations at the input and output frequencies open

Input Filtering

and matching network

Output Filtering

and matching network

Zs

Vs

ZL

Pin

fin

Pout

nfout

210

circuited terminations at the higher harmonics and optimum reactive terminations (an

inductance which resonates with the junction capacitance) at the output frequencies

Figure 76 Empirical modelling of ASPAT doubler utilising Keysight ADS tool

The circuit in Figure 76 is the simplest way of constructing a frequency doubler

circuit which consists of the signal source (P_1Tone) input matching circuit (Stubs)

filter (Short stub) ASPAT diode low pass filter output matching circuit(Stubs) and ZL

(load resistance) The utilisation of the stubs is an ideal case of simulation since in the

real fabrication stubs are normally formed in large sizes Therefore a proper design such

as using CPW instead of stubs is essential in real fabrication

Again this simulation works for this particular circuit in this chapter as all the diode

parameters were extracted from on-wafer one port S-Parameter measurement described

in chapter 5 To apply them in such two ports applications may not very accurate

however it still provides adequate parameters to build and design particular frequency

multipliers but not in general applications These simulations and results are adequate for

high frequency applications due to the fact that the actual RF measurements on the

diodes were carried out up to 40GHz and the target operating frequency in this multiplier

design does not exceed 40GHz

Since the on-wafer measurement that were carried out were limited to one port

characterization applying them to two port network applications may have extra

consequences which are unknown Therefore actual MMIC frequency multiplier is

needed to be built and test to validate this work

Input Matching

Output Matching

211

To find the optimum output power initial simulation without matching circuit

was performed This simulation was run by varying the input power from -35dBm to

20dBm but fixing the input frequency at 20GHz As can be seen in Figure 77 the lowest

point in the conversion loss (CL) and the highest point of the conversion efficiency (CE)

are obtained from an input power of -1dBm However this amount of input power is too

high for the ASPAT diode The lowest CL at -1dBm may not be accurate since it was

applied without matching Note that it is difficult to achieve a matching between source

impedance and load impedance when varying the input power Therefore a lower input

power of -10dBm is chosen for this frequency doubler operation Figure 77 shows the

Conversion Loss and Conversion Efficiency as a function of the available power of the

given input source

Figure 77 Conversion loss and conversion efficiency as a function of input power

The circuit in Figure 76 works with a -10dBm input power and 20GHz centre

frequency input signal is pumped from the power source (P1_tone) to the ASPAT diode

and distortionabrupt change of input waveform occurs at the fundamental frequency (f0

in this case 20GHz) Such abrupt change produces harmonics and these harmonics can

be classified into desired frequency component by placing two-quarter wavelength (λ4)

stubs (90˚) at both sides of the ASPAT diode At the input side of the diode short circuit

stubs are utilised to permit the f0 tone to reach the ASPAT diode and block the second

harmonics (2f0 in this case 40GHz) back to the input side and pushes it towards the

load resistance On the contrary at the output side of the diode the open circuit stubs are

used to ldquoopen circuitrdquo the 2f0 signals while ldquoshort circuitrdquo the f0 component Thus 2f0

212

signal will not be affected due to the open circuit stubs being half wave (λ2) long The

function of both stubs is basically to isolate the input and output signal from mixing each

other Therefore the design of input and output matching circuit can be achieved easily

The input matching circuit was designed based on the mentioned input frequency

(f0=20GHz) for an available input power (Ps=-10dBm) which is set up at the power

source by using two stubs with the same configuration as used in Chapter 6 Such

configuration is purposely deployed to increase the 50Ω coming from the P1_tone

source impedance to the conjugate thus reducing the reflection coefficient to the

ASPAT diode From the simulation without matching the input impedance to the diode

is 551Ω in magnitude for an available input power of -10dBm

On the output side of the diode output matching circuit is available to transform the 50Ω

port impedance in the optimum load impedance which provides minimum conversion

loss for the ASPAT diode The output matching circuit is designed based on expected

output frequency which is in this case 2f0 =40GHz Other than this optimum

impedance between load impedance and ASPAT will not be achieved thus resulting in

higher conversion loss

To ensure the proposed circuit is valid and suitable for the specific ASPAT diode

mesa size the response of conversion loss and efficiency are plotted as a function of

output frequency from 20GHz to 100GHz The results of both conversions are illustrated

in Figure 78

Figure 78 Conversion loss and Conversion efficiency as a function of output frequency

213

As can be seen in Figure 78 the conversion loss is obtained at the lowest point where

the output frequency is needed Meanwhile the conversion efficiency is maximum at the

same output frequency Therefore this indicates that the first attempt of an ASPAT

Doubler frequency source works well However the values obtained for Ln from this

simulation is 28dB which is rather high On the contrary the η achieved in this study is

very low with a value less than 1

Since the ASPAT is in varistor mode with no bias applied the conversion

efficiency is expected to be low due to resistive losses Another factor that may

contribute to lower η is the diode model itself as it is taken from a 10th

order polynomial

equation not from a diode model provided in the ADS software tools Thus some

properties of such tunnelling diode may not be included Hence it is necessary in due

course to fabricate and build such a compact frequency doubler in the future to verify

the simulation results

From the simulation point of view the less than 1 Conversion Efficiency

obtained is still good enough for a first attempt at a frequency doubler which utilises the

new ASPAT tunnelling diode The frequency doubler obtained from this work is suitable

for use in zero bias varistor modes for low power application The varistor mode doubler

performances from this simulation work are gathered and compared to other in the

literature as summarized in Table 71 below where fout is 40GHz η is 015 Ln is 28dB

and Pi is -10dBm

Table 71 Comparison parameter obtained from this simulation to the state-of-the-art multiplier

diode

Device fout (GHz) η() Ln(dB) Pi (dBm)

ASPAT 40 015 28 -10

SBD (Si)[177] 104 2 134 -10

SBD (GaAs)[178] 13 2 137 -16

214

The performance of the 2040GHz is compared to the literature based on their input

power below -8dBm since the ASPAT is only capable of working at low power To the

best of the author knowledge very few diodes operating in varistor mode at low power

excitation can be found in the field of research and industry Therefore this 2040GHz

ASPAT Doubler might a first for tunnel diodes if it can be fabricated and test at Ka band

and above

76 Conclusions

In this chapter another alternative application based on non-linear features of

GaAsAlAs ASPAT has been presented The simulation of a frequency multiplier

(doubler) was carried out utilizing the 4times4microm2 size ASPAT diode The theory and the

ability of the ASPAT diode to operate as a frequency source were explained in detail

A unique varistor mode frequency multiplier circuit topology for the 2040GHz

ASPAT doubler has been demonstrated and briefly discussed The details and step by

step simulation technique utilizing harmonic balance from Keysight ADS has been

presented Even though the conversion efficiency is very small at 015 and large

conversion loss of 28dB there is still space for improvement in term of design ie

different circuit topology optimized input and output matching circuit etc This design

can be a good reference for a doubler operating at very low power but produce high

frequencies in Ka band

215

8 CONCLUSION AND FUTURE WORK

81 Conclusion

The main focus of this research was the development of a new tunneling diode

namely the asymmetrical spacer layer tunnel (ASPAT) diode for process repeatability

manufacturability and reproducibility The broad study undertaken was to improve the

microwave performance technology by introducing a new type of tunneling diode

For years the asymmetrical spacer layer tunnel diode was unable to be manufactured

due to the high sensitivity of the tunneling current to the barrier thickness This changed

dramatically when the MBE method was carefully optimized to precisely control the

growth to sub-monolayer precisions When stability repeatability and reproducibility in

the epitaxial growth was achieved the next step was to qualify the fabrication process of

the diodes themselves thus ensuring high performance device can be delivered to the

market

For this purpose GaAsAlAs ASPAT diodes made of two different types of

substrates were grown The first batch was grown in the Riber V100HU SSMBE and

used doped substrates Samples XMBE307 and XMBE368 were successfully grown

and fabricated from that batch The DC characterization obtained from measurement

proved that this first batch had fully functional reproducible and manufacturable

devices Later a second batch using semi insulating substrates improvement in spacer

layer and doping concentration were grown This set of samples (9 x 2rdquo wafers grown

simultaneously) and denoted as XMBE304 also showed fully functional DC

characteristic and was used for RF characterization and detector integrated circuits

The conventional GaAsAlAs ASPAT diode structures grown on doped

substrates and developed previously in our lab were not suitable for high frequency RF

characterizations Therefore a major contribution of this work was to develop a new

fabrication technique for a new GaAsAlAs ASPAT structure using semi insulating

substrates to achieve repeatability manufacturability and reproducibility in term of

process flow DC characteristics and ultimately RF characteristics Apart from the

enhancement of the epitaxial layer the other important contribution of this research was

216

the optimization of the small 4times4microm2 emitter size by incorporating both vertical and

lateral structure based purely on low cost I-line optical lithography

To obtain a repeatable and manufacturable fabrication process of lateral

GaAsAlAs ASPAT structures the key issue was to solve the over etching of the

effective mesa area when qualifying the Air Bridge technique This issue caused all

semiconductors under the metal contact to be completely lost thus leaving the metal

contact hanging without connection to any bond pad area However the developments of

the Dielectric Bridge technique realized the true performance of the GaAsAlAs ASPAT

diode structures The samplersquos surface cleanliness as well as DC and RF performance

showed significant improvement when using Si3N4 as a dielectric layer Due to the

highly isotropic etching profile and thicker GaAs layer in XMBE304 samples and

although the smallest emitter size designed on the mask was 2times2microm2 only 4times4microm

2 were

reproducible and showed good uniformity in I-V characteristics

Upon successful optimization in the fabrication process flow of the small emitter

size diodes (4times4microm2 6times6microm

2 and 10times10microm

2) a good uniformity of better than 91

was obtained for DC measurement results within a tile containing over 1000 devices

This confirmed that the lateral GaAsAlAs ASPAT diode structure can only be realized

through the Dielectric Bridge technique These devices were further characterized with

S-parameter measurements and their intrinsic and extrinsic parameter values and

junction capacitances series resistances and junction resistances were extracted leading

to intrinsic cut-off frequencies of 600GHz 429GHz and 100GHz for the three device

sizes respectively

Temperature dependence measurements and simulations were also carried out in

order to confirm that the ASPAT diodersquos characteristics were temperature insensitive

showing less than 5 change in current at both extremes of temperatures 77K to 400K

By comparison the SBD I-V characteristics variations with temperature span orders of

magnitude Physical modelling agreed very well with measured data confirming good

and validated models that can also describe temperature effects

For the realization of the integrated ASPAT millimeter wave detector empirical

modelling using ADS simulation tools was carried out This was performed to predict

the detector performance at 100GHz to comply with the initial objective to develop a

217

millimeter wave detector circuit The simulations using the 4times4microm2 diode models led to

a successful 100GHz circuit design able to detect 100GHz incoming frequency with

2100VW voltage sensitivity

The first ever GaAsAlAs ASPAT diode frequency multiplier design was also

attempted A reasonably good result was obtained for a 20 to 40GHz frequency doubler

operating in varistor mode However the conversion efficiency obtained was less than

1 Further research on this is required to improve the efficiency by using other circuit

topology ie using a balun or other Co-planar waveguides Ultimately fabricating and

testing the actual multiplier circuits are essential to validate the simulation data

82 Future Work

This work has provided a foundation for a reproducible and repeatable GaAsAlAs

ASPAT (SI substrate) wafer fabrication process and recommendations for design and

simulation of ASPAT diode MMIC detectors and frequency source has also been

provided However the GaAsAlAs ASPAT diodes still remain immature and there are

many ways to improve its DC and RF performances both experimentally and in

simulations which will directly affect the detection performances

In term of fabrication process smaller size diodes ie submicron level can be

achieved using dry etching technique with proper calibration For wet etch technique the

etched profile still can be improved by thinning the doped layers so that etching time

will be reduced and hence dimensions down to 2times2microm2 or even 15times15microm

2 can be

reproducibly made

For simulations advanced AC and RF modelling utilizing physical device

simulation available software (SILVACO) must be include in future research hence

holistic study can be conducted to improve the understanding of the ASPAT

For MMIC detector and multiplier design it is vital to produce actual MMIC

devices so that the simulation results can be validated Ultimately tested devices with

good performances can be realized and manufactured

218

REFERENCES

1 Laeri F U Simon and M Wark Host-Guest-Systems Based on Nanoporous

Crystals 2006 John Wiley amp Sons

2 Łukasiak L and A Jakubowski History of semiconductors Journal of

Telecommunications and information technology 2010 p 3-9

3 Song H-J and T Nagatsuma Present and future of terahertz communications

IEEE Transactions on Terahertz Science and Technology 2011 1(1) p 256-

263

4 Hu B and M Nuss Imaging with terahertz waves Optics letters 1995 20(16)

p 1716-1718

5 Smith PR DH Auston and MC Nuss Subpicosecond photoconducting

dipole antennas IEEE Journal of Quantum Electronics 1988 24(2) p 255-260

6 Nagatsuma T Terahertz technologies present and future IEICE Electronics

Express 2011 8(14) p 1127-1142

7 Kumar S et al A 18-THz quantum cascade laser operating significantly above

the temperature of [planck][omega]kB Nature Physics 2011 7(2) p 166-171

8 Phillips T and D Woody Millimeter-and submillimeter-wave receivers Annual

Review of Astronomy and Astrophysics 1982 20(1) p 285-321

9 Whitmer HCTaCA Crystal Rectifiers McGraw-Hill book Company

London 1948

10 Young D and J Irvin Millimeter frequency conversion using Au-n-type GaAs

Schottky barrier epitaxial diodes with a novel contacting technique Proceedings

of the Ieee 1965 53(12) p 2130-2131

11 Liu L et al A broadband quasi-optical terahertz detector utilizing a zero bias

Schottky diode IEEE microwave and wireless components letters 2010 20(9) p

504-506

12 Sankaran S Schottky barrier diodes for millimeter wave detection in a foundry

CMOS process IEEE Electron Device Letters 2005 26(7) p 492-494

13 Chattopadhyay G Submillimeter-wave coherent and incoherent sensors for

space applications in Sensors 2008 Springer p 387-414

14 Anand Y and WJ Moroney Microwave mixer and detector diodes

Proceedings of the Ieee 1971 59(8) p 1182-1190

15 Syme RT Microwave Detection Using GaasAlas Tunnel Structures Gec

Journal of Research 1993 11(1) p 12-23

16 Syme RT et al Asymmetric superlattices for microwave detection in Physical

Concepts of Materials for Novel Optoelectronic Device Applications 1991

International Society for Optics and Photonics

17 Missous M MJ Kelly and J Sexton Extremely uniform tunnel barriers for

low-cost device manufacture IEEE Electron Device Letters 2015 36(6) p 543-

545

18 Syme RT et al Novel GaAsAlAs tunnel structures as microwave detectors in

Semiconductors 92 1992 International Society for Optics and Photonics

19 Schwierz F and JJ Liou Semiconductor devices for RF applications evolution

and current status Microelectronics Reliability 2001 41(2) p 145-168

219

20 HayasHi H Development of Compound Semiconductor DevicesmdashIn Search of

Immense Possibilitiesmdash SEI TECHNICAL REVIEW 2011(72) p 5

21 Mead C Schottky barrier gate field effect transistor Proceedings of the Ieee

1966 54(2) p 307-308

22 Hooper W and W Lehrer An epitaxial GaAs field-effect transistor Proceedings

of the Ieee 1967 55(7) p 1237-1238

23 Drangeid K R Sommerhalder and W Walter High-speed gallium-arsenide

Schottky-barrier field-effect transistors Electronics Letters 1970 6(8) p 228-

229

24 Pillarisetty R Academic and industry research progress in germanium

nanodevices Nature 2011 479(7373) p 324-328

25 Oxley TH 50 years development of the microwave mixer for heterodyne

reception IEEE transactions on microwave theory and techniques 2002 50(3)

p 867-876

26 Baca AG and CI Ashby Fabrication of GaAs devices 2005 IET

27 Cho AY and J Arthur Molecular beam epitaxy Progress in solid state

chemistry 1975 10 p 157-191

28 Cho A Growth of IIIndashV semiconductors by molecular beam epitaxy and their

properties Thin Solid Films 1983 100(4) p 291-317

29 Kiehl RA and TG Sollner High speed heterostructure devices 1994

Academic Press

30 Feiginov M et al Resonant-tunnelling-diode oscillators operating at

frequencies above 11 THz Applied Physics Letters 2011 99(23) p 233506

31 Chang LL L Esaki and R Tsu Resonant tunneling in semiconductor double

barriers Applied Physics Letters 1974 24(12) p 593-595

32 Kasjoo SR Novel Electronic Nanodevices Operating in the Terahertz Region

2012

33 Kanaya H et al Structure dependence of oscillation characteristics of

resonant-tunneling-diode terahertz oscillators associated with intrinsic and

extrinsic delay times Japanese Journal of Applied Physics 2015 54(9) p

094103

34 Chattopadhyay G Technology capabilities and performance of low power

terahertz sources IEEE Transactions on Terahertz Science and Technology

2011 1(1) p 33-53

35 Betz A and R Boreiko A practical Schottky mixer for 5 THz in Proceedings of

the 7th International Symposium on Space Terahertz Technology 1996

36 Yu D et al Ultra high-speed 025-spl mum emitter InP-InGaAs SHBTs with

fsub maxof 687 GHz in Electron Devices Meeting 2004 IEDM Technical

Digest IEEE International 2004 IEEE

37 Das MB Optoelectronic detectors and receivers speed and sensitivity limits in

Optoelectronic and Microelectronic Materials Devices 1998 Proceedings 1998

Conference on 1999 IEEE

38 Rodwell MJ et al Submicron scaling of HBTs IEEE Transactions on Electron

Devices 2001 48(11) p 2606-2624

220

39 Bouloukou A and S Missous Novel High-breakdown Low-noise InGaAs-

InA1As Transistors for Radio Astronomy Applications 2006 University of

Manchester

40 Bean J Materials and technologies 1990 John Wiley amp Sons New York p

13

41 Swaminathan V and A Macrander Materials aspects of GaAs and InP based

structures 1991 Prentice-Hall Inc

42 Vurgaftman I J Meyer and L Ram-Mohan Band parameters for IIIndashV

compound semiconductors and their alloys Journal of applied physics 2001

89(11) p 5815-5875

43 Dingle R W Wiegmann and CH Henry Quantum states of confined carriers

in very thin Al x Ga 1minus x As-GaAs-Al x Ga 1minus x As heterostructures Physical

Review Letters 1974 33(14) p 827

44 Sze SM and KK Ng Physics of semiconductor devices 2006 John wiley amp

sons

45 Tyagi MS Introduction to semiconductor materials and devices 2008 John

Wiley amp Sons

46 Schubert E Delta doping of IIIndashV compound semiconductors Fundamentals

and device applications Journal of Vacuum Science amp Technology A Vacuum

Surfaces and Films 1990 8(3) p 2980-2996

47 Rhoderick EH Metal-semiconductor contacts IEE Proceedings I-Solid-State

and Electron Devices 1982 129(1) p 1

48 Piotrowska A A Guivarch and G Pelous Ohmic contacts to IIIndashV compound

semiconductors A review of fabrication techniques Solid-State Electronics

1983 26(3) p 179-197

49 Rideout V A review of the theory and technology for ohmic contacts to group

IIIndashV compound semiconductors Solid-State Electronics 1975 18(6) p 541-

550

50 Baca A et al A survey of ohmic contacts to III-V compound semiconductors

Thin Solid Films 1997 308 p 599-606

51 Higman T et al Structural analysis of AundashNindashGe and AundashAgndashGe alloyed

ohmic contacts on modulation‐doped AlGaAsndashGaAs heterostructures Journal of

applied physics 1986 60(2) p 677-680

52 Chen KJ et al High-performance enhancement-mode InAlAsInGaAs HEMTs

using non-alloyed ohmic contact and Pt-based buried-gate in Indium Phosphide

and Related Materials 1995 Conference Proceedings Seventh International

Conference on 1995 IEEE

53 Berlin L The man behind the microchip Robert Noyce and the invention of

Silicon Valley 2005 Oxford University Press

54 Goodhue W et al Large room‐temperature effects from resonant tunneling

through AlAs barriers Applied Physics Letters 1986 49(17) p 1086-1088

55 Kerr A and Y Anand Schottky diode MM detectors with improved sensitivity

and dynamic range Microwave Journal 1981 24 p 67-71

56 Davies R Simulations of the current-voltage characteristics of semiconductor

tunnel structures Gec Journal of Research 1987 5(2) p 65-75

221

57 Kelly M Tunnel structures and devices over the coming decade Philosophical

Transactions of the Royal Society of London A Mathematical Physical and

Engineering Sciences 1996 354(1717) p 2291-2293

58 Wilkinson V and M Kelly Manufacturability of quantum semiconductor

devices in High Performance Electron Devices for Microwave and

Optoelectronic Applications 1995 EDMO IEEE 1995 Workshop on 1995

IEEE

59 Wilkinson V M Kelly and M Carr Tunnel devices are not yet

manufacturable Semiconductor Science and Technology 1997 12(1) p 91

60 Eaves L and MJ Kelly The current status of semiconductor tunnelling devices

Philos trans of the Roy soc of London Ser A Math phys and eng sciences

1996 354(1717)

61 Billen K V Wilkinson and M Kelly Manufacturability of heterojunction

tunnel devices further progress Semiconductor Science and Technology 1997

12(7) p 894

62 Kelly M The engineering of quantumndashdot devices Philosophical Transactions

of the Royal Society of London A Mathematical Physical and Engineering

Sciences 2003 361(1803) p 393-401

63 Kelly M New statistical analysis of tunnel diode barriers Semiconductor

Science and Technology 2000 15(1) p 79

64 Hayden R et al Ex situ re-calibration method for low-cost precision epitaxial

growth of heterostructure devices Semiconductor Science and Technology

2002 17(2) p 135

65 Dasmahapatra P et al Thickness control of molecular beam epitaxy grown

layers at the 001ndash01 monolayer level Semiconductor Science and Technology

2012 27(8) p 085007

66 Hayden R M Missous and M Kelly Precision growth for the manufacture of

semiconductor heterostructure devices Semiconductor Science and Technology

2001 16(8) p 676

67 Shao C et al Highly reproducible tunnel currents in MBE-grown

semiconductor multilayers Electronics Letters 2012 48(13) p 792-794

68 Abdullah MR et al GaAsAlAs tunnelling structure Temperature dependence

of ASPAT detectors in Millimeter Waves and THz Technology Workshop

(UCMMT) 2015 8th UK Europe China 2015 IEEE

69 Ariffin KZ et al Asymmetric Spacer Layer Tunnel In0 18Ga0 82AsAlAs

(ASPAT) Diode using double quantum wells for dual functions Detection and

oscillation in Millimeter Waves and THz Technology Workshop (UCMMT)

2015 8th UK Europe China 2015 IEEE

70 Liboff RL Introductory quantum mechanics 2003 Addison-Wesley

71 Esaki L Discovery of the tunnel diode IEEE Transactions on Electron Devices

1976 23(7) p 644-647

72 Landau LD LEM Quantum Mechanics Non-relativistic Theory Pergamon 3

73 Landau LD et al Quantum Mechanics Non‐Relativistic Theory Vol 3 of

Course of Theoretical Physics 1958 AIP

222

74 Syme R Tunnelling devices as microwave mixers and detectors Philosophical

Transactions of the Royal Society of London A Mathematical Physical and

Engineering Sciences 1996 354(1717) p 2351-2364

75 Syme R et al Tunnel diode with asymmetric spacer layers for use as

microwave detector Electronics Letters 1991 27(23) p 2192-2194

76 Brown E W Goodhue and T Sollner Fundamental oscillations up to 200 GHz

in resonant tunneling diodes and new estimates of their maximum oscillation

frequency from stationary‐state tunneling theory Journal of applied physics

1988 64(3) p 1519-1529

77 Reddy M Schottky-collector resonant tunnel diodes for sub-millimeter-wave

applications 1997 University of California Santa Barbara

78 Cox R and H Strack Ohmic contacts for GaAs devices Solid-State Electronics

1967 10(12) p 1213IN71215-1214IN81218

79 Valdes LB Resistivity measurements on germanium for transistors

Proceedings of the IRE 1954 42(2) p 420-427

80 Schroder DK Semiconductor material and device characterization 2006 John

Wiley amp Sons

81 Klootwijk J and C Timmering Merits and Limitations of Circular TLM

structures for contact resistance determination for novel 111-V HBTs Proc

fEEE 2004

82 Marlow GS and MB Das The effects of contact size and non-zero metal

resistance on the determination of specific contact resistance Solid-State

Electronics 1982 25(2) p 91-94

83 Murrmann H and D Widmann Current crowding on metal contacts to planar

devices IEEE Transactions on Electron Devices 1969 16(12) p 1022-1024

84 Berger H Models for contacts to planar devices Solid-State Electronics 1972

15(2) p 145-158

85 Reeves G and H Harrison Obtaining the specific contact resistance from

transmission line model measurements IEEE Electron Device Letters 1982

3(5) p 111-113

86 Shur MS GaAs devices and circuits 2013 Springer Science amp Business

Media

87 Popescu D and B Odbert The Advantages Of Remote Labs In Engineering

Education Educatorrsquos Corner-Agilent Technologies-application note 2011 p

11

88 DataSheet Karl Suss- PM5 Probe System Datasheet 2013

89 Keysight IC-CAP Device Modeling Software 2016 Available from

httpwwwkeysightcomenpc-1297149ic-cap-device-modeling-software-

measurement-control-and-parameter-extractioncc=USamplc=eng

90 DataSheet Anritsu 37369A Vector Network Analyzer Datasheet 2016 Available

from httpwwwtestequipmenthqcomdatasheetsANRITSU-37397D-

Datasheetpdf

91 Packard H HP 4142B Modular DC SourceManual Operation Manual 1992

Available from httpcpliteratureagilentcomlitwebpdf04142-90010pdf

223

92 Microtech C Cascade Microtech- Wincal High Performence RF calaibration

Software (Official Website) 2016 Available from

httpswwwcascademicrotechcom

93 Singh J Electronic and optoelectronic properties of semiconductor structures

2007 Cambridge University Press

94 Whyte W Cleanroom design 1999 Wiley Online Library

95 Vieu C et al Electron beam lithography resolution limits and applications

Applied Surface Science 2000 164(1) p 111-117

96 La Fontaine B Lasers and Moorersquos law SPIE Professional October 2010 p

20

97 Madou MJ Fundamentals of microfabrication the science of miniaturization

2002 CRC press

98 Serway R Physics for Scientists and Engineers 1996 Saunders Publ

Philadelphia

99 Jalali B and S Pearton InP HBTs growth processing and applications 1995

Artech House Publishers

100 Shih YC et al Effects of interfacial microstructure on uniformity and thermal

stability of AuNiGe ohmic contact to n‐type GaAs Journal of applied physics

1987 62(2) p 582-590

101 Zawawi M Advanced In0 8Ga0 2AsAlAs Resonant Tunneling Diodes

forApplications in Integrated mm-waves MMIC Oscillators 2015

102 Zawawi MAM et al Fabrication of Submicrometer InGaAsAlAs Resonant

Tunneling Diode Using a Trilayer Soft Reflow Technique With Excellent

Scalability IEEE Transactions on Electron Devices 2014 61(7) p 2338-2342

103 Silvaco I ATLAS Users Manual Device Simulation Software 2010 Santa

Clara CA

104 Kyono C et al Dependence of apparent barrier height on barrier thickness for

perpendicular transport in AlAsGaAs single‐barrier structures grown by

molecular beam epitaxy Applied Physics Letters 1989 54(6) p 549-551

105 Yang K JR East and GI Haddad Numerical modeling of abrupt

heterojunctions using a thermionic-field emission boundary condition Solid-

State Electronics 1993 36(3) p 321-330

106 Varshni YP Temperature dependence of the energy gap in semiconductors

Physica 1967 34(1) p 149-154

107 Handbook LLM Precision DC Current Voltage and Resistance

Measurements Keithley Instruments Inc[online] 6th revision Ohio 2004

108 Lipsky SE Microwave passive direction finding 2004 SciTech Publishing

109 Howell CM and SJ Parisi Principles Applications and Selection of Receiving

Diodes MACOM Semiconductor Products Division Application note AG314

110 Schulman J D Chow and D Jang InGaAs zero bias backward diodes for

millimeter wave direct detection IEEE Electron Device Letters 2001 22(5) p

200-202

111 Zhang Z et al Sub-Micron Area Heterojunction Backward Diode Millimeter-

Wave Detectors With 018$ rm pWHz^12 $ Noise Equivalent Power IEEE

microwave and wireless components letters 2011 21(5) p 267-269

224

112 Jin N et al High sensitivity Si-based backward diodes for zero-biased square-

law detection and the effect of post-growth annealing on performance IEEE

Electron Device Letters 2005 26(8) p 575-578

113 Shashkin VI et al Millimeter-wave detectors based on antenna-coupled low-

barrier Schottky diodes International journal of infrared and millimeter waves

2007 28(11) p 945-952

114 Zhao P et al GaN Heterostructure Barrier Diodes Exploiting Polarization-

Induced $delta $-Doping IEEE Electron Device Letters 2014 35(6) p 615-

617

115 Pozar DM Microwave engineering 2009 John Wiley amp Sons

116 Koolen M J Geelen and M Versleijen An improved de-embedding technique

for on-wafer high-frequency characterization in Bipolar Circuits and

Technology Meeting 1991 Proceedings of the 1991 1991 IEEE

117 Cao M et al RF characteristics uniformity of GaAsAlAs tunnel diodes in

Infrared Millimeter and Terahertz waves (IRMMW-THz) 2016 41st

International Conference on 2016 IEEE

118 Gao J RF and microwave modeling and measurement techniques for field effect

transistors 2010 SciTec

119 Ren T et al A 340-400 GHz Zero-Biased Waveguide Detector Using an Self-

Consistent Method to Extract the Parameters of Schottky Barrier Diode Applied

Computational Electromagnetics Society Journal 2015 30(12)

120 Fobelets K et al High‐frequency capacitances in resonant interband tunneling

diodes Applied Physics Letters 1994 64(19) p 2523-2525

121 Diebold S et al Modeling and Simulation of Terahertz Resonant Tunneling

Diode-Based Circuits IEEE Transactions on Terahertz Science and Technology

2016 6(5) p 716-723

122 Yong Z et al Design of a 220 GHz frequency tripler based on EM model of

Schottky diodes JOURNAL OF INFRARED AND MILLIMETER WAVES

2014 33(4) p 405-411

123 Louhi JT and AV Raisanen On the modeling and optimization of Schottky

varactor frequency multipliers at submillimeter wavelengths IEEE transactions

on microwave theory and techniques 1995 43(4) p 922-926

124 Guo J Z Zhang and C Qian Modeling of commercial millimeter wave

Schottky diodes in Microwave and Millimeter Wave Technology (ICMMT) 2016

IEEE International Conference on 2016 IEEE

125 Schneider M Metal-semiconductor junctions as frequency converters Infrared

and Millimeter Waves 1982 6 p 209

126 Muth C et al Advanced technology microwave sounder on NPOESS and NPP

in Geoscience and Remote Sensing Symposium 2004 IGARSS04 Proceedings

2004 IEEE International 2004 IEEE

127 Putley E Thermal detectors in Optical and Infrared Detectors 1977 Springer

p 71-100

128 Martin DH Spectroscopic techniques for far infra-red submillimetre and

millimetre waves in Spectroscopic Techniques for Far Infra-red Submillimetre

and Millimetre Waves 1967

225

129 Lucas W Tangential sensitivity of a detector video system with RF

preamplification in Proceedings of the Institution of Electrical Engineers 1966

IET

130 Balocco C et al Low-frequency noise of unipolar nanorectifiers Applied

Physics Letters 2011 99(11) p 113511

131 Benford D T Hunter and TG Phillips Noise equivalent power of background

limited thermal detectors at submillimeter wavelengths International journal of

infrared and millimeter waves 1998 19(7) p 931-938

132 Papoušek D Vibration-rotational Spectroscopy and Molecular Dynamics

Advances in Quantum Chemical and Spectroscopical Studies of Molecular

Structures and Dynamics Vol 9 1997 World Scientific

133 Nyquist H Thermal agitation of electric charge in conductors Physical review

1928 32(1) p 110

134 Turner CS Johnson-Nyquist Noise url httpwww claysturner

comdspJohnson-NyquistNoise pdf(Letzter Abruf Juli 2012)

135 Voss RF 1f (flicker) noise A brief review in 33rd Annual Symposium on

Frequency Control 1979 1979 IEEE

136 McWhorter AL 1f noise and related surface effects in germanium 1955

137 Hooge FN 1ƒ noise is no surface effect Physics letters A 1969 29(3) p 139-

140

138 Van der Ziel A Noise Sources characterization measurement Prentice-Hall

Information and System Sciences Series Englewood Cliffs Prentice-Hall 1970

1970

139 Hooge F 1f noise sources IEEE Transactions on Electron Devices 1994

41(11) p 1926-1935

140 Der Ziel A Theory of shot noise in junction diodes and junction transistors

Proceedings of the IRE 1955 43(11) p 1639-1646

141 Schottky W Small-shot effect and flicker effect Physical review 1926 28(1)

p 74

142 Cowley A and H Sorensen Quantitative comparison of solid-state microwave

detectors IEEE transactions on microwave theory and techniques 1966 14(12)

p 588-602

143 Schulman J et al 1$f $ Noise of Sb-Heterostructure Diodes for Pre-Amplified

Detection IEEE microwave and wireless components letters 2007 17(5) p

355-357

144 Lynch JJ et al Passive millimeter-wave imaging module with preamplified

zero-bias detection IEEE transactions on microwave theory and techniques

2008 56(7) p 1592-1600

145 Su N et al Sb-heterostructure millimeter-wave detectors with reduced

capacitance and noise equivalent power IEEE Electron Device Letters 2008

29(6) p 536-539

146 Westlund A Self-Switching Diodes for Zero-Bias Terahertz Detection 2015

Chalmers University of Technology

147 Yajima T and L Esaki Excess noise in narrow germanium pn junctions

Journal of the physical society of Japan 1958 13(11) p 1281-1287

226

148 Sommers H Tunnel diodes as high-frequency devices Proceedings of the IRE

1959 47(7) p 1201-1206

149 Miraftab V and A Abdipour Harmonic balance analysis of a microwave

balanced power amplifier in Electrical and Computer Engineering 2001

Canadian Conference on 2001 IEEE

150 Patrashin M et al GaAsSbInAlAsInGaAs Tunnel Diodes for Millimeter Wave

Detection in 220ndash330-GHz Band IEEE Transactions on Electron Devices 2015

62(3) p 1068-1071

151 Hesler JL and TW Crowe Responsivity and noise measurements of zero-bias

Schottky diode detectors Proc ISSTT 2007 p 89-92

152 Su N et al Temperature dependence of high frequency and noise performance

of Sb-heterostructure millimeter-wave detectors IEEE Electron Device Letters

2007 28(5) p 336-339

153 Lynch J et al Unamplified direct detection sensor for passive millimeter wave

imaging in Proc of SPIE Vol 2006

154 ZHANG Z et al A physics-based tunneling model for Sb-heterostructure

backward tunnel diode millimeter-wave detectors International Journal of High

Speed Electronics and Systems 2011 20(03) p 589-596

155 Bahl IJ and P Bhartia Microwave solid state circuit design 2003 John Wiley

amp Sons

156 Yeom K-W Microwave Circuit Design A Practical Approach Using ADS

2015 Prentice Hall Press

157 Xie L et al A W-band detector with high tangential signal sensitivity and

voltage sensitivity in Microwave and Millimeter Wave Technology (ICMMT)

2010 International Conference on 2010 IEEE

158 Fay P et al High-performance antimonide-based heterostructure backward

diodes for millimeter-wave detection IEEE Electron Device Letters 2002

23(10) p 585-587

159 Hrobak M et al Planar zero bias Schottky diode detector operating in the E-

and W-band in Microwave Conference (EuMC) 2013 European 2013 IEEE

160 Maas SA Nonlinear microwave and RF circuits 2003 Artech House

161 Appleby R and RN Anderton Millimeter-wave and submillimeter-wave

imaging for security and surveillance Proceedings of the Ieee 2007 95(8) p

1683-1690

162 Crowe TW et al Opening the terahertz window with integrated diode circuits

IEEE Journal of Solid-State Circuits 2005 40(10) p 2104-2110

163 Raisanen AV Frequency multipliers for millimeter and submillimeter

wavelengths Proceedings of the Ieee 1992 80(11) p 1842-1852

164 Erickson NR Diode frequency multipliers for terahertz local-oscillator

applications in Astronomical Telescopes amp Instrumentation 1998 International

Society for Optics and Photonics

165 Mehdi I et al Terahertz local oscillator sources performance and capabilities

in Astronomical Telescopes and Instrumentation 2003 International Society for

Optics and Photonics

166 Tonouchi M Cutting-edge terahertz technology Nature photonics 2007 1(2)

p 97-105

227

167 Nilsen SM et al Single barrier varactors for submillimeter wave power

generation IEEE transactions on microwave theory and techniques 1993 41(4)

p 572-580

168 Xiao Q et al A 270-GHz tuner-less heterostructure barrier varactor frequency

tripler IEEE microwave and wireless components letters 2007 17(4) p 241-

243

169 David T et al Monolithic integrated circuits incorporating InP-based

heterostructure barrier varactors IEEE microwave and wireless components

letters 2002 12(8) p 281-283

170 Lieneweg U B Hancock and J Maserjian Barrier-intrinsic-N+(BIN) diodes

for near-millimeter wave generation in Conference Digest for the Twelft

International Conference on Infrared and Millimeter Waves 1987

171 Lieneweg U et al Modeling of planar varactor frequency multiplier devices

with blocking barriers IEEE transactions on microwave theory and techniques

1992 40(5) p 839-845

172 Chattopadhyay G et al An all-solid-state broad-band frequency multiplier

chain at 1500 GHz IEEE transactions on microwave theory and techniques

2004 52(5) p 1538-1547

173 Maestrini A et al A 17-19 THz local oscillator source IEEE microwave and

wireless components letters 2004 14(6) p 253-255

174 Faber MT J Chramiec and ME Adamski Microwave and millimeter-wave

diode frequency multipliers 1995 Artech House Publishers

175 Frerking MA and JR East Novel heterojunction varactors Proceedings of the

Ieee 1992 80(11) p 1853-1860

176 Penfield P and RP Rafuse Varactor applications 1962

177 Palazzi V et al Low-power frequency doubler in cellulose-based materials for

harmonic RFID applications IEEE microwave and wireless components letters

2014 24(12) p 896-898

178 Presas SM Microwave frequency doubler integrated with miniaturized planar

antennas 2008

228

APPENDICES

Appendix I Doped substrate process details

Mask Stage Process Stage Process step Process detail Equipment

Mask 1 (Mesa

Etch)

Sample clean NMP 10 min 80˚C Beaker

Acetone 5 min Beaker

Isopropanol (IPA) 5 min Beaker

Apply Resist Prebake Bake for 5mins 150C Hotplate

1st Resist S1805 - Program 4 Laurell Spinner

Post bake 110C for 1 Min Hotplate

Photolithography Mask Align mask to wafer

MA4 Mask

aligner

Expose

1 min 09mW iline

(Compensation error set to 1)

MA4 Mask

aligner

Develop MIF 319 for 1 min Beaker

Post Exposure

Bake Oven or Bake 120C for 15

minutes Hotplate

Etch Etch Cal Cal Sample - etch for 2 minutes Beaker

Etch

H3PO4H2O2H2O 3150

time is determined by the etch cal Beaker

Resist Strip Acetone - 5 min and IPA - 5 min Beaker

Mask2 (Top

Contact)

Sample clean NMP ultrasonic for 15 Min Power 3 Beaker

Acetone ultrasonic for 5 Min Power 3 Beaker

Isopropanol (IPA) ultrasonic for 5 Min Power 3 Beaker

Apply Resist Prebake Bake for 5mins 150C Hotplate

1st Resist AznLoF - 2um grade - Program 6 Laurell Spinner

Post bake 110C for 1 Min Hotplate

Photolithography Mask Align mask to wafer

MA4 Mask

aligner

Expose

55 seconds 09mW i-line

(Compensation error set to 1)

MA4 Mask

aligner

Post Exposure bake 110C for 1 Min Hotplate

Develop MIF 326 for 1 mins Beaker

Clean O2 Plasma Etch

20sec 100W forward (340)

60mTorr Plasma Tech

Surface De-oxide HCLH2O 11 40 sec Beaker

Metallisation Evaporation Clean trays + metals Beaker

Load sample and metals

Edward

Evaporator

AuGe (100mg) deposits 55nm

Edward

Evaporator

Ni (1cm) deposits 15nm

Edward

Evaporator

Au (10cm) deposits 200nm

Edward

Evaporator

Lift-off

NMP1165 for 24 hours (25C) or

20 min 80C Beaker

Anneal 420˚C 2min Furnace

229

TLM

Measurement ICCAP

Measurement

Bench

Mask 3

(BottomBacks

ide Contact)

Sample clean Acetone Optional Beaker

Isopropanol (IPA) Optional Beaker

Apply Resist Prebake

Bake for 5mins 150C to dry

the sample Hotplate

Resist (top side) S1813 - Program 6 Laurell Spinner

Soft bake 110C for 1 Min Hotplate

Photolithography Mask Align mask to wafer

MA4 Mask

aligner

Exposure

10 seconds 09mW I-line

(Compensation error set to 1)

MA4 Mask

aligner

Post Exposure bake 110C for 1 Min Hotplate

Develop MIF 319 for 2 mins Beaker

De-scum O2 Plasma Etch

20sec 100W forward (340)

60mTorr Plasma Tech

De-oxidise

Surface De-oxide

HCLH2O 11 40

sec Beaker

metallisation Evaporation Clean trays + metals Beaker

Load sample and metals

Edward

Evaporator

AuGe (100mg) deposits 55nm

Edward

Evaporator

Ni (1cm) deposits 15nm

Edward

Evaporator

Au (10cm) deposits 200nm

Edward

Evaporator

Lift-off

NMP1165 for 24 hours (25C) or

20 min 80C Beaker

Clean Water 3min Beaker

TLM

Measurement ICCAP

230

Appendix II Four Mask step Process Flow

Mask Stage Process Stage Process step Process detail

Mask1 Top

Contact)

Sample clean NMP ultrasonic for 15 Min Power 3

Acetone ultrasonic for 5 Min Power 3

Isopropanol (IPA) ultrasonic for 5 Min Power 3

Apply Resist Prebake Bake for 5mins 150C

1st Resist AznLoF - 2um grade - Program 6

Hot Plate 110C for 1 Min

Photolithography Mask Align mask to wafer

Expose 55 seconds 09mW iline (Compensation

error set to 1)

Post Exposure bake 110C for 1 Min

Develop MIF 326 for 1 mins

Clean O2 Plasma Etch 20sec 100W forward (340) 60mTorr

Surface De-oxide HCLH2O 11 40 sec

Metalisation Evaporation Clean trays + metals

Load sample and metals

AuGe (100mg) deposits 55nm

Ni (1cm) deposits 15nm

Au (10cm) deposits 200nm

Lift-off NMP1165 for 24 hours (25C) or 20 mnt 80C

Mask 2 (Mesa

Etch)

Sample clean NMP Optional

Acetone Optional

Isopropanol (IPA) Optional

Apply Resist Prebake Bake for 5mins 150C

1st Resist S1805 - Program 4

HotPlate 110C for 1 Min

Photolithography Mask Align mask to wafer

Expose 1 min 09mW i-line (Compensation error set

to 1)

Develop MIF 319 for 1 min

Post Exposure Bake Oven or Bake 120C for 15 minutes

Etch Etch Cal Cal Sample - etch for 2 minutes

Etch H3PO4H2O2H2O 3150 time is

determined by the etch cal

Measure TLM

Resist Strip Acetone - 5 min and IPA - 5 min

Mask 3(Isolation) Sample clean Acetone Optional

Isopropanol (IPA) Optional

231

Apply Resist Prebake Bake for 5mins 150C

Resist S1828 - Program 4

Hot Plate 115C for 1 mins

Photolithography Mask Align mask to wafer

Expose 9 mins 09mW iline (Compensation error set

to 1)

Develop MF 319 3 mins

Post Bake Oven Bake 120C for 15mins

Etch Etch Cal Refer to Etch Cal instr tab

Sub-collector Etch H3PO4H2O2H2O 3150 time is

determined by the etch cal

Resist Strip Acetone 5mins + IPA 5 mins in ultrasonic bath

power 1

inspection Microscope

Sample clean Acetone Optional

Mask 4 (Bottom

Contact)

Isopropanol (IPA) Optional

Apply Resist Prebake Bake for 5mins 150C 1st

Resist AznLoF - 2um grade - Program 6

HotPlate 110C for 1 Min

Photolithography Mask Align mask to wafer

Expose 10 seconds 09mW i-line (Compensation

error set to 1)

Post Exposure bake 110C for 1 Min

Develop MIF 326 for 2 mins

Clean O2 Plasma Etch 20sec 100W forward (340) 60mTorr

Surface De-oxide

HCLH2O 11 40 sec

Metalisation Evaporation Clean trays + metals

Load sample and metals

AuGe (100mg) deposits 55nm

Ni (1cm) deposits 15nm

Au (10cm) deposits 200nm

Lift-off NMP1165 for 24 hours (25C) or 20 mnt 80C

232

Appendix III Epitaxial Layer XMBE277

TABLE I The epitaxial structure for sample XMBE277

Layer Thickness (nm) Doping Concentration (cm-3)

n+- In053Ga047As 45 200 x 1019

n- In053Ga047As 25 300 x 1018

In053Ga047As 20 undoped

AlAs 13 undoped

In08Ga02As 45 undoped

AlAs 13 undoped

In053Ga047As 20 undoped

n- In053Ga047As 25 300 x 1018

n+- In053Ga047As 400 100 x 1019

Semi-insulating InP

233

Appendix IV SilVaco (Atlas) Simulation Code

go atlas

---------------------------------------------------------

Structure parameter definition (Constants) values in um

---------------------------------------------------------

Thicknesses

set t_contact1=0

set t_ohmic1=03

set t_emitter=004

set t_spacer1=0005

set t_barrier=000283

set t_spacer2=02

set t_collector=004

set t_ohmic2=045

Doping concentrations

set d_ohmic1=4e18

set d_emitter=2e17

set d_collector=2e17

set d_ohmic2=4e18

set d_gap=2

set d_mesa=4

set d_device=10

set d_etch=008

Layers

set I=$t_contact1

set A=$I+$t_ohmic1

set B=$A+$t_emitter

set C=$B+$t_spacer1

set D=$C+$t_barrier

set E=$D+$t_spacer2

set F=$E+$t_collector

set G=$F+$t_ohmic2

-------------------------------------

Mesh generator

-------------------------------------

mesh diagflip width=45

xmesh location=0 s=1

xmesh location=1 s=1

xmesh location=2 s=1

xmesh location=4 s=1

xmesh location=5 s=1

xmesh location=6 s=1

xmesh location=7 s=1

xmesh location=8 s=1

xmesh location=$d_mesa s=1

xmesh location=$d_mesa+$d_gap s=1

xmesh location=$d_device s=1

Ohmic1

ymesh l=0000 s=005

ymesh l=$I s=005

234

ymesh l=$A s=0005

ymesh l=$B s=0005

ymesh l=$C s=00005

ymesh l=$D s=00005

ymesh l=$E s=0009

ymesh l=$F s=0005

ymesh l=$G s=0005

-----------------------------------

SECTION 2 Regions Structure definition

-----------------------------------

region num=1 name=contact1 material=Gold ymin=0 ymax=$I

region num=2 name=ohmic1 material=GaAs ymin=$I ymax=$A

region num=3 name=emitter material=GaAs ymin=$A ymax=$B

region num=4 name=spacer1 material=GaAs ymin=$B ymax=$C

region num=5 name=barrier material=AlAs ymin=$C ymax=$D xmin=0 xmax=$d_mesa

calcstrain qtregion=1

region num=6 name=spacer2 material=GaAs ymin=$D ymax=$E

region num=7 name=collector material=GaAs ymin=$E ymax=$F

region num=8 name=ohmic2 material=GaAs ymin=$F ymax=$G

region num=9

name=etch material=Air ymin=0 ymax=$F+$d_etch xmin=$d_mesa xmax=$d

_device

---------------------------------

Electrodes

---------------------------------

electrode num=1 name=anode xmin=0 xmax=$d_mesa ymin=0 ymax

=$I material=Gold

electrode num=2 name=cathode xmin=$d_mesa+$d_gap xmax=$d_device ymin=$F+

$d_etch ymax=$F+$d_etch material=Gold

--------------------------------

Doping

--------------------------------

doping uniform ntype conc=$d_ohmic1 Region=2 ymin=$I ymax=$A

doping uniform ntype conc=$d_emitter Region=3 ymin=$A ymax=$B

doping uniform ntype conc=$d_collector Region=7 ymin=$E ymax=$F

doping uniform ntype conc=$d_ohmic2 Region=8 ymin=$F ymax=$G

--------------------------

Contacts

--------------------------

interface sc region=1

interface ss region=2

interface ss region=3

interface si region=4

interface si region=5

interface ss region=6

interface ss region=7

interface sc ymin=$F ymax=$F xmin=$d_mesa+$d_gap xmax=$d_device

interface tunnel region=5 dytunnel=0001

contact name=cathode

contact name=anode

235

------------------------------------------

SECTION 3 Material amp Models Definitions

------------------------------------------

material material=AlAs

permittivity=10 eg300=28 mc=004 affinity=305 nc300=4e19 nimin=1e1

material material=GaAs permittivity=139 eg300=14 mc=0067

affinity=407 nc300=09e17 nimin=1e6

BAND DIAGRAM

output tquantum bandparam qfn qfp valband conband charge polarcharge flowlines

STRUCTURE GRAPHIC

solve init

save outf=XMBE304+real2str

tonyplot XMBE304+real2str

------------------------------------------

SECTION 4 ANALYSIS

------------------------------------------

trap acceptor structure=top elevel=03 density=48e17 degenfac=10 sign=284e-17 sigp=1e-17

ymin=$A ymax=$C xmin=0 xmax=$d_mesa

trap acceptor structure=BOTTOM elevel=035 density=47e17 degenfac=10 sign=284e-17 sigp=1e-17

ymin=$E ymax=$F xmin=0 xmax=$d_mesa

models sisel sisnlderivs qtregion=1 print

method climit=1e-4 itlimit=50 maxtraps=20

DC ANALYSIS

log outf=XMBE304log

solve init

solve vanode=0 name=anode vstep=001 vfinal=15

save outf=XMBE304str

log off

tonyplot XMBE304str

tonyplot XMBE304log

Page 2: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre

2

LIST OF CONTENTS

LIST OF CONTENTS 2

LIST OF TABLES 6

LIST OF FIGURES 8

LIST OF PUBLICATIONS CONFERENCE PRESENTATIONS 13

PUBLICATIONS 13

CONFERENCE PRESENTATIONS 14

ABSTRACT 15

DECLARATION AND COPYRIGHT STATEMENT 16

ACKNOWLEDGEMENTS 17

DEDICATION 18

1 INTRODUCTION 19

11 Background 19

12 Aims and objectives 26

13 Outline of this Thesis 27

2 LITERATURE REVIEW 29

21 Introduction 29

22 Historical review of III-V Compound Semiconductor for RF applications 30

23 The Concept of Heterostructures 33

231 Homojunctions Heterojunctions and Band Discontinuities 34

232 Lattice-Matched and Pseudomorphic Materials 36

233 Quantum well and 2DEG 41

24 Metal-Semiconductor Contact 42

241 Schottky Contact 43

242 Ohmic Contact 46

25 Asymmetrical Spacer-layer Tunnel (ASPAT) diode Background work 48

26 Asymmetrical Spacer-layer Tunnel (ASPAT) diode The Basics 50

261 Principle of Quantum Tunneling 51

262 ASPAT Structural Parameters of GaAsAlAs materials System 55

Figure 214 Epitaxial layer structure of device XMBE 304 (lateral structure)

used in this study 55

3

263 ASPAT Electrical Parameters 61

27 Characterization of Ohmic Contacts 68

271 Transmission Line Measurement (TLM) 69

28 Basic Characterization Techniques and procedures 72

281 Measuring tools and apparatuses 72

282 Measurement steps using a VNA 75

283 Measurement Practice and Flowchart 76

3 RESULTS OF FABRICATION PROCESS OF ASPAT DIODES 79

31 Introduction 79

32 Epitaxial Layer Growth Techniques 80

321 Molecular Beam Epitaxy (MBE) 80

33 Basic Principles of Common Fabrication techniques 81

331 Sample cleaning 81

332 Photolithography 82

333 Etching Process 86

334 Sputtering (dielectric deposition) 88

335 Metallization Process Lift-off and Annealing 88

34 GaAsAlAs ASPAT Process Optimization 92

341 ASPAT Devices used in Fabrication 93

342 Fabrication process of GaAsAlAs ASPAT diode toward repeatability

reproducibility and manufacturability 95

343 Fabrication process of GaAsAlAs ASPAT diode toward High

frequency Applications 117

35 Conclusions 125

4 PHYSICAL MODELLING OF THE GAASALAS ASPAT DIODE USING

SILVACO 126

41 Introduction 126

42 SILVACO modelling Tools 127

43 SILVACO Implementation GaAs AlAs ASPAT Modelling 130

44 Simulation Result and Analysis 132

45 Structure Analysis of ASPAT Diode 135

451 Dependencies of current on AlAs Barrier thickness 135

452 Dependence of current on Spacer I length l1 137

4

453 Dependence of current on Spacer II length l2 138

46 Temperature Dependent Simulation 140

47 Temperature Dependence characteristics of GaAsAlAs ASPAT Diodes 142

48 Conclusions 147

5 DC amp RF CHARACTERISTIC OF ASPAT DIODES 148

51 Introduction 148

52 General Analysis of Current-Voltage (I-V) characteristics of GaAsAlAs

ASPAT diodes 149

53 RF Test Fixture Theory and Experiment 153

531 On-Wafer Measurement and Small Signal One-Port Characterizations

154

54 Device Calibration 155

541 Open and Short De-Embedding Technique 155

55 S-Parameter Measurement Result and Analysis 157

551 Diode to diode uniformity 158

552 Wafer to wafer uniformity 160

553 Small devices RF measurements 161

56 Extracting RF models of ASPAT at Zero Bias Voltage 164

561 Extraction of ASPAT parasitic element 165

562 Extraction of ASPAT intrinsic elements 168

563 Capacitances -Voltage (C-V) Extraction 172

57 Conclusions 173

6 MILLIMETRE WAVE GAASALAS ASPAT DETECTOR DESIGN USING

ADS 175

61 Introduction 175

62 Detection Theory 176

63 Definition of Characteristics of Interest Detection Parameters in small signal

analysis 179

64 Noise Consideration in a Detector diode 182

65 Modelling of a 100GHz Zero-biased ASPAT Detector 184

66 Conclusions 199

7 FREQUENCY MULTIPLIERS (DOUBLER) DESIGN USING GAASALAS

ASPAT DIODES 201

71 Introduction 201

5

72 Motivation and Background 202

73 Frequency Multiplier Architecture the Basics 203

731 Types of frequency multipliers 205

74 Parameters of interest for Frequency Multipliers 206

75 20GHz40GHz Varistor Mode ASPAT Diode Frequency Doubler 208

76 Conclusions 214

8 CONCLUSION AND FUTURE WORK 215

81 Conclusion 215

82 Future Work 217

REFERENCES 218

APPENDICES 228

Word count including footnotes and endnotes 61500(approximately)

6

LIST OF TABLES

Table 11 Sample XMBE304 XMBE307 and XMBE368 GaAsAlAs materials

structure grown on GaAs Substrates by MBE 23

Table 12 Quantum wells sandwiching the quantum barrier for sample XMBE314

grown on a GaAs substrate by MBE 24

Table 13 Novel InGaAsAlAs ASPAT diode grown on InP substrate by MBE on InP

substrate 24

Table 21 Lattice constant (a) and band gap for group III-V binary and ternary

compound semiconductors a room temperature [41 42] 38

Table 22 epitaxial layer of sample XMBE304 which is the main focus of this work 56

Table 31 Chemical wet etch characteristics using Orthophosphoric-based and

Ammonia on GaAs and InGaAs materials 87

Table 32 Epitaxial layer of Doped substrate samples 93

Table 33 The epitaxial layers of sample XMBE304 with barrier 10ML~28nm 94

Table 34 Generic fabrication steps established by Dr Md Adzhar [101] 97

Table 35 Standard process flow for Air-Bridge design fabrication 101

Table 36 Standard fabrication process flow for Dielectric-Bridge design 107

Table 37 New arrangement of the mask number and step in Second Run 110

Table 38 New arrangement for the Third run using Dielectric-Bridge mask 113

Table 39 The outcome of the spreading resistance before and after using LOR

technique 115

Table 310 DC and RF characteristics for XMBE304 118

Table 311 3rd

Gen Mask process step 119

Table 312 Standard deviation at two different voltages 124

Table 41 The parameter values used in this simulation 134

Table 42 The calculated values of bandgap at different temperatures 140

Table 43 The calculated effective masses for each temperature used in this simulation

141

Table 44 Epitaxial layer profile for Schottky Barrier Diode (XMBE104) 145

Table 51 Performance of the ASPAT diode obtained from measured IV characteristics

in this work 153

Table 52 Device to device uniformity check for large ASPAT diode 159

Table 53 The variation of reflection coefficient (S11) for GaAsAlAs ASPAT diode at

four different frequencies[117] 159

7

Table 54 Wafer to wafer uniformity check on XMBE304A and XMBE304B 161

Table 55 Comparison between calculated (fully Depleted) and extracted (different

biases) values from equivalent circuit parameters for different ASPAT mesa sizes at zero

bias voltage 170

Table 61 A summary of all the important parameters of the 4x4 microm2 diode 185

Table 62 Noise Equivalent Power (NEP) for zero bias detector diode 190

Table 63 Summary of characteristic of interest for GaAsAlAs ASPAT detector 198

Table 64 Comparison 4times4microm2 GaAsAlAs ASPAT diode to the-state-of-the-art zero

bias detector at W-band (75GHz-110GHz) 199

Table 71 Comparison parameter obtained from this simulation to the state-of-the-art

multiplier diode 213

8

LIST OF FIGURES

Figure 21 III-V compound semiconductors mobility and band gap[24] 31 Figure 22 illustration of Homojunctions band structure material before (left) and after

(right) equilibrium 34 Figure 23 Energy band diagrams before (left) and after (after) interface combination at

equilibrium 35

Figure 24 Lattice Matching for both materials when aL=aS 37 Figure 25 Energy gap and lattice constant for direct and indirect band gap of compound

semiconductor at ambient temperature[40] 37 Figure 26 Lattice mismatched material 39

Figure 27 The formation of Pseudomorphic-layer in two situations (a) compressive and

(b) tensile strain [1] 40 Figure 28 The band diagram of an ideal quantum well formed by a narrow band gap

semiconductor eg InGaAs sandwiched between two large band gap semiconductors eg

GaAs (a) Structure (b) energy band diagram and (c) Conduction band diagram when

AlGaAs is n-doped[43] 41 Figure 29 Metal and semiconductor in two conditions (a) separation (b) in contact 43

Figure 210 Energy band diagram of Schottky contact on n-type material under (a)

reverse and (b) forward bias 45 Figure 211 Figure 10 N-type semiconductor Ohmic contact with different doping

concentration ND (a) Low (b) Intermediate and (c) high 47 Figure 212 Classical view of whether an electron is can surmount a barrier or not

Quantum mechanical view allows an electron to tunnel through a barrier The probability

(blue) is related to the barrier thickness 51

Figure 213 Rectangular tunnelling with incident reflected and transmitted wave

function[70] 52

Figure 214 Epitaxial layer structure of device XMBE 304 (lateral structure) used in

this study 55 Figure 215 The layer profile of ASPAT with 28nm barrier thickness [27] 57

Figure 216 Conduction band diagram showing band bending and 2DEG formation at

the L1 spacer 60 Figure 217 I-V characteristics of a fabricated ASPAT diode 63 Figure 218 Intrinsic Elements of the Asymmetric Spacer Tunnel Layer (ASPAT) Diode

63 Figure 219 Cross-sectional view of ASPAT vertical structure the dimensions of d b

and h are not drawn to scale 65 Figure 220 Cross-section view of the complete process ASPAT diode (XMBE304)

The dimensions are not drawn to scale 67 Figure 221 A simple TLM structure with effective length and sheet resistance

underneath 69 Figure 222 Top view of TLM ladder structure use in this work 71 Figure 223 Typical plot of resistance versus TLM spacing 71

Figure 224 The set of DC measurement apparatus arrangement to measure the TLM 73

9

Figure 225 The set of RF measurement for on-wafer DC and RF one-port measurement

75 Figure 226 Actual VNA system that was used for RF characterization 75 Figure 227 Block diagram of the ASPAT measurement step 77

Figure 31 3D illustration of Optical lithography process used in this research 85 Figure 32 Actual picture of thermal evaporator used in this study 89 Figure 33 Single layer lift-off process using negative photoresist 91 Figure 34 Current-Voltage characteristic of sample XMBE368 used in this study at

two different locations on the wafer tile 96

Figure 35 ASPAT mask design for (a) 100times100microm2 (b) 30times30microm

2 and (c) 15times15 microm

2

diode dimensions designed in the 1st Gen Mask 96

Figure 36 Air Bridge design (red circle) for large mesa area (100times100microm2) device with

option for doped substrates 100 Figure 37 The layout of 1

st design of Dielectric Bridge (green circle) mask design for

100 times 100microm2 emitter size with option for doped substrate processing 100

Figure 38 Dry Etching for the first run in this study 102 Figure 39 Severe undercut of 2times2 microm

2 and 6times6 microm

2 devices 103

Figure 310 Device with emitter area of 2times2 microm2 and 6times6 microm

2 after dry etch and wet

etched 104 Figure 311 SEM Images of the GaAs sample 104

Figure 312 The cross-sectional view of InGaAs (left) and GaAs (right) samples used in

this study 105

Figure 313 SEM images taken from the XMBE304 samples and a GaAs test sample

105

Figure 314 SEM images for InP and InGaAs taken from [56] 105 Figure 315 Short circuit behaviour on one of the fabricated device in this run 106

Figure 316 The surface of the sample after final processing 108 Figure 317 IV characteristics of sample XMBE304 for emitter sizes of 1000 microm2

2500 microm2 900 microm2 400 microm2 225 microm2 100 microm2 and 36 microm2 109

Figure 318 Optical images after MESA etching (a) 1 microm Tolerance (b) 05 microm

Tolerance 110

Figure 319 After lift-off processing 111 Figure 320 The current density of XMBE304 processed using Si3N4 as dielectric layer

112

Figure 321 side view of lateral ASPAT structure 113

Figure 322 The measured size of the emitter area and the length D (blue color marked)

114 Figure 323 Summary of LOR technique steps 115 Figure 324 IV characteristic in 3rd run (reduced d-Gap) for each diode size on the mask

116

Figure 325 3rd

Gen Mask layout showing actual devices de-embedding TLM and

alignment mark structures used in this study 118 Figure 326 Step by step device fabrication (in 3D and cross-sectional view) for semi-

insulating substrate device type used in this study 121 Figure 327 Example finished process device with bond pad using 3

rd Gen mask 121

Figure 328 XMBE304 TLM measurement for the top contact after annealing 122

10

Figure 329 XMBE304 TLM measurement for the bottom contact after annealing 122 Figure 330 XMBE304 GaAsAlAs ASPAT measured IV characteristics at room

temperature for 4times4microm2 mesa size 123

Figure 331 XMBE304 GaAsAlAs ASPAT measured IV characteristics at room

temperature for 6times6microm2 mesa size 124

Figure 332 XMBE304 GaAsAlAs ASPAT measured IV characteristics at room

temperature for 10times10microm2 mesa size 124

Figure 41 SILVACO Atlas simulation process flow 128 Figure 42 The GaAsAlAs ASPAT structures produced from simulation based on the

diode multilayer heterostructures on the right 131 Figure 43 The Energy-momentum diagram for AlAs heterojunction semiconductor 131 Figure 44 (a) The conduction band and the valence band of the ASPAT diode structure

(b) the energy band diagram of the ASPAT diode structure when under three different

biases 132 Figure 45 The current-voltage characteristic of an ASPAT diode (100times100microm

2) and

(4times4microm2) using SILVACO Atlas simulator for structure device XMBE304 showing

excellent agreement between simulated and experimental data 134

Figure 46 IV characteristics of the dependencies of current on AlAs barrier 136 Figure 47 Example of analysis at -1 and 1V to the current 136 Figure 48 I-V characteristic of the dependencies current to Spacer I layer 137

Figure 49 Current changes with layer thickness l1 138 Figure 410 IV characteristic of the dependencies current to Spacer 1 layer 139

Figure 411Current change with layer thickness l2 139

Figure 412 Measurement and simulation comparison result as a function of temperature

range from 100K to 398K 142 Figure 413 Log Current vs voltage as a function of temperature for ASPAT sample

XMBE304 143

Figure 414 The effective barrier high of the Schottky barrier diode and the ASPAT

Diode [3] 144

Figure 415 Log Current vs voltage as a function of temperature for SBD sample

XMBE104 145

Figure 416 Influence of temperature on IV characteristic for both ASPAT diode and

SBD 146

Figure 51 IV Characteristics of measured ASPAT diodes for emitter sizes of 4x4um2

6x6um2 and 10x10um

2 Note the good scalability 149

Figure 52 Junction resistance versus voltage 151

Figure 53 The rectifying action of the diode at zero voltage which gives the ASPAT

high sensitivity near zero bias detection 152

Figure 54 One port S-parameter measurements 155

Figure 55 Fabricated open short de-embedding structure and real device (zoomed) use

for RF calibration and measurements (Note Images are not to scale) 156

11

Figure 56 Real S11 RF measurements for (XMBE304A) were done for four devices

from 15times15microm2 to 100times100microm2 device sizes for within-wafer uniformity check 158

Figure 57 Imaginary S11 measurements (XMBE304A) to qualify the repeatability and

reproducibility were done on four devices from 15times15 microm2 to 100times100 microm2 at zero

bias 158

Figure 58 Wafer to wafer uniformity test on 1st and 2

nd RF measurement(XMBE304B)

to qualify the repeatability and reproducibility were done for 30times30microm2 to 15times15 microm

2

device sizes (Real and Imaginary) Note blue colour is XMBE304A and red colour is

XMBE304B 160

Figure 59 RF measurement performed on device sizes of 10times10microm2 6times6mmicro

2 and

4times4microm2

(Real and Imaginary) Note that green red and blue colour represents 4times4microm2

6times6mmicro2 and 10times10microm

2 diodes respectively 162

Figure 510 RF measurement performed on device sizes of 10times10 microm2 6times6 microm2 and

4times4 microm2 (Smith Chart) Note that green red and blue colour represents 4times4microm2

6times6mmicro2 and 10times10microm2 diodes respectively 162

Figure 511 Cross-sectional of fabricated ASPAT diode and the corresponding

equivalent circuit model 164

Figure 512 The S-parameter Touchstone file is used to read the measured files 165

Figure 513 Equivalent Circuit Model of Open CPW or de-embedded structure 166

Figure 514 Equivalent circuit model for short de-embedded structure 166

Figure 515 Smith chart representative S-parameter measurement for short (left) and

open (right) CPW The blue lines represent simulated data and the red is measured data

167

Figure 516 Equivalent circuit of the ASPAT diode 169

Figure 517 Excellent fitting between measured (red colour) and simulated (blue colour)

for S11 (Real and Imaginary) results for various small device designs 169

Figure 518 Excellent fitting between measured (red colour) and simulated (blue colour)

for S11 results (Smith Chart) for various small device designs 170

Figure 519 Capacitance extraction showing full depletion at -025 volts (modelled

capacitance vs Voltage) 172

Figure 61 Block diagram represent a complete direct receiver system 177

Figure 62 The detection process of a single wave through a non-linear IV characteristic

of a diode 177

Figure 63 Lumped element illustration of microwave detector circuit 178

Figure 64 The mixing process where the signals are processed by the non-linear I-V

characteristic to produce an intermediate frequency (fIF=fLO-fRF) when signals at fLO and

fRF are applied to the diode 179

Figure 65 Measurement of Tangential Sensitivity[108 129] 181

12

Figure 66 The configuration circuit to verify 10th order polynomial equation extracted

from MATLAB to realize a virtual GaAsAlAs ASPAT diode 186

Figure 67 Verification of actual (blue measured) and virtual (red_10th order

polynomial) I-V characteristic of the 4times4 microm2 diode used in this study 186

Figure 68 Direct detector circuit topology using an ASPAT diode 187

Figure 69 Output voltage and detector sensitivity over wide range of input power 188

Figure 610 Simulated dependence of Voltage Sensitivity on incident power and load

resistance of the ASPAT detector 189

Figure 611 Junction resistance as a function of forward voltage 189

Figure 612 Curvature coefficient for the measured I-V characteristic with a device size

of 4times4μm2 191

Figure 613 Voltage Sensitivity against frequency at -20dBm fixed input power 192

Figure 614 Reflection Coefficient versus operating frequency without matching

circuitry 193

Figure 615 Detector circuit with impedance matching circuit placed in between diode

and source 194

Figure 616 Reflection Coefficient over wide frequency band with matching 195

Figure 617 Simulated voltage sensitivity of GaAsAlAs ASPAT detector at W-band

frequency 195

Figure 618 Lowest detectable signal at 100GHz operating frequency 196

Figure 619 The dynamic range of the ASPAT diode obtained from linear region of

diode operation 197

Figure 620 Simulation result of diode sensitivity for three devices geometries obtained

from the fabricated ASPAT in this work 198

Figure 71 performance of state-of the-art millimetre wave source [166] 202

Figure 72 Frequency multiplier has the property that foutput is an integer multiple of fin

204

Figure 73 Principle of operation for frequency multiplier utilising a non-linear

resistance [10] 204

Figure 74 A standard system for two port frequency multiplier circuit 207

Figure 75 Block diagram of frequency doubler with series ASPAT mounted diode 209

Figure 76 Empirical modelling of ASPAT doubler utilising Keysight ADS tool 210

Figure 77 Conversion loss and conversion efficiency as a function of input power 211

Figure 78 Conversion loss and Conversion efficiency as a function of output frequency

212

13

LIST OF PUBLICATIONS CONFERENCE PRESENTATIONS

PUBLICATIONS

1 MRR Abdullah Y K Wang J Sexton M Missous and M J Kelly ldquoGaAsAlAs

Tunnelling Structure Temperature Dependence of ASPAT Detectorsrdquo 8th UK-Europe-

China Workshop on mm-waves and THz Technologies 2015 Cardiff University IEEE

proceedings DOI 101109UCMMT20157460591

2 Yuekun Wang Mohd Rashid Redza Abdullah James Sexton and M Missous

ldquoInGaAs-AlAs asymmetric space layer tunnel (ASPAT) diodes for THz electronicsrdquo 8th

UK-Europe-China Workshop on mm-waves and THz Technologies 2015 Cardiff

University IEEE proceedings DOI 101109UCMMT20157460589

3 K N Zainul Ariffin S G Muttlak M Abdullah M R R Abdullah Y Wang and M

Missous ldquoAsymmetric Spacer Layer Tunnel In018Ga082AsAlAs (ASPAT) Diode using

Double Quantum Wells for Dual Functions Detection and Oscillationrdquo 8th UK-

Europe-China Workshop on mm-waves and THz Technologies 2015 Cardiff

University IEEE proceedings Doi 101109UCMMT20157460599

4 K N Zainul Ariffin M R R Abdullah Y K Wang S G Muttlak O S

Abdulwahid J Sexton MJ Kelly and M Missous ldquoAsymmetric Spacer Layer Tunnel

Diode (ASPAT) Quantum Structure Design Linked to Current-Voltage Characteristics

A Physical Simulation Studyrdquo UK-China Millimetre Waves and Terahertz Technology

Workshop September 2017 Submitted 14 July 2017 Conference held on 11th -13th

September 2017 DOI 101109UCMMT20178068358

5 K N Zainul Ariffin Y Wang M R R Abdullah S G Muttlak Omar S

Abdulwahid J Sexton Ka Wa Ian Michael J Kelly and M Missous ldquoInvestigations of

Asymmetric Spacer Tunnel Layer (ASPAT) Diode for High-Frequency Applicationsrdquo

DOI 101109TED20172777803

6 Omar S Abdulwahid S G Muttlak M R R Abdullah K N Zainul Ariffin J

Sexton Ka Wa Ian Michael J Kelly and M Missous ldquoA 100GHz Zero-Biased

Quantum Tunnelling ASPAT Detectorrdquo Submitted to IEEE TED on DEC 2016 under

correctionamendment Pending fabrication data

14

CONFERENCE PRESENTATIONS

1 Mohd Rashid Redza Abdullah J Sexton Kawa Ian MJKelly and M Missousldquo

G2040GHz Frequency Doubler Varistor Mode using ASPAT diodesrdquo UK

Semiconductors 2017 2017 University of Sheffield Oral presentation

2 M R R Abdullah YueKun Wang J Sexton Kawa Ian and M Missousldquo

Microwave Performance of GaAsAlAs Asymmetric Spacer Layer Tunnel (ASPAT)

Diodesrdquo UK Semiconductors 2016 2016 University of Sheffield Oral presentation

3 M R R Abdullah J Sexton and M Missousldquo GaAsAlAs Tunnelling Structures

THz RTD oscillators and ASPAT detectorsrdquo UK Semiconductors 2015 2015

University of Sheffield Oral presentation

4 Yuekun Wang Mohd Rashid Redza Abdullah and M MissousldquoInGaAs-AlAs

asymmetric space layer tunnel (ASPAT) diodes for THz electronicsrdquo UK

Semiconductors 2015 2015 University of Sheffield Oral presentation

5 Mohd Rashid Redza Abdullah and M Missousldquo GaAsAlAs Tunnelling

Structure Temperature Dependence of ASPAT Detectorsrdquo PGR Conference2016

2016 University of Manchester Poster presentation

6 YueKun Wang KNZainul Ariffin Mohd Rashid Redza Abdullah J Sexton

Kawa Ian and M Missous ldquoPhysical Modelling and Experimental Studies of

InGaAsAlAs Asymmetric spacer Layer Tunnel Diodesrdquo UK Semiconductors 2016

2016 University of Sheffield Oral presentation

7 K N Zainul Ariffin S G Muttlak M R R Abdullah Y Wang Omar S

Abdulwahid M Missous ldquoExperimental and Physical Modelling of Temperature

Dependence of a Double Quantum Well In018Ga082AsAlAs ASPAT Dioderdquo UK

Semiconductor Conference 2016 Sheffield 6 ndash 7 July 2016 Oral Presentation

8 Omar S Abdulwahid Mohd Rashid Redza Abdullah S G Muttlak K N Zainul

Ariffin Mohamed Missous ldquoTunnelling Barrier Diode for Millimetre Wave

Mixingrdquo UK Semiconductor Conference 2016 Sheffield 6 ndash 7 July 2016 Oral

Presentation

9 M Abdullah K N Zainul Ariffin MRR Abdullah J Sexton M Missous and

MJ Kelly ldquoA Novel In18Ga82As-AlAs Asymmetric Spacer Layer Tunnel (ASPAT)

Diode with Double Quantum Wells for Microwave Detectionrdquo UK Semiconductor

Conference 2015 Sheffield 1 ndash 2 July 2015 Oral Presentation

15

ABSTRACT

Thesis Title GaAsAlAs ASPAT Diodes for Millimetre and Sub-Millimetre Wave

Applications

Institute School of Electrical and Electronic Engineering the University of Manchester

Candidate Mohd Rashid Redza bin Abdullah

Degree Doctor of Philosophy (PhD)

Date 3 October 2017

The Asymmetric Spacer layer Tunnel (ASPAT) diode is a new diode invented in

the early 90s as an alternative to the Schottky barrier diode (SBD) technology for

microwave detector applications due to its highly stable temperature characteristics The

ASPAT features a strong non-linear I-V characteristic as a result of tunnelling through a

thin barrier which enables RF detection at zero bias from microwaves up to

submillimetre wave frequencies In this work two heavily doped GaAs contact layer on

top and bottom layers adjacent to lightly doped GaAs intermediate layers enclose

undoped GaAs spacers with different lengths sandwiching an undoped AlAs layer that

acts as a tunnel barrier The ultimate ambition of this work was to develop a MMIC

detector as well as a frequency source based on optimized ASPAT diodes for millimetre

wave (100GHz) applications The effect of material parameter and dimensions on the

ASPAT source performances was described using an empirical model for the first time

Since this is a new device keys challenges in this work were to improve DC and

RF characteristic as well as to develop a repeatable reproducible and ultimately

manufacturable fabrication process flow This was investigated using two approaches

namely air-bridge and dielectric-bridge fabrication process flows Through this work it

was found that the GaAsAlAs heterostructures ASPAT diode are more amenable to the

dielectric-bridge technique as large-scale fabrication of mesa area up to 4times4microm2 with

device yields exceeding 80 routinely produced The fabrication of the ASPAT using i-

line optical lithography which has the capability to reduce emitter area to 4times4microm2 to

lower down the device capacitance for millimetre wave application has been made

feasible in this work The former challenge was extensively studied through materials

and structural characterisations by a SILVACO physical modelling and confirmed by

comparison with experimental data The I-V characteristic of the fabricated ASPAT

demonstrated outstanding scalability demonstrating robust processing A fair

comparison has been made between ASPAT and SBD fabricated in-house indicating

ASPAT is extremely stable to the temperature The RF characterisations were carried out

with the aid of Keysight ADS software

The DC characteristic from fabricated GaAsAlAs ASPAT diodes were absorbed

into an ADS simulation tool and utilized to demonstrate the performance of MMIC

100GHz detector as well as 20GHz40GHz signal generators Zero bias ASPAT with

mesa area of 4times4microm2 with video resistance of 90KΩ junction capacitance of 23fF and

curvature coefficient of 23V-1

has demonstrated detector voltage sensitivity above

2000VW while the signal source conversion loss and conversion efficiency are 28dB

and 03 respectively An estimate noise equivalent power (NEP) for this particular

device is 188pWHz12

16

DECLARATION AND COPYRIGHT STATEMENT

No portion of the work referred to in the dissertation has been submitted in support of an

application for another degree or qualification of this or any other university or other

institute of learning

COPYRIGHT STATEMENT

i The author of this thesis (including any appendices andor schedules to this thesis) owns

certain copyright or related rights in it (the ldquoCopyrightrdquo) and he has given The University of

Manchester certain rights to use such Copyright including for administrative purposes

ii Copies of this thesis either in full or in extracts and whether in hard or electronic copy

may be made only in accordance with the Copyright Designs and Patents Act 1988 (as

amended) and regulations issued under it or where appropriate in accordance with licensing

agreements which the University has from time to time This page must form part of any

such copies made

iii The ownership of certain Copyright patents designs trademarks and other intellectual

property (the ldquoIntellectual Propertyrdquo) and any reproductions of copyright works in the thesis

for example graphs and tables (ldquoReproductionsrdquo) which may be described in this thesis

may not be owned by the author and may be owned by third parties Such Intellectual

Property and Reproductions cannot and must not be made available for use without the prior

written permission of the owner(s) of the relevant Intellectual Property andor

Reproductions

iv Further information on the conditions under which disclosure publication and

commercialisation of this thesis the Copyright and any Intellectual Property andor

Reproductions described in it may take place is available in the University IP Policy

(httpdocumentsmanchesteracukDocuInfoaspxDocID=487) in any relevant Thesis

restriction declarations deposited in the University Library The University Libraryrsquos

regulations (httpwwwmanchesteracuklibraryaboutusregulations) and in The

Universityrsquos policy on Presentation of Theses

17

ACKNOWLEDGEMENTS

First and foremost all gratefulness and praise is to Allah swt for everything in my

life He is the one and the only one who granted me knowledge health patience and

ability to complete this thesis as well as colouring the whole journey of my PhD

I give my deepest and sincere gratitude to my PhD supervisor Professor Mohamed

Missous for his time support patience and guidance throughout the journey of this PhD

studies His encouragements valuable advice precious ideas and a wealth of knowledge

amp experiences have had a direct inspiration on this research Special thanks also to our

experimental officer Dr James Sexton for not only sharing his knowledge advice and

semiconductor fabrication skills but also his effort in maintaining our clean room

facilities to a great level My gratitude also extends to Mr Mallachi McGowan for his

help and assist in the lab-related issue

I am also obligated to Prof MJ Kelly from University of Cambridge and Dr Kawa

Ian from ICS limited for their measurement of the ASPAT samples on realizing the RF

characteristics This collaboration effort can hopefully last longer in designing and

implementing the ASPAT MMIC detectors

My deepest appreciation also goes to my PhD colleagues Khairul Nabilah Saad

GMuttlak Omar AbdulWahid and Yuekun Wang for their support as well as working

together with me to realize this exciting project directly and indirectly A sincere

thankfulness similarly to my seniors Dr Md Adzhar Zawawi and Dr Fauzi Packeer for

their support during the first and second year of my research For other friends and staff

members under Prof Missous and Dr M Migliorato I will always remember the strong

bond and friendship we made

I am really fortunate that I been blessed with my motherrsquos care who always make doarsquo

for my success every day during my studies As for my beloved wife Dr Nik Maryam

Anisah Nik Mursquotasim who had always encouraged me supported me and gave me

patience through all the hardship in this journey thank you very much

Finally I also would like to thank and acknowledge my sponsor Majlis Amanah

Rakyat (MARA) for financially supporting me during this studies I am greatly indebted

with your kind support which was vital to my study

18

DEDICATION

This thesis dedicated to

My respected and beloved parentshellip

My loving wife dearest siblings and in-lawshellip

19

1 INTRODUCTION

11 Background

It is an undeniable fact that semiconductors have changed the world much further

than anything people could have predicted in the last 60 or 70 years ie after the lsquocats

whisker and vacuum tube eras This field of research has been expanding from year to

year starting from the discovery of the first semiconductor (silver sulfide) in 1833 by

Michael Faraday [1 2] and it still remains very active to the present Semiconductors

have a large range of applications and are not just limited to use in communications they

can be found everywhere in other applications from Earth to space The widespread

usage and sheer number of applications have led to it growing very quickly and

contributing greatly to the growth of World Economics Over time the successful

development of semiconductor growth techniques such as Molecular Beam Epitaxy

(MBE) has enabled researchers to tailor and precisely control the semiconductor

material for new electronic devices with extra functionalities This has led to the

development of advanced devices such as high electron mobility transistors (HEMTs)

and Heterojunction bipolar transistors (HBTs) for use in wireless communication

technology Given this development today electronic devices such as computers

handheld smartphone tablets etc are no longer perceived as luxury and attractive items

but rather have become crucial in everyday life Such devices provide the means to allow

for people to remain connected to each other via the sending and receiving of

information electronically The huge demand for such types of devices has resulted in

competition in both the electronic market and technologies which only goes on to

advance the semiconductor industry

Nowadays the demand for electronic devices characterised by high speed high

efficiency ultra-low power and low manufacturing cost has increased exponentially To

fulfil this growth in demand high data rate systems are required in other words the

system must work at a higher frequency for both the transmitter and receiver The

frequency of interest for advanced wireless communication is in the Millimetre and sub-

20

millimetre wave region which is around 30-300GHz and 300 - 3000GHz respectively

The second frequency region is also sometime known as the terahertz (THz)

electromagnetic region This band lies between the microwave and infrared frequency

bands From the first time it was revealed in the late 80s[3-5] the THz region has gained

a lot of international attention due to its unique properties and since then the motivation

to develop these devices has increased significantly To date the THz frequencies region

has shown its ability to fulfil various applications such as high-resolution imaging in

medical security and surveillance field atmospheric monitoring and environment radio

astronomy as well as compact range radars[3] to name a few

However despite these developments not much effort has been made in exploring

alternative compact THz devices As a result electronic THz devices are still in the state

of immaturity as compared to microwave and photonics devices This is due to their high

cost and absence of compact amp solid-state THz sources (oscillators) and receivers

(detectors) that are capable of operating at both room and extreme temperatures[6 7] A

great deal of work still continues to fill up the lsquoTHz gaprsquo (between 300GHz and 3THz)

used for the most important part of a communication system namely the front-end

receiver or first stage Such a system is responsible for receiving detecting and

processing the received signal to be translated into useful information Furthermore THz

receivers systems still require the best-integrated components such as source mixer and

detector to reach their complete competencies[8] The detector which remains the

critical part of the receiver system requires devices or components that are able to fulfil

the THz gap requirement Studies conducted over a number of years have found out that

the key element in improving THz detection relies upon the use of passive devices ie

diodes Based on these findings many types of diode ie tunnelling diode point-contact

diode and Schottky barrier diode (SBD) have been proposed for detection applications

Amongst microwave and millimetre wave detector diode devices the Schottky

Barrier Diode (SBD) is the dominant detector that has been used since the 1940s[9] The

reason for this dominance is the ease of fabrication of a SBD (by either a point-contact

or evaporated semiconductor-metal structure) and its ability to produce a non-linear

current-voltage (I-V) characteristic which is necessary for rectifyingdetecting diodes [9

10] SBDs also have high cut-off frequency good dynamic range and are low cost To

21

date the SBD has been able to detect signals up to 100GHz [11] 1THz[12] and as high

as 10 THz[13] However the current transport mechanism in a SBD relies on thermionic

emission and therefore is strongly dependent on temperature and means that using them

in extreme conditions ie military and automotive applications is complex The SBD

also suffers from high noise figure[14] and is susceptible to burnout at a modest pulse

power level this will limit the use of ultra-high frequencies and low power signal

applications Other diodes that share the same characteristics are Planar Doped Barrier

(PDB) Germanium Backward Diode (GBD) ie a type of Esaki tunnel diode These

diodes are well known and are reliably used as millimetre wave detectors However it

still proves inefficient to substitute the SDB with any of the previously mentioned

diodes This is due to some drawbacks such as strong temperature dependence limited

dynamic range fabrication complications and hence poor reproducibility (ie GBD) and

other circuit complexities

Hence there is strong compulsion to study examine and produce new detector

diode structures that are able to solve the mentioned diodes limitations and which have

high sensitivity larger dynamic range low noise strong independence to temperature as

well as being able to work efficiently in the high-frequency band and at zero bias The

advantages of working at zero-bias relates very much to the need for a system with less

power consumption so that the device (ie mobile communication) is able to run off

small batteries for a reasonable length of time eliminating extra biasing circuit as well as

noise Therefore a new tunnelling device namely the Asymmetrical Spacer Layer

Tunnel diode (ASPAT) developed by RT Syme [15 16] and refined by Missous et

al[17] has been examined in this work The ASPAT which is in essence a

Semiconductor-insulator-semiconductor structure relies on tunnelling through a barrier

to provide current compared to conventional thermionic emission in SBDs The ASPAT

diode has many advantages a zero bias turn-on voltage very weak sensitivity to changes

in temperature (due to tunnelling) very low noise large dynamic range high resistance

to pulse burn-out [18] and as demonstrated recently can be reproducibly

manufactured[17] The growing interest in THz frequencies nowadays makes the

ASPAT an excellent choice to fulfil all requirements for ultra-high speed applications

22

ie communication (mobile computer networking) radar (military equipment) scalar

analyser and built-in test equipment

In this work an ASPAT diode based on group III-V elements of the periodic table

comprising compound semiconductors of large band gap material Aluminium Arsenide

(AlAs) sandwiched between two lower bandgap Gallium Arsenide (GaAs) are used and

intensively examined The AlAs semiconductor which is ten-monolayer thick has

almost the same lattice constant as GaAs but has a larger bandgap Consequently in the

conduction band a thin barrier of the AlAs is formed from the arrangement of such

structure The structure is made up of GaAs and AlAs both materials are grown on

GaAs substrate using Solid Source Molecular Beam Epitaxy (SSMBE) Therefore in

this study the ASPAT diode will be referred to as ldquoGaAsAlAs ASPATrdquo diode The

conventional GaAsAlAs ASPAT diode has been developed and successfully fabricated

in two different stages This work was the first carried out using facilities provided by

the University of Manchester The first stage of the work was to qualify the

reproducibility and repeatability of growth and fabrication technique which is mostly

performed on larger emitteranode size The second was to develop conventional

ASPAT diodes that can perform at Millimetre and sub-millimetre wave frequencies and

which are comprised of small emitter area

Prior to this work full physical modelling using SILVACO design software was

undertaken to generate models and to fully characterise and identify the fundamental

physical phenomenon of multi-junction ASPAT diode Therefore insight into and

performance based on diode structure and electron movement can be understood and

predicted which lead to the crucial idea in helping and advising iterations to epitaxial

growth as well as diode fabrication The verification of the physical models must be set

as a priority goal by comparing the results of statistically fabricated measured data The

advantage of physical modelling is that it can help reduce materials resources cost and

fabrication time

Further research into the field has led to the development of two other types of

ASPAT diodes that are used to compare with the conventional GaAsAlAs ASPAT

diode Their configuration involved the use of a more advanced semiconductor

technology which comprises InGaAsAlAs materials and GaAsAlAs with InGaAs

23

quantum wells The latter was a novel ASPAT diode and the former is identified as

advanced ASPAT diode However these two advanced ASPAT diodes have not been

extensively studied in this thesis as they will be covered by other co-workers at

Manchester Hence due to these some important parameters are compared to the

conventional one as it is the main focus of this work In the case of temperature

dependent studies the DC characteristic of conventional ASPAT is compared to in-

house fabrication AuGaAs SBD All the ASPATs epitaxial layer materials structures are

shown in the following tables

Table 11 Sample XMBE304 XMBE307 and XMBE368 GaAsAlAs materials structure grown

on GaAs Substrates by MBE

Thickness (Aring)

Layer Material Doping (cm-3

) XMBE368 XMBE304 XMBE307

Top Ohmic1 GaAs (Si) 4times1018

~3000 ~3000 ~3000

Buffer 1 GaAs(Si) 4times1017

~50 ~400 ~50

Spacer1 GaAs NID 50 50 50

Barrier AlAs NID 28 28 28

Spacer 2 GaAs NID 1000 2000 1000

Buffer GaAs(Si) 4times1017

50 400 50

Bottom Ohmic GaAs(Si) 5times1018

~3500 ~4500 ~3500

Substrate GaAs (Si) 50000 50000 50000

Note that sample XMBE368 and XMBE304 are grown on doped GaAs

substrates Sample XMBE368 was grown un-rotated to study the effect of barrier

thickness variation

24

Table 12 Quantum wells sandwiching the quantum barrier for sample XMBE314 grown on a

GaAs substrate by MBE

XMBE314

Layer Material Doping (119836119846minus120785) Thickness (Aring) Bandgap (eV)

Ohmic Layer GaAs (Si) 4times1018

3000 142

Emitter GaAs (Si) 1times1017

400 142

Spacer GaAs Undoped 50 142

Quantum Well In18Ga82As Undoped 60 116

Barrier AlAs Undoped 28 283

Quantum Well In18Ga82As Undoped 60 116

Spacer GaAs Undoped 2000 142

Collector GaAs (Si) 1times1017

400 142

Ohmic Layer GaAs (Si) 4times1018

4500 142

Substrate GaAs

Table 13 Novel InGaAsAlAs ASPAT diode grown on InP substrate by MBE on InP substrate

XMBE326

Layer Material Doping (cm-3

) Thickness (Aring) Bandgap (eV)

Top Ohmic1 In053Ga047As(Si) 5times1019

3000 075

Buffer 1 In053Ga047As(Si) 1times1017

350 075

Spacer1 In053Ga047As NID 50 073

Barrier AlAs NID 283 283

Spacer 2 In053Ga047As NID 2000 075

Buffer In053Ga047As(Si) 1times1017

350 075

Bottom Ohmic In053Ga047As(Si) 15times1019

4200 075

Substrate InP (Si) NID 620000

From the above tables it can be noted that XMBE307 is the first batch that was grown

in-house using a Riber V100HU MBE machine followed by XMBE368 XMBE 304

XMBE314 and finally XMBE326 The two earlier batches were grown on n+

substrate hence their fabrication process flow is simpler On the other hand the three

other batches were grown on semi insulating substrate thus requiring the development

of new repeatable reproducible and robust process flow which will be covered in this

thesis

25

In general the fabrication of the ASPAT diode is based on top-down processes this

is because the ASPAT is a vertical structure device and the junction capacitance of the

ASPAT is directly associated with the size of the anodeemitter area Further to these

the capacitance directly influences the diode cut-off frequency Therefore the simplest

way to reduce the capacitance is by reducing the lateral area of the device of the ASPAT

structure since it can be represented by a parallel plate capacitor where the capacitance

is inversely proportional to the area of the device In order to achieve high cut-off

frequencies minimising capacitance via small dimensions ie sub-micrometre level is

essential However this will also increase the series resistance of the diode As a result

the cut-off frequency will be degraded Thus there is a trade-off between small

dimension of device and high cut-off frequency to be achieved

Finally successful growth and fabrication for small area GaAsAlAs ASPAT diode

in this work has led to carefully extracted RF characteristics This becomes a stepping

stone to designing a millimetre wave integrated circuit (MMIC) detector using empirical

modelling in Keysight ADS tools Therefore a predicted performance for a 4times4 um2

fabricated ASPAT is that can operate at 100GHz ASPAT as a zero-bias detector with a

voltage sensitivity of over 2000VW Additionally the design of a millimetre wave

source using similar ASPAT diodes was also carried out The performance of a 2040

GHz doubler using GaAsAlAs ASPAT in varistor mode is demonstrated for the first

time with a conversion loss of 33dB and conversion efficiency of ~ 02

26

12 Aims and objectives

The aim of this study is to further improve the performance of microwave and

millimetre wave technology by incorporating the Asymmetrical Spacer Layer Tunnel

Diode (ASPAT) for ultimate operation near THz frequencies by designing a range of

low power high-speed devices enhancing the methods of Simulation layout and

materials amp structural characterisations with fabrication process optimization using the

facilities available at the University of Manchester

There are three main objectives in this research firstly to streamline the physical

device design and modelling using the GaAsAlAsGaAs materials systems in order to

produce a zero bias detector which is basically a rectifier of a microwave signal by

using the SILVACO Atlas simulation tools

Secondly to achieve reproducibility and manufacturability of the fabrication

process for new type of GaAsAlAs ASPAT structure (lateral structure) hence small size

ASPAT emitter by improving the device processing technique and maximising the

capability limit of the conventional i-line optical lithography that is available in Prof

Missousrsquos group laboratory

Thirdly to optimise DC parameters through electrical properties investigation as a

stepping stone to the next objective that is to characterise the RF performance of the

GaAsAlAs ASPAT detector circuit The detection properties of microwave and

millimetre wave diode will also be investigated with different ASPAT diode size at

100GHz Further to these the properties of microwave signal source will also be

developed by way of utilizing the non-linearity feature of the diode Therefore this new

type of tunnelling diode can be applied to both applications of signal detection and

signal source in the microwave and millimetre wave ranges

27

13 Outline of this Thesis

This thesis is organized into eight chapters The first chapter discusses the

contextual information that motivates the undertaking of the study An overview of the

work which includes the details of the studied samples the aim and objectives of the

whole research project are also outlined in this chapter

Chapter 2 deals with the literature review of the basic principles and concepts of

the group III-V compound semiconductors The historical background of such

semiconductors which is essential to the development of ternary structures etc and the

advancement of semiconductor materials engineering is presented The types of existing

tunneling diode as well as conventional microwave diodes are also discussed and

compared The fundamentals of ASPAT diode which includes structural parameters and

its operation are then explained in detail Finally discussions of the ASPAT key DC

characteristics which are important for detection purposes are presented

Chapter 3 focusses on the development of the experimental techniques which can

be divided into two stages In the first stage the development is towards repeatability

reproducibility and manufacturability of the ASPAT grown in-house by MBE and

fabricated by conventional i-line optical lithography The second stage involves

optimisation and fine tuning such fabrication method for GaAsAlAs ASPAT samples

that can operate at high frequency ie 100GHz detector For both stages of the

fabrication process all techniques including mask design generic and special process

flow are presented The chapter ends with discussions on issues related to sample

processing and improvements that are proposed to solve these issues

Chapter 4 dwells on the modelling of the GaAsAlAs ASPAT using the SILVACO

simulation package The discussions are expected to offer a better understanding or

insight into each layer that forms the ASPAT diode structure The chapter begins with

discussions of the operation of the SILVACO Atlas tool A validation of physical

modelling is essential and presented according to the fabricated mesa sizes of the diode

28

Thereafter towards the end of the chapter the analyses of the relationship between

device current-voltage (I-V) characteristics the structural parameter including various

temperatures dependent simulations with a comparison to an in-house fabricated SBD

are offered

Chapter 5 presents relevant DC results based on optimized fabrication process and

RF characterization which enable obtaining an intrinsic and extrinsic element of the

GaAsAlAs ASPAT diode The discussions also highlight the analysis of DC zero bias

equivalent circuit and de-embedding extraction using ADS The chapter ends with

discussions on the RF reproducibility performance which includes the performance as

well for millimeter-wave and sub-millimeter wave applications

Chapter 6 discusses the main applications of ASPAT diodes The chapter begins

with discussions on detection theory followed by the parameters of interest and ends

with circuit design as well as the performance of a 100GHz detector The circuit design

was conducted using Keysight ADS software via harmonics balance simulation tool The

performance in term of sensitivity depending on measured ASPAT emitter size is

demonstrated Finally a comparison with conventional Schottky diode is presented

towards the end of the chapter

Chapter 7 discusses a secondary application that can be applied to the ASPAT by

utilizing the nonlinearity feature of the diode to create a signal source namely a 20 to

40GHz frequency doubler in varistor mode The doubler performance of ASPAT will be

explored through circuit design constructed via Keysight ADS simulation software Each

key parameter is highlighted and discussed in detail

The final chapter of this thesis that is Chapter 8 discusses the conclusions of the

study with emphasis on the overall key research findings The chapter also highlights

suggestions for further research in this particular field of study

29

2 LITERATURE REVIEW

21 Introduction

Since 1940s the development in the technology of semiconductor electronics has

been expanding and now has led to the establishment one of the most astonishing

industries of the 3rd

-millennium era Leading this advancement is the integrated circuit

(IC) or chip which was driven mostly by silicon (Si) Overtime the IC has undergone

substantial revolution in term of power economics size and efficient energy

consumption Currently it covers every aspect of human life ie from desktop personal

computers in the office and house to the compact smartphone in the pocket and from a

gigantic satellite in space to small satellite navigation in cars In other words

semiconductor technology is crucial to human life Without developments in

semiconductor materials engineering and shrinking of device size such accomplishment

may not have been realised today Therefore this chapter presents a macro view of the

development in compound semiconductor technology especially in radio frequency (RF)

towards Millimetre and submillimetre wave applications with regard to the improvement

of material and device structures

The essentials of group III-V compound semiconductor will be emphasised for its

points of interest and application in this field (RF technology) This chapter comprises

five main sections The first section is an overview of the semiconductor history with

concentration on its advantages and applications in the RF field while the second and

third discuss the effects of III-V compounds when the interface occurs between

semiconductor-semiconductor and semiconductor-metal respectively which leads to a

basic understanding of hetero-structures device as well as contacts namely Schottky and

Ohmic The fourth section is predominantly concerned with high-speed devices ie

diodes and materials in this field which leads to the exploitation of the main researchrsquos

device Then the following section describes in detail the background works basic

principle and intrinsic amp extrinsic parameters of the Asymmetric spacer Tunnel Diode

(ASPAT) Finally the basic way of characterising the device will also include giving an

overview of how the device is measured and what parameters are needed

30

22 Historical review of III-V Compound Semiconductor for RF applications

The beginning of commercial electronic devices was marked with the first point-

contact semiconductor transistor developed in 1947 by William Shockley at Bell

Laboratories in New Jersey Shockley developed a device based on a Germanium

Bipolar Junction Transistor (Ge BJT) structure [19] with operating frequency above 1

GHz Since then and until early 1950s the development of Ge BJT was fast and it

became foremost in the market of semiconductor technology However the emergence

of Silicon (Si) challenged Ge in the market in the 1960s Si has the upper hand primarily

because it has better electron transport and low manufacturing costs compared to Ge

[19] By the 1970s almost all RF transistors were based on Si BJT Additionally the

development of Si which forms a new material from the formation of native oxide

namely Silicon Dioxide (SiO2) led to the invention of the Metal Oxide Semiconductor

Field Effect Transistor (MOSFET) [19] The future of digital electronic industries has

been ldquobrightrdquo ever since the MOSFET was ldquobornrdquo as it has become a fundamental

building block component in complex microprocessors and flash memories Despite this

development the exploitation of Si at RF frequencies did not last long since Si is not an

optimum semiconductor for RF electronic devices The emergence of GaAs has

improved RF applications for high-speed transistors

Ge and Si which are categorised as single element semiconductor are the earliest

materials used to build the first transistor devices These devices played a crucial role

towards the development of more advanced material such as GaAs of the compound

semiconductor type[20] A compound semiconductor is a semiconductor formed by the

ionic bond of different types of semiconductor material most widely known as the group

III-V compound semiconductors The main reason for the progression of the III-V

compound semiconductors is due to their better electron mobility compared to the single

element semiconductors The term ldquomobilityrdquo in the semiconductor industry refers to the

easiness of movement of charges in many directions inside a crystal In fact it is

determined by the access resistances values with saturated velocity under certain values

of electric fields (bias) the higher the electric field the faster is the carrier movement in

the crystal Figure 21 shows the electron mobility and band gap for the most common

31

group III-V compound semiconductors Besides higher mobility III-V compound

semiconductors also have light-emission capability and are suitable for bandgap-

engineering techniques

The work on III-V compound semiconductors mainly on GaAs FETs led to a new

change for the whole RF electronics industry For example in 1966 the first GaAs

MESFET was invented[21] and achieved a maximum operating frequency of operation

of 3GHz [22] Three years later the frequency increased to 30GHz [23]

Figure 21 III-V compound semiconductors mobility and band gap[24]

With better features in terms of having a higher electron mobility compared to Si

electronic devices based on III-V materials developed rapidly This attracted attention in

many aspects especially in military radar application electronic warfare system missile

guidance control electronic for smart warfare system and secure communication To be

specific those demands were fulfilled through the application of microwave mixer and

detectors [25] which were achieved based on Schottky barrier diodes and FETs

However these applications remained largely as niche markets for use only in military

and exotic scientific projects until 1980 In addition to the microwave industry two

important diodes that played a large role in very high-frequency power source namely

the Gunn diode and the Impact Avalanche and Transit Time (IMPATT) diode which

were discovered in the 1960s[26]

100

1000

10000

100000

0 05 1 15

Bu

lk M

ob

ility

(cm

2 V

-1 s

-1

BandGap (eV)

InS

b

InA

s

Ge

Ga

Sb

In

GaA

s

Si

GaA

s

InP

32

Furthermore the invention of Molecular Beam Epitaxy (MBE) growth technique at

the beginning of 1970s has enhanced the full potential of the III-V compound

semiconductors[27] This technique has led to the formation of a new class of materials

and heterojunction device with high-quality interfaces and accurate control of the

thickness during growth[28] The advancement of material engineering that tailored the

III-V compound semiconductor with MBE effect has been beneficial for both three-

terminal and two-terminal devices As a result of this more advanced devices in both

electronics and optics were developed such as quantum well (QW) laser Resonant

tunnelling diode (RTD) high electron mobility transistor (HEMT) and many more[29]

The aim was to achieve high-speed devices transporting data at high data rates and

robust devices These devices promised an excellent option to conventional transistor

(three terminal devices) in high-frequency systems especially in the terahertz (THz) or

Millimetre and sub-millimetre wave regions [30]

One of the promising diodes that received a lot of attention is the resonant

tunnelling diode (RTD) which was first described in 1974 by Chang [31] This device

which consists of a double barrier and one quantum well is the classical tunnelling diode

Due to its good symmetrical non-linearity in its current- voltage characteristic it can be

exploited for signal generation and detection However the main focus of RTD to date

has been in the generation of continuous wave (CW) ultra-high frequency and to a lesser

extent in detection Therefore other tunnelling based diodes were developed specifically

for detection purposes which are the main foci of this work The PDB and ASPAT

diodes are the workhorse candidates for detection purposes Most of these are built based

on group III-V compound semiconductors [32]

Unlike the Very Large Scale Integration (VLSI) market ie CMOS for personal

computer (PCs) the RF electronic device for civilian application reached the consumer

market only in the late 1980s through satellite television with operating frequencies

around 12GHz [19] Since then many RF application have been deployed on the mass

market depending on their operating frequency such as 09GHz ndash 25GHz for wireless

communication 20GHz to 30GHz for satellite communication 77GHz for car radar

systems and above 90GHz for different sensor applications Utilising GaAs as the main

material RF devices have become the key underpinning components for modern

33

communication systems As a result in 1998 the volume production of mobile phones

was greater than that of PCs for the first time in history Presently production is being

made for devices like smartphones cellular phones mobile internet access and new

communication services and tablets

The development of the RF field is never ending More and more improvements are

being made especially through the design and fabrication of oscillators and detectors

which are mainly built based on group III-V compound semiconductors When RF

devices were used by the military (in the 1970s to 1980s) cost was not a concern

However after getting into civil application market (ie in the 1990s) the most frequent

issues highlighted were performance and cost[19]

The ability to generate or receive high operating frequencies with high power large

bandwidth and high sensitivity is an indicator for a good performance of RF devices

(depending on specific applications) For example the highest room temperature based

oscillator of up to 186 THz was achieved in thin well AlAs-InGaAs RTD by Professor

Masahiro Asada from Tokyo Institute of Technology [33] An excellent review on THz

sources can be found in [34] For ultra-high frequency detector and mixer applications

the two terminals RF device that is mostly used is the SBD In 1996 the highest cut off

frequency achieved by a mixer utilizing the SBD was about 5THz[35] and this has kept

increasing ever since The factor that motivates the development of THz devices is the

requirement to have a compact coherent source in the THz range Undoubtedly in the

future there will be very exciting times for enthusiasts of terahertz sources and receiver

as new generations of compact broadband and tuneable solid source device based on

advanced compound semiconductor are developed

23 The Concept of Heterostructures

A III-V compound semiconductor is mostly grown on a single semiconductor

substrate forming a layer called epitaxial heterojunction layer It is a starting point and

the key feature that brings the idea of realising the most advanced semiconductor

devices currently being developed and manufactured by combining several epitaxial

semiconductors [36-38] Heterojunctions have the capability of manipulating carrier

transport ie electron and holes transport in crystal separately unlike homojunctions

34

This has resulted in the successful development of new devices for high-speed and high-

frequency applications as well as optical sources and detectors [37] This section will

discuss lattice matched material pseudomorphic material hetero-junction band

discontinuities and quantum wells

231 Homojunctions Heterojunctions and Band Discontinuities

The term homo-junction refers to the interface between identical semiconductor

materials that have different polarity ie p-type or n-type but similar in energy gap This

phenomenon is usually applied in forming p-n junction diodes and can be understood by

referring to Figure 22 below

Figure 22 illustration of Homojunctions band structure material before (left) and after (right)

equilibrium

The materials A and B which have similar bandgap (Eg) and different dopant

types ie p-type and n-type will have their Fermi levels (Ef) closer to the valence band

(EV) and conduction band (EC) respectively before ldquothermal equilibriumrdquo Once

equilibrium is achieved Ef of both p-type and n-type will be aligned causing band

bending of EC and EV As a result a built-in electric field is introduced (via diffusion of

carriers) for both holes and electrons and forcing them to move in one direction

On the contrary a heterojunction occurs when the interface between two

semiconductor materials with different bandgap energy are brought together (ie large

energy band gap material combined with a low band gap one eg wide band gap AlAs

and narrow bandgap GaAs) This results in a steep band bending which leads to the

formation of energy band discontinuities at the junction as shown in Figure 23 In a

semiconductor heterojunction the most important parameter is the band gap energy

EC

EV

Ef

Eg

p-type E

g

n-type

EC

EV

Ef

Material A

p-type

n-type

Material B Material A Material B

35

associated with each material in the structure where the degree of discontinuity can be

utilised in varying the carriers transport properties as well as the quality of the junction

depending on the interest of the designer This leads to flexibility in tailoring device

characteristics leading to vastly improved performance of the device

Figure 23 Energy band diagrams before (left) and after (after) interface combination at

equilibrium

Based on Figure 23 above Material A indicated with blue line is a large band

gap energy material and Material B highlighted with the red line is a low band gap

material EV represents the valence band EC the conduction band and Ef is the Fermi

level of the materials Alternatively the two materials band discontinuities are denoted

by ΔEC for the conduction band and ΔEV for the valence band χ and Eg represent the

electron affinity and band gap energy respectively

At some point where by the Fermi energy of both semiconductor materials are levelled

the structure would have reached its thermal equilibrium The band gap of materials A

and B have a discontinuity at the interface (ΔEg) of these two materials In general this

is given by

120549119864119892 = 119864119892119860 minus 119864119892

119861 (21)

Furthermore when thermal equilibrium is achieved ΔEg is then divided between

conduction band and valence band discontinuities (ΔEC and ΔEV respectively) at the

material A and B junction interfaces Their relationships can be expressed as

EC

A

EV

A

Ef

A

Eg

A

Material A Material B

χA

χ

B

ΔEC

ΔEV

Ef

B

E

g

B

EC

B

EV

B

EV

A

χA

ΔE

C

ΔEV

EC

A

Ef

EC

B

EV

B

χB

Material A Material B

Vacuum

Level Vacuum

Level

36

∆119864119862 = 120594119860 minus 120594119861 (22)

∆119864119881 = (119864119892119861 minus 119864119892

119860) minus (120594119860 minus 120594119861) (23)

∆119864119892 = 119864119892119860 minus 119864119892

119861 = ∆119864119862 + ∆119864119881 (24)

However these relationships which were introduced by Anderson can only offer an

approximation In practice the results are always different since dislocation and

interface strain occur at the junction Therefore precise control during epitaxial growth is

always required and growth technologies such as MBE are employed In due course the

band gap discontinuity can be further exploited by using different types of material

combination Examples are GaAsAl052Ga048As and In053Ga047As In052Al048As [39]

232 Lattice-Matched and Pseudomorphic Materials

As discussed earlier a heterojunction happens when any two different

semiconductor materials that have different bandgap are joined together At the atomic

level both materials often differ in lattice constant The easiest way to explain this is by

setting the formation of heterojunction which can be separated into two types lattice

matched and lattice mismatched (pseudomorphic)

2321 Lattice Matched Systems

To create discontinuities for use as a high-performance device the combination of

semiconductor materials is essential Selecting the appropriate materials that have

similar or very close lattice constants to combine is crucial to avoid disruption at the

atomic level heterojunction interface Figure 24 shows that the lattice constant of a

material A ie substrate (aS) and material B ie deposited over layer (aL) are identical

or very close and their surface atoms are perfectly matched This scenario is known as

lattice matching

37

Figure 24 Lattice Matching for both materials when aL=aS

As can be seen in Figure 25 while there are restricted binary materials available to

form good heterojunction interfaces it is possible to combine semiconductor materials in

binary ternary and quaternary forms to allow the formation of a variety of lattice-

matched heterojunction interfaces The examples of materials that have successfully

been alloyed are In053Ga047As In052Al048AsInP and GaAsAlxGa(1-x)As (x=0 to 1)

Even though the materials system hetero-junction of these materials has close lattice

constant value their band-gap will experience an abrupt variation

Figure 25 Energy gap and lattice constant for direct and indirect band gap of compound

semiconductor at ambient temperature[40]

Material A(layer)

Material B(substrate)

aL

aL

aS

aS

38

The existence of binary ternary and quaternary semiconductors formed by alloying

semiconductors has expanded the opportunity for heterojunction formation in devices

The alloy semiconductor which is produced by the combinations of two semiconductors

A and B has a lattice constant that obeys Vegardrsquos Law as follows

119886(119886119897119897119900119910) = 119909119886119860 + (1 minus 119909)119886(119861) (25)

For the alloy the band gap normally follows the virtual crystal approximation

119864119892(119886119897119897119900119910) = 119909119864119892(119860) + (1 minus 119909)119864119892(119861) (26)

Table 21 shows the list of the semiconductor alloy band gap and lattice constant for

common binary and ternary for group III-V compound semiconductors[41]

Table 21 Lattice constant (a) and band gap for group III-V binary and ternary compound

semiconductors a room temperature [41 42]

Alloy Lattice constant a(Aring) Band gap EgeV)

AlAs 5661 2239

AlSb 6136 1581

GaAs 5653 1424

GaN 3189 34

InAs 6058 0417

InP 5869 1344

Al052Ga048As 5657 2072

In053Ga047As 5868 0773

In052Al048As 5852 1543

39

2322 Pseudo-morphic Materials

The other scenario is when two different materials with different lattice constants

are brought into contact The observation can be made at the atomic level where the

atom will try to match each other as shown in Figure 26 below

Figure 26 Lattice mismatched material

In fact for both situations (ie lattice matched and lattice mismatched) the atom of

the material at the hetero-interface will change their position to maintain the geometry of

the lattice Due to distortion at this atomic level a strain is then induced at the hetero-

interface In order to form a good hetero-junction interface the strain must not exceed a

certain specific critical value which will cause crystal dislocations to occur The result of

crystal dislocation is generally bad as it will affect the carriers which will be

concentrated in the defect area thus degrading the carriersrsquo mobility This then makes

the overall function or performance of the device to become poor

Nowadays the Molecular Beam Epitaxial (MBE) technique is able to grow epitaxial

layers of mismatched semiconductor layers profile ie mismatched in their lattice

constant (aLneaS) The growth method works when the grown epitaxial layer assumes the

lattice parameters of the layer it is deposited on Nonetheless the layers must be kept

within a certain limit and the deposited layer must be thin enough to avoid defect or

dislocation formations This new layer known as a ldquopseudomorphicrdquo material will alter

its original crystal structure and physical properties

Material A(layer)

Material B(substrate)

aL

aL

aS

aS

40

Figure 27 The formation of Pseudomorphic-layer in two situations (a) compressive and (b) tensile

strain [1]

Figure 27 shows material A in which the pseudomorphic materials can be related to

two situations compressive and tensile strain The compressive strain occurs when the

deposited layer has a larger lattice constant than the substrate (aL gtaS) while tensile strain

happens when the deposited layer has a smaller lattice constant than the substrate

(aLltaS) These leads to aL either to compress or stretch to fit aS respectively Note that

the pseudomorphic layers can only be grown to a certain critical thickness hc From

Figure 27 the strain between the substrate and the deposited epitaxial layer is given by

휀 =119886119871 minus 119886119878

119886119878 (27)

Where Ɛ is strain between two layers aL is lattice constant of the deposited layer and

aS is lattice constant of the substrate layer The concern in deposition of the over layer is

to avoid dislocation occurring at the interface if there is too much strain at the junction

The strain is naturally influenced by the thickness of the deposited layer and thus the

thickness of growth must be controlled below the critical thickness hc which is

expresses as

(b)

41

ℎ119888 =119886119904

(28)

Moreover one needs to appreciate that even though the crystal structure and their

physical properties change the total energy within the unit cell is maintained This is

possible by distortion of the deposited layer in the direction perpendicular to the growth

direction while leading to lattice matching in the lateral plane Example of lattice

matched materials is GaAsAlAs and pseudomorphic material is In08Ga02AsInP

233 Quantum well and 2DEG

A typical application of heterojunction interface is one in which utilises ΔEC and ΔEV to

form barriers for electrons and holes One example of barriers that confines these

carriers is known as a Quantum Well (QW) A QW is a layered semiconductor usually

very thin ie about ~ 100 Aring thicknesses in which many quantum mechanical effects can

occur It is formed by a thin layer of a low bandgap energy semiconductor material eg

GaAs sandwiched between two similar large bandgap energy semiconductors eg AlAs

or AlGaAs The growth technique to achieve thin layers of QW is usually MBE The

benefit of this method is that it allows the formation of heterojunction with very thin

epitaxial layer

Figure 28 The band diagram of an ideal quantum well formed by a narrow band gap

semiconductor eg InGaAs sandwiched between two large band gap semiconductors eg GaAs (a)

Structure (b) energy band diagram and (c) Conduction band diagram when AlGaAs is n-doped[43]

42

The thickness of the layer that can be achieved can be as thin as the electron mean free

path (De Broglie wavelength) which is around 100 Aring to 300 Aring [44] The expression for

the De Broglie wavelength is given by

120582 = ℎ120588frasl (29)

Here h and ρ are Planckrsquos constant and momentum of the electron respectively Figure

28(b) illustrates a quantum well formation in abrupt semiconductor interfaces It can be

observed that the heterojunction boundary will experience discontinuities at the edges of

the conduction band and valence band with a quantum well generated for the carriers

(both electron and holes) The quantised energy sub-bands in the quantum well structure

in Figure 28(b) can be determined from [43]

119864 = 119864119899 + (

ℏ2

2119898lowast) (119896119909

2 + 1198961199102)

(210)

Where 119864119899 = (ℏ21205872

2119898lowast ) (119899

119871)2

and n is the energy level index that can be n=1 2 3hellip

The dopants in a semiconductor with large band gap layers may supply the

carrier to the quantum well and this occurs when the base or bottom of the quantum well

is lower than the Fermi Level and hence the high energy donors will go down to the

well therefore creating a Two-Dimensional Electron Gas (2DEG) In the 2DEG the

electrons and holes move freely in the quantum well in the plane perpendicular to the

growth direction however they are not capable of moving in the crystal growth

direction (confinement direction)[45 46] The 2DEG phenomena can be seen in Figure

28(c)

24 Metal-Semiconductor Contact

A semiconductor device is incomplete if there is no connection between the

semiconductors and the outside world A metal which is usually gold (Au) or gold

germanium (AuGe) is diffused into the semiconductor to allow for electrical connection

from the outside world to the semiconductor and vice versa The metal-semiconductor

contact can be either a Schottky contact or an Ohmic contact The Schottky contact is a

43

rectifying contact while the Ohmic contact is a contact that provides a low resistance

path between semiconductor and metal

241 Schottky Contact

The Schottky contact is basically a metal contact to the gate to enter a region or

channel in a transistor Figure 29 shows a schematic band diagram of a metal-

semiconductor contact before and after contact (Schottky-Mott concept)[47]

Figure 29 Metal and semiconductor in two conditions (a) separation (b) in contact

In Figure 29 the work function of the metal is represented by qm while the

semiconductor work function is qS The qχ is the energy difference of an electron

between the vacuum level and conduction band edge ie known as the electron affinity

and qVn is the difference between the conduction band and Fermi level in the

semiconductor EV EC Ef is the valence band energy conduction band energy and the

Fermi level respectively

The metal and semiconductor are brought together as showed in Figure 29(a)

both materials are at steady state However when the metal and semiconductors are in

contact as illustrated in Figure 29(b) the electrons that flow from the conduction band

in the semiconductor into the lower energy state of the metal will cause the Fermi level

to be aligned in thermal equilibrium Due to this process the positive charge donor is

trapped in the semiconductor interface hence forming a depletion region Xdep

Thereafter the upward bending of the energy in the semiconductor takes place On the

qχ(s)

Eg(s)

Vacuum Level

EV

EF(m)

E

F

EC

qϕ(m)

qϕ(s)

Metal Semiconductor

qVn

qϕB

qϕ(s)

qVbi

X

qχ(s)

Eg(s)

Vacuum Level

EV

EF

EC

qϕ(m)

Metal Semiconductor

Xdep

(a) (b)

44

other hand the negative charge (electron) will be accumulated within a narrow region in

the metal interface The existence of two different charges at the metal-semiconductor

boundary generates an electric field This leads to a potential barrier qB as seen by

electrons in the metal moving into the semiconductor and a built-in potential qVbi as

seen by electrons in the semiconductor trying to move into the metal

The built-in potential qVbi is defined as follows

119902119881119887119894 = 119902empty119861 minus 119902119881119899 (211)

The barrier height empty119861 in the ideal case is specified by the dissimilarity between a metal

work function empty119898 and electron affinity of the semiconductor

119902empty119861 = 119902empty119898 minus 119902120594 (212)

Referring to Eq (212) above the barrier height empty119861 rises linearly with the metalrsquos work

function empty119898 Nevertheless this is only in theory as the presence of localised surface

stated at the edges causes empty119861 to become unresponsive to the metal work function

Consequently Eq(211) is then reordered to match the difference in metal and

semiconductor work function Thus the new equation becomes

119902119881119887119894

= 119902(120601119898 minus 120601119904) (213)

A Schottky contact appears when a metal-semiconductor contact has a large

barrier height (B ge kT) and low doping concentration in the semiconductor (ND le NC) In

the case when the metal-semiconductor contact is under some bias eg reverse bias the

semiconductor will react to a positive bias according to the metal by a voltage V=-VR

This condition will affect the built-in potential and leads to increase from Vbi to

(Vbi+VR) thus increasing the barrier height empty119861 in the semiconductor as well

Consequently electrons are less able to flow from the semiconductor and cross into the

metal Therefore the current flow will be very small This phenomena is shown in

Figure 210(a)

45

Figure 210 Energy band diagram of Schottky contact on n-type material under (a) reverse and (b)

forward bias

As can be seen from Figure 210(b) when a forward bias is applied the semiconductor

is biased negatively with respect to the metal by a voltage V=Vf This will result in a

reduction in built-in potential from Vbi to Vbi-Vf The electrons in the semiconductor will

lower the barrier height and a lot of electrons will escape into the metal causing a large

current to flow Thus a large current flow in the forward direction compared to the

reverse direction Essentially this is the origin why the Schottky contact is named a

rectifying contact [48 49] For the metal side both forward and reverse biases applied

do not affect the barrier high empty119861 because there is no voltage drop there

In this system the electron and holes are transported by a phenomenon called

Thermionic Emission (TE) which happens when the semiconductor layer is lightly

doped Nd lt 1x1017119888119898minus3 The electron will only be thermionically emitted into the metal

when the energy is higher than the potential barrier[50] There is another phenomenon

called Thermionic Field Emission (TFE) which happens when the potential barrier

thickness is very thin (thin enough) to allow the electron to tunnel through the barrier

This will be discussed in the next section as this phenomenon leads to the formation of

an ohmic contact

(a) (b)

46

242 Ohmic Contact

Basically an ohmic contactrsquos purpose is to provide a low resistance path from

the semiconductor to the outside world It is different to a Schottky contact as it is a non-

rectifying contact and does not control the current flow the I-V characteristic of an

ohmic contact is linear in both forward and reverse directions (equality in current flow)

The ohmic contact also has a small voltage drop across it compared to the voltage drop

across the device

If a metal and semiconductor are bought together unavoidably a Schottky

contact will be formed Therefore to create an ohmic contact some techniques to reduce

barrier height and width of the depletion region must be used ie increase Nd In carrier

transport theory there are three mechanisms of carrier transport across the barrier

Firstly the Thermionic Emission (TE) which happens when the carries are excited to

overcome the barrier when the thermal energy is present Secondly the Thermionic

Field Emission (TFE) occurs when the electronholes have enough energy to tunnel

through an adequately thin barrier and some has overcome the low barrier at the top

Finally the Field Emission (FE) which results when carriers can tunnel through the

entire barrier The FE is the most favoured mechanism in the ohmic contact approach

[51]

From the three mechanisms above the current can be determined by the following

equations

1) Current in Thermionic emission (Figure 211(a))

exp (empty119861)

119896119879

(214)

2) Current in Thermionic field emission (Figure 211(b))

exp [

(empty119861)

11986400119888119900119905ℎ11986400

119896119879

] (215)

3) Current in Field Emission (Figure 211(c))

exp (empty119861)

11986400

(216)

47

Where k is the Boltzmann constant T is the temperature 11986400 is the tunnelling parameter

and is related to the doping concentration radic119873119863 The barrier height is denoted by empty119861

Figure 211 Figure 10 N-type semiconductor Ohmic contact with different doping concentration

ND (a) Low (b) Intermediate and (c) high

Figure 211 shows that the carrier transport mechanism is varied by the doping

concentration (ND) As can be seen from Figure 211(c) the doping concentration here is

the highest and influences the depletion region width Xdep to become smaller Therefore

Field Emission (FE) becomes dominant This FE method is the favourite method for

ohmic contact formation [51] and will be utilised in the fabrication carried out in this

work

In fabricating practical devices the ohmic contact is often split into two types

alloyed and Non-Alloyed The difference between the two is that the alloyed type is used

when the semiconductor is doped with a low doping concentration ie less than

1x1018119888119898minus3 while the Non-Alloyed is designed for heavily doped semiconductors with

more than1 times 1019119888119898minus3 doping

The alloyed ohmic contact requires thermal annealing to have a good performance

for electron transport In multi-layer metals one of the metals has the role of donor or

acceptor which is used to increase the doping concentration of the semiconductor If a

temperature anneals eg 420˚ Celsius is applied the metal will diffuse into the

semiconductor and carry the dopant into the semiconductor Therefore a heavily doped

region will be formed and the depletion width becomes narrow establishing the ohmic

contact The key example of this is the usage of the Gold-Germanium-Nickel (Au-Ge-

Ni) alloy where the Ge is the n-type dopant [52] which diffuses into the semiconductor

(a) (b) (c)

48

and perform atom replacement in the semiconductor ie in GaAs it replaces Ga On the

other hand the Non-Alloyed does not require any thermal annealing as it already has a

very high doping concentration and will automatically reduce the depletion region width

The Non-Alloyed ohmic contact has some advantages such as reproducible contact

reduced processing time and good uniformity [53]

25 Asymmetrical Spacer-layer Tunnel (ASPAT) diode Background work

In this section some historical background of the Asymmetrical Spacer-layer

Tunnel (ASPAT) diode is given Since this is the first thesis reporting about this new

device it is worth to mention some historical background about this tunnelling diode

The ASPAT was first proposed by a group of scientists from General Electrical

Company (GEC) in 1990[16] The works led by Richard T Syme and assisted by

Michael JKelly Angus Condie and Ian Dale initiated the idea of launching a new type

of tunnel diode The idea managed to attract the interest of many parties following the

development of resonant tunnelling diode (RTD) which earlier had shown a promising

weak temperature dependence [54] However the interest in RTD is mainly limited to

microwavesub-millimetre wave generation For THz detection the requirement is to

have a significantly asymmetric IV characteristic Given this the ASPAT which has

only a single energy barrier and most importantly weak temperature dependence and

large dynamic range would be a promising candidate for this application

The development of ASPAT is a kind of reverse engineering since it was built

purposely to replace the earliest receiver diode especially the Schottky Barrier Diode

(SBD) which has strong dependence on operating temperature [55] From the time when

it was first revealed a lot of works have been done to realise this most sophisticated

tunnel diode The first attempt which was reported in [56] was meant to gather some

insights into the device by using the well-known Schrodinger and Poissonrsquos equations

for simulation The second attempt on the other hand was directed to physically grow

and fabricate the device Here the real problem occurs At the first stage of qualifying

this device it was found not to be manufacturable Since then a new tunnel diode

structure based on GaAsAlAs materials system was built by both MBE and MOCVD

Its microwave performance was then tested at 94GHz [18] The same paper also

49

reported performance comparisons between ASPAT and another microwave diode ie

Germanium Backward Diode (GBD) PDB and SBD

Work on these devices stopped due to the inability to commercialise the ASPAT

and other tunnel based devices [57-60] The problems associated with low-cost

manufacture of tunnel diodes are due to firstly the thickness of the AlAs barrier layer

the dependence of tunnelling probability (electron) through a single barrier is

exponential and varies by a factor of more than 350 for one monolayer change in the

AlAs barrier thickness[61] The tunnelling of the electron through a barrier is

proportional to the current through a barrier as a function of a bias across the AlAs

barrier[62] Secondly the bandgap which is predominantly happens to be a ternary alloy

with relative composition ie AlxGa1-xAs Here the x can vary the bandgap in the

semiconductor layer For the ASPAT a 1 change in x results in a 30 change in the

current[62] To design an ASPAT for microwave and THz applications the designer

often allows at most plusmn10 variation of the absolute current through a specific diode at a

pre-identified bias This implies that within a wafer the uniformity that the ASPAT must

achieve is less than plusmn01 monolayers while between wafer to wafer the reproducibility

in barrier thickness in average must be identically controlled[63] This explains why at

the qualification stage of investigating ASPAT there was a need to focus on

GaAsAlAs-based material to diminish further errors because of the change in x This

type of work on ASPATs has been carried out by other co-workers at Manchester and

Cambridge

Thereafter the work then focused on repeatability and manufacturability tests

These result in many attempts being carried but failed with unacceptable between wafer

to wafer reproducibility [61 64] The development of reproducibility and repeatability of

the ASPAT was pursued for over 10 years until precise control of the growth of the

thickness AlAs layer was finally achieved using MBE[65] [66 67] This achievement is

confirmed by a current density produced which varied by less than plusmn30 indicating that

the reproducibility of AlAs barrier of the order of plusmn 02 monolayers A final step to

achieve the level control for the ASPATrsquos AlAs barrier thickness was carried out and

resulted in a 1 standard deviation of the IV characteristics for both within a wafer and

different wafer (2 inches wafer size)[17] In the early stage of this work some

50

repeatability and manufacturability test was also carried out Once this vital step is

accomplished further investigation on the ASPAT was made most recently and which

will be covered in this thesis ie temperature independence[68] and new ASPAT

types[69] characterisation to achieve smaller device RF measurement and development

of THz detectors The material systems that have been investigated so far are

GaAsAlAs and InGaAsAlAs both in the ManchesterCambridge group

Recently the ASPAT was commercialised by Linwave Technology as a wideband

zero-bias detector diode This was done in April 2016 Although it is now on the market

the ASPAT remains immature in term of research and development A lot of work is still

required to enhance the device ie working at the sub-millimetre wave using ternary

material etc

26 Asymmetrical Spacer-layer Tunnel (ASPAT) diode The Basics

The basic building block of the ASPAT diode is based on heterojunction of three

multilayer semiconductor structures which have two different band gaps The structure

comprises a thin layer of wider-gap semiconductor sandwiched between two

semiconductors with narrower-gap forming a tunnel barrier The basic principle of the

ASPAT device is based on the exploitation of quantum mechanics theory using

heterojunctions interface According to quantum mechanics theory moving particles

(electrons) with less energy than the barrier height have a probability of appearing on the

other side of the barrier by a tunneling through it This can be achieved in conditions

where the barrier must be very thin (~ 10 monolayer ) This is in contrast to classical

physics where a particle must have kinetic energy at least slightly greater than the

potential barrier height in order to overcome the barrier otherwise the probability of the

particle to appear on the side of the barrier is zero

Since the ASPAT diode operation is based on tunnelling through a barrier one

needs to know that the tunnelling mechanisms can be classified into two types[44]

intraband and interband The latter is described as tunnelling that occurs from

conduction band to valence band (electron) and valence band to conduction band (holes)

This normally happens in bipolar device ie p-n junction diode which has n-type and p-

type doped regions On the other hand intraband refers to tunnelling which occurs when

51

electron tunnel from the conduction band of a semiconductor to the conduction band of

its neighbouring semiconductor The same thing happens to the holes in the valence

band The device with this type of tunnelling is normally a unipolar device which is

either p-type or n-type doped The ASPAT diode can be considered as a device that is

based on intraband tunnelling mechanism Therefore the focus will be entirely based on

its principles

261 Principle of Quantum Tunneling

Generally all tunneling diodes obey the concept of quantum mechanical

tunneling Quantum mechanical tunneling is a phenomenon where a particle is able to go

through an energy barrier higher than the kinetic energy of the particle and if it is thin

enough compared to the de Broglie electron wavelength (λ) If the electron wave is

greater than the barrier the probability of the wave to occur at both side of the barrier is

higher

Figure 212 Classical view of whether an electron is can surmount a barrier or not Quantum

mechanical view allows an electron to tunnel through a barrier The probability (blue) is related to

the barrier thickness

For the case of classical physics (Figure 212(a)) the particles can be confined by

energy barriers of a semiconductor if their kinetic energy is less than the barrier energy

The particles thus require higher kinetic to escape to other states this phenomenon is

called thermal emission In quantum mechanics (Figure 212(b)) the particle is

described in two ways as a wave and as a particle If the particle moves like a wave it

will carry all the waversquos properties Therefore it will not brusquely end up at the

En

erg

y

(a) Classical view (b) Quantum mechanical view

En

erg

y

En

erg

y

52

boundary of the energy barrier Hence when the particles collide with the barrier

(incidence) there will be a probability of penetrating the barrier if the barrier is thin

enough and has a finite height For thicker barrier the probability of a wave that can be

found on the other side of the barrier is very small However the possibility of the

electron wave to appear on the other side of the barrier is increased by thinning the

barrier The potential barrier of semiconductor material technology can be determined by

using Homojunctions structures with different doping profile This will result in a

difference in band alignment and multilayer heterojunction structure (different

semiconductors have different band gap) which includes semiconductors insulators and

conductors (metals)

The easiest way to explain the phenomenon is by considering a potential barrier

Epot(x) with barrier height E0 energy bigger than the total energy E as shown in Figure

213 the potential energy occurs in a finite space between 0ltxlta and is 0 outside

Figure 213 Rectangular tunnelling with incident reflected and transmitted wave function[70]

The electron outside the region of the potential barrier (xlt0 and xgta) is free to

move The effective mass of the electron is different inside and outside of the barrier in

real tunneling devices when implemented using semiconductor heterostructures The

quantum mechanical equations predict the wave nature of matter which states that matter

unveils wavelike properties under some conditions and particle-like properties under

other conditions The wavelike properties as described by the Schrodinger Formula of

E0 Transmitted Ψ = 119862119890minus119894119896119909

Incident Ψ = 119860119890119894119896119909

Reflected Ψ = 119861119890minus119894119896119909 Ψ = 119863119890minus119894119896119909

0 a

E

x

V(x

)

53

quantum mechanics represent a particle penetrating through a potential barrier most

likely as an evanescent wave coupling of electromagnetic waves[44]

To start the calculation of the tunnelling probability the Schrodinger equation is given

by

119894ℏ

120597

120597119905Ψ(119903 119905) = ΨΗ(r t)

(217)

Where ℏ is Planckrsquos constant (662606957 times 10119890minus341198982119896119892119904

2120587frasl ) Ψ(119903 119905)is the wave

function at position r and time t Η is the Hamiltonian operator given by

Η = minus

ℏ2

2119898nabla2 + 119881(119903 119905)

(218)

Where 119881(119903 119905)is the potential energy which is dependent on space and time 119881(119903 119905) is

considered zero for a particle traveling in free space without any potentials The plane

wave with vector r and t is given by

Ψ(119903 119905) = 119890119894(119896119903minus120596119905) (219)

This equation satisfies Eq (217) above under the condition where the particle is

travelling in free space without potential k is the wave vector which is equivalent

to2120587120582frasl and the angular frequency 120596 is 2120587 multiplied by the frequency

In the case of tunneling through a potential barrier the method of separation of

variables is used to simplify the problem as in the equation below

Ψ(119903 119905) = 119877(119903)119879(119905) (220)

It is assumed that the problem above is divided into time-dependent and time-

independent parts 119877(119903) is the spatial component and 119879(119905) is the time-based component

of the wave function The time dependent problem as shown above in the Schroumldinger

Equation (1) can easily be solved by filling up all the finite parameters The solution of

54

the time-independent part gives the tunneling probability For the one-dimensional (1D)

time-independent Schroumldinger equation[44]

119864120595(119909) = minus

ℏ2

2119898

1198892

1198891199092120595(119909) + 119881(119909)120595(119909)

(221)

E is the total energy and 120595(119909)is the spatial component of the wave function along the x

axis The combination to the wave function is given by

120595(119903 119905) = 120595(119903)119890minus(

119894119864119905ℏ

)

(222)

The time-independent plane wave solution of 120595 = 119890119894119896119909 satisfies the equation (221) for

any constant potential V0 in space Plugging in the wave solution yields the condition

that

119896 = radic2119898lowast(119864 minus 1198810)

ℏ2

(223)

Referring to Figure 213 also the barrier with exact rectangular shape with height E0 and

width W the solution of the wave functions and tunneling probability can be extracted

by using the below equation [44]

119879119905 =

[119862119890minus119894119896119909]2

[119860119890119894119896119909]2= [1 +

11986402 sin ℎ2 (119896119882)

4119864(1198640 minus 119864)]minus1 asymp

16119864(1198640 minus 119864)

11986402 exp(minus2radic

2119898lowast(1198640 minus 119864)

ℎ2119882)

(224)

For more complex barrier shape Wentzel-Kramers-Brillouin has simplified the

Schrodinger equation for tunneling probability of carrier which becomes[44]

119879119905 =

[119862119890minus119894119896119909]2

[119860119890119894119896119909]2asymp 119890minus2int |119896(119886)|119889119886

1198860 asymp 119890

minus2int radic2119898lowast

ℎ2 [119880(119886)minus119864]1198891198861198860

(225)

55

This means that the incident electron has a finite probability T of tunneling

through the potential barrier and this leads to the concept of tunneling probability as well

as a tunnelling current Therefore this becomes the basis of tunneling phenomena and

thus all devices which are related to tunneling can be modelled and analysed based on

this basic example The tunneling phenomenon is a majority-carrier effect and the

tunneling time is set by the quantum transition probability per unit time (which is on the

order of picoseconds) rather than the transit time concept [44 71] This enables the

tunneling devices to work at a much higher switching speed They can also be used in

high-frequency applications such as microwave circuit and high-speed oscillators

262 ASPAT Structural Parameters of GaAsAlAs materials System

Figure 214 Epitaxial layer structure of device XMBE 304 (lateral structure) used in this study

Conduction band profile

56

The core materials that make up the ASPAT diode in this work is based on

heterostructure of group III-V compound semiconductors Such materials are chosen due

their mature excellent properties and their band gap which can be tailored to fit the

desired design as well as to improve the carrier mobility In this work the primary layers

that form an ASPAT diode are very thin pure Aluminium Arsenic (AlAs) of thickness

ten monolayers buried in between dissimilar thickness of pure Gallium Arsenic (GaAs)

layers as can be seen in the red circle in Figure 214 above These two GaAs layers are

known as spacer layer which normally have a ratio of 401 or 201 in thickness The

asymmetrical spacers layer and the thin barrier in such arrangement lead to an

asymmetric current-voltage characteristics as proposed firstly by Syme and Kelly[15]

To examine and investigate this GaAsAlAs ASPAT structure in term of electrical and

RF characteristics the device have been grown according to Table 22 below

Table 22 epitaxial layer of sample XMBE304 which is the main focus of this work

Material Doping (cmminus3) Thickness (Aring) Bandgap (eV)

Ohmic Layer GaAs (Si) 4times1018 3000 142

Emitter GaAs (Si) 1times1017 400 142

Spacer GaAs Undoped 50 142

Barrier AlAs Undoped 28 283

Spacer GaAs Undoped 2000 142

Collector GaAs (Si) 1times1017 400 142

Ohmic Layer GaAs (Si) 4times1018 4500 142

Substrate GaAs - 650 microm 142

The arrangement of the multi-layers that form a lattice matched GaAsAlAs ASPAT

diode can be transformed into band structure profile view for easy understanding The

conduction band profile at equilibrium is as sketched in Figure 214 As can be seen in

the Figure 214 the ASPAT diode is generally a heterojunction multilayer structure

tunnelling diode

57

Figure 215 The layer profile of ASPAT with 28nm barrier thickness [27]

The generic structure of ASPAT diode which is shown in Figure 215 above with a

schematic band structure comprises the following (starting from the top)

(1) A thick layer of heavily doped n++

about 4e+18cm-3

of GaAs with a thickness of

approximately 300nm

(2) An intermediate layer of lightly doped n-type about 1e+17cm-3

of GaAs with

thickness of approximately 40nm

(3) A spacer layer not intentionally doped (NID) GaAs with thickness of approximately

5nm

(4) An ultra-thin layer of NID AlAs with thickness of approximately 28nm

(5) A spacer layer of NID GaAs with thickness of approximately 200nm

(6) An intermediate layer of lightly doped n+ ~1e+17

GaAs with thickness of

approximately 40nm

(7) A thick layer of heavily doped n+ about 4e+18 of GaAs with a thickness of

approximately 750nm

Each layer has its own role For instance layer (1) and (7) are used as ohmic

contacts via connection to a AuGeNiAu metal stack This explains why they are

purposely heavily doped (gt 1018

cm-3

) for better low resistance ohmic contacts Two

intermediate layers (layers 2 and 6) are used to prevent the carrier in the contact layers

from diffusing into the undoped layers The two unequal length spacer layers with ratio

58

1198971 1198972 of about 401 are used as voltage arms to yield an asymmetric current-voltage

characteristic The asymmetry means that after a positive bias is applied from the long

spacer region an accumulation layer is formed and it is deeper than that formed by the

negative bias The thin layer positioned in the middle (Layer 4) is the tunneling barrier

The performance of a single barrier ASPAT diode can be optimised depending on

the applications by appropriate selection of the material system so that the band gap and

barrier height of such material can be modified Furthermore the mobility of electron

and doping concentration of the contacts region can be tuned The parameters to tune

during the growth for instant growth interrupt time and growth temperature will also

affect the performance of this diode The key layers that will affect the performance are

the barrier thickness and the two spacer layers enclosing it The study has shown that a

one monolayer change in thickness results in 300 change in in tunneling current for a

fixed voltage point[65]

The following discussions account for the effect of the main structure of the ASPAT

which is related to their performance

2621 Barrier Thickness and height

The probability of an electron tunnelling through a barrier depends exponentially on

the width and height of the barrier as well as the energy that is incident on the barrier[72

73] All these will affect the I-V characteristic of the ASPAT diode The tunnel current is

obtained by summing over all incidents electrons energies with tunnelling probabilities

through the barrier The tunnelling varies approximately as[74]

119879 prop 119890minus120581119889 (226)

Where d is the barrier thickness and κ is defined by the expression below

120581 =

radic2119898lowast(1198810 minus 119864)2

ℏfrasl

(227)

From textbook the tunnelling probability is given by

119879(119864) = 412058121198702

[(1205812 + 1198702)2119904119894119899ℎ2119870119897 + 412058121198702]frasl (228)

Where K is expressed as

59

119870 = radic2119898lowast119864

ℏfrasl (229)

Where ℏ represents the reduced Planckrsquos constant (h2π) E and m are the electron

energy and effective mass respectively Thus by inputting appropriate value into these

equations one finds that reducing the barrier thickness by one monolayer increases the

tunnelling probability by a factor of nearly three for every electron that tunnels through

the barrier As a result the current will also increase Further it indicates that the

tunnelling strongly depends upon barrier thickness and height By contrast the current

does not strongly depend on temperature

2622 Spacer Thickness

The reason for having two dissimilar undoped spacer lengths is mainly to avoid

diffusion of dopant to the barrier and subsequent layers during growth but in the case of

the ASPAT diode the spacer can also act as a voltage arm Varying the thick spacer

layer (1198971) results in the reverse current decreasing as the layer thickness increases and

varying the thin spacer layer (1198972) will affect the forward current which increases as the

thickness reduces To obtain appropriate asymmetrical I-V characteristics one needs to

maintain an adequate ratio between these two spacer thicknesses While a too thin

1198971 results in high leakage current at reverse bias a too thick 1198972 results in low forward

current One also needs to keep it thick enough to prevent carrier diffusion

These two spacers must be kept undoped or very low doped to allow the electron

moving in the electron mean free path region as it is clear from ionised donors Under

large forward bias an accumulation layer is formed between the spacer and barrier

segment and it is more noticeable compared to the accumulation layer that is formed if a

negative bias is applied

60

Figure 216 Conduction band diagram showing band bending and 2DEG formation at the L1

spacer

Consequently a triangular well is formed which creates an emitter 2D electron gas

(2DEG) population The electrons in this 2DEG occupy the quasi-bound states which

mean high excitation energy thus allowing the electron to tunnel through the barrier as

depicted in Figure 216

In term of RF performance it is important to highlight that a thicker spacer layer

will affect the depletion region which gets wider and thus will reduce the junction

capacitance of the device as per the following expression

119862 = 휀0휀119903

119860

119889

(230)

Where A is the area of the device d is the thickness of the main device structure which

consists of spacers barrier and well layers 휀0 is the permittivity of free space and the

relative permittivity of the spacer material is denoted by εr However the intrinsic delay

time will also increase and hence degrade the device high frequency performance

Therefore optimisation through spacer thickness requires trade-off between reducing

leakage current at reverse bias and degrading device junction capacitance

GaAs

GaAs

AlAs

2DEG

Γ

X

61

263 ASPAT Electrical Parameters

The classical approach in determining the current flow through an ASPAT diode is

by solving the Schrodinger and Poison equations Prior work had been done by Syme et

al in 1991 Due to the fact that AlAs barrier is very thin tunnelling is assumed to occur

at the Gamma valley ie AlAs bandgap= 283eV (rather than X valley)[18 75] and only

from accumulation layer (2DEG)[59] Here the DC characteristic of the ASPAT diode

can be calculated by solving Schroumldinger equation with the position vector represented

by z (in this case) Thus the equation is expressed as

minus

ћ2

2nabla

1

119898lowast(119911)nabla120569 + |119890|120593(119911)120569 = 119864120569(119911)

(231)

Supposing the current is uniform across x and y planes then this can be simplified to one

dimension Therefore the 1 D Schroumldinger equations becomes

minusћ2

2119898lowast

1198892

1198891199112120595 + |119890|(120595 minus ∆120595) = 119864119911120595

(232)

Where

120595 =

120569

exp (119894119896119911119911)

(233)

where ∆120595 is the correction term which reduces the effective barrier height The

Schroumldinger equation is solved using different values for Ez thus the quantum

mechanical current density in the z-direction is now expressed as[15]

119895119911 =

minus|119890|ћ

2119898lowast(120595lowast

119889120595

119889119911minus 120595

119889120595lowast

119889119911 )

(234)

It is different for a heavily doped contact which can describe as below the envelope

functions in the left and right contacts respectively can be described by plane wave

120595119897 = exp(119894119896119897119911) + 119877 exp(minus119894119896119897119911) (235)

120595119903 = 119879119890119909119901[119894119896119903(119911 minus 120580119873)] (236)

62

In this case the left contact covers the region zlt0 while the right contact covers the

regions zgt 120580119873 Equations (35 and 36) are then inserted into Eq (34) to form the

following expression

119895119911 =

|119890|ћ1198961

119898lowast(1 minus |119877|2) =

|119890|ћ1198961

119898lowast|119879|2

(237)

Where R (Ez) and T (Ez) are the complex reflection and transmission coefficients

respectively and they are solved by using the transfer matrix method This method has

been described in reference [56] The next step is to integrate the current in the z-

direction 119895119911 for all possible Ez values Thus the expression for the current density

becomes

119895119911 =|119890|119898lowast119896119861119879

2120587ћ2int(1 minus |119877|2)

infin

0

119897119899 |1 + exp (

119864119891 minus 119864119911

119896119861119879)

1 + exp (119864119891 minus 119881|119890| minus 119864119911

119896119861119879

|

(238)

The equations above are used to calculate the current density approximation from

the ASPAT main structure (two spacer layers and one barrier) only and based on

intraband tunnelling from the conduction band profile For real fabricated structure the

calculation must take into account both intrinsic and extrinsic elements of the diode

While the latter is mostly related to the pad and probe that is used to extract the I-V

characteristic the first element mostly comes from the epitaxial layer of the ASPAT

diode itself The ASPAT I-V characteristic is shown in Figure 217 which clearly

indicates nonlinear characteristics and thus can be used for detection applications

63

Figure 217 I-V characteristics of a fabricated ASPAT diode

2631 Intrinsic Elements of the ASPAT diode

In order to extract the intrinsic electrical characteristic of the ASPAT diode a

generic structure as shown in Figure 218 is essential Two main sources of contribution

to the electrical characteristics are the interfaces of each layer and properties of the

materials themselves

Figure 218 Intrinsic Elements of the Asymmetric Spacer Tunnel Layer (ASPAT) Diode

64

The electrical current flowing from the top contact to the bottom contact will go

through each epitaxial layer producing a close loop Each junction interface limits the

current flow and sum up the total resistance resulting in what is known as the diode

series resistance (Rs) In the ASPAT main structure there is a junction capacitance (Cj)

due to the undoped regions surrounded by heavily doped contacts thus acting as a

parallel plate capacitor The fully depleted capacitance (Cj) of the diode can be

expressed as in Eq (230)

2632 Series resistance of the ASPAT diode

The total series resistance (Rs) of an ASPAT diode can be calculated based on the

finished fabricated diode structure In general Rs depends on three contributors namely

the non-uniformities in the contact metallization the un-depleted epitaxial layer (total

thickness) on both side of the heterostructures and the resistance caused by the

spreading current from the Mesa into the much wider second contact layer ie doped

substrate or 2nd

ohmic layer[76] In fact the contribution toward building up the total Rs

solely depends on how the structure is designed In this work two types of structures

were deployed namely lateral structures and vertical structures These will be described

in the next subsection

For both types of structures the series resistance (Rs) of the ASPAT diode consists

of aspecific Ohmic contact resistance (ρcA) contact epitaxial layer resistance (Repi-

Layers) and spreading resistance (Rspr)[33 77] where Rspr is influenced by the type of

structure The specific contact resistance is obtained from Transmission Line

Measurements (TLM) of the sample The theory of the TLM will be discussed in detail

in the next section The expression for the specific contact resistance is

120588119888 = 119877119888119871119879119908sinh119889

119871119879frasl

cosh 119889119871119879

frasl

(239)

Here Rc is the contact resistance LT is the transfer length (effective length) w is the

contact pad width and d is the length of the contact pad

65

Repi-Layer is the sum of all doped layers that sandwich the main ASPAT device For

each doped layer the resistance is given by

119877119890119901119894minus119897119886119910119890119903 = 120588

119871

119860

(240)

Where ρ is the resistivity which is given by 120588 = 1

120583119899119902119873119863 L is the epitaxial layer thickness

in cm A is the device area (emitter size) micron denotes the mobility of the electron q is the

electron charge and ND is the donor concentration The spreading resistance depends on

the structure design of the device This will be elaborated in detail in the following

section

26321 Vertical structure (doped Substrate)

The XMBE307 structures were grown on n+ GaAs substrate to provide the

simplest fabrication process The cross section of the finished single diode can be seen in

Figure 219 below

Figure 219 Cross-sectional view of ASPAT vertical structure the dimensions of d b and h are not

drawn to scale

66

At low frequency and in a mesa that is etched into a doped substrate material the

spreading resistance can be approximated by[76]

119877119904119901119903 =120588119904

2119889

(241)

Where ρs is the substrate resistivity and d is the ASPAT diode mesa length

However the spreading resistance is increased at high enough frequencies as the skin

depth (δ) in the substrate is much lower than the effective mesa length (d) of the diode

A new spreading resistance is then calculated also based on the assumption that the skin

depth is much lower than the chip thickness (h) but much larger than the mesa length

Thus the spreading resistance at high frequency is given by

119877119904119901119903(119891) =

120588119904

120587120575[05 ln (

119887

119889) +

119887]

(242)

Where the skin depth (δ) is taken from standard planar formula and can be expressed as

120575 = [

2120588

(120583120596)]

12frasl

(243)

Where micro is the permeability and ω is the angular frequency During DC measurement

this type of structure requires having good suction on the stage for a good contact

However for small die (15mm times15mm) the suction sometimes is not strong enough to

provide a very good adhesion to the sample Therefore another type of ASPAT diode is

deployed which is based on the lateral structure by utilising semi-insulating substrate

and both contacts are connected to probes

67

26322 Lateral structure (Semi-insulating Substrate)

Figure 220 Cross-section view of the complete process ASPAT diode (XMBE304) The dimensions

are not drawn to scale

In order to obtain accurate measurement results so as to avoid contact errors to the

substrate between stage and Device Under Test (DUT) a lateral structure as shown in

Figure 220 above is deployed This type of design offers many advantages ie it

provides a path for on-wafer RF measurement However the proper design has to take

into account the increase in RS due to improper attention to the spreading resistance

This spreading resistance is different from the vertical structure that was discussed

above In this case it is mainly caused by a gap at the bottom contact The gap in the

horizontal direction between epitaxial layer and metal at bottom contact is denoted as D

gap in in Figure 220 Therefore Rspr for the lateral design can be expressed as [77]

119877119904119901119903 =

1

120587120590119889119866119886119860119904ln (

119886

119886119898119890119904119886)

(244)

Where σ is the conductivity between two coaxial half-cylindrical electrodes with inner

(amesa) and outer (a) rectangular length or bottom ohmic layer which is given by (σ =1

ρ) a d and amesa are the length and thickness indicated in Figure 220 above Noticeably

68

the D gap will have direct effect on the outer length (a) of the device which is also

proportional to Rspr For high-frequency operation where the skin depth is less than

d(GaAs) σ becomes

120590(120596) =

120590(0)

[1 + (120596120591119903119890119897)2]

(245)

Where τrel =microme and micro m as well as e are the mobility effective mass and electron

charge respectively It is recommended that for high-frequency applications the device

series resistance must be as low as possible

Hence for both type of structure the ASPAT series resistance is calculated based on

all the above-stated resistances and these are set by

119877119904 =120588119862

119860+ 119877119890119901119894minus119897119886119910119890119903119904 + 119877119904119901119903

(246)

The total RS can be decreased by increasing the emitter area of the diode However a

large device will not able to reach millimetre and submillimeter wave region (THz) as

the capacitance will also increase (Eq 230) Therefore both parameters will have a

trade-off between them to be able to work at ultra-high frequencies

27 Characterization of Ohmic Contacts

The semiconductor Ohmic contact can be characterised using techniques that will

be described in the following section First is the Cox-Strack technique which is

specially designed to characterise bulk type semiconductor (thick) contact resistance on

two opposite sides The detailed description of this technique can be found in [78]

Second is a technique called Four Point Probe This technique was developed in 1954 by

Valdes etel [79] to characterise semiconductor resistivity It can also be used to

characterise the contact resistance for planar type devices As this research does not

cover this method the details of the measurement can be referred to [80] Finally the

most common method which is also extensively used in this research is the standard

Transmission Line Measurement (TLM) The details of this method will be explained in

the next section There are simplified versions of the TLM method which require just

one lithography step but are nevertheless very powerful in characterising and optimising

the contact resistance known as Circular Transmission Line Measurement (CTLM)[81

69

82] However in this research this is not to be covered as the standard TLM is already

adequate for planar type devices

271 Transmission Line Measurement (TLM)

The formation of metal and semiconductor interfaces will create a contact that

becomes very important for the characterisations of any fabricated device Additionally

it enables the quality of certain process flow to be determined This interface must be

evaluated by a technique known as the Transmission Line Measurement (TLM) TLM

which was first introduced by Murrmann and Widmand [83] in 1969 underwent some

refinements by Berger [84] in 1971 The theory of the TLM can be described by

constructing a TLM structure which comprises a set of metals contact pads placed in

series on a highly doped semiconductor layer as depicted in Figure 221 The structure is

designed like a series of the islands to permit current flow in parallel to the contact pads

[80] which is a direction defined by etched patterns Each contact metal pad behaves

like a MESA which has a thickness (t) and width (W) The distance between each metal

contact pads is defined by d1 the gap between two neighbouring contact pads which are

beneath each contact pad is defined as effective length LT This will allow current flow

in and out of the subsequent neighbouring metal pad The resistance elements that will

be extracted are RA and RB which sit under the contact and in between two metal pads

These two elements represent the sheet resistance under the metal contact pad area and a

sheet resistance of the material between two metal pads

Figure 221 A simple TLM structure with effective length and sheet resistance underneath

t L

T

LT

RA R

A R

B

dn

Probe Probe

Metal Pad

GaAs

MESA

70

The basic relationship of resistance R with respect to the size of the metal contact

or in the standard transmission line can be expressed as [44]

119877 = 120588

119871

119860= 120588

119871

119905 times 119882=

120588

119905times

119871

119882= 119877119904ℎ

119871

119882

(247)

Where ρ is the materialrsquos resistivity L is the length t is the thickness W is the width

Rsh is the sheet resistance and A is the cross-sectional area of the transmission line The

unit for ρ and Rsh are Ωm and Ωm2 respectively

The total resistance RT of this structure can be taken from the sum of the two

neighbouring padrsquos resistance RA and RB In order to relate with Eq (247) above this

RT will be substituted into Eq(247) to become

119877119879 = 2119877119860 + 119877119861 = 2119877119904ℎ119860

119871119879

119882+ 119877119904ℎ119861

119889119899

119882

(248)

As suggested in [85] RshA and RshB are assumed to be identical Therefore Eq(248)

can be reorganised into specific contact resistance RC and semiconductor sheet

resistance Rsh above as 119877119862 = 119877119904ℎ119860119871119879

119882 the new equation is then

119877119879 = 2119877119862 + 119877119904ℎ119861

119889119899

119882

(249)

The common practice throughout this research is to design a TLM structure that

has a ladder structure consisting of 10 metal pads with each one measuring to a size of

100microm width and 50microm length and the space between the first and second metal pad

5microm The gap is increased after the second metal pad by a further 5microm until ten metal

pads are completed produce a separation between the ninth and tenth pad of 45microm The

TLM ladder structure as depicted in Figure 222 is supplied by a constant current of

1mA at the very left and right metal contacts by two probes This allows the extraction

of RC (Ωmm) and Rsh (Ω) instantly from such structure The potential difference

between the two adjacent metal pads is measured by another two probes and the reading

of a voltmeter is recorded The total resistance is obtained by using Ohm law where

voltage is divided by current (VI) Another voltage reading is taken for the next two

neighbouring metal pads until the largest gap is reached The readings of (conversion

71

VI) which result in the corresponding resistance RT are then plotted against spacing and

the result can be seen in the graph in Figure 223

Figure 222 Top view of TLM ladder structure use in this work

Additionally in the measurement the voltmeter used in work has a very high resistance

Otherwise there will be leakage of current occurring through the probes and cables

Therefore the parasitic resistance of the cable or connector and the probe contact can be

ignored The key parameters that can be extracted from the graph will be discussed in

the next paragraph

Figure 223 Typical plot of resistance versus TLM spacing

y = 02592x + 02375

0

2

4

6

8

10

12

14

0 10 20 30 40 50Res

ista

nce

Rn (

Oh

ms)

TLM Spacing dn (um)

LT

d1 d

2 d

3 d

n

I (1mA)

V

V

LT

ME

SA

Su

bst

rate

72

The straight-line graph plotted in Figure 223 can be referred to Eq (249) and

this must be done by assuming the sheet resistance Rsh of the material is constant If the

straight line of the graph is extended up to initial gap (d=0) the intercept on the y-axis

provides the 2RC value To extract the 2LT further extrapolation is made until the

interception at RT = 0 is reached Therefore an important parameter which is the specific

contact resistance ρC can be found from this expression[85 86]

120588119888 = 119877119862119871119879119882sinh

119897119871119879

cosh119897

119871119879

(250)

Where l represents the total conductive semiconductor thickness The final part that can

be extracted from the graph is the slope of the line This is obtained by dividing the sheet

resistance with the width of the metal pad ie represented by expression (119877119904ℎ

119882frasl )

28 Basic Characterization Techniques and procedures

281 Measuring tools and apparatuses

The success of every experiment is determined by the backend results that are

obtained from the measurements It is important to choose an appropriate instrument

which will provide the required data for a valid and detail analysis Thus this section

will give a brief explanation of the measuring instruments and methods that were

exploited in this research The measurement apparatus systems that are available and

have been utilised in completing this thesis are ldquoset of DCrdquo and ldquoSet of RFrdquo

measurements The DC set measurement consists of room temperature and variable

temperature system This system is built to perform process monitoring during device

fabrication and it comprises of five main components

The fundamental component in the ldquoDCrsquos set toolrdquo is the Agilent B1500A

Parameter Analyser [87] used to provide fixed currentbias during testing The other

component is a Karl Suss PM5 Cascade Prober [88] which is used to receive fixed

currentvoltage and to supply its to the semiconductor contact via probe tip The PM5

Prober has at least four probe arms and each of them is fitted with ldquoneedlerdquo called probe

73

tip The size of the tip that is normally used here is 2microm and in some cases the tip size

of 1microm is also utilised The currentbias supplied to the sample must go through the two

of Source Measurement Units (SMU) namely SMU1 and SMU2 to ensure no mismatch

occurs between parameter analyser and diode All the testing are controlled using a

software called Integrated Circuit Characterization and Analysis Program (IC-CAP

2009) brought from Keysight Technologies[89] The software is installed on a standard

Personal Computer (PC) and the PC is connected to a General Purpose Bus Interface

(GPIB) to link with the B1500A Parameter Analyser This system can be organised

based on the purpose of measurements ie IV characteristics TLM Schottky Diode

and transistor as it is very flexible to change the configuration For examples

Transmission Line Measurement (TLM) configuration requires the addition of a digital

multi-meter and four-point probe tip while diode measurement just needs two point

probe tip without a digital multi-meter Figure 224 illustrates the measurement system

for a set of DC to test the TLM structure

Figure 224 The set of DC measurement apparatus arrangement to measure the TLM

On the other hand the set of RF measurement consists of five block elements

configuration in the system namely the Vector Network Analyser (VNA) DC

sourcemonitor Cascade Microtech Prober SMU and Control PC via GPIB The RF set

Agilent B1500A

Parameter analyser

DUT

(TLM Diode

Capacitance)

Karl Suss

PM5 Cascade

Prober

PC

(MS windows 2000)

Digital Multi-meter

ICCAP 2014 Provide current

source

Measure and read the

voltage Stage and Probe

Signal

Current Source

SMU1 SMU2

General purpose bus interface

74

performs RF characterisation after the device fabrication is completed This system can

also perform DC characterisation as its basic instrument has this function too The VNA

machine used in this research is the Anritsu 37369A [90] which can perform the

Scattering Parameter (S-Parameter) measurement with a frequency range of 40MHz to

40GHz The DC sourceMonitor utilised in this experiment is the HP 4142B Modular

DC SourceMonitor [91] Both of these sub-systems are controlled by a standard PC

which exploits GPIB port to link them During operation the HP 4142B is connected to

the VNA by an internal bridge network and two SMUs to the Cascade Microtech Prober

which is then connected to the bond pads of the device It is identical to the set of DC

measurement where the SMU setting and data assembly are accomplished by IC-CAP

software package therefore the data that was obtained before and after completing the

fabrication can be compared This will enhance the validation of the data

However the stage and probes in both sets of measurement are different The

Cascade Microtech Prober has only two probe arms with each arm fitted with a 3-

fingers probe tip in the arrangement of Ground-Signal-Ground (GSG) as shown in

Figure 225 Each finger (pitch) is separated by 100microm thus to fit in with the pitch and

to reduce the mismatch in resistance the bond pad design must follow this separation

between each contact The GSG is configured by connecting the outer pads (Collector)

to the Ground probe tip while the inner pads (Emitter) are attached to the Signal probe

tip where the RF signal is sent and received through it Figure 226 shows the actual set-

up for RF measurement used in this research

75

Figure 225 The set of RF measurement for on-wafer DC and RF one-port measurement

Figure 226 Actual VNA system that was used for RF characterization

282 Measurement steps using a VNA

The RF measurement steps can be summarised in the following

- Step 1 Select or find the suitable VNA depending on applications

PC

(MS windows 2000)

ICCAP 2014

VNA

Anristru 32379A

40MHz to

40GHz

HP4142B

Modular DC

source

DC Bias Source

Cascade

Microtech

Probe Station

One 50microm pitch G-

S-G probe tips

Ground

SMU1

Cathode

Gen

eral

purp

ose

bus

inte

rfac

e

Lo

w P

ow

er H

igh P

ow

er

Cathode

Anode

SMU2

Ground

Ground

Signal

76

There are few factors that need to be well-thought-out before starting to use a VNA

especially for S-parameter measurements The factors are the availability of the VNA in

term of operational frequency and measurement port types air-filled metallic waveguide

or on-wafer Not all VNA can have an operational frequency for banded measurement

ie W-band Ku-band etc All these may require external signal sources to extend the

operational frequency

- Step 2 Properly setting up the VNA

The VNA can be set up depending on its application and the goal of measurement for an

instant number of point requires a desired frequency span IF bandwidth and the supplied

power level This very important to ensure the desired measurements are correct and

appropriate

- Step 3 Appropriate calibration system

In order to have a valid calibration appropriate calibration method has to be chosen

depending on the applications This will determine the accuracy and standards of

calibration As for on-wafer calibrations the de-embedding is normally used while for

the off-wafer the SOLR technique is more suitable to employ

- Step 4 Validation or verification of the calibration results

It is vital to validate the calibration results to ensure that the system has been properly

calibrated

- Step 5 Proper measurement

Proper alignment positioning and touching from probes tip to DUT is necessary to

guarantee a good repeatability and reproducibility of measurement results Normally

when positioning the probe an alignment marker is used as an aid By doing this similar

travel distance for the probes can be achieved The measuring plane will also be equally

well-defined

283 Measurement Practice and Flowchart

Essentially the device characterisation is performed in two stages ie during

fabrication as a process monitoring and after completion for data collection and

analysis Figure 227 shows the block diagram of a flow chart for testing a 15mm times

15mm wafer processing performance In the wafer processing after reaching the top and

77

bottom contactrsquos step the sample can be examined by measuring the TLM structure

according to the TLM procedure The measurement is conducted by exploiting a set of

DC measurement apparatus as mentioned above analogous to the TLM ladder structure

on the wafer surface Based on the TLM results the presence of any process issue during

the fabrication can be identified by examining the parameter such as contact resistance

and sheet resistance Thus a decision can be made either to proceed or to terminate the

fabrication should any issue is found early on As a result no materials will be wasted

further When a dielectric layer is involved Capacitance Dielectric measurement can be

tested This practise can be used to obtain the quality of the dielectric layer To connect

between a diode effective area and probe a bridge is requires It can be attached to the

diode emitter (to determine the diode size) to a bond pad for probe tip to touch This

bridge can either be left hanging in air or sticking to the isolated dielectric layer For the

GaAsAlAs material system the latter technique is preferable since it avoids issues with

the air-bridge which will be discussed in detail in Chapter 3 The opening area (via) for

metal connection can be checked by measuring the resistance on a special design pad

after a plasma dry etching step

Figure 227 Block diagram of the ASPAT measurement step

The next stage of characterising the device is the on-wafer diode measurement

which takes place after completion of all processing (including bond padsco-planar

pads) The work is carried out using the above set for RF measurement and employed

Ohmic Contact

Opening Via

TLM

Qualified

Device

Bad device

Dielectric

capacitance

Bonding pad DC and RF

Good

Good Good

Fail

Fail

Fail

78

purposely to access the DC and RF performance of each diode where the current-voltage

(I-V) characteristic and S-Parameter results are obtained In fact DC measurements are

first performed using a set of DC measurement and a rough IV characteristic can be

obtained to ensure the diode is working properly Usually the yield of any fabricated

device on 15mm times 15mm wafer in this research is between 70 to 90 In this research

the outcomes that will be discussed in the subsequent chapters are in term of the average

of measured values Thus it is very important to have a meaningful data to compare with

physical modelling and simulation in the future The diode IV characteristics are studied

by applying different DC bias at the emitter to collector terminals to extract its keys

parameters ie turn on voltage (supposedly zero bias) non-linear characteristic Rj RS

Cj etc

The device that has been measured by DC and having produced a valid result

will be marked for the next investigation ie the S-Parameter measurement This

measurement is executed using a similar system but with different probe types The

three fingers probe type is used and the device frequency response is measured via the

one-port network from the VNA The important parameters extracted from this

measurement are usually S11 (depending on how many ports are measured) Although

working devices are selected to measure accordingly there is a need to ensure the VNA

RF cable and probe tips are calibrated so that only valid data without errors will be

obtained A calibration technique called SHORT-OPEN-LOAD (SOL) is performed

prior to each daily measurement by exploiting a calibration sample with WinCal

software (Cascade Microtech)[92]

To avoid confusion it is worth mentioning that this technique used to calibrate

the device structure is different from what is used in SOL calibration to the equipment

Furthermore the de-embedding calibration is made on-wafer with the same substrate of

the actual device whereas the SOL is performed on a special calibration substrate

Normally the de-embedding results are not constantly automated with the VNA

equipment However the measurement is done separately starting with the special DUT

substrate then followed by the OPEN and SHORT de-embedding structures

79

3 RESULTS OF FABRICATION PROCESS OF ASPAT DIODES

31 Introduction

This chapter presents in detail the general fabrication techniques for a generic and

development work of micron scale ASPAT diodes The discussions will focus on the

semiconductor growth technique used in this work ie MBE and the fabrication process

steps which include sample cleaning photolithography etching and contact

metallization These techniques are ample to build and deliver commercially marketable

fabricated structure Hence all photolithography techniques used to complete this

project are based on conventional i-line optical lithography which is adaptable to

industry and commercial purposes

The fabrication process in this work can be fragmented into two major works

firstly the fabrication towards reproducibility repeatability and manufacturability in

term of device structure process flow and DC amp RF characteristics This work will

involve relatively larger emitter area which varies from 15times15microm2 to 100times100microm

2 The

larger area provides for ease and fast fabrication as well as DCamp RF measurements

Once repeatability and reproducibility of the process flow and performance is confirmed

the second part of this work which is concerned with applications in millimetre and sub-

millimetre-wave then took place In this work small emitter designs varying from

2times2microm2 up to 10times10microm

2 were considered with appropriate measurement pads The

successful fabrication of smaller diode geometries in the second part of the programme

naturally leads to a further investigations to understand its epitaxial layer structure and

extracting intrinsic components ie junction resistance junction capacitance and series

resistance which will determine the device performance in high-frequency applications

The I-V characteristic is obtained from DC measurement which is usually performed at

room temperature Its results are then compiled and compared with advanced simulation

It is worth mentioning that all samples that are investigated in this project are grown by

means of MBE and the activities related to the epitaxial layer growth using MBE in the

University of Manchester were done by the Materials Growth Team (Prof Missous) and

the authors has no responsibility for this particular task

80

32 Epitaxial Layer Growth Techniques

Before discussing the principle of the common lithography technique it is

important to discuss the wafer growth technique as it comes first before the fabrication

The growth technique that is extensively used in this study as well contributing a lot in

the electronic semiconductor industry is Molecular Beam Epitaxy (MBE) The following

section will discuss the basic operating principle of solid source MBE

321 Molecular Beam Epitaxy (MBE)

The MBE technique was developed in the early 1970s [27] and is purposely used for

growing high purity epitaxial layers of compound semiconductors Such sophisticated

growth technique provides significant functionality ie precise control of the thickness

(to one atomic layer) and contributes to the growth of various types of complex

semiconductor multilayers high quality and advanced materials This level of control is

vital for an assortment of heterostructures devices that are being utilized as part of the

development of the advanced electronics devices especially for the ASPAT diode which

require 01ML control over the AlAs barrier to attain acceptable variability in device

characteristics Additionally the accurate doping profile and excellent junction

abruptness also can be achieved by using this technique

Practically the MBE system used in this work is a solid source MBE which

utilises beams produced by heating up various sources The sources can be Si Al Ga

As and other group III-V compound semiconductors When the crucibles which contain

the sources are heated atoms or molecules of the various elements are evaporated and

travel in straight lines paths like beams directed toward a target (heated and rotating

substrate surface) The condition of the vacuum during evaporation is ultra-high vacuum

(UHV) ~10-11

torr in order to have high quality crystals The substrate is heated and

rotated to provide good growth uniformity over large areas ( up to 4times4rdquo wafers) [93]

The growth rate in typical MBE growth is ~1ML second and can be controlled

by the source temperature in the crucible The abruptness at the heterojunctions interface

and switching of the growth compositions can be obtained by precise control of shutters

that are placed in front of the crucibles Therefore an abrupt junction at GaAs and AlAs

81

interface can be formed to realize the barrier in the ASPAT diode In order to monitor

the quality of the growing crystal and measure the layer by layer growth mode

Reflection High Energy Electron diffraction (RHEED) technique is utilised This

technique works based on the diffraction of electrons from the crystal surface [93]

Additionally given that the ASPAT current density is very sensitive to the barrier

thickness a study has been made using different growth techniques namely MBE and

Metal-Organic Chemical Vapour Deposition (MOCVD)[59] From this investigation it

was concluded that the percentage local variability of current density produced by MBE

grown diodes is better than those grown by MOCVD Thus in this study to get benefit

from its performance all wafers are MBE grown

33 Basic Principles of Common Fabrication techniques

This section covers the generic fabrication process which underpins

reproducibility repeatability and process optimisations for high-frequency applications

331 Sample cleaning

Essentially semiconductor processing requires a ldquoclean environmentrdquo to produce

devices The clean environment is classified according to how many ldquounwantedrdquo

particles are present in a cubic meter There are four categories of clean room available

in the industry Class 10 Class 100 Class 1000 and Class 10000 [94]

The fabrication of all ASPAT diodes in this project was performed in a clean

room environment of Class 1000 equipped with laminar air flow and filter system to

give Class 100 or better during processing Although the sample was processed in a

highly controlled particle environment there is still a high chance for a sample to get

contaminated when handled by a human Besides this the source of particle which

contributes to the contamination can be from the apparatuses and processing equipment

used in the laboratory themselves Thus the process of cleaning the sample wafer

surface before the start of each step is vital

Generally in a clean room the standard solutions that are used to clean a sample are

N-Methyl Pyrrolidone (NMP) Acetone Propan-3-ol (Iso-Propane-ol) (IPA) and

82

deionized (DI) water The sample which is cut up into 15times15mm2 size tiles is cleaned

based on the following procedures

1 Hot NMP- The sample is dipped into the solution at 80˚C for 10 minutes This

solution acts as an organic type of nature pollutants removal

2 DI water- acts as NMP remover The sample is then washed by flowing DI water

throughout the samplersquos surface

3 Acetone- to ensure any remaining NMP is completely removed from the sample

The sample is dipped into Acetone for 5 minutes in a low power ultrasonic bath

at ambient temperature

4 IPA- is used to remove the Acetone from the samplersquos surface The sample is

then dipped into the IPA for 5 minutes in a low power ultrasonic bath at room

temperature The use of low power for the ultrasonic bath is to avoid the sample

cracking or breaking

5 Once done the sample is then blow-dried using nitrogen (N2) gun to remove any

moisture coming from the IPA Fortunately it is easy to remove the IPA

completely from the sample surface given its higher rate of evaporation

Once all these steps are accomplished a visual inspection using a high magnification

optical microscope is conducted to ensure the sample surface is clean The cleanliness of

the sample is determined by the (lack) of particles or other spots (liquid mark) that can

sometime be observed during the inspection Obviously the lower the number is the

better the sample is as it is impossible to totally remove dirt especially marks

Sometimes it is hard to remove the particles in one go There is always a need to repeat

each cleaning step for a few times However this will not affect the sample in term of

electrical performance as the cleaning solutions used are non-destructive

332 Photolithography

Lithography or sometimes called pattern transfer is the most important step in

realizing microelectronic devices The designed geometry and dimensions on a quartz

glass plate called a ldquophoto-maskrdquo must be done prior to the fabrication process While

the photo mask can be designed by using various software tools in this work the design

83

is done via the Advance Design System (ADS) by Keysight The details of the design

which includes three different mask designs will be covered in Section 34 The mask

consists of the desired patterns (master) that can be printed onto solid materialrsquos surface

by means of an electrochemically sensitive polymer (photoresist) using

photolithography In fact this type of optical lithography technique can be performed

with and without a mask due to its simplicity It is also easy and cheaper compared to

other techniques such as x-ray lithography or Electron-beam lithography (EBL)

Although the latter is expensive it is still worth to have since it provides a higher

resolution which is preferable when developing sub-micrometre technology

processing[95] The special feature about EBL is that it does not need a mask to pattern

samples but can be produced by the movement of an electron beam point source and

hence writing the patterns directly on the surface

The ultra-violet (UV) based light sources are more popular among researchers

and development workers because they are cheaper and have modest resolution Usually

photolithography operates at wavelengths (λ) from 193 nm to 436nm The source that

provides the UV light is a Mercury (Hg) arc lamp which uses narrowband filters to select

single emission lines First is the i-line at λ= 365nm then the h-line which is of lower

resolution and has λ of 405nm and thirdly the g-line with a λ= 436nm[96] The

conventional optical lithography used in this research uses the ldquoi-linerdquo at a wavelength of

365nm For shorter wavelengths than these excimer laser or krypton fluoride laser with

a λ= 248nm and argon-Fluoride with a λ = 193nm are also used in the industry The

higher power levels enable higher productivity (throughput) while narrower spectral

widths reduce chromatic aberration provide better resolution and larger depth-of-focus

In this research all the fabrication process are done by utilizing a conventional optical

lithography (i-line) using a Karl Suss MA4 mask aligner Before starting any UV

exposure it is important to check the UV light intensity as it will affect the resolution

and thus desired device dimensions Therefore every corner that is exposed to the UV

light is calibrated to be at 09mWatt power exposure

The complete set of photolithography components consists of photoresist

(photosensitive polymer) photo-mask (chromium) which is used to block the UV to

form a pattern mask-aligner and developer (chemical solutions) Standard fabrication

84

process usually practiced at the University of Manchester starts with sample cleaning as

mentioned earlier Then the sample needs to go through heat treatment to remove all the

moisture with a temperature set to be 150˚C and bake for 5-10 minutes After having

cooled down the samplersquos surface is coated with a thin layer of photoresist via a

technique called spin-coating using a Laurell CZ-650 series spinner The spinning speed

is set depending on the type of resist ie 3000rpm for negative photoresist and 5000rpm

for positive photoresist The rotating speed of the spinner does not have much effect on

the coated photoresist thickness but will have consequence on the uniformity distribution

over the sample surface The coated thickness photoresist however depends on the

concentration of the specific photoresist Once spin-coated is done another short heat

treatment (1 minute) is required to ensure the resist is hard enough to contact a

photomask and to remove any excess solvent The temperature is set on the hot plate to

about 115˚C and 110˚C for positive and negative photoresists respectively

The important segments contained in the photoresist are a polymer (base resin)

a sensitizer and a casting solvent [97] The polymer will react by changing its structure

when exposed to the radiation While sensitizers will govern the reaction of the

photochemical in the polymeric phase the casting solvent will permit the spin-coated

application on the wafer surface The photoresist consists of positive and negative

photoresist both of which were used in this research Their basic difference is with

respect to the area that is exposed by UV light ie whether it will remain on the

semiconductor surface or will be removed In the case of a positive resist the exposed

area will be removed by the developer and the covered area will remain In other words

whatever is displayed on the photomask goes onto the sample surface On the other

hand for the negative photoresist the area that is covered from UV exposure will be

dissolved by the developer Figure 31 shows a 3D picture of both processes used in this

research

85

Figure 31 3D illustration of Optical lithography process used in this research

To obtain the desired pattern on the surface after UV exposure for a certain

duration the sample is required to go through a development procedure by using a

developer The developer is used to expel the dissolvable part of the photoresist after

being exposed to UV The usefulness of the developer depends on the photoresist ie

MF319 developer is specifically for positive photoresist while MIF326 is suitable for

negative photoresist Both types of developers will not harm the devices as they are a

kind of metal free ion solution Thus no free ion will change the characteristic of the

device The common practice in the clean room at the University of Manchester is to use

positive photoresist namely Shipley Microposit S1800 supplied by The Dow Chemical

and negative resist AZnLOF2020 (AZ2microm) which is supplied by MicroChem For the

S1800 series the thickness of the photoresist is determined by the last two digits ie

13microm thick for S1813 and 05microm thick for S1805 Both positive photoresists are used

Resist

GaAs

Dielectric

Photo Mask

Negative Resist Positive Resist

After Etching

86

mostly as protective area during wet chemical etching and as sacrificial dielectric layer if

higher temperature is applied ie 190˚C On the contrary AZ2microm which normally has a

thickness of 2microm is useful for patterning small dimension which leaves small gaparea

for the metallization process to fill It has good aspect ratio and useful in single layer lift-

off (post metallization process) In fact this type of resist can be thinned by diluting into

an Edge Bead Removal (EBR) solution and smaller device feature size can be obtained

The final dimension of certain devices (mesa size) is governed by the exposure

time the distance between photomask and samplersquos surface and the development time

The appropriate UV exposure time is required to avoid over-exposure which will cause

the spreading of light into the purportedly dark-field area[98] However the effect of the

exposure time differs between the photoresist types smaller opening area than the mask

pattern for negative resist and larger opening area for positive resist[45] The gap

between the photomask and wafer surface must be reduced as much as possible to avoid

the UV light going through the unwanted area To obtain a good gap value an applied

pressure from stage to the mask is required The normal pressure used in this research is

between 04 to 1 Pascal depending on the type of photoresist (negative and positive) as

well as its thickness Lastly the development time also influences the dimension of the

device Appropriate development time is required because ie over-development will

cause the polymerized photoresist to etch laterally resulting in bad patterned geometries

On the other hand under-development will cause non-uniformity in the surface after wet

chemical etching (positive resist) as well as causing lift-off problem (negative resist)

333 Etching Process

The etching process is a process of removing unwanted semiconductor layers to

define device geometry and isolate each individual device in one sample There are two

types of etching technique used in this research ie wet chemical etching and plasma

dry etching The wet chemical etching is based on Orthophosphoric(H3PO4) solution

which is a selective etchant to materials like GaAs InGaAs and AlAs The selective

etchant is referred to a solution that can etch away a specific semiconductor with a

specific etch rate The etch rate depends on the mixture ratio and concentration of the

solutions ie the higher the concentration the higher the etching rate In practice the

87

temperature humidity and epitaxial layer doping level also have an impact on the etch

rate Hence to minimize variation in the etch rate both temperature and humidity in the

clean room are constantly monitored and regulated Ambient temperature between 18degC

to 19degC is usually suitable while humidity is kept within 30 to 40 The advantage of

the wet chemical etching is that it is inexpensive controllable and with high throughput

highly selective and simple The mixture solution that is used in this work is

Orthophosphoric (H3PO4) Hydrogen Peroxide (H2O2) and Di-ionised water (H2O) with

ratios of 3150 3110 and 212 The ratio of 3150 provides an etch rate of about

600Aminute and is good to define the area and opening the under-cut in air bridges for

InGaAs samples[99] The 3110 ratio results with highly anisotropic shape but is easy to

control as the etch rate is about 1500Aminute for GaAs material system However the

212 etchant solution will provide extremely anisotropic etch rate of about

1000Asecond and is quite difficult to control For GaAsAlAs ASPAT diode it can still

be controlled since the thickness that needs to be removed is about 7000A (refer to

section 342 for details on the fabrication of ASPAT) Table 31 summarizes the etch

rate with different ratios and different selective materials Common to all the ratios

mentioned above are the isotropic side walls with lateral and vertical etch rate of 11

Table 31 Chemical wet etch characteristics using Orthophosphoric-based and Ammonia on GaAs

and InGaAs materials

Material Etchant Ratio Etch Rate (Aringminute)

GaAs H3PO4H2O2H2O 3150 600

GaAs H3PO4H2O2H2O 3110 1500

GaAs H3PO4H2O2H2O 212 60000

GaAs NH4OHH2O2H2O 118 2000~3500

InGaAs H3PO4H2O2H2O 3150 850

The plasma dry etch is purposely run to obtain extremely high anisotropic etch

profile vertically and horizontally It can be done via mask with positive photoresist and

self-align mask which is metal contact as a mesa protector The precursors that are used

to etch away GaAs InGaAs and AlAs layer in this technique are Methane and

Hydrogen (CH4+H2) On the other hand Carbon Tetrafluoromethane (CF4) and H2 is

88

used in removing Si3N4 The etch rate is determined by how much power is applied to

the plasma to hit the sample surface the pressure inside the chamber and amount of the

precursor In this research plasma Technology is used for dry etching This machine

which is a conventional OXFORD INSTRUMENTS 1990 machine can produce an etch

rate on average of 100Aminute for an RF power of 100mWatt

334 Sputtering (dielectric deposition)

In this work a Kurt JLesker PVD 75 is used to deposit Si3N4 layer on the sample

surface The deposition rate depends on the RF power that is applied To avoid surface

damage on the sample surface a sacrificial layer formed by SiO is deposited using the

Bio-Rad Thermal evaporator before transferring to the PVD 75 to start deposition with a

low (75Watt) RF power Once the deposition time reaches 30 minutes the power is

increased up to 200Watt As a result good uniformity of the dielectric layer is obtained

335 Metallization Process Lift-off and Annealing

The metallization process is a process in which metal contacts on semiconductor

devices are created The purpose of this process is to make a proper interconnection

between the semiconductor devices to other parts of the circuit elements In other words

it is to connect the semiconductor device to the outside world This process will allow

the device to be examined electrically so that all electrical characteristics can be obtained

ie resistance I-V curve capacitance conductance etc The metal scheme used in this

process depends on the semiconductor material In the case of GaAsAlAs ASPAT

diode Gold-Germanium (AuGe) Nickel (Ni) and Gold (Au) are used However in the

case of InGaAsAlAs ASPAT diode the metal scheme used is Titanium (Ti) and Au

The technique used in this process is resistive thermal evaporation

The metallization process starts by cleaning all the metallic sources and boats

ie tungsten boat Au Ni and Ti metals by using Trichloroethylene Acetone and IPA

consecutively for 5 minutes each in a high power ultrasonic bath This step is very

important to reduce the risk of contamination during thermal evaporation Once done all

the metallic materials are dried using a high-pressure nitrogen gun and then dipped in

89

Hydrochloric acid (HCL) solution for 2 minutes to de-oxidize the metals so that it has

minimal effect on the series resistance of the ASPAT diode

Two types of thermal evaporators were used extensively in this study both

Edwards Auto 306 (one denoted as Junior Auto 306) The latter is used to deposit alloy

type of metals while the former is used for the non-alloyed type of metals In the case of

GaAsAlAs ASPAT alloy type metals are used while for InGaAsAlAs non-alloyed

metals are used The cleaned metals are then loaded in the thermal evaporators and

placed on a resistive tungsten boat Each metal is placed on its specific tungsten boat to

avoid unnecessary mixture during the evaporation

Figure 32 Actual picture of thermal evaporator used in this study

Prior to loading the sample into the evaporator a standard fabrication process for

ASPAT diodes takes place by patterning the samples with AZ2microm negative photoresist

At the end of this step an opening area is created for the metal contact to be filled and

connected to the ohmic contact of the semiconductor The sample is also deoxidised

using a mixture of HCL and water in the ratio of 11 prior to evaporation This has to be

done in a very short time in order to avoid re-development of the native oxide layer

Inside the chamber the sample is securely placed on a chuck upside down facing the

filled tungsten boat The distance between the sample and metallic source boat is about

90

40 cm Figure 32 illustrates the actual thermal evaporation system used in this study

The thermal evaporator chamber is pumped down to reach a minimum pressure below

1times10-5

mbar before vaporizing the metal It is important that the mean free path between

metal amp sample is created and each vaporized metal stick firmly on the samplersquos surface

The normal practice in this study is to keep the vacuum pressure under 1 times10-6

bar so

that better device performance can be obtained The amount of current required to melt

down the metal is between 4 Amps to 6 Amp This amount of current is forced through

the tungsten boat and generates very high heat melting down the metallic source and

vaporizing it towards the sample surface The deposition rate for each metal can be

monitored by using a built-in film thickness monitoring (FTM) on the thermal

evaporator which proportionally depends on the amount of materials deposited The

GaAsAlAs sample is started with deposit of 55nm AuGe 13nm Ni and 500nm of Au

The reason of depositing AuGe first is due to the fact that the ohmic layer of the

ASPAT only can only be doped with a maximum doping of 4 times1018

cm-3

which is not

high enough for good conductivity Thus here Ge atoms will diffuse into the GaAs and

replaces Ga atoms during annealing process leading to higher doping levels (gt1 times1019

cm-3

) and hence improved conductivity

3351 Lift-off

The use of AZ2microm allows for the exact patterning of metals without the need for etching

using a single layer lift-off technique The negative photoresist also provides an undercut

profile which will create disjointedness between the desired metal pattern (on the

semiconductor) and undesired metal (on photoresist) The process of getting rid of the

unwanted metal from a sample surface is called lift-off The process starts once the

evaporation process is accomplished The sample is soaked into 80˚C N-Methyl-2-

Pyrrolidone (NMP) solution for usually 20 minutes (fast lift-off process) In most

instances the sample is in NMP for more than 12 hours in ambient temperature (slow

lift-off process) as the NMP solution is not destructive to the sample In this solution

the negative resist will be softened and the metal part which sticks on it will also be

eliminated from the sample As depicted in Figure 33 the lift-off process for a single

device shows the usual photoresist undercut profile observed To ensure that NMP

91

residues on the sample surface are completely removed DI water is used to rinse the

sample which is then blown dry with a nitrogen gun

Figure 33 Single layer lift-off process using negative photoresist

3352 Annealing

The alloyed (AuGeNiAu) metal stack requires a thermal treatment called annealing to

improve the ohmic contact between metal and semiconductor The sample which has

deposited top and bottom contacts is loaded into an annealing furnace at a temperature of

420˚C for 2 minutes In the case of GaAs during thermal annealing Ge atoms penetrates

into the GaAs crystal for approximately 70nm-250nm depending on evaporated

thickness of the metal layers annealing temperature as well as time [100] In this work

the total metal thickness evaporated for each contact layer is around 500nm This

thermal annealing treatment will also melt down the Au if it is too thick and is subjected

to too long a heat treatment Therefore it is not advisable to do annealing after the

sample is coated with bond pad metals as it can result in short circuited devices

sometimes

Az2micro Negative

photoresist profile

after UV exposure

and development

Evaporation to

form metal layer

(AuGe Ni Au)

Desired metal

contact

Metallisation

process

Lift-off undesired metal

92

34 GaAsAlAs ASPAT Process Optimization

As mentioned earlier this section present details two major process flows of the

fabrication process for the ASPAT diodes which utilised three stages of development of

photomasks design namely a ldquoFirst generation mask designrdquo (1st Gen) a ldquoSecond

Generation mask (2nd

Gen)rdquo and a ldquoThird Generation mask (3rd

Gen)rdquo design The 1st

and 2nd

generations mask designs are the designs that were produced in the first stage of

this work to develop the fabrication process know-how and to get familiarized with the

actual fabrication techniques in the cleanroom The difference between the 1st Gen and

2nd

Gen masks is in term of the development towards realizing Air Bridges and

Dielectric Bridges which were mostly covered by the 2nd

Gen mask design The

analyses in term of reproducibility repeatability and manufacturability for process

control and current-voltage characteristic as well as ultimately RF measurement are

obtained on relatively large size devices via these two mask designs The large area

emitter dimensions range in size from 15times15 microm2 to 100times100microm

2

The 3rd

Gen mask design was designed based on the optimization of the 2nd

Gen

mask which was to realize ASPAT diodes that are able to work at very high operating

frequencies Therefore in such design the ASPAT devices have to have a minimal

amount of capacitance and low series resistance this is can be achieved by shrinking the

emitter size of the diode to the smallest area possible as well as optimising the

fabrication process ton reduce parasitics The smallest fabricated devices designed on the

new mask has an 2times2 microm2 MESA area which also includes Ground Signal Ground bond

pads for both device and de-embedding structures (open and Short) for RF

measurements

Before the commencement of any fabrications and designing any layouts on eg

Si GaAs InP etc the epitaxial layer must be grown first to a desired design In the

University of Manchester the epitaxial layer structures are grown using one of the two

Molecular Beam Epitaxy (MBE) machines which are either the RIBER V90H or the

V100HU system Both systems are managed by Professor Missous Epitaxial wafers or

sample grown by each system are identified by a prefix and numbers that are prefix

VMBE for the V90H system and XMBE for the V100 system The epitaxial layers are

93

grown on four-inch wafer diameter The maximum diameter that can be grown on using

the V100H is 8 inches but generally single 4rdquo or 4x4rdquo wafers are used The wafers are

then diced and cut using a diamond scriber into 15mm times 15mm tiles for easy handling

and fabrication in the D12 cleanroom lab in the University of Manchester

In this section the structures fabrication and performance of the various ASPAT

diodes for both repeatability and high frequency applications will be discussed further

The ASPAT is manufactured on wafer sample XMBE304 which is a GaAsAlAsGaAs

lattice matched to a GaAs semi insulating substrate is the main focus The ground works

on this ASPAT such as the initial design and fabrication process flow optimization had

been conducted by fellow PhD colleagues in the group led by Professor Missous at the

University of Manchester

341 ASPAT Devices used in Fabrication

3411 XMBE368 and XMBE307

Table 32 Epitaxial layer of Doped substrate samples

Thickness (Aring)

Layer Material Doping (cm-3

) XMBE368 XMBE307

Top Ohmic1 GaAs (Si) 4times1018

~3000 ~3000

Buffer 1 GaAs(Si) 4times1017

~50 ~50

Spacer1 GaAs NID 50 50

Barrier AlAs NID 28 28

Spacer 2 GaAs NID 1000 1000

Buffer GaAs(Si) 4times1017

50 50

Bottom Ohmic GaAs(Si) 5times1018

~3500 ~3500

Substrate GaAs (Si) Doped 50000 50000

These two samples were the first batch of diode structures used in this work and

were grown using the RIBER V100 MBE machine A great deal of work was expanded

to ensure that it is able to produce appropriate non-linear I-V characteristics The work

carried out including finding suitable fabrication process steps mask designs process

control limitations ie etching rates etc The results obtained from processing these

94

samples mostly on large area anode and cathode sizes and their analysis included both

growth profiles and fabrication process flows These samples were grown on doped

GaAs substrates As such the finished diodes were vertical structures and the fabrication

process has marked differences compared to undoped substrate samples While

XMBE368 had similar epitaxial layer profile to XMBE307 during growth the AlAs

layer was set to be stagnant (ie no-rotation of the substrate during growth) to

investigate the effects of slight variations in barrier thickness

3412 XMBE304

The next sets of samples were all grown on semi-insulating GaAs substrates

Table 33 details the epitaxial layer profile of sample XMBE304 the main work horse

of this research work The growth of this structure was performed on a multi-wafer

platen and consisted of 9 x 2rdquo wafers (from XMBE304A to XMBE304I)

Table 33 The epitaxial layers of sample XMBE304 with barrier 10ML~28nm

Epitaxial layer Material Doping(cm-3

) Thickness(Aring)

Emitter GaAs(Si) 4e+18

3000

Emitter 2 GaAs(Si) 1e+17

400

Spacer GaAs NID 50

Barrier AlAs NID 28

Spacer 2 GaAs NID 2000

Collector GaAs(Si) 1e+17

400

Collector 2 GaAs(Si) 4e+18

4500

Substrate GaAs(SI) Semi-Insulating

For a typical ASPAT structure the emitter is essentially highly doped to

4times10+18

cm-3

to provide accumulation of electron in the emitter contact region It was

purposely highly doped to also achieve low ohmic contact with the metal The spacers

are used to avoid diffusion of dopants to the subsequent layers The ASPAT structure as

mentioned earlier has a single AlAs barrier with a very small thickness sandwiched

between two different length GaAs spacer layers

Batch XMBE304 is the main focus in these studies All activities required for

repeatability reproducibility process flow and devices as well as qualifying new or

95

optimization fabrication technique for small devices and RF performance which were

then used for high frequency were based on the set of wafers grow in this batch

342 Fabrication process of GaAsAlAs ASPAT diode toward repeatability

reproducibility and manufacturability

As the new wafer structures have to be tested and evaluated to gauge the

performance of the ASPAT diodes their uniformity also needs to be tested so as to

ensure it exhibited fully functional diode with zero bias detection in minimal variation in

IV characteristics between diodes As described previously 4rdquo wafers are always diced

up into 15mm times15 mm size for ease of handling in the cleanroom and for masks cost

purposes

3421 Doped n+ Substrate Wafers

For wafers grown on n+ substrates (XMBE368 and XMBE307) a two-level mask

set is sufficient to complete the fabrication process In this case the devices are designed

with mesa structures and top metal contact on the upper surface of the wafer and a

bottom contact on the backside of the wafer One mask plate is used for defining the

mesa structure (wet etch) and the other mask plate is used for defining the metal pad

Since the mesa is relatively large (30times30microm2

to 100times100microm2) there is no requirement to

define bond pads for measurement purposes A prior RTD mask designed by fellow

colleague Dr Md Adzhar for his PhD work was used for processing the ASPAT diodes

as well in the first instance The detail fabrication process is summarized in appendix I

for the doped substrate wafers Figure 34 shows typical IV characteristics for 30x30microm2

emitter size diodes obtained from sample XMBE368

96

The I-V measurement was taken from two different tiles (15mm times15 mm) located

on top and bottom of a 4rdquo XMBE368 wafer The location 1 marked as a blue line in

figure 34 refers to a device on a tile taken from the top of the 4rdquo wafer while location

2 (red) refers to a device on a tile taken from the bottom of 4rdquo wafer At 07V the

separation difference in current is ~228 for both tiles This shows that the device in

tile 1 is less resistive than the sample in tile 2 implying that the AlAs layer for tile

located on the top corner of the wafer is thinner and thus allows more electrons to tunnel

through it at a given bias

3422 First Generation Mask Design (1st Gen)

-0015

0005

0025

0045

0065

0085

-15 -1 -05 0 05 1 15

I C

urr

en

t (A

mp

)

V Voltage (Volt)

Current vs Voltage (XMBE368_1) for 30x30microm2

MidLeft2

MidRight3

Figure 34 Current-Voltage characteristic of sample XMBE368 used

in this study at two different locations on the wafer tile

(a) (b) (c)

Figure 35 ASPAT mask design for (a) 100times100microm2 (b) 30times30microm

2 and (c) 15times15 microm

2 diode dimensions

designed in the 1st Gen Mask

Location 2

Location 1

97

The fabrication of the GaAsAlAs ASPAT conducted in this 1st Gen mask is

followed the established generic process flow for InGaAs material which was done by

former co-worker Dr Md Adzhar Md Zawawi[101] The generic process flow is fairly

simple consisting of four mask steps as shown in Table 34 below The fabrication

process starts with the sample being cleaned in a NMP solution at a temperature of 80˚C

with Acetone and Propan-2-ol (IPA) The purpose of cleaning using NMP and Acetone

is to remove any organic material The IPA is used to remove Acetone residues

Table 34 Generic fabrication steps established by Dr Md Adzhar [101]

Step number Process

1 Top Contact

2 Mesa Etch

3 Isolation

4 Bottom Contact

In the 1st Gen mask the first step in fabricating the GaAsAlAs ASPAT diodes is

to use the first mask to define the emitter contact area The emitter has three different

sizes 100times100microm2 30times30microm

2 and 15times15microm

2 The lithography technique uses the

negative photo-resists AZ2micron with 55 second UV-photolithography to pattern the

top contact Then it is developed using MC319 developer to clear and define the

exposed area for the metals to stick to The sample is then subjected to plasma etching to

remove all organic residues and contaminants Then it is dipped into a mixture of

diluted Hydrochloric acid and water HCL H2O with a concentration of 11 for de-

oxidation This must be done in a short time right before the evaporation in order to

ensure good contact between metal and semiconductor surface with very low oxide

formation in between The ASPAT device performance depends on this step as contact

to the channel is by means of current flowing through the anode to the cathode terminals

In our lab the approach taken to achieve the Ohmic contact is by evaporation of Gold

Germanium (AuGe) Nickel (Ni) and Gold (Au) metals layers on top of the cap layer

Subsequently the metal is defined via a lift-off process using NMP

98

The next critical step is to define the MESA or island The mesa etch mask is

designed with two options with 05 microm or 10 microm tolerance The different mesa

tolerances are introduced to act as a safeguard for the emitter from producing excessive

undercut caused by lateral etching This step is to isolate and eliminate the unwanted

GaAs which will electrically link the active layers as many devices will be fabricated at

the same time on the same wafer tile (15mm times 15mm size) The lithography process in

this step uses positive resist and is developed using MC326 developer This step is

achieved using a wet etch process where a non-selective etchants mixture of

H3PO4H2O2H2O etches down the epitaxial layers until it slightly exceeds the AlAs

barrier with an etch rate of about 600Aring to 900Aring per minute The outcome of the MESA

or island step is the formation MESA active layers which are surrounded by inactive

layers of semi-insulating substrate (when using semi-insulating substrates)

The isolation mask is a step that is basically the same as the MESA etch step

eg using similar etchants mixture same lithography process but a different mask This

step is used to etch down until the GaAs substrate is reached which means that the

etching time is longer than that in the MESA etching step The purpose of this step is to

fully isolate the device from other neighboring devices hence ensuring no electrical

connection exists between each device within one sample Since these ASPAT diodes

employ an air bridge structure of size 1times5 microm the two minutes mesa etches will

simultaneously provide an initial undercut through lateral etching for the air bridge

formation

Based on the results so far obtained using the first-generation mask which

provided large mesa areas the current voltage characteristics of the ASPAT as far as the

non-linear zero bias is concerned did work very well However there was a need to

reduce the ASPAT mesa area down to very small dimensions to achieve mm-wave or

THz detection frequencies It is certainly a general rule for semiconductor device which

operate at very high frequency to have extremely small lateral dimension which

minimises the capacitance within the ASPAT device Furthermore for wider adoption of

the technology it is also important to develop a simple reproducible and low-cost

fabrication method for ASPAT diodes Details of this fabrication process are attached in

appendix II

99

3423 Second Generation (2nd

Gen) mask (ASPAT-GSG)

This work intended to enhance the 1st Gen-Large Area ASPAT photomask by

adding many features including 2 times 2microm2 mesa areas ground-signal-ground (GSG) bond

pads to enable RF measurement and three options device processing eg Air Bridge

Dielectric Bridge for semi-insulating doped substrate and dry etch-mesa Other reasons

for designing this mask was also to qualify process steps when deploying thin (1microm

width) bridge to connect small mesa area ASPAT diodes to the co-planar GSG bond pad

for DC and ultimately RF measurements

The mask was designed to fulfil the basic rules of fabricating various types of

tunnelling diode for instance RTD and ASPAT The diodes layouts were designed using

the commercial software Advanced Design Software (ADS) from Keysight

Technologies Ltd Once the design was completed it was ready to be sent to Compu-

Graphics Company for printing and patterned on a special chrome coated glass plate

The new 2nd

Gen mask design was termed ASPAT with Ground-Signal-Ground

(ASPAT-GSG) and consisted of two main designs ldquoAir Bridge and Dielectric Bridgerdquo

where each contained eight diodes with different emitter sizes (100times100μm2 50times50μm

2

30times30μm2 20times20μm

2 15times15μm

2 10times10μm

2 6times6μm

2 and 2times2μm

2) The Air Bridge

design is comprised of Design 1 (doped substrate) and Design 2 (undoped substrate)

This can be selected by changing the order of each individual layer of the mask steps

The same options are applied for Dielectric Bridge design which included processing for

doped and undoped substrates The details of the masks will be explained in the

following paragraph

(1) Air Bridge

Design 1 Contains five steps or layers mask of size 15mm times 15mm

suitable for air-bridge for undoped semi-insulating substrate

Design 2 Consist of seven layers steps mask of 15mm times 15mm size

There are 413 die chips in about 6mm times 6mm sizes in this design Figure 36 shows both

type of design for Air Bridge mask processing

(i) Mask 1- Top Contact

(ii) Mask 2- MESA

100

(iii) Mask 3-Isolation

(iv) Mask 4- Bottom Contact

(v) Mask 5- Collector Bond pad with GSG

(vi) Mask 2A-Dielectric

(vii) Mask 6A-Collector Bond Pad with GSG

(2) Dielectric Bridge

Design 1 Consists of seven identical steplayers masks with lateral

lengths of 15mm times 15mm each suitable for fabrication on semi-insulating

substrate

Design 2 Also has seven steps layers masked with dimension of 15mm

times15mm each suitable for doped substrate processing

There are 357 die chips in this type of mask design with an estimated size of 6mm times

6mm separately The smallest emitter size that is connected with a dielectric as the

bridge is 6times6microm2 The smallest 2times2microm

2 diodes was designed with the air-bridge

connected to the emitter bond pad due to the difficulties of opening viascavity for less

than 2 microm2 devices Figure 37 show both options for Dielectric Bridge mask processing

(i) Mask 1- Top Contact

(ii)Mask 2- MESA

(iii)Mask 3-Isolation

(iv) Mask 4- Bottom Contact

(v) Mask 5-GSG bond pad

(vi) Mask 6- Via Dielectric

(vii) Mask 7- Dielectric Bridge

(vii)Mask 5A- Via Dielectric

(ix) Mask 6A-GSG bond Pad

(x) Dielectric Bridge

Figure 36 Air Bridge design (red circle) for large mesa area (100times100microm2) device with

option for doped substrates

Figure 37 The layout of 1st design of Dielectric Bridge (green circle) mask design for 100 times

100microm2 emitter size with option for doped substrate processing

101

34231 Fabrication Process of the Air-Bridge Design

The fabrication using 1st Gen mask as mentioned in Section 3422 is less

complex however the fabrication process for Air-Bridge mask design contains a few

additional steps which are to add bond pads for the Ground-Signal-Ground radio

frequency (RF) layouts If the sample is on a doped substrate another layer needs to be

added leading to a total of six steps all together

Table 35 Standard process flow for Air-Bridge design fabrication

Mask Number Step number Process

1 1 Top Contact

22A 2 Mesa EtchDielectric

3 3 Isolation

4

5

4

5

Bottom Contact

Bond Pad

In this work the samples that have been processed so far are XMBE304

XMBE314 (GaAsAlAs base material system) and XMBE326 (InGaAsAlAs base

material system) All of them are built up on semi-insulating substrate thus the normal

process flow was followed Step 1 to step 4 follow exactly the same route as the 1st Gen-

ASPAT Large Area mask formerly discussed in Section 3422 This process is then

continued by spin coating AZ2microm on the surface Mask 5 is used to pattern the bond

pads The metallisation scheme used for bond pads is TiAu and the thickness must be at

least 1 micrometre thick This is to minimise series resistance at the pads and ensure a

robust surface is created when used for DC and RF measurements

342311 Air-Bridge Process Optimization

Since the Air Bridge design approach is focused more on developing air-bridges

which have a width of 1 microm and as the smallest device is 2 times2 microm2 the negative resist

(Az2microm) which was used in the 1st Gen design type was changed to Az1microm After spin

coating an Edge Bead Remover (EBR) is required to remove the beads at the edge of the

sample this is to ensure no gap is created between samples and mask when ldquohard-

102

contactrdquo is applied during the photolithography step The use of Az1microm and EBR are

critical to enable the fabrication of small air-bridges and emitter sizes

The exposure time during the photolithography technique also needs to be

increased to achieve a 1microm size air bridge A longer time than normal is required thus

95 seconds is used for exposure under UV light Once finished appropriate

development time must be applied to ensure the line for the air bridge is perfectly

opened for the metal to fill up To get impeccable result in this step both combination of

exposure and development have to be tuned naturally leading to trade off in both Too

long exposure and developing time will break the bridge whereas for too short exposure

time and developing the air bridge will not open

There are two types of etching method used in this fabrication wet

(H3PO4H2O2H2O) and dry (NH4H2) etching Both have their own advantages and

disadvantages Wet etching is faster than dry etching However the etching profile is

highly isotropic and causes serious undercut The smaller devices which have

dimensional sizes of 6times6microm2 and below will ldquoshrinkrdquo the effective area under the top

contact metal Even though dry etching can have highly anisotropic etching profile

which can prevent excessive undercuts it requires a lot of time for etching the

semiconductor layers The NH4 and H2 plasma that are used in this technique do not only

etch the GaAs and AlAs layers but also etch the metal contact at the same rate as the

semiconductors Thus the metal area must be covered with a photoresist or thicker metal

(~1microm) must be used to counter this issue Figure 38 shows that the emitter bond pad

which is not covered by the resist will eventually be etched away by the dry etching

Figure 38 Dry Etching for the first run in this study

As mentioned in section 333 the advantage of H3PO4H2O2H2O wet etching is

that it can be made using two solutions fast and slow The fast solution is based on a

concentration ratio of 212 while the slow solution is based on a 3150 mixture The

103

212 fast solution will produce good anisotropic etch profile but is tricky to control due

to the rather high etch rate of 1000Aring per second Both fast and slow wet etching

solutions being carried out in this experiment are replicated from successful recipes that

were developed by other co-worker from Prof Missousrsquos group

342312 Issue of Over Etch under Top Metal (Wet Etching for Air-Bridge

Design)

The first run using the solutions mentioned above started with the deposited

metal as a top contact (Top contact must be defined first to make a bridge) then Mesa

etching to remove ~7000Aring GaAs using the fast etch solution The next process to be

followed was to employ the slow etch solution to etch down the epitaxial layer to the

substrate Doing this ensured that air-bridge was open and the devices isolated

individually Unfortunately unexpected results were observed where large undercuts

still appeared for both 2times2 microm2 and 6times6microm

2 devices Figure 39 shows severe undercut

under the big device that can clearly be seen in the digital scope Most probably this

problem occurs due to excessive time used for the wet etching (10 seconds for fast

solution and 10 minutes for the slow etch solution)

Figure 39 Severe undercut of 2times2 microm2 and 6times6 microm

2 devices

The second run was employed to reduce the wet etching time in total thus dry

etching was needed As discussed above CH4 and H2 are used to etch away the GaAs

and AlAs semiconductor material while O2 is used to remove polymer residues To

reduce redundant generated polymer during dry etching the process must be separated

into several goes (runs) In this run at the mesa etch step the metal area is covered by

S1805 resist and then the plasma etches down until the heavily doped Ohmic layer is

reached For the isolation step the slow wet etching solution is used to ensure the

semiconductor under the bridge is removed The etching time is still long ie over 5

Zoomed Zoomed

104

minutes Hence this will still generate appreciable undercuts Although this run

managed to reduce the wet etching time it still did not solve the undercut Figure 310

shows undercuts under the effective mesa area still occurring in this run

Figure 310 Device with emitter area of 2times2 microm2 and 6times6 microm

2 after dry etch and wet etched

Further investigation has been made to find the root cause of this issue Scanning

Electron Microscopy (SEM) picture were taken to see deep into the completed devices to

investigate what is actually happening Figure 311 shows clearly that the semiconductor

under the bridge and effective area under the 2times2 microm2 device went missing The same

goes to for the large device where about half the size of the designed effective area under

to metal was also unintentionally removed by the solution

Figure 311 SEM Images of the GaAs sample

Since this work was also run together with an InGaAs based sample

(XMBE326) new knowledge regarding etching profile was acquired The lesson learnt

from this run is that the wet etch profile of the Gallium Arsenic (GaAs) material is

totally different from that of InGaAs Figure 312 shows the cross section of GaAs and

InGaAs materials after wet etching The evidences of these issues are also indicated by

the SEM images in Figure 313 and Figure 314 This more or less explained the reason

why the semiconductor under the top metal contact is always missing when wet

chemical etching is applied

Zoomed Zoomed

105

Figure 312 The cross-sectional view of InGaAs (left) and GaAs (right) samples used in this study

Figure 313 SEM images taken from the XMBE304 samples and a GaAs test sample

Figure 314 SEM images for InP and InGaAs taken from [56]

Furthermore previous work done by co-worker Dr Md Adzhar Md Zawawi has

optimised the process for RTD samples based on InGaAs and InAlAs heterojuction

semiconductor materials In this work Dr Zawawi found out that achieving submicron

dimensions is possible when using the soft reflow technique on InGaAs[102] However

it does not apply to GaAs when using the same technique There are few reasons to

106

believe that the work that was carried out by the previous student is not repeatable to the

XMBE304 GaAsAlAs material The main reason is due to the much thicker Ohmic

layers used in the GaAs of ASPAT structures which requires longer etch time The

sample that was used in the previous submicron work namely XMBE277 (RTD) [101]

and corresponding epitaxial layer structure is attached in appendix III The thickness

needed to be removed for the MESA step is 1421Aring compared to XMBE304 (ASPAT)

which is ~6900Aring

The fabrication results shown from the Air-Bridge design does not seem

favourable to the GaAsAlAs heterojunction sample These include unreproducible IV

characteristics and excessive undercut under the metal Therefore the GaAsAlAs

ASPAT sample cannot be processed using this type of mask (Air-Bridge approach) The

fabrication efforts were then moved to the Dielectric Bridge design described below

342313 DC measurement

Fully functional ASPAT I-V characteristics were not obtainable in this run due to

severe damages caused by the undercuts that happened underneath the top metal contact

As can be seen from Figure 315 the behaviour of the current response suggests a very

leaky diode This confirms that the air bridge approach does not work with the

GaAsAlAs ASPAT diode

Figure 315 Short circuit behaviour on one of the fabricated device in this run

-0015

-001

-0005

0

0005

001

0015

002

0025

003

0035

-3 -2 -1 0 1 2 3

Cu

rren

t (A

mp

)

Voltage (Volt)

30x30BP

107

34232 Fabrication Process of the Dielectric-Bridge Design

Since the Air-Bridge design was not successful an alternative Dielectric-bridge

process flow was then designed to solve this issue The first run in this Dielectric-bridge

process was performed to ensure that when the fabrication was completed it was able to

produce the correct and reproducible ASPAT current-voltage characteristics The

process flow in this run is to follow the initial design which identifies the steps according

to the mask number Step 1 to step 4 in the process are exactly the same as in the 1st Gen

mask design Step 5 which is the bond pad process is then continued with spin coating

the AZ2microm After exposure and development a TiAu metal scheme with a minimum

thickness of approximately 1 microm is deposited

Table 36 Standard fabrication process flow for Dielectric-Bridge design

Mask Number Step number Process

1 1 Top Contact

22A 2 Mesa Etch

3 3 Isolation

4

5

6

7

4

5

6

7

Bottom Contact

Bond Pad

Via Etch

Bridge

Step 6 is quite complicated compared to the other steps where the introduction of

S1805 resist is used as a dielectric layer to prevent short circuit between top and bottom

contacts This includes opening the smallest via (holes within the dielectric layer) of 2times2

microm2 and 6times6 microm

2 emitter sizes This step started by spin-coating the S1805 resist and

was then followed by baking it at 150˚C for 30 minutes The longer baking time is to

ensure it is hard enough to be deployed as a dielectric layer Once the S1805 had

hardened as dielectric layer the sample was then spin-coated again with S1813 (thicker

resist) on the dielectric layer to act as a mask for vias opening (to cover the dielectric

layer from via etching damages) After the S813 was developed O2 plasma etches was

108

applied to remove the S1805 that covered the top metal and bond pads The hole on the

metal was then exposed This required only two minutersquos plasma etches time

As soon as the holes were created within the dielectric layer a quick clean to

remove S1813 was performed using Acetone and IPA This has to be fast enough to

remove the S1813 without removing the S1805 dielectric layer Mask 7 then defined the

area for the metal that will fill up into the open vias This metal connected the top

contact metal of the device to the bond pad (GSG pad) The same metal scheme as

mentioned in Section 3422 was used as the bond pad

The fabrication using the dielectric bridge approach in the 2nd

Gen mask design

appeared to work successfully when initial I-V characteristics were taken and it also

showed that the ASPAT diode was fully functional However using hardened S1805 as

a dielectric layer is not a good practise for manufacturing device since there were left

over residue on the sample surface as depicted in Figure 316 This residue comes from

the non-uniformity of S1805 resit that was formed during heat treatment Therefore the

next run was to avoid using S1805 but replacing it with a standard dielectric layer based

on Silicon nitrite (Si3N4)

Figure 316 The surface of the sample after final processing

342321 DC measurements

Given that the process of qualifying mask steps for GaAsAlAs ASPAT diodes

looked promising using the dielectric bridge approaches initial I-V characteristics

measurements were carried out for XMBE304A Figure 317 shows that the I-V

characteristics are comparable to those of the other runs ie in the 1st Gen mask

Noticeably the IV characteristic (Figure 317) between each device size does not scale

109

on a single line This is attributed to the unintentional variation in the size of the emitter

during wet etching within each sample

Figure 317 IV characteristics of sample XMBE304 for emitter sizes of 1000 microm2 2500 microm2 900

microm2 400 microm2 225 microm2 100 microm2 and 36 microm2

The current densities shown in Figure 317 are calculated based on a reduced

effective area of the devices by a factor of 08 This assumption is made as actual area

(without metal) is un-measurable Theoretically devices with small areas will have

higher resistances compared to larger devices However the opposite occurred in these

devices This problem is still under investigation It could be due to the spreading

resistance producing different values according to device sizes (smaller devices have

longer D while bigger devices have shorter D)

34233 Dielectric-Bridge Process Optimization (Si3N4)

Due to poor surface roughness of the sample when using S1805 as dielectric

layer this run was employed to improve the surface quality by using Si3N4 In order to

get the actual size of the emitter the processing needed to start by defining the

semiconductor area using either top contact mask or mesa etch mask and not to deposit

the metal first Smaller effective area will be obtained if the top contact mask is used to

define the area compared to the MESA mask since there are 05 microm and 1 microm tolerances

designed in the MESA mask

-000001

0

000001

000002

000003

000004

000005

-2 -1 0 1 2

Cu

rren

t D

en

sity

(A

mp

microm

2 )

Voltage (Volt)

Current Density vs Voltage

density36

density100

density225

density400

density900

density2500

density10000

110

Table 37 New arrangement of the mask number and step in Second Run

Mask Number Step number Process

3 1 Isolation

22A 2 Mesa Etch

4 3 Bottom Contact

6

1

7

5

4

5

6

7

Via Etch

Top Contact

Bridge

Bond Pad

The fabrication was initiated by isolating each device using Mask 3 to define

them individually Then MESA mask (Mask 2) was used to cover the area from wet

etching optical measurement were used to obtain the actual size of the emitter without

metal Figure 318 shows the process results after H3P04H2O2H2O etching for a ratio of

3150

Figure 318 Optical images after MESA etching (a) 1 microm Tolerance (b) 05 microm Tolerance

Although the mask was designed with smaller size of 6times6microm2 using the slow etchant

solution the effective emitter size can still shrink to ~3times3 microm2 and ~2times2 microm

2 as can be

seen in Figure 318 The next step was to define the bottom contact using a metal scheme

that is suitable for GaAs ohmic layer ie AuGeNiAu After lift-off the samples were

then cleaned and spin-coated with AZ2microm Then the area was defined by using Mask 6

allowing the exposed area to be filled by Si3N4 which is the dielectric layer used in the

second run However a tricky issue happened here as the Si3N4 was difficult to lift-off

especially on the smallest emitter area in this mask In this run the lift-off process was

successfully done after three days Step five which uses Mask 1 was to define the top

111

contact The same metal scheme will fill up the tiny holes within the dielectric layer to

attach on the emitter semiconductor Figure 319 below shows an optical image of a

6times6microm2

device after the top contact lift-off process Once lifted-off annealing took place

to ensure Ge diffuse into the GaAs contact layer to lower the resistance

Figure 319 After lift-off processing

Step six uses Mask 7 to define the area of metal connection between active areas

to the bond pad In this step and step seven a different metal scheme from the top and

bottom contact was used to connect metal to metal Here a TiAu scheme was used The

final step (step 7) was to define the bond pad area In this context mask 5 is used The

bond pad requires thicker metal thickness to reduce the series resistance and to increase

the robustness of the metal surface when probed with needles during measurements

This second run of 2nd

Gen mask dielectric approach went well with better samplersquos

surface when using Si3N4 as the dielectric layer compared to the previous run using

S1805 Once the bond pad were defined for each device as done in previous run the

preliminary current-voltage characteristics of the Si3N4 run were obtained This is

described in the next section

342331 DC measurements

As mentioned in the previous section the Si3N4 run started with a defined emitter

area This allows actual ASPAT dimension to be measured thus the current density

versus voltage that are plotted in the following figures are based on actual measured

sizes

112

Figure 320 The current density of XMBE304 processed using Si3N4 as dielectric layer

Although the actual area of each device is obtainable from measurement the current

density of this sample does not scale well either Figure 320 show the current density is

not scalable from 03V to 1V This probably happened due to the same issue related to

the spreading resistance for each diode varying To investigate this issue new runs were

required with intention of reducing Rsprd and were expected to lead to more scalable

devices Therefore a study of reducing the spreading resistance (Rsprd ) due to

contribution of the large D-gap was carried out using the 2nd

Gen mask with the Si3N4

dielectric bridge approach The following section discusses in detail the steps used in

reducing the gap between the top and bottom contact for the GaAsAlAs ASPAT diode

34234 Dielectric-Bridge (Si3N4) Process Optimization by varying the D-Gap

The series resistance as discussed in detail in Section 26322 is generally due to

the total contribution of specific contact resistance in particular the diode size the sum

of all doped layer resistances that sandwich the main ASPAT layer and the spreading

resistance which comes from the lateral structure diode design[33] Thus in order to

acquire better performance at high frequency these contributors must be controlled One

parameter that can be controlled through this fabrication process is to reduce the

spreading resistance by controlling the separation between top and bottom contact ie

the D-gap as indicated in Figure 321

-000001

0

000001

000002

000003

000004

000005

-1 -05 0 05 1

Cu

rren

t D

ensi

ty

(Am

pm

2)

Voltage (Volt)

Current Density vs Voltage

D36

D100

D225

D400

D900

D2500

D10000

113

Figure 321 side view of lateral ASPAT structure

Without designing a new mask at this stage the same 2nd

Gen Dielectric-Bridge

mask was used but the process flows did not follow the sequence order of mask

numbers The reason for deciding not to design a new mask to reduce the D-Gap was

because the fabrication process flow had not yet confirmed it repeatability and

reproducibility Thus the available photo mask at that moment was fully utilised

Furthermore the feature of the 2nd

Gen mask that included the mesa tolerance can be

exploited in this study Hence the initial idea was to reduce the length of D by using the

tolerance that was designed in with Mask 2 (MESA etch mask)

Table 38 New arrangement for the Third run using Dielectric-Bridge mask

Mask Number Step number Process

1 1 Mesa Etch

3 2 Isolation

3

2

3

4

Bottom Contact

Mesa Cover

6

1

7

5

5

6

7

8

Via Etch

Top Contact

Bridge

Bond Pad

The process therefore was initiated by cleaning the sample and was then

followed by spin-coating it with S1805 to cover the emitter contact and define the actual

D-gap

114

size of the diode This first step used Mask 1 with wet chemical etching using the slow

solution (H3P04H2O2H2O etch with a ratio of 3150) Once the resist was striped

optical measurements were performed to obtain the effective area of the emitter Figure

322 below shows the smallest emitter area obtained after etching

Figure 322 The measured size of the emitter area and the length D (blue color marked)

Step two was to isolate the devices individually by using Mask 3 This was

performed using the fast etch solution (H3P04H2O2H2O etch with ratio of 212) This

took about 10 seconds to remove about 10000Aring of material

Step three which involved two masks was the most complicated in this process

Firstly Mask 3 was used to define the bottom contact by covering the sample with

AZ2microm and then the sample was hard- baked at 190 ˚C for 4- 5 minutes to ensure it had

totally dried before applying Polydimethylglutarimide (PMGI) The PMGI type used in

this process is Lift-off Resist (LOR) SF11 After spin-coating the SF11 at 7000RPM for

45 seconds it was post-baked at 190 ˚C for 4 minutes This was done to ensure that the

LOR SF11 was hard enough for the S1805 to stick on it This was followed by spin-

coating S1805 at 4000RPM for 30 seconds and exposing it under i-line UV for 20

seconds and developed using Micro Dev mix with DI water (11 ratio) for one minute

When the correct shape of S1805 was formed the sample was exposed under flood UV

for 15 minutes The SF11 was developed with 101A developer for 1 minute Figure 323

below summarizes the whole process in step three The step three processes concluded

by depositing the bottom contact with alloyed metal scheme and once lift-off had taken

place the next step continues as usual

115

Figure 323 Summary of LOR technique steps

Step four as well as the subsequent steps were completed in this run by copying

from the previous run (ie second run) which was to deposit the Si3N4 as the dielectric

layer and so on All the devices that underwent fabrication using this technique were

measured under optical microscope MC60 assisted by the Walsall software tool in the

lab From the measured value obtained it was expected that reducing the length of D in

this technique would improve the spreading resistance by up to 70 from the original

2nd

Gen mask design Table 39 summarizes the calculation of original spreading

resistance and improvement using this technique The calculation method and equation

are taken from Section 26322

Table 39 The outcome of the spreading resistance before and after using LOR technique

Device Size(microm2) D-original(microm) RSprd(Ω) D-new(microm) RSprd(Ω)

100x100 4 056 163 023

50x50 14 327 194 054

30x30 19 602 163 097

20x20 21 832 212 135

15x15 22 1007 202 18

10x10 24 1293 205 252

6x6 25 1643 230 418

116

Technically from the experiment in this run it was observed that the development time

also controlled the final length of D longer development time produced shorter D

lengths while shorter time yielded larger D gap lengths Therefore in this run the

development time was kept constant because of the desired D length from the mesa

tolerance in general is only ~2microm

342341 DC measurement

As usual when wafer processing was completed early DC measurement took

place to check the diodes performance Each diode size was measured and compiled

into one graph as shown in Figure 324 It is clear that the performance was better than

that in the previous run in term of scalability and current conductivity

Figure 324 IV characteristic in 3rd run (reduced d-Gap) for each diode size on the mask

To compare the current conductivity from this run and the previous run current

density at 05V was measured for both 2nd

and 3rd

runs Assuming the contact resistances

(TLM) for both runs were constant for 50times50microm2 device size the improvement of

current in the 3rd

run is about 92 This showed that such approach to reduce the D-gap

-500E-06

000E+00

500E-06

100E-05

150E-05

200E-05

-2 -15 -1 -05 0 05 1 15 2Cu

rre

nt

De

nsi

ty (

Amicro

m2)

Voltage (Volt)

Den_100microm2

Den_225microm2

Den_400microm2

Den_900microm2

Den_2500microm2

Den_10000microm2

Den2_36microm2

117

was successful in this run A new mask design was ready to take the challenge for

processing toward millimetre and sub-millimetre wave application

343 Fabrication process of GaAsAlAs ASPAT diode toward High frequency

Applications

So far the information that can be gathered from previous processing is that the

optimum process flow is achieved through dielectric approach design The effort in

reducing the series resistance by lowering its biggest contribution was attained through

lowering the D-Gap in the structure Once everything had been optimized ie the

process flow series resistance junction capacitance etc it was time to develop a new

mask design which only focused on the development of small ASPAT devices for use in

the millimetre and sub-millimetre wave regions

3431 Third Generation (3rd

Gen) mask design

The 3rd

Gen mask design was developed by taking into the account every aspect

of parameters that can contribute to the device by means of robust devices that are able

to function properly at ultra-high frequency The device cut-off frequency is given by

119891119888119906119905minus119900119891119891 =

1

2120587119877119904119862119895

(31)

Here RS is the series resistance and Cj is the junction capacitance To obtain high cut-off

frequencies Rs and Cj must be kept as low as possible From the fabrication point of

view two parameters that can be directly and easily controlled are the device area (A)

and the D-gap (which contributes to Rs)

Therefore calculations were made to find the best option Table 310 shows the

calculated value of capacitance (eq230 in page 44) and cut-off frequencies (eq31

above) for the ASPAT diodes studied These two equations extract the cut-off frequency

of the ASPAT assuming no external effects and for fully depleted devices The

XMBE304 ASPAT sample is expected to be suitable for millimetre wave applications

if small devices are successfully made In real devices sub-millimetrewave operation

can be hard to achieve due to increased series resistance and other process related

parasitics

118

Table 310 DC and RF characteristics for XMBE304

Device Size (microm2) Fully depleted

Capacitance(fF)

Calculated Series

Resistance (Ω)

Fully depleted Cut-Off

Frequency(GHz)

10000 5490 04 72

900 490 15 216

36 198 7 1148

16 879 11 1646

4 22 29 2500

Therefore in the 3rd

Gen mask design the smallest device that can possibly be obtained

in the GaAs based material fabricated using i-line lithography which is available at the

University of Manchester is 2times2microm2 and the gap between top and bottom is 15microm at

least The connecting bridge technique applied only utilised dielectric bridge method for

GSG features Figure 325 shows the layout of actual 3rd

Gen mask used in this study

There are 344 die chips on this type of mask design It also includes the Ground-signal

ground pad with 50um pitch for each chip and six de-embedding test structures as well

as eight TLM structures

Figure 325 3rd

Gen Mask layout showing actual devices de-embedding TLM and alignment mark

structures used in this study

119

This mask is designed generally from optimizations from 2nd

Gen Mask which

deploys a dielectric bridge for connection between the devices to the bond pads

Consequently the processing steps are not being much different but mostly follow what

is shown in Table 311 below The difference only applies to the much smaller mesa

size Other features included in this mask are de-embedding structures for RF

measurements via-hole test structure TLM structures and parallel plate capacitance test

structures

Table 311 3rd

Gen Mask process step

Mask Number Step number Process

1 1 Mesa Etch

2 2 Isolation

3

4

3

4

Bottom Contact

Mesa Cover

5

6

7

5

6

7

Via Etch

Top Contact

Bond Pad

The high frequency fabrication process flow is summarized in the following section

which shows illustrations in three dimensional and cross sectional (Figure 326) views

for easy understanding

34311 Step by Step Processing

0 Wafer preparation

Cleaning using NMP and DI

water or

Acetone and IPA

120

1 Mesa Etch

2 Isolation

3 Bottom Contact

4 Mesa Cover (Dielectric Deposition)

5 Via Etch

Etching using etchant solution

H3P04H2O2H2O etch with ratio

of 3110

Etching using etchant solution

H3P04H2O2H2O etch with ratio

of 212

Metal deposition with metal

scheme AuGeNiAu for

~500nm thick

Dielectric deposition 500nm

thick Si3N4

Via etch to open holes for metal

contact using reactive ion

etching using CF4

121

6 Top Contact

7 Bridge and Bond Pad

Figure 326 Step by step device fabrication (in 3D and cross-sectional view) for semi-insulating

substrate device type used in this study

Figure 327 Example finished process device with bond pad using 3rd

Gen mask

3432 TLM measurements

Transmission Line model (TLM) measurements for this run were carried out

after annealing (ie at 420˚C for 2 minutes) for process control monitoring by extracting

contact resistance (Rc) values These values are a measure of the quality of ohmic metal

contacts for a given process As discussed in Chapter 2 the TLM technique used four-

Metal deposition with metal

scheme AuGeNiAu for ~500nm

thickness and thermal annealing

420˚C for 2 minutes

Metal deposition with metal

scheme TiAu (1microm thick)

122

point measurement on TLM test structures located around the 15times15mm2 tiles as shown

in Figure 222 (in page 55) Normally five TLM structures are measured across the tile

for both top and bottom contacts Figures 328 and 329 display graphical TLM results

for top and bottom contact respectively As can be seen both graphs exhibit excellent

uniformity

Figure 328 XMBE304 TLM measurement for the top contact after annealing

Figure 329 XMBE304 TLM measurement for the bottom contact after annealing

Based on the graphs above for the top metal contact the average contact

resistance (RC) value using the metal scheme of AuGeNiAu is found to be ~005Ωmicrom

and the sheet resistance (RSH) is 22Ω However for the bottom contact the average

value for contact resistance is 012Ωmicrom and sheet resistance obtained is 26Ω Both

values are in a very good agreement with the known doping of both ohmic contact

0

2

4

6

8

10

0 10 20 30 40 50

Res

ista

nce

(O

hm

s)

Spacing (microm)

After 420C for 2mins top contact

y = 02592x + 02375

0

2

4

6

8

10

12

14

0 10 20 30 40 50

Res

ista

nce

(O

hm

s)

Spacing (microm)

After 420C for 2mins bottom contact

y = 02157x + 01053

123

layers Therefore the specific contact resistance (Eq 250) that contributes to the total

series resistance can be calculated and the value obtained is 15Ωmicrom2 and 54Ωmicrom

2 for

top and bottom TLM structures respectively

3433 DC characteristic measurements

Again once the wafer processing is completed room temperature DC

characteristics are taken using an HP 414B or HP500B parameter analyser and its actual

setup is as described in Chapter 2 This initial measurement was to ensure the

functionality of the diodes For this run using the 3rd

Gen mask the I-V measurements

for mesa active area of 4times4microm2 6times6 microm

2 and 10times10 microm

2 were taken and are depicted in

Figure 330 Figure 331 and Figure 332 respectively Nine diodes were measured for

each three device sizes to check their uniformity Thus the average current and standard

deviation are taken at two voltage steps of 1V and 15V Table 312 summarizes the

standard deviation for each measured data obtained from this final run

Figure 330 XMBE304 GaAsAlAs ASPAT measured IV characteristics at room temperature for

4times4microm2 mesa size

-00005

0

00005

0001

00015

0002

00025

0003

00035

-3 -2 -1 0 1 2 3

Cu

rren

t (A

)

Voltage (V)

C01_4um_1P_SI

C22_4um_1P_SI

C40_4um_1P_SI

W01_4um_1P_SI

W22_4um_1P_SI

W40_4um_1P_SI

AI01_4um_1P_SI

AI22_4um_1P_SI

AI40_4um_1P_SI

124

Figure 331 XMBE304 GaAsAlAs ASPAT measured IV characteristics at room temperature for

6times6microm2 mesa size

Figure 332 XMBE304 GaAsAlAs ASPAT measured IV characteristics at room temperature

for 10times10microm2 mesa size

Table 312 Standard deviation at two different voltages

Device size (um2) Standard Deviation 1V () Standard Deviation 15V ()

4times4 14 14

6times6 65 72

10times10 44 39

Noticeably the standard deviation of the device increases for the smaller size

devices This trend happened probably because the active mesa area is not uniform

causing different series resistances Since the smallest mesa active area which achieved

-0001

0

0001

0002

0003

0004

0005

0006

0007

0008

-3 -2 -1 0 1 2 3

Cu

rren

t (A

)

Voltage (V)

C02_6um_1P_SI

C23_6um_1P_SI

C41_6um_1P_SI

W02_6um_1P_SI

W23_6um_1P_SI

W41_6um_1P_SI

AI02_6um_1P_SI

AI23_6um_1P_SI

AI41_6um_1P_SI

-0005

0

0005

001

0015

002

0025

-3 -2 -1 0 1 2 3

Cu

rren

t (A

)

Voltage (V)

C03_10um_1P_SI

C24_10um_1P_SI

C42_10um_1P_SI

W03_10um_1P_SI

W24_10um_1P_SI

W42_10um_1P_SI

AI03_10um_1P_SI

AI24_10um_1P_SI

AI42_10um_1P_SI

125

a good I-V characteristic is 4times4microm2

in this run these devices were used for the next step

of characterisation which is the S-parameter measurement to extract their behaviour at

different frequencies

35 Conclusions

In this chapter basic fabrication techniques of GaAsAlAs ASPAT on both doped

and semi insulating substrates using standard I-line lithography as well as step by step

descriptions to achieve reproducibility in the process fabrication flow has been

demonstrated with relevant initial measured results Two major outcomes have been

demonstrated firstly related to repeatability reproducibility and manufacturability

mostly done on large device areas (15times15microm2 to 100times100 microm

2) Secondly a successful

process flow for small emitter size devices (2times2 microm2 to 10times10 microm

2) has been

developed

Subsequently two types of designs during process optimization were developed

namely Air-Bridge and Dielectric-Bridge approaches The latter approach seems

favourable for GaAsAlAs materials but was only successful when reducing the series

resistance of the device This problem was addressed by optimising the D-gap between

top and bottom contact which resulted in good scalability of each ASPAT dimensions

improving current conductivity by 92 and achieving a reproducible process On the

other hand in the former approach issues were encountered with over etching underneath

the diode effective area and thus it was hard to achieve reproducible devices

Repeatability reproducibility and manufacturability fabrication processes based on

dielectric bridge method were successfully developed This new process provides a

highly efficient and economical solution for the fabrication of GaAsAlAs ASPAT

diode Emitter sizes down to 4times4 microm2 dimensions are routinely and reproducibly

achieved in this process Series resistance which is an important parameter in

determining high frequency application are greatly reduced by changing the gap between

top and bottom contact

With all the new improvements implemented the possibility of the proposed Dielectric

Bridge method fabrication process was successfully applied for the fabrication of

ASPAT diodes

126

4 PHYSICAL MODELLING OF THE GAASALAS ASPAT

DIODE USING SILVACO

41 Introduction

Fabricating any device or circuit requires a lot of time resources cost etc

especially elements related to the production of semiconductor devices Furthermore in

realising such a device a clean room is required Hence fabrication and processing in

wafer fab need full attention to get it done with a higher rate of success reproducibility

and manufacturability One solution that can be highlighted which will be able to reduce

all the resources that are mentioned above is by using computer simulation approaches

or to be precise physical modelling For epitaxial layer based devices physical

modelling is a good choice as it would give a better understanding and insight into each

layer and how electrical characteristics are derived A software that is most suitable for

epitaxial layer physically modelled is SILVACO This software is a very comprehensive

tool to simulate epitaxial based devices and to predict their behaviour Such software

covers many aspects starting from the first principles of physic epitaxial layer

definition as well as device layout thus making it the most powerful virtual wafer fab

tool in the market

In this chapter the SILVACO packages are discussed The discussions include the

method of defining a new material defining models and constructing the AC or DC

supply to obtain the output characteristics of the virtual device Apart from these the

focus will also be on the ASPAT diode modelling simulation and analyses of the results

which will include the ASPAT structure suitable models and DC current-voltage

characteristics The dependencies of individual structure on the I-V curves will also be

highlighted Finally the discussion of results and analyses involving a range of

operational temperatures dependency as well as from comparison made to conventional

SBD used in this study will be examined

127

42 SILVACO modelling Tools

SILVACO is a modelling software introduced in 1984 by Dr Ivan Pesic It is

purposely created for electronicsrsquo devices physical modelling and characterization This

software company has become the major supplier for most of the Electronic Design

Automation (EDA) for circuit simulation amp design of analogue Mixed-Signal and RF

circuit market This Technology Computer Aided Design (TCAD) software can predict

the simulated device performances starting from first principles It has a package which

can provide Virtual Wafer Fabrication (VWF) simulation to the device designer and

which has the capability to perform two or three dimensional physical device modelling

by using the ATLAS simulator [103] SILVACO allows all parameters such as

electrical thermal and optical characteristics of a device to be simulated under desired

bias conditions It offers cost effectiveness as well as quick prediction of results for

many semiconductor devices compared to real experiments Some hands-on experiment

may not always perform hence SILVACO can be used as an alternative

The core of SILVACO is Atlas itself which provides a platform to perform DC AC

and transient analysis for such dimensional device structure regardless of the

heterojunction material type ie binary ternary quaternary etc As the brain of

SILVACO Atlas which receives input command files containing instruction text for

execution from a runtime environment known as Decbuild will process the instruction

text and display progression error and warning via Runtime Output All the calculations

of the resultsoutcomes of the simulation are plotted via a tool called TonyPlot which is a

tool to visualise the output Figure 41 below shows how precisely the physical

modelling takes place the process of building the structure how its parameters and

variables are defined how an appropriate model statement is selected how performance

is analysed and lastly how the outcomes are displayed

128

Figure 41 SILVACO Atlas simulation process flow

As can be seen in Figure 41 the structure specification statement is used to define

any desired structure by setting the command in Deckbuid of the following parameters

a) Mesh where the structure can be defined either in 2D or 3D Cartesian grids The

unit of the coordinates used is in microns and the spacing parameters which

define the netting size can be used to improve the accuracy of the analysis at any

given position The density of netting size in this statement determines the

processing time

b) Region where the multi-layers in a structure are defined and this statement has

to outline each layer that represents a separated region independently The mesh

must be assigned to a region and the sequence of the region is arranged from low

to high

-Mesh -Region -Electrode

-Contact -Material

Structure Specification

-Model -Interface

Model Specification

Method

Numerical Method

-Log -Solve -Load -Save

Solution Specification

-Tony Plot

Display Results

129

c) Electrode are used to define the location of bias point for a designed structure

when performing the electrical analysis In this work or for the case of a diode

two electrodes are allocated as an anode and a cathode In a vertical device the

latter electrode is placed at the bottom of the device while the anode is at the top

d) Doping this statement refers to doping concentration injected into the desired

region and it normally depends on the material types

e) Materials since SILVACO was developed specifically for Silicon-based devices

default parameters are set up for Silicon properties However the use of

materials statement allows SILVACO to run for different material ie GaAs

InGaAs etc In order to make it work this parameter is defined first and then

followed by the material name and its properties such as bandgap permittivity

conductionamp valence band discontinuities mobility etc

The most crucial part is to determine whether the simulated structure of a particular

device is correct or incorrect This is done by properly choosing the specific model

statement The Specific model statement is employed to express the physic equations

that are used during the device analysis The models statement depends solely on the

structure definition Examples are device structure with double barrier use Non-

equilibrium Green function (NEGF) model and single barrier uses Semiconductor-

insulator-Semiconductor (SIS) model for effective and accurate analysis process For the

case of a single barrier in SILVACO there are many model statements that can be used

for such analysis Therefore it is recommended to check model by model in order to

ensure that all the needed parameters are defined in the material statements and results

produced are valid

SILVACO is able to calculate such models by using different numerical methods

which means semiconductor device problematic is computed to make successive

solutions by random discretisation There are three different numerical methods that are

regularly used by SILVACO to perform its calculations which are Newton Gummel and

Block Basically this is solved by using a non-linear iteration procedure which begins

from an initial guess and which then uses an iterative process to find the predicted

solution The detail of these can be found in reference [103]

130

In order to turn on the problem solution the solution specification is defined This

include log solve save and load statement All these work together to provide data for

analysis by other functions The log statement is a file type that saves in memory and can

be loaded by Atlas Any solved device will be stored in the log file Therefore it is

necessary to define the LOG before SOLVE statement and close it after the calculations

While the save statement is used to store all data point to a node in the output file the

load statement is utilised to recall all the saved data to be read by Atlas

Finally all these files can be displayed or plotted on TonyPlot it is recommended to

make SET files at plotting point for a better visualisation The plotted or displayed files

in TonyPlot can be manipulated for scaling graphs overlaying different curves and most

importantly to export the data to other files formats

43 SILVACO Implementation GaAs AlAs ASPAT Modelling

The structures play an important role in determining the terminal output

characteristics Therefore to start simulating the ASPAT device the first thing that must

be specified is its structure As mentioned earlier in Section 3412 the ASPAT diode is

a top down multilayer structure This structure which is adopted from that of Section

3412 will be used as a basis to perform the ASPAT simulation Figure 42 shows the

structure of the ASPAT in this real simulation which is exactly the same as been

discussed in Section 26322 The ASPAT diode consists of two heavily doped (up to

5x1018119888119898minus3) GaAs contact layers on top and bottom slices adjacent to lightly doped (up

to 3x1017119888119898minus3) GaAs intermediate layers In between these layers is a sandwiched

structure consisting of two different lengths of undoped GaAs spacer layers and a thin

layer of AlAs that act as a tunnel barrier In this work the simulation result will be

compared with the fabricated measurement result depending on the size of the diode

The actual device that will be used to validate the simulation is the main device used in

this study (XMBE304) which is based on a lattice matched GaAsAlAs grown on a

semi-insulating substrate Therefore the design structures that are proposed in the

fabrication are compatible with the fabricated devices and are based on lateral structures

as can be seen in Figure 42

131

Figure 42 The GaAsAlAs ASPAT structures produced from simulation based on the diode

multilayer heterostructures on the right

In the consequent simulations a key observation regarding the AlAsGaAs

heterojunction is that there are two types of tunnelling processes direct tunnelling and

indirect tunnelling Figure 43 shows the Energy-Momentum (Ek) diagram depicting the

three valleys for the AlAs conduction band namely L Γ and X points In normal

circumstances the transition of electrons happens to the X-point which is the lowest

energy in the conduction band In the case of very thin barrier the tunnelling process

occurs at the Γ point in both AlAs and GaAs materials which is the direct tunnelling

process[18] It has been reported that in the case of an ASPAT diode tunnelling which

occurs at the Γ point will be the dominant component in the tunnelling current

Therefore the actual band gap will be different from the one at the X-point which is

which 216eV[15] By contrast the energy band gap at the Γ-point is around

283eV[104] Thus this simulation uses this band gap value

Figure 43 The Energy-momentum diagram for AlAs heterojunction semiconductor

216eV

Γ

X

E

K

L

AlAs

CB

VB

283e

132

The simulation code as attached in Appendix IV and the output of the simulation with all

the input mentioned above are shown in Figure 44 (a) and Figure 44 (b) Figure 44 (a)

is the band diagram at equilibrium and Figure 44 (b) is the band diagram when a bias is

applied

Figure 44 (a) The conduction band and the valence band of the ASPAT diode structure (b) the

energy band diagram of the ASPAT diode structure when under three different biases

44 Simulation Result and Analysis

Basically the SILVACOrsquos Atlas simulation package is used to calculate the I-V

characteristics from multilayers structures In the case of GaAsAlAs heterojunctions all

details of the structure as shown in Figure 44 above are calculated based on solving the

Schrodinger time-independent equation in each layer taking into account the variation of

effective mass and conduction band offset between GaAs and AlAs

For thermionic emission and tunnelling mechanism across an abrupt heterojunction

interface the general method used in SILVACO is taken from K Y Yang work [105]

The tunnelling current of the ASPAT diode uses the equation below[74]

119869 = sum2119898lowast119864119894(1 minus 119877)119896119879

1205872ħ3

119873

119894=1ln [

1 + exp (119864119865 minus 119864119894

119896119879)

1 + exp (119864119865 minus 119881 minus 119864119894

119896119879)]

(41)

Where m is the electron effective mass E denotes the energy of the electron R is the

total resistance k represent Boltzmann constant ħ is the reduced Planck constant EF is

the fermi level V is applied voltage and Ei is the electron energy perpendicular to the

-2

-15

-1

-05

0

05

1

15

1

82

16

3

24

4

32

5

40

6

48

7

56

8

64

9

73

0

81

1

89

2

97

3

Ene

rgy(

eV

)

Thickness (microm)

VB

CB

133

barrier The important parameters that enable SILVACO Atlas to perform correct

calculations and analysis are the choice of appropriate models The suitable model that is

available for evaluating the GaAsAlAs ASPAT is based on the Semiconductor-

insulator-semiconductor (SIS) model

Thus in this run Non-local Quantum Barrier Tunnelling Model (SISEL and

SISHO) are utilized specifically semiconductor-insulator-semiconductor mode This

model enables the tunnelling current between two semiconducting regions separated by a

quantum barrier to be calculated [103] It is assumed that the charge tunnels across the

whole barrier with the source or sinks at the interface with the semiconductor regions

Under the Non-Local Quantum Barrier tunnelling model another model that can be used

is semiconductor-semiconductor-semiconductor (SS) tunnelling model if the materials

are specified By correctly inserting all parameters with the right model an excellent DC

IV characteristic match between simulation and measurement can be produced as shown

in Figure 45

441 DC Current-Voltage Characteristic

In this simulation the current at each bias step and each mesh point can be set up by the

user however the detailed calculation such as formula usage methodology and

approach that is adopted by SILVACO Atlas is unknown As mentioned above in order

to produce the energy band diagram of the ASPAT the DC characteristic of the structure

can be solved by using the Schroumldinger and Poisson equation self-consistently

To ensure that these simulations are valid one fabrication was performed on an

ASPAT diode sample XMBE304 For this sample the structure parameters are the

same as has been set in this simulation The result of the measurement and simulation

are then compared Figure 45 shows that the simulation result is in excellent agreement

with the measured data for this sample

134

Figure 45 The current-voltage characteristic of an ASPAT diode (100times100microm2) and (4times4microm

2)

using SILVACO Atlas simulator for structure device XMBE304 showing excellent agreement

between simulated and experimental data

These fitted results were performed on both a large 100times100microm2 device and the

smallest obtainable from fabrication (4times4microm2) which was to be used for the repeatability

amp reproducibility studies as well as for high-frequency applications study respectively at

room temperature In order to get a good fit a few parameters had been modified in the

SILVACO software via the Deck-built tool for example energy band effective masses

and bandgap discontinuity of GaAs spacer and AlAs barrier (The mentioned parameters

values are summarized in Table 41) The percentages of bandgap discontinuity in

SILVACO using the ALIGN parameter is given by[103]

119860119897119894119892119899 =

Δ119864119862

Δ119864119862 + Δ119864119881

(42)

Where ΔEc and ΔEv are the conduction band discontinuity and valence band

discontinuity respectively The m0 in the tables denotes the electron rest mass Once all

agreement between measurement and simulation has been met the simulation is then

carried out with structure analysis at room temperature and different temperatures

simulation

Table 41 The parameter values used in this simulation

Material Bandgap(eV) ΔEg(eV) Effective mass(kg)

GaAs 1424 03 0067m0

AlAs 2835 071 0126m0

-00002

0

00002

00004

00006

00008

0001

00012

-2 -1 0 1 2

Cu

rre

nt

I (A

mp

)

Voltage V (Volt)

Current vs Voltage

Measurment

Simulation

135

45 Structure Analysis of ASPAT Diode

Once the device structure was modelled and having successfully produced a

precise band diagram as well as validated the simulation results with experimental I-V

characteristics the next step is to further analyse the relationship between basic device

structure and its I-V characteristics This approach is used to predict what would happen

to the DC output if some of the parameters were varied especially with regards to the

AlAs barrier thickness In the subsequence simulations the thickness of each main

ASPAT (unequal spacers and barrier) layer will be studied independently as a variable in

order to determine how each parameter affects the I-V characteristic in both magnitude

and curvature The analysis will also include manufacturing tolerance where the

structure parameter which will result in a 10 difference in their I-V characteristics is

examined[59 75] This will provide an overview of how precisely to manufacture each

layer of the device The following simulation is based on the XMBE304 structure with

emitter size of 4x4microm2

451 Dependencies of current on AlAs Barrier thickness

Since the AlAs barrier is what limits the transportation of electron flow and

hence the current (which depends exponentially on the tunnelling barrier thickness)

therefore the first analysis to run on the simulation is the variation on barrier thickness

In the simulation the barrier thickness is measured in term of the monolayer Generally

one monolayer can be calculated by dividing the lattice constant of the material by two

In the case of AlAs one monolayer is calculated as follows

119900119899119890 119898119900119899119900119897119886119910119890119903(1119872119871) =

119860119897119860119904 119897119886119905119905119894119888119890 119898119886119905119888ℎ119890119889 (5666Å)

2

(43)

= 283Å

The nominal value of the AlAs barrier thickness for sample XMBE304 is 283nm

ie ten monolayers In the first simulation test the current change due to the barrier

thickness variation from 9ML to 11ML was examined first followed by the amount of

barrier thickness change that would produce a 10 change in current The simulation is

setup by fixing all other parameters and varying the AlAs barrier thickness as mentioned

136

above with a step of 02ML In order to determine what fraction of a ML would yield a

10 difference in the current the barrier thickness is slightly changed to fit the curve for

both 5 above and 5 below the original curve

Figure 46 IV characteristics of the dependencies of current on AlAs barrier

The current-voltage characteristics of the ASPAT diode do change dramatically with

barrier thickness in forward bias but not that much in reverse bias (Figure 46) The

current decreases as the barrier thickness increases For a 1ML change in layer thickness

(from 9ML to 11ML) the current changes by over ~300 at 05V

Figure 47 Example of analysis at -1 and 1V to the current

-00002

0

00002

00004

00006

00008

0001

00012

00014

-15 -1 -05 0 05 1 15

Cu

rre

nt

(A)

Voltage(V)

9ML

92ML

94ML

96ML

98ML

10ML

102ML

104ML

106ML

108ML

11ML

973E-06

0

0000005

000001

0000015

000002

0000025

000003

0000035

85 95 105 115

Cu

rren

t a

t 1V

B

ias

(Am

p)

Barrier Thickness (ML)

Current change with tunnel barrier thickness

Forward Current

5

-5

137

From the simulation result shown in Figure 46 a 9945ML barrier thickness will

give a 5 higher current and a 10056ML barrier will give a 5 lower current (Figure

47) Therefore in total 01 ML difference yields 10 current difference These indicate

that in order to control the current within 10 barrier thickness difference the growth

precision in the barrier must be precise to better than 01ML Extensive studies have

also shown that the I-V characteristic of a GaAsAlAsGaAs diode is very sensitive to

the thickness of AlAs barrier This work has been reported elsewhere[17]

452 Dependence of current on Spacer I length l1

For the longer spacer length (l1) five different values are chosen from 01microm to

03microm The lengths are changed in the order of 005microm Therefore the arrangement of

length is as follows l1=01microm l1=015microm l1=02microm l1=0 25microm and l1=03microm

respectively The results are plotted from -15V to 15V anode voltage in Figure 48 The

I-V characteristic of the ASPAT diode does not change much with the length in the

forward bias region but in the reverse bias region the current decreases as the layer

thickness increases Here the l1 layer acts as a voltage arm and a small size device

cannot sustain big changes in spacer length Changing the length at the forward region

will also change the energy states on the anode side as well changing the states

distribution on the cathode side in reverse bias

Figure 48 I-V characteristic of the dependencies current to Spacer I layer

-00002

0

00002

00004

00006

00008

0001

-2 -1 0 1 2

Cu

rren

t (A

)

Voltage (V)

L=01

L=015

L=02

L=025

L=03

138

By fixing the current at -1V and 1V the current and the layer thickness relationship is

illustrated in Figure 49

Figure 49 Current changes with layer thickness l1

It is noticeable that there is a dramatic change in reverse current from 01 microm to

05 microm layer thicknesses However the forward current only falls slightly from 01microm to

015microm and is stable afterwards Hence for a small size device a large change in spacer

layer at the cathode will allow more current to pass

453 Dependence of current on Spacer II length l2

Finally is the variation in the spacer II Similar to spacer I above five values are

chosen for the shorter undoped GaAs layer length l2 the thickness is varied from

00025microm to 00075 microm (steps are l2=00025microm l2=000375microm l2=0005microm

l2=000625microm and l2=00075microm respectively The results are plotted from -15V to

15V anode voltage as shown in Figure 410 In this case a slight change in I-V

characteristic in the forward bias can be seen clearly which means the I-V

characteristic depends on the length of the shorter undoped layer Therefore the l2 layer

also acts as another voltage arm due to the asymmetrical length The effect is quite

similar to the spacer l1 but this times the forward current only slight changes The reason

for the small change in current is that the length change is small and it linearly affect the

states distribution

-000002

-0000018

-0000016

-0000014

-0000012

-000001

-0000008

-0000006

-0000004

-0000002

0

0

000005

00001

000015

00002

000025

00003

000035

01 015 02 025 03

Rev

erse

Cu

rren

t (A

)

Forw

ard

Cu

rren

t (A

)

Spacer Lenght (microm)

Forward Current (1V)Reverse Current(-1V)

139

Figure 410 IV characteristic of the dependencies current to Spacer 1 layer

Fixing the current at -1V and 1V the current versus layer thickness relationship is

illustrated in Figure 411

Figure 411Current change with layer thickness l2

The I-V curve depends on the length of the shorter undoped spacer layer quite linearly

The forward current changes in increase to the layer compared to the backward current

The layer thickness l2 should be small as long as it prevents carrier diffusion Therefore

all these three layers must be kept within limit to ensure that the high performance of the

ASPAT diode can be fully utilised

-00002

0

00002

00004

00006

00008

0001

-2 -1 0 1 2

Cu

rren

t (A

)

Voltage (V)

L=75n

L=25n

L=625n

L=5n

L=375n

-2E-07

-18E-07

-16E-07

-14E-07

-12E-07

-1E-07

-8E-08

-6E-08

-4E-08

-2E-08

0

0

00002

00004

00006

00008

0001

00012

0002 0004 0006 0008

Rev

erse

Cu

rren

t (A

)

Forw

ard

Cu

rren

t (A

)

Spacer Lenght (microm)

Forward Current (1V)

Reverse Current(-1V)

140

46 Temperature Dependent Simulation

This section will discuss in detail how changes in operating temperatures influence

the IV characteristics of the ASPAT diode The same codes as in the previous

simulation with fitted results are used for this temperature dependence study but a few

parameters were changed for different temperatures

Theoretically the material parameters that are influenced by the change of

temperature are band gap electron effective mass the density of state (NC NV) light

hole mass heavy hole mass permittivity and electron amp hole mobilities However not

all mentioned parameters will have a large impact on the IV characteristic in the

SILVACO Atlas simulation The most significant factors that give appreciable impact on

the DC output current-voltage were the energy bandgap and the effective mass The

GaAs bandgap as a function of temperature is given by the equation below [106]

119864119892 = 1198641198920 minus

120572 1198791198712

120573 + 119879119871

(44)

Here Eg is the bandgap Eg0 denotes the bandgap at 0K TL is the Temperature α=

Constant (Varshni Parameter) AlAs6e-4 GaAs5405e-4 β= Constant (Varshni

Parameter) AlAs408 and GaAs204 The calculated parameters that are used in this

simulation are shown in Table 42

Table 42 The calculated values of bandgap at different temperatures

Temperature (K) GaAs Eg(eV) AlAs Eg(eV)

77 1506 2903

100 1500 2899

125 149 2893

150 1486 2887

175 1478 2879

200 1470 2871

225 1461 2863

250 1452 2854

275 1443 2844

300 1424 2835

325 1419 2824

350 1414 2814

375 1404 2803

398 1394 2793

141

The effective mass of the materials used in this simulation can be expressed by

119898119899 = 1198980119899 + 11989810 (

119879119871

300119870)

119898119901 = 1198980119901 + 1198981119901 (119879119871

300119870) + 1198982119901 (

119879119871

300119870)2

(45)

Where mn is the effective electron mass mp represents the effective hole mass m1n and

m1p are constant number for the basic GaAs material m0n

for 119898119883119898119871 119886119903119890 119892119894119907119890119899 119887119910 119905ℎ119890 119890119902119906119886119905119894119900119899 (1198981198991199052 lowast 119898119899119897)

13 while m0p is based on the

expression 1198980119901 = (11989811990111989732

+ 119898119901ℎ32

)23 The calculated parameters are shown in Table

43

Table 43 The calculated effective masses for each temperature used in this simulation

Temperatures (K) GaAs Effective Mass (kg) AlAs Effective Mass (kg)

77 00660m0 03790 m0

100 00658 m0 03788 m0

125 00655 m0 03785 m0

150 00652 m0 03782 m0

175 00649 m0 03779 m0

200 00646 m0 03776 m0

225 00643 m0 03773 m0

250 0064 m0 0377 m0

275 00637 m0 03767 m0

300 00634 m0 03764 m0

325 00631 m0 03761 m0

350 00628 m0 03758 m0

375 00625 m0 03755 m0

398 00622 m0 03752 m0

142

Figure 412 Measurement and simulation comparison result as a function of temperature range

from 100K to 398K

Figure 412 above shows excellent agreement between simulation and

measurement results at various temperatures The IV characteristics correspond to a

device of size 100times100microm2 as presented in Chapter 3

47 Temperature Dependence characteristics of GaAsAlAs ASPAT Diodes

As mentioned earlier in chapter one the tunnelling diode has many advantages

over conventional Schottky barrier diodes some of which are a large dynamic range

low power consumption and very weak temperature dependence This section will

discuss the effect of variable temperature applied to the GaAsAlAs ASPAT diode and a

similarly processed TiAu Schottky diode Two samples were fabricated together for

these studies (XMBE304 and XMBE104 representing an ASPAT and a SBD

respectively) The fabrication technique is exactly the same as has been discussed in

Chapter 3 In order to make it fair for direct comparisons as well as easy probing both

diodes were fabricated with the same emitter size (100times100microm2) The DC measurements

at different temperature were carried out using a Lakeshore Cryogenic probe station over

the range of 77K to 398K in 25K step interval

-002

-001

0

001

002

003

004

005

006

-2 -15 -1 -05 0 05 1 15 2

Cu

rren

t I

(Am

p)

Voltage V (Volt)

T=100K_Simu

T=100K_Meas

T=398K_Simu

T=398K_Meas

T=200K_Simu

T=200K_Meas

T=300K_Simu

T=300K_Meas

143

471 GaAsAlAs ASPAT diode vs TiAu SBD

Once fabrication and measurement were completed both DC outputs of the diodes

were characterised and analysed Figure 413 shows a semi-logarithmic plot for

measured current versus voltage as a function of temperature for ASPAT sample

XMBE304 In forwards bias the current changes for different temperatures from 77K

to 398K are less than 5 percent This confirms the very weak temperature dependence of

current transport as it is dominated by tunnelling through the barrier On the other hand

the backward bias shows the current changes at different temperature are slightly bigger

than in forward this due to band bending occurring faster (making the effective barrier

lower) and allowing thermionic emission to significantly contribute to transport of the

current The only other study of temperature dependence for the ASPAT was made by et

el RT Syme[15] but details were not stated in their report

Figure 413 Log Current vs voltage as a function of temperature for ASPAT sample XMBE304

The effective barrier height for the GaAsAlAs ASPAT diode is higher than that of

the SBD (See Figure 414) therefore there is an expectation of more limited thermionic

current flow in the ASPAT than the SBD As mentioned earlier the conventional

Schottky Barrier diode that is used in this study consists of a Gold Titanium and GaAs

(AuTiGaAs) interface which is the baseline for the temperature dependence study

0000001

000001

00001

0001

001

-15 -1 -05 0 05 1 15

Log

Cu

rren

t (A

)

Voltage (V)

T=77K

T=100K

T=125K

T=150K

T=175K

T=200K

T=225K

T=250K

T=275K

T=300K

T=325K

T=350K

T=375K

T=398K

144

Figure 414 The effective barrier high of the Schottky barrier diode and the ASPAT Diode [3]

The SBD epitaxial layers profile is as shown in Table 44 below Theoretically the SBD

obey thermionic emission transport[44] and its I-V characteristic is given by

119868 = 1198680[exp (

119902119881

119899119896119879) minus 1]

(46)

Where q is the electron charge V is the applied voltage across the diode n denotes the

diode ideality factor k is the Boltzmann Constant T is the absolute temperature in

Kelvin and I0 is the diode saturation current which is given by the expression

1198680 = 119860119860lowast1198792 exp (minus

119902empty1198870

119899119896119879) exp (minus120572120594119890

12120575)

(47)

here A is the area of the diode A denotes the effective Richardson constant Oslashb0 is the

barrier height at zero bias δ represents the thickness of interfacial insulator layer χ

denotes the mean tunnelling barrier and α = radic(4120587

ℎ)(2119898lowast) is a constant value The

ideality factor n is taken from the slope of the SBD current-voltage characteristic and in

this study its value varies from 1 to 2 (depending on temperature) In the case of the

ASPAT diode thermionic emission can also happen if a thicker barrier is used (~ 100Aring

or thicker) as shown by CS Kyono et el [104] who concluded that when a thicker

barrier of AlAs barrier is used the current transport is dominated by thermionic emission

145

Table 44 Epitaxial layer profile for Schottky Barrier Diode (XMBE104)

Layer Material Doping(cm-3

) Thickness(Aring) Bandgap (eV)

Schottky GaAs(Si) 500E+15 7500 14

Semiconductor GaAs(Si) 300E+16 7500 14

Semiconductor GaAs(Si) 100E+17 7500 14

Ohmic GaAs(Si) 500E+17 7500 14

Buffer GaAs(Si) 300E+18 7500 14

Substrate GaAs(Si) N+ 3000 14

The fabricated SBD was also measured and its I-V characteristic is plotted as a

function of temperature in Figure 415 Unlike the ASPAT diode the current at forward

bias for the SBD change enormously with temperature from 77K to 398K and at all

biases For the ASPAT diode the slight change in current only started after 08V bias as

the current starts to have some component of thermionic emission over barrier

Figure 415 Log Current vs voltage as a function of temperature for SBD sample XMBE104

In order to see clearly how much the current is changing in forward bias for both

ASPAT and SBDs diode a log current at different voltages versus 1000temperature is

plotted as shown in Figure 416

1E-08

00000001

0000001

000001

00001

0001

001

01

-2 -1 0 1 2

Log

Cu

rren

t (A

)

Voltage (V)

T=398K

T=375K

T=350K

T=325K

T=300K

T=275K

T=250K

T=225K

T=200K

T=175K

T=150K

T=125K

T=100K

T=77K

146

Figure 416 Influence of temperature on IV characteristic for both ASPAT diode and SBD

Semi-logarithmic plots of current (at V= 05 06 07 08V) versus inverse

temperature for both SBD and ASPAT are shown in Figure 416 When the temperature

is increased the current also increases in the SBD as a result of thermionic emission

over the barrier for sample XMBE104 This is in contrast to the temperature-

independent tunnelling through the thin AlAs barrier of sample XMBE304 ASPAT

diode where when the temperature is increased the current is almost constant

At low and high temperatures the ASPAT shows excellent temperature

independence with a constant current flow It exhibits a tunnelling current in excess of

values expected by the elastic tunnelling current calculation equation suggested by RT

Syme [16 18] above (Eq 1) using a Oslash value of 105eV (ΓGaAs to Γ AlAs tunnelling) By

contrast for the SBD at low temperature (77K-275K) the changes of currents were very

high and for every 02V there is an exponential change of more than 40 This

temperature dependent study was also reported in[68]

147

48 Conclusions

This chapter demonstrated the establishment of an excellent physical model and

comparison of room temperature I-V characteristics of GaAsAlAs ASPAT diodes for

different emitter sizes their scalability as well as an investigation of their characteristics

at different temperatures from 77K to 398K Simulation are validated on well-

characterized experimental data and excellent fitting which had been achieved in this

work permit the designer to extract all related parameters of heterojunctionmultilayer

ASPAT structures thus creating modification for future growth specification in order to

achieved precise designs

It is clear that the work which had been carried out in this chapter is able to

achieve with adequate accuracy a claim of reverse engineering capability The ability of

the GaAsAlAs ASPAT to act as a zero-bias detector has been analysed and compared

with the SBD It is clear that the temperature stability which is shown by the

GaAsAlAs ASPAT is much better than that of the SBD thus demonstrating that the

tunnelling current is dominant over the thermionic emission in ASPAT diodes

148

5 DC amp RF CHARACTERISTIC OF ASPAT DIODES

51 Introduction

To assist in circuit designs for any type of high-frequency circuits such as

millimetre wave detectors frequency multipliers and mixer circuits which are built

based on non-linear devices (diodes) an equivalent-circuit model for the diode is

required This is among the simplest and most effective method for analysing

semiconductor devices which work at high frequency where the electrical characteristics

measured obtained from the devices are extracted and presented in a circuit consisting of

lumped elements components (resistor inductor capacitor etc) However accurate DC

and RF measurement data is essential to extract the equivalent-circuit elements quickly

and correctly The extracted parameters values from the circuit that are taken into

account usually depends on bias and frequency associated with the device physically

which is also interrelated to the semiconductor material parameters device structure as

well as fabrication process flows

In this work the DC and RF data were derived from DC and S-parameter

measurements respectively These measurements were carried out both in-house and at

the University of Cambridge by a collaborator partner (Prof MJ Kelly) The I-V

characteristics of the diode obtained from DC measurements were measured from -2V to

2V while the S-parameters were carried out over a wide frequency range from 40MHz to

40GHz with eight different biases In this chapter the DC measurements for various

sizes of the diodes with analysis of their IV characteristics will be discussed The one-

port on-wafer ASPAT measurement setup as well as the de-embedding method will

also be explained Thereafter the equivalent circuit models with all lumped element

effect will be discussed This work is carried out with the help of the VNA which

principle has been described in Chapter 2 and Keysight ADS simulation tool All

technical details regarding the equivalent circuit models will be explained together with

the method used for the ASPAT diode evaluation The equivalent circuit model also will

cover the diode intrinsic elements such as Cj Rj and Rs and extrinsic elements ie CP

149

and RP Finally an equivalent circuit model with the small signal characterization of the

fabricated ASPAT diodes will be presented

52 General Analysis of Current-Voltage (I-V) characteristics of GaAsAlAs

ASPAT diodes

The recent development of state-of-art for DC measurement apparatus has led to

capabilities for high-level accuracy of measuring voltages to a few nano Volts and

current signals in the femto Amp range[107] This can easily be obtained by exploitation

of proper connections and high-quality cables connecting the equipment to the Device

under Test (DUT) In this work the DC measurements were carried out using an Agilent

B1500A Parameter analyser whose description was covered in Chapter 2

As was discussed in Chapter 3 the GaAsAlAs ASPAT diodes have been

fabricated with different mesa areas between 2times2microm2 to 100times100microm

2 but the smallest

size obtained with good I-Vs was 4times4microm2 In this section the focus will be on how the

extracted data can be expanded further for empirical modelling Figure 51 shows typical

results for measured ASPAT diodes with various dimensions to check for their

uniformity According to our standard procedure the DC measurement has to be

conducted prior to the RF to ensure the diode is in fully working order as this will later

save a lot of time during RF characterization

Figure 51 IV Characteristics of measured ASPAT diodes for emitter sizes of 4x4um2 6x6um

2 and

10x10um2 Note the good scalability

1E-10

1E-09

1E-08

00000001

0000001

000001

00001

0001-2 -1 0 1 2

Cu

rre

nt

De

nsi

ti (

Amicro

m2)

Voltage (V)

4x4microm^2

6x6microm^2

10x10microm^2

150

Figure 51 above demonstrates the IV Characteristics of measured GaAsAlAs

ASPAT diodes (XMBE304) for emitter sizes of 4times4um2 6times6um

2 and 10times10um

2 This

sample was processed using the dielectric bridge technique developed in this work It

can be observed that current per unit area for each dimension fits and scales to each

other The scalability of each diodes measurement is very important to ensure no process

related issues are hampering the devicersquos proper operations This also confirms that the

diodes are completely functional and can be used for the next stage of measurements

The advantage of having an excellent scalability of those diode sizes is that a prediction

of smaller emitter size can be made

This type of IV characteristic shows asymmetric behaviour which results from

the unequal spacer lengths of the device This behaviour is very useful for detection

application as it obeys a square law model The square law predicts that the current is

proportional to the square of the applied bias

119868 = 1198861198812 119908ℎ119890119899 119881 gt 0

119868 = 0 119908ℎ119890119899 119881 lt 0

(51)

To extract the first order effects of ASPAT diodes DC measurements which

result in asymmetric I-V characteristics are analysed The slope of the non-linear region

is used to determine the junction resistance (Rj) which is obtained from the first

derivative of voltage versus current (dVdI) The expression of Rj is given by

119877119895 =

120597119881

120597119868

(52)

In order to understand the relationship between Rj and diode sizes of the ASPAT the IV

characteristic for each diode displayed in Figure 51 is used to extract the Rj This has

been done by using the expression in equation (52) above and their response is plotted

against bias as displayed in Figure 52

151

Figure 52 Junction resistance versus voltage

As can be seen in the Figure 52 above the Rj for each device decreases strongly

when the voltage increases At zero bias the 4times4um2 devices show the highest Rj value

followed by 6times6um2 and 10times10um

2 devices The junction resistance at zero bias obtained

from the 4times4um2 diode is around 86KΩ while reducing by a third for the 6times6um

2 and

10times10um2 diodes with Rj of 27KΩ and 10KΩ respectively A diode with a smaller

forward current under the same applied voltage will exhibit a larger Rj For a good

millimetre wave detector a device with a large value of Rj is desirable since it will

provide high detection sensitivity

The slope at the IV characteristic contributes to an important parameter that is

commonly used by electronic manufacturers to describe diode specification namely the

video impedance (RV) which is also known as the non-linear resistance The RV which is

extracted from the real part of the diode small signal impedance is highly dependent on

the DC bias current and only weakly depends on the series resistance of the diode (RS)

Therefore the video impedance is given by

119877119881 = 119877119895 + 119877119878 (53)

Where RS is the series resistance of the diode whose value is normally very small and

does not contribute much to the whole slope and hence RV is dominated by Rj The RV

changes in behaviour if any DC current is flowing through the diode Practically small

DC current in the range of 1 to 10 microAmp or total zero bias is used to maintain the

appropriate RV value (1-2KΩ to several MΩ) RV will also determine the voltage

-10

10

30

50

70

90

110

130

150

-01 0 01 02 03

Rj(

)

Voltage (V)

6x6um^2

10x10um^2

4x4um^2

152

sensitivity of the whole detector circuit This will be explained further in the next

chapter In the case of a detector with an amplifier RV of the diode acts as the RF

impedance which needs to be matched with the video amplifier ( impedance looking into

the diode from the amplifier)[108 109]

The quotient of the second order derivative to the first derivative

((d2IdV

2)dIdV)) when calculated from the whole I-V characteristic translates directly

into a curvature coefficient (k) This is the most commonly used figure-of-merit to

quantify diode nonlinearity at zero bias Figure 53 below shows the variation of k with

bias and more importantly the zero bias rectifying action for device sizes of 4times4 um2

6times6 um2 10times10 um

2 This parameter which represents the small-signal rectifying

action of the diode will affect the performance of the detector (voltage sensitivity)

Detailed discussions on how this parameter effect the detector performance will also be

discussed in the next chapter

Figure 53 The rectifying action of the diode at zero voltage which gives the ASPAT high sensitivity

near zero bias detection

Figure 53 above shows calculated curvature coefficient of the measured I-V

characteristics from the same diodes shown in Figure 51 The highest k value is

obtained from the diodes with size of 4times4um2 followed by 6times6um

2 then 10times10um

2

The curvature coefficient decreases sharply as the bias increases for each diode This can

be attributed to a significantly increasing number of electrons that tunnel through the

thin barrier which were accumulated in the 2DEG formed in the intrinsic spacer region

-5

0

5

10

15

20

25

30

-001 004 009

Cu

rvat

ure

Co

effi

cien

t(V

-1)

Voltage (V)

k(10x10 um^2)

k(6x6 um^2)

k(4x4 um^2)

153

An ASPAT diode with a smaller size will have a larger Rj with a corresponding smaller

current under the same bias condition and hence will demonstrate a larger k value In

this calculation the curvature coefficient at zero bias obtained from 4times4um2 6times6um

2

and 10times10um2 diode is 23V

-1 17V

-1 and 16V

-1 respectively

A summary of the ASPAT diodes parameters obtained from measured I-V

characteristics that have been translated into first and second order differentials are

gathered in Table 51 below and compared to other diodes in the literature

Table 51 Performance of the ASPAT diode obtained from measured IV characteristics in this work

Sample Rj(Ω) k(1volt)

ASPAT 10times10 microm2 10K 16

ASPAT 6times6 microm2 27K 17

ASPAT 4times4 microm2 86K 23

Ge Backward diode 182K[110] 159[110]

InGaAs Backward diode 154[110] 23[110]

Sb Backward diode 5K[111] 47[111]

Si-Backward diode 135K[112] 31[112]

PDB 15K[8] -

AlGaAs SBD 20-100K[113] 34-38[113]

GaN HBD - 16[114]

From Table 51 above it is clearly that the ASPAT diode has a comparable value of Rj

and k to existing detector diode in the research community and in the commercial

market Based on literature of each diodes stated in the table the key to obtaining a high

value of k at zero bias is to minimize any forward tunnelling current Furthermore the

largest ASPAT diode used here (10times10microm2) has very close performances to that of a

commercial diode ie discrete Ge backward diode (ref[110]) where both Rj and k value

are close to each other

53 RF Test Fixture Theory and Experiment

RF measurements differ from DC measurement as they are more complicated

and it is necessary to comprehend the basic measurement principles to achieve

meaningful data This is obligatory especially for on-wafer RF characterization and

154

analysis to attain precise results Most of the electronics component measurements

which have input and output for instance antenna amplifier cables etc are based on a

two-port network configuration The characteristics which can be extracted from these

components are usually used to define their impact on a more complicated system

The performance of the two-port network can be described by a few parameters

ie scattering (S-Parameter) admittance (Y-Parameter) Impedance (Z-parameter) and

Hybrid (H-Parameter) However the S-parameter approach is favoured for high-

frequency measurements as it is relatively easier to characterize the microwave

performance and is able to convert to other parameters when necessary The advantage

of S-parameters is that they can straightforwardlydirectly convert into other two-port

parameters as mentioned above in term of currents and voltages[115] In fact to obtain

the device capacitance the appropriate S-parameters needs to be transformed into Y-

parameters using specific equations Furthermore the devicersquos cut-off frequency can

also be obtained when S-parameter measurements are performed over a wide frequency

range

531 On-Wafer Measurement and Small Signal One-Port Characterizations

In this work the arrangement of the RF measurement setup is assumed to be a

linear system as small voltage amplitude signals are used this means that the signals

have only a linear effect on the network without any gain compression or attenuation

The assumption is still acceptable even though the typical ASPAT is characterised as

non-linear in nature because it is a passive device which will act linearly at any input

power level

Generally the S-Parameter measurements on a diode can be adequately and

suitably performed using a one port measurement The technique used to characterise the

output is similar to the two ports technique but only incident and reflected waves are

used to characterise the input and output ports of the device Essentially this is because

the ASPAT has only two terminals and it is a passive device like other diodes

Therefore the analysis will revolve around the S11 parameter Figure 54 below show the

S11 is a ratio of reflected wave to the incident wave

155

11987811 =

119877119890119891119897119890119888119905119890119889

119868119899119888119894119889119890119899119905=

1198871

1198861 119908ℎ119890119899 1198862 = 0

(54)

A VNA as described in Chapter 2 is used to measure the ASPAT diodes This

powerful equipment is able to measure S-parameters up to 40 GHz To conduct accurate

S-parameter measurement at the diode the measurement setup must be calibrated prior

to the actual measurements taking place

54 Device Calibration

541 Open and Short De-Embedding Technique

Further calibration to be made involves anything related or attached to the

device The co-planar waveguide (CPW) bond pad and interconnect line that are

attached to the intrinsic diode are the main contributors of the errors also are required to

be calibrated In general the bonds pad could generate a capacitance (parasitic) in

parallel with the intrinsic diode and its contribution depends on the size of the bonds pad

as well as the operating frequency Meanwhile the CPW and interconnect line may

cause a parasitic inductance in series with the diode

The method that is used to get rid of this parasitic is called de-embedding and the

most common technique to realise it is by introducing OPEN-SHORT structure[116]

This method is based on a lumped-elements model Parasitic elements of the diode

De-Embedding Structure

Incident wave

Reflected wave

One-port device

a1

b1

Figure 54 One port S-parameter measurements

156

equivalent circuit correlate directly to the access section of the CPW hence can be

derived from de-embedding structures The aim of the de-embedding technique is to

represent these parasitic elements so that the one-port characteristic of the actual diode

can be determined

The two types of de-embedding structure OPEN and SHORT are conventional

techniques that are widely used in this study The design of all structure must be

identical (in size) to the device to avoid any discrepancy It is very simple to design all

these three structures for example open structures are obtained by eliminating the diode

layout and keeping the bond pad layer only The short structure just adds a bridge and

ensures ground and signal pad are connected to each other Through structures are

realised by disconnecting both ground pad and leaving the signal pad to connect to each

other

To gain more accuracy this external effect must be removed by the implementation of

de-embedding structures on the same tile as the actual device Figure 55 shows the

fabricated de-embedding structures used in this study

Figure 55 Fabricated open short de-embedding structure and real device (zoomed) use for RF

calibration and measurements (Note Images are not to scale)

In summary the de-embedding which is used to extract out the parasitic elements

from entire single diode measurement is a very important step as normally on-wafer

measurement requires coplanar waveguide (CPW) to access the diode structure (active

region) The CPW will have some effects which will disturb the accuracy of the device

characteristics

157

55 S-Parameter Measurement Result and Analysis

This section will only present RF measurement results after all VNA setup and

calibration were performed The S-Parameter measurements were carried out on ASPAT

diodes at five different DC biases from -2 to 05 volt with a sweep frequency from

40MHz to 40GHz using a calibrated VNA and the input power was fixed at -30dBm

The measurement procedure as described in Chapter 2 was performed on the device

(on-wafer) equipped with the appropriate bond pads This is important to ensure the

results obtained are valid The reason for using different biases is to find at what voltage

the device capacitance is fully depleted This is also very important in determining the

cut- off frequency of the devices

In this research two phases of the experiment on the S-parameter measurements

were carried out The first phase is to qualify the process flow ie for manufacturability

and repeatability which can be obtained from the consistency of the result The S-

parameter measurements taken on the same wafer dies are repeated several times on

different GaAsAlAs ASPAT diodes There are three different tiles taken from 3

different wafers namely XMBE304A XMBE304B and XMBE304C carried out in

this experiment The repeatability tests are done mostly on the large devices (15times15microm2

up to 100times100 microm2) and the results are analysed based on the reflection coefficient (S11)

on Real and Imaginary measurements

In the second phase the measurement is toward producing devices that can

perform at high-frequencies This can be realised by utilising small emitter size devices

(4times4microm2 6times6 microm

2 and 10times10 microm

2) The measurement results of these devices will be

used to build the equivalent circuit models while both the intrinsic as well as the

parasitics of the device will be evaluated Hence all the values obtained from these S-

parameter measurements will be used to design the device that can be used in

millimetres-wave applications As can be seen in Figure 56 below the extracted S-

parameter measurement results comprise of a reflection coefficient (S11) for real

imaginary and Smith chart for XMBE304A While these measurement results are

extracted at zero bias voltage the other bias voltages will be used to extract the

capacitance This will be described in the final section of this chapter

158

551 Diode to diode uniformity

In order to study within tile uniformity and reproducibility statistics of the RF

performance five devices of different mesa sizes in the same tile (XMBE304A) were

measured at zero bias and represented in term of Real and Imaginary reflection

coefficient (S11) the uniformity check is carried out at three different frequencies step

under 15GHz since the cut-off frequency for these big devices is relatively low at about

~20GHz on average The variation of the reflection coefficient is taken from the

percentage of the (standard deviationmean values) for all five diodes from this run The

following figures show Real and Imaginary S11 of large mesa area ASPAT diode from

15times15microm2 up to 100times100microm

2 which are represented by lines graph in a few different

colours

Figure 56 Real S11 RF measurements for (XMBE304A) were done for four devices from

15times15microm2 to 100times100microm2 device sizes for within-wafer uniformity check

Figure 57 Imaginary S11 measurements (XMBE304A) to qualify the repeatability and

reproducibility were done on four devices from 15times15 microm2 to 100times100 microm2 at zero bias

159

The variance data extracted from the graph (Real S11) above for each device within-

wafer (device to device) uniformity study is summarize in Table 52 below

Table 52 Device to device uniformity check for large ASPAT diode

Device Size 100times100 microm2 50times50 microm

2 30times30 microm

2 20times20 microm

2 15times15 microm

2

Variation 5GHz 181 115 405 151 145

Variation 10GHz 106 133 522 424 293

Variation 15GHz 119 198 281 76 509

The majority of diodes show that the variations of S11 measurements are below

3 and only a few are below 8 These finding still can be considered as good for

manufacturing control since absolute I-V characteristics reported in [63] is set by

designerrsquos specification to be not more than plusmn10 variation Further extensive RF

measurements were carried out by the research collaboration with the University of

Cambridge on the same GaAsAlAs ASPAT diodes wafer [117] In their study they

focused on 50times50 microm2

mesa size 17 of diodes were chosen to be measured The study of

uniformity of RF characteristic only focused on frequencies below 20GHz The same

approach was used to get the variation of the reflection coefficient for all 17 diodes but

this work was carried at four different frequencies Table 53 shows the zero bias S11

result for four different frequencies and standard deviation of the devices

Table 53 The variation of reflection coefficient (S11) for GaAsAlAs ASPAT diode at four different

frequencies[117]

Frequency (GHz) 5 10 15 20

Variation () 197 243 26 276

From the results the variations of 50times50 microm2 mesa sizes measured in-house and

at the University of Cambridge are comparable with all variations showing good

uniformity ie recording variations below 3 This indicates that the RF performance

of the GaAs AlAs diode is valid and reproducible and is thus considered as a good

achievement for manufacturing Once the reproducibility and repeatability of the large

devices showed stable results the fabrication process then continued to obtain smaller

emitter size for work at high-frequencies

160

552 Wafer to wafer uniformity

Other RF measurements were conducted on sample XMBE304B which was

fabricated in-house using the same process steps but the only difference from

XMBE304A was the use of SiN3 as dielectric In this run three different mesa sizes

were measured (15times15mmicro2 20times20microm

2 and 30times30 microm

2) and Real and Imaginary S11

plotted against frequency The RF performances of both samples are gathered in one

graph as shown in Figure 58 below

Figure 58 Wafer to wafer uniformity test on 1st and 2

nd RF measurement(XMBE304B) to qualify

the repeatability and reproducibility were done for 30times30microm2 to 15times15 microm

2 device sizes (Real and

Imaginary) Note blue colour is XMBE304A and red colour is XMBE304B

For this wafer to wafer uniformity study four diodes with three different sizes as

specified previously were measured from sample XMBE304B and four diodes from

previous measurements of XMBE304A The blue line in Figure 58 represent

measurement result of real and imaginary for sample XMBE304A while the red line

161

represents XMBE304B The uniformity data is compared at three different frequencies

and gathered in the table below

Table 54 Wafer to wafer uniformity check on XMBE304A and XMBE304B

Device Size 30times30 microm2 20times20 microm

2 15times15 microm

2

Wafer A vs wafer B variation 5GHz 305 31 1 314

Wafer A vs wafer B Variation 10GHz 352 344 329

Wafer A vs wafer B Variation 15GHz 321 376 359

As can be seen in the Table 54 above the wafer to wafer uniformity is rather

large (30) on average The main reasons being that sample XMBE304A was

processed by utilizing S1805 as a dielectric layer while sample XMBE304B used

Si4N3 Although the process steps are similar for both wafer processing the use of

different dielectric layer will influence the diode parameters especially resistance and

capacitance as the dielectric constant for each materials is different Secondly the wafer

processing is not run concurrently at the same time thus the moisture and temperature in

the clean room might differ for both processing Although the wafer to wafer uniformity

test for this run might not be favourable for manufacturing tolerance at least the use of

different dielectric layer shows some significant result in term of capacitance resistance

effect to the GaAsAlAs ASPAT diode

553 Small devices RF measurements

The first objective of this study was to make smaller size mesa devices ie

1times2microm2 1times3 microm

2 2times2 microm

2 and 3times3 microm

2 However for GaAsAlAs ASPAT type this is

difficult to achieve in practise These issues were discussed in detail in Chapter 3

Hence the smallest emitter size that yields repeatable and reproducible results was

4times4um2 The final measurement which was done on sample XMBE304C focused on

small devices The measurements were done on four devices two with the diode bond

pads sitting on substrate (GaAs SI) and the other two sitting on dielectric layer (Si4N3)

Figure 59 below shows three measured results obtained from sample XMBE304C

using the 3rd

Gen Mask

162

Figure 59 RF measurement performed on device sizes of 10times10microm2 6times6mmicro

2 and 4times4microm

2 (Real and

Imaginary) Note that green red and blue colour represents 4times4microm2 6times6mmicro

2 and 10times10microm

2 diodes

respectively

Figure 510 RF measurement performed on device sizes of 10times10 microm2 6times6 microm2 and 4times4 microm2

(Smith Chart) Note that green red and blue colour represents 4times4microm2 6times6mmicro2 and 10times10microm2

diodes respectively

30MHz

40GHz

163

As can be seen in Figure 59 (Real Imaginary) and Figure 510 (Smith Chart) are

obtained from measurement of four diodes in the same tile The diode to diode

uniformity that is sitting on the same platform obtained at 35GHz frequency in this run

on average is ~15 25 and 1 for 4times4microm2 6times6mmicro

2 and 10times10microm

2 diodes

respectively On the other hand the uniformity between diode to diode sitting on

dielectric and substrate is quite high due to different capacitance value of devices on

average ~7 are attained from three different sizes of diode

From the measurement results above the Real S11 measurement of four different

sizes show the same trend for each frequency At low frequency resistances for each

diode is high as the S11 value is large At intermediate frequency the values drop

tremendously for big devices (30times30 mmicro2 and 20times20 mmicro

2) ie in Figure 57 At high

frequency all diode reach saturation limit as the value are constant Small devices

(4times4mmicro2 6times6mmicro

2 and 10times10microm

2) ie Figure 59 show S11 values that are higher than

those of large devices as smaller emitter diode have larger resistance value

The imaginary S11 value also shows the same trend as for big devices However

for small devices in this run (4times4mmicro2 6times6mmicro

2 and 10times10microm

2) the S11 value keep

dropping toward negative values at increasing frequencies This indicates that bigger

devices with positive value at high frequency are more capacitive than the smaller

devices It is worth mentioned that the capacitance and inductance values for device

sizes of 4times4microm2and 6times6microm

2 come from the CPW layouts and these are dominant while

for device sizes of 15times15 mmicro2 and above the device capacitance itself is dominant

The Smith Chart shows the reflection coefficient (S11) as a function of the

applied frequency (30MHz to 40GHz) All measurements from each mesa size follow

unique impedance circle which is that most of the lines are at the lower right outer ring

This means that the diode capacitance value is frequency dependent For the case of

10times10microm2

devices these impedance circles are mostly toward the outer ring meaning a

higher capacitance than the other two mesa dimensions All the device constantly follow

the outer ring without crossing any real axis at any frequency point meaning that the Cj

is not shorted at the maximum 40GHz measurement frequency (not reached cut-off

frequency) Therefore the entire small GaAsAlAs diodes in this run have capability to

work in the millimetre wave frequencies range

164

56 Extracting RF models of ASPAT at Zero Bias Voltage

The methodology used in the S-Parameters measurement for high-frequency

analysis must ensure that the derivation of the equivalent circuit corresponds to their port

characteristics In other words the component representing the ASPAT in the equivalent

circuit model must have physical significance otherwise the circuit will be meaningless

The fabricated ASPAT diodes as discussed in Chapter 3 have the cross section shown in

Figure 511

There are two main components that can be extracted from the fabricated

ASPAT depicted above ie intrinsic and parasitic The intrinsic refers to the main

structure of the diode itself and are represented by three bias dependent elements

namely Junction Resistance (Rj) Junction Capacitance (Cj) and diode Series Resistance

(Rs) The parasitic is the elements related to the bond pad of the anode and cathode as

well as interconnects They are represented by parasitic inductance (LP) resistance (RP)

and capacitance (CP)

The diode parameter extraction is different from the three terminal devices

(FETs) in the sense that FETs are a kind of direct extraction in which all the elements in

the transistor have linear functions to the port characteristics ie S-parameter Y-

Parameter Z-Parameter and can easily be solved by the matrix calculation method for

those particular parameters[118] The same extraction method cannot be applied to the

diodes because its elements will embroil with each other Therefore only one method is

used to extract the diode element which is optimisation by tuning the initial value toward

the measured S-Parameter values

Figure 511 Cross-sectional of fabricated ASPAT diode and the corresponding equivalent circuit

model

165

The strategy used to model the ASPAT is based on an initial fitting value of the

lumped elements to the extracted value from measurement on three S-parameter graphs

(real imaginary and Smith chart) for the reflection coefficient (S11) The refinement is

accomplished by optimisation and fine tuning of the values which result in minimum

error between extracted and modelled values Figure 517 (on page 153) shows fitted S-

parameter result for extracted and model numbers with each one fitted in a single line as

an example However to achieve this excellent fitting key prior steps have to be used

de-embedding fitting the intrinsic value and optimisation

561 Extraction of ASPAT parasitic element

Once the S-parameter measurements achieve stability repeatability and

reproducibility for each measurement in term of S11 results as mentioned above the

results of the de-embedding structure which had been measured prior to the device

structure are then extracted to form a well-defined equivalent circuit In order to build

and analyse the equivalent circuit firstly the measured data is imported into the ADS

software prior to any fitting This can be realised via the ldquoStart The Data File Toolrdquo

features provided by this particular software When successfully imported the data is

read by the function S2PMDIF (These files are a natural extension of two-port S-

parameter Touchstone files) as depicted in Figure 512 below

Figure 512 The S-parameter Touchstone file is used to read the measured files

166

For the open and short techniques after de-embedding the equivalent circuit

model which is represented by mainly a capacitor and an inductor is built The open

structure requires resistance and the capacitance values of 20KΩ and 26fF respectively

connected in parallel to be well fitted to the real imaginary and Smith chart (S11) output

On the other hand for the short structure the Real imaginary and Smith chart (S11) have

to satisfy the values of resistance and inductance elements of 1Ω and ~47pH respectively

connected in series These values strongly rely on the bond pad or CPW dimension and

length The Equivalent circuit models and fitted data as well as measurement can be

seen in Figure 513 and Figure 514 below

Figure 513 Equivalent Circuit Model of Open CPW or de-embedded structure

Figure 514 Equivalent circuit model for short de-embedded structure

To satisfy the equivalent circuit a self-consistence method introduced by

Ren[119] is utilised This approach accurately extracts the CPW capacitance (Cpad) and

inductance (Lpad) as well as intrinsic Junction capacitor (Cj) which is attained from the

one-port S-parameter measurements Therefore the pad capacitance introduced by the

self-consistence method for the open structure can be expressed by

167

119862119875 =

119868119898(11988411119874119901119890119899)

120596

(55)

Lpad which represents the short structure is given by

119871119875 =

1

120596(119868119898 (11988411119874119901119890119899minus 11988411119878ℎ119900119903119905

))

(56)

Here Y is the admittance parameter (Y-parameter) converted from the S-parameter

measurement data and ω is the angular frequency The extracted measurement data

represented in the equivalent circuit fits with the simulated data in three S11 graphs as

can be seen in Figure 515 below From the Smith Chart it can be clearly seen that both

open and short S11 results are on the circumference which means the resistance of the

short structure is very small while in the open it is very large Additionally the

calculated Cpad and Lpad using Equations 55 and 56 above produce results similar to

those obtained in the equivalent circuit model for the open and short structure The

values are ~25fF and ~45pH respectively These data completely verify and validate

both results

Figure 515 Smith chart representative S-parameter measurement for short (left) and open (right)

CPW The blue lines represent simulated data and the red is measured data

Short

Open

168

562 Extraction of ASPAT intrinsic elements

Once the parasitic elements are determined it is easy to build a complete ASPAT

equivalent circuit The ASPAT is not like other tunnelling diode which their equivalent

circuit models widely studied ie RTD [120] IMPATT and PDB The only literature

which reports ASPAT equivalent circuits can be found in [15] and other RT Symersquos

journal paper[16] Fortunately its equivalent circuit model is not much different

compared to other diode video detectors Thus other literature which is based on

Schottky diode equivalent circuit model used for detector application can be referred to

The simplest form of ASPAT equivalent circuit and other video detectors intrinsically

consist of junction capacitance (Cj) series resistance (RS) and junction resistance (Rj)

First and foremost to extract the equivalent circuit one must know the theory behind

each parameter that is developedbuilt as a spine to become a complete element This is

vital to ensure the equivalent circuit is correct In the case of the ASPAT Cj is predicted

from a simple fully depleted parallel plate capacitor approximation which was discussed

previously in Eq (230)[15] Additionally for the S-parameter measurements the Cj can

also be validated by the self-consistence method mentioned earlier and thus can be

expressed by

119862119895 =

[

(1

120596)

1

1

119868119898 (11988411119905119900119905119886119897minus 11988411119874119901119890119899

)+

1

119868119898 (11988411119874119901119890119899minus 11988411119878ℎ119900119903119905

)]

(57)

This approach helps to verify both the fully depleted parallel plate capacitor in S-

parameter measurements The basic component which is responsible for the ASPAT

series resistance RS was discussed in detail in Chapter 2 RS and Cj are key contributors

to the high-frequency operation as expressed by the device cut-off frequency Equation

(58) below

119891119888119906119905minus119900119891119891 =

1

2120587119877119904119862119895

(58)

The R and C parameters must be kept as low as possible in order to obtain high cut-off

frequencies for millimetre wave applications From the fabrication point of view Cj can

169

be reduced by making as small a diode emitter size as possible while for RS reducing

the D gap is of paramount importance as it dominates the series resistance The ASPAT

contact resistance in the electrodes (contacts between metal and semiconductor) can be

reduced by using high doping in the ohmic layers

The junction resistance (Rj) of the ASPAT is taken from the 1st derivative or

slope of the current-voltage characteristics Normally the value of Rj is very large

(several kilo Ω) compared to Rs The small signals ASPAT equivalent circuit built with

intrinsic and extrinsic components is shown in Figure 516 below while the fitting

results is shown in Figure 517 and Figure 518

Figure 517 Excellent fitting between measured (red colour) and simulated (blue colour) for S11

(Real and Imaginary) results for various small device designs

Rj

Cj

Cpad Rpad

Rs Lpad

Figure 516 Equivalent circuit of the ASPAT diode

170

Figure 518 Excellent fitting between measured (red colour) and simulated (blue colour) for S11

results (Smith Chart) for various small device designs

The equivalent circuit that was built for the ASPAT is taken from sample

XMBE304C with emitter dimensions of 4times4 microm2 6times6 microm

2 and 10times10 microm

2 These

devices are expected to work in the millimetre-wave region and have cut-off frequencies

(intrinsic) of ~650GHz ~200GHz and ~100GHz respectively

Table 55 Comparison between calculated (fully Depleted) and extracted (different biases) values

from equivalent circuit parameters for different ASPAT mesa sizes at zero bias voltage

Parameters 4times4microm2 6times6 microm

2 10times10 microm

2

Calculated

(fully

depleted)

Extracted Calculated

(fully

depleted)

Extracted Calculated

(fully

depleted)

Extracted

0V -05V -1V 0V -05V -1V 0V -05V -1V

Cj(fF) 879 23 7 6 198 55 148 139 549 171 486 46

Cpad(fF) - 15 15 15 - 15 152 152 - 15 15 15

Lpad(pH) - 45 43 42 - 50 473 473 - 51 51 46

Rj(KΩ) - 90 833 522 - 35 392 392 - 12 125 13

Rs(Ω) 99 11 11 11 67 95 8 7 41 95 45 37

fcut-off

Cj(GHz)

1828 629 2066 241

1

1208 192 1344 163

5

710 107 728 935

fcut-off

Cj+Cp(GHz)

- 380 658 688 - 151 663 781 - 98 556 705

171

The focus in this study is purposely to build ASPATs as zero bias detectors that are able

to work in the millimetre and sub-millimetre frequency range therefore all the

parameters which are obtained from equivalent circuit were extracted at zero bias

voltage Theoretically the calculations which are derived from both self-consistence amp

theory can only be solved for fully depleted device capacitance (using a parallel plate

configuration) Hence both extracted and calculated results are compiled in Table 55

Noticeably the calculation can only produce the intrinsic parameters of the

ASPAT for fully depleted capacitance On the other hand both intrinsic and extrinsic

parameters of GaAsAlAs ASPAT are obtainable from extraction and thus help to

determine at what bias the diode is start to deplete The junction and series resistance (Rj

and RS) of each dimensions shown in Table 55 above were achieved by fitting the

elements of the equivalent circuit with the three measured S11 graphs whereas the Cpad

and Lpad were extracted by utilising the self-consistent method from the S-parameter

measurements which is fitting the de-embedding structure Additionally the Cj values

are obtained via fitting the measured S11 data and employing the self-consistence

approach Results obtained from both techniques are identical

At zero bias all extracted junction capacitance from each device sizes are very

different from the calculated one while the extracted series resistance are closer to the

calculation This means that Cj is a highly voltage dependent parameter and Rs is

voltage independent but solely dependent on device structure and material used to

fabricate it The extraction at -05V and -1V shows that the values of junction resistance

is changing for most of the devices which means this parameter also rely on bias voltage

as discussed earlier in Section 52

The cut-off frequency for each calculated devices are near the THz range even

for the 100microm2 emitter area However with the introduction of parasitics elements ie

pad capacitances fcut-off is degraded tremendously Therefore it is important to make sure

all the intrinsic elements have optimum values so that the target operating frequency of

the ASPAT diode can be met Due to this it is advisable to operate the devices at no

more than 13 of fcut-off when designing detector systems

The parameters extraction at -05 and -1V also show that Cj values are closer the

calculated ones which means the ASPAT diode is reaching full depletion

172

563 Capacitances -Voltage (C-V) Extraction

Theoretically the junction capacitance of the ASPAT is calculated from the fully

depleted formula 119862119895 = 휀119900휀119903119860119889 which was also discussed in Equation (230) in Chapter

2 in page 43 Its value depends on the change of voltages to depletion at the emitter

contact[15] and make it one of the voltage dependent parameter for the diode[121]

Therefore the C-V characteristic of the GaAsAlAs must be precisely extracted

In practice the capacitance is difficult to measure due to the very low resistance at zero-

bias However alternatively it can be measured and extracted by applying different

voltage and identifying the point at which there is change which essentially represents

full-depletion Apart from these it can also be extracted from S-parameters measurement

which is then converted to Y-parameters A C-V characteristic of the GaAsAlAs

ASPAT from XMBE304C is extracted and plotted as depicted in Figure 518 below

Figure 519 Capacitance extraction showing full depletion at -025 volts (modelled capacitance vs

Voltage)

From the graph shown in Figure 518 the capacitance is extracted at eight

different biases for 4times4 microm2 6times6 microm

2 and 10times10 microm

2 The devicersquos junction

capacitance for each dimension increases and reaches a maximum value at 025V There

are additional quantum capacitance effect which comes from an increase in the negative

charges in the 2DEG region (when band bending happens creating an accumulation

0

50

100

150

200

250

-2 -15 -1 -05 0 05

Cap

acit

ance

(fF

)

Voltage (V)

Cj(4x4microm^2)

Cj(6x6microm^2)

Cj(10x10microm^2)

173

region at the barrier) This charge is imaged by the positive charge in the whole

depletion region Increasing the voltage toward positive values leads to a lowering of

the AlAs barrier and thus allowing thermionic emission to take place after certain bias

values leaving only the depletion capacitance and making the quantum capacitance

negligibly small

In the reverse bias case the device junction capacitance reaches a saturation

(fully depleted capacitance) at a voltage of -025V and remain constant up to -2V If the

reverse bias voltage is increased further the ASPAT may reach breakdown Therefore it

is important to know how far the diode can withstand applied reverse bias to ensure it

can still give full performance

57 Conclusions

In this chapter scalable DC characteristics of GaAsAlAs ASPAT diode derived

from three different emitter sizes of 4times4 microm2 6times6 microm

2 and 10times10 microm

2 was

demonstrated The current density obtained at zero bias is several microAmicrom2 These allow

1st order differential effect to exhibit high value of Junction resistance (Rj) at zero bias

However Rj is highly bias dependent The 2nd

order differential effect on IV

characteristics display a high value of curvature coefficient leading to high voltage

sensitivity when applied in millimeter wave detector applications These two parameters

are vital in the design of millimeter wave detectors and especially those operating at zero

bias

Subsequently RF measurement up to 40GHz of uniformity study for both within

wafer and wafer to wafer variance were undertaken An average uniformity below 7

was obtained for within wafer study on large device area ( 15times15 microm2 to 100times100 microm

2)

while for small device area ( 4times4 microm2 to 10times10 microm

2) a smaller 3 uniformity variance

was achieved in average However for wafer to wafer study the variant uniformity was

quite high at around 30 on average for relatively large device (15x15 microm2 to 30x30

microm2) This was mainly attributed to different dielectric layers used in the process flows

of the sample rather than fundamental MBE control of the AlAs barrier thickness

174

It was demonstrated in this chapter that careful on-wafer RF measurements of small

size GaAsAlAs ASPAT diodes allow accurate device parameter extraction of both

extrinsic and intrinsic parameters The extrinsic parameters are namely pad capacitance

and inductance with obtained values of 26fF and 47pH respectively These values were

obtained from de-embedding structure fabricated on the same tile as the real devices

The intrinsic parameters such as junction capacitor junction resistor and series

resistance had different values according to device dimensions The smallest zero bias

value of Cj obtained from 4times4 microm2 diodes was 23fF ensuring a high cut-off frequency of

380GHz and hence in the next chapter a 100GHz detector will be presented working at

slightly less than 13 of this cut off frequency

The C-V data extraction confirmed that the fully depleted capacitance started to

happen at around -025V The maximum junction resistance occurs at +025V largely

caused by the depletion region and an additional quantum capacitance effect CQ This

effect is strongly related to the size of a 2DEG which occurs under forward bias (01V to

025V) and can be reduced by having a smaller thickness AlAs barrier

175

6 MILLIMETRE WAVE GAASALAS ASPAT DETECTOR

DESIGN USING ADS

61 Introduction

The ASPAT diode having features of non-linear IV characteristics at zero voltage

make it useful for signals rectification ie detector and mixer for millimetre-wave

applications Additionally ASPAT diodes have a range of advantages such as large

dynamic range strong temperature insensitivity etc[15]over other rectifier diode This

makes ASPAT an appropriate candidate in RF detection applications Since 1940 the

only two terminals device that has been the workhorse for RF applications is the

Schottky Barrier Diode (SBD)[9] In its earliest form the SBD was built based on a

point-contact device which could not perform at high frequencies It was then developed

to work at higher frequencies by exploitation of epitaxial structures [10] and to date the

SBD remains the mainstay of two terminal devices that are able to work in the

millimetre and submillimetre-wave regions However as discussed in Chapter 4 the

performance of SBD is degraded at extremes of temperatures and these circuits

employing SBDs require temperature compensating circuitry Thus there is additional

complexity associated with technologies and applications related to SBDs

Before this work was carried out no model for the ASPAT diode as detector had

been available or developed especially using the empirical modelling ADS software

When the ASPAT diode was first introduced its function was conceptually explained it

was then built and tested to compare with other microwave detectors at X-band

frequency (95GHz) The comparisons were made in terms of detector parameters ie

sensitivity dynamic range temperature dependence etc[15] These early works lead by

RT Syme et al supplied the basic knowledge to model the ASPAT diode as a zero-bias

detector for mm-wave frequency gt100GHz For the SBD many models and equivalent

circuit approaches have been reported [122 123] The modelling of conventional SBD is

mostly implemented through fitting the S-parameter curves of the model to the

experimental one This approach is also carried out in this research since it is accurate to

predict the performance of the device under test [124]

176

This Chapter aims to introduce low cost reliable and sophisticated detector design

based on ASPAT diodes which is believed to be able to improvereplace SBD in

millimetre-wave applications especially in imaging The focus was on developing and

establishing an appropriate circuit design that suit the new ASPAT diode for such

applications The detector sensitivity as its key parameter ultimately limits the quality

and acquisition time of the detector In the subsequence section the theory of detection

including both direct and heterodyne will be discussed This is followed by definition of

detector characteristics of interest as well as noise consideration Section 65 present the

main focus of this chapter which is the development of 100GHz ASPAT detectors and

their result will be explained in term of all detector characteristic of interest

62 Detection Theory

Any incoming signal such as RF microwave or mm-wave in the form of envelope

function or single wave can be detected by rectification of the signal using a nonlinear

device ie transistor or diode The input and output signal (RF) signals are normally in

the form of amplitude as a function of time Typically the detector output is a low-

frequency signal known as the video signal which has amplitudes that are proportional to

the square of the input RF signalrsquos voltage amplitude

A complete receiver system as shown in Figure 61 below consists of receiver

antenna and a circuit designed to extract the signal and then amplify it The function of

the receiver system is the converse of the function of the transmitter side At the receiver

side the antenna is used to receive the signal it then conveys the signal to the extraction

circuit for detection as the information-bearing part of the signal (using the nonlinear

device as its heart) The signal is finally amplified to avoid any information strength

decay Additionally in a digital system the output signal which had been processed by

the detector circuit has to maintain an optimum input signal conveyed to in-phase and

quadrature (IQ) demodulators The output signal will go through a low pass filter then

to an analogue-to-digital converter (ADC) and thus a digital baseband output signals

will be produced

177

Figure 61 Block diagram represent a complete direct receiver system

There are two types of millimetre wave integrated circuit (MMIC) used for

detection purposes namely direct detectors and heterodyne detectors The direct detector

MMIC is the simplest circuit used for detector applications and has the simplest way of

extracting the RF information Due to its simplicity the direct detection method is

inexpensive and most attractive method used for measuring power in RF Laboratories

and Industries This detection scheme is also sometimes known as video detection[125]

The simplest way of explaining the detection process is that the incoming RF or

microwave signals depicted in Figure 62 below with an appropriate input power (Pin) is

rectified by using a diode and results in a corresponding output voltage (Vout) A detector

IC designed based on diodes is normally able to rectify very low levels of RF power (lt-

40dBm) then produces an output DC voltage that is proportional to the RF power A

rectifier diode can function at zero bias (which is very good for reducing noise) at very

small DC bias (003mA) and relatively high RF impedance which will produce around

600Ω This will affect the capacitance value and a low capacitance is needed to realise a

high detection sensitivity

Figure 62 The detection process of a single wave through a non-linear IV characteristic of a diode

Detector

Speakerdisplay unit Amplifier

Antenna

Vout

Pin

Tuner Amplifier

178

However this type of detector has a drawback which is itrsquos relatively low signal to

noise ratio Thus it will also rectify any incoming electrical noise at all frequencies and

up to the cut-off frequency (fC) The basics lumped components circuit as shown in

Figure 63 is used to build such detector which consists of Source impedance (Zo)

Rectifier diode (ASPAT or Schottky) wire or pad inductance and capacitance and Load

impedance (RL)

Figure 63 Lumped element illustration of microwave detector circuit

Another type of detector is the heterodyne method which mixes incoming RF or

microwave signal (fRF) with another constant signal produced by a secondary circuit

called the Local Oscillator (LO) The LO frequency (fLO) must be slightly higher than fRF

to enhance the RF signal This mixing between fRF and fLO happens in the nonlinear

device as depicted in the Figure 64 This will produce a signal at a different frequency

called the intermediate frequency (fIF) which can then be amplified and detected as

explained in the previous paragraph Theoretically a basic requirement of the mixer is to

have fIF as efficient as possible while practically the minimum conversion efficiency

obtained is around 20 The main reason for using a mixer is due to the fact that

selective amplifiers at RF frequencies are costly and hard to achieve Hence a mixer is a

good technique as it only convert the signal to a lower frequency in which good

selectivity and high gain can be more effortlessly realised[14] A good mixer diode is the

one that can produce a high cut-off frequency and reduce conversion losses (Lc) A

mixer and detector diode with a low driven input power result in reducing overall noise

figure and thus in the ideal case the fIF amplifier also should have a low noise figure for

better performance The advantage of the heterodyne method is that it has a higher

179

sensitivity compared to the direct detection method this is achieved by producing an fIF

which has a lower frequency than the incoming RF signal[14] Obviously a zero-bias

voltage diode is more favourable to be used in mixer and detector applications

Figure 64 The mixing process where the signals are processed by the non-linear I-V characteristic

to produce an intermediate frequency (fIF=fLO-fRF) when signals at fLO and fRF are applied to the

diode

To have good detection efficiency for both types of detectors the operating

frequency (fO) must be several times smaller than fC In the case of an incoming

maximum modulated signal fM the frequency that can be acquired is in the range of fO

plusmnfM and will normally come with noise The standard method that is used to reduce the

noise is using a filter of bandwidth about 2fM at the centre of fO with the condition that fO

must be smaller than fM (f0lefM) Otherwise it would be difficult to attain However in

most cases fM is smaller than fO and fO is smaller than fC thus this will make the video

impedance (RV) (or nonlinear impedance) very close to the differential resistance of the

diode (at fO) in the equivalent circuit[74]

63 Definition of Characteristics of Interest Detection Parameters in small signal

analysis

Theoretically a transfer function measurement is preferable prior to any empirical

modelling since no assumption can be made due to the detector non-linearity

Furthermore measuring voltage output at high frequency can be very low while

measuring the power incident on the detector is hard to achieve where the linearity of

180

the typical power meter is normally less than 3 over its operating range[126]

Therefore the modelling of detector output voltage vs input power (ie transfer

function) can help to determine both nonlinearity correction and appropriate operating

range for the detector itself

The performance of the diode that is often taken into account is the transfer function

(output voltage Vout versus incident power Pin) and the main parameters that is used to

characterise and determine the quality of any detector diode are the voltage sensitivity

(βV) tangential sensitivity (TSS) dynamic range (P1dB) ie under 1 dB roll off power and

variation of output voltage when examined in extreme temperature situation (ΔV(T))

The voltage sensitivity in small signal analysis can use the approach introduced by

Torrey and Whitmer [9] then βV can be expressed as

120573119881(119894119889119890119886119897) =

119877119895119877119871120581

2(119877119881 + 119877119871) (1 +119877119904

119877119895)

2

[1 +1205962119862119895

2119877119878119877119895

1 + 119877119904 119877119895frasl]

(61)

Where ω is the angular frequency (2πf) Cj is the junction capacitance of the diode

active region RL is the load resistance RV is the video impedance taken from the

expression of Rj + Rs and κ is the curvature efficiency that give small signal rectifying

action of the ASPAT diode which is given by the second order term and itrsquos expression

is

κ =

11988921198681198891198812frasl

119889119868119889119881frasl

(62)

The curvature coefficient or responsivity (κ) is translated directly from the non-linearity

of the IV characteristics of the ASPAT diode for detector application Both RV and κ are

the parameters that can be extracted directly from diode DC measurement as discussed

in Chapter 5

The voltage sensitivity is actually a quantitative relationship between input

power and detector response Meaning that it is a change in signal output over change in

input power Normally output power is measured in Volts and input signal is measured

in Watts Therefore the unit of responsivity is VW[127 128]

181

TSS is referred to the lowest or minimum signal that the detector could detect it is

determined by the diodersquos βV and total noise available in the system (from the diode and

any amplifiers in the detector circuit) For any diode with fO=10 GHz and low noise of

1MHz bandwidth amplifier the TSS is typically less than -55 dBm[74] The TSS equation

is given by

119879119878119878 =

radic[4119884119870119861119879119861(119865119886 + 119905 minus 1)]

119872

(63)

Where M is a figure of merits and is derived based on the expression M= 120573119881 radic119877119881frasl t

denotes the diode noise temperature Fa is the noise figure B T and Y are the amplifierrsquos

bandwidth temperature and power for signal-to-noise ratio respectively For a low-level

video detector ie lt10GHz the sensitivity mainly depends upon three factors firstly on

RF matching structure secondly on the rectification efficiency output impedance and

noise properties of the diode and finally the input impedance bandwidth and noise

properties of the video amplifier at the detector output The RF matching structure

controls the quantity of overall energy at the active junction for rectification The second

factor controls the reaction of the diode to incident microwave radiation and the last

factor will influence the detector sensitivity in general[109]

In practical the Tss is a direct measure of the signal-to-noise ratio of a detector and

is achieved by varying the amplitude of the input pulse (RF signal) until a point in which

the top of the noise level with no signal applied is at the same level of noise at the

bottom level of RF signal It is commonly measured on an oscilloscope as depicted in

Figure 65 below It is defined as the input power at which a signal to noise ratio of 251

is produced[109]

Figure 65 Measurement of Tangential Sensitivity[108 129]

182

The transfer function in many detector diodes is often divided into three sections

Firstly at low incident power secondly at higher input power and finally at very high

power static (continuous) In the first region the detector diode performs as a square-law

detector in which Vout is proportional to Pin This region normally is used to extract the

dynamic range of the diode detector In the second region Vout is approximately

proportional to Vin and this region is known as the linear regime Finally at higher Pin

still the transfer function or response rolls off and thus Vout ultimately become saturated

This roll-off point where Vout has dropped by 1dB below an extrapolation of the

dependence at low Pin is termed the ldquo1dB roll-off pointrdquo and this value is usually in the

range of -11 to 12 dBm[15] Therefore a dynamic range of the detector diode can be

obtained by taking the interval between TSS and 1dB roll-off point (in dBm)

Finally the temperature dependence of Vout for a detector is normally taken from

two extreme points of the temperature (-40C˚ to +80C˚) and thus can be determined

from

Δ119881(119879) = 10 11989711990011989210 |

119881(1198791)

119881(1198792)|

(64)

This Vout variation between -40C˚ and +80C˚ normally expressed in dB

64 Noise Consideration in a Detector diode

The existence of noise in a system limits the accuracy of device performance and

the precision of measurements In a detector system specifically using a diode the noise

which can reduce the sensitivity of signal encryption is called the Noise Equivalent

Power (NEP) By definition the NEP is a noise power density over the detection

sensitivity and it can be exploited to determine the overall noise performance of a

detector[130 131] In other words NEP is defined as the power from the input source

(Pin) that is required to supply a voltage output (Vout) equal to the root means square

noise at Vout [132] For an ideal lossless match and assuming only Johnson-Nyquist is

present the NEP of a zero-bias detector can be expressed as

119873119864119875119900119901119905 = radic4119896119879119861119877119881120573119900119901119905 (65)

183

Where βopt is responsivity with an optimum match which is given by1 2frasl 119877119895120581 This type

of noise appears when changing voltages across a diode and a noise voltage (Vn)

normally will arise Theoretically the NEP has units of Watts (as it is actually a power)

but it often normalized to 1Hz as it is independent of bandwidth and thus the unit

becomes WHz12

Additionally there are also several noise sources that contribute to Vn in a

semiconductor diode which are Johnson-Nyquist noise Flicker Noise and shot noise

Johnson-Nyquist noise [133 134] appears across any conductor or semiconductor at

thermal equilibrium this is due to the thermal agitation of the carriers or charges It can

be expressed in root mean square voltage as below

119881119869minus119873 = radic4119896119879119861119877119895

(66)

Where Rj is the differential intrinsic resistance B denotes the post-detection bandwidth

T is the device temperature and k is the Boltzmann constant[134]

The second noise that is taken into consideration when dealing with semiconductor

devices is Flicker noise more commonly known as 1f noise It is a group of known and

unknown noise sources that can be observed in the frequency spectrum and normally

display an opposite to the frequency power density curve[135] It comes from a variety

of different causes ie recombination effects at a defect in semiconductor mobility

fluctuation and flow of direct current as well as interface phenomena [136-138] In term

of voltage source Flicker noise can be expressed as[139]

1198811119891 = 119870119891119881119909119891119910 (67)

Where Kf denotes a device-specific constant V is the voltage and f is the frequency The

value of x and y typically used are 2 and -1 respectively This type of noise (1f noise)

will be neglected at frequencies high enough due to the fact that the NEP of the diode is

proportional to the thermal noise and resistance of the diode

Finally the noise that causes time-dependent fluctuation in a flow of electrical

current because of the carrier or electron charge crossing a potential barrier is called shot

noise Shot noise is due to the randomness in the diffusion and recombination of both

majority and minority carriers[140] The equation of shot noise term at random time is

given by[141]

184

119881119878ℎ119900119905 = 2119902119868119861 (68)

Where q is the electron charge I is the current and B is the bandwidth This type of noise

is not affected by changes in temperature or device parameters Therefore the total noise

voltage appearing in the semiconductor is found to be [141]

1198811198992 = 119881119869minus119873

2 + 11988111198912 + 119881119904ℎ119900119905

2 (69)

However a zero-bias device will greatly eliminate both shot and flicker noise

compared to a biased device This has been explained by Equation (66) and Equation

(67) above where both noises are significantly related to the current and voltage Thus

if V and I =0 in the nonappearance of incident power then Vn will also become zero For

a detection process with bias the diode will be self-biased by ΔV which causes both

flicker and shot noise to appear But the shot noise in practical situation is much smaller

and thus normally ignored [142-144] Usually the noise in a zero-bias detector is

estimated by considering the low power limit as good first order estimation in which the

presence of only Johnson-Nyquist noise and ΔV is arbitrarily low [11 145 146]

Additionally it has been reported that the noise in tunnelling type diodes displays very

low or no excess noise in the bias region of the current-voltage characteristic[147 148]

Therefore in general most of the noise specifically in tunnelling type of diode will be

neglected this is a great advantage compared to SBD or transistors

65 Modelling of a 100GHz Zero-biased ASPAT Detector

Once the DC and RF characteristics of the ASPAT diode had been accurately

obtained the next step is to model and design a detector circuit based on S-parameter

measurement results as was explained in the previous chapter The aim is to realise a

detector circuit design which can be operated at millimetre and sub-millimetre wave

regions from an accurate diode model prior to the circuit design A diode detector model

puts experimental observations into context and offers insight into future experiment

results Consequently an electrical model based on lumped element component is vital

for a deeper understanding of how and to what extent a new device like the ASPAT

diode can affect all the key detector parameters that were previously discussed The

prediction of the detector parameters mostly depends upon the ASPAT diode

185

geometrical emitter size and material parameters However for millimetre wave

operating frequency the accuracy of the model is more sensitive not only to diode size

but also to the diode periphery ie substrate as well as coplanar amp transmission line

adopted in the circuit Therefore both extracted intrinsic and parasitic element of such a

device must be taken into account

In this work an ASPAT diode with an emitter size of 4times4microm2 which is the

smallest size that could be fabricated so far was chosen to be exploited for detector

designs The important parameters related to the 4times4microm2 GaAsAlAs ASPAT which

works at 0V is summarize in table 61 below

Table 61 A summary of all the important parameters of the 4x4 microm2 diode

Device Rj(Ω) Rs(Ω) Cj(fF) Cp(fF) κ(V-1) Intrinsic_fcut-off(GHz)

4times4microm2 90K 11 21 15 23 629

The actual measured I-V characteristic is used to model the diode since the library

in the ADS simulation tool does not have an ASPAT diode model or any tunnelling

diode for that matter The procedure of realizing the diode model is by taking the I-V

characteristic obtained from the 4times4microm2 emitter size measurement results and

converting it into a10th

orders polynomial equation via MATLAB software to create a

virtual I-V characteristic Thereafter this equation is then defined as a two terminals

device namely Symbolically-defined Device (SDD1P) ie a component of the non-

linear equation provided by ADS (Figure 66) to represent the ASPAT Figure 67

shows the measured data and 10th

orders polynomial equation fit very well to each other

Hence this new component used to represent the whole ASPAT diode will be used in

this research for MMIC detector and Frequency Multiplier designs The device chosen to

be modelled (4times4microm2) has measured junction capacitance of 21fF (at 0V at 40GHz) The

detector circuit is designed to operate at 100GHz for a safe side due to the extrinsic

calculated fcut-off is around 380GHz Since there are a lot of advantages in using unbiased

detectors compared to biased one this work will discuss the performance of millimetre-

wave detector at zero-bias and their result will be compared to the current performance

of other diodes reported in the literature

186

Figure 66 The configuration circuit to verify 10th order polynomial equation extracted from

MATLAB to realize a virtual GaAsAlAs ASPAT diode

Figure 67 Verification of actual (blue measured) and virtual (red_10th order polynomial) I-V

characteristic of the 4times4 microm2 diode used in this study

To realize the ASPAT detector circuit a simple detector circuit topology as

depicted in Figure 68 was constructed Initial simulation was run to perform a

functionality check of the detector circuit utilizing the Harmonic Balance (HB)

simulation tool embedded in that particular software Such simulation tools will analyse

the detector performances in the frequency-domain as it is mostly beneficial and fully

compatible with microwave and millimetre wave problems The frequency domain is

also suitable for single and multi-tone power excitation The importance of harmonic

balance are described in [149]

-00005

0

00005

0001

00015

0002

00025

0003

-3 -2 -1 0 1 2 3

Cu

rre

nt

(A)

Voltage (V)

ADS

4x4

187

Figure 68 Direct detector circuit topology using an ASPAT diode

Initially the circuit topology that consists of P1_Tone power supply ASPAT

bypass capacitor and load resistance is simulated by setting up a fixed input frequency at

100GHz The ASPAT diode provides a DC output voltage proportional to the input

power strength depending on the absolute values of the DC terms associated with the

nonlinearity of the I-V characteristics The capacitance in the output part is a bypass

capacitor used to prevent millimetre-waves from leaking to the output The load

resistance is large enough to ensure the voltage divider between load impedance and

device impedance gives maximum voltage sensitivity This large load resistance is

achieved by creating an open circuit at the end of detector circuit terminals

Noted that this simulation was run using diode parameters that were extracted from on-

wafer one port S-Parameter measurements as described in chapter 5 To apply them in a

two port application ie detector circuit may or may not provide a very accurate

outcome it however worked adequately in the particular circuit described in this chapter

but may not work in other circuits in general Thus the one port extractions in this work

still provide adequate parameters to build and design specific MMIC detector circuits

but not in general applications

The main reason for these simulations and their results to be used in high frequency

applications is due to the fact that actual RF measurement were done up to 40GHz

Additionally the 100GHz operating frequency was obtained from extrapolation of each

ASPATrsquos component Since the on-wafer measurement that were carried out were

limited to one port characterization applying them to two port network applications may

188

have extra consequences which are unknown Therefore actual MMIC detectors are

needed to be built and test to validate this work

To find out what power the 4times4microm2 ASPAT diode can withstand the input power

is varied from -40dBm to 10dBm via control by the P1_tone As can be seen in Figure

69 the diode starts to saturate when the received input power is about -8 dBm Above

this power limit both output voltage and sensitivity drop dramatically

Figure 69 Output voltage and detector sensitivity over wide range of input power

This diode detector circuit can thus operate adequately at given input powers from -30

dBm to -8 dBm with a sensitivity of 950VW However for the best possible sensitivity

over a range of input frequencies only one optimized input power needs to be chosen

The parameters that directly influence the voltage sensitivity are the curvature

coefficients load resistance and video resistance as can be seen from Equation (61)

Therefore in the following simulation the values RL will be optimized according to the

diodes optimum input power with regards to the highest possible voltage sensitivity

Consequently five values of load resistance were chosen from few ohms to infinity

and with the same applied input power as depicted in Figure 610 For most load

resistors the sensitivity is constant at low input power and drop at the diode saturation

region (-8dBm and greater) However for an RL value of 100KΩ and below the loaded

voltage sensitivity shows a peak near 0dBm input power which corresponds to the

maximum slope of the ASPAT detector transfer function The highest voltage sensitivity

is obtained by using an Open circuit load impedance as shown in the graph The Open

circuit load impedance gives the highest voltage sensitivity due to the voltage divider

189

between source and load impedance therefore RL must be at least 5-10 times larger than

Rj to give a better sensitivity

Figure 610 Simulated dependence of Voltage Sensitivity on incident power and load resistance of

the ASPAT detector

Furthermore the value of voltage sensitivity (βV) depends on the junction

resistance of the diode and thus the large value of Rj of the ASPAT diode yields the high

βV observed Rj which was taken from the non-linear measured IV characteristic is a

voltage dependent parameter [142]and is inversely proportional to the forward bias

voltage as depicted in Figure 611 below

Figure 611 Junction resistance as a function of forward voltage

1

10

100

1000

-40 -30 -20 -10 0 10

Vo

lta

ge

Sen

siti

vit

y (

VW

)

Incident Power Pin (dBm)

Infinity

1KΩ

10KΩ

1MΩ

100Ω

100KΩ

0

20

40

60

80

100

0 001 002 003 004 005 006 007 008 009 01

Ju

nct

ion

Res

ista

nce

(k

Ω)

Bias Voltage (V)

190

In fact the expression of (120597119881120597119868frasl ) contribute to the video impedance expression via the

expression RV=Rj+RS (nonlinear resistance) As RS is very small compared to Rj thus it

was ignored when calculating RV Although the large value of Rj will increase the

voltage sensitivity it will also make matching difficult to achieve Therefore there will

be a compromise between the size of the matching circuit and voltage sensitivity to

attain the correct value of Rj Additionally a very high Rj (around ~1MΩ) will also

increase the detectorrsquos noise equivalent power (NEP in Eq(65)) Thus an average value

of RJ typically around 100kΩ is satisfactory[150] By having a large value of RJ one can

benefit from a low input power to drive the diode into the non-linear region and thus the

detector can work at very low RF input power The NEP for GaAsAlAs ASPAT diode

is then calculated based on parameters obtained from this simulation at room

temperature The values are compared to other diode detector available in the literature

as shown in Table 62 below

Table 62 Noise Equivalent Power (NEP) for zero bias detector diode

Device NEP (pWHz12) RJ(KΩ) Frequency(GHz)

ASPAT (4times4 microm2) 188 92 100

Tunnel Diode (08times08 microm2)[150] 370 26 220-330

Zero bias SBD[11] 15 3 150

Sb-Heterojunction Backward Diode[145] 024 32 94

VDI Zero bias SBD[151] 2 18 110

From the Table 62 above the NEP of the ASPAT is calculated based on RF input power

which is required to obtain an output signal-to-noise ratio of unity in a 1Hz at detector

output[142] and also the assumption of only Johnson-Nyquist (thermal noise) is

dominant for small incident power (-25dBm)[152 153] The prediction of NEP for the

ASPAT is comparable to the VDI Zero bias detector since the value of their junction

resistance is much lower than that of the ASPAT Therefore it is very important to

obtain a reasonable value of junction resistance From this it is clear that a trade-off of

high voltage sensitivity and low junction resistance is best to obtain low noise

191

Finally the curvature coefficient at specific operating voltages also influences

the voltage sensitivity of the detector Figure 612 shows the calculated curvature

coefficient of the 4times4microm2 ASPAT diode used in this work

Figure 612 Curvature coefficient for the measured I-V characteristic with a device size of 4times4μm2

The high zero-bias curvature (23V) is reached from the mutual effect of the intra-band

tunnelling in the GaAs-AlAs-GaAs and the highly doped GaAs at the anode and cathode

(Rs amp Rj) This was shown in the numerical simulation in[154] where the combined

effects of the optimum anode AlAs composition increases the curvature coefficients by

thinning the energy tunnelling window (intraband tunnelling process)

Other approaches that can lead to a large curvature is using smaller device area

with minimum series resistance [155] as was also discussed in Section 52 Having a

better curvature coefficient leads to increased voltage sensitivity as seen from Equation

(61) above and βV is also proportional to κ [111] The curvature coefficient calculated

using the above formula is nearly 23V at 1mV peak and high voltage sensitivity can be

achieved by having large value of curvature coefficient But one needs to remember that

this will also decrease with increasing input power because RJ which will also decrease

Therefore for a safe operating region the incident power that can be applied through the

diode is in between -30dBm to -8dBm for a 100GHz operating frequency

In this simulation -25dBm is chosen to simulate the GaAsAlAs ASPTAT diode

detector working at 100GHz input frequency By fixing the P1_Tone to this input power

-5

0

5

10

15

20

25

30

-001 001 003 005 007 009

Cu

rva

ture

Co

effi

cien

t(V

-1)

Voltage (V)

k(4x4 um^2)

192

the frequency is varied from 90GHz to 110GHz and the results are depicted in Figure

613 below

Figure 613 Voltage Sensitivity against frequency at -20dBm fixed input power

Noticeably the voltage sensitivity of the diode detector decreases linearly with

increasing input frequency At 100GHz a sensitivity of around 540VW is obtained

However the sensitivity in this case is roughly estimated from the Equation (61) above

and will not be sufficiently accurate because of the effect of other key factors such as

reflective power which was not included This parameter must be taken into the account

due the P_1Tone power source which provides a 50 Ω impedance source (Zin) which

does not match the load impedance (ZL) which consists of JωL 1JωC and Z and which

mainly comes from the ASPAT diode itself In order to determine the load impedance at

the diode a typical ohmrsquos law (Z=VI) equation must be used at the input side of the

diode with regards to the applied frequency (100GHz)

As a result the total load impedance obtained from the simulation is 11055-

j69057Ω which is clearly not matched to the 50Ω impedance source The mismatch

between source and load leads the available power from the source to be not fully

delivered to the load and hence there is loss of power leading to a lower detector

sensitivity Therefore actual calculation must take into the account the reflection impact

as expressed in the equation below

193

120573119881(119886119888119905119906119886119897) =

119877119869119877119871120581(1 minus |Γ|2)

2(119877119881 + 119877119871) (1 +119877119904

119877119895)2

[1 +1205962119862119895

2119877119878119877119895

1 + 119877119904 119877119895frasl]

(610)

Where the term (1-|Г|2) refers to the normalized power absorption by the ASPAT diode

and Г is the reflection coefficient due to discrepancy between 50Ω input impedance (Zin)

of the input port and the diode Consequently the calculation of reflection coefficient

(eq611) is carried out by using this expression and the result is shown in Figure 614

Γ =

119885119871 minus 119885119894119899

119885119871 + 119885119894119899

(611)

Figure 614 Reflection Coefficient versus operating frequency without matching circuitry

As can be seen in Figure 614 without the matching circuit the S11 is decrease

linearly with frequency but only very slowly This means that most of the RF power

transmitted from the source is reflected back (by that ratio) when it went through the

diode Therefore in order to resolve the mismatch a matching circuit is introduced in

between the source and load of the detector circuit as shown in Figure 615 (red

rectangular) This matching circuit works by transforming the load impedance into an

impedance that is identical to the source or input impedance Note that for any

impedance matching circuit the main purpose is usually to obtain maximum power

transfer to the load however in some cases (ie oscillators) the matching circuit is to

achieve a lower noise figure Hence in a broad sense the introduction of the matching

194

circuit in a detector circuit can be defined as a circuit that convert available impedance

into wanted impedance by obeying the maximum power transfer theorem [156]

Figure 615 Detector circuit with impedance matching circuit placed in between diode and source

There are many type of matching circuit that can be used to achieve both

objectives above such as circuits using lumped element transmission-line-impedance

matching circuit single and double-stub tuners as well as a quarter-wavelength In this

design the technique used to match source and load impedance is the single open and

short stub (in red rectangular) as it is simple convenient and very efficient in ADS

simulation The important parameter that needs to be tuned in both stubs is the electrical

length (E) at any designed frequency ie 100GHz in this case The E tuning is realized

by using the Smith Chart features available in the ADS simulation tools Figure 616

shows the reflection coefficient with matching circuit modelled over a broad frequency

band and it is clearly shown that at the desired operating frequency the reflection is very

low Note that it is difficult to obtain wide frequency band matching

Matching circuit

195

Figure 616 Reflection Coefficient over wide frequency band with matching

The simulation is continued to find the effect of the matching circuit placed in

the detector circuit on the voltage sensitivity The same input power (-25dBm) is applied

to the diode and the voltage sensitivity is plotted against frequency as depicted in Figure

617 Obviously at the desired operating frequency (100GHz) the sensitivity rises up to

a maximum value of 2100VW with this value obtained without any reflection and the

input port being completely matched with the ASPAT diode model used in this work

Figure 617 Simulated voltage sensitivity of GaAsAlAs ASPAT detector at W-band frequency

Once the matching circuit to be used for 100GHz operating frequency was confirmed

further simulations were made by applying a series of low input power to find the

tangential sensitivity (Tss) of the detector diode When determining the Tss it is very

important to include the matching circuit as it will minimize any power losses through

the diode thus a very small input power can be detected

196

Figure 618 Lowest detectable signal at 100GHz operating frequency

The transfer function depicted in Figure 618 shows incident power of -80 dBm

to -50 dBm applied and the lowest detectable signal that can be obtained with 4times4microm2

mesa size ASPAT diode is around 138microV at -68dBm Although a typical value of Tss is

normally not more than ~-55dBm as in ref [74] the lower value obtained in the

simulation is because the device is operated at zero bias operation and does not use any

amplifier therefore the noise and values related to amplifier as in Equation 63 have been

neglected Even though the TSS appeared very low it is most likely very dependant to

the 10th

order polynomial equation embedded into SDD in ADS software Therefore in

near future the TSS value has to be determined in real fabricated MMIC ASPAT detector

As discusses in Section 63 other important parameter that can be extracted from

the diode transfer function is the Dynamic Range of the diode It can be obtained in a

region called square law region which is a region where the Vout of the ASPAT diode is

proportional to the square law of the input power signal From Figure 619 the square

law region is in between -68dBm and -12dBm and the linear region or in this case

saturated region is above -12dBm Taking to the account the roll-off point where Vout

has dropped 1dB below the extrapolation of the dependence at low input power and

therefore the dynamic range of the detector diode can be obtained by taking the interval

between TSS and the 1dB roll-off point (in dBm) which is ~55dBm

197

Figure 619 The dynamic range of the ASPAT diode obtained from linear region of diode operation

The figure of Merit (M) of the detector (ie equation 63) is 2100 (90K) 12

where 2100 is the sensitivity and 90K is the value of RJ at zero bias which is equivalent

to 652 W The M value should be large however in this case due to RJ being very

large it has dropped tremendously when compared to the voltage sensitivity obtained in

this simulation The results obtained indicate a reasonably successful design of the

MMIC detector using the 4times4microm2

emitter size GaAsAlAs ASPAT diode The results

obtained lead to the design of other MMIC detector using the other fabricated diode

sizes (6times6microm2 and 10times10microm

2) A Similar procedure to the one used for the 4times4microm

2

diode was followed The only difference was the use a slight higher input power (-

20dBm) than in the 4times4microm2 design Hence the simulation results obtained are then

compiled and compared as depicted in Figure 620

Saturated

Region

198

Figure 620 Simulation result of diode sensitivity for three devices geometries obtained from the

fabricated ASPAT in this work

The graph plotted for each dimension was taken after matching circuits were included

As can be seen in Figure 620 the highest sensitivity is achieved using the smallest

device size as this has the highest cut off frequency Table 63 below summarises the

performances of the 100GHz ASPAT detectors obtained from the simulation in this

work

Table 63 Summary of characteristic of interest for GaAsAlAs ASPAT detector

Device Tss(dBm) fcut-

off(GHz)

Bv

(VW)

Dynamic range dB(dBm) M(W-12

)

4times4microm2 -68 380 2100 55 65

6times6microm2 -50 151 1445 48 45

10times10microm2 -40 98 247 40 21

From table 63 above it is clear that a lower cut-off frequency will affect the

voltage sensitivity The dynamic range between each ASPAT is different because larger

size area will allow more power to go through the diode as a result of high current that

such a device can handle before reaching the saturation their lowest detectable is higher

Small diode size will detect lowest voltage but cannot handle high power On the other

hand a large diode size is able to receive high power however can only offer lower

voltage output Therefore there is a trade-off between small diode size and receiving

input power which will directly affect Tss and 1dB roll off Once again all the

100

1000

10000

90 95 100 105 110

Vo

lta

ge

Sen

siti

vit

y (

VW

)

Frequency (GHz)

10times10microm^2

6times6microm^2

4times4microm^2

199

parameters obtained in this simulation are just estimation from the 10th

order polynomial

equation thus real ASPAT detector has to be fabricated for verification

The best device performance among all GaAsAlAs ASPAT diodes was obtained

with the 4times4microm2 mesa area size diode which was compared to other exiting millimetre

wave detector diode available in the literature Since the 100GHz is located in the W-

band spectrum frequency therefore the comparison will be performed in this frequency

band but with low input power The parameters for the-state-of-the-art zero bias

detectors are gathered in Table 64 below

Table 64 Comparison 4times4microm2 GaAsAlAs ASPAT diode to the-state-of-the-art zero bias detector

at W-band (75GHz-110GHz)

Device Size(microm2) Tss(dBm) βv (VW) Pin (dBm) f (GHz)

GaAsAlAs ASPAT 4times4 -68 2100 -25 100

GaAs SBD HSCH-

9161[157]

- -49 2200 94

HBD[158] 15x15 2540 -20 95

Planar SBD[159] - -68 2100 -25 100

Note that the ASPAT diode retains its favourable temperature stable characteristics

which are not the case for all the diodes used for comparison in Table 64

66 Conclusions

In this chapter all theory regarding RF detection using diodes ie parameters of

interest noise consideration etc have been discussed The aim to design and develop a

low cost reliable and sophisticated zero bias 100GHz detector circuit was achieved

through exploitation of a 4times4microm2 GaAsAlAs ASPAT diode The design was performed

with the aid of Keysight ADS modelling software utilizing harmonic balance simulation

The effect of load resistance junction resistor to the detector voltage sensitivity was also

discussed in details The 90KΩ Rj value and open circuit load resistance was chosen for

high sensitivity

200

A step by step design of a W-band ASPAT detector was presented The effect of

matching circuit was discussed in detail and where an unmatched sensitivity of 843VW

is obtained which then increases to 2100VW after matching Through RF

characterization simulation a detection at 100GHz (W-band) was successfully achieved

with a relatively large device mesa area (4times4microm2) at an input power of -25dBm

(8microWatt) leading to a 2100VW voltage sensitivity a -68dBm TSS and 55dBm dynamic

range All these values are comparable to others fabricated diodes in the literature

The zero bias ASPAT detectors based on the GaAsAlAs material system in this

work are still at an early stage of development a lot of work is still required to realize

high yielding integrated millimeter and sub-millimeter wave (MMIC) detector circuits

However as this work is on-going at Manchester it is expected that fabricated ASPAT

MMICs with even higher voltage sensitivity will be fabricated in the near future through

collaborating bodies involved in this research especially the University of Cambridge

and ICS Limited

201

7 FREQUENCY MULTIPLIERS (DOUBLER) DESIGN USING

GAASALAS ASPAT DIODES

71 Introduction

Originally the key application for the ASPAT diode was for use as microwave and

millimetre wave detectors[18] This is due to the fact that such diode demonstrates

strong non-linearity low noise and high cut-off frequency features as described in the

previous chapters However these features are not only beneficial for detection purposes

but also allow them to be used and designed as microwave and millimetre wave sources

The only way to generateenhance continuous wave (CW) power using a non-linear

device is through frequency multiplication techniques It is known that the frequency

multiplier is the alternative approach (to 3 terminal transistors) using non-linear devices

that are used to generate high frequency low phase noise signals Any high quality low

frequency signal that goes through a frequency multiplier circuit can be generated to any

desired high output frequency[160] Therefore the main objective of this chapter is to

demonstrate the feasibility of the ASPAT diode as a compact source of microwave and

millimeter-wave receiver for imaging applications[161]

The study of the ASPAT diode as a power source begins with a brief explanation

of the importance of a frequency source and the lack of compact device and technologies

at high-frequency signals The state-of-the-art for frequency multiplier will also be

discussed In the next section (Section 74) the fundamentals of the frequency multiplier

architecture ie the principle of operation and appropriate devices will be presented

Since this is the first attempt at using GaAsAlAs ASPAT diodes a simple multiplier

circuit design and topology was built This will be discussed in detail in the subsequent

sections where simulation results are discussed The focus of the discussion will be to

demonstrate the possibility of a GaAsAlAs ASPAT diode functioning as a frequency

multiplier and comparison with other state-of-the-art varistor mode frequency

multipliers

202

72 Motivation and Background

Typically continuous wave (CW) sources generating below 100 GHz can be

obtained through oscillators amplifiers and pin diode comb generator Below 10THz

the sources can be made from RTD IMPATT diodes and Gunn oscillators and above 10

THz it is commonly done by photonic mixing quantum cascade laser (QCL) and gas

lasers [162] Both types of sources and their performance are plotted in Figure 71

However these conventional ways of generating millimetre and sub-millimetre waves

have their own limitations ie high cost complexity and sometimes requirement for

cryonic cooling The most effective way to tackle the limitations of conventional mm-

wave and THz sources is by implementing frequency multiplication technique using

solid state nonlinear diodes [163-165] such as SBD and ASPAT diodes

Figure 71 performance of state-of the-art millimetre wave source [166]

Twenty years ago there were only two types of diodes (SBD and P-N junction

diodes) often used for frequency multiplication To date besides the SBD there are

many types of diode that have been used as frequency multipliers These include the

high electron mobility varactor (HEMV) single barrier varactor (SBV) [167] and hetero-

structure barrier varactor (HBV) (270GHz with 90mW input power and Conversion

Efficiency of 72)[168 169] Other variants that have developed to enhance the

frequency multiplier performance of the classic SBD [170] include the Barrier-intrinsic-

203

n+ (BIN)diode and Barrier N-layer N+ (BNN) diode [171] Other diodes for use in such

applications are the planar doped barrier diode (PDB) Resonant tunnelling diode (RTD)

amp it families ie Quantum well diode (QWD) and step recovery diode which is a

modification of the P-N junction diode

Although three terminal devices ie FET GaAs MESFET and HEMT had

shown better performance and are capable of achieving greater efficiency and

bandwidth as well as having additional conversion gain features [160] two terminal

devices (ie varactor diodes) which are passive multiplier are still preferred This is due

to their simplicity and most importantly their ability to generate very little noise Among

these types of device technologies the SBD is preferable as it is mature and has been

shown to be very suitable for high-frequency applications [172 173]

The ASPAT diode is exploited to investigate the possibility and the feasibility of

generating microwave and millimetre-wave power through well-known frequency

multiplication methods The utilisation of the ASPAT in frequency multiplication will

also aid in generating local oscillator sources which are critical components in

heterodyne receivers The ASPAT diode will work in resistive I-V mode (varistor mode)

and has features to work also at zero bias condition thus offering low power handling

than traditional high-efficiency varactor diode since the varactor diode requires a large

reverse bias supply of several tens of volts

73 Frequency Multiplier Architecture the Basics

In principle a frequency multiplier is an electronic circuit that gives an output

frequency that is a multiple integer of its input frequency signal pumped from a local

oscillator as depicted in Figure 72 The ability to generate any desired multiple output

signals is realised by a nonlinear device ie diode or transistor Such devices though

also can give distortion or cause sudden change to the input frequency Additionally

these devices generate multiples of the input frequency (fout) The distortion of the

sinusoidal signal refers to an abruptsudden change versus amplitude or time which thus

generates higher frequency with lower amplitudes of the input signal Usually a

frequency multiplier circuit will include a bandpass filter to select the desired harmonic

204

frequencies and deselect undesired harmonic frequencies especially fundamental

harmonic at the output for further processing

Any non-linear device either in symmetricalantisymmetric current-voltage or

capacitance-voltage can be utilised to realise a frequency multiplier source [168 174]

Figure 73 describes the method where a nonlinear resistance is utilised to convert a

harmonic input signal into periodic output signal containing components at multiples of

the input frequency Both non-linear resistance and reactance characteristics can be

extended into power series methods

Figure 73 Principle of operation for frequency multiplier utilising a non-linear resistance [10]

The operating principle of the frequency multiplier is shown in Figure 73 where

the I-V curve converts a harmonic frequency input into a periodic frequency output

including components at multiples of the input frequency The non-linear I-V

characteristic can be explained in term of a power series at the operating fixed point of

bias voltage (VB) [174]

Frequency

Multiplier Circuit

finput

foutput

= nfinput

Figure 72 Frequency multiplier has the property that foutput is an integer multiple of fin

205

119868(119881119861 + ∆119881) = 1198860 + 1198861∆119881 + 1198862∆1198812 + 1198863∆1198813 helliphellip (71)

For a given input voltage as below

∆119881 = 119881119904 cos120596119878119905 (72)

The input signal harmonics will become

119868(119905) = 1198680 + 1198681 cos120596119904119905 + 1198682 cos 2120596119904119905 + 1198683 cos 3120596119904119905 (73)

Where t and ω are the time and angular frequency respectively based on equation (73)

the output contains both signal source and harmonics Therefore a complete frequency

multiplier circuit has to have non-linear device and filter to allow the selection of any

frequency components needed

731 Types of frequency multipliers

Frequency multipliers can be classified into passive and active multipliers This

classification is based on the ability of the frequency multiplier to yield any conversion

gainlosses The passive multiplier is the one that only produces conversion losses In

other words it can be described as a multiplier that generates an output power level

lower than the excitation input power and it is mostly dominated by passive nonlinear

devices ie Diodes On the other hand the active multipliers refer to a device that would

produce an output signal with a power level that is greater than the input signal power

This conversion of power is termed as conversion gain These types of multipliers attract

much attention as they do not only increase the frequency at the output but also the

signal power

Passive frequency multiplier can be formed by using diodes that are classified as

being of the varistor (non-linear I-V) or varactor (non-linear C-V) type [160 174] The

varistor type will influence the frequency multiplication with a non-linear resistance or

conductance (resistive diodes) and this results in a very large potential bandwidth at the

output but poor conversion efficiency The varactor diode type where the frequency

multiplications are affected by the non-linear capacitance (reactive diode) as their

reactive element typical result is high conversion efficiency A diode that is used in this

206

application must have strong nonlinearity stable electrical characteristic repeatable and

has fast enough response to an applied frequency Therefore multipliers are classified

into Doubler Tripler quintuple and so forth depending on the highest power of output

harmonic signal

In general all varactor type diodes with such characteristics will produce high

power at odd-order harmonic oscillation if any microwave signal is pumped into them

The benefit of having odd-order in multiplier design is that it reduces the complexity of

the overall circuit ie it eliminates even-order idler frequencies [175] The varactor type

diode had been shown by Manley-Rowe to a get maximum 100 conversion efficiency

for generating an ideal harmonic [176] compared to the varistor type where the

maximum efficiency achievable ideally is 1 1198992frasl where n is the multiplication factor

(output harmonics number) [174] In the case of power handling (input excitation power)

for multipliers varactor mode diode required greater power (several milli Watt) than the

varistor mode due to the fact that reverse applied voltages are very large (many tens of

volts) Therefore these types of frequency multipliers may not be suitable for the case of

high input power excitation There no report in the literature of varistor based

multipliers working with high power excitations

74 Parameters of interest for Frequency Multipliers

The simplest way to describe an equivalent circuit for a complete frequency

multiplier is by setting a Source impedance (ZS) at the input side and load impedance

(ZL) at the output side as depicted in Figure 74 below This circuit usually has the same

properties as described in the previous chapter and most of the others two ports

networks However in this case the purpose is different and is the conversion of a sine

wave signal source (Vs) with angular frequency ωs to an output signal with frequency

nωs where n is the multiplication integer or the order of multiplication

207

Figure 74 A standard system for two port frequency multiplier circuit

Referring to Figure 74 above there are few sets of parameters for the frequency

multiplier to be taken out and compared Examples are the conversion loss maximum

input signal power Impedance at source and load Bandwidth multiplication factor or

harmonic amp subharmonic content and noise conversion properties The conversion loss

(CL) is described by the ratio of available power at source (Ps) to the output harmonic

power delivered to load resistance (PL) and is normally expressed in dB It occurs due to

the nature of passive semiconductor diodes and the electronic circuit itself that are lossy

and dissipate energy On the other hand the conversion efficiency (ηn) is a ratio of the

output power at load (PL) to the available power at the input (Ps) This is often expressed

in percentage () The conversion efficiency can be determined as

120578 =

119875119878

119875119871

(74)

while the conversion loss is expressed as[174]

119871119899[119889119861] = 10 log

119875119904

119875119871= 10 log (

|1198811199042|

4 119877119890 |119885119904||1198681198712|119877119890|119885119871|

) (75)

Where Vs is the input voltage and IL is the output current amplitude Besides this the

conversion efficiency is often referred to as the inverted value of the Ln In designing a

frequency multiplier it is crucial to minimise the conversion loss and maximise the

conversion efficiency value

To achieve a perfect multiplier with minimum conversion loss the impedance of

source and load must be at an optimum level This implies that the source impedance

Frequency

Multiplier Z

L

Zout

Z

in

V

Z

SWR Г

208

(Zs) must be very close to the complex conjugate of the multiplier input impedance

(Zin) hence minimum reflection loss will occur at the input side This can be realised by

introducing an impedance matching circuit between the diode and source The power

transfer between the source and the multiplier is quantitatively described by the value of

the multiplier input reflection coefficient (Ѓ) with source Zs assumed to represent a

reference impedance This specification also can be explained in the standing wave ratio

(SWR) Both relationships are described below respectively [174]

Γ =

119885119894119899 minus 119885119904lowast

119885119894119899 + 119885119904

(76)

119878119882119877 =

1 + |Γ|

1 minus |Γ|

(77)

Where the asterisk () represents the complex conjugate of the Zs impedance

On the other hand the situation of the load impedance is different when a standard

or an optimum value is provided by the designer This will either increase the conversion

loss or decrease the output power Thus one has to keep in mind that frequency

multipliers are non-linear devices and power transfer condition both at the input and the

output depend on each other and the input signal level[174]

75 20GHz40GHz Varistor Mode ASPAT Diode Frequency Doubler

In this work a similar ASPAT diode (4times4microm2) to that in designing millimetre

wave detector in the previous chapter is used The main objective of designing the

frequency multiplier circuit was first to investigate the performance of the ASPAT

diode as a microwave or millimetre wave signal source A design is deemed successful

when the diode physical parameters are optimised and the suitable impedance matching

network is produced for each desired harmonic as well as maximising the output power

These goals however are hard to achieve when a higher frequency operation is targeted

for use

There are many types of multiplier circuit topologies that can be implemented

using GaAsAlAs ASPAT diode in varistor mode to achieve high order of multiplication

209

Examples are single diode multiplier series or parallel connected diode multiplier anti-

parallel amp anti-series connection diode pair multiplier anti-parallel-series connected

diode multiplier and bridge frequency multiplier as well as nonlinear transmission line

frequency multiplier [174] Before designing a circuit there is one most important

consideration to make Prior to choosing any mentioned circuits to be used for frequency

doubler the design considerations are made based on the capability of ASPAT diode to

receive an optimum amount of input excitation RF power From the discussions in

Chapter 6 the ASPAT diode will reach saturation level (linear regime) at power ~

-10dBm for a device size of 4x4microm2 Once the optimum input power was confirmed the

circuit topology was carefully chosen to balance between the requirements of the

ASPAT to work at high frequencies ie low Rs and Cj amp high diode cut-off frequency

as well as the desired output signal frequency that needs to be produced

To realise the first attempt of an ASPAT diode as a signal source a simple circuit

topology of a frequency doubler was deployed as depicted in Figure 75 below The

frequency doubler circuit consists of a voltage source (can be power source) input

filtering with matching network ASPAT diode output filtering with matching network

and load impedance (ZL)

Figure 75 Block diagram of frequency doubler with series ASPAT mounted diode

In order to investigate the doubler performance the Keysight ADS simulation

tool and similar procedure to obtain accurate ASPAT model using a 10th

order

polynomial equation as in Section 65 was used The circuit in Figure 75 is translated

into ADS format as illustrated in Figure 76 Once the circuit was constructed the

analysis was performed using the Harmonics Balance (HB) simulator The circuit

requirements are matched terminations at the input and output frequencies open

Input Filtering

and matching network

Output Filtering

and matching network

Zs

Vs

ZL

Pin

fin

Pout

nfout

210

circuited terminations at the higher harmonics and optimum reactive terminations (an

inductance which resonates with the junction capacitance) at the output frequencies

Figure 76 Empirical modelling of ASPAT doubler utilising Keysight ADS tool

The circuit in Figure 76 is the simplest way of constructing a frequency doubler

circuit which consists of the signal source (P_1Tone) input matching circuit (Stubs)

filter (Short stub) ASPAT diode low pass filter output matching circuit(Stubs) and ZL

(load resistance) The utilisation of the stubs is an ideal case of simulation since in the

real fabrication stubs are normally formed in large sizes Therefore a proper design such

as using CPW instead of stubs is essential in real fabrication

Again this simulation works for this particular circuit in this chapter as all the diode

parameters were extracted from on-wafer one port S-Parameter measurement described

in chapter 5 To apply them in such two ports applications may not very accurate

however it still provides adequate parameters to build and design particular frequency

multipliers but not in general applications These simulations and results are adequate for

high frequency applications due to the fact that the actual RF measurements on the

diodes were carried out up to 40GHz and the target operating frequency in this multiplier

design does not exceed 40GHz

Since the on-wafer measurement that were carried out were limited to one port

characterization applying them to two port network applications may have extra

consequences which are unknown Therefore actual MMIC frequency multiplier is

needed to be built and test to validate this work

Input Matching

Output Matching

211

To find the optimum output power initial simulation without matching circuit

was performed This simulation was run by varying the input power from -35dBm to

20dBm but fixing the input frequency at 20GHz As can be seen in Figure 77 the lowest

point in the conversion loss (CL) and the highest point of the conversion efficiency (CE)

are obtained from an input power of -1dBm However this amount of input power is too

high for the ASPAT diode The lowest CL at -1dBm may not be accurate since it was

applied without matching Note that it is difficult to achieve a matching between source

impedance and load impedance when varying the input power Therefore a lower input

power of -10dBm is chosen for this frequency doubler operation Figure 77 shows the

Conversion Loss and Conversion Efficiency as a function of the available power of the

given input source

Figure 77 Conversion loss and conversion efficiency as a function of input power

The circuit in Figure 76 works with a -10dBm input power and 20GHz centre

frequency input signal is pumped from the power source (P1_tone) to the ASPAT diode

and distortionabrupt change of input waveform occurs at the fundamental frequency (f0

in this case 20GHz) Such abrupt change produces harmonics and these harmonics can

be classified into desired frequency component by placing two-quarter wavelength (λ4)

stubs (90˚) at both sides of the ASPAT diode At the input side of the diode short circuit

stubs are utilised to permit the f0 tone to reach the ASPAT diode and block the second

harmonics (2f0 in this case 40GHz) back to the input side and pushes it towards the

load resistance On the contrary at the output side of the diode the open circuit stubs are

used to ldquoopen circuitrdquo the 2f0 signals while ldquoshort circuitrdquo the f0 component Thus 2f0

212

signal will not be affected due to the open circuit stubs being half wave (λ2) long The

function of both stubs is basically to isolate the input and output signal from mixing each

other Therefore the design of input and output matching circuit can be achieved easily

The input matching circuit was designed based on the mentioned input frequency

(f0=20GHz) for an available input power (Ps=-10dBm) which is set up at the power

source by using two stubs with the same configuration as used in Chapter 6 Such

configuration is purposely deployed to increase the 50Ω coming from the P1_tone

source impedance to the conjugate thus reducing the reflection coefficient to the

ASPAT diode From the simulation without matching the input impedance to the diode

is 551Ω in magnitude for an available input power of -10dBm

On the output side of the diode output matching circuit is available to transform the 50Ω

port impedance in the optimum load impedance which provides minimum conversion

loss for the ASPAT diode The output matching circuit is designed based on expected

output frequency which is in this case 2f0 =40GHz Other than this optimum

impedance between load impedance and ASPAT will not be achieved thus resulting in

higher conversion loss

To ensure the proposed circuit is valid and suitable for the specific ASPAT diode

mesa size the response of conversion loss and efficiency are plotted as a function of

output frequency from 20GHz to 100GHz The results of both conversions are illustrated

in Figure 78

Figure 78 Conversion loss and Conversion efficiency as a function of output frequency

213

As can be seen in Figure 78 the conversion loss is obtained at the lowest point where

the output frequency is needed Meanwhile the conversion efficiency is maximum at the

same output frequency Therefore this indicates that the first attempt of an ASPAT

Doubler frequency source works well However the values obtained for Ln from this

simulation is 28dB which is rather high On the contrary the η achieved in this study is

very low with a value less than 1

Since the ASPAT is in varistor mode with no bias applied the conversion

efficiency is expected to be low due to resistive losses Another factor that may

contribute to lower η is the diode model itself as it is taken from a 10th

order polynomial

equation not from a diode model provided in the ADS software tools Thus some

properties of such tunnelling diode may not be included Hence it is necessary in due

course to fabricate and build such a compact frequency doubler in the future to verify

the simulation results

From the simulation point of view the less than 1 Conversion Efficiency

obtained is still good enough for a first attempt at a frequency doubler which utilises the

new ASPAT tunnelling diode The frequency doubler obtained from this work is suitable

for use in zero bias varistor modes for low power application The varistor mode doubler

performances from this simulation work are gathered and compared to other in the

literature as summarized in Table 71 below where fout is 40GHz η is 015 Ln is 28dB

and Pi is -10dBm

Table 71 Comparison parameter obtained from this simulation to the state-of-the-art multiplier

diode

Device fout (GHz) η() Ln(dB) Pi (dBm)

ASPAT 40 015 28 -10

SBD (Si)[177] 104 2 134 -10

SBD (GaAs)[178] 13 2 137 -16

214

The performance of the 2040GHz is compared to the literature based on their input

power below -8dBm since the ASPAT is only capable of working at low power To the

best of the author knowledge very few diodes operating in varistor mode at low power

excitation can be found in the field of research and industry Therefore this 2040GHz

ASPAT Doubler might a first for tunnel diodes if it can be fabricated and test at Ka band

and above

76 Conclusions

In this chapter another alternative application based on non-linear features of

GaAsAlAs ASPAT has been presented The simulation of a frequency multiplier

(doubler) was carried out utilizing the 4times4microm2 size ASPAT diode The theory and the

ability of the ASPAT diode to operate as a frequency source were explained in detail

A unique varistor mode frequency multiplier circuit topology for the 2040GHz

ASPAT doubler has been demonstrated and briefly discussed The details and step by

step simulation technique utilizing harmonic balance from Keysight ADS has been

presented Even though the conversion efficiency is very small at 015 and large

conversion loss of 28dB there is still space for improvement in term of design ie

different circuit topology optimized input and output matching circuit etc This design

can be a good reference for a doubler operating at very low power but produce high

frequencies in Ka band

215

8 CONCLUSION AND FUTURE WORK

81 Conclusion

The main focus of this research was the development of a new tunneling diode

namely the asymmetrical spacer layer tunnel (ASPAT) diode for process repeatability

manufacturability and reproducibility The broad study undertaken was to improve the

microwave performance technology by introducing a new type of tunneling diode

For years the asymmetrical spacer layer tunnel diode was unable to be manufactured

due to the high sensitivity of the tunneling current to the barrier thickness This changed

dramatically when the MBE method was carefully optimized to precisely control the

growth to sub-monolayer precisions When stability repeatability and reproducibility in

the epitaxial growth was achieved the next step was to qualify the fabrication process of

the diodes themselves thus ensuring high performance device can be delivered to the

market

For this purpose GaAsAlAs ASPAT diodes made of two different types of

substrates were grown The first batch was grown in the Riber V100HU SSMBE and

used doped substrates Samples XMBE307 and XMBE368 were successfully grown

and fabricated from that batch The DC characterization obtained from measurement

proved that this first batch had fully functional reproducible and manufacturable

devices Later a second batch using semi insulating substrates improvement in spacer

layer and doping concentration were grown This set of samples (9 x 2rdquo wafers grown

simultaneously) and denoted as XMBE304 also showed fully functional DC

characteristic and was used for RF characterization and detector integrated circuits

The conventional GaAsAlAs ASPAT diode structures grown on doped

substrates and developed previously in our lab were not suitable for high frequency RF

characterizations Therefore a major contribution of this work was to develop a new

fabrication technique for a new GaAsAlAs ASPAT structure using semi insulating

substrates to achieve repeatability manufacturability and reproducibility in term of

process flow DC characteristics and ultimately RF characteristics Apart from the

enhancement of the epitaxial layer the other important contribution of this research was

216

the optimization of the small 4times4microm2 emitter size by incorporating both vertical and

lateral structure based purely on low cost I-line optical lithography

To obtain a repeatable and manufacturable fabrication process of lateral

GaAsAlAs ASPAT structures the key issue was to solve the over etching of the

effective mesa area when qualifying the Air Bridge technique This issue caused all

semiconductors under the metal contact to be completely lost thus leaving the metal

contact hanging without connection to any bond pad area However the developments of

the Dielectric Bridge technique realized the true performance of the GaAsAlAs ASPAT

diode structures The samplersquos surface cleanliness as well as DC and RF performance

showed significant improvement when using Si3N4 as a dielectric layer Due to the

highly isotropic etching profile and thicker GaAs layer in XMBE304 samples and

although the smallest emitter size designed on the mask was 2times2microm2 only 4times4microm

2 were

reproducible and showed good uniformity in I-V characteristics

Upon successful optimization in the fabrication process flow of the small emitter

size diodes (4times4microm2 6times6microm

2 and 10times10microm

2) a good uniformity of better than 91

was obtained for DC measurement results within a tile containing over 1000 devices

This confirmed that the lateral GaAsAlAs ASPAT diode structure can only be realized

through the Dielectric Bridge technique These devices were further characterized with

S-parameter measurements and their intrinsic and extrinsic parameter values and

junction capacitances series resistances and junction resistances were extracted leading

to intrinsic cut-off frequencies of 600GHz 429GHz and 100GHz for the three device

sizes respectively

Temperature dependence measurements and simulations were also carried out in

order to confirm that the ASPAT diodersquos characteristics were temperature insensitive

showing less than 5 change in current at both extremes of temperatures 77K to 400K

By comparison the SBD I-V characteristics variations with temperature span orders of

magnitude Physical modelling agreed very well with measured data confirming good

and validated models that can also describe temperature effects

For the realization of the integrated ASPAT millimeter wave detector empirical

modelling using ADS simulation tools was carried out This was performed to predict

the detector performance at 100GHz to comply with the initial objective to develop a

217

millimeter wave detector circuit The simulations using the 4times4microm2 diode models led to

a successful 100GHz circuit design able to detect 100GHz incoming frequency with

2100VW voltage sensitivity

The first ever GaAsAlAs ASPAT diode frequency multiplier design was also

attempted A reasonably good result was obtained for a 20 to 40GHz frequency doubler

operating in varistor mode However the conversion efficiency obtained was less than

1 Further research on this is required to improve the efficiency by using other circuit

topology ie using a balun or other Co-planar waveguides Ultimately fabricating and

testing the actual multiplier circuits are essential to validate the simulation data

82 Future Work

This work has provided a foundation for a reproducible and repeatable GaAsAlAs

ASPAT (SI substrate) wafer fabrication process and recommendations for design and

simulation of ASPAT diode MMIC detectors and frequency source has also been

provided However the GaAsAlAs ASPAT diodes still remain immature and there are

many ways to improve its DC and RF performances both experimentally and in

simulations which will directly affect the detection performances

In term of fabrication process smaller size diodes ie submicron level can be

achieved using dry etching technique with proper calibration For wet etch technique the

etched profile still can be improved by thinning the doped layers so that etching time

will be reduced and hence dimensions down to 2times2microm2 or even 15times15microm

2 can be

reproducibly made

For simulations advanced AC and RF modelling utilizing physical device

simulation available software (SILVACO) must be include in future research hence

holistic study can be conducted to improve the understanding of the ASPAT

For MMIC detector and multiplier design it is vital to produce actual MMIC

devices so that the simulation results can be validated Ultimately tested devices with

good performances can be realized and manufactured

218

REFERENCES

1 Laeri F U Simon and M Wark Host-Guest-Systems Based on Nanoporous

Crystals 2006 John Wiley amp Sons

2 Łukasiak L and A Jakubowski History of semiconductors Journal of

Telecommunications and information technology 2010 p 3-9

3 Song H-J and T Nagatsuma Present and future of terahertz communications

IEEE Transactions on Terahertz Science and Technology 2011 1(1) p 256-

263

4 Hu B and M Nuss Imaging with terahertz waves Optics letters 1995 20(16)

p 1716-1718

5 Smith PR DH Auston and MC Nuss Subpicosecond photoconducting

dipole antennas IEEE Journal of Quantum Electronics 1988 24(2) p 255-260

6 Nagatsuma T Terahertz technologies present and future IEICE Electronics

Express 2011 8(14) p 1127-1142

7 Kumar S et al A 18-THz quantum cascade laser operating significantly above

the temperature of [planck][omega]kB Nature Physics 2011 7(2) p 166-171

8 Phillips T and D Woody Millimeter-and submillimeter-wave receivers Annual

Review of Astronomy and Astrophysics 1982 20(1) p 285-321

9 Whitmer HCTaCA Crystal Rectifiers McGraw-Hill book Company

London 1948

10 Young D and J Irvin Millimeter frequency conversion using Au-n-type GaAs

Schottky barrier epitaxial diodes with a novel contacting technique Proceedings

of the Ieee 1965 53(12) p 2130-2131

11 Liu L et al A broadband quasi-optical terahertz detector utilizing a zero bias

Schottky diode IEEE microwave and wireless components letters 2010 20(9) p

504-506

12 Sankaran S Schottky barrier diodes for millimeter wave detection in a foundry

CMOS process IEEE Electron Device Letters 2005 26(7) p 492-494

13 Chattopadhyay G Submillimeter-wave coherent and incoherent sensors for

space applications in Sensors 2008 Springer p 387-414

14 Anand Y and WJ Moroney Microwave mixer and detector diodes

Proceedings of the Ieee 1971 59(8) p 1182-1190

15 Syme RT Microwave Detection Using GaasAlas Tunnel Structures Gec

Journal of Research 1993 11(1) p 12-23

16 Syme RT et al Asymmetric superlattices for microwave detection in Physical

Concepts of Materials for Novel Optoelectronic Device Applications 1991

International Society for Optics and Photonics

17 Missous M MJ Kelly and J Sexton Extremely uniform tunnel barriers for

low-cost device manufacture IEEE Electron Device Letters 2015 36(6) p 543-

545

18 Syme RT et al Novel GaAsAlAs tunnel structures as microwave detectors in

Semiconductors 92 1992 International Society for Optics and Photonics

19 Schwierz F and JJ Liou Semiconductor devices for RF applications evolution

and current status Microelectronics Reliability 2001 41(2) p 145-168

219

20 HayasHi H Development of Compound Semiconductor DevicesmdashIn Search of

Immense Possibilitiesmdash SEI TECHNICAL REVIEW 2011(72) p 5

21 Mead C Schottky barrier gate field effect transistor Proceedings of the Ieee

1966 54(2) p 307-308

22 Hooper W and W Lehrer An epitaxial GaAs field-effect transistor Proceedings

of the Ieee 1967 55(7) p 1237-1238

23 Drangeid K R Sommerhalder and W Walter High-speed gallium-arsenide

Schottky-barrier field-effect transistors Electronics Letters 1970 6(8) p 228-

229

24 Pillarisetty R Academic and industry research progress in germanium

nanodevices Nature 2011 479(7373) p 324-328

25 Oxley TH 50 years development of the microwave mixer for heterodyne

reception IEEE transactions on microwave theory and techniques 2002 50(3)

p 867-876

26 Baca AG and CI Ashby Fabrication of GaAs devices 2005 IET

27 Cho AY and J Arthur Molecular beam epitaxy Progress in solid state

chemistry 1975 10 p 157-191

28 Cho A Growth of IIIndashV semiconductors by molecular beam epitaxy and their

properties Thin Solid Films 1983 100(4) p 291-317

29 Kiehl RA and TG Sollner High speed heterostructure devices 1994

Academic Press

30 Feiginov M et al Resonant-tunnelling-diode oscillators operating at

frequencies above 11 THz Applied Physics Letters 2011 99(23) p 233506

31 Chang LL L Esaki and R Tsu Resonant tunneling in semiconductor double

barriers Applied Physics Letters 1974 24(12) p 593-595

32 Kasjoo SR Novel Electronic Nanodevices Operating in the Terahertz Region

2012

33 Kanaya H et al Structure dependence of oscillation characteristics of

resonant-tunneling-diode terahertz oscillators associated with intrinsic and

extrinsic delay times Japanese Journal of Applied Physics 2015 54(9) p

094103

34 Chattopadhyay G Technology capabilities and performance of low power

terahertz sources IEEE Transactions on Terahertz Science and Technology

2011 1(1) p 33-53

35 Betz A and R Boreiko A practical Schottky mixer for 5 THz in Proceedings of

the 7th International Symposium on Space Terahertz Technology 1996

36 Yu D et al Ultra high-speed 025-spl mum emitter InP-InGaAs SHBTs with

fsub maxof 687 GHz in Electron Devices Meeting 2004 IEDM Technical

Digest IEEE International 2004 IEEE

37 Das MB Optoelectronic detectors and receivers speed and sensitivity limits in

Optoelectronic and Microelectronic Materials Devices 1998 Proceedings 1998

Conference on 1999 IEEE

38 Rodwell MJ et al Submicron scaling of HBTs IEEE Transactions on Electron

Devices 2001 48(11) p 2606-2624

220

39 Bouloukou A and S Missous Novel High-breakdown Low-noise InGaAs-

InA1As Transistors for Radio Astronomy Applications 2006 University of

Manchester

40 Bean J Materials and technologies 1990 John Wiley amp Sons New York p

13

41 Swaminathan V and A Macrander Materials aspects of GaAs and InP based

structures 1991 Prentice-Hall Inc

42 Vurgaftman I J Meyer and L Ram-Mohan Band parameters for IIIndashV

compound semiconductors and their alloys Journal of applied physics 2001

89(11) p 5815-5875

43 Dingle R W Wiegmann and CH Henry Quantum states of confined carriers

in very thin Al x Ga 1minus x As-GaAs-Al x Ga 1minus x As heterostructures Physical

Review Letters 1974 33(14) p 827

44 Sze SM and KK Ng Physics of semiconductor devices 2006 John wiley amp

sons

45 Tyagi MS Introduction to semiconductor materials and devices 2008 John

Wiley amp Sons

46 Schubert E Delta doping of IIIndashV compound semiconductors Fundamentals

and device applications Journal of Vacuum Science amp Technology A Vacuum

Surfaces and Films 1990 8(3) p 2980-2996

47 Rhoderick EH Metal-semiconductor contacts IEE Proceedings I-Solid-State

and Electron Devices 1982 129(1) p 1

48 Piotrowska A A Guivarch and G Pelous Ohmic contacts to IIIndashV compound

semiconductors A review of fabrication techniques Solid-State Electronics

1983 26(3) p 179-197

49 Rideout V A review of the theory and technology for ohmic contacts to group

IIIndashV compound semiconductors Solid-State Electronics 1975 18(6) p 541-

550

50 Baca A et al A survey of ohmic contacts to III-V compound semiconductors

Thin Solid Films 1997 308 p 599-606

51 Higman T et al Structural analysis of AundashNindashGe and AundashAgndashGe alloyed

ohmic contacts on modulation‐doped AlGaAsndashGaAs heterostructures Journal of

applied physics 1986 60(2) p 677-680

52 Chen KJ et al High-performance enhancement-mode InAlAsInGaAs HEMTs

using non-alloyed ohmic contact and Pt-based buried-gate in Indium Phosphide

and Related Materials 1995 Conference Proceedings Seventh International

Conference on 1995 IEEE

53 Berlin L The man behind the microchip Robert Noyce and the invention of

Silicon Valley 2005 Oxford University Press

54 Goodhue W et al Large room‐temperature effects from resonant tunneling

through AlAs barriers Applied Physics Letters 1986 49(17) p 1086-1088

55 Kerr A and Y Anand Schottky diode MM detectors with improved sensitivity

and dynamic range Microwave Journal 1981 24 p 67-71

56 Davies R Simulations of the current-voltage characteristics of semiconductor

tunnel structures Gec Journal of Research 1987 5(2) p 65-75

221

57 Kelly M Tunnel structures and devices over the coming decade Philosophical

Transactions of the Royal Society of London A Mathematical Physical and

Engineering Sciences 1996 354(1717) p 2291-2293

58 Wilkinson V and M Kelly Manufacturability of quantum semiconductor

devices in High Performance Electron Devices for Microwave and

Optoelectronic Applications 1995 EDMO IEEE 1995 Workshop on 1995

IEEE

59 Wilkinson V M Kelly and M Carr Tunnel devices are not yet

manufacturable Semiconductor Science and Technology 1997 12(1) p 91

60 Eaves L and MJ Kelly The current status of semiconductor tunnelling devices

Philos trans of the Roy soc of London Ser A Math phys and eng sciences

1996 354(1717)

61 Billen K V Wilkinson and M Kelly Manufacturability of heterojunction

tunnel devices further progress Semiconductor Science and Technology 1997

12(7) p 894

62 Kelly M The engineering of quantumndashdot devices Philosophical Transactions

of the Royal Society of London A Mathematical Physical and Engineering

Sciences 2003 361(1803) p 393-401

63 Kelly M New statistical analysis of tunnel diode barriers Semiconductor

Science and Technology 2000 15(1) p 79

64 Hayden R et al Ex situ re-calibration method for low-cost precision epitaxial

growth of heterostructure devices Semiconductor Science and Technology

2002 17(2) p 135

65 Dasmahapatra P et al Thickness control of molecular beam epitaxy grown

layers at the 001ndash01 monolayer level Semiconductor Science and Technology

2012 27(8) p 085007

66 Hayden R M Missous and M Kelly Precision growth for the manufacture of

semiconductor heterostructure devices Semiconductor Science and Technology

2001 16(8) p 676

67 Shao C et al Highly reproducible tunnel currents in MBE-grown

semiconductor multilayers Electronics Letters 2012 48(13) p 792-794

68 Abdullah MR et al GaAsAlAs tunnelling structure Temperature dependence

of ASPAT detectors in Millimeter Waves and THz Technology Workshop

(UCMMT) 2015 8th UK Europe China 2015 IEEE

69 Ariffin KZ et al Asymmetric Spacer Layer Tunnel In0 18Ga0 82AsAlAs

(ASPAT) Diode using double quantum wells for dual functions Detection and

oscillation in Millimeter Waves and THz Technology Workshop (UCMMT)

2015 8th UK Europe China 2015 IEEE

70 Liboff RL Introductory quantum mechanics 2003 Addison-Wesley

71 Esaki L Discovery of the tunnel diode IEEE Transactions on Electron Devices

1976 23(7) p 644-647

72 Landau LD LEM Quantum Mechanics Non-relativistic Theory Pergamon 3

73 Landau LD et al Quantum Mechanics Non‐Relativistic Theory Vol 3 of

Course of Theoretical Physics 1958 AIP

222

74 Syme R Tunnelling devices as microwave mixers and detectors Philosophical

Transactions of the Royal Society of London A Mathematical Physical and

Engineering Sciences 1996 354(1717) p 2351-2364

75 Syme R et al Tunnel diode with asymmetric spacer layers for use as

microwave detector Electronics Letters 1991 27(23) p 2192-2194

76 Brown E W Goodhue and T Sollner Fundamental oscillations up to 200 GHz

in resonant tunneling diodes and new estimates of their maximum oscillation

frequency from stationary‐state tunneling theory Journal of applied physics

1988 64(3) p 1519-1529

77 Reddy M Schottky-collector resonant tunnel diodes for sub-millimeter-wave

applications 1997 University of California Santa Barbara

78 Cox R and H Strack Ohmic contacts for GaAs devices Solid-State Electronics

1967 10(12) p 1213IN71215-1214IN81218

79 Valdes LB Resistivity measurements on germanium for transistors

Proceedings of the IRE 1954 42(2) p 420-427

80 Schroder DK Semiconductor material and device characterization 2006 John

Wiley amp Sons

81 Klootwijk J and C Timmering Merits and Limitations of Circular TLM

structures for contact resistance determination for novel 111-V HBTs Proc

fEEE 2004

82 Marlow GS and MB Das The effects of contact size and non-zero metal

resistance on the determination of specific contact resistance Solid-State

Electronics 1982 25(2) p 91-94

83 Murrmann H and D Widmann Current crowding on metal contacts to planar

devices IEEE Transactions on Electron Devices 1969 16(12) p 1022-1024

84 Berger H Models for contacts to planar devices Solid-State Electronics 1972

15(2) p 145-158

85 Reeves G and H Harrison Obtaining the specific contact resistance from

transmission line model measurements IEEE Electron Device Letters 1982

3(5) p 111-113

86 Shur MS GaAs devices and circuits 2013 Springer Science amp Business

Media

87 Popescu D and B Odbert The Advantages Of Remote Labs In Engineering

Education Educatorrsquos Corner-Agilent Technologies-application note 2011 p

11

88 DataSheet Karl Suss- PM5 Probe System Datasheet 2013

89 Keysight IC-CAP Device Modeling Software 2016 Available from

httpwwwkeysightcomenpc-1297149ic-cap-device-modeling-software-

measurement-control-and-parameter-extractioncc=USamplc=eng

90 DataSheet Anritsu 37369A Vector Network Analyzer Datasheet 2016 Available

from httpwwwtestequipmenthqcomdatasheetsANRITSU-37397D-

Datasheetpdf

91 Packard H HP 4142B Modular DC SourceManual Operation Manual 1992

Available from httpcpliteratureagilentcomlitwebpdf04142-90010pdf

223

92 Microtech C Cascade Microtech- Wincal High Performence RF calaibration

Software (Official Website) 2016 Available from

httpswwwcascademicrotechcom

93 Singh J Electronic and optoelectronic properties of semiconductor structures

2007 Cambridge University Press

94 Whyte W Cleanroom design 1999 Wiley Online Library

95 Vieu C et al Electron beam lithography resolution limits and applications

Applied Surface Science 2000 164(1) p 111-117

96 La Fontaine B Lasers and Moorersquos law SPIE Professional October 2010 p

20

97 Madou MJ Fundamentals of microfabrication the science of miniaturization

2002 CRC press

98 Serway R Physics for Scientists and Engineers 1996 Saunders Publ

Philadelphia

99 Jalali B and S Pearton InP HBTs growth processing and applications 1995

Artech House Publishers

100 Shih YC et al Effects of interfacial microstructure on uniformity and thermal

stability of AuNiGe ohmic contact to n‐type GaAs Journal of applied physics

1987 62(2) p 582-590

101 Zawawi M Advanced In0 8Ga0 2AsAlAs Resonant Tunneling Diodes

forApplications in Integrated mm-waves MMIC Oscillators 2015

102 Zawawi MAM et al Fabrication of Submicrometer InGaAsAlAs Resonant

Tunneling Diode Using a Trilayer Soft Reflow Technique With Excellent

Scalability IEEE Transactions on Electron Devices 2014 61(7) p 2338-2342

103 Silvaco I ATLAS Users Manual Device Simulation Software 2010 Santa

Clara CA

104 Kyono C et al Dependence of apparent barrier height on barrier thickness for

perpendicular transport in AlAsGaAs single‐barrier structures grown by

molecular beam epitaxy Applied Physics Letters 1989 54(6) p 549-551

105 Yang K JR East and GI Haddad Numerical modeling of abrupt

heterojunctions using a thermionic-field emission boundary condition Solid-

State Electronics 1993 36(3) p 321-330

106 Varshni YP Temperature dependence of the energy gap in semiconductors

Physica 1967 34(1) p 149-154

107 Handbook LLM Precision DC Current Voltage and Resistance

Measurements Keithley Instruments Inc[online] 6th revision Ohio 2004

108 Lipsky SE Microwave passive direction finding 2004 SciTech Publishing

109 Howell CM and SJ Parisi Principles Applications and Selection of Receiving

Diodes MACOM Semiconductor Products Division Application note AG314

110 Schulman J D Chow and D Jang InGaAs zero bias backward diodes for

millimeter wave direct detection IEEE Electron Device Letters 2001 22(5) p

200-202

111 Zhang Z et al Sub-Micron Area Heterojunction Backward Diode Millimeter-

Wave Detectors With 018$ rm pWHz^12 $ Noise Equivalent Power IEEE

microwave and wireless components letters 2011 21(5) p 267-269

224

112 Jin N et al High sensitivity Si-based backward diodes for zero-biased square-

law detection and the effect of post-growth annealing on performance IEEE

Electron Device Letters 2005 26(8) p 575-578

113 Shashkin VI et al Millimeter-wave detectors based on antenna-coupled low-

barrier Schottky diodes International journal of infrared and millimeter waves

2007 28(11) p 945-952

114 Zhao P et al GaN Heterostructure Barrier Diodes Exploiting Polarization-

Induced $delta $-Doping IEEE Electron Device Letters 2014 35(6) p 615-

617

115 Pozar DM Microwave engineering 2009 John Wiley amp Sons

116 Koolen M J Geelen and M Versleijen An improved de-embedding technique

for on-wafer high-frequency characterization in Bipolar Circuits and

Technology Meeting 1991 Proceedings of the 1991 1991 IEEE

117 Cao M et al RF characteristics uniformity of GaAsAlAs tunnel diodes in

Infrared Millimeter and Terahertz waves (IRMMW-THz) 2016 41st

International Conference on 2016 IEEE

118 Gao J RF and microwave modeling and measurement techniques for field effect

transistors 2010 SciTec

119 Ren T et al A 340-400 GHz Zero-Biased Waveguide Detector Using an Self-

Consistent Method to Extract the Parameters of Schottky Barrier Diode Applied

Computational Electromagnetics Society Journal 2015 30(12)

120 Fobelets K et al High‐frequency capacitances in resonant interband tunneling

diodes Applied Physics Letters 1994 64(19) p 2523-2525

121 Diebold S et al Modeling and Simulation of Terahertz Resonant Tunneling

Diode-Based Circuits IEEE Transactions on Terahertz Science and Technology

2016 6(5) p 716-723

122 Yong Z et al Design of a 220 GHz frequency tripler based on EM model of

Schottky diodes JOURNAL OF INFRARED AND MILLIMETER WAVES

2014 33(4) p 405-411

123 Louhi JT and AV Raisanen On the modeling and optimization of Schottky

varactor frequency multipliers at submillimeter wavelengths IEEE transactions

on microwave theory and techniques 1995 43(4) p 922-926

124 Guo J Z Zhang and C Qian Modeling of commercial millimeter wave

Schottky diodes in Microwave and Millimeter Wave Technology (ICMMT) 2016

IEEE International Conference on 2016 IEEE

125 Schneider M Metal-semiconductor junctions as frequency converters Infrared

and Millimeter Waves 1982 6 p 209

126 Muth C et al Advanced technology microwave sounder on NPOESS and NPP

in Geoscience and Remote Sensing Symposium 2004 IGARSS04 Proceedings

2004 IEEE International 2004 IEEE

127 Putley E Thermal detectors in Optical and Infrared Detectors 1977 Springer

p 71-100

128 Martin DH Spectroscopic techniques for far infra-red submillimetre and

millimetre waves in Spectroscopic Techniques for Far Infra-red Submillimetre

and Millimetre Waves 1967

225

129 Lucas W Tangential sensitivity of a detector video system with RF

preamplification in Proceedings of the Institution of Electrical Engineers 1966

IET

130 Balocco C et al Low-frequency noise of unipolar nanorectifiers Applied

Physics Letters 2011 99(11) p 113511

131 Benford D T Hunter and TG Phillips Noise equivalent power of background

limited thermal detectors at submillimeter wavelengths International journal of

infrared and millimeter waves 1998 19(7) p 931-938

132 Papoušek D Vibration-rotational Spectroscopy and Molecular Dynamics

Advances in Quantum Chemical and Spectroscopical Studies of Molecular

Structures and Dynamics Vol 9 1997 World Scientific

133 Nyquist H Thermal agitation of electric charge in conductors Physical review

1928 32(1) p 110

134 Turner CS Johnson-Nyquist Noise url httpwww claysturner

comdspJohnson-NyquistNoise pdf(Letzter Abruf Juli 2012)

135 Voss RF 1f (flicker) noise A brief review in 33rd Annual Symposium on

Frequency Control 1979 1979 IEEE

136 McWhorter AL 1f noise and related surface effects in germanium 1955

137 Hooge FN 1ƒ noise is no surface effect Physics letters A 1969 29(3) p 139-

140

138 Van der Ziel A Noise Sources characterization measurement Prentice-Hall

Information and System Sciences Series Englewood Cliffs Prentice-Hall 1970

1970

139 Hooge F 1f noise sources IEEE Transactions on Electron Devices 1994

41(11) p 1926-1935

140 Der Ziel A Theory of shot noise in junction diodes and junction transistors

Proceedings of the IRE 1955 43(11) p 1639-1646

141 Schottky W Small-shot effect and flicker effect Physical review 1926 28(1)

p 74

142 Cowley A and H Sorensen Quantitative comparison of solid-state microwave

detectors IEEE transactions on microwave theory and techniques 1966 14(12)

p 588-602

143 Schulman J et al 1$f $ Noise of Sb-Heterostructure Diodes for Pre-Amplified

Detection IEEE microwave and wireless components letters 2007 17(5) p

355-357

144 Lynch JJ et al Passive millimeter-wave imaging module with preamplified

zero-bias detection IEEE transactions on microwave theory and techniques

2008 56(7) p 1592-1600

145 Su N et al Sb-heterostructure millimeter-wave detectors with reduced

capacitance and noise equivalent power IEEE Electron Device Letters 2008

29(6) p 536-539

146 Westlund A Self-Switching Diodes for Zero-Bias Terahertz Detection 2015

Chalmers University of Technology

147 Yajima T and L Esaki Excess noise in narrow germanium pn junctions

Journal of the physical society of Japan 1958 13(11) p 1281-1287

226

148 Sommers H Tunnel diodes as high-frequency devices Proceedings of the IRE

1959 47(7) p 1201-1206

149 Miraftab V and A Abdipour Harmonic balance analysis of a microwave

balanced power amplifier in Electrical and Computer Engineering 2001

Canadian Conference on 2001 IEEE

150 Patrashin M et al GaAsSbInAlAsInGaAs Tunnel Diodes for Millimeter Wave

Detection in 220ndash330-GHz Band IEEE Transactions on Electron Devices 2015

62(3) p 1068-1071

151 Hesler JL and TW Crowe Responsivity and noise measurements of zero-bias

Schottky diode detectors Proc ISSTT 2007 p 89-92

152 Su N et al Temperature dependence of high frequency and noise performance

of Sb-heterostructure millimeter-wave detectors IEEE Electron Device Letters

2007 28(5) p 336-339

153 Lynch J et al Unamplified direct detection sensor for passive millimeter wave

imaging in Proc of SPIE Vol 2006

154 ZHANG Z et al A physics-based tunneling model for Sb-heterostructure

backward tunnel diode millimeter-wave detectors International Journal of High

Speed Electronics and Systems 2011 20(03) p 589-596

155 Bahl IJ and P Bhartia Microwave solid state circuit design 2003 John Wiley

amp Sons

156 Yeom K-W Microwave Circuit Design A Practical Approach Using ADS

2015 Prentice Hall Press

157 Xie L et al A W-band detector with high tangential signal sensitivity and

voltage sensitivity in Microwave and Millimeter Wave Technology (ICMMT)

2010 International Conference on 2010 IEEE

158 Fay P et al High-performance antimonide-based heterostructure backward

diodes for millimeter-wave detection IEEE Electron Device Letters 2002

23(10) p 585-587

159 Hrobak M et al Planar zero bias Schottky diode detector operating in the E-

and W-band in Microwave Conference (EuMC) 2013 European 2013 IEEE

160 Maas SA Nonlinear microwave and RF circuits 2003 Artech House

161 Appleby R and RN Anderton Millimeter-wave and submillimeter-wave

imaging for security and surveillance Proceedings of the Ieee 2007 95(8) p

1683-1690

162 Crowe TW et al Opening the terahertz window with integrated diode circuits

IEEE Journal of Solid-State Circuits 2005 40(10) p 2104-2110

163 Raisanen AV Frequency multipliers for millimeter and submillimeter

wavelengths Proceedings of the Ieee 1992 80(11) p 1842-1852

164 Erickson NR Diode frequency multipliers for terahertz local-oscillator

applications in Astronomical Telescopes amp Instrumentation 1998 International

Society for Optics and Photonics

165 Mehdi I et al Terahertz local oscillator sources performance and capabilities

in Astronomical Telescopes and Instrumentation 2003 International Society for

Optics and Photonics

166 Tonouchi M Cutting-edge terahertz technology Nature photonics 2007 1(2)

p 97-105

227

167 Nilsen SM et al Single barrier varactors for submillimeter wave power

generation IEEE transactions on microwave theory and techniques 1993 41(4)

p 572-580

168 Xiao Q et al A 270-GHz tuner-less heterostructure barrier varactor frequency

tripler IEEE microwave and wireless components letters 2007 17(4) p 241-

243

169 David T et al Monolithic integrated circuits incorporating InP-based

heterostructure barrier varactors IEEE microwave and wireless components

letters 2002 12(8) p 281-283

170 Lieneweg U B Hancock and J Maserjian Barrier-intrinsic-N+(BIN) diodes

for near-millimeter wave generation in Conference Digest for the Twelft

International Conference on Infrared and Millimeter Waves 1987

171 Lieneweg U et al Modeling of planar varactor frequency multiplier devices

with blocking barriers IEEE transactions on microwave theory and techniques

1992 40(5) p 839-845

172 Chattopadhyay G et al An all-solid-state broad-band frequency multiplier

chain at 1500 GHz IEEE transactions on microwave theory and techniques

2004 52(5) p 1538-1547

173 Maestrini A et al A 17-19 THz local oscillator source IEEE microwave and

wireless components letters 2004 14(6) p 253-255

174 Faber MT J Chramiec and ME Adamski Microwave and millimeter-wave

diode frequency multipliers 1995 Artech House Publishers

175 Frerking MA and JR East Novel heterojunction varactors Proceedings of the

Ieee 1992 80(11) p 1853-1860

176 Penfield P and RP Rafuse Varactor applications 1962

177 Palazzi V et al Low-power frequency doubler in cellulose-based materials for

harmonic RFID applications IEEE microwave and wireless components letters

2014 24(12) p 896-898

178 Presas SM Microwave frequency doubler integrated with miniaturized planar

antennas 2008

228

APPENDICES

Appendix I Doped substrate process details

Mask Stage Process Stage Process step Process detail Equipment

Mask 1 (Mesa

Etch)

Sample clean NMP 10 min 80˚C Beaker

Acetone 5 min Beaker

Isopropanol (IPA) 5 min Beaker

Apply Resist Prebake Bake for 5mins 150C Hotplate

1st Resist S1805 - Program 4 Laurell Spinner

Post bake 110C for 1 Min Hotplate

Photolithography Mask Align mask to wafer

MA4 Mask

aligner

Expose

1 min 09mW iline

(Compensation error set to 1)

MA4 Mask

aligner

Develop MIF 319 for 1 min Beaker

Post Exposure

Bake Oven or Bake 120C for 15

minutes Hotplate

Etch Etch Cal Cal Sample - etch for 2 minutes Beaker

Etch

H3PO4H2O2H2O 3150

time is determined by the etch cal Beaker

Resist Strip Acetone - 5 min and IPA - 5 min Beaker

Mask2 (Top

Contact)

Sample clean NMP ultrasonic for 15 Min Power 3 Beaker

Acetone ultrasonic for 5 Min Power 3 Beaker

Isopropanol (IPA) ultrasonic for 5 Min Power 3 Beaker

Apply Resist Prebake Bake for 5mins 150C Hotplate

1st Resist AznLoF - 2um grade - Program 6 Laurell Spinner

Post bake 110C for 1 Min Hotplate

Photolithography Mask Align mask to wafer

MA4 Mask

aligner

Expose

55 seconds 09mW i-line

(Compensation error set to 1)

MA4 Mask

aligner

Post Exposure bake 110C for 1 Min Hotplate

Develop MIF 326 for 1 mins Beaker

Clean O2 Plasma Etch

20sec 100W forward (340)

60mTorr Plasma Tech

Surface De-oxide HCLH2O 11 40 sec Beaker

Metallisation Evaporation Clean trays + metals Beaker

Load sample and metals

Edward

Evaporator

AuGe (100mg) deposits 55nm

Edward

Evaporator

Ni (1cm) deposits 15nm

Edward

Evaporator

Au (10cm) deposits 200nm

Edward

Evaporator

Lift-off

NMP1165 for 24 hours (25C) or

20 min 80C Beaker

Anneal 420˚C 2min Furnace

229

TLM

Measurement ICCAP

Measurement

Bench

Mask 3

(BottomBacks

ide Contact)

Sample clean Acetone Optional Beaker

Isopropanol (IPA) Optional Beaker

Apply Resist Prebake

Bake for 5mins 150C to dry

the sample Hotplate

Resist (top side) S1813 - Program 6 Laurell Spinner

Soft bake 110C for 1 Min Hotplate

Photolithography Mask Align mask to wafer

MA4 Mask

aligner

Exposure

10 seconds 09mW I-line

(Compensation error set to 1)

MA4 Mask

aligner

Post Exposure bake 110C for 1 Min Hotplate

Develop MIF 319 for 2 mins Beaker

De-scum O2 Plasma Etch

20sec 100W forward (340)

60mTorr Plasma Tech

De-oxidise

Surface De-oxide

HCLH2O 11 40

sec Beaker

metallisation Evaporation Clean trays + metals Beaker

Load sample and metals

Edward

Evaporator

AuGe (100mg) deposits 55nm

Edward

Evaporator

Ni (1cm) deposits 15nm

Edward

Evaporator

Au (10cm) deposits 200nm

Edward

Evaporator

Lift-off

NMP1165 for 24 hours (25C) or

20 min 80C Beaker

Clean Water 3min Beaker

TLM

Measurement ICCAP

230

Appendix II Four Mask step Process Flow

Mask Stage Process Stage Process step Process detail

Mask1 Top

Contact)

Sample clean NMP ultrasonic for 15 Min Power 3

Acetone ultrasonic for 5 Min Power 3

Isopropanol (IPA) ultrasonic for 5 Min Power 3

Apply Resist Prebake Bake for 5mins 150C

1st Resist AznLoF - 2um grade - Program 6

Hot Plate 110C for 1 Min

Photolithography Mask Align mask to wafer

Expose 55 seconds 09mW iline (Compensation

error set to 1)

Post Exposure bake 110C for 1 Min

Develop MIF 326 for 1 mins

Clean O2 Plasma Etch 20sec 100W forward (340) 60mTorr

Surface De-oxide HCLH2O 11 40 sec

Metalisation Evaporation Clean trays + metals

Load sample and metals

AuGe (100mg) deposits 55nm

Ni (1cm) deposits 15nm

Au (10cm) deposits 200nm

Lift-off NMP1165 for 24 hours (25C) or 20 mnt 80C

Mask 2 (Mesa

Etch)

Sample clean NMP Optional

Acetone Optional

Isopropanol (IPA) Optional

Apply Resist Prebake Bake for 5mins 150C

1st Resist S1805 - Program 4

HotPlate 110C for 1 Min

Photolithography Mask Align mask to wafer

Expose 1 min 09mW i-line (Compensation error set

to 1)

Develop MIF 319 for 1 min

Post Exposure Bake Oven or Bake 120C for 15 minutes

Etch Etch Cal Cal Sample - etch for 2 minutes

Etch H3PO4H2O2H2O 3150 time is

determined by the etch cal

Measure TLM

Resist Strip Acetone - 5 min and IPA - 5 min

Mask 3(Isolation) Sample clean Acetone Optional

Isopropanol (IPA) Optional

231

Apply Resist Prebake Bake for 5mins 150C

Resist S1828 - Program 4

Hot Plate 115C for 1 mins

Photolithography Mask Align mask to wafer

Expose 9 mins 09mW iline (Compensation error set

to 1)

Develop MF 319 3 mins

Post Bake Oven Bake 120C for 15mins

Etch Etch Cal Refer to Etch Cal instr tab

Sub-collector Etch H3PO4H2O2H2O 3150 time is

determined by the etch cal

Resist Strip Acetone 5mins + IPA 5 mins in ultrasonic bath

power 1

inspection Microscope

Sample clean Acetone Optional

Mask 4 (Bottom

Contact)

Isopropanol (IPA) Optional

Apply Resist Prebake Bake for 5mins 150C 1st

Resist AznLoF - 2um grade - Program 6

HotPlate 110C for 1 Min

Photolithography Mask Align mask to wafer

Expose 10 seconds 09mW i-line (Compensation

error set to 1)

Post Exposure bake 110C for 1 Min

Develop MIF 326 for 2 mins

Clean O2 Plasma Etch 20sec 100W forward (340) 60mTorr

Surface De-oxide

HCLH2O 11 40 sec

Metalisation Evaporation Clean trays + metals

Load sample and metals

AuGe (100mg) deposits 55nm

Ni (1cm) deposits 15nm

Au (10cm) deposits 200nm

Lift-off NMP1165 for 24 hours (25C) or 20 mnt 80C

232

Appendix III Epitaxial Layer XMBE277

TABLE I The epitaxial structure for sample XMBE277

Layer Thickness (nm) Doping Concentration (cm-3)

n+- In053Ga047As 45 200 x 1019

n- In053Ga047As 25 300 x 1018

In053Ga047As 20 undoped

AlAs 13 undoped

In08Ga02As 45 undoped

AlAs 13 undoped

In053Ga047As 20 undoped

n- In053Ga047As 25 300 x 1018

n+- In053Ga047As 400 100 x 1019

Semi-insulating InP

233

Appendix IV SilVaco (Atlas) Simulation Code

go atlas

---------------------------------------------------------

Structure parameter definition (Constants) values in um

---------------------------------------------------------

Thicknesses

set t_contact1=0

set t_ohmic1=03

set t_emitter=004

set t_spacer1=0005

set t_barrier=000283

set t_spacer2=02

set t_collector=004

set t_ohmic2=045

Doping concentrations

set d_ohmic1=4e18

set d_emitter=2e17

set d_collector=2e17

set d_ohmic2=4e18

set d_gap=2

set d_mesa=4

set d_device=10

set d_etch=008

Layers

set I=$t_contact1

set A=$I+$t_ohmic1

set B=$A+$t_emitter

set C=$B+$t_spacer1

set D=$C+$t_barrier

set E=$D+$t_spacer2

set F=$E+$t_collector

set G=$F+$t_ohmic2

-------------------------------------

Mesh generator

-------------------------------------

mesh diagflip width=45

xmesh location=0 s=1

xmesh location=1 s=1

xmesh location=2 s=1

xmesh location=4 s=1

xmesh location=5 s=1

xmesh location=6 s=1

xmesh location=7 s=1

xmesh location=8 s=1

xmesh location=$d_mesa s=1

xmesh location=$d_mesa+$d_gap s=1

xmesh location=$d_device s=1

Ohmic1

ymesh l=0000 s=005

ymesh l=$I s=005

234

ymesh l=$A s=0005

ymesh l=$B s=0005

ymesh l=$C s=00005

ymesh l=$D s=00005

ymesh l=$E s=0009

ymesh l=$F s=0005

ymesh l=$G s=0005

-----------------------------------

SECTION 2 Regions Structure definition

-----------------------------------

region num=1 name=contact1 material=Gold ymin=0 ymax=$I

region num=2 name=ohmic1 material=GaAs ymin=$I ymax=$A

region num=3 name=emitter material=GaAs ymin=$A ymax=$B

region num=4 name=spacer1 material=GaAs ymin=$B ymax=$C

region num=5 name=barrier material=AlAs ymin=$C ymax=$D xmin=0 xmax=$d_mesa

calcstrain qtregion=1

region num=6 name=spacer2 material=GaAs ymin=$D ymax=$E

region num=7 name=collector material=GaAs ymin=$E ymax=$F

region num=8 name=ohmic2 material=GaAs ymin=$F ymax=$G

region num=9

name=etch material=Air ymin=0 ymax=$F+$d_etch xmin=$d_mesa xmax=$d

_device

---------------------------------

Electrodes

---------------------------------

electrode num=1 name=anode xmin=0 xmax=$d_mesa ymin=0 ymax

=$I material=Gold

electrode num=2 name=cathode xmin=$d_mesa+$d_gap xmax=$d_device ymin=$F+

$d_etch ymax=$F+$d_etch material=Gold

--------------------------------

Doping

--------------------------------

doping uniform ntype conc=$d_ohmic1 Region=2 ymin=$I ymax=$A

doping uniform ntype conc=$d_emitter Region=3 ymin=$A ymax=$B

doping uniform ntype conc=$d_collector Region=7 ymin=$E ymax=$F

doping uniform ntype conc=$d_ohmic2 Region=8 ymin=$F ymax=$G

--------------------------

Contacts

--------------------------

interface sc region=1

interface ss region=2

interface ss region=3

interface si region=4

interface si region=5

interface ss region=6

interface ss region=7

interface sc ymin=$F ymax=$F xmin=$d_mesa+$d_gap xmax=$d_device

interface tunnel region=5 dytunnel=0001

contact name=cathode

contact name=anode

235

------------------------------------------

SECTION 3 Material amp Models Definitions

------------------------------------------

material material=AlAs

permittivity=10 eg300=28 mc=004 affinity=305 nc300=4e19 nimin=1e1

material material=GaAs permittivity=139 eg300=14 mc=0067

affinity=407 nc300=09e17 nimin=1e6

BAND DIAGRAM

output tquantum bandparam qfn qfp valband conband charge polarcharge flowlines

STRUCTURE GRAPHIC

solve init

save outf=XMBE304+real2str

tonyplot XMBE304+real2str

------------------------------------------

SECTION 4 ANALYSIS

------------------------------------------

trap acceptor structure=top elevel=03 density=48e17 degenfac=10 sign=284e-17 sigp=1e-17

ymin=$A ymax=$C xmin=0 xmax=$d_mesa

trap acceptor structure=BOTTOM elevel=035 density=47e17 degenfac=10 sign=284e-17 sigp=1e-17

ymin=$E ymax=$F xmin=0 xmax=$d_mesa

models sisel sisnlderivs qtregion=1 print

method climit=1e-4 itlimit=50 maxtraps=20

DC ANALYSIS

log outf=XMBE304log

solve init

solve vanode=0 name=anode vstep=001 vfinal=15

save outf=XMBE304str

log off

tonyplot XMBE304str

tonyplot XMBE304log

Page 3: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre

3

263 ASPAT Electrical Parameters 61

27 Characterization of Ohmic Contacts 68

271 Transmission Line Measurement (TLM) 69

28 Basic Characterization Techniques and procedures 72

281 Measuring tools and apparatuses 72

282 Measurement steps using a VNA 75

283 Measurement Practice and Flowchart 76

3 RESULTS OF FABRICATION PROCESS OF ASPAT DIODES 79

31 Introduction 79

32 Epitaxial Layer Growth Techniques 80

321 Molecular Beam Epitaxy (MBE) 80

33 Basic Principles of Common Fabrication techniques 81

331 Sample cleaning 81

332 Photolithography 82

333 Etching Process 86

334 Sputtering (dielectric deposition) 88

335 Metallization Process Lift-off and Annealing 88

34 GaAsAlAs ASPAT Process Optimization 92

341 ASPAT Devices used in Fabrication 93

342 Fabrication process of GaAsAlAs ASPAT diode toward repeatability

reproducibility and manufacturability 95

343 Fabrication process of GaAsAlAs ASPAT diode toward High

frequency Applications 117

35 Conclusions 125

4 PHYSICAL MODELLING OF THE GAASALAS ASPAT DIODE USING

SILVACO 126

41 Introduction 126

42 SILVACO modelling Tools 127

43 SILVACO Implementation GaAs AlAs ASPAT Modelling 130

44 Simulation Result and Analysis 132

45 Structure Analysis of ASPAT Diode 135

451 Dependencies of current on AlAs Barrier thickness 135

452 Dependence of current on Spacer I length l1 137

4

453 Dependence of current on Spacer II length l2 138

46 Temperature Dependent Simulation 140

47 Temperature Dependence characteristics of GaAsAlAs ASPAT Diodes 142

48 Conclusions 147

5 DC amp RF CHARACTERISTIC OF ASPAT DIODES 148

51 Introduction 148

52 General Analysis of Current-Voltage (I-V) characteristics of GaAsAlAs

ASPAT diodes 149

53 RF Test Fixture Theory and Experiment 153

531 On-Wafer Measurement and Small Signal One-Port Characterizations

154

54 Device Calibration 155

541 Open and Short De-Embedding Technique 155

55 S-Parameter Measurement Result and Analysis 157

551 Diode to diode uniformity 158

552 Wafer to wafer uniformity 160

553 Small devices RF measurements 161

56 Extracting RF models of ASPAT at Zero Bias Voltage 164

561 Extraction of ASPAT parasitic element 165

562 Extraction of ASPAT intrinsic elements 168

563 Capacitances -Voltage (C-V) Extraction 172

57 Conclusions 173

6 MILLIMETRE WAVE GAASALAS ASPAT DETECTOR DESIGN USING

ADS 175

61 Introduction 175

62 Detection Theory 176

63 Definition of Characteristics of Interest Detection Parameters in small signal

analysis 179

64 Noise Consideration in a Detector diode 182

65 Modelling of a 100GHz Zero-biased ASPAT Detector 184

66 Conclusions 199

7 FREQUENCY MULTIPLIERS (DOUBLER) DESIGN USING GAASALAS

ASPAT DIODES 201

71 Introduction 201

5

72 Motivation and Background 202

73 Frequency Multiplier Architecture the Basics 203

731 Types of frequency multipliers 205

74 Parameters of interest for Frequency Multipliers 206

75 20GHz40GHz Varistor Mode ASPAT Diode Frequency Doubler 208

76 Conclusions 214

8 CONCLUSION AND FUTURE WORK 215

81 Conclusion 215

82 Future Work 217

REFERENCES 218

APPENDICES 228

Word count including footnotes and endnotes 61500(approximately)

6

LIST OF TABLES

Table 11 Sample XMBE304 XMBE307 and XMBE368 GaAsAlAs materials

structure grown on GaAs Substrates by MBE 23

Table 12 Quantum wells sandwiching the quantum barrier for sample XMBE314

grown on a GaAs substrate by MBE 24

Table 13 Novel InGaAsAlAs ASPAT diode grown on InP substrate by MBE on InP

substrate 24

Table 21 Lattice constant (a) and band gap for group III-V binary and ternary

compound semiconductors a room temperature [41 42] 38

Table 22 epitaxial layer of sample XMBE304 which is the main focus of this work 56

Table 31 Chemical wet etch characteristics using Orthophosphoric-based and

Ammonia on GaAs and InGaAs materials 87

Table 32 Epitaxial layer of Doped substrate samples 93

Table 33 The epitaxial layers of sample XMBE304 with barrier 10ML~28nm 94

Table 34 Generic fabrication steps established by Dr Md Adzhar [101] 97

Table 35 Standard process flow for Air-Bridge design fabrication 101

Table 36 Standard fabrication process flow for Dielectric-Bridge design 107

Table 37 New arrangement of the mask number and step in Second Run 110

Table 38 New arrangement for the Third run using Dielectric-Bridge mask 113

Table 39 The outcome of the spreading resistance before and after using LOR

technique 115

Table 310 DC and RF characteristics for XMBE304 118

Table 311 3rd

Gen Mask process step 119

Table 312 Standard deviation at two different voltages 124

Table 41 The parameter values used in this simulation 134

Table 42 The calculated values of bandgap at different temperatures 140

Table 43 The calculated effective masses for each temperature used in this simulation

141

Table 44 Epitaxial layer profile for Schottky Barrier Diode (XMBE104) 145

Table 51 Performance of the ASPAT diode obtained from measured IV characteristics

in this work 153

Table 52 Device to device uniformity check for large ASPAT diode 159

Table 53 The variation of reflection coefficient (S11) for GaAsAlAs ASPAT diode at

four different frequencies[117] 159

7

Table 54 Wafer to wafer uniformity check on XMBE304A and XMBE304B 161

Table 55 Comparison between calculated (fully Depleted) and extracted (different

biases) values from equivalent circuit parameters for different ASPAT mesa sizes at zero

bias voltage 170

Table 61 A summary of all the important parameters of the 4x4 microm2 diode 185

Table 62 Noise Equivalent Power (NEP) for zero bias detector diode 190

Table 63 Summary of characteristic of interest for GaAsAlAs ASPAT detector 198

Table 64 Comparison 4times4microm2 GaAsAlAs ASPAT diode to the-state-of-the-art zero

bias detector at W-band (75GHz-110GHz) 199

Table 71 Comparison parameter obtained from this simulation to the state-of-the-art

multiplier diode 213

8

LIST OF FIGURES

Figure 21 III-V compound semiconductors mobility and band gap[24] 31 Figure 22 illustration of Homojunctions band structure material before (left) and after

(right) equilibrium 34 Figure 23 Energy band diagrams before (left) and after (after) interface combination at

equilibrium 35

Figure 24 Lattice Matching for both materials when aL=aS 37 Figure 25 Energy gap and lattice constant for direct and indirect band gap of compound

semiconductor at ambient temperature[40] 37 Figure 26 Lattice mismatched material 39

Figure 27 The formation of Pseudomorphic-layer in two situations (a) compressive and

(b) tensile strain [1] 40 Figure 28 The band diagram of an ideal quantum well formed by a narrow band gap

semiconductor eg InGaAs sandwiched between two large band gap semiconductors eg

GaAs (a) Structure (b) energy band diagram and (c) Conduction band diagram when

AlGaAs is n-doped[43] 41 Figure 29 Metal and semiconductor in two conditions (a) separation (b) in contact 43

Figure 210 Energy band diagram of Schottky contact on n-type material under (a)

reverse and (b) forward bias 45 Figure 211 Figure 10 N-type semiconductor Ohmic contact with different doping

concentration ND (a) Low (b) Intermediate and (c) high 47 Figure 212 Classical view of whether an electron is can surmount a barrier or not

Quantum mechanical view allows an electron to tunnel through a barrier The probability

(blue) is related to the barrier thickness 51

Figure 213 Rectangular tunnelling with incident reflected and transmitted wave

function[70] 52

Figure 214 Epitaxial layer structure of device XMBE 304 (lateral structure) used in

this study 55 Figure 215 The layer profile of ASPAT with 28nm barrier thickness [27] 57

Figure 216 Conduction band diagram showing band bending and 2DEG formation at

the L1 spacer 60 Figure 217 I-V characteristics of a fabricated ASPAT diode 63 Figure 218 Intrinsic Elements of the Asymmetric Spacer Tunnel Layer (ASPAT) Diode

63 Figure 219 Cross-sectional view of ASPAT vertical structure the dimensions of d b

and h are not drawn to scale 65 Figure 220 Cross-section view of the complete process ASPAT diode (XMBE304)

The dimensions are not drawn to scale 67 Figure 221 A simple TLM structure with effective length and sheet resistance

underneath 69 Figure 222 Top view of TLM ladder structure use in this work 71 Figure 223 Typical plot of resistance versus TLM spacing 71

Figure 224 The set of DC measurement apparatus arrangement to measure the TLM 73

9

Figure 225 The set of RF measurement for on-wafer DC and RF one-port measurement

75 Figure 226 Actual VNA system that was used for RF characterization 75 Figure 227 Block diagram of the ASPAT measurement step 77

Figure 31 3D illustration of Optical lithography process used in this research 85 Figure 32 Actual picture of thermal evaporator used in this study 89 Figure 33 Single layer lift-off process using negative photoresist 91 Figure 34 Current-Voltage characteristic of sample XMBE368 used in this study at

two different locations on the wafer tile 96

Figure 35 ASPAT mask design for (a) 100times100microm2 (b) 30times30microm

2 and (c) 15times15 microm

2

diode dimensions designed in the 1st Gen Mask 96

Figure 36 Air Bridge design (red circle) for large mesa area (100times100microm2) device with

option for doped substrates 100 Figure 37 The layout of 1

st design of Dielectric Bridge (green circle) mask design for

100 times 100microm2 emitter size with option for doped substrate processing 100

Figure 38 Dry Etching for the first run in this study 102 Figure 39 Severe undercut of 2times2 microm

2 and 6times6 microm

2 devices 103

Figure 310 Device with emitter area of 2times2 microm2 and 6times6 microm

2 after dry etch and wet

etched 104 Figure 311 SEM Images of the GaAs sample 104

Figure 312 The cross-sectional view of InGaAs (left) and GaAs (right) samples used in

this study 105

Figure 313 SEM images taken from the XMBE304 samples and a GaAs test sample

105

Figure 314 SEM images for InP and InGaAs taken from [56] 105 Figure 315 Short circuit behaviour on one of the fabricated device in this run 106

Figure 316 The surface of the sample after final processing 108 Figure 317 IV characteristics of sample XMBE304 for emitter sizes of 1000 microm2

2500 microm2 900 microm2 400 microm2 225 microm2 100 microm2 and 36 microm2 109

Figure 318 Optical images after MESA etching (a) 1 microm Tolerance (b) 05 microm

Tolerance 110

Figure 319 After lift-off processing 111 Figure 320 The current density of XMBE304 processed using Si3N4 as dielectric layer

112

Figure 321 side view of lateral ASPAT structure 113

Figure 322 The measured size of the emitter area and the length D (blue color marked)

114 Figure 323 Summary of LOR technique steps 115 Figure 324 IV characteristic in 3rd run (reduced d-Gap) for each diode size on the mask

116

Figure 325 3rd

Gen Mask layout showing actual devices de-embedding TLM and

alignment mark structures used in this study 118 Figure 326 Step by step device fabrication (in 3D and cross-sectional view) for semi-

insulating substrate device type used in this study 121 Figure 327 Example finished process device with bond pad using 3

rd Gen mask 121

Figure 328 XMBE304 TLM measurement for the top contact after annealing 122

10

Figure 329 XMBE304 TLM measurement for the bottom contact after annealing 122 Figure 330 XMBE304 GaAsAlAs ASPAT measured IV characteristics at room

temperature for 4times4microm2 mesa size 123

Figure 331 XMBE304 GaAsAlAs ASPAT measured IV characteristics at room

temperature for 6times6microm2 mesa size 124

Figure 332 XMBE304 GaAsAlAs ASPAT measured IV characteristics at room

temperature for 10times10microm2 mesa size 124

Figure 41 SILVACO Atlas simulation process flow 128 Figure 42 The GaAsAlAs ASPAT structures produced from simulation based on the

diode multilayer heterostructures on the right 131 Figure 43 The Energy-momentum diagram for AlAs heterojunction semiconductor 131 Figure 44 (a) The conduction band and the valence band of the ASPAT diode structure

(b) the energy band diagram of the ASPAT diode structure when under three different

biases 132 Figure 45 The current-voltage characteristic of an ASPAT diode (100times100microm

2) and

(4times4microm2) using SILVACO Atlas simulator for structure device XMBE304 showing

excellent agreement between simulated and experimental data 134

Figure 46 IV characteristics of the dependencies of current on AlAs barrier 136 Figure 47 Example of analysis at -1 and 1V to the current 136 Figure 48 I-V characteristic of the dependencies current to Spacer I layer 137

Figure 49 Current changes with layer thickness l1 138 Figure 410 IV characteristic of the dependencies current to Spacer 1 layer 139

Figure 411Current change with layer thickness l2 139

Figure 412 Measurement and simulation comparison result as a function of temperature

range from 100K to 398K 142 Figure 413 Log Current vs voltage as a function of temperature for ASPAT sample

XMBE304 143

Figure 414 The effective barrier high of the Schottky barrier diode and the ASPAT

Diode [3] 144

Figure 415 Log Current vs voltage as a function of temperature for SBD sample

XMBE104 145

Figure 416 Influence of temperature on IV characteristic for both ASPAT diode and

SBD 146

Figure 51 IV Characteristics of measured ASPAT diodes for emitter sizes of 4x4um2

6x6um2 and 10x10um

2 Note the good scalability 149

Figure 52 Junction resistance versus voltage 151

Figure 53 The rectifying action of the diode at zero voltage which gives the ASPAT

high sensitivity near zero bias detection 152

Figure 54 One port S-parameter measurements 155

Figure 55 Fabricated open short de-embedding structure and real device (zoomed) use

for RF calibration and measurements (Note Images are not to scale) 156

11

Figure 56 Real S11 RF measurements for (XMBE304A) were done for four devices

from 15times15microm2 to 100times100microm2 device sizes for within-wafer uniformity check 158

Figure 57 Imaginary S11 measurements (XMBE304A) to qualify the repeatability and

reproducibility were done on four devices from 15times15 microm2 to 100times100 microm2 at zero

bias 158

Figure 58 Wafer to wafer uniformity test on 1st and 2

nd RF measurement(XMBE304B)

to qualify the repeatability and reproducibility were done for 30times30microm2 to 15times15 microm

2

device sizes (Real and Imaginary) Note blue colour is XMBE304A and red colour is

XMBE304B 160

Figure 59 RF measurement performed on device sizes of 10times10microm2 6times6mmicro

2 and

4times4microm2

(Real and Imaginary) Note that green red and blue colour represents 4times4microm2

6times6mmicro2 and 10times10microm

2 diodes respectively 162

Figure 510 RF measurement performed on device sizes of 10times10 microm2 6times6 microm2 and

4times4 microm2 (Smith Chart) Note that green red and blue colour represents 4times4microm2

6times6mmicro2 and 10times10microm2 diodes respectively 162

Figure 511 Cross-sectional of fabricated ASPAT diode and the corresponding

equivalent circuit model 164

Figure 512 The S-parameter Touchstone file is used to read the measured files 165

Figure 513 Equivalent Circuit Model of Open CPW or de-embedded structure 166

Figure 514 Equivalent circuit model for short de-embedded structure 166

Figure 515 Smith chart representative S-parameter measurement for short (left) and

open (right) CPW The blue lines represent simulated data and the red is measured data

167

Figure 516 Equivalent circuit of the ASPAT diode 169

Figure 517 Excellent fitting between measured (red colour) and simulated (blue colour)

for S11 (Real and Imaginary) results for various small device designs 169

Figure 518 Excellent fitting between measured (red colour) and simulated (blue colour)

for S11 results (Smith Chart) for various small device designs 170

Figure 519 Capacitance extraction showing full depletion at -025 volts (modelled

capacitance vs Voltage) 172

Figure 61 Block diagram represent a complete direct receiver system 177

Figure 62 The detection process of a single wave through a non-linear IV characteristic

of a diode 177

Figure 63 Lumped element illustration of microwave detector circuit 178

Figure 64 The mixing process where the signals are processed by the non-linear I-V

characteristic to produce an intermediate frequency (fIF=fLO-fRF) when signals at fLO and

fRF are applied to the diode 179

Figure 65 Measurement of Tangential Sensitivity[108 129] 181

12

Figure 66 The configuration circuit to verify 10th order polynomial equation extracted

from MATLAB to realize a virtual GaAsAlAs ASPAT diode 186

Figure 67 Verification of actual (blue measured) and virtual (red_10th order

polynomial) I-V characteristic of the 4times4 microm2 diode used in this study 186

Figure 68 Direct detector circuit topology using an ASPAT diode 187

Figure 69 Output voltage and detector sensitivity over wide range of input power 188

Figure 610 Simulated dependence of Voltage Sensitivity on incident power and load

resistance of the ASPAT detector 189

Figure 611 Junction resistance as a function of forward voltage 189

Figure 612 Curvature coefficient for the measured I-V characteristic with a device size

of 4times4μm2 191

Figure 613 Voltage Sensitivity against frequency at -20dBm fixed input power 192

Figure 614 Reflection Coefficient versus operating frequency without matching

circuitry 193

Figure 615 Detector circuit with impedance matching circuit placed in between diode

and source 194

Figure 616 Reflection Coefficient over wide frequency band with matching 195

Figure 617 Simulated voltage sensitivity of GaAsAlAs ASPAT detector at W-band

frequency 195

Figure 618 Lowest detectable signal at 100GHz operating frequency 196

Figure 619 The dynamic range of the ASPAT diode obtained from linear region of

diode operation 197

Figure 620 Simulation result of diode sensitivity for three devices geometries obtained

from the fabricated ASPAT in this work 198

Figure 71 performance of state-of the-art millimetre wave source [166] 202

Figure 72 Frequency multiplier has the property that foutput is an integer multiple of fin

204

Figure 73 Principle of operation for frequency multiplier utilising a non-linear

resistance [10] 204

Figure 74 A standard system for two port frequency multiplier circuit 207

Figure 75 Block diagram of frequency doubler with series ASPAT mounted diode 209

Figure 76 Empirical modelling of ASPAT doubler utilising Keysight ADS tool 210

Figure 77 Conversion loss and conversion efficiency as a function of input power 211

Figure 78 Conversion loss and Conversion efficiency as a function of output frequency

212

13

LIST OF PUBLICATIONS CONFERENCE PRESENTATIONS

PUBLICATIONS

1 MRR Abdullah Y K Wang J Sexton M Missous and M J Kelly ldquoGaAsAlAs

Tunnelling Structure Temperature Dependence of ASPAT Detectorsrdquo 8th UK-Europe-

China Workshop on mm-waves and THz Technologies 2015 Cardiff University IEEE

proceedings DOI 101109UCMMT20157460591

2 Yuekun Wang Mohd Rashid Redza Abdullah James Sexton and M Missous

ldquoInGaAs-AlAs asymmetric space layer tunnel (ASPAT) diodes for THz electronicsrdquo 8th

UK-Europe-China Workshop on mm-waves and THz Technologies 2015 Cardiff

University IEEE proceedings DOI 101109UCMMT20157460589

3 K N Zainul Ariffin S G Muttlak M Abdullah M R R Abdullah Y Wang and M

Missous ldquoAsymmetric Spacer Layer Tunnel In018Ga082AsAlAs (ASPAT) Diode using

Double Quantum Wells for Dual Functions Detection and Oscillationrdquo 8th UK-

Europe-China Workshop on mm-waves and THz Technologies 2015 Cardiff

University IEEE proceedings Doi 101109UCMMT20157460599

4 K N Zainul Ariffin M R R Abdullah Y K Wang S G Muttlak O S

Abdulwahid J Sexton MJ Kelly and M Missous ldquoAsymmetric Spacer Layer Tunnel

Diode (ASPAT) Quantum Structure Design Linked to Current-Voltage Characteristics

A Physical Simulation Studyrdquo UK-China Millimetre Waves and Terahertz Technology

Workshop September 2017 Submitted 14 July 2017 Conference held on 11th -13th

September 2017 DOI 101109UCMMT20178068358

5 K N Zainul Ariffin Y Wang M R R Abdullah S G Muttlak Omar S

Abdulwahid J Sexton Ka Wa Ian Michael J Kelly and M Missous ldquoInvestigations of

Asymmetric Spacer Tunnel Layer (ASPAT) Diode for High-Frequency Applicationsrdquo

DOI 101109TED20172777803

6 Omar S Abdulwahid S G Muttlak M R R Abdullah K N Zainul Ariffin J

Sexton Ka Wa Ian Michael J Kelly and M Missous ldquoA 100GHz Zero-Biased

Quantum Tunnelling ASPAT Detectorrdquo Submitted to IEEE TED on DEC 2016 under

correctionamendment Pending fabrication data

14

CONFERENCE PRESENTATIONS

1 Mohd Rashid Redza Abdullah J Sexton Kawa Ian MJKelly and M Missousldquo

G2040GHz Frequency Doubler Varistor Mode using ASPAT diodesrdquo UK

Semiconductors 2017 2017 University of Sheffield Oral presentation

2 M R R Abdullah YueKun Wang J Sexton Kawa Ian and M Missousldquo

Microwave Performance of GaAsAlAs Asymmetric Spacer Layer Tunnel (ASPAT)

Diodesrdquo UK Semiconductors 2016 2016 University of Sheffield Oral presentation

3 M R R Abdullah J Sexton and M Missousldquo GaAsAlAs Tunnelling Structures

THz RTD oscillators and ASPAT detectorsrdquo UK Semiconductors 2015 2015

University of Sheffield Oral presentation

4 Yuekun Wang Mohd Rashid Redza Abdullah and M MissousldquoInGaAs-AlAs

asymmetric space layer tunnel (ASPAT) diodes for THz electronicsrdquo UK

Semiconductors 2015 2015 University of Sheffield Oral presentation

5 Mohd Rashid Redza Abdullah and M Missousldquo GaAsAlAs Tunnelling

Structure Temperature Dependence of ASPAT Detectorsrdquo PGR Conference2016

2016 University of Manchester Poster presentation

6 YueKun Wang KNZainul Ariffin Mohd Rashid Redza Abdullah J Sexton

Kawa Ian and M Missous ldquoPhysical Modelling and Experimental Studies of

InGaAsAlAs Asymmetric spacer Layer Tunnel Diodesrdquo UK Semiconductors 2016

2016 University of Sheffield Oral presentation

7 K N Zainul Ariffin S G Muttlak M R R Abdullah Y Wang Omar S

Abdulwahid M Missous ldquoExperimental and Physical Modelling of Temperature

Dependence of a Double Quantum Well In018Ga082AsAlAs ASPAT Dioderdquo UK

Semiconductor Conference 2016 Sheffield 6 ndash 7 July 2016 Oral Presentation

8 Omar S Abdulwahid Mohd Rashid Redza Abdullah S G Muttlak K N Zainul

Ariffin Mohamed Missous ldquoTunnelling Barrier Diode for Millimetre Wave

Mixingrdquo UK Semiconductor Conference 2016 Sheffield 6 ndash 7 July 2016 Oral

Presentation

9 M Abdullah K N Zainul Ariffin MRR Abdullah J Sexton M Missous and

MJ Kelly ldquoA Novel In18Ga82As-AlAs Asymmetric Spacer Layer Tunnel (ASPAT)

Diode with Double Quantum Wells for Microwave Detectionrdquo UK Semiconductor

Conference 2015 Sheffield 1 ndash 2 July 2015 Oral Presentation

15

ABSTRACT

Thesis Title GaAsAlAs ASPAT Diodes for Millimetre and Sub-Millimetre Wave

Applications

Institute School of Electrical and Electronic Engineering the University of Manchester

Candidate Mohd Rashid Redza bin Abdullah

Degree Doctor of Philosophy (PhD)

Date 3 October 2017

The Asymmetric Spacer layer Tunnel (ASPAT) diode is a new diode invented in

the early 90s as an alternative to the Schottky barrier diode (SBD) technology for

microwave detector applications due to its highly stable temperature characteristics The

ASPAT features a strong non-linear I-V characteristic as a result of tunnelling through a

thin barrier which enables RF detection at zero bias from microwaves up to

submillimetre wave frequencies In this work two heavily doped GaAs contact layer on

top and bottom layers adjacent to lightly doped GaAs intermediate layers enclose

undoped GaAs spacers with different lengths sandwiching an undoped AlAs layer that

acts as a tunnel barrier The ultimate ambition of this work was to develop a MMIC

detector as well as a frequency source based on optimized ASPAT diodes for millimetre

wave (100GHz) applications The effect of material parameter and dimensions on the

ASPAT source performances was described using an empirical model for the first time

Since this is a new device keys challenges in this work were to improve DC and

RF characteristic as well as to develop a repeatable reproducible and ultimately

manufacturable fabrication process flow This was investigated using two approaches

namely air-bridge and dielectric-bridge fabrication process flows Through this work it

was found that the GaAsAlAs heterostructures ASPAT diode are more amenable to the

dielectric-bridge technique as large-scale fabrication of mesa area up to 4times4microm2 with

device yields exceeding 80 routinely produced The fabrication of the ASPAT using i-

line optical lithography which has the capability to reduce emitter area to 4times4microm2 to

lower down the device capacitance for millimetre wave application has been made

feasible in this work The former challenge was extensively studied through materials

and structural characterisations by a SILVACO physical modelling and confirmed by

comparison with experimental data The I-V characteristic of the fabricated ASPAT

demonstrated outstanding scalability demonstrating robust processing A fair

comparison has been made between ASPAT and SBD fabricated in-house indicating

ASPAT is extremely stable to the temperature The RF characterisations were carried out

with the aid of Keysight ADS software

The DC characteristic from fabricated GaAsAlAs ASPAT diodes were absorbed

into an ADS simulation tool and utilized to demonstrate the performance of MMIC

100GHz detector as well as 20GHz40GHz signal generators Zero bias ASPAT with

mesa area of 4times4microm2 with video resistance of 90KΩ junction capacitance of 23fF and

curvature coefficient of 23V-1

has demonstrated detector voltage sensitivity above

2000VW while the signal source conversion loss and conversion efficiency are 28dB

and 03 respectively An estimate noise equivalent power (NEP) for this particular

device is 188pWHz12

16

DECLARATION AND COPYRIGHT STATEMENT

No portion of the work referred to in the dissertation has been submitted in support of an

application for another degree or qualification of this or any other university or other

institute of learning

COPYRIGHT STATEMENT

i The author of this thesis (including any appendices andor schedules to this thesis) owns

certain copyright or related rights in it (the ldquoCopyrightrdquo) and he has given The University of

Manchester certain rights to use such Copyright including for administrative purposes

ii Copies of this thesis either in full or in extracts and whether in hard or electronic copy

may be made only in accordance with the Copyright Designs and Patents Act 1988 (as

amended) and regulations issued under it or where appropriate in accordance with licensing

agreements which the University has from time to time This page must form part of any

such copies made

iii The ownership of certain Copyright patents designs trademarks and other intellectual

property (the ldquoIntellectual Propertyrdquo) and any reproductions of copyright works in the thesis

for example graphs and tables (ldquoReproductionsrdquo) which may be described in this thesis

may not be owned by the author and may be owned by third parties Such Intellectual

Property and Reproductions cannot and must not be made available for use without the prior

written permission of the owner(s) of the relevant Intellectual Property andor

Reproductions

iv Further information on the conditions under which disclosure publication and

commercialisation of this thesis the Copyright and any Intellectual Property andor

Reproductions described in it may take place is available in the University IP Policy

(httpdocumentsmanchesteracukDocuInfoaspxDocID=487) in any relevant Thesis

restriction declarations deposited in the University Library The University Libraryrsquos

regulations (httpwwwmanchesteracuklibraryaboutusregulations) and in The

Universityrsquos policy on Presentation of Theses

17

ACKNOWLEDGEMENTS

First and foremost all gratefulness and praise is to Allah swt for everything in my

life He is the one and the only one who granted me knowledge health patience and

ability to complete this thesis as well as colouring the whole journey of my PhD

I give my deepest and sincere gratitude to my PhD supervisor Professor Mohamed

Missous for his time support patience and guidance throughout the journey of this PhD

studies His encouragements valuable advice precious ideas and a wealth of knowledge

amp experiences have had a direct inspiration on this research Special thanks also to our

experimental officer Dr James Sexton for not only sharing his knowledge advice and

semiconductor fabrication skills but also his effort in maintaining our clean room

facilities to a great level My gratitude also extends to Mr Mallachi McGowan for his

help and assist in the lab-related issue

I am also obligated to Prof MJ Kelly from University of Cambridge and Dr Kawa

Ian from ICS limited for their measurement of the ASPAT samples on realizing the RF

characteristics This collaboration effort can hopefully last longer in designing and

implementing the ASPAT MMIC detectors

My deepest appreciation also goes to my PhD colleagues Khairul Nabilah Saad

GMuttlak Omar AbdulWahid and Yuekun Wang for their support as well as working

together with me to realize this exciting project directly and indirectly A sincere

thankfulness similarly to my seniors Dr Md Adzhar Zawawi and Dr Fauzi Packeer for

their support during the first and second year of my research For other friends and staff

members under Prof Missous and Dr M Migliorato I will always remember the strong

bond and friendship we made

I am really fortunate that I been blessed with my motherrsquos care who always make doarsquo

for my success every day during my studies As for my beloved wife Dr Nik Maryam

Anisah Nik Mursquotasim who had always encouraged me supported me and gave me

patience through all the hardship in this journey thank you very much

Finally I also would like to thank and acknowledge my sponsor Majlis Amanah

Rakyat (MARA) for financially supporting me during this studies I am greatly indebted

with your kind support which was vital to my study

18

DEDICATION

This thesis dedicated to

My respected and beloved parentshellip

My loving wife dearest siblings and in-lawshellip

19

1 INTRODUCTION

11 Background

It is an undeniable fact that semiconductors have changed the world much further

than anything people could have predicted in the last 60 or 70 years ie after the lsquocats

whisker and vacuum tube eras This field of research has been expanding from year to

year starting from the discovery of the first semiconductor (silver sulfide) in 1833 by

Michael Faraday [1 2] and it still remains very active to the present Semiconductors

have a large range of applications and are not just limited to use in communications they

can be found everywhere in other applications from Earth to space The widespread

usage and sheer number of applications have led to it growing very quickly and

contributing greatly to the growth of World Economics Over time the successful

development of semiconductor growth techniques such as Molecular Beam Epitaxy

(MBE) has enabled researchers to tailor and precisely control the semiconductor

material for new electronic devices with extra functionalities This has led to the

development of advanced devices such as high electron mobility transistors (HEMTs)

and Heterojunction bipolar transistors (HBTs) for use in wireless communication

technology Given this development today electronic devices such as computers

handheld smartphone tablets etc are no longer perceived as luxury and attractive items

but rather have become crucial in everyday life Such devices provide the means to allow

for people to remain connected to each other via the sending and receiving of

information electronically The huge demand for such types of devices has resulted in

competition in both the electronic market and technologies which only goes on to

advance the semiconductor industry

Nowadays the demand for electronic devices characterised by high speed high

efficiency ultra-low power and low manufacturing cost has increased exponentially To

fulfil this growth in demand high data rate systems are required in other words the

system must work at a higher frequency for both the transmitter and receiver The

frequency of interest for advanced wireless communication is in the Millimetre and sub-

20

millimetre wave region which is around 30-300GHz and 300 - 3000GHz respectively

The second frequency region is also sometime known as the terahertz (THz)

electromagnetic region This band lies between the microwave and infrared frequency

bands From the first time it was revealed in the late 80s[3-5] the THz region has gained

a lot of international attention due to its unique properties and since then the motivation

to develop these devices has increased significantly To date the THz frequencies region

has shown its ability to fulfil various applications such as high-resolution imaging in

medical security and surveillance field atmospheric monitoring and environment radio

astronomy as well as compact range radars[3] to name a few

However despite these developments not much effort has been made in exploring

alternative compact THz devices As a result electronic THz devices are still in the state

of immaturity as compared to microwave and photonics devices This is due to their high

cost and absence of compact amp solid-state THz sources (oscillators) and receivers

(detectors) that are capable of operating at both room and extreme temperatures[6 7] A

great deal of work still continues to fill up the lsquoTHz gaprsquo (between 300GHz and 3THz)

used for the most important part of a communication system namely the front-end

receiver or first stage Such a system is responsible for receiving detecting and

processing the received signal to be translated into useful information Furthermore THz

receivers systems still require the best-integrated components such as source mixer and

detector to reach their complete competencies[8] The detector which remains the

critical part of the receiver system requires devices or components that are able to fulfil

the THz gap requirement Studies conducted over a number of years have found out that

the key element in improving THz detection relies upon the use of passive devices ie

diodes Based on these findings many types of diode ie tunnelling diode point-contact

diode and Schottky barrier diode (SBD) have been proposed for detection applications

Amongst microwave and millimetre wave detector diode devices the Schottky

Barrier Diode (SBD) is the dominant detector that has been used since the 1940s[9] The

reason for this dominance is the ease of fabrication of a SBD (by either a point-contact

or evaporated semiconductor-metal structure) and its ability to produce a non-linear

current-voltage (I-V) characteristic which is necessary for rectifyingdetecting diodes [9

10] SBDs also have high cut-off frequency good dynamic range and are low cost To

21

date the SBD has been able to detect signals up to 100GHz [11] 1THz[12] and as high

as 10 THz[13] However the current transport mechanism in a SBD relies on thermionic

emission and therefore is strongly dependent on temperature and means that using them

in extreme conditions ie military and automotive applications is complex The SBD

also suffers from high noise figure[14] and is susceptible to burnout at a modest pulse

power level this will limit the use of ultra-high frequencies and low power signal

applications Other diodes that share the same characteristics are Planar Doped Barrier

(PDB) Germanium Backward Diode (GBD) ie a type of Esaki tunnel diode These

diodes are well known and are reliably used as millimetre wave detectors However it

still proves inefficient to substitute the SDB with any of the previously mentioned

diodes This is due to some drawbacks such as strong temperature dependence limited

dynamic range fabrication complications and hence poor reproducibility (ie GBD) and

other circuit complexities

Hence there is strong compulsion to study examine and produce new detector

diode structures that are able to solve the mentioned diodes limitations and which have

high sensitivity larger dynamic range low noise strong independence to temperature as

well as being able to work efficiently in the high-frequency band and at zero bias The

advantages of working at zero-bias relates very much to the need for a system with less

power consumption so that the device (ie mobile communication) is able to run off

small batteries for a reasonable length of time eliminating extra biasing circuit as well as

noise Therefore a new tunnelling device namely the Asymmetrical Spacer Layer

Tunnel diode (ASPAT) developed by RT Syme [15 16] and refined by Missous et

al[17] has been examined in this work The ASPAT which is in essence a

Semiconductor-insulator-semiconductor structure relies on tunnelling through a barrier

to provide current compared to conventional thermionic emission in SBDs The ASPAT

diode has many advantages a zero bias turn-on voltage very weak sensitivity to changes

in temperature (due to tunnelling) very low noise large dynamic range high resistance

to pulse burn-out [18] and as demonstrated recently can be reproducibly

manufactured[17] The growing interest in THz frequencies nowadays makes the

ASPAT an excellent choice to fulfil all requirements for ultra-high speed applications

22

ie communication (mobile computer networking) radar (military equipment) scalar

analyser and built-in test equipment

In this work an ASPAT diode based on group III-V elements of the periodic table

comprising compound semiconductors of large band gap material Aluminium Arsenide

(AlAs) sandwiched between two lower bandgap Gallium Arsenide (GaAs) are used and

intensively examined The AlAs semiconductor which is ten-monolayer thick has

almost the same lattice constant as GaAs but has a larger bandgap Consequently in the

conduction band a thin barrier of the AlAs is formed from the arrangement of such

structure The structure is made up of GaAs and AlAs both materials are grown on

GaAs substrate using Solid Source Molecular Beam Epitaxy (SSMBE) Therefore in

this study the ASPAT diode will be referred to as ldquoGaAsAlAs ASPATrdquo diode The

conventional GaAsAlAs ASPAT diode has been developed and successfully fabricated

in two different stages This work was the first carried out using facilities provided by

the University of Manchester The first stage of the work was to qualify the

reproducibility and repeatability of growth and fabrication technique which is mostly

performed on larger emitteranode size The second was to develop conventional

ASPAT diodes that can perform at Millimetre and sub-millimetre wave frequencies and

which are comprised of small emitter area

Prior to this work full physical modelling using SILVACO design software was

undertaken to generate models and to fully characterise and identify the fundamental

physical phenomenon of multi-junction ASPAT diode Therefore insight into and

performance based on diode structure and electron movement can be understood and

predicted which lead to the crucial idea in helping and advising iterations to epitaxial

growth as well as diode fabrication The verification of the physical models must be set

as a priority goal by comparing the results of statistically fabricated measured data The

advantage of physical modelling is that it can help reduce materials resources cost and

fabrication time

Further research into the field has led to the development of two other types of

ASPAT diodes that are used to compare with the conventional GaAsAlAs ASPAT

diode Their configuration involved the use of a more advanced semiconductor

technology which comprises InGaAsAlAs materials and GaAsAlAs with InGaAs

23

quantum wells The latter was a novel ASPAT diode and the former is identified as

advanced ASPAT diode However these two advanced ASPAT diodes have not been

extensively studied in this thesis as they will be covered by other co-workers at

Manchester Hence due to these some important parameters are compared to the

conventional one as it is the main focus of this work In the case of temperature

dependent studies the DC characteristic of conventional ASPAT is compared to in-

house fabrication AuGaAs SBD All the ASPATs epitaxial layer materials structures are

shown in the following tables

Table 11 Sample XMBE304 XMBE307 and XMBE368 GaAsAlAs materials structure grown

on GaAs Substrates by MBE

Thickness (Aring)

Layer Material Doping (cm-3

) XMBE368 XMBE304 XMBE307

Top Ohmic1 GaAs (Si) 4times1018

~3000 ~3000 ~3000

Buffer 1 GaAs(Si) 4times1017

~50 ~400 ~50

Spacer1 GaAs NID 50 50 50

Barrier AlAs NID 28 28 28

Spacer 2 GaAs NID 1000 2000 1000

Buffer GaAs(Si) 4times1017

50 400 50

Bottom Ohmic GaAs(Si) 5times1018

~3500 ~4500 ~3500

Substrate GaAs (Si) 50000 50000 50000

Note that sample XMBE368 and XMBE304 are grown on doped GaAs

substrates Sample XMBE368 was grown un-rotated to study the effect of barrier

thickness variation

24

Table 12 Quantum wells sandwiching the quantum barrier for sample XMBE314 grown on a

GaAs substrate by MBE

XMBE314

Layer Material Doping (119836119846minus120785) Thickness (Aring) Bandgap (eV)

Ohmic Layer GaAs (Si) 4times1018

3000 142

Emitter GaAs (Si) 1times1017

400 142

Spacer GaAs Undoped 50 142

Quantum Well In18Ga82As Undoped 60 116

Barrier AlAs Undoped 28 283

Quantum Well In18Ga82As Undoped 60 116

Spacer GaAs Undoped 2000 142

Collector GaAs (Si) 1times1017

400 142

Ohmic Layer GaAs (Si) 4times1018

4500 142

Substrate GaAs

Table 13 Novel InGaAsAlAs ASPAT diode grown on InP substrate by MBE on InP substrate

XMBE326

Layer Material Doping (cm-3

) Thickness (Aring) Bandgap (eV)

Top Ohmic1 In053Ga047As(Si) 5times1019

3000 075

Buffer 1 In053Ga047As(Si) 1times1017

350 075

Spacer1 In053Ga047As NID 50 073

Barrier AlAs NID 283 283

Spacer 2 In053Ga047As NID 2000 075

Buffer In053Ga047As(Si) 1times1017

350 075

Bottom Ohmic In053Ga047As(Si) 15times1019

4200 075

Substrate InP (Si) NID 620000

From the above tables it can be noted that XMBE307 is the first batch that was grown

in-house using a Riber V100HU MBE machine followed by XMBE368 XMBE 304

XMBE314 and finally XMBE326 The two earlier batches were grown on n+

substrate hence their fabrication process flow is simpler On the other hand the three

other batches were grown on semi insulating substrate thus requiring the development

of new repeatable reproducible and robust process flow which will be covered in this

thesis

25

In general the fabrication of the ASPAT diode is based on top-down processes this

is because the ASPAT is a vertical structure device and the junction capacitance of the

ASPAT is directly associated with the size of the anodeemitter area Further to these

the capacitance directly influences the diode cut-off frequency Therefore the simplest

way to reduce the capacitance is by reducing the lateral area of the device of the ASPAT

structure since it can be represented by a parallel plate capacitor where the capacitance

is inversely proportional to the area of the device In order to achieve high cut-off

frequencies minimising capacitance via small dimensions ie sub-micrometre level is

essential However this will also increase the series resistance of the diode As a result

the cut-off frequency will be degraded Thus there is a trade-off between small

dimension of device and high cut-off frequency to be achieved

Finally successful growth and fabrication for small area GaAsAlAs ASPAT diode

in this work has led to carefully extracted RF characteristics This becomes a stepping

stone to designing a millimetre wave integrated circuit (MMIC) detector using empirical

modelling in Keysight ADS tools Therefore a predicted performance for a 4times4 um2

fabricated ASPAT is that can operate at 100GHz ASPAT as a zero-bias detector with a

voltage sensitivity of over 2000VW Additionally the design of a millimetre wave

source using similar ASPAT diodes was also carried out The performance of a 2040

GHz doubler using GaAsAlAs ASPAT in varistor mode is demonstrated for the first

time with a conversion loss of 33dB and conversion efficiency of ~ 02

26

12 Aims and objectives

The aim of this study is to further improve the performance of microwave and

millimetre wave technology by incorporating the Asymmetrical Spacer Layer Tunnel

Diode (ASPAT) for ultimate operation near THz frequencies by designing a range of

low power high-speed devices enhancing the methods of Simulation layout and

materials amp structural characterisations with fabrication process optimization using the

facilities available at the University of Manchester

There are three main objectives in this research firstly to streamline the physical

device design and modelling using the GaAsAlAsGaAs materials systems in order to

produce a zero bias detector which is basically a rectifier of a microwave signal by

using the SILVACO Atlas simulation tools

Secondly to achieve reproducibility and manufacturability of the fabrication

process for new type of GaAsAlAs ASPAT structure (lateral structure) hence small size

ASPAT emitter by improving the device processing technique and maximising the

capability limit of the conventional i-line optical lithography that is available in Prof

Missousrsquos group laboratory

Thirdly to optimise DC parameters through electrical properties investigation as a

stepping stone to the next objective that is to characterise the RF performance of the

GaAsAlAs ASPAT detector circuit The detection properties of microwave and

millimetre wave diode will also be investigated with different ASPAT diode size at

100GHz Further to these the properties of microwave signal source will also be

developed by way of utilizing the non-linearity feature of the diode Therefore this new

type of tunnelling diode can be applied to both applications of signal detection and

signal source in the microwave and millimetre wave ranges

27

13 Outline of this Thesis

This thesis is organized into eight chapters The first chapter discusses the

contextual information that motivates the undertaking of the study An overview of the

work which includes the details of the studied samples the aim and objectives of the

whole research project are also outlined in this chapter

Chapter 2 deals with the literature review of the basic principles and concepts of

the group III-V compound semiconductors The historical background of such

semiconductors which is essential to the development of ternary structures etc and the

advancement of semiconductor materials engineering is presented The types of existing

tunneling diode as well as conventional microwave diodes are also discussed and

compared The fundamentals of ASPAT diode which includes structural parameters and

its operation are then explained in detail Finally discussions of the ASPAT key DC

characteristics which are important for detection purposes are presented

Chapter 3 focusses on the development of the experimental techniques which can

be divided into two stages In the first stage the development is towards repeatability

reproducibility and manufacturability of the ASPAT grown in-house by MBE and

fabricated by conventional i-line optical lithography The second stage involves

optimisation and fine tuning such fabrication method for GaAsAlAs ASPAT samples

that can operate at high frequency ie 100GHz detector For both stages of the

fabrication process all techniques including mask design generic and special process

flow are presented The chapter ends with discussions on issues related to sample

processing and improvements that are proposed to solve these issues

Chapter 4 dwells on the modelling of the GaAsAlAs ASPAT using the SILVACO

simulation package The discussions are expected to offer a better understanding or

insight into each layer that forms the ASPAT diode structure The chapter begins with

discussions of the operation of the SILVACO Atlas tool A validation of physical

modelling is essential and presented according to the fabricated mesa sizes of the diode

28

Thereafter towards the end of the chapter the analyses of the relationship between

device current-voltage (I-V) characteristics the structural parameter including various

temperatures dependent simulations with a comparison to an in-house fabricated SBD

are offered

Chapter 5 presents relevant DC results based on optimized fabrication process and

RF characterization which enable obtaining an intrinsic and extrinsic element of the

GaAsAlAs ASPAT diode The discussions also highlight the analysis of DC zero bias

equivalent circuit and de-embedding extraction using ADS The chapter ends with

discussions on the RF reproducibility performance which includes the performance as

well for millimeter-wave and sub-millimeter wave applications

Chapter 6 discusses the main applications of ASPAT diodes The chapter begins

with discussions on detection theory followed by the parameters of interest and ends

with circuit design as well as the performance of a 100GHz detector The circuit design

was conducted using Keysight ADS software via harmonics balance simulation tool The

performance in term of sensitivity depending on measured ASPAT emitter size is

demonstrated Finally a comparison with conventional Schottky diode is presented

towards the end of the chapter

Chapter 7 discusses a secondary application that can be applied to the ASPAT by

utilizing the nonlinearity feature of the diode to create a signal source namely a 20 to

40GHz frequency doubler in varistor mode The doubler performance of ASPAT will be

explored through circuit design constructed via Keysight ADS simulation software Each

key parameter is highlighted and discussed in detail

The final chapter of this thesis that is Chapter 8 discusses the conclusions of the

study with emphasis on the overall key research findings The chapter also highlights

suggestions for further research in this particular field of study

29

2 LITERATURE REVIEW

21 Introduction

Since 1940s the development in the technology of semiconductor electronics has

been expanding and now has led to the establishment one of the most astonishing

industries of the 3rd

-millennium era Leading this advancement is the integrated circuit

(IC) or chip which was driven mostly by silicon (Si) Overtime the IC has undergone

substantial revolution in term of power economics size and efficient energy

consumption Currently it covers every aspect of human life ie from desktop personal

computers in the office and house to the compact smartphone in the pocket and from a

gigantic satellite in space to small satellite navigation in cars In other words

semiconductor technology is crucial to human life Without developments in

semiconductor materials engineering and shrinking of device size such accomplishment

may not have been realised today Therefore this chapter presents a macro view of the

development in compound semiconductor technology especially in radio frequency (RF)

towards Millimetre and submillimetre wave applications with regard to the improvement

of material and device structures

The essentials of group III-V compound semiconductor will be emphasised for its

points of interest and application in this field (RF technology) This chapter comprises

five main sections The first section is an overview of the semiconductor history with

concentration on its advantages and applications in the RF field while the second and

third discuss the effects of III-V compounds when the interface occurs between

semiconductor-semiconductor and semiconductor-metal respectively which leads to a

basic understanding of hetero-structures device as well as contacts namely Schottky and

Ohmic The fourth section is predominantly concerned with high-speed devices ie

diodes and materials in this field which leads to the exploitation of the main researchrsquos

device Then the following section describes in detail the background works basic

principle and intrinsic amp extrinsic parameters of the Asymmetric spacer Tunnel Diode

(ASPAT) Finally the basic way of characterising the device will also include giving an

overview of how the device is measured and what parameters are needed

30

22 Historical review of III-V Compound Semiconductor for RF applications

The beginning of commercial electronic devices was marked with the first point-

contact semiconductor transistor developed in 1947 by William Shockley at Bell

Laboratories in New Jersey Shockley developed a device based on a Germanium

Bipolar Junction Transistor (Ge BJT) structure [19] with operating frequency above 1

GHz Since then and until early 1950s the development of Ge BJT was fast and it

became foremost in the market of semiconductor technology However the emergence

of Silicon (Si) challenged Ge in the market in the 1960s Si has the upper hand primarily

because it has better electron transport and low manufacturing costs compared to Ge

[19] By the 1970s almost all RF transistors were based on Si BJT Additionally the

development of Si which forms a new material from the formation of native oxide

namely Silicon Dioxide (SiO2) led to the invention of the Metal Oxide Semiconductor

Field Effect Transistor (MOSFET) [19] The future of digital electronic industries has

been ldquobrightrdquo ever since the MOSFET was ldquobornrdquo as it has become a fundamental

building block component in complex microprocessors and flash memories Despite this

development the exploitation of Si at RF frequencies did not last long since Si is not an

optimum semiconductor for RF electronic devices The emergence of GaAs has

improved RF applications for high-speed transistors

Ge and Si which are categorised as single element semiconductor are the earliest

materials used to build the first transistor devices These devices played a crucial role

towards the development of more advanced material such as GaAs of the compound

semiconductor type[20] A compound semiconductor is a semiconductor formed by the

ionic bond of different types of semiconductor material most widely known as the group

III-V compound semiconductors The main reason for the progression of the III-V

compound semiconductors is due to their better electron mobility compared to the single

element semiconductors The term ldquomobilityrdquo in the semiconductor industry refers to the

easiness of movement of charges in many directions inside a crystal In fact it is

determined by the access resistances values with saturated velocity under certain values

of electric fields (bias) the higher the electric field the faster is the carrier movement in

the crystal Figure 21 shows the electron mobility and band gap for the most common

31

group III-V compound semiconductors Besides higher mobility III-V compound

semiconductors also have light-emission capability and are suitable for bandgap-

engineering techniques

The work on III-V compound semiconductors mainly on GaAs FETs led to a new

change for the whole RF electronics industry For example in 1966 the first GaAs

MESFET was invented[21] and achieved a maximum operating frequency of operation

of 3GHz [22] Three years later the frequency increased to 30GHz [23]

Figure 21 III-V compound semiconductors mobility and band gap[24]

With better features in terms of having a higher electron mobility compared to Si

electronic devices based on III-V materials developed rapidly This attracted attention in

many aspects especially in military radar application electronic warfare system missile

guidance control electronic for smart warfare system and secure communication To be

specific those demands were fulfilled through the application of microwave mixer and

detectors [25] which were achieved based on Schottky barrier diodes and FETs

However these applications remained largely as niche markets for use only in military

and exotic scientific projects until 1980 In addition to the microwave industry two

important diodes that played a large role in very high-frequency power source namely

the Gunn diode and the Impact Avalanche and Transit Time (IMPATT) diode which

were discovered in the 1960s[26]

100

1000

10000

100000

0 05 1 15

Bu

lk M

ob

ility

(cm

2 V

-1 s

-1

BandGap (eV)

InS

b

InA

s

Ge

Ga

Sb

In

GaA

s

Si

GaA

s

InP

32

Furthermore the invention of Molecular Beam Epitaxy (MBE) growth technique at

the beginning of 1970s has enhanced the full potential of the III-V compound

semiconductors[27] This technique has led to the formation of a new class of materials

and heterojunction device with high-quality interfaces and accurate control of the

thickness during growth[28] The advancement of material engineering that tailored the

III-V compound semiconductor with MBE effect has been beneficial for both three-

terminal and two-terminal devices As a result of this more advanced devices in both

electronics and optics were developed such as quantum well (QW) laser Resonant

tunnelling diode (RTD) high electron mobility transistor (HEMT) and many more[29]

The aim was to achieve high-speed devices transporting data at high data rates and

robust devices These devices promised an excellent option to conventional transistor

(three terminal devices) in high-frequency systems especially in the terahertz (THz) or

Millimetre and sub-millimetre wave regions [30]

One of the promising diodes that received a lot of attention is the resonant

tunnelling diode (RTD) which was first described in 1974 by Chang [31] This device

which consists of a double barrier and one quantum well is the classical tunnelling diode

Due to its good symmetrical non-linearity in its current- voltage characteristic it can be

exploited for signal generation and detection However the main focus of RTD to date

has been in the generation of continuous wave (CW) ultra-high frequency and to a lesser

extent in detection Therefore other tunnelling based diodes were developed specifically

for detection purposes which are the main foci of this work The PDB and ASPAT

diodes are the workhorse candidates for detection purposes Most of these are built based

on group III-V compound semiconductors [32]

Unlike the Very Large Scale Integration (VLSI) market ie CMOS for personal

computer (PCs) the RF electronic device for civilian application reached the consumer

market only in the late 1980s through satellite television with operating frequencies

around 12GHz [19] Since then many RF application have been deployed on the mass

market depending on their operating frequency such as 09GHz ndash 25GHz for wireless

communication 20GHz to 30GHz for satellite communication 77GHz for car radar

systems and above 90GHz for different sensor applications Utilising GaAs as the main

material RF devices have become the key underpinning components for modern

33

communication systems As a result in 1998 the volume production of mobile phones

was greater than that of PCs for the first time in history Presently production is being

made for devices like smartphones cellular phones mobile internet access and new

communication services and tablets

The development of the RF field is never ending More and more improvements are

being made especially through the design and fabrication of oscillators and detectors

which are mainly built based on group III-V compound semiconductors When RF

devices were used by the military (in the 1970s to 1980s) cost was not a concern

However after getting into civil application market (ie in the 1990s) the most frequent

issues highlighted were performance and cost[19]

The ability to generate or receive high operating frequencies with high power large

bandwidth and high sensitivity is an indicator for a good performance of RF devices

(depending on specific applications) For example the highest room temperature based

oscillator of up to 186 THz was achieved in thin well AlAs-InGaAs RTD by Professor

Masahiro Asada from Tokyo Institute of Technology [33] An excellent review on THz

sources can be found in [34] For ultra-high frequency detector and mixer applications

the two terminals RF device that is mostly used is the SBD In 1996 the highest cut off

frequency achieved by a mixer utilizing the SBD was about 5THz[35] and this has kept

increasing ever since The factor that motivates the development of THz devices is the

requirement to have a compact coherent source in the THz range Undoubtedly in the

future there will be very exciting times for enthusiasts of terahertz sources and receiver

as new generations of compact broadband and tuneable solid source device based on

advanced compound semiconductor are developed

23 The Concept of Heterostructures

A III-V compound semiconductor is mostly grown on a single semiconductor

substrate forming a layer called epitaxial heterojunction layer It is a starting point and

the key feature that brings the idea of realising the most advanced semiconductor

devices currently being developed and manufactured by combining several epitaxial

semiconductors [36-38] Heterojunctions have the capability of manipulating carrier

transport ie electron and holes transport in crystal separately unlike homojunctions

34

This has resulted in the successful development of new devices for high-speed and high-

frequency applications as well as optical sources and detectors [37] This section will

discuss lattice matched material pseudomorphic material hetero-junction band

discontinuities and quantum wells

231 Homojunctions Heterojunctions and Band Discontinuities

The term homo-junction refers to the interface between identical semiconductor

materials that have different polarity ie p-type or n-type but similar in energy gap This

phenomenon is usually applied in forming p-n junction diodes and can be understood by

referring to Figure 22 below

Figure 22 illustration of Homojunctions band structure material before (left) and after (right)

equilibrium

The materials A and B which have similar bandgap (Eg) and different dopant

types ie p-type and n-type will have their Fermi levels (Ef) closer to the valence band

(EV) and conduction band (EC) respectively before ldquothermal equilibriumrdquo Once

equilibrium is achieved Ef of both p-type and n-type will be aligned causing band

bending of EC and EV As a result a built-in electric field is introduced (via diffusion of

carriers) for both holes and electrons and forcing them to move in one direction

On the contrary a heterojunction occurs when the interface between two

semiconductor materials with different bandgap energy are brought together (ie large

energy band gap material combined with a low band gap one eg wide band gap AlAs

and narrow bandgap GaAs) This results in a steep band bending which leads to the

formation of energy band discontinuities at the junction as shown in Figure 23 In a

semiconductor heterojunction the most important parameter is the band gap energy

EC

EV

Ef

Eg

p-type E

g

n-type

EC

EV

Ef

Material A

p-type

n-type

Material B Material A Material B

35

associated with each material in the structure where the degree of discontinuity can be

utilised in varying the carriers transport properties as well as the quality of the junction

depending on the interest of the designer This leads to flexibility in tailoring device

characteristics leading to vastly improved performance of the device

Figure 23 Energy band diagrams before (left) and after (after) interface combination at

equilibrium

Based on Figure 23 above Material A indicated with blue line is a large band

gap energy material and Material B highlighted with the red line is a low band gap

material EV represents the valence band EC the conduction band and Ef is the Fermi

level of the materials Alternatively the two materials band discontinuities are denoted

by ΔEC for the conduction band and ΔEV for the valence band χ and Eg represent the

electron affinity and band gap energy respectively

At some point where by the Fermi energy of both semiconductor materials are levelled

the structure would have reached its thermal equilibrium The band gap of materials A

and B have a discontinuity at the interface (ΔEg) of these two materials In general this

is given by

120549119864119892 = 119864119892119860 minus 119864119892

119861 (21)

Furthermore when thermal equilibrium is achieved ΔEg is then divided between

conduction band and valence band discontinuities (ΔEC and ΔEV respectively) at the

material A and B junction interfaces Their relationships can be expressed as

EC

A

EV

A

Ef

A

Eg

A

Material A Material B

χA

χ

B

ΔEC

ΔEV

Ef

B

E

g

B

EC

B

EV

B

EV

A

χA

ΔE

C

ΔEV

EC

A

Ef

EC

B

EV

B

χB

Material A Material B

Vacuum

Level Vacuum

Level

36

∆119864119862 = 120594119860 minus 120594119861 (22)

∆119864119881 = (119864119892119861 minus 119864119892

119860) minus (120594119860 minus 120594119861) (23)

∆119864119892 = 119864119892119860 minus 119864119892

119861 = ∆119864119862 + ∆119864119881 (24)

However these relationships which were introduced by Anderson can only offer an

approximation In practice the results are always different since dislocation and

interface strain occur at the junction Therefore precise control during epitaxial growth is

always required and growth technologies such as MBE are employed In due course the

band gap discontinuity can be further exploited by using different types of material

combination Examples are GaAsAl052Ga048As and In053Ga047As In052Al048As [39]

232 Lattice-Matched and Pseudomorphic Materials

As discussed earlier a heterojunction happens when any two different

semiconductor materials that have different bandgap are joined together At the atomic

level both materials often differ in lattice constant The easiest way to explain this is by

setting the formation of heterojunction which can be separated into two types lattice

matched and lattice mismatched (pseudomorphic)

2321 Lattice Matched Systems

To create discontinuities for use as a high-performance device the combination of

semiconductor materials is essential Selecting the appropriate materials that have

similar or very close lattice constants to combine is crucial to avoid disruption at the

atomic level heterojunction interface Figure 24 shows that the lattice constant of a

material A ie substrate (aS) and material B ie deposited over layer (aL) are identical

or very close and their surface atoms are perfectly matched This scenario is known as

lattice matching

37

Figure 24 Lattice Matching for both materials when aL=aS

As can be seen in Figure 25 while there are restricted binary materials available to

form good heterojunction interfaces it is possible to combine semiconductor materials in

binary ternary and quaternary forms to allow the formation of a variety of lattice-

matched heterojunction interfaces The examples of materials that have successfully

been alloyed are In053Ga047As In052Al048AsInP and GaAsAlxGa(1-x)As (x=0 to 1)

Even though the materials system hetero-junction of these materials has close lattice

constant value their band-gap will experience an abrupt variation

Figure 25 Energy gap and lattice constant for direct and indirect band gap of compound

semiconductor at ambient temperature[40]

Material A(layer)

Material B(substrate)

aL

aL

aS

aS

38

The existence of binary ternary and quaternary semiconductors formed by alloying

semiconductors has expanded the opportunity for heterojunction formation in devices

The alloy semiconductor which is produced by the combinations of two semiconductors

A and B has a lattice constant that obeys Vegardrsquos Law as follows

119886(119886119897119897119900119910) = 119909119886119860 + (1 minus 119909)119886(119861) (25)

For the alloy the band gap normally follows the virtual crystal approximation

119864119892(119886119897119897119900119910) = 119909119864119892(119860) + (1 minus 119909)119864119892(119861) (26)

Table 21 shows the list of the semiconductor alloy band gap and lattice constant for

common binary and ternary for group III-V compound semiconductors[41]

Table 21 Lattice constant (a) and band gap for group III-V binary and ternary compound

semiconductors a room temperature [41 42]

Alloy Lattice constant a(Aring) Band gap EgeV)

AlAs 5661 2239

AlSb 6136 1581

GaAs 5653 1424

GaN 3189 34

InAs 6058 0417

InP 5869 1344

Al052Ga048As 5657 2072

In053Ga047As 5868 0773

In052Al048As 5852 1543

39

2322 Pseudo-morphic Materials

The other scenario is when two different materials with different lattice constants

are brought into contact The observation can be made at the atomic level where the

atom will try to match each other as shown in Figure 26 below

Figure 26 Lattice mismatched material

In fact for both situations (ie lattice matched and lattice mismatched) the atom of

the material at the hetero-interface will change their position to maintain the geometry of

the lattice Due to distortion at this atomic level a strain is then induced at the hetero-

interface In order to form a good hetero-junction interface the strain must not exceed a

certain specific critical value which will cause crystal dislocations to occur The result of

crystal dislocation is generally bad as it will affect the carriers which will be

concentrated in the defect area thus degrading the carriersrsquo mobility This then makes

the overall function or performance of the device to become poor

Nowadays the Molecular Beam Epitaxial (MBE) technique is able to grow epitaxial

layers of mismatched semiconductor layers profile ie mismatched in their lattice

constant (aLneaS) The growth method works when the grown epitaxial layer assumes the

lattice parameters of the layer it is deposited on Nonetheless the layers must be kept

within a certain limit and the deposited layer must be thin enough to avoid defect or

dislocation formations This new layer known as a ldquopseudomorphicrdquo material will alter

its original crystal structure and physical properties

Material A(layer)

Material B(substrate)

aL

aL

aS

aS

40

Figure 27 The formation of Pseudomorphic-layer in two situations (a) compressive and (b) tensile

strain [1]

Figure 27 shows material A in which the pseudomorphic materials can be related to

two situations compressive and tensile strain The compressive strain occurs when the

deposited layer has a larger lattice constant than the substrate (aL gtaS) while tensile strain

happens when the deposited layer has a smaller lattice constant than the substrate

(aLltaS) These leads to aL either to compress or stretch to fit aS respectively Note that

the pseudomorphic layers can only be grown to a certain critical thickness hc From

Figure 27 the strain between the substrate and the deposited epitaxial layer is given by

휀 =119886119871 minus 119886119878

119886119878 (27)

Where Ɛ is strain between two layers aL is lattice constant of the deposited layer and

aS is lattice constant of the substrate layer The concern in deposition of the over layer is

to avoid dislocation occurring at the interface if there is too much strain at the junction

The strain is naturally influenced by the thickness of the deposited layer and thus the

thickness of growth must be controlled below the critical thickness hc which is

expresses as

(b)

41

ℎ119888 =119886119904

(28)

Moreover one needs to appreciate that even though the crystal structure and their

physical properties change the total energy within the unit cell is maintained This is

possible by distortion of the deposited layer in the direction perpendicular to the growth

direction while leading to lattice matching in the lateral plane Example of lattice

matched materials is GaAsAlAs and pseudomorphic material is In08Ga02AsInP

233 Quantum well and 2DEG

A typical application of heterojunction interface is one in which utilises ΔEC and ΔEV to

form barriers for electrons and holes One example of barriers that confines these

carriers is known as a Quantum Well (QW) A QW is a layered semiconductor usually

very thin ie about ~ 100 Aring thicknesses in which many quantum mechanical effects can

occur It is formed by a thin layer of a low bandgap energy semiconductor material eg

GaAs sandwiched between two similar large bandgap energy semiconductors eg AlAs

or AlGaAs The growth technique to achieve thin layers of QW is usually MBE The

benefit of this method is that it allows the formation of heterojunction with very thin

epitaxial layer

Figure 28 The band diagram of an ideal quantum well formed by a narrow band gap

semiconductor eg InGaAs sandwiched between two large band gap semiconductors eg GaAs (a)

Structure (b) energy band diagram and (c) Conduction band diagram when AlGaAs is n-doped[43]

42

The thickness of the layer that can be achieved can be as thin as the electron mean free

path (De Broglie wavelength) which is around 100 Aring to 300 Aring [44] The expression for

the De Broglie wavelength is given by

120582 = ℎ120588frasl (29)

Here h and ρ are Planckrsquos constant and momentum of the electron respectively Figure

28(b) illustrates a quantum well formation in abrupt semiconductor interfaces It can be

observed that the heterojunction boundary will experience discontinuities at the edges of

the conduction band and valence band with a quantum well generated for the carriers

(both electron and holes) The quantised energy sub-bands in the quantum well structure

in Figure 28(b) can be determined from [43]

119864 = 119864119899 + (

ℏ2

2119898lowast) (119896119909

2 + 1198961199102)

(210)

Where 119864119899 = (ℏ21205872

2119898lowast ) (119899

119871)2

and n is the energy level index that can be n=1 2 3hellip

The dopants in a semiconductor with large band gap layers may supply the

carrier to the quantum well and this occurs when the base or bottom of the quantum well

is lower than the Fermi Level and hence the high energy donors will go down to the

well therefore creating a Two-Dimensional Electron Gas (2DEG) In the 2DEG the

electrons and holes move freely in the quantum well in the plane perpendicular to the

growth direction however they are not capable of moving in the crystal growth

direction (confinement direction)[45 46] The 2DEG phenomena can be seen in Figure

28(c)

24 Metal-Semiconductor Contact

A semiconductor device is incomplete if there is no connection between the

semiconductors and the outside world A metal which is usually gold (Au) or gold

germanium (AuGe) is diffused into the semiconductor to allow for electrical connection

from the outside world to the semiconductor and vice versa The metal-semiconductor

contact can be either a Schottky contact or an Ohmic contact The Schottky contact is a

43

rectifying contact while the Ohmic contact is a contact that provides a low resistance

path between semiconductor and metal

241 Schottky Contact

The Schottky contact is basically a metal contact to the gate to enter a region or

channel in a transistor Figure 29 shows a schematic band diagram of a metal-

semiconductor contact before and after contact (Schottky-Mott concept)[47]

Figure 29 Metal and semiconductor in two conditions (a) separation (b) in contact

In Figure 29 the work function of the metal is represented by qm while the

semiconductor work function is qS The qχ is the energy difference of an electron

between the vacuum level and conduction band edge ie known as the electron affinity

and qVn is the difference between the conduction band and Fermi level in the

semiconductor EV EC Ef is the valence band energy conduction band energy and the

Fermi level respectively

The metal and semiconductor are brought together as showed in Figure 29(a)

both materials are at steady state However when the metal and semiconductors are in

contact as illustrated in Figure 29(b) the electrons that flow from the conduction band

in the semiconductor into the lower energy state of the metal will cause the Fermi level

to be aligned in thermal equilibrium Due to this process the positive charge donor is

trapped in the semiconductor interface hence forming a depletion region Xdep

Thereafter the upward bending of the energy in the semiconductor takes place On the

qχ(s)

Eg(s)

Vacuum Level

EV

EF(m)

E

F

EC

qϕ(m)

qϕ(s)

Metal Semiconductor

qVn

qϕB

qϕ(s)

qVbi

X

qχ(s)

Eg(s)

Vacuum Level

EV

EF

EC

qϕ(m)

Metal Semiconductor

Xdep

(a) (b)

44

other hand the negative charge (electron) will be accumulated within a narrow region in

the metal interface The existence of two different charges at the metal-semiconductor

boundary generates an electric field This leads to a potential barrier qB as seen by

electrons in the metal moving into the semiconductor and a built-in potential qVbi as

seen by electrons in the semiconductor trying to move into the metal

The built-in potential qVbi is defined as follows

119902119881119887119894 = 119902empty119861 minus 119902119881119899 (211)

The barrier height empty119861 in the ideal case is specified by the dissimilarity between a metal

work function empty119898 and electron affinity of the semiconductor

119902empty119861 = 119902empty119898 minus 119902120594 (212)

Referring to Eq (212) above the barrier height empty119861 rises linearly with the metalrsquos work

function empty119898 Nevertheless this is only in theory as the presence of localised surface

stated at the edges causes empty119861 to become unresponsive to the metal work function

Consequently Eq(211) is then reordered to match the difference in metal and

semiconductor work function Thus the new equation becomes

119902119881119887119894

= 119902(120601119898 minus 120601119904) (213)

A Schottky contact appears when a metal-semiconductor contact has a large

barrier height (B ge kT) and low doping concentration in the semiconductor (ND le NC) In

the case when the metal-semiconductor contact is under some bias eg reverse bias the

semiconductor will react to a positive bias according to the metal by a voltage V=-VR

This condition will affect the built-in potential and leads to increase from Vbi to

(Vbi+VR) thus increasing the barrier height empty119861 in the semiconductor as well

Consequently electrons are less able to flow from the semiconductor and cross into the

metal Therefore the current flow will be very small This phenomena is shown in

Figure 210(a)

45

Figure 210 Energy band diagram of Schottky contact on n-type material under (a) reverse and (b)

forward bias

As can be seen from Figure 210(b) when a forward bias is applied the semiconductor

is biased negatively with respect to the metal by a voltage V=Vf This will result in a

reduction in built-in potential from Vbi to Vbi-Vf The electrons in the semiconductor will

lower the barrier height and a lot of electrons will escape into the metal causing a large

current to flow Thus a large current flow in the forward direction compared to the

reverse direction Essentially this is the origin why the Schottky contact is named a

rectifying contact [48 49] For the metal side both forward and reverse biases applied

do not affect the barrier high empty119861 because there is no voltage drop there

In this system the electron and holes are transported by a phenomenon called

Thermionic Emission (TE) which happens when the semiconductor layer is lightly

doped Nd lt 1x1017119888119898minus3 The electron will only be thermionically emitted into the metal

when the energy is higher than the potential barrier[50] There is another phenomenon

called Thermionic Field Emission (TFE) which happens when the potential barrier

thickness is very thin (thin enough) to allow the electron to tunnel through the barrier

This will be discussed in the next section as this phenomenon leads to the formation of

an ohmic contact

(a) (b)

46

242 Ohmic Contact

Basically an ohmic contactrsquos purpose is to provide a low resistance path from

the semiconductor to the outside world It is different to a Schottky contact as it is a non-

rectifying contact and does not control the current flow the I-V characteristic of an

ohmic contact is linear in both forward and reverse directions (equality in current flow)

The ohmic contact also has a small voltage drop across it compared to the voltage drop

across the device

If a metal and semiconductor are bought together unavoidably a Schottky

contact will be formed Therefore to create an ohmic contact some techniques to reduce

barrier height and width of the depletion region must be used ie increase Nd In carrier

transport theory there are three mechanisms of carrier transport across the barrier

Firstly the Thermionic Emission (TE) which happens when the carries are excited to

overcome the barrier when the thermal energy is present Secondly the Thermionic

Field Emission (TFE) occurs when the electronholes have enough energy to tunnel

through an adequately thin barrier and some has overcome the low barrier at the top

Finally the Field Emission (FE) which results when carriers can tunnel through the

entire barrier The FE is the most favoured mechanism in the ohmic contact approach

[51]

From the three mechanisms above the current can be determined by the following

equations

1) Current in Thermionic emission (Figure 211(a))

exp (empty119861)

119896119879

(214)

2) Current in Thermionic field emission (Figure 211(b))

exp [

(empty119861)

11986400119888119900119905ℎ11986400

119896119879

] (215)

3) Current in Field Emission (Figure 211(c))

exp (empty119861)

11986400

(216)

47

Where k is the Boltzmann constant T is the temperature 11986400 is the tunnelling parameter

and is related to the doping concentration radic119873119863 The barrier height is denoted by empty119861

Figure 211 Figure 10 N-type semiconductor Ohmic contact with different doping concentration

ND (a) Low (b) Intermediate and (c) high

Figure 211 shows that the carrier transport mechanism is varied by the doping

concentration (ND) As can be seen from Figure 211(c) the doping concentration here is

the highest and influences the depletion region width Xdep to become smaller Therefore

Field Emission (FE) becomes dominant This FE method is the favourite method for

ohmic contact formation [51] and will be utilised in the fabrication carried out in this

work

In fabricating practical devices the ohmic contact is often split into two types

alloyed and Non-Alloyed The difference between the two is that the alloyed type is used

when the semiconductor is doped with a low doping concentration ie less than

1x1018119888119898minus3 while the Non-Alloyed is designed for heavily doped semiconductors with

more than1 times 1019119888119898minus3 doping

The alloyed ohmic contact requires thermal annealing to have a good performance

for electron transport In multi-layer metals one of the metals has the role of donor or

acceptor which is used to increase the doping concentration of the semiconductor If a

temperature anneals eg 420˚ Celsius is applied the metal will diffuse into the

semiconductor and carry the dopant into the semiconductor Therefore a heavily doped

region will be formed and the depletion width becomes narrow establishing the ohmic

contact The key example of this is the usage of the Gold-Germanium-Nickel (Au-Ge-

Ni) alloy where the Ge is the n-type dopant [52] which diffuses into the semiconductor

(a) (b) (c)

48

and perform atom replacement in the semiconductor ie in GaAs it replaces Ga On the

other hand the Non-Alloyed does not require any thermal annealing as it already has a

very high doping concentration and will automatically reduce the depletion region width

The Non-Alloyed ohmic contact has some advantages such as reproducible contact

reduced processing time and good uniformity [53]

25 Asymmetrical Spacer-layer Tunnel (ASPAT) diode Background work

In this section some historical background of the Asymmetrical Spacer-layer

Tunnel (ASPAT) diode is given Since this is the first thesis reporting about this new

device it is worth to mention some historical background about this tunnelling diode

The ASPAT was first proposed by a group of scientists from General Electrical

Company (GEC) in 1990[16] The works led by Richard T Syme and assisted by

Michael JKelly Angus Condie and Ian Dale initiated the idea of launching a new type

of tunnel diode The idea managed to attract the interest of many parties following the

development of resonant tunnelling diode (RTD) which earlier had shown a promising

weak temperature dependence [54] However the interest in RTD is mainly limited to

microwavesub-millimetre wave generation For THz detection the requirement is to

have a significantly asymmetric IV characteristic Given this the ASPAT which has

only a single energy barrier and most importantly weak temperature dependence and

large dynamic range would be a promising candidate for this application

The development of ASPAT is a kind of reverse engineering since it was built

purposely to replace the earliest receiver diode especially the Schottky Barrier Diode

(SBD) which has strong dependence on operating temperature [55] From the time when

it was first revealed a lot of works have been done to realise this most sophisticated

tunnel diode The first attempt which was reported in [56] was meant to gather some

insights into the device by using the well-known Schrodinger and Poissonrsquos equations

for simulation The second attempt on the other hand was directed to physically grow

and fabricate the device Here the real problem occurs At the first stage of qualifying

this device it was found not to be manufacturable Since then a new tunnel diode

structure based on GaAsAlAs materials system was built by both MBE and MOCVD

Its microwave performance was then tested at 94GHz [18] The same paper also

49

reported performance comparisons between ASPAT and another microwave diode ie

Germanium Backward Diode (GBD) PDB and SBD

Work on these devices stopped due to the inability to commercialise the ASPAT

and other tunnel based devices [57-60] The problems associated with low-cost

manufacture of tunnel diodes are due to firstly the thickness of the AlAs barrier layer

the dependence of tunnelling probability (electron) through a single barrier is

exponential and varies by a factor of more than 350 for one monolayer change in the

AlAs barrier thickness[61] The tunnelling of the electron through a barrier is

proportional to the current through a barrier as a function of a bias across the AlAs

barrier[62] Secondly the bandgap which is predominantly happens to be a ternary alloy

with relative composition ie AlxGa1-xAs Here the x can vary the bandgap in the

semiconductor layer For the ASPAT a 1 change in x results in a 30 change in the

current[62] To design an ASPAT for microwave and THz applications the designer

often allows at most plusmn10 variation of the absolute current through a specific diode at a

pre-identified bias This implies that within a wafer the uniformity that the ASPAT must

achieve is less than plusmn01 monolayers while between wafer to wafer the reproducibility

in barrier thickness in average must be identically controlled[63] This explains why at

the qualification stage of investigating ASPAT there was a need to focus on

GaAsAlAs-based material to diminish further errors because of the change in x This

type of work on ASPATs has been carried out by other co-workers at Manchester and

Cambridge

Thereafter the work then focused on repeatability and manufacturability tests

These result in many attempts being carried but failed with unacceptable between wafer

to wafer reproducibility [61 64] The development of reproducibility and repeatability of

the ASPAT was pursued for over 10 years until precise control of the growth of the

thickness AlAs layer was finally achieved using MBE[65] [66 67] This achievement is

confirmed by a current density produced which varied by less than plusmn30 indicating that

the reproducibility of AlAs barrier of the order of plusmn 02 monolayers A final step to

achieve the level control for the ASPATrsquos AlAs barrier thickness was carried out and

resulted in a 1 standard deviation of the IV characteristics for both within a wafer and

different wafer (2 inches wafer size)[17] In the early stage of this work some

50

repeatability and manufacturability test was also carried out Once this vital step is

accomplished further investigation on the ASPAT was made most recently and which

will be covered in this thesis ie temperature independence[68] and new ASPAT

types[69] characterisation to achieve smaller device RF measurement and development

of THz detectors The material systems that have been investigated so far are

GaAsAlAs and InGaAsAlAs both in the ManchesterCambridge group

Recently the ASPAT was commercialised by Linwave Technology as a wideband

zero-bias detector diode This was done in April 2016 Although it is now on the market

the ASPAT remains immature in term of research and development A lot of work is still

required to enhance the device ie working at the sub-millimetre wave using ternary

material etc

26 Asymmetrical Spacer-layer Tunnel (ASPAT) diode The Basics

The basic building block of the ASPAT diode is based on heterojunction of three

multilayer semiconductor structures which have two different band gaps The structure

comprises a thin layer of wider-gap semiconductor sandwiched between two

semiconductors with narrower-gap forming a tunnel barrier The basic principle of the

ASPAT device is based on the exploitation of quantum mechanics theory using

heterojunctions interface According to quantum mechanics theory moving particles

(electrons) with less energy than the barrier height have a probability of appearing on the

other side of the barrier by a tunneling through it This can be achieved in conditions

where the barrier must be very thin (~ 10 monolayer ) This is in contrast to classical

physics where a particle must have kinetic energy at least slightly greater than the

potential barrier height in order to overcome the barrier otherwise the probability of the

particle to appear on the side of the barrier is zero

Since the ASPAT diode operation is based on tunnelling through a barrier one

needs to know that the tunnelling mechanisms can be classified into two types[44]

intraband and interband The latter is described as tunnelling that occurs from

conduction band to valence band (electron) and valence band to conduction band (holes)

This normally happens in bipolar device ie p-n junction diode which has n-type and p-

type doped regions On the other hand intraband refers to tunnelling which occurs when

51

electron tunnel from the conduction band of a semiconductor to the conduction band of

its neighbouring semiconductor The same thing happens to the holes in the valence

band The device with this type of tunnelling is normally a unipolar device which is

either p-type or n-type doped The ASPAT diode can be considered as a device that is

based on intraband tunnelling mechanism Therefore the focus will be entirely based on

its principles

261 Principle of Quantum Tunneling

Generally all tunneling diodes obey the concept of quantum mechanical

tunneling Quantum mechanical tunneling is a phenomenon where a particle is able to go

through an energy barrier higher than the kinetic energy of the particle and if it is thin

enough compared to the de Broglie electron wavelength (λ) If the electron wave is

greater than the barrier the probability of the wave to occur at both side of the barrier is

higher

Figure 212 Classical view of whether an electron is can surmount a barrier or not Quantum

mechanical view allows an electron to tunnel through a barrier The probability (blue) is related to

the barrier thickness

For the case of classical physics (Figure 212(a)) the particles can be confined by

energy barriers of a semiconductor if their kinetic energy is less than the barrier energy

The particles thus require higher kinetic to escape to other states this phenomenon is

called thermal emission In quantum mechanics (Figure 212(b)) the particle is

described in two ways as a wave and as a particle If the particle moves like a wave it

will carry all the waversquos properties Therefore it will not brusquely end up at the

En

erg

y

(a) Classical view (b) Quantum mechanical view

En

erg

y

En

erg

y

52

boundary of the energy barrier Hence when the particles collide with the barrier

(incidence) there will be a probability of penetrating the barrier if the barrier is thin

enough and has a finite height For thicker barrier the probability of a wave that can be

found on the other side of the barrier is very small However the possibility of the

electron wave to appear on the other side of the barrier is increased by thinning the

barrier The potential barrier of semiconductor material technology can be determined by

using Homojunctions structures with different doping profile This will result in a

difference in band alignment and multilayer heterojunction structure (different

semiconductors have different band gap) which includes semiconductors insulators and

conductors (metals)

The easiest way to explain the phenomenon is by considering a potential barrier

Epot(x) with barrier height E0 energy bigger than the total energy E as shown in Figure

213 the potential energy occurs in a finite space between 0ltxlta and is 0 outside

Figure 213 Rectangular tunnelling with incident reflected and transmitted wave function[70]

The electron outside the region of the potential barrier (xlt0 and xgta) is free to

move The effective mass of the electron is different inside and outside of the barrier in

real tunneling devices when implemented using semiconductor heterostructures The

quantum mechanical equations predict the wave nature of matter which states that matter

unveils wavelike properties under some conditions and particle-like properties under

other conditions The wavelike properties as described by the Schrodinger Formula of

E0 Transmitted Ψ = 119862119890minus119894119896119909

Incident Ψ = 119860119890119894119896119909

Reflected Ψ = 119861119890minus119894119896119909 Ψ = 119863119890minus119894119896119909

0 a

E

x

V(x

)

53

quantum mechanics represent a particle penetrating through a potential barrier most

likely as an evanescent wave coupling of electromagnetic waves[44]

To start the calculation of the tunnelling probability the Schrodinger equation is given

by

119894ℏ

120597

120597119905Ψ(119903 119905) = ΨΗ(r t)

(217)

Where ℏ is Planckrsquos constant (662606957 times 10119890minus341198982119896119892119904

2120587frasl ) Ψ(119903 119905)is the wave

function at position r and time t Η is the Hamiltonian operator given by

Η = minus

ℏ2

2119898nabla2 + 119881(119903 119905)

(218)

Where 119881(119903 119905)is the potential energy which is dependent on space and time 119881(119903 119905) is

considered zero for a particle traveling in free space without any potentials The plane

wave with vector r and t is given by

Ψ(119903 119905) = 119890119894(119896119903minus120596119905) (219)

This equation satisfies Eq (217) above under the condition where the particle is

travelling in free space without potential k is the wave vector which is equivalent

to2120587120582frasl and the angular frequency 120596 is 2120587 multiplied by the frequency

In the case of tunneling through a potential barrier the method of separation of

variables is used to simplify the problem as in the equation below

Ψ(119903 119905) = 119877(119903)119879(119905) (220)

It is assumed that the problem above is divided into time-dependent and time-

independent parts 119877(119903) is the spatial component and 119879(119905) is the time-based component

of the wave function The time dependent problem as shown above in the Schroumldinger

Equation (1) can easily be solved by filling up all the finite parameters The solution of

54

the time-independent part gives the tunneling probability For the one-dimensional (1D)

time-independent Schroumldinger equation[44]

119864120595(119909) = minus

ℏ2

2119898

1198892

1198891199092120595(119909) + 119881(119909)120595(119909)

(221)

E is the total energy and 120595(119909)is the spatial component of the wave function along the x

axis The combination to the wave function is given by

120595(119903 119905) = 120595(119903)119890minus(

119894119864119905ℏ

)

(222)

The time-independent plane wave solution of 120595 = 119890119894119896119909 satisfies the equation (221) for

any constant potential V0 in space Plugging in the wave solution yields the condition

that

119896 = radic2119898lowast(119864 minus 1198810)

ℏ2

(223)

Referring to Figure 213 also the barrier with exact rectangular shape with height E0 and

width W the solution of the wave functions and tunneling probability can be extracted

by using the below equation [44]

119879119905 =

[119862119890minus119894119896119909]2

[119860119890119894119896119909]2= [1 +

11986402 sin ℎ2 (119896119882)

4119864(1198640 minus 119864)]minus1 asymp

16119864(1198640 minus 119864)

11986402 exp(minus2radic

2119898lowast(1198640 minus 119864)

ℎ2119882)

(224)

For more complex barrier shape Wentzel-Kramers-Brillouin has simplified the

Schrodinger equation for tunneling probability of carrier which becomes[44]

119879119905 =

[119862119890minus119894119896119909]2

[119860119890119894119896119909]2asymp 119890minus2int |119896(119886)|119889119886

1198860 asymp 119890

minus2int radic2119898lowast

ℎ2 [119880(119886)minus119864]1198891198861198860

(225)

55

This means that the incident electron has a finite probability T of tunneling

through the potential barrier and this leads to the concept of tunneling probability as well

as a tunnelling current Therefore this becomes the basis of tunneling phenomena and

thus all devices which are related to tunneling can be modelled and analysed based on

this basic example The tunneling phenomenon is a majority-carrier effect and the

tunneling time is set by the quantum transition probability per unit time (which is on the

order of picoseconds) rather than the transit time concept [44 71] This enables the

tunneling devices to work at a much higher switching speed They can also be used in

high-frequency applications such as microwave circuit and high-speed oscillators

262 ASPAT Structural Parameters of GaAsAlAs materials System

Figure 214 Epitaxial layer structure of device XMBE 304 (lateral structure) used in this study

Conduction band profile

56

The core materials that make up the ASPAT diode in this work is based on

heterostructure of group III-V compound semiconductors Such materials are chosen due

their mature excellent properties and their band gap which can be tailored to fit the

desired design as well as to improve the carrier mobility In this work the primary layers

that form an ASPAT diode are very thin pure Aluminium Arsenic (AlAs) of thickness

ten monolayers buried in between dissimilar thickness of pure Gallium Arsenic (GaAs)

layers as can be seen in the red circle in Figure 214 above These two GaAs layers are

known as spacer layer which normally have a ratio of 401 or 201 in thickness The

asymmetrical spacers layer and the thin barrier in such arrangement lead to an

asymmetric current-voltage characteristics as proposed firstly by Syme and Kelly[15]

To examine and investigate this GaAsAlAs ASPAT structure in term of electrical and

RF characteristics the device have been grown according to Table 22 below

Table 22 epitaxial layer of sample XMBE304 which is the main focus of this work

Material Doping (cmminus3) Thickness (Aring) Bandgap (eV)

Ohmic Layer GaAs (Si) 4times1018 3000 142

Emitter GaAs (Si) 1times1017 400 142

Spacer GaAs Undoped 50 142

Barrier AlAs Undoped 28 283

Spacer GaAs Undoped 2000 142

Collector GaAs (Si) 1times1017 400 142

Ohmic Layer GaAs (Si) 4times1018 4500 142

Substrate GaAs - 650 microm 142

The arrangement of the multi-layers that form a lattice matched GaAsAlAs ASPAT

diode can be transformed into band structure profile view for easy understanding The

conduction band profile at equilibrium is as sketched in Figure 214 As can be seen in

the Figure 214 the ASPAT diode is generally a heterojunction multilayer structure

tunnelling diode

57

Figure 215 The layer profile of ASPAT with 28nm barrier thickness [27]

The generic structure of ASPAT diode which is shown in Figure 215 above with a

schematic band structure comprises the following (starting from the top)

(1) A thick layer of heavily doped n++

about 4e+18cm-3

of GaAs with a thickness of

approximately 300nm

(2) An intermediate layer of lightly doped n-type about 1e+17cm-3

of GaAs with

thickness of approximately 40nm

(3) A spacer layer not intentionally doped (NID) GaAs with thickness of approximately

5nm

(4) An ultra-thin layer of NID AlAs with thickness of approximately 28nm

(5) A spacer layer of NID GaAs with thickness of approximately 200nm

(6) An intermediate layer of lightly doped n+ ~1e+17

GaAs with thickness of

approximately 40nm

(7) A thick layer of heavily doped n+ about 4e+18 of GaAs with a thickness of

approximately 750nm

Each layer has its own role For instance layer (1) and (7) are used as ohmic

contacts via connection to a AuGeNiAu metal stack This explains why they are

purposely heavily doped (gt 1018

cm-3

) for better low resistance ohmic contacts Two

intermediate layers (layers 2 and 6) are used to prevent the carrier in the contact layers

from diffusing into the undoped layers The two unequal length spacer layers with ratio

58

1198971 1198972 of about 401 are used as voltage arms to yield an asymmetric current-voltage

characteristic The asymmetry means that after a positive bias is applied from the long

spacer region an accumulation layer is formed and it is deeper than that formed by the

negative bias The thin layer positioned in the middle (Layer 4) is the tunneling barrier

The performance of a single barrier ASPAT diode can be optimised depending on

the applications by appropriate selection of the material system so that the band gap and

barrier height of such material can be modified Furthermore the mobility of electron

and doping concentration of the contacts region can be tuned The parameters to tune

during the growth for instant growth interrupt time and growth temperature will also

affect the performance of this diode The key layers that will affect the performance are

the barrier thickness and the two spacer layers enclosing it The study has shown that a

one monolayer change in thickness results in 300 change in in tunneling current for a

fixed voltage point[65]

The following discussions account for the effect of the main structure of the ASPAT

which is related to their performance

2621 Barrier Thickness and height

The probability of an electron tunnelling through a barrier depends exponentially on

the width and height of the barrier as well as the energy that is incident on the barrier[72

73] All these will affect the I-V characteristic of the ASPAT diode The tunnel current is

obtained by summing over all incidents electrons energies with tunnelling probabilities

through the barrier The tunnelling varies approximately as[74]

119879 prop 119890minus120581119889 (226)

Where d is the barrier thickness and κ is defined by the expression below

120581 =

radic2119898lowast(1198810 minus 119864)2

ℏfrasl

(227)

From textbook the tunnelling probability is given by

119879(119864) = 412058121198702

[(1205812 + 1198702)2119904119894119899ℎ2119870119897 + 412058121198702]frasl (228)

Where K is expressed as

59

119870 = radic2119898lowast119864

ℏfrasl (229)

Where ℏ represents the reduced Planckrsquos constant (h2π) E and m are the electron

energy and effective mass respectively Thus by inputting appropriate value into these

equations one finds that reducing the barrier thickness by one monolayer increases the

tunnelling probability by a factor of nearly three for every electron that tunnels through

the barrier As a result the current will also increase Further it indicates that the

tunnelling strongly depends upon barrier thickness and height By contrast the current

does not strongly depend on temperature

2622 Spacer Thickness

The reason for having two dissimilar undoped spacer lengths is mainly to avoid

diffusion of dopant to the barrier and subsequent layers during growth but in the case of

the ASPAT diode the spacer can also act as a voltage arm Varying the thick spacer

layer (1198971) results in the reverse current decreasing as the layer thickness increases and

varying the thin spacer layer (1198972) will affect the forward current which increases as the

thickness reduces To obtain appropriate asymmetrical I-V characteristics one needs to

maintain an adequate ratio between these two spacer thicknesses While a too thin

1198971 results in high leakage current at reverse bias a too thick 1198972 results in low forward

current One also needs to keep it thick enough to prevent carrier diffusion

These two spacers must be kept undoped or very low doped to allow the electron

moving in the electron mean free path region as it is clear from ionised donors Under

large forward bias an accumulation layer is formed between the spacer and barrier

segment and it is more noticeable compared to the accumulation layer that is formed if a

negative bias is applied

60

Figure 216 Conduction band diagram showing band bending and 2DEG formation at the L1

spacer

Consequently a triangular well is formed which creates an emitter 2D electron gas

(2DEG) population The electrons in this 2DEG occupy the quasi-bound states which

mean high excitation energy thus allowing the electron to tunnel through the barrier as

depicted in Figure 216

In term of RF performance it is important to highlight that a thicker spacer layer

will affect the depletion region which gets wider and thus will reduce the junction

capacitance of the device as per the following expression

119862 = 휀0휀119903

119860

119889

(230)

Where A is the area of the device d is the thickness of the main device structure which

consists of spacers barrier and well layers 휀0 is the permittivity of free space and the

relative permittivity of the spacer material is denoted by εr However the intrinsic delay

time will also increase and hence degrade the device high frequency performance

Therefore optimisation through spacer thickness requires trade-off between reducing

leakage current at reverse bias and degrading device junction capacitance

GaAs

GaAs

AlAs

2DEG

Γ

X

61

263 ASPAT Electrical Parameters

The classical approach in determining the current flow through an ASPAT diode is

by solving the Schrodinger and Poison equations Prior work had been done by Syme et

al in 1991 Due to the fact that AlAs barrier is very thin tunnelling is assumed to occur

at the Gamma valley ie AlAs bandgap= 283eV (rather than X valley)[18 75] and only

from accumulation layer (2DEG)[59] Here the DC characteristic of the ASPAT diode

can be calculated by solving Schroumldinger equation with the position vector represented

by z (in this case) Thus the equation is expressed as

minus

ћ2

2nabla

1

119898lowast(119911)nabla120569 + |119890|120593(119911)120569 = 119864120569(119911)

(231)

Supposing the current is uniform across x and y planes then this can be simplified to one

dimension Therefore the 1 D Schroumldinger equations becomes

minusћ2

2119898lowast

1198892

1198891199112120595 + |119890|(120595 minus ∆120595) = 119864119911120595

(232)

Where

120595 =

120569

exp (119894119896119911119911)

(233)

where ∆120595 is the correction term which reduces the effective barrier height The

Schroumldinger equation is solved using different values for Ez thus the quantum

mechanical current density in the z-direction is now expressed as[15]

119895119911 =

minus|119890|ћ

2119898lowast(120595lowast

119889120595

119889119911minus 120595

119889120595lowast

119889119911 )

(234)

It is different for a heavily doped contact which can describe as below the envelope

functions in the left and right contacts respectively can be described by plane wave

120595119897 = exp(119894119896119897119911) + 119877 exp(minus119894119896119897119911) (235)

120595119903 = 119879119890119909119901[119894119896119903(119911 minus 120580119873)] (236)

62

In this case the left contact covers the region zlt0 while the right contact covers the

regions zgt 120580119873 Equations (35 and 36) are then inserted into Eq (34) to form the

following expression

119895119911 =

|119890|ћ1198961

119898lowast(1 minus |119877|2) =

|119890|ћ1198961

119898lowast|119879|2

(237)

Where R (Ez) and T (Ez) are the complex reflection and transmission coefficients

respectively and they are solved by using the transfer matrix method This method has

been described in reference [56] The next step is to integrate the current in the z-

direction 119895119911 for all possible Ez values Thus the expression for the current density

becomes

119895119911 =|119890|119898lowast119896119861119879

2120587ћ2int(1 minus |119877|2)

infin

0

119897119899 |1 + exp (

119864119891 minus 119864119911

119896119861119879)

1 + exp (119864119891 minus 119881|119890| minus 119864119911

119896119861119879

|

(238)

The equations above are used to calculate the current density approximation from

the ASPAT main structure (two spacer layers and one barrier) only and based on

intraband tunnelling from the conduction band profile For real fabricated structure the

calculation must take into account both intrinsic and extrinsic elements of the diode

While the latter is mostly related to the pad and probe that is used to extract the I-V

characteristic the first element mostly comes from the epitaxial layer of the ASPAT

diode itself The ASPAT I-V characteristic is shown in Figure 217 which clearly

indicates nonlinear characteristics and thus can be used for detection applications

63

Figure 217 I-V characteristics of a fabricated ASPAT diode

2631 Intrinsic Elements of the ASPAT diode

In order to extract the intrinsic electrical characteristic of the ASPAT diode a

generic structure as shown in Figure 218 is essential Two main sources of contribution

to the electrical characteristics are the interfaces of each layer and properties of the

materials themselves

Figure 218 Intrinsic Elements of the Asymmetric Spacer Tunnel Layer (ASPAT) Diode

64

The electrical current flowing from the top contact to the bottom contact will go

through each epitaxial layer producing a close loop Each junction interface limits the

current flow and sum up the total resistance resulting in what is known as the diode

series resistance (Rs) In the ASPAT main structure there is a junction capacitance (Cj)

due to the undoped regions surrounded by heavily doped contacts thus acting as a

parallel plate capacitor The fully depleted capacitance (Cj) of the diode can be

expressed as in Eq (230)

2632 Series resistance of the ASPAT diode

The total series resistance (Rs) of an ASPAT diode can be calculated based on the

finished fabricated diode structure In general Rs depends on three contributors namely

the non-uniformities in the contact metallization the un-depleted epitaxial layer (total

thickness) on both side of the heterostructures and the resistance caused by the

spreading current from the Mesa into the much wider second contact layer ie doped

substrate or 2nd

ohmic layer[76] In fact the contribution toward building up the total Rs

solely depends on how the structure is designed In this work two types of structures

were deployed namely lateral structures and vertical structures These will be described

in the next subsection

For both types of structures the series resistance (Rs) of the ASPAT diode consists

of aspecific Ohmic contact resistance (ρcA) contact epitaxial layer resistance (Repi-

Layers) and spreading resistance (Rspr)[33 77] where Rspr is influenced by the type of

structure The specific contact resistance is obtained from Transmission Line

Measurements (TLM) of the sample The theory of the TLM will be discussed in detail

in the next section The expression for the specific contact resistance is

120588119888 = 119877119888119871119879119908sinh119889

119871119879frasl

cosh 119889119871119879

frasl

(239)

Here Rc is the contact resistance LT is the transfer length (effective length) w is the

contact pad width and d is the length of the contact pad

65

Repi-Layer is the sum of all doped layers that sandwich the main ASPAT device For

each doped layer the resistance is given by

119877119890119901119894minus119897119886119910119890119903 = 120588

119871

119860

(240)

Where ρ is the resistivity which is given by 120588 = 1

120583119899119902119873119863 L is the epitaxial layer thickness

in cm A is the device area (emitter size) micron denotes the mobility of the electron q is the

electron charge and ND is the donor concentration The spreading resistance depends on

the structure design of the device This will be elaborated in detail in the following

section

26321 Vertical structure (doped Substrate)

The XMBE307 structures were grown on n+ GaAs substrate to provide the

simplest fabrication process The cross section of the finished single diode can be seen in

Figure 219 below

Figure 219 Cross-sectional view of ASPAT vertical structure the dimensions of d b and h are not

drawn to scale

66

At low frequency and in a mesa that is etched into a doped substrate material the

spreading resistance can be approximated by[76]

119877119904119901119903 =120588119904

2119889

(241)

Where ρs is the substrate resistivity and d is the ASPAT diode mesa length

However the spreading resistance is increased at high enough frequencies as the skin

depth (δ) in the substrate is much lower than the effective mesa length (d) of the diode

A new spreading resistance is then calculated also based on the assumption that the skin

depth is much lower than the chip thickness (h) but much larger than the mesa length

Thus the spreading resistance at high frequency is given by

119877119904119901119903(119891) =

120588119904

120587120575[05 ln (

119887

119889) +

119887]

(242)

Where the skin depth (δ) is taken from standard planar formula and can be expressed as

120575 = [

2120588

(120583120596)]

12frasl

(243)

Where micro is the permeability and ω is the angular frequency During DC measurement

this type of structure requires having good suction on the stage for a good contact

However for small die (15mm times15mm) the suction sometimes is not strong enough to

provide a very good adhesion to the sample Therefore another type of ASPAT diode is

deployed which is based on the lateral structure by utilising semi-insulating substrate

and both contacts are connected to probes

67

26322 Lateral structure (Semi-insulating Substrate)

Figure 220 Cross-section view of the complete process ASPAT diode (XMBE304) The dimensions

are not drawn to scale

In order to obtain accurate measurement results so as to avoid contact errors to the

substrate between stage and Device Under Test (DUT) a lateral structure as shown in

Figure 220 above is deployed This type of design offers many advantages ie it

provides a path for on-wafer RF measurement However the proper design has to take

into account the increase in RS due to improper attention to the spreading resistance

This spreading resistance is different from the vertical structure that was discussed

above In this case it is mainly caused by a gap at the bottom contact The gap in the

horizontal direction between epitaxial layer and metal at bottom contact is denoted as D

gap in in Figure 220 Therefore Rspr for the lateral design can be expressed as [77]

119877119904119901119903 =

1

120587120590119889119866119886119860119904ln (

119886

119886119898119890119904119886)

(244)

Where σ is the conductivity between two coaxial half-cylindrical electrodes with inner

(amesa) and outer (a) rectangular length or bottom ohmic layer which is given by (σ =1

ρ) a d and amesa are the length and thickness indicated in Figure 220 above Noticeably

68

the D gap will have direct effect on the outer length (a) of the device which is also

proportional to Rspr For high-frequency operation where the skin depth is less than

d(GaAs) σ becomes

120590(120596) =

120590(0)

[1 + (120596120591119903119890119897)2]

(245)

Where τrel =microme and micro m as well as e are the mobility effective mass and electron

charge respectively It is recommended that for high-frequency applications the device

series resistance must be as low as possible

Hence for both type of structure the ASPAT series resistance is calculated based on

all the above-stated resistances and these are set by

119877119904 =120588119862

119860+ 119877119890119901119894minus119897119886119910119890119903119904 + 119877119904119901119903

(246)

The total RS can be decreased by increasing the emitter area of the diode However a

large device will not able to reach millimetre and submillimeter wave region (THz) as

the capacitance will also increase (Eq 230) Therefore both parameters will have a

trade-off between them to be able to work at ultra-high frequencies

27 Characterization of Ohmic Contacts

The semiconductor Ohmic contact can be characterised using techniques that will

be described in the following section First is the Cox-Strack technique which is

specially designed to characterise bulk type semiconductor (thick) contact resistance on

two opposite sides The detailed description of this technique can be found in [78]

Second is a technique called Four Point Probe This technique was developed in 1954 by

Valdes etel [79] to characterise semiconductor resistivity It can also be used to

characterise the contact resistance for planar type devices As this research does not

cover this method the details of the measurement can be referred to [80] Finally the

most common method which is also extensively used in this research is the standard

Transmission Line Measurement (TLM) The details of this method will be explained in

the next section There are simplified versions of the TLM method which require just

one lithography step but are nevertheless very powerful in characterising and optimising

the contact resistance known as Circular Transmission Line Measurement (CTLM)[81

69

82] However in this research this is not to be covered as the standard TLM is already

adequate for planar type devices

271 Transmission Line Measurement (TLM)

The formation of metal and semiconductor interfaces will create a contact that

becomes very important for the characterisations of any fabricated device Additionally

it enables the quality of certain process flow to be determined This interface must be

evaluated by a technique known as the Transmission Line Measurement (TLM) TLM

which was first introduced by Murrmann and Widmand [83] in 1969 underwent some

refinements by Berger [84] in 1971 The theory of the TLM can be described by

constructing a TLM structure which comprises a set of metals contact pads placed in

series on a highly doped semiconductor layer as depicted in Figure 221 The structure is

designed like a series of the islands to permit current flow in parallel to the contact pads

[80] which is a direction defined by etched patterns Each contact metal pad behaves

like a MESA which has a thickness (t) and width (W) The distance between each metal

contact pads is defined by d1 the gap between two neighbouring contact pads which are

beneath each contact pad is defined as effective length LT This will allow current flow

in and out of the subsequent neighbouring metal pad The resistance elements that will

be extracted are RA and RB which sit under the contact and in between two metal pads

These two elements represent the sheet resistance under the metal contact pad area and a

sheet resistance of the material between two metal pads

Figure 221 A simple TLM structure with effective length and sheet resistance underneath

t L

T

LT

RA R

A R

B

dn

Probe Probe

Metal Pad

GaAs

MESA

70

The basic relationship of resistance R with respect to the size of the metal contact

or in the standard transmission line can be expressed as [44]

119877 = 120588

119871

119860= 120588

119871

119905 times 119882=

120588

119905times

119871

119882= 119877119904ℎ

119871

119882

(247)

Where ρ is the materialrsquos resistivity L is the length t is the thickness W is the width

Rsh is the sheet resistance and A is the cross-sectional area of the transmission line The

unit for ρ and Rsh are Ωm and Ωm2 respectively

The total resistance RT of this structure can be taken from the sum of the two

neighbouring padrsquos resistance RA and RB In order to relate with Eq (247) above this

RT will be substituted into Eq(247) to become

119877119879 = 2119877119860 + 119877119861 = 2119877119904ℎ119860

119871119879

119882+ 119877119904ℎ119861

119889119899

119882

(248)

As suggested in [85] RshA and RshB are assumed to be identical Therefore Eq(248)

can be reorganised into specific contact resistance RC and semiconductor sheet

resistance Rsh above as 119877119862 = 119877119904ℎ119860119871119879

119882 the new equation is then

119877119879 = 2119877119862 + 119877119904ℎ119861

119889119899

119882

(249)

The common practice throughout this research is to design a TLM structure that

has a ladder structure consisting of 10 metal pads with each one measuring to a size of

100microm width and 50microm length and the space between the first and second metal pad

5microm The gap is increased after the second metal pad by a further 5microm until ten metal

pads are completed produce a separation between the ninth and tenth pad of 45microm The

TLM ladder structure as depicted in Figure 222 is supplied by a constant current of

1mA at the very left and right metal contacts by two probes This allows the extraction

of RC (Ωmm) and Rsh (Ω) instantly from such structure The potential difference

between the two adjacent metal pads is measured by another two probes and the reading

of a voltmeter is recorded The total resistance is obtained by using Ohm law where

voltage is divided by current (VI) Another voltage reading is taken for the next two

neighbouring metal pads until the largest gap is reached The readings of (conversion

71

VI) which result in the corresponding resistance RT are then plotted against spacing and

the result can be seen in the graph in Figure 223

Figure 222 Top view of TLM ladder structure use in this work

Additionally in the measurement the voltmeter used in work has a very high resistance

Otherwise there will be leakage of current occurring through the probes and cables

Therefore the parasitic resistance of the cable or connector and the probe contact can be

ignored The key parameters that can be extracted from the graph will be discussed in

the next paragraph

Figure 223 Typical plot of resistance versus TLM spacing

y = 02592x + 02375

0

2

4

6

8

10

12

14

0 10 20 30 40 50Res

ista

nce

Rn (

Oh

ms)

TLM Spacing dn (um)

LT

d1 d

2 d

3 d

n

I (1mA)

V

V

LT

ME

SA

Su

bst

rate

72

The straight-line graph plotted in Figure 223 can be referred to Eq (249) and

this must be done by assuming the sheet resistance Rsh of the material is constant If the

straight line of the graph is extended up to initial gap (d=0) the intercept on the y-axis

provides the 2RC value To extract the 2LT further extrapolation is made until the

interception at RT = 0 is reached Therefore an important parameter which is the specific

contact resistance ρC can be found from this expression[85 86]

120588119888 = 119877119862119871119879119882sinh

119897119871119879

cosh119897

119871119879

(250)

Where l represents the total conductive semiconductor thickness The final part that can

be extracted from the graph is the slope of the line This is obtained by dividing the sheet

resistance with the width of the metal pad ie represented by expression (119877119904ℎ

119882frasl )

28 Basic Characterization Techniques and procedures

281 Measuring tools and apparatuses

The success of every experiment is determined by the backend results that are

obtained from the measurements It is important to choose an appropriate instrument

which will provide the required data for a valid and detail analysis Thus this section

will give a brief explanation of the measuring instruments and methods that were

exploited in this research The measurement apparatus systems that are available and

have been utilised in completing this thesis are ldquoset of DCrdquo and ldquoSet of RFrdquo

measurements The DC set measurement consists of room temperature and variable

temperature system This system is built to perform process monitoring during device

fabrication and it comprises of five main components

The fundamental component in the ldquoDCrsquos set toolrdquo is the Agilent B1500A

Parameter Analyser [87] used to provide fixed currentbias during testing The other

component is a Karl Suss PM5 Cascade Prober [88] which is used to receive fixed

currentvoltage and to supply its to the semiconductor contact via probe tip The PM5

Prober has at least four probe arms and each of them is fitted with ldquoneedlerdquo called probe

73

tip The size of the tip that is normally used here is 2microm and in some cases the tip size

of 1microm is also utilised The currentbias supplied to the sample must go through the two

of Source Measurement Units (SMU) namely SMU1 and SMU2 to ensure no mismatch

occurs between parameter analyser and diode All the testing are controlled using a

software called Integrated Circuit Characterization and Analysis Program (IC-CAP

2009) brought from Keysight Technologies[89] The software is installed on a standard

Personal Computer (PC) and the PC is connected to a General Purpose Bus Interface

(GPIB) to link with the B1500A Parameter Analyser This system can be organised

based on the purpose of measurements ie IV characteristics TLM Schottky Diode

and transistor as it is very flexible to change the configuration For examples

Transmission Line Measurement (TLM) configuration requires the addition of a digital

multi-meter and four-point probe tip while diode measurement just needs two point

probe tip without a digital multi-meter Figure 224 illustrates the measurement system

for a set of DC to test the TLM structure

Figure 224 The set of DC measurement apparatus arrangement to measure the TLM

On the other hand the set of RF measurement consists of five block elements

configuration in the system namely the Vector Network Analyser (VNA) DC

sourcemonitor Cascade Microtech Prober SMU and Control PC via GPIB The RF set

Agilent B1500A

Parameter analyser

DUT

(TLM Diode

Capacitance)

Karl Suss

PM5 Cascade

Prober

PC

(MS windows 2000)

Digital Multi-meter

ICCAP 2014 Provide current

source

Measure and read the

voltage Stage and Probe

Signal

Current Source

SMU1 SMU2

General purpose bus interface

74

performs RF characterisation after the device fabrication is completed This system can

also perform DC characterisation as its basic instrument has this function too The VNA

machine used in this research is the Anritsu 37369A [90] which can perform the

Scattering Parameter (S-Parameter) measurement with a frequency range of 40MHz to

40GHz The DC sourceMonitor utilised in this experiment is the HP 4142B Modular

DC SourceMonitor [91] Both of these sub-systems are controlled by a standard PC

which exploits GPIB port to link them During operation the HP 4142B is connected to

the VNA by an internal bridge network and two SMUs to the Cascade Microtech Prober

which is then connected to the bond pads of the device It is identical to the set of DC

measurement where the SMU setting and data assembly are accomplished by IC-CAP

software package therefore the data that was obtained before and after completing the

fabrication can be compared This will enhance the validation of the data

However the stage and probes in both sets of measurement are different The

Cascade Microtech Prober has only two probe arms with each arm fitted with a 3-

fingers probe tip in the arrangement of Ground-Signal-Ground (GSG) as shown in

Figure 225 Each finger (pitch) is separated by 100microm thus to fit in with the pitch and

to reduce the mismatch in resistance the bond pad design must follow this separation

between each contact The GSG is configured by connecting the outer pads (Collector)

to the Ground probe tip while the inner pads (Emitter) are attached to the Signal probe

tip where the RF signal is sent and received through it Figure 226 shows the actual set-

up for RF measurement used in this research

75

Figure 225 The set of RF measurement for on-wafer DC and RF one-port measurement

Figure 226 Actual VNA system that was used for RF characterization

282 Measurement steps using a VNA

The RF measurement steps can be summarised in the following

- Step 1 Select or find the suitable VNA depending on applications

PC

(MS windows 2000)

ICCAP 2014

VNA

Anristru 32379A

40MHz to

40GHz

HP4142B

Modular DC

source

DC Bias Source

Cascade

Microtech

Probe Station

One 50microm pitch G-

S-G probe tips

Ground

SMU1

Cathode

Gen

eral

purp

ose

bus

inte

rfac

e

Lo

w P

ow

er H

igh P

ow

er

Cathode

Anode

SMU2

Ground

Ground

Signal

76

There are few factors that need to be well-thought-out before starting to use a VNA

especially for S-parameter measurements The factors are the availability of the VNA in

term of operational frequency and measurement port types air-filled metallic waveguide

or on-wafer Not all VNA can have an operational frequency for banded measurement

ie W-band Ku-band etc All these may require external signal sources to extend the

operational frequency

- Step 2 Properly setting up the VNA

The VNA can be set up depending on its application and the goal of measurement for an

instant number of point requires a desired frequency span IF bandwidth and the supplied

power level This very important to ensure the desired measurements are correct and

appropriate

- Step 3 Appropriate calibration system

In order to have a valid calibration appropriate calibration method has to be chosen

depending on the applications This will determine the accuracy and standards of

calibration As for on-wafer calibrations the de-embedding is normally used while for

the off-wafer the SOLR technique is more suitable to employ

- Step 4 Validation or verification of the calibration results

It is vital to validate the calibration results to ensure that the system has been properly

calibrated

- Step 5 Proper measurement

Proper alignment positioning and touching from probes tip to DUT is necessary to

guarantee a good repeatability and reproducibility of measurement results Normally

when positioning the probe an alignment marker is used as an aid By doing this similar

travel distance for the probes can be achieved The measuring plane will also be equally

well-defined

283 Measurement Practice and Flowchart

Essentially the device characterisation is performed in two stages ie during

fabrication as a process monitoring and after completion for data collection and

analysis Figure 227 shows the block diagram of a flow chart for testing a 15mm times

15mm wafer processing performance In the wafer processing after reaching the top and

77

bottom contactrsquos step the sample can be examined by measuring the TLM structure

according to the TLM procedure The measurement is conducted by exploiting a set of

DC measurement apparatus as mentioned above analogous to the TLM ladder structure

on the wafer surface Based on the TLM results the presence of any process issue during

the fabrication can be identified by examining the parameter such as contact resistance

and sheet resistance Thus a decision can be made either to proceed or to terminate the

fabrication should any issue is found early on As a result no materials will be wasted

further When a dielectric layer is involved Capacitance Dielectric measurement can be

tested This practise can be used to obtain the quality of the dielectric layer To connect

between a diode effective area and probe a bridge is requires It can be attached to the

diode emitter (to determine the diode size) to a bond pad for probe tip to touch This

bridge can either be left hanging in air or sticking to the isolated dielectric layer For the

GaAsAlAs material system the latter technique is preferable since it avoids issues with

the air-bridge which will be discussed in detail in Chapter 3 The opening area (via) for

metal connection can be checked by measuring the resistance on a special design pad

after a plasma dry etching step

Figure 227 Block diagram of the ASPAT measurement step

The next stage of characterising the device is the on-wafer diode measurement

which takes place after completion of all processing (including bond padsco-planar

pads) The work is carried out using the above set for RF measurement and employed

Ohmic Contact

Opening Via

TLM

Qualified

Device

Bad device

Dielectric

capacitance

Bonding pad DC and RF

Good

Good Good

Fail

Fail

Fail

78

purposely to access the DC and RF performance of each diode where the current-voltage

(I-V) characteristic and S-Parameter results are obtained In fact DC measurements are

first performed using a set of DC measurement and a rough IV characteristic can be

obtained to ensure the diode is working properly Usually the yield of any fabricated

device on 15mm times 15mm wafer in this research is between 70 to 90 In this research

the outcomes that will be discussed in the subsequent chapters are in term of the average

of measured values Thus it is very important to have a meaningful data to compare with

physical modelling and simulation in the future The diode IV characteristics are studied

by applying different DC bias at the emitter to collector terminals to extract its keys

parameters ie turn on voltage (supposedly zero bias) non-linear characteristic Rj RS

Cj etc

The device that has been measured by DC and having produced a valid result

will be marked for the next investigation ie the S-Parameter measurement This

measurement is executed using a similar system but with different probe types The

three fingers probe type is used and the device frequency response is measured via the

one-port network from the VNA The important parameters extracted from this

measurement are usually S11 (depending on how many ports are measured) Although

working devices are selected to measure accordingly there is a need to ensure the VNA

RF cable and probe tips are calibrated so that only valid data without errors will be

obtained A calibration technique called SHORT-OPEN-LOAD (SOL) is performed

prior to each daily measurement by exploiting a calibration sample with WinCal

software (Cascade Microtech)[92]

To avoid confusion it is worth mentioning that this technique used to calibrate

the device structure is different from what is used in SOL calibration to the equipment

Furthermore the de-embedding calibration is made on-wafer with the same substrate of

the actual device whereas the SOL is performed on a special calibration substrate

Normally the de-embedding results are not constantly automated with the VNA

equipment However the measurement is done separately starting with the special DUT

substrate then followed by the OPEN and SHORT de-embedding structures

79

3 RESULTS OF FABRICATION PROCESS OF ASPAT DIODES

31 Introduction

This chapter presents in detail the general fabrication techniques for a generic and

development work of micron scale ASPAT diodes The discussions will focus on the

semiconductor growth technique used in this work ie MBE and the fabrication process

steps which include sample cleaning photolithography etching and contact

metallization These techniques are ample to build and deliver commercially marketable

fabricated structure Hence all photolithography techniques used to complete this

project are based on conventional i-line optical lithography which is adaptable to

industry and commercial purposes

The fabrication process in this work can be fragmented into two major works

firstly the fabrication towards reproducibility repeatability and manufacturability in

term of device structure process flow and DC amp RF characteristics This work will

involve relatively larger emitter area which varies from 15times15microm2 to 100times100microm

2 The

larger area provides for ease and fast fabrication as well as DCamp RF measurements

Once repeatability and reproducibility of the process flow and performance is confirmed

the second part of this work which is concerned with applications in millimetre and sub-

millimetre-wave then took place In this work small emitter designs varying from

2times2microm2 up to 10times10microm

2 were considered with appropriate measurement pads The

successful fabrication of smaller diode geometries in the second part of the programme

naturally leads to a further investigations to understand its epitaxial layer structure and

extracting intrinsic components ie junction resistance junction capacitance and series

resistance which will determine the device performance in high-frequency applications

The I-V characteristic is obtained from DC measurement which is usually performed at

room temperature Its results are then compiled and compared with advanced simulation

It is worth mentioning that all samples that are investigated in this project are grown by

means of MBE and the activities related to the epitaxial layer growth using MBE in the

University of Manchester were done by the Materials Growth Team (Prof Missous) and

the authors has no responsibility for this particular task

80

32 Epitaxial Layer Growth Techniques

Before discussing the principle of the common lithography technique it is

important to discuss the wafer growth technique as it comes first before the fabrication

The growth technique that is extensively used in this study as well contributing a lot in

the electronic semiconductor industry is Molecular Beam Epitaxy (MBE) The following

section will discuss the basic operating principle of solid source MBE

321 Molecular Beam Epitaxy (MBE)

The MBE technique was developed in the early 1970s [27] and is purposely used for

growing high purity epitaxial layers of compound semiconductors Such sophisticated

growth technique provides significant functionality ie precise control of the thickness

(to one atomic layer) and contributes to the growth of various types of complex

semiconductor multilayers high quality and advanced materials This level of control is

vital for an assortment of heterostructures devices that are being utilized as part of the

development of the advanced electronics devices especially for the ASPAT diode which

require 01ML control over the AlAs barrier to attain acceptable variability in device

characteristics Additionally the accurate doping profile and excellent junction

abruptness also can be achieved by using this technique

Practically the MBE system used in this work is a solid source MBE which

utilises beams produced by heating up various sources The sources can be Si Al Ga

As and other group III-V compound semiconductors When the crucibles which contain

the sources are heated atoms or molecules of the various elements are evaporated and

travel in straight lines paths like beams directed toward a target (heated and rotating

substrate surface) The condition of the vacuum during evaporation is ultra-high vacuum

(UHV) ~10-11

torr in order to have high quality crystals The substrate is heated and

rotated to provide good growth uniformity over large areas ( up to 4times4rdquo wafers) [93]

The growth rate in typical MBE growth is ~1ML second and can be controlled

by the source temperature in the crucible The abruptness at the heterojunctions interface

and switching of the growth compositions can be obtained by precise control of shutters

that are placed in front of the crucibles Therefore an abrupt junction at GaAs and AlAs

81

interface can be formed to realize the barrier in the ASPAT diode In order to monitor

the quality of the growing crystal and measure the layer by layer growth mode

Reflection High Energy Electron diffraction (RHEED) technique is utilised This

technique works based on the diffraction of electrons from the crystal surface [93]

Additionally given that the ASPAT current density is very sensitive to the barrier

thickness a study has been made using different growth techniques namely MBE and

Metal-Organic Chemical Vapour Deposition (MOCVD)[59] From this investigation it

was concluded that the percentage local variability of current density produced by MBE

grown diodes is better than those grown by MOCVD Thus in this study to get benefit

from its performance all wafers are MBE grown

33 Basic Principles of Common Fabrication techniques

This section covers the generic fabrication process which underpins

reproducibility repeatability and process optimisations for high-frequency applications

331 Sample cleaning

Essentially semiconductor processing requires a ldquoclean environmentrdquo to produce

devices The clean environment is classified according to how many ldquounwantedrdquo

particles are present in a cubic meter There are four categories of clean room available

in the industry Class 10 Class 100 Class 1000 and Class 10000 [94]

The fabrication of all ASPAT diodes in this project was performed in a clean

room environment of Class 1000 equipped with laminar air flow and filter system to

give Class 100 or better during processing Although the sample was processed in a

highly controlled particle environment there is still a high chance for a sample to get

contaminated when handled by a human Besides this the source of particle which

contributes to the contamination can be from the apparatuses and processing equipment

used in the laboratory themselves Thus the process of cleaning the sample wafer

surface before the start of each step is vital

Generally in a clean room the standard solutions that are used to clean a sample are

N-Methyl Pyrrolidone (NMP) Acetone Propan-3-ol (Iso-Propane-ol) (IPA) and

82

deionized (DI) water The sample which is cut up into 15times15mm2 size tiles is cleaned

based on the following procedures

1 Hot NMP- The sample is dipped into the solution at 80˚C for 10 minutes This

solution acts as an organic type of nature pollutants removal

2 DI water- acts as NMP remover The sample is then washed by flowing DI water

throughout the samplersquos surface

3 Acetone- to ensure any remaining NMP is completely removed from the sample

The sample is dipped into Acetone for 5 minutes in a low power ultrasonic bath

at ambient temperature

4 IPA- is used to remove the Acetone from the samplersquos surface The sample is

then dipped into the IPA for 5 minutes in a low power ultrasonic bath at room

temperature The use of low power for the ultrasonic bath is to avoid the sample

cracking or breaking

5 Once done the sample is then blow-dried using nitrogen (N2) gun to remove any

moisture coming from the IPA Fortunately it is easy to remove the IPA

completely from the sample surface given its higher rate of evaporation

Once all these steps are accomplished a visual inspection using a high magnification

optical microscope is conducted to ensure the sample surface is clean The cleanliness of

the sample is determined by the (lack) of particles or other spots (liquid mark) that can

sometime be observed during the inspection Obviously the lower the number is the

better the sample is as it is impossible to totally remove dirt especially marks

Sometimes it is hard to remove the particles in one go There is always a need to repeat

each cleaning step for a few times However this will not affect the sample in term of

electrical performance as the cleaning solutions used are non-destructive

332 Photolithography

Lithography or sometimes called pattern transfer is the most important step in

realizing microelectronic devices The designed geometry and dimensions on a quartz

glass plate called a ldquophoto-maskrdquo must be done prior to the fabrication process While

the photo mask can be designed by using various software tools in this work the design

83

is done via the Advance Design System (ADS) by Keysight The details of the design

which includes three different mask designs will be covered in Section 34 The mask

consists of the desired patterns (master) that can be printed onto solid materialrsquos surface

by means of an electrochemically sensitive polymer (photoresist) using

photolithography In fact this type of optical lithography technique can be performed

with and without a mask due to its simplicity It is also easy and cheaper compared to

other techniques such as x-ray lithography or Electron-beam lithography (EBL)

Although the latter is expensive it is still worth to have since it provides a higher

resolution which is preferable when developing sub-micrometre technology

processing[95] The special feature about EBL is that it does not need a mask to pattern

samples but can be produced by the movement of an electron beam point source and

hence writing the patterns directly on the surface

The ultra-violet (UV) based light sources are more popular among researchers

and development workers because they are cheaper and have modest resolution Usually

photolithography operates at wavelengths (λ) from 193 nm to 436nm The source that

provides the UV light is a Mercury (Hg) arc lamp which uses narrowband filters to select

single emission lines First is the i-line at λ= 365nm then the h-line which is of lower

resolution and has λ of 405nm and thirdly the g-line with a λ= 436nm[96] The

conventional optical lithography used in this research uses the ldquoi-linerdquo at a wavelength of

365nm For shorter wavelengths than these excimer laser or krypton fluoride laser with

a λ= 248nm and argon-Fluoride with a λ = 193nm are also used in the industry The

higher power levels enable higher productivity (throughput) while narrower spectral

widths reduce chromatic aberration provide better resolution and larger depth-of-focus

In this research all the fabrication process are done by utilizing a conventional optical

lithography (i-line) using a Karl Suss MA4 mask aligner Before starting any UV

exposure it is important to check the UV light intensity as it will affect the resolution

and thus desired device dimensions Therefore every corner that is exposed to the UV

light is calibrated to be at 09mWatt power exposure

The complete set of photolithography components consists of photoresist

(photosensitive polymer) photo-mask (chromium) which is used to block the UV to

form a pattern mask-aligner and developer (chemical solutions) Standard fabrication

84

process usually practiced at the University of Manchester starts with sample cleaning as

mentioned earlier Then the sample needs to go through heat treatment to remove all the

moisture with a temperature set to be 150˚C and bake for 5-10 minutes After having

cooled down the samplersquos surface is coated with a thin layer of photoresist via a

technique called spin-coating using a Laurell CZ-650 series spinner The spinning speed

is set depending on the type of resist ie 3000rpm for negative photoresist and 5000rpm

for positive photoresist The rotating speed of the spinner does not have much effect on

the coated photoresist thickness but will have consequence on the uniformity distribution

over the sample surface The coated thickness photoresist however depends on the

concentration of the specific photoresist Once spin-coated is done another short heat

treatment (1 minute) is required to ensure the resist is hard enough to contact a

photomask and to remove any excess solvent The temperature is set on the hot plate to

about 115˚C and 110˚C for positive and negative photoresists respectively

The important segments contained in the photoresist are a polymer (base resin)

a sensitizer and a casting solvent [97] The polymer will react by changing its structure

when exposed to the radiation While sensitizers will govern the reaction of the

photochemical in the polymeric phase the casting solvent will permit the spin-coated

application on the wafer surface The photoresist consists of positive and negative

photoresist both of which were used in this research Their basic difference is with

respect to the area that is exposed by UV light ie whether it will remain on the

semiconductor surface or will be removed In the case of a positive resist the exposed

area will be removed by the developer and the covered area will remain In other words

whatever is displayed on the photomask goes onto the sample surface On the other

hand for the negative photoresist the area that is covered from UV exposure will be

dissolved by the developer Figure 31 shows a 3D picture of both processes used in this

research

85

Figure 31 3D illustration of Optical lithography process used in this research

To obtain the desired pattern on the surface after UV exposure for a certain

duration the sample is required to go through a development procedure by using a

developer The developer is used to expel the dissolvable part of the photoresist after

being exposed to UV The usefulness of the developer depends on the photoresist ie

MF319 developer is specifically for positive photoresist while MIF326 is suitable for

negative photoresist Both types of developers will not harm the devices as they are a

kind of metal free ion solution Thus no free ion will change the characteristic of the

device The common practice in the clean room at the University of Manchester is to use

positive photoresist namely Shipley Microposit S1800 supplied by The Dow Chemical

and negative resist AZnLOF2020 (AZ2microm) which is supplied by MicroChem For the

S1800 series the thickness of the photoresist is determined by the last two digits ie

13microm thick for S1813 and 05microm thick for S1805 Both positive photoresists are used

Resist

GaAs

Dielectric

Photo Mask

Negative Resist Positive Resist

After Etching

86

mostly as protective area during wet chemical etching and as sacrificial dielectric layer if

higher temperature is applied ie 190˚C On the contrary AZ2microm which normally has a

thickness of 2microm is useful for patterning small dimension which leaves small gaparea

for the metallization process to fill It has good aspect ratio and useful in single layer lift-

off (post metallization process) In fact this type of resist can be thinned by diluting into

an Edge Bead Removal (EBR) solution and smaller device feature size can be obtained

The final dimension of certain devices (mesa size) is governed by the exposure

time the distance between photomask and samplersquos surface and the development time

The appropriate UV exposure time is required to avoid over-exposure which will cause

the spreading of light into the purportedly dark-field area[98] However the effect of the

exposure time differs between the photoresist types smaller opening area than the mask

pattern for negative resist and larger opening area for positive resist[45] The gap

between the photomask and wafer surface must be reduced as much as possible to avoid

the UV light going through the unwanted area To obtain a good gap value an applied

pressure from stage to the mask is required The normal pressure used in this research is

between 04 to 1 Pascal depending on the type of photoresist (negative and positive) as

well as its thickness Lastly the development time also influences the dimension of the

device Appropriate development time is required because ie over-development will

cause the polymerized photoresist to etch laterally resulting in bad patterned geometries

On the other hand under-development will cause non-uniformity in the surface after wet

chemical etching (positive resist) as well as causing lift-off problem (negative resist)

333 Etching Process

The etching process is a process of removing unwanted semiconductor layers to

define device geometry and isolate each individual device in one sample There are two

types of etching technique used in this research ie wet chemical etching and plasma

dry etching The wet chemical etching is based on Orthophosphoric(H3PO4) solution

which is a selective etchant to materials like GaAs InGaAs and AlAs The selective

etchant is referred to a solution that can etch away a specific semiconductor with a

specific etch rate The etch rate depends on the mixture ratio and concentration of the

solutions ie the higher the concentration the higher the etching rate In practice the

87

temperature humidity and epitaxial layer doping level also have an impact on the etch

rate Hence to minimize variation in the etch rate both temperature and humidity in the

clean room are constantly monitored and regulated Ambient temperature between 18degC

to 19degC is usually suitable while humidity is kept within 30 to 40 The advantage of

the wet chemical etching is that it is inexpensive controllable and with high throughput

highly selective and simple The mixture solution that is used in this work is

Orthophosphoric (H3PO4) Hydrogen Peroxide (H2O2) and Di-ionised water (H2O) with

ratios of 3150 3110 and 212 The ratio of 3150 provides an etch rate of about

600Aminute and is good to define the area and opening the under-cut in air bridges for

InGaAs samples[99] The 3110 ratio results with highly anisotropic shape but is easy to

control as the etch rate is about 1500Aminute for GaAs material system However the

212 etchant solution will provide extremely anisotropic etch rate of about

1000Asecond and is quite difficult to control For GaAsAlAs ASPAT diode it can still

be controlled since the thickness that needs to be removed is about 7000A (refer to

section 342 for details on the fabrication of ASPAT) Table 31 summarizes the etch

rate with different ratios and different selective materials Common to all the ratios

mentioned above are the isotropic side walls with lateral and vertical etch rate of 11

Table 31 Chemical wet etch characteristics using Orthophosphoric-based and Ammonia on GaAs

and InGaAs materials

Material Etchant Ratio Etch Rate (Aringminute)

GaAs H3PO4H2O2H2O 3150 600

GaAs H3PO4H2O2H2O 3110 1500

GaAs H3PO4H2O2H2O 212 60000

GaAs NH4OHH2O2H2O 118 2000~3500

InGaAs H3PO4H2O2H2O 3150 850

The plasma dry etch is purposely run to obtain extremely high anisotropic etch

profile vertically and horizontally It can be done via mask with positive photoresist and

self-align mask which is metal contact as a mesa protector The precursors that are used

to etch away GaAs InGaAs and AlAs layer in this technique are Methane and

Hydrogen (CH4+H2) On the other hand Carbon Tetrafluoromethane (CF4) and H2 is

88

used in removing Si3N4 The etch rate is determined by how much power is applied to

the plasma to hit the sample surface the pressure inside the chamber and amount of the

precursor In this research plasma Technology is used for dry etching This machine

which is a conventional OXFORD INSTRUMENTS 1990 machine can produce an etch

rate on average of 100Aminute for an RF power of 100mWatt

334 Sputtering (dielectric deposition)

In this work a Kurt JLesker PVD 75 is used to deposit Si3N4 layer on the sample

surface The deposition rate depends on the RF power that is applied To avoid surface

damage on the sample surface a sacrificial layer formed by SiO is deposited using the

Bio-Rad Thermal evaporator before transferring to the PVD 75 to start deposition with a

low (75Watt) RF power Once the deposition time reaches 30 minutes the power is

increased up to 200Watt As a result good uniformity of the dielectric layer is obtained

335 Metallization Process Lift-off and Annealing

The metallization process is a process in which metal contacts on semiconductor

devices are created The purpose of this process is to make a proper interconnection

between the semiconductor devices to other parts of the circuit elements In other words

it is to connect the semiconductor device to the outside world This process will allow

the device to be examined electrically so that all electrical characteristics can be obtained

ie resistance I-V curve capacitance conductance etc The metal scheme used in this

process depends on the semiconductor material In the case of GaAsAlAs ASPAT

diode Gold-Germanium (AuGe) Nickel (Ni) and Gold (Au) are used However in the

case of InGaAsAlAs ASPAT diode the metal scheme used is Titanium (Ti) and Au

The technique used in this process is resistive thermal evaporation

The metallization process starts by cleaning all the metallic sources and boats

ie tungsten boat Au Ni and Ti metals by using Trichloroethylene Acetone and IPA

consecutively for 5 minutes each in a high power ultrasonic bath This step is very

important to reduce the risk of contamination during thermal evaporation Once done all

the metallic materials are dried using a high-pressure nitrogen gun and then dipped in

89

Hydrochloric acid (HCL) solution for 2 minutes to de-oxidize the metals so that it has

minimal effect on the series resistance of the ASPAT diode

Two types of thermal evaporators were used extensively in this study both

Edwards Auto 306 (one denoted as Junior Auto 306) The latter is used to deposit alloy

type of metals while the former is used for the non-alloyed type of metals In the case of

GaAsAlAs ASPAT alloy type metals are used while for InGaAsAlAs non-alloyed

metals are used The cleaned metals are then loaded in the thermal evaporators and

placed on a resistive tungsten boat Each metal is placed on its specific tungsten boat to

avoid unnecessary mixture during the evaporation

Figure 32 Actual picture of thermal evaporator used in this study

Prior to loading the sample into the evaporator a standard fabrication process for

ASPAT diodes takes place by patterning the samples with AZ2microm negative photoresist

At the end of this step an opening area is created for the metal contact to be filled and

connected to the ohmic contact of the semiconductor The sample is also deoxidised

using a mixture of HCL and water in the ratio of 11 prior to evaporation This has to be

done in a very short time in order to avoid re-development of the native oxide layer

Inside the chamber the sample is securely placed on a chuck upside down facing the

filled tungsten boat The distance between the sample and metallic source boat is about

90

40 cm Figure 32 illustrates the actual thermal evaporation system used in this study

The thermal evaporator chamber is pumped down to reach a minimum pressure below

1times10-5

mbar before vaporizing the metal It is important that the mean free path between

metal amp sample is created and each vaporized metal stick firmly on the samplersquos surface

The normal practice in this study is to keep the vacuum pressure under 1 times10-6

bar so

that better device performance can be obtained The amount of current required to melt

down the metal is between 4 Amps to 6 Amp This amount of current is forced through

the tungsten boat and generates very high heat melting down the metallic source and

vaporizing it towards the sample surface The deposition rate for each metal can be

monitored by using a built-in film thickness monitoring (FTM) on the thermal

evaporator which proportionally depends on the amount of materials deposited The

GaAsAlAs sample is started with deposit of 55nm AuGe 13nm Ni and 500nm of Au

The reason of depositing AuGe first is due to the fact that the ohmic layer of the

ASPAT only can only be doped with a maximum doping of 4 times1018

cm-3

which is not

high enough for good conductivity Thus here Ge atoms will diffuse into the GaAs and

replaces Ga atoms during annealing process leading to higher doping levels (gt1 times1019

cm-3

) and hence improved conductivity

3351 Lift-off

The use of AZ2microm allows for the exact patterning of metals without the need for etching

using a single layer lift-off technique The negative photoresist also provides an undercut

profile which will create disjointedness between the desired metal pattern (on the

semiconductor) and undesired metal (on photoresist) The process of getting rid of the

unwanted metal from a sample surface is called lift-off The process starts once the

evaporation process is accomplished The sample is soaked into 80˚C N-Methyl-2-

Pyrrolidone (NMP) solution for usually 20 minutes (fast lift-off process) In most

instances the sample is in NMP for more than 12 hours in ambient temperature (slow

lift-off process) as the NMP solution is not destructive to the sample In this solution

the negative resist will be softened and the metal part which sticks on it will also be

eliminated from the sample As depicted in Figure 33 the lift-off process for a single

device shows the usual photoresist undercut profile observed To ensure that NMP

91

residues on the sample surface are completely removed DI water is used to rinse the

sample which is then blown dry with a nitrogen gun

Figure 33 Single layer lift-off process using negative photoresist

3352 Annealing

The alloyed (AuGeNiAu) metal stack requires a thermal treatment called annealing to

improve the ohmic contact between metal and semiconductor The sample which has

deposited top and bottom contacts is loaded into an annealing furnace at a temperature of

420˚C for 2 minutes In the case of GaAs during thermal annealing Ge atoms penetrates

into the GaAs crystal for approximately 70nm-250nm depending on evaporated

thickness of the metal layers annealing temperature as well as time [100] In this work

the total metal thickness evaporated for each contact layer is around 500nm This

thermal annealing treatment will also melt down the Au if it is too thick and is subjected

to too long a heat treatment Therefore it is not advisable to do annealing after the

sample is coated with bond pad metals as it can result in short circuited devices

sometimes

Az2micro Negative

photoresist profile

after UV exposure

and development

Evaporation to

form metal layer

(AuGe Ni Au)

Desired metal

contact

Metallisation

process

Lift-off undesired metal

92

34 GaAsAlAs ASPAT Process Optimization

As mentioned earlier this section present details two major process flows of the

fabrication process for the ASPAT diodes which utilised three stages of development of

photomasks design namely a ldquoFirst generation mask designrdquo (1st Gen) a ldquoSecond

Generation mask (2nd

Gen)rdquo and a ldquoThird Generation mask (3rd

Gen)rdquo design The 1st

and 2nd

generations mask designs are the designs that were produced in the first stage of

this work to develop the fabrication process know-how and to get familiarized with the

actual fabrication techniques in the cleanroom The difference between the 1st Gen and

2nd

Gen masks is in term of the development towards realizing Air Bridges and

Dielectric Bridges which were mostly covered by the 2nd

Gen mask design The

analyses in term of reproducibility repeatability and manufacturability for process

control and current-voltage characteristic as well as ultimately RF measurement are

obtained on relatively large size devices via these two mask designs The large area

emitter dimensions range in size from 15times15 microm2 to 100times100microm

2

The 3rd

Gen mask design was designed based on the optimization of the 2nd

Gen

mask which was to realize ASPAT diodes that are able to work at very high operating

frequencies Therefore in such design the ASPAT devices have to have a minimal

amount of capacitance and low series resistance this is can be achieved by shrinking the

emitter size of the diode to the smallest area possible as well as optimising the

fabrication process ton reduce parasitics The smallest fabricated devices designed on the

new mask has an 2times2 microm2 MESA area which also includes Ground Signal Ground bond

pads for both device and de-embedding structures (open and Short) for RF

measurements

Before the commencement of any fabrications and designing any layouts on eg

Si GaAs InP etc the epitaxial layer must be grown first to a desired design In the

University of Manchester the epitaxial layer structures are grown using one of the two

Molecular Beam Epitaxy (MBE) machines which are either the RIBER V90H or the

V100HU system Both systems are managed by Professor Missous Epitaxial wafers or

sample grown by each system are identified by a prefix and numbers that are prefix

VMBE for the V90H system and XMBE for the V100 system The epitaxial layers are

93

grown on four-inch wafer diameter The maximum diameter that can be grown on using

the V100H is 8 inches but generally single 4rdquo or 4x4rdquo wafers are used The wafers are

then diced and cut using a diamond scriber into 15mm times 15mm tiles for easy handling

and fabrication in the D12 cleanroom lab in the University of Manchester

In this section the structures fabrication and performance of the various ASPAT

diodes for both repeatability and high frequency applications will be discussed further

The ASPAT is manufactured on wafer sample XMBE304 which is a GaAsAlAsGaAs

lattice matched to a GaAs semi insulating substrate is the main focus The ground works

on this ASPAT such as the initial design and fabrication process flow optimization had

been conducted by fellow PhD colleagues in the group led by Professor Missous at the

University of Manchester

341 ASPAT Devices used in Fabrication

3411 XMBE368 and XMBE307

Table 32 Epitaxial layer of Doped substrate samples

Thickness (Aring)

Layer Material Doping (cm-3

) XMBE368 XMBE307

Top Ohmic1 GaAs (Si) 4times1018

~3000 ~3000

Buffer 1 GaAs(Si) 4times1017

~50 ~50

Spacer1 GaAs NID 50 50

Barrier AlAs NID 28 28

Spacer 2 GaAs NID 1000 1000

Buffer GaAs(Si) 4times1017

50 50

Bottom Ohmic GaAs(Si) 5times1018

~3500 ~3500

Substrate GaAs (Si) Doped 50000 50000

These two samples were the first batch of diode structures used in this work and

were grown using the RIBER V100 MBE machine A great deal of work was expanded

to ensure that it is able to produce appropriate non-linear I-V characteristics The work

carried out including finding suitable fabrication process steps mask designs process

control limitations ie etching rates etc The results obtained from processing these

94

samples mostly on large area anode and cathode sizes and their analysis included both

growth profiles and fabrication process flows These samples were grown on doped

GaAs substrates As such the finished diodes were vertical structures and the fabrication

process has marked differences compared to undoped substrate samples While

XMBE368 had similar epitaxial layer profile to XMBE307 during growth the AlAs

layer was set to be stagnant (ie no-rotation of the substrate during growth) to

investigate the effects of slight variations in barrier thickness

3412 XMBE304

The next sets of samples were all grown on semi-insulating GaAs substrates

Table 33 details the epitaxial layer profile of sample XMBE304 the main work horse

of this research work The growth of this structure was performed on a multi-wafer

platen and consisted of 9 x 2rdquo wafers (from XMBE304A to XMBE304I)

Table 33 The epitaxial layers of sample XMBE304 with barrier 10ML~28nm

Epitaxial layer Material Doping(cm-3

) Thickness(Aring)

Emitter GaAs(Si) 4e+18

3000

Emitter 2 GaAs(Si) 1e+17

400

Spacer GaAs NID 50

Barrier AlAs NID 28

Spacer 2 GaAs NID 2000

Collector GaAs(Si) 1e+17

400

Collector 2 GaAs(Si) 4e+18

4500

Substrate GaAs(SI) Semi-Insulating

For a typical ASPAT structure the emitter is essentially highly doped to

4times10+18

cm-3

to provide accumulation of electron in the emitter contact region It was

purposely highly doped to also achieve low ohmic contact with the metal The spacers

are used to avoid diffusion of dopants to the subsequent layers The ASPAT structure as

mentioned earlier has a single AlAs barrier with a very small thickness sandwiched

between two different length GaAs spacer layers

Batch XMBE304 is the main focus in these studies All activities required for

repeatability reproducibility process flow and devices as well as qualifying new or

95

optimization fabrication technique for small devices and RF performance which were

then used for high frequency were based on the set of wafers grow in this batch

342 Fabrication process of GaAsAlAs ASPAT diode toward repeatability

reproducibility and manufacturability

As the new wafer structures have to be tested and evaluated to gauge the

performance of the ASPAT diodes their uniformity also needs to be tested so as to

ensure it exhibited fully functional diode with zero bias detection in minimal variation in

IV characteristics between diodes As described previously 4rdquo wafers are always diced

up into 15mm times15 mm size for ease of handling in the cleanroom and for masks cost

purposes

3421 Doped n+ Substrate Wafers

For wafers grown on n+ substrates (XMBE368 and XMBE307) a two-level mask

set is sufficient to complete the fabrication process In this case the devices are designed

with mesa structures and top metal contact on the upper surface of the wafer and a

bottom contact on the backside of the wafer One mask plate is used for defining the

mesa structure (wet etch) and the other mask plate is used for defining the metal pad

Since the mesa is relatively large (30times30microm2

to 100times100microm2) there is no requirement to

define bond pads for measurement purposes A prior RTD mask designed by fellow

colleague Dr Md Adzhar for his PhD work was used for processing the ASPAT diodes

as well in the first instance The detail fabrication process is summarized in appendix I

for the doped substrate wafers Figure 34 shows typical IV characteristics for 30x30microm2

emitter size diodes obtained from sample XMBE368

96

The I-V measurement was taken from two different tiles (15mm times15 mm) located

on top and bottom of a 4rdquo XMBE368 wafer The location 1 marked as a blue line in

figure 34 refers to a device on a tile taken from the top of the 4rdquo wafer while location

2 (red) refers to a device on a tile taken from the bottom of 4rdquo wafer At 07V the

separation difference in current is ~228 for both tiles This shows that the device in

tile 1 is less resistive than the sample in tile 2 implying that the AlAs layer for tile

located on the top corner of the wafer is thinner and thus allows more electrons to tunnel

through it at a given bias

3422 First Generation Mask Design (1st Gen)

-0015

0005

0025

0045

0065

0085

-15 -1 -05 0 05 1 15

I C

urr

en

t (A

mp

)

V Voltage (Volt)

Current vs Voltage (XMBE368_1) for 30x30microm2

MidLeft2

MidRight3

Figure 34 Current-Voltage characteristic of sample XMBE368 used

in this study at two different locations on the wafer tile

(a) (b) (c)

Figure 35 ASPAT mask design for (a) 100times100microm2 (b) 30times30microm

2 and (c) 15times15 microm

2 diode dimensions

designed in the 1st Gen Mask

Location 2

Location 1

97

The fabrication of the GaAsAlAs ASPAT conducted in this 1st Gen mask is

followed the established generic process flow for InGaAs material which was done by

former co-worker Dr Md Adzhar Md Zawawi[101] The generic process flow is fairly

simple consisting of four mask steps as shown in Table 34 below The fabrication

process starts with the sample being cleaned in a NMP solution at a temperature of 80˚C

with Acetone and Propan-2-ol (IPA) The purpose of cleaning using NMP and Acetone

is to remove any organic material The IPA is used to remove Acetone residues

Table 34 Generic fabrication steps established by Dr Md Adzhar [101]

Step number Process

1 Top Contact

2 Mesa Etch

3 Isolation

4 Bottom Contact

In the 1st Gen mask the first step in fabricating the GaAsAlAs ASPAT diodes is

to use the first mask to define the emitter contact area The emitter has three different

sizes 100times100microm2 30times30microm

2 and 15times15microm

2 The lithography technique uses the

negative photo-resists AZ2micron with 55 second UV-photolithography to pattern the

top contact Then it is developed using MC319 developer to clear and define the

exposed area for the metals to stick to The sample is then subjected to plasma etching to

remove all organic residues and contaminants Then it is dipped into a mixture of

diluted Hydrochloric acid and water HCL H2O with a concentration of 11 for de-

oxidation This must be done in a short time right before the evaporation in order to

ensure good contact between metal and semiconductor surface with very low oxide

formation in between The ASPAT device performance depends on this step as contact

to the channel is by means of current flowing through the anode to the cathode terminals

In our lab the approach taken to achieve the Ohmic contact is by evaporation of Gold

Germanium (AuGe) Nickel (Ni) and Gold (Au) metals layers on top of the cap layer

Subsequently the metal is defined via a lift-off process using NMP

98

The next critical step is to define the MESA or island The mesa etch mask is

designed with two options with 05 microm or 10 microm tolerance The different mesa

tolerances are introduced to act as a safeguard for the emitter from producing excessive

undercut caused by lateral etching This step is to isolate and eliminate the unwanted

GaAs which will electrically link the active layers as many devices will be fabricated at

the same time on the same wafer tile (15mm times 15mm size) The lithography process in

this step uses positive resist and is developed using MC326 developer This step is

achieved using a wet etch process where a non-selective etchants mixture of

H3PO4H2O2H2O etches down the epitaxial layers until it slightly exceeds the AlAs

barrier with an etch rate of about 600Aring to 900Aring per minute The outcome of the MESA

or island step is the formation MESA active layers which are surrounded by inactive

layers of semi-insulating substrate (when using semi-insulating substrates)

The isolation mask is a step that is basically the same as the MESA etch step

eg using similar etchants mixture same lithography process but a different mask This

step is used to etch down until the GaAs substrate is reached which means that the

etching time is longer than that in the MESA etching step The purpose of this step is to

fully isolate the device from other neighboring devices hence ensuring no electrical

connection exists between each device within one sample Since these ASPAT diodes

employ an air bridge structure of size 1times5 microm the two minutes mesa etches will

simultaneously provide an initial undercut through lateral etching for the air bridge

formation

Based on the results so far obtained using the first-generation mask which

provided large mesa areas the current voltage characteristics of the ASPAT as far as the

non-linear zero bias is concerned did work very well However there was a need to

reduce the ASPAT mesa area down to very small dimensions to achieve mm-wave or

THz detection frequencies It is certainly a general rule for semiconductor device which

operate at very high frequency to have extremely small lateral dimension which

minimises the capacitance within the ASPAT device Furthermore for wider adoption of

the technology it is also important to develop a simple reproducible and low-cost

fabrication method for ASPAT diodes Details of this fabrication process are attached in

appendix II

99

3423 Second Generation (2nd

Gen) mask (ASPAT-GSG)

This work intended to enhance the 1st Gen-Large Area ASPAT photomask by

adding many features including 2 times 2microm2 mesa areas ground-signal-ground (GSG) bond

pads to enable RF measurement and three options device processing eg Air Bridge

Dielectric Bridge for semi-insulating doped substrate and dry etch-mesa Other reasons

for designing this mask was also to qualify process steps when deploying thin (1microm

width) bridge to connect small mesa area ASPAT diodes to the co-planar GSG bond pad

for DC and ultimately RF measurements

The mask was designed to fulfil the basic rules of fabricating various types of

tunnelling diode for instance RTD and ASPAT The diodes layouts were designed using

the commercial software Advanced Design Software (ADS) from Keysight

Technologies Ltd Once the design was completed it was ready to be sent to Compu-

Graphics Company for printing and patterned on a special chrome coated glass plate

The new 2nd

Gen mask design was termed ASPAT with Ground-Signal-Ground

(ASPAT-GSG) and consisted of two main designs ldquoAir Bridge and Dielectric Bridgerdquo

where each contained eight diodes with different emitter sizes (100times100μm2 50times50μm

2

30times30μm2 20times20μm

2 15times15μm

2 10times10μm

2 6times6μm

2 and 2times2μm

2) The Air Bridge

design is comprised of Design 1 (doped substrate) and Design 2 (undoped substrate)

This can be selected by changing the order of each individual layer of the mask steps

The same options are applied for Dielectric Bridge design which included processing for

doped and undoped substrates The details of the masks will be explained in the

following paragraph

(1) Air Bridge

Design 1 Contains five steps or layers mask of size 15mm times 15mm

suitable for air-bridge for undoped semi-insulating substrate

Design 2 Consist of seven layers steps mask of 15mm times 15mm size

There are 413 die chips in about 6mm times 6mm sizes in this design Figure 36 shows both

type of design for Air Bridge mask processing

(i) Mask 1- Top Contact

(ii) Mask 2- MESA

100

(iii) Mask 3-Isolation

(iv) Mask 4- Bottom Contact

(v) Mask 5- Collector Bond pad with GSG

(vi) Mask 2A-Dielectric

(vii) Mask 6A-Collector Bond Pad with GSG

(2) Dielectric Bridge

Design 1 Consists of seven identical steplayers masks with lateral

lengths of 15mm times 15mm each suitable for fabrication on semi-insulating

substrate

Design 2 Also has seven steps layers masked with dimension of 15mm

times15mm each suitable for doped substrate processing

There are 357 die chips in this type of mask design with an estimated size of 6mm times

6mm separately The smallest emitter size that is connected with a dielectric as the

bridge is 6times6microm2 The smallest 2times2microm

2 diodes was designed with the air-bridge

connected to the emitter bond pad due to the difficulties of opening viascavity for less

than 2 microm2 devices Figure 37 show both options for Dielectric Bridge mask processing

(i) Mask 1- Top Contact

(ii)Mask 2- MESA

(iii)Mask 3-Isolation

(iv) Mask 4- Bottom Contact

(v) Mask 5-GSG bond pad

(vi) Mask 6- Via Dielectric

(vii) Mask 7- Dielectric Bridge

(vii)Mask 5A- Via Dielectric

(ix) Mask 6A-GSG bond Pad

(x) Dielectric Bridge

Figure 36 Air Bridge design (red circle) for large mesa area (100times100microm2) device with

option for doped substrates

Figure 37 The layout of 1st design of Dielectric Bridge (green circle) mask design for 100 times

100microm2 emitter size with option for doped substrate processing

101

34231 Fabrication Process of the Air-Bridge Design

The fabrication using 1st Gen mask as mentioned in Section 3422 is less

complex however the fabrication process for Air-Bridge mask design contains a few

additional steps which are to add bond pads for the Ground-Signal-Ground radio

frequency (RF) layouts If the sample is on a doped substrate another layer needs to be

added leading to a total of six steps all together

Table 35 Standard process flow for Air-Bridge design fabrication

Mask Number Step number Process

1 1 Top Contact

22A 2 Mesa EtchDielectric

3 3 Isolation

4

5

4

5

Bottom Contact

Bond Pad

In this work the samples that have been processed so far are XMBE304

XMBE314 (GaAsAlAs base material system) and XMBE326 (InGaAsAlAs base

material system) All of them are built up on semi-insulating substrate thus the normal

process flow was followed Step 1 to step 4 follow exactly the same route as the 1st Gen-

ASPAT Large Area mask formerly discussed in Section 3422 This process is then

continued by spin coating AZ2microm on the surface Mask 5 is used to pattern the bond

pads The metallisation scheme used for bond pads is TiAu and the thickness must be at

least 1 micrometre thick This is to minimise series resistance at the pads and ensure a

robust surface is created when used for DC and RF measurements

342311 Air-Bridge Process Optimization

Since the Air Bridge design approach is focused more on developing air-bridges

which have a width of 1 microm and as the smallest device is 2 times2 microm2 the negative resist

(Az2microm) which was used in the 1st Gen design type was changed to Az1microm After spin

coating an Edge Bead Remover (EBR) is required to remove the beads at the edge of the

sample this is to ensure no gap is created between samples and mask when ldquohard-

102

contactrdquo is applied during the photolithography step The use of Az1microm and EBR are

critical to enable the fabrication of small air-bridges and emitter sizes

The exposure time during the photolithography technique also needs to be

increased to achieve a 1microm size air bridge A longer time than normal is required thus

95 seconds is used for exposure under UV light Once finished appropriate

development time must be applied to ensure the line for the air bridge is perfectly

opened for the metal to fill up To get impeccable result in this step both combination of

exposure and development have to be tuned naturally leading to trade off in both Too

long exposure and developing time will break the bridge whereas for too short exposure

time and developing the air bridge will not open

There are two types of etching method used in this fabrication wet

(H3PO4H2O2H2O) and dry (NH4H2) etching Both have their own advantages and

disadvantages Wet etching is faster than dry etching However the etching profile is

highly isotropic and causes serious undercut The smaller devices which have

dimensional sizes of 6times6microm2 and below will ldquoshrinkrdquo the effective area under the top

contact metal Even though dry etching can have highly anisotropic etching profile

which can prevent excessive undercuts it requires a lot of time for etching the

semiconductor layers The NH4 and H2 plasma that are used in this technique do not only

etch the GaAs and AlAs layers but also etch the metal contact at the same rate as the

semiconductors Thus the metal area must be covered with a photoresist or thicker metal

(~1microm) must be used to counter this issue Figure 38 shows that the emitter bond pad

which is not covered by the resist will eventually be etched away by the dry etching

Figure 38 Dry Etching for the first run in this study

As mentioned in section 333 the advantage of H3PO4H2O2H2O wet etching is

that it can be made using two solutions fast and slow The fast solution is based on a

concentration ratio of 212 while the slow solution is based on a 3150 mixture The

103

212 fast solution will produce good anisotropic etch profile but is tricky to control due

to the rather high etch rate of 1000Aring per second Both fast and slow wet etching

solutions being carried out in this experiment are replicated from successful recipes that

were developed by other co-worker from Prof Missousrsquos group

342312 Issue of Over Etch under Top Metal (Wet Etching for Air-Bridge

Design)

The first run using the solutions mentioned above started with the deposited

metal as a top contact (Top contact must be defined first to make a bridge) then Mesa

etching to remove ~7000Aring GaAs using the fast etch solution The next process to be

followed was to employ the slow etch solution to etch down the epitaxial layer to the

substrate Doing this ensured that air-bridge was open and the devices isolated

individually Unfortunately unexpected results were observed where large undercuts

still appeared for both 2times2 microm2 and 6times6microm

2 devices Figure 39 shows severe undercut

under the big device that can clearly be seen in the digital scope Most probably this

problem occurs due to excessive time used for the wet etching (10 seconds for fast

solution and 10 minutes for the slow etch solution)

Figure 39 Severe undercut of 2times2 microm2 and 6times6 microm

2 devices

The second run was employed to reduce the wet etching time in total thus dry

etching was needed As discussed above CH4 and H2 are used to etch away the GaAs

and AlAs semiconductor material while O2 is used to remove polymer residues To

reduce redundant generated polymer during dry etching the process must be separated

into several goes (runs) In this run at the mesa etch step the metal area is covered by

S1805 resist and then the plasma etches down until the heavily doped Ohmic layer is

reached For the isolation step the slow wet etching solution is used to ensure the

semiconductor under the bridge is removed The etching time is still long ie over 5

Zoomed Zoomed

104

minutes Hence this will still generate appreciable undercuts Although this run

managed to reduce the wet etching time it still did not solve the undercut Figure 310

shows undercuts under the effective mesa area still occurring in this run

Figure 310 Device with emitter area of 2times2 microm2 and 6times6 microm

2 after dry etch and wet etched

Further investigation has been made to find the root cause of this issue Scanning

Electron Microscopy (SEM) picture were taken to see deep into the completed devices to

investigate what is actually happening Figure 311 shows clearly that the semiconductor

under the bridge and effective area under the 2times2 microm2 device went missing The same

goes to for the large device where about half the size of the designed effective area under

to metal was also unintentionally removed by the solution

Figure 311 SEM Images of the GaAs sample

Since this work was also run together with an InGaAs based sample

(XMBE326) new knowledge regarding etching profile was acquired The lesson learnt

from this run is that the wet etch profile of the Gallium Arsenic (GaAs) material is

totally different from that of InGaAs Figure 312 shows the cross section of GaAs and

InGaAs materials after wet etching The evidences of these issues are also indicated by

the SEM images in Figure 313 and Figure 314 This more or less explained the reason

why the semiconductor under the top metal contact is always missing when wet

chemical etching is applied

Zoomed Zoomed

105

Figure 312 The cross-sectional view of InGaAs (left) and GaAs (right) samples used in this study

Figure 313 SEM images taken from the XMBE304 samples and a GaAs test sample

Figure 314 SEM images for InP and InGaAs taken from [56]

Furthermore previous work done by co-worker Dr Md Adzhar Md Zawawi has

optimised the process for RTD samples based on InGaAs and InAlAs heterojuction

semiconductor materials In this work Dr Zawawi found out that achieving submicron

dimensions is possible when using the soft reflow technique on InGaAs[102] However

it does not apply to GaAs when using the same technique There are few reasons to

106

believe that the work that was carried out by the previous student is not repeatable to the

XMBE304 GaAsAlAs material The main reason is due to the much thicker Ohmic

layers used in the GaAs of ASPAT structures which requires longer etch time The

sample that was used in the previous submicron work namely XMBE277 (RTD) [101]

and corresponding epitaxial layer structure is attached in appendix III The thickness

needed to be removed for the MESA step is 1421Aring compared to XMBE304 (ASPAT)

which is ~6900Aring

The fabrication results shown from the Air-Bridge design does not seem

favourable to the GaAsAlAs heterojunction sample These include unreproducible IV

characteristics and excessive undercut under the metal Therefore the GaAsAlAs

ASPAT sample cannot be processed using this type of mask (Air-Bridge approach) The

fabrication efforts were then moved to the Dielectric Bridge design described below

342313 DC measurement

Fully functional ASPAT I-V characteristics were not obtainable in this run due to

severe damages caused by the undercuts that happened underneath the top metal contact

As can be seen from Figure 315 the behaviour of the current response suggests a very

leaky diode This confirms that the air bridge approach does not work with the

GaAsAlAs ASPAT diode

Figure 315 Short circuit behaviour on one of the fabricated device in this run

-0015

-001

-0005

0

0005

001

0015

002

0025

003

0035

-3 -2 -1 0 1 2 3

Cu

rren

t (A

mp

)

Voltage (Volt)

30x30BP

107

34232 Fabrication Process of the Dielectric-Bridge Design

Since the Air-Bridge design was not successful an alternative Dielectric-bridge

process flow was then designed to solve this issue The first run in this Dielectric-bridge

process was performed to ensure that when the fabrication was completed it was able to

produce the correct and reproducible ASPAT current-voltage characteristics The

process flow in this run is to follow the initial design which identifies the steps according

to the mask number Step 1 to step 4 in the process are exactly the same as in the 1st Gen

mask design Step 5 which is the bond pad process is then continued with spin coating

the AZ2microm After exposure and development a TiAu metal scheme with a minimum

thickness of approximately 1 microm is deposited

Table 36 Standard fabrication process flow for Dielectric-Bridge design

Mask Number Step number Process

1 1 Top Contact

22A 2 Mesa Etch

3 3 Isolation

4

5

6

7

4

5

6

7

Bottom Contact

Bond Pad

Via Etch

Bridge

Step 6 is quite complicated compared to the other steps where the introduction of

S1805 resist is used as a dielectric layer to prevent short circuit between top and bottom

contacts This includes opening the smallest via (holes within the dielectric layer) of 2times2

microm2 and 6times6 microm

2 emitter sizes This step started by spin-coating the S1805 resist and

was then followed by baking it at 150˚C for 30 minutes The longer baking time is to

ensure it is hard enough to be deployed as a dielectric layer Once the S1805 had

hardened as dielectric layer the sample was then spin-coated again with S1813 (thicker

resist) on the dielectric layer to act as a mask for vias opening (to cover the dielectric

layer from via etching damages) After the S813 was developed O2 plasma etches was

108

applied to remove the S1805 that covered the top metal and bond pads The hole on the

metal was then exposed This required only two minutersquos plasma etches time

As soon as the holes were created within the dielectric layer a quick clean to

remove S1813 was performed using Acetone and IPA This has to be fast enough to

remove the S1813 without removing the S1805 dielectric layer Mask 7 then defined the

area for the metal that will fill up into the open vias This metal connected the top

contact metal of the device to the bond pad (GSG pad) The same metal scheme as

mentioned in Section 3422 was used as the bond pad

The fabrication using the dielectric bridge approach in the 2nd

Gen mask design

appeared to work successfully when initial I-V characteristics were taken and it also

showed that the ASPAT diode was fully functional However using hardened S1805 as

a dielectric layer is not a good practise for manufacturing device since there were left

over residue on the sample surface as depicted in Figure 316 This residue comes from

the non-uniformity of S1805 resit that was formed during heat treatment Therefore the

next run was to avoid using S1805 but replacing it with a standard dielectric layer based

on Silicon nitrite (Si3N4)

Figure 316 The surface of the sample after final processing

342321 DC measurements

Given that the process of qualifying mask steps for GaAsAlAs ASPAT diodes

looked promising using the dielectric bridge approaches initial I-V characteristics

measurements were carried out for XMBE304A Figure 317 shows that the I-V

characteristics are comparable to those of the other runs ie in the 1st Gen mask

Noticeably the IV characteristic (Figure 317) between each device size does not scale

109

on a single line This is attributed to the unintentional variation in the size of the emitter

during wet etching within each sample

Figure 317 IV characteristics of sample XMBE304 for emitter sizes of 1000 microm2 2500 microm2 900

microm2 400 microm2 225 microm2 100 microm2 and 36 microm2

The current densities shown in Figure 317 are calculated based on a reduced

effective area of the devices by a factor of 08 This assumption is made as actual area

(without metal) is un-measurable Theoretically devices with small areas will have

higher resistances compared to larger devices However the opposite occurred in these

devices This problem is still under investigation It could be due to the spreading

resistance producing different values according to device sizes (smaller devices have

longer D while bigger devices have shorter D)

34233 Dielectric-Bridge Process Optimization (Si3N4)

Due to poor surface roughness of the sample when using S1805 as dielectric

layer this run was employed to improve the surface quality by using Si3N4 In order to

get the actual size of the emitter the processing needed to start by defining the

semiconductor area using either top contact mask or mesa etch mask and not to deposit

the metal first Smaller effective area will be obtained if the top contact mask is used to

define the area compared to the MESA mask since there are 05 microm and 1 microm tolerances

designed in the MESA mask

-000001

0

000001

000002

000003

000004

000005

-2 -1 0 1 2

Cu

rren

t D

en

sity

(A

mp

microm

2 )

Voltage (Volt)

Current Density vs Voltage

density36

density100

density225

density400

density900

density2500

density10000

110

Table 37 New arrangement of the mask number and step in Second Run

Mask Number Step number Process

3 1 Isolation

22A 2 Mesa Etch

4 3 Bottom Contact

6

1

7

5

4

5

6

7

Via Etch

Top Contact

Bridge

Bond Pad

The fabrication was initiated by isolating each device using Mask 3 to define

them individually Then MESA mask (Mask 2) was used to cover the area from wet

etching optical measurement were used to obtain the actual size of the emitter without

metal Figure 318 shows the process results after H3P04H2O2H2O etching for a ratio of

3150

Figure 318 Optical images after MESA etching (a) 1 microm Tolerance (b) 05 microm Tolerance

Although the mask was designed with smaller size of 6times6microm2 using the slow etchant

solution the effective emitter size can still shrink to ~3times3 microm2 and ~2times2 microm

2 as can be

seen in Figure 318 The next step was to define the bottom contact using a metal scheme

that is suitable for GaAs ohmic layer ie AuGeNiAu After lift-off the samples were

then cleaned and spin-coated with AZ2microm Then the area was defined by using Mask 6

allowing the exposed area to be filled by Si3N4 which is the dielectric layer used in the

second run However a tricky issue happened here as the Si3N4 was difficult to lift-off

especially on the smallest emitter area in this mask In this run the lift-off process was

successfully done after three days Step five which uses Mask 1 was to define the top

111

contact The same metal scheme will fill up the tiny holes within the dielectric layer to

attach on the emitter semiconductor Figure 319 below shows an optical image of a

6times6microm2

device after the top contact lift-off process Once lifted-off annealing took place

to ensure Ge diffuse into the GaAs contact layer to lower the resistance

Figure 319 After lift-off processing

Step six uses Mask 7 to define the area of metal connection between active areas

to the bond pad In this step and step seven a different metal scheme from the top and

bottom contact was used to connect metal to metal Here a TiAu scheme was used The

final step (step 7) was to define the bond pad area In this context mask 5 is used The

bond pad requires thicker metal thickness to reduce the series resistance and to increase

the robustness of the metal surface when probed with needles during measurements

This second run of 2nd

Gen mask dielectric approach went well with better samplersquos

surface when using Si3N4 as the dielectric layer compared to the previous run using

S1805 Once the bond pad were defined for each device as done in previous run the

preliminary current-voltage characteristics of the Si3N4 run were obtained This is

described in the next section

342331 DC measurements

As mentioned in the previous section the Si3N4 run started with a defined emitter

area This allows actual ASPAT dimension to be measured thus the current density

versus voltage that are plotted in the following figures are based on actual measured

sizes

112

Figure 320 The current density of XMBE304 processed using Si3N4 as dielectric layer

Although the actual area of each device is obtainable from measurement the current

density of this sample does not scale well either Figure 320 show the current density is

not scalable from 03V to 1V This probably happened due to the same issue related to

the spreading resistance for each diode varying To investigate this issue new runs were

required with intention of reducing Rsprd and were expected to lead to more scalable

devices Therefore a study of reducing the spreading resistance (Rsprd ) due to

contribution of the large D-gap was carried out using the 2nd

Gen mask with the Si3N4

dielectric bridge approach The following section discusses in detail the steps used in

reducing the gap between the top and bottom contact for the GaAsAlAs ASPAT diode

34234 Dielectric-Bridge (Si3N4) Process Optimization by varying the D-Gap

The series resistance as discussed in detail in Section 26322 is generally due to

the total contribution of specific contact resistance in particular the diode size the sum

of all doped layer resistances that sandwich the main ASPAT layer and the spreading

resistance which comes from the lateral structure diode design[33] Thus in order to

acquire better performance at high frequency these contributors must be controlled One

parameter that can be controlled through this fabrication process is to reduce the

spreading resistance by controlling the separation between top and bottom contact ie

the D-gap as indicated in Figure 321

-000001

0

000001

000002

000003

000004

000005

-1 -05 0 05 1

Cu

rren

t D

ensi

ty

(Am

pm

2)

Voltage (Volt)

Current Density vs Voltage

D36

D100

D225

D400

D900

D2500

D10000

113

Figure 321 side view of lateral ASPAT structure

Without designing a new mask at this stage the same 2nd

Gen Dielectric-Bridge

mask was used but the process flows did not follow the sequence order of mask

numbers The reason for deciding not to design a new mask to reduce the D-Gap was

because the fabrication process flow had not yet confirmed it repeatability and

reproducibility Thus the available photo mask at that moment was fully utilised

Furthermore the feature of the 2nd

Gen mask that included the mesa tolerance can be

exploited in this study Hence the initial idea was to reduce the length of D by using the

tolerance that was designed in with Mask 2 (MESA etch mask)

Table 38 New arrangement for the Third run using Dielectric-Bridge mask

Mask Number Step number Process

1 1 Mesa Etch

3 2 Isolation

3

2

3

4

Bottom Contact

Mesa Cover

6

1

7

5

5

6

7

8

Via Etch

Top Contact

Bridge

Bond Pad

The process therefore was initiated by cleaning the sample and was then

followed by spin-coating it with S1805 to cover the emitter contact and define the actual

D-gap

114

size of the diode This first step used Mask 1 with wet chemical etching using the slow

solution (H3P04H2O2H2O etch with a ratio of 3150) Once the resist was striped

optical measurements were performed to obtain the effective area of the emitter Figure

322 below shows the smallest emitter area obtained after etching

Figure 322 The measured size of the emitter area and the length D (blue color marked)

Step two was to isolate the devices individually by using Mask 3 This was

performed using the fast etch solution (H3P04H2O2H2O etch with ratio of 212) This

took about 10 seconds to remove about 10000Aring of material

Step three which involved two masks was the most complicated in this process

Firstly Mask 3 was used to define the bottom contact by covering the sample with

AZ2microm and then the sample was hard- baked at 190 ˚C for 4- 5 minutes to ensure it had

totally dried before applying Polydimethylglutarimide (PMGI) The PMGI type used in

this process is Lift-off Resist (LOR) SF11 After spin-coating the SF11 at 7000RPM for

45 seconds it was post-baked at 190 ˚C for 4 minutes This was done to ensure that the

LOR SF11 was hard enough for the S1805 to stick on it This was followed by spin-

coating S1805 at 4000RPM for 30 seconds and exposing it under i-line UV for 20

seconds and developed using Micro Dev mix with DI water (11 ratio) for one minute

When the correct shape of S1805 was formed the sample was exposed under flood UV

for 15 minutes The SF11 was developed with 101A developer for 1 minute Figure 323

below summarizes the whole process in step three The step three processes concluded

by depositing the bottom contact with alloyed metal scheme and once lift-off had taken

place the next step continues as usual

115

Figure 323 Summary of LOR technique steps

Step four as well as the subsequent steps were completed in this run by copying

from the previous run (ie second run) which was to deposit the Si3N4 as the dielectric

layer and so on All the devices that underwent fabrication using this technique were

measured under optical microscope MC60 assisted by the Walsall software tool in the

lab From the measured value obtained it was expected that reducing the length of D in

this technique would improve the spreading resistance by up to 70 from the original

2nd

Gen mask design Table 39 summarizes the calculation of original spreading

resistance and improvement using this technique The calculation method and equation

are taken from Section 26322

Table 39 The outcome of the spreading resistance before and after using LOR technique

Device Size(microm2) D-original(microm) RSprd(Ω) D-new(microm) RSprd(Ω)

100x100 4 056 163 023

50x50 14 327 194 054

30x30 19 602 163 097

20x20 21 832 212 135

15x15 22 1007 202 18

10x10 24 1293 205 252

6x6 25 1643 230 418

116

Technically from the experiment in this run it was observed that the development time

also controlled the final length of D longer development time produced shorter D

lengths while shorter time yielded larger D gap lengths Therefore in this run the

development time was kept constant because of the desired D length from the mesa

tolerance in general is only ~2microm

342341 DC measurement

As usual when wafer processing was completed early DC measurement took

place to check the diodes performance Each diode size was measured and compiled

into one graph as shown in Figure 324 It is clear that the performance was better than

that in the previous run in term of scalability and current conductivity

Figure 324 IV characteristic in 3rd run (reduced d-Gap) for each diode size on the mask

To compare the current conductivity from this run and the previous run current

density at 05V was measured for both 2nd

and 3rd

runs Assuming the contact resistances

(TLM) for both runs were constant for 50times50microm2 device size the improvement of

current in the 3rd

run is about 92 This showed that such approach to reduce the D-gap

-500E-06

000E+00

500E-06

100E-05

150E-05

200E-05

-2 -15 -1 -05 0 05 1 15 2Cu

rre

nt

De

nsi

ty (

Amicro

m2)

Voltage (Volt)

Den_100microm2

Den_225microm2

Den_400microm2

Den_900microm2

Den_2500microm2

Den_10000microm2

Den2_36microm2

117

was successful in this run A new mask design was ready to take the challenge for

processing toward millimetre and sub-millimetre wave application

343 Fabrication process of GaAsAlAs ASPAT diode toward High frequency

Applications

So far the information that can be gathered from previous processing is that the

optimum process flow is achieved through dielectric approach design The effort in

reducing the series resistance by lowering its biggest contribution was attained through

lowering the D-Gap in the structure Once everything had been optimized ie the

process flow series resistance junction capacitance etc it was time to develop a new

mask design which only focused on the development of small ASPAT devices for use in

the millimetre and sub-millimetre wave regions

3431 Third Generation (3rd

Gen) mask design

The 3rd

Gen mask design was developed by taking into the account every aspect

of parameters that can contribute to the device by means of robust devices that are able

to function properly at ultra-high frequency The device cut-off frequency is given by

119891119888119906119905minus119900119891119891 =

1

2120587119877119904119862119895

(31)

Here RS is the series resistance and Cj is the junction capacitance To obtain high cut-off

frequencies Rs and Cj must be kept as low as possible From the fabrication point of

view two parameters that can be directly and easily controlled are the device area (A)

and the D-gap (which contributes to Rs)

Therefore calculations were made to find the best option Table 310 shows the

calculated value of capacitance (eq230 in page 44) and cut-off frequencies (eq31

above) for the ASPAT diodes studied These two equations extract the cut-off frequency

of the ASPAT assuming no external effects and for fully depleted devices The

XMBE304 ASPAT sample is expected to be suitable for millimetre wave applications

if small devices are successfully made In real devices sub-millimetrewave operation

can be hard to achieve due to increased series resistance and other process related

parasitics

118

Table 310 DC and RF characteristics for XMBE304

Device Size (microm2) Fully depleted

Capacitance(fF)

Calculated Series

Resistance (Ω)

Fully depleted Cut-Off

Frequency(GHz)

10000 5490 04 72

900 490 15 216

36 198 7 1148

16 879 11 1646

4 22 29 2500

Therefore in the 3rd

Gen mask design the smallest device that can possibly be obtained

in the GaAs based material fabricated using i-line lithography which is available at the

University of Manchester is 2times2microm2 and the gap between top and bottom is 15microm at

least The connecting bridge technique applied only utilised dielectric bridge method for

GSG features Figure 325 shows the layout of actual 3rd

Gen mask used in this study

There are 344 die chips on this type of mask design It also includes the Ground-signal

ground pad with 50um pitch for each chip and six de-embedding test structures as well

as eight TLM structures

Figure 325 3rd

Gen Mask layout showing actual devices de-embedding TLM and alignment mark

structures used in this study

119

This mask is designed generally from optimizations from 2nd

Gen Mask which

deploys a dielectric bridge for connection between the devices to the bond pads

Consequently the processing steps are not being much different but mostly follow what

is shown in Table 311 below The difference only applies to the much smaller mesa

size Other features included in this mask are de-embedding structures for RF

measurements via-hole test structure TLM structures and parallel plate capacitance test

structures

Table 311 3rd

Gen Mask process step

Mask Number Step number Process

1 1 Mesa Etch

2 2 Isolation

3

4

3

4

Bottom Contact

Mesa Cover

5

6

7

5

6

7

Via Etch

Top Contact

Bond Pad

The high frequency fabrication process flow is summarized in the following section

which shows illustrations in three dimensional and cross sectional (Figure 326) views

for easy understanding

34311 Step by Step Processing

0 Wafer preparation

Cleaning using NMP and DI

water or

Acetone and IPA

120

1 Mesa Etch

2 Isolation

3 Bottom Contact

4 Mesa Cover (Dielectric Deposition)

5 Via Etch

Etching using etchant solution

H3P04H2O2H2O etch with ratio

of 3110

Etching using etchant solution

H3P04H2O2H2O etch with ratio

of 212

Metal deposition with metal

scheme AuGeNiAu for

~500nm thick

Dielectric deposition 500nm

thick Si3N4

Via etch to open holes for metal

contact using reactive ion

etching using CF4

121

6 Top Contact

7 Bridge and Bond Pad

Figure 326 Step by step device fabrication (in 3D and cross-sectional view) for semi-insulating

substrate device type used in this study

Figure 327 Example finished process device with bond pad using 3rd

Gen mask

3432 TLM measurements

Transmission Line model (TLM) measurements for this run were carried out

after annealing (ie at 420˚C for 2 minutes) for process control monitoring by extracting

contact resistance (Rc) values These values are a measure of the quality of ohmic metal

contacts for a given process As discussed in Chapter 2 the TLM technique used four-

Metal deposition with metal

scheme AuGeNiAu for ~500nm

thickness and thermal annealing

420˚C for 2 minutes

Metal deposition with metal

scheme TiAu (1microm thick)

122

point measurement on TLM test structures located around the 15times15mm2 tiles as shown

in Figure 222 (in page 55) Normally five TLM structures are measured across the tile

for both top and bottom contacts Figures 328 and 329 display graphical TLM results

for top and bottom contact respectively As can be seen both graphs exhibit excellent

uniformity

Figure 328 XMBE304 TLM measurement for the top contact after annealing

Figure 329 XMBE304 TLM measurement for the bottom contact after annealing

Based on the graphs above for the top metal contact the average contact

resistance (RC) value using the metal scheme of AuGeNiAu is found to be ~005Ωmicrom

and the sheet resistance (RSH) is 22Ω However for the bottom contact the average

value for contact resistance is 012Ωmicrom and sheet resistance obtained is 26Ω Both

values are in a very good agreement with the known doping of both ohmic contact

0

2

4

6

8

10

0 10 20 30 40 50

Res

ista

nce

(O

hm

s)

Spacing (microm)

After 420C for 2mins top contact

y = 02592x + 02375

0

2

4

6

8

10

12

14

0 10 20 30 40 50

Res

ista

nce

(O

hm

s)

Spacing (microm)

After 420C for 2mins bottom contact

y = 02157x + 01053

123

layers Therefore the specific contact resistance (Eq 250) that contributes to the total

series resistance can be calculated and the value obtained is 15Ωmicrom2 and 54Ωmicrom

2 for

top and bottom TLM structures respectively

3433 DC characteristic measurements

Again once the wafer processing is completed room temperature DC

characteristics are taken using an HP 414B or HP500B parameter analyser and its actual

setup is as described in Chapter 2 This initial measurement was to ensure the

functionality of the diodes For this run using the 3rd

Gen mask the I-V measurements

for mesa active area of 4times4microm2 6times6 microm

2 and 10times10 microm

2 were taken and are depicted in

Figure 330 Figure 331 and Figure 332 respectively Nine diodes were measured for

each three device sizes to check their uniformity Thus the average current and standard

deviation are taken at two voltage steps of 1V and 15V Table 312 summarizes the

standard deviation for each measured data obtained from this final run

Figure 330 XMBE304 GaAsAlAs ASPAT measured IV characteristics at room temperature for

4times4microm2 mesa size

-00005

0

00005

0001

00015

0002

00025

0003

00035

-3 -2 -1 0 1 2 3

Cu

rren

t (A

)

Voltage (V)

C01_4um_1P_SI

C22_4um_1P_SI

C40_4um_1P_SI

W01_4um_1P_SI

W22_4um_1P_SI

W40_4um_1P_SI

AI01_4um_1P_SI

AI22_4um_1P_SI

AI40_4um_1P_SI

124

Figure 331 XMBE304 GaAsAlAs ASPAT measured IV characteristics at room temperature for

6times6microm2 mesa size

Figure 332 XMBE304 GaAsAlAs ASPAT measured IV characteristics at room temperature

for 10times10microm2 mesa size

Table 312 Standard deviation at two different voltages

Device size (um2) Standard Deviation 1V () Standard Deviation 15V ()

4times4 14 14

6times6 65 72

10times10 44 39

Noticeably the standard deviation of the device increases for the smaller size

devices This trend happened probably because the active mesa area is not uniform

causing different series resistances Since the smallest mesa active area which achieved

-0001

0

0001

0002

0003

0004

0005

0006

0007

0008

-3 -2 -1 0 1 2 3

Cu

rren

t (A

)

Voltage (V)

C02_6um_1P_SI

C23_6um_1P_SI

C41_6um_1P_SI

W02_6um_1P_SI

W23_6um_1P_SI

W41_6um_1P_SI

AI02_6um_1P_SI

AI23_6um_1P_SI

AI41_6um_1P_SI

-0005

0

0005

001

0015

002

0025

-3 -2 -1 0 1 2 3

Cu

rren

t (A

)

Voltage (V)

C03_10um_1P_SI

C24_10um_1P_SI

C42_10um_1P_SI

W03_10um_1P_SI

W24_10um_1P_SI

W42_10um_1P_SI

AI03_10um_1P_SI

AI24_10um_1P_SI

AI42_10um_1P_SI

125

a good I-V characteristic is 4times4microm2

in this run these devices were used for the next step

of characterisation which is the S-parameter measurement to extract their behaviour at

different frequencies

35 Conclusions

In this chapter basic fabrication techniques of GaAsAlAs ASPAT on both doped

and semi insulating substrates using standard I-line lithography as well as step by step

descriptions to achieve reproducibility in the process fabrication flow has been

demonstrated with relevant initial measured results Two major outcomes have been

demonstrated firstly related to repeatability reproducibility and manufacturability

mostly done on large device areas (15times15microm2 to 100times100 microm

2) Secondly a successful

process flow for small emitter size devices (2times2 microm2 to 10times10 microm

2) has been

developed

Subsequently two types of designs during process optimization were developed

namely Air-Bridge and Dielectric-Bridge approaches The latter approach seems

favourable for GaAsAlAs materials but was only successful when reducing the series

resistance of the device This problem was addressed by optimising the D-gap between

top and bottom contact which resulted in good scalability of each ASPAT dimensions

improving current conductivity by 92 and achieving a reproducible process On the

other hand in the former approach issues were encountered with over etching underneath

the diode effective area and thus it was hard to achieve reproducible devices

Repeatability reproducibility and manufacturability fabrication processes based on

dielectric bridge method were successfully developed This new process provides a

highly efficient and economical solution for the fabrication of GaAsAlAs ASPAT

diode Emitter sizes down to 4times4 microm2 dimensions are routinely and reproducibly

achieved in this process Series resistance which is an important parameter in

determining high frequency application are greatly reduced by changing the gap between

top and bottom contact

With all the new improvements implemented the possibility of the proposed Dielectric

Bridge method fabrication process was successfully applied for the fabrication of

ASPAT diodes

126

4 PHYSICAL MODELLING OF THE GAASALAS ASPAT

DIODE USING SILVACO

41 Introduction

Fabricating any device or circuit requires a lot of time resources cost etc

especially elements related to the production of semiconductor devices Furthermore in

realising such a device a clean room is required Hence fabrication and processing in

wafer fab need full attention to get it done with a higher rate of success reproducibility

and manufacturability One solution that can be highlighted which will be able to reduce

all the resources that are mentioned above is by using computer simulation approaches

or to be precise physical modelling For epitaxial layer based devices physical

modelling is a good choice as it would give a better understanding and insight into each

layer and how electrical characteristics are derived A software that is most suitable for

epitaxial layer physically modelled is SILVACO This software is a very comprehensive

tool to simulate epitaxial based devices and to predict their behaviour Such software

covers many aspects starting from the first principles of physic epitaxial layer

definition as well as device layout thus making it the most powerful virtual wafer fab

tool in the market

In this chapter the SILVACO packages are discussed The discussions include the

method of defining a new material defining models and constructing the AC or DC

supply to obtain the output characteristics of the virtual device Apart from these the

focus will also be on the ASPAT diode modelling simulation and analyses of the results

which will include the ASPAT structure suitable models and DC current-voltage

characteristics The dependencies of individual structure on the I-V curves will also be

highlighted Finally the discussion of results and analyses involving a range of

operational temperatures dependency as well as from comparison made to conventional

SBD used in this study will be examined

127

42 SILVACO modelling Tools

SILVACO is a modelling software introduced in 1984 by Dr Ivan Pesic It is

purposely created for electronicsrsquo devices physical modelling and characterization This

software company has become the major supplier for most of the Electronic Design

Automation (EDA) for circuit simulation amp design of analogue Mixed-Signal and RF

circuit market This Technology Computer Aided Design (TCAD) software can predict

the simulated device performances starting from first principles It has a package which

can provide Virtual Wafer Fabrication (VWF) simulation to the device designer and

which has the capability to perform two or three dimensional physical device modelling

by using the ATLAS simulator [103] SILVACO allows all parameters such as

electrical thermal and optical characteristics of a device to be simulated under desired

bias conditions It offers cost effectiveness as well as quick prediction of results for

many semiconductor devices compared to real experiments Some hands-on experiment

may not always perform hence SILVACO can be used as an alternative

The core of SILVACO is Atlas itself which provides a platform to perform DC AC

and transient analysis for such dimensional device structure regardless of the

heterojunction material type ie binary ternary quaternary etc As the brain of

SILVACO Atlas which receives input command files containing instruction text for

execution from a runtime environment known as Decbuild will process the instruction

text and display progression error and warning via Runtime Output All the calculations

of the resultsoutcomes of the simulation are plotted via a tool called TonyPlot which is a

tool to visualise the output Figure 41 below shows how precisely the physical

modelling takes place the process of building the structure how its parameters and

variables are defined how an appropriate model statement is selected how performance

is analysed and lastly how the outcomes are displayed

128

Figure 41 SILVACO Atlas simulation process flow

As can be seen in Figure 41 the structure specification statement is used to define

any desired structure by setting the command in Deckbuid of the following parameters

a) Mesh where the structure can be defined either in 2D or 3D Cartesian grids The

unit of the coordinates used is in microns and the spacing parameters which

define the netting size can be used to improve the accuracy of the analysis at any

given position The density of netting size in this statement determines the

processing time

b) Region where the multi-layers in a structure are defined and this statement has

to outline each layer that represents a separated region independently The mesh

must be assigned to a region and the sequence of the region is arranged from low

to high

-Mesh -Region -Electrode

-Contact -Material

Structure Specification

-Model -Interface

Model Specification

Method

Numerical Method

-Log -Solve -Load -Save

Solution Specification

-Tony Plot

Display Results

129

c) Electrode are used to define the location of bias point for a designed structure

when performing the electrical analysis In this work or for the case of a diode

two electrodes are allocated as an anode and a cathode In a vertical device the

latter electrode is placed at the bottom of the device while the anode is at the top

d) Doping this statement refers to doping concentration injected into the desired

region and it normally depends on the material types

e) Materials since SILVACO was developed specifically for Silicon-based devices

default parameters are set up for Silicon properties However the use of

materials statement allows SILVACO to run for different material ie GaAs

InGaAs etc In order to make it work this parameter is defined first and then

followed by the material name and its properties such as bandgap permittivity

conductionamp valence band discontinuities mobility etc

The most crucial part is to determine whether the simulated structure of a particular

device is correct or incorrect This is done by properly choosing the specific model

statement The Specific model statement is employed to express the physic equations

that are used during the device analysis The models statement depends solely on the

structure definition Examples are device structure with double barrier use Non-

equilibrium Green function (NEGF) model and single barrier uses Semiconductor-

insulator-Semiconductor (SIS) model for effective and accurate analysis process For the

case of a single barrier in SILVACO there are many model statements that can be used

for such analysis Therefore it is recommended to check model by model in order to

ensure that all the needed parameters are defined in the material statements and results

produced are valid

SILVACO is able to calculate such models by using different numerical methods

which means semiconductor device problematic is computed to make successive

solutions by random discretisation There are three different numerical methods that are

regularly used by SILVACO to perform its calculations which are Newton Gummel and

Block Basically this is solved by using a non-linear iteration procedure which begins

from an initial guess and which then uses an iterative process to find the predicted

solution The detail of these can be found in reference [103]

130

In order to turn on the problem solution the solution specification is defined This

include log solve save and load statement All these work together to provide data for

analysis by other functions The log statement is a file type that saves in memory and can

be loaded by Atlas Any solved device will be stored in the log file Therefore it is

necessary to define the LOG before SOLVE statement and close it after the calculations

While the save statement is used to store all data point to a node in the output file the

load statement is utilised to recall all the saved data to be read by Atlas

Finally all these files can be displayed or plotted on TonyPlot it is recommended to

make SET files at plotting point for a better visualisation The plotted or displayed files

in TonyPlot can be manipulated for scaling graphs overlaying different curves and most

importantly to export the data to other files formats

43 SILVACO Implementation GaAs AlAs ASPAT Modelling

The structures play an important role in determining the terminal output

characteristics Therefore to start simulating the ASPAT device the first thing that must

be specified is its structure As mentioned earlier in Section 3412 the ASPAT diode is

a top down multilayer structure This structure which is adopted from that of Section

3412 will be used as a basis to perform the ASPAT simulation Figure 42 shows the

structure of the ASPAT in this real simulation which is exactly the same as been

discussed in Section 26322 The ASPAT diode consists of two heavily doped (up to

5x1018119888119898minus3) GaAs contact layers on top and bottom slices adjacent to lightly doped (up

to 3x1017119888119898minus3) GaAs intermediate layers In between these layers is a sandwiched

structure consisting of two different lengths of undoped GaAs spacer layers and a thin

layer of AlAs that act as a tunnel barrier In this work the simulation result will be

compared with the fabricated measurement result depending on the size of the diode

The actual device that will be used to validate the simulation is the main device used in

this study (XMBE304) which is based on a lattice matched GaAsAlAs grown on a

semi-insulating substrate Therefore the design structures that are proposed in the

fabrication are compatible with the fabricated devices and are based on lateral structures

as can be seen in Figure 42

131

Figure 42 The GaAsAlAs ASPAT structures produced from simulation based on the diode

multilayer heterostructures on the right

In the consequent simulations a key observation regarding the AlAsGaAs

heterojunction is that there are two types of tunnelling processes direct tunnelling and

indirect tunnelling Figure 43 shows the Energy-Momentum (Ek) diagram depicting the

three valleys for the AlAs conduction band namely L Γ and X points In normal

circumstances the transition of electrons happens to the X-point which is the lowest

energy in the conduction band In the case of very thin barrier the tunnelling process

occurs at the Γ point in both AlAs and GaAs materials which is the direct tunnelling

process[18] It has been reported that in the case of an ASPAT diode tunnelling which

occurs at the Γ point will be the dominant component in the tunnelling current

Therefore the actual band gap will be different from the one at the X-point which is

which 216eV[15] By contrast the energy band gap at the Γ-point is around

283eV[104] Thus this simulation uses this band gap value

Figure 43 The Energy-momentum diagram for AlAs heterojunction semiconductor

216eV

Γ

X

E

K

L

AlAs

CB

VB

283e

132

The simulation code as attached in Appendix IV and the output of the simulation with all

the input mentioned above are shown in Figure 44 (a) and Figure 44 (b) Figure 44 (a)

is the band diagram at equilibrium and Figure 44 (b) is the band diagram when a bias is

applied

Figure 44 (a) The conduction band and the valence band of the ASPAT diode structure (b) the

energy band diagram of the ASPAT diode structure when under three different biases

44 Simulation Result and Analysis

Basically the SILVACOrsquos Atlas simulation package is used to calculate the I-V

characteristics from multilayers structures In the case of GaAsAlAs heterojunctions all

details of the structure as shown in Figure 44 above are calculated based on solving the

Schrodinger time-independent equation in each layer taking into account the variation of

effective mass and conduction band offset between GaAs and AlAs

For thermionic emission and tunnelling mechanism across an abrupt heterojunction

interface the general method used in SILVACO is taken from K Y Yang work [105]

The tunnelling current of the ASPAT diode uses the equation below[74]

119869 = sum2119898lowast119864119894(1 minus 119877)119896119879

1205872ħ3

119873

119894=1ln [

1 + exp (119864119865 minus 119864119894

119896119879)

1 + exp (119864119865 minus 119881 minus 119864119894

119896119879)]

(41)

Where m is the electron effective mass E denotes the energy of the electron R is the

total resistance k represent Boltzmann constant ħ is the reduced Planck constant EF is

the fermi level V is applied voltage and Ei is the electron energy perpendicular to the

-2

-15

-1

-05

0

05

1

15

1

82

16

3

24

4

32

5

40

6

48

7

56

8

64

9

73

0

81

1

89

2

97

3

Ene

rgy(

eV

)

Thickness (microm)

VB

CB

133

barrier The important parameters that enable SILVACO Atlas to perform correct

calculations and analysis are the choice of appropriate models The suitable model that is

available for evaluating the GaAsAlAs ASPAT is based on the Semiconductor-

insulator-semiconductor (SIS) model

Thus in this run Non-local Quantum Barrier Tunnelling Model (SISEL and

SISHO) are utilized specifically semiconductor-insulator-semiconductor mode This

model enables the tunnelling current between two semiconducting regions separated by a

quantum barrier to be calculated [103] It is assumed that the charge tunnels across the

whole barrier with the source or sinks at the interface with the semiconductor regions

Under the Non-Local Quantum Barrier tunnelling model another model that can be used

is semiconductor-semiconductor-semiconductor (SS) tunnelling model if the materials

are specified By correctly inserting all parameters with the right model an excellent DC

IV characteristic match between simulation and measurement can be produced as shown

in Figure 45

441 DC Current-Voltage Characteristic

In this simulation the current at each bias step and each mesh point can be set up by the

user however the detailed calculation such as formula usage methodology and

approach that is adopted by SILVACO Atlas is unknown As mentioned above in order

to produce the energy band diagram of the ASPAT the DC characteristic of the structure

can be solved by using the Schroumldinger and Poisson equation self-consistently

To ensure that these simulations are valid one fabrication was performed on an

ASPAT diode sample XMBE304 For this sample the structure parameters are the

same as has been set in this simulation The result of the measurement and simulation

are then compared Figure 45 shows that the simulation result is in excellent agreement

with the measured data for this sample

134

Figure 45 The current-voltage characteristic of an ASPAT diode (100times100microm2) and (4times4microm

2)

using SILVACO Atlas simulator for structure device XMBE304 showing excellent agreement

between simulated and experimental data

These fitted results were performed on both a large 100times100microm2 device and the

smallest obtainable from fabrication (4times4microm2) which was to be used for the repeatability

amp reproducibility studies as well as for high-frequency applications study respectively at

room temperature In order to get a good fit a few parameters had been modified in the

SILVACO software via the Deck-built tool for example energy band effective masses

and bandgap discontinuity of GaAs spacer and AlAs barrier (The mentioned parameters

values are summarized in Table 41) The percentages of bandgap discontinuity in

SILVACO using the ALIGN parameter is given by[103]

119860119897119894119892119899 =

Δ119864119862

Δ119864119862 + Δ119864119881

(42)

Where ΔEc and ΔEv are the conduction band discontinuity and valence band

discontinuity respectively The m0 in the tables denotes the electron rest mass Once all

agreement between measurement and simulation has been met the simulation is then

carried out with structure analysis at room temperature and different temperatures

simulation

Table 41 The parameter values used in this simulation

Material Bandgap(eV) ΔEg(eV) Effective mass(kg)

GaAs 1424 03 0067m0

AlAs 2835 071 0126m0

-00002

0

00002

00004

00006

00008

0001

00012

-2 -1 0 1 2

Cu

rre

nt

I (A

mp

)

Voltage V (Volt)

Current vs Voltage

Measurment

Simulation

135

45 Structure Analysis of ASPAT Diode

Once the device structure was modelled and having successfully produced a

precise band diagram as well as validated the simulation results with experimental I-V

characteristics the next step is to further analyse the relationship between basic device

structure and its I-V characteristics This approach is used to predict what would happen

to the DC output if some of the parameters were varied especially with regards to the

AlAs barrier thickness In the subsequence simulations the thickness of each main

ASPAT (unequal spacers and barrier) layer will be studied independently as a variable in

order to determine how each parameter affects the I-V characteristic in both magnitude

and curvature The analysis will also include manufacturing tolerance where the

structure parameter which will result in a 10 difference in their I-V characteristics is

examined[59 75] This will provide an overview of how precisely to manufacture each

layer of the device The following simulation is based on the XMBE304 structure with

emitter size of 4x4microm2

451 Dependencies of current on AlAs Barrier thickness

Since the AlAs barrier is what limits the transportation of electron flow and

hence the current (which depends exponentially on the tunnelling barrier thickness)

therefore the first analysis to run on the simulation is the variation on barrier thickness

In the simulation the barrier thickness is measured in term of the monolayer Generally

one monolayer can be calculated by dividing the lattice constant of the material by two

In the case of AlAs one monolayer is calculated as follows

119900119899119890 119898119900119899119900119897119886119910119890119903(1119872119871) =

119860119897119860119904 119897119886119905119905119894119888119890 119898119886119905119888ℎ119890119889 (5666Å)

2

(43)

= 283Å

The nominal value of the AlAs barrier thickness for sample XMBE304 is 283nm

ie ten monolayers In the first simulation test the current change due to the barrier

thickness variation from 9ML to 11ML was examined first followed by the amount of

barrier thickness change that would produce a 10 change in current The simulation is

setup by fixing all other parameters and varying the AlAs barrier thickness as mentioned

136

above with a step of 02ML In order to determine what fraction of a ML would yield a

10 difference in the current the barrier thickness is slightly changed to fit the curve for

both 5 above and 5 below the original curve

Figure 46 IV characteristics of the dependencies of current on AlAs barrier

The current-voltage characteristics of the ASPAT diode do change dramatically with

barrier thickness in forward bias but not that much in reverse bias (Figure 46) The

current decreases as the barrier thickness increases For a 1ML change in layer thickness

(from 9ML to 11ML) the current changes by over ~300 at 05V

Figure 47 Example of analysis at -1 and 1V to the current

-00002

0

00002

00004

00006

00008

0001

00012

00014

-15 -1 -05 0 05 1 15

Cu

rre

nt

(A)

Voltage(V)

9ML

92ML

94ML

96ML

98ML

10ML

102ML

104ML

106ML

108ML

11ML

973E-06

0

0000005

000001

0000015

000002

0000025

000003

0000035

85 95 105 115

Cu

rren

t a

t 1V

B

ias

(Am

p)

Barrier Thickness (ML)

Current change with tunnel barrier thickness

Forward Current

5

-5

137

From the simulation result shown in Figure 46 a 9945ML barrier thickness will

give a 5 higher current and a 10056ML barrier will give a 5 lower current (Figure

47) Therefore in total 01 ML difference yields 10 current difference These indicate

that in order to control the current within 10 barrier thickness difference the growth

precision in the barrier must be precise to better than 01ML Extensive studies have

also shown that the I-V characteristic of a GaAsAlAsGaAs diode is very sensitive to

the thickness of AlAs barrier This work has been reported elsewhere[17]

452 Dependence of current on Spacer I length l1

For the longer spacer length (l1) five different values are chosen from 01microm to

03microm The lengths are changed in the order of 005microm Therefore the arrangement of

length is as follows l1=01microm l1=015microm l1=02microm l1=0 25microm and l1=03microm

respectively The results are plotted from -15V to 15V anode voltage in Figure 48 The

I-V characteristic of the ASPAT diode does not change much with the length in the

forward bias region but in the reverse bias region the current decreases as the layer

thickness increases Here the l1 layer acts as a voltage arm and a small size device

cannot sustain big changes in spacer length Changing the length at the forward region

will also change the energy states on the anode side as well changing the states

distribution on the cathode side in reverse bias

Figure 48 I-V characteristic of the dependencies current to Spacer I layer

-00002

0

00002

00004

00006

00008

0001

-2 -1 0 1 2

Cu

rren

t (A

)

Voltage (V)

L=01

L=015

L=02

L=025

L=03

138

By fixing the current at -1V and 1V the current and the layer thickness relationship is

illustrated in Figure 49

Figure 49 Current changes with layer thickness l1

It is noticeable that there is a dramatic change in reverse current from 01 microm to

05 microm layer thicknesses However the forward current only falls slightly from 01microm to

015microm and is stable afterwards Hence for a small size device a large change in spacer

layer at the cathode will allow more current to pass

453 Dependence of current on Spacer II length l2

Finally is the variation in the spacer II Similar to spacer I above five values are

chosen for the shorter undoped GaAs layer length l2 the thickness is varied from

00025microm to 00075 microm (steps are l2=00025microm l2=000375microm l2=0005microm

l2=000625microm and l2=00075microm respectively The results are plotted from -15V to

15V anode voltage as shown in Figure 410 In this case a slight change in I-V

characteristic in the forward bias can be seen clearly which means the I-V

characteristic depends on the length of the shorter undoped layer Therefore the l2 layer

also acts as another voltage arm due to the asymmetrical length The effect is quite

similar to the spacer l1 but this times the forward current only slight changes The reason

for the small change in current is that the length change is small and it linearly affect the

states distribution

-000002

-0000018

-0000016

-0000014

-0000012

-000001

-0000008

-0000006

-0000004

-0000002

0

0

000005

00001

000015

00002

000025

00003

000035

01 015 02 025 03

Rev

erse

Cu

rren

t (A

)

Forw

ard

Cu

rren

t (A

)

Spacer Lenght (microm)

Forward Current (1V)Reverse Current(-1V)

139

Figure 410 IV characteristic of the dependencies current to Spacer 1 layer

Fixing the current at -1V and 1V the current versus layer thickness relationship is

illustrated in Figure 411

Figure 411Current change with layer thickness l2

The I-V curve depends on the length of the shorter undoped spacer layer quite linearly

The forward current changes in increase to the layer compared to the backward current

The layer thickness l2 should be small as long as it prevents carrier diffusion Therefore

all these three layers must be kept within limit to ensure that the high performance of the

ASPAT diode can be fully utilised

-00002

0

00002

00004

00006

00008

0001

-2 -1 0 1 2

Cu

rren

t (A

)

Voltage (V)

L=75n

L=25n

L=625n

L=5n

L=375n

-2E-07

-18E-07

-16E-07

-14E-07

-12E-07

-1E-07

-8E-08

-6E-08

-4E-08

-2E-08

0

0

00002

00004

00006

00008

0001

00012

0002 0004 0006 0008

Rev

erse

Cu

rren

t (A

)

Forw

ard

Cu

rren

t (A

)

Spacer Lenght (microm)

Forward Current (1V)

Reverse Current(-1V)

140

46 Temperature Dependent Simulation

This section will discuss in detail how changes in operating temperatures influence

the IV characteristics of the ASPAT diode The same codes as in the previous

simulation with fitted results are used for this temperature dependence study but a few

parameters were changed for different temperatures

Theoretically the material parameters that are influenced by the change of

temperature are band gap electron effective mass the density of state (NC NV) light

hole mass heavy hole mass permittivity and electron amp hole mobilities However not

all mentioned parameters will have a large impact on the IV characteristic in the

SILVACO Atlas simulation The most significant factors that give appreciable impact on

the DC output current-voltage were the energy bandgap and the effective mass The

GaAs bandgap as a function of temperature is given by the equation below [106]

119864119892 = 1198641198920 minus

120572 1198791198712

120573 + 119879119871

(44)

Here Eg is the bandgap Eg0 denotes the bandgap at 0K TL is the Temperature α=

Constant (Varshni Parameter) AlAs6e-4 GaAs5405e-4 β= Constant (Varshni

Parameter) AlAs408 and GaAs204 The calculated parameters that are used in this

simulation are shown in Table 42

Table 42 The calculated values of bandgap at different temperatures

Temperature (K) GaAs Eg(eV) AlAs Eg(eV)

77 1506 2903

100 1500 2899

125 149 2893

150 1486 2887

175 1478 2879

200 1470 2871

225 1461 2863

250 1452 2854

275 1443 2844

300 1424 2835

325 1419 2824

350 1414 2814

375 1404 2803

398 1394 2793

141

The effective mass of the materials used in this simulation can be expressed by

119898119899 = 1198980119899 + 11989810 (

119879119871

300119870)

119898119901 = 1198980119901 + 1198981119901 (119879119871

300119870) + 1198982119901 (

119879119871

300119870)2

(45)

Where mn is the effective electron mass mp represents the effective hole mass m1n and

m1p are constant number for the basic GaAs material m0n

for 119898119883119898119871 119886119903119890 119892119894119907119890119899 119887119910 119905ℎ119890 119890119902119906119886119905119894119900119899 (1198981198991199052 lowast 119898119899119897)

13 while m0p is based on the

expression 1198980119901 = (11989811990111989732

+ 119898119901ℎ32

)23 The calculated parameters are shown in Table

43

Table 43 The calculated effective masses for each temperature used in this simulation

Temperatures (K) GaAs Effective Mass (kg) AlAs Effective Mass (kg)

77 00660m0 03790 m0

100 00658 m0 03788 m0

125 00655 m0 03785 m0

150 00652 m0 03782 m0

175 00649 m0 03779 m0

200 00646 m0 03776 m0

225 00643 m0 03773 m0

250 0064 m0 0377 m0

275 00637 m0 03767 m0

300 00634 m0 03764 m0

325 00631 m0 03761 m0

350 00628 m0 03758 m0

375 00625 m0 03755 m0

398 00622 m0 03752 m0

142

Figure 412 Measurement and simulation comparison result as a function of temperature range

from 100K to 398K

Figure 412 above shows excellent agreement between simulation and

measurement results at various temperatures The IV characteristics correspond to a

device of size 100times100microm2 as presented in Chapter 3

47 Temperature Dependence characteristics of GaAsAlAs ASPAT Diodes

As mentioned earlier in chapter one the tunnelling diode has many advantages

over conventional Schottky barrier diodes some of which are a large dynamic range

low power consumption and very weak temperature dependence This section will

discuss the effect of variable temperature applied to the GaAsAlAs ASPAT diode and a

similarly processed TiAu Schottky diode Two samples were fabricated together for

these studies (XMBE304 and XMBE104 representing an ASPAT and a SBD

respectively) The fabrication technique is exactly the same as has been discussed in

Chapter 3 In order to make it fair for direct comparisons as well as easy probing both

diodes were fabricated with the same emitter size (100times100microm2) The DC measurements

at different temperature were carried out using a Lakeshore Cryogenic probe station over

the range of 77K to 398K in 25K step interval

-002

-001

0

001

002

003

004

005

006

-2 -15 -1 -05 0 05 1 15 2

Cu

rren

t I

(Am

p)

Voltage V (Volt)

T=100K_Simu

T=100K_Meas

T=398K_Simu

T=398K_Meas

T=200K_Simu

T=200K_Meas

T=300K_Simu

T=300K_Meas

143

471 GaAsAlAs ASPAT diode vs TiAu SBD

Once fabrication and measurement were completed both DC outputs of the diodes

were characterised and analysed Figure 413 shows a semi-logarithmic plot for

measured current versus voltage as a function of temperature for ASPAT sample

XMBE304 In forwards bias the current changes for different temperatures from 77K

to 398K are less than 5 percent This confirms the very weak temperature dependence of

current transport as it is dominated by tunnelling through the barrier On the other hand

the backward bias shows the current changes at different temperature are slightly bigger

than in forward this due to band bending occurring faster (making the effective barrier

lower) and allowing thermionic emission to significantly contribute to transport of the

current The only other study of temperature dependence for the ASPAT was made by et

el RT Syme[15] but details were not stated in their report

Figure 413 Log Current vs voltage as a function of temperature for ASPAT sample XMBE304

The effective barrier height for the GaAsAlAs ASPAT diode is higher than that of

the SBD (See Figure 414) therefore there is an expectation of more limited thermionic

current flow in the ASPAT than the SBD As mentioned earlier the conventional

Schottky Barrier diode that is used in this study consists of a Gold Titanium and GaAs

(AuTiGaAs) interface which is the baseline for the temperature dependence study

0000001

000001

00001

0001

001

-15 -1 -05 0 05 1 15

Log

Cu

rren

t (A

)

Voltage (V)

T=77K

T=100K

T=125K

T=150K

T=175K

T=200K

T=225K

T=250K

T=275K

T=300K

T=325K

T=350K

T=375K

T=398K

144

Figure 414 The effective barrier high of the Schottky barrier diode and the ASPAT Diode [3]

The SBD epitaxial layers profile is as shown in Table 44 below Theoretically the SBD

obey thermionic emission transport[44] and its I-V characteristic is given by

119868 = 1198680[exp (

119902119881

119899119896119879) minus 1]

(46)

Where q is the electron charge V is the applied voltage across the diode n denotes the

diode ideality factor k is the Boltzmann Constant T is the absolute temperature in

Kelvin and I0 is the diode saturation current which is given by the expression

1198680 = 119860119860lowast1198792 exp (minus

119902empty1198870

119899119896119879) exp (minus120572120594119890

12120575)

(47)

here A is the area of the diode A denotes the effective Richardson constant Oslashb0 is the

barrier height at zero bias δ represents the thickness of interfacial insulator layer χ

denotes the mean tunnelling barrier and α = radic(4120587

ℎ)(2119898lowast) is a constant value The

ideality factor n is taken from the slope of the SBD current-voltage characteristic and in

this study its value varies from 1 to 2 (depending on temperature) In the case of the

ASPAT diode thermionic emission can also happen if a thicker barrier is used (~ 100Aring

or thicker) as shown by CS Kyono et el [104] who concluded that when a thicker

barrier of AlAs barrier is used the current transport is dominated by thermionic emission

145

Table 44 Epitaxial layer profile for Schottky Barrier Diode (XMBE104)

Layer Material Doping(cm-3

) Thickness(Aring) Bandgap (eV)

Schottky GaAs(Si) 500E+15 7500 14

Semiconductor GaAs(Si) 300E+16 7500 14

Semiconductor GaAs(Si) 100E+17 7500 14

Ohmic GaAs(Si) 500E+17 7500 14

Buffer GaAs(Si) 300E+18 7500 14

Substrate GaAs(Si) N+ 3000 14

The fabricated SBD was also measured and its I-V characteristic is plotted as a

function of temperature in Figure 415 Unlike the ASPAT diode the current at forward

bias for the SBD change enormously with temperature from 77K to 398K and at all

biases For the ASPAT diode the slight change in current only started after 08V bias as

the current starts to have some component of thermionic emission over barrier

Figure 415 Log Current vs voltage as a function of temperature for SBD sample XMBE104

In order to see clearly how much the current is changing in forward bias for both

ASPAT and SBDs diode a log current at different voltages versus 1000temperature is

plotted as shown in Figure 416

1E-08

00000001

0000001

000001

00001

0001

001

01

-2 -1 0 1 2

Log

Cu

rren

t (A

)

Voltage (V)

T=398K

T=375K

T=350K

T=325K

T=300K

T=275K

T=250K

T=225K

T=200K

T=175K

T=150K

T=125K

T=100K

T=77K

146

Figure 416 Influence of temperature on IV characteristic for both ASPAT diode and SBD

Semi-logarithmic plots of current (at V= 05 06 07 08V) versus inverse

temperature for both SBD and ASPAT are shown in Figure 416 When the temperature

is increased the current also increases in the SBD as a result of thermionic emission

over the barrier for sample XMBE104 This is in contrast to the temperature-

independent tunnelling through the thin AlAs barrier of sample XMBE304 ASPAT

diode where when the temperature is increased the current is almost constant

At low and high temperatures the ASPAT shows excellent temperature

independence with a constant current flow It exhibits a tunnelling current in excess of

values expected by the elastic tunnelling current calculation equation suggested by RT

Syme [16 18] above (Eq 1) using a Oslash value of 105eV (ΓGaAs to Γ AlAs tunnelling) By

contrast for the SBD at low temperature (77K-275K) the changes of currents were very

high and for every 02V there is an exponential change of more than 40 This

temperature dependent study was also reported in[68]

147

48 Conclusions

This chapter demonstrated the establishment of an excellent physical model and

comparison of room temperature I-V characteristics of GaAsAlAs ASPAT diodes for

different emitter sizes their scalability as well as an investigation of their characteristics

at different temperatures from 77K to 398K Simulation are validated on well-

characterized experimental data and excellent fitting which had been achieved in this

work permit the designer to extract all related parameters of heterojunctionmultilayer

ASPAT structures thus creating modification for future growth specification in order to

achieved precise designs

It is clear that the work which had been carried out in this chapter is able to

achieve with adequate accuracy a claim of reverse engineering capability The ability of

the GaAsAlAs ASPAT to act as a zero-bias detector has been analysed and compared

with the SBD It is clear that the temperature stability which is shown by the

GaAsAlAs ASPAT is much better than that of the SBD thus demonstrating that the

tunnelling current is dominant over the thermionic emission in ASPAT diodes

148

5 DC amp RF CHARACTERISTIC OF ASPAT DIODES

51 Introduction

To assist in circuit designs for any type of high-frequency circuits such as

millimetre wave detectors frequency multipliers and mixer circuits which are built

based on non-linear devices (diodes) an equivalent-circuit model for the diode is

required This is among the simplest and most effective method for analysing

semiconductor devices which work at high frequency where the electrical characteristics

measured obtained from the devices are extracted and presented in a circuit consisting of

lumped elements components (resistor inductor capacitor etc) However accurate DC

and RF measurement data is essential to extract the equivalent-circuit elements quickly

and correctly The extracted parameters values from the circuit that are taken into

account usually depends on bias and frequency associated with the device physically

which is also interrelated to the semiconductor material parameters device structure as

well as fabrication process flows

In this work the DC and RF data were derived from DC and S-parameter

measurements respectively These measurements were carried out both in-house and at

the University of Cambridge by a collaborator partner (Prof MJ Kelly) The I-V

characteristics of the diode obtained from DC measurements were measured from -2V to

2V while the S-parameters were carried out over a wide frequency range from 40MHz to

40GHz with eight different biases In this chapter the DC measurements for various

sizes of the diodes with analysis of their IV characteristics will be discussed The one-

port on-wafer ASPAT measurement setup as well as the de-embedding method will

also be explained Thereafter the equivalent circuit models with all lumped element

effect will be discussed This work is carried out with the help of the VNA which

principle has been described in Chapter 2 and Keysight ADS simulation tool All

technical details regarding the equivalent circuit models will be explained together with

the method used for the ASPAT diode evaluation The equivalent circuit model also will

cover the diode intrinsic elements such as Cj Rj and Rs and extrinsic elements ie CP

149

and RP Finally an equivalent circuit model with the small signal characterization of the

fabricated ASPAT diodes will be presented

52 General Analysis of Current-Voltage (I-V) characteristics of GaAsAlAs

ASPAT diodes

The recent development of state-of-art for DC measurement apparatus has led to

capabilities for high-level accuracy of measuring voltages to a few nano Volts and

current signals in the femto Amp range[107] This can easily be obtained by exploitation

of proper connections and high-quality cables connecting the equipment to the Device

under Test (DUT) In this work the DC measurements were carried out using an Agilent

B1500A Parameter analyser whose description was covered in Chapter 2

As was discussed in Chapter 3 the GaAsAlAs ASPAT diodes have been

fabricated with different mesa areas between 2times2microm2 to 100times100microm

2 but the smallest

size obtained with good I-Vs was 4times4microm2 In this section the focus will be on how the

extracted data can be expanded further for empirical modelling Figure 51 shows typical

results for measured ASPAT diodes with various dimensions to check for their

uniformity According to our standard procedure the DC measurement has to be

conducted prior to the RF to ensure the diode is in fully working order as this will later

save a lot of time during RF characterization

Figure 51 IV Characteristics of measured ASPAT diodes for emitter sizes of 4x4um2 6x6um

2 and

10x10um2 Note the good scalability

1E-10

1E-09

1E-08

00000001

0000001

000001

00001

0001-2 -1 0 1 2

Cu

rre

nt

De

nsi

ti (

Amicro

m2)

Voltage (V)

4x4microm^2

6x6microm^2

10x10microm^2

150

Figure 51 above demonstrates the IV Characteristics of measured GaAsAlAs

ASPAT diodes (XMBE304) for emitter sizes of 4times4um2 6times6um

2 and 10times10um

2 This

sample was processed using the dielectric bridge technique developed in this work It

can be observed that current per unit area for each dimension fits and scales to each

other The scalability of each diodes measurement is very important to ensure no process

related issues are hampering the devicersquos proper operations This also confirms that the

diodes are completely functional and can be used for the next stage of measurements

The advantage of having an excellent scalability of those diode sizes is that a prediction

of smaller emitter size can be made

This type of IV characteristic shows asymmetric behaviour which results from

the unequal spacer lengths of the device This behaviour is very useful for detection

application as it obeys a square law model The square law predicts that the current is

proportional to the square of the applied bias

119868 = 1198861198812 119908ℎ119890119899 119881 gt 0

119868 = 0 119908ℎ119890119899 119881 lt 0

(51)

To extract the first order effects of ASPAT diodes DC measurements which

result in asymmetric I-V characteristics are analysed The slope of the non-linear region

is used to determine the junction resistance (Rj) which is obtained from the first

derivative of voltage versus current (dVdI) The expression of Rj is given by

119877119895 =

120597119881

120597119868

(52)

In order to understand the relationship between Rj and diode sizes of the ASPAT the IV

characteristic for each diode displayed in Figure 51 is used to extract the Rj This has

been done by using the expression in equation (52) above and their response is plotted

against bias as displayed in Figure 52

151

Figure 52 Junction resistance versus voltage

As can be seen in the Figure 52 above the Rj for each device decreases strongly

when the voltage increases At zero bias the 4times4um2 devices show the highest Rj value

followed by 6times6um2 and 10times10um

2 devices The junction resistance at zero bias obtained

from the 4times4um2 diode is around 86KΩ while reducing by a third for the 6times6um

2 and

10times10um2 diodes with Rj of 27KΩ and 10KΩ respectively A diode with a smaller

forward current under the same applied voltage will exhibit a larger Rj For a good

millimetre wave detector a device with a large value of Rj is desirable since it will

provide high detection sensitivity

The slope at the IV characteristic contributes to an important parameter that is

commonly used by electronic manufacturers to describe diode specification namely the

video impedance (RV) which is also known as the non-linear resistance The RV which is

extracted from the real part of the diode small signal impedance is highly dependent on

the DC bias current and only weakly depends on the series resistance of the diode (RS)

Therefore the video impedance is given by

119877119881 = 119877119895 + 119877119878 (53)

Where RS is the series resistance of the diode whose value is normally very small and

does not contribute much to the whole slope and hence RV is dominated by Rj The RV

changes in behaviour if any DC current is flowing through the diode Practically small

DC current in the range of 1 to 10 microAmp or total zero bias is used to maintain the

appropriate RV value (1-2KΩ to several MΩ) RV will also determine the voltage

-10

10

30

50

70

90

110

130

150

-01 0 01 02 03

Rj(

)

Voltage (V)

6x6um^2

10x10um^2

4x4um^2

152

sensitivity of the whole detector circuit This will be explained further in the next

chapter In the case of a detector with an amplifier RV of the diode acts as the RF

impedance which needs to be matched with the video amplifier ( impedance looking into

the diode from the amplifier)[108 109]

The quotient of the second order derivative to the first derivative

((d2IdV

2)dIdV)) when calculated from the whole I-V characteristic translates directly

into a curvature coefficient (k) This is the most commonly used figure-of-merit to

quantify diode nonlinearity at zero bias Figure 53 below shows the variation of k with

bias and more importantly the zero bias rectifying action for device sizes of 4times4 um2

6times6 um2 10times10 um

2 This parameter which represents the small-signal rectifying

action of the diode will affect the performance of the detector (voltage sensitivity)

Detailed discussions on how this parameter effect the detector performance will also be

discussed in the next chapter

Figure 53 The rectifying action of the diode at zero voltage which gives the ASPAT high sensitivity

near zero bias detection

Figure 53 above shows calculated curvature coefficient of the measured I-V

characteristics from the same diodes shown in Figure 51 The highest k value is

obtained from the diodes with size of 4times4um2 followed by 6times6um

2 then 10times10um

2

The curvature coefficient decreases sharply as the bias increases for each diode This can

be attributed to a significantly increasing number of electrons that tunnel through the

thin barrier which were accumulated in the 2DEG formed in the intrinsic spacer region

-5

0

5

10

15

20

25

30

-001 004 009

Cu

rvat

ure

Co

effi

cien

t(V

-1)

Voltage (V)

k(10x10 um^2)

k(6x6 um^2)

k(4x4 um^2)

153

An ASPAT diode with a smaller size will have a larger Rj with a corresponding smaller

current under the same bias condition and hence will demonstrate a larger k value In

this calculation the curvature coefficient at zero bias obtained from 4times4um2 6times6um

2

and 10times10um2 diode is 23V

-1 17V

-1 and 16V

-1 respectively

A summary of the ASPAT diodes parameters obtained from measured I-V

characteristics that have been translated into first and second order differentials are

gathered in Table 51 below and compared to other diodes in the literature

Table 51 Performance of the ASPAT diode obtained from measured IV characteristics in this work

Sample Rj(Ω) k(1volt)

ASPAT 10times10 microm2 10K 16

ASPAT 6times6 microm2 27K 17

ASPAT 4times4 microm2 86K 23

Ge Backward diode 182K[110] 159[110]

InGaAs Backward diode 154[110] 23[110]

Sb Backward diode 5K[111] 47[111]

Si-Backward diode 135K[112] 31[112]

PDB 15K[8] -

AlGaAs SBD 20-100K[113] 34-38[113]

GaN HBD - 16[114]

From Table 51 above it is clearly that the ASPAT diode has a comparable value of Rj

and k to existing detector diode in the research community and in the commercial

market Based on literature of each diodes stated in the table the key to obtaining a high

value of k at zero bias is to minimize any forward tunnelling current Furthermore the

largest ASPAT diode used here (10times10microm2) has very close performances to that of a

commercial diode ie discrete Ge backward diode (ref[110]) where both Rj and k value

are close to each other

53 RF Test Fixture Theory and Experiment

RF measurements differ from DC measurement as they are more complicated

and it is necessary to comprehend the basic measurement principles to achieve

meaningful data This is obligatory especially for on-wafer RF characterization and

154

analysis to attain precise results Most of the electronics component measurements

which have input and output for instance antenna amplifier cables etc are based on a

two-port network configuration The characteristics which can be extracted from these

components are usually used to define their impact on a more complicated system

The performance of the two-port network can be described by a few parameters

ie scattering (S-Parameter) admittance (Y-Parameter) Impedance (Z-parameter) and

Hybrid (H-Parameter) However the S-parameter approach is favoured for high-

frequency measurements as it is relatively easier to characterize the microwave

performance and is able to convert to other parameters when necessary The advantage

of S-parameters is that they can straightforwardlydirectly convert into other two-port

parameters as mentioned above in term of currents and voltages[115] In fact to obtain

the device capacitance the appropriate S-parameters needs to be transformed into Y-

parameters using specific equations Furthermore the devicersquos cut-off frequency can

also be obtained when S-parameter measurements are performed over a wide frequency

range

531 On-Wafer Measurement and Small Signal One-Port Characterizations

In this work the arrangement of the RF measurement setup is assumed to be a

linear system as small voltage amplitude signals are used this means that the signals

have only a linear effect on the network without any gain compression or attenuation

The assumption is still acceptable even though the typical ASPAT is characterised as

non-linear in nature because it is a passive device which will act linearly at any input

power level

Generally the S-Parameter measurements on a diode can be adequately and

suitably performed using a one port measurement The technique used to characterise the

output is similar to the two ports technique but only incident and reflected waves are

used to characterise the input and output ports of the device Essentially this is because

the ASPAT has only two terminals and it is a passive device like other diodes

Therefore the analysis will revolve around the S11 parameter Figure 54 below show the

S11 is a ratio of reflected wave to the incident wave

155

11987811 =

119877119890119891119897119890119888119905119890119889

119868119899119888119894119889119890119899119905=

1198871

1198861 119908ℎ119890119899 1198862 = 0

(54)

A VNA as described in Chapter 2 is used to measure the ASPAT diodes This

powerful equipment is able to measure S-parameters up to 40 GHz To conduct accurate

S-parameter measurement at the diode the measurement setup must be calibrated prior

to the actual measurements taking place

54 Device Calibration

541 Open and Short De-Embedding Technique

Further calibration to be made involves anything related or attached to the

device The co-planar waveguide (CPW) bond pad and interconnect line that are

attached to the intrinsic diode are the main contributors of the errors also are required to

be calibrated In general the bonds pad could generate a capacitance (parasitic) in

parallel with the intrinsic diode and its contribution depends on the size of the bonds pad

as well as the operating frequency Meanwhile the CPW and interconnect line may

cause a parasitic inductance in series with the diode

The method that is used to get rid of this parasitic is called de-embedding and the

most common technique to realise it is by introducing OPEN-SHORT structure[116]

This method is based on a lumped-elements model Parasitic elements of the diode

De-Embedding Structure

Incident wave

Reflected wave

One-port device

a1

b1

Figure 54 One port S-parameter measurements

156

equivalent circuit correlate directly to the access section of the CPW hence can be

derived from de-embedding structures The aim of the de-embedding technique is to

represent these parasitic elements so that the one-port characteristic of the actual diode

can be determined

The two types of de-embedding structure OPEN and SHORT are conventional

techniques that are widely used in this study The design of all structure must be

identical (in size) to the device to avoid any discrepancy It is very simple to design all

these three structures for example open structures are obtained by eliminating the diode

layout and keeping the bond pad layer only The short structure just adds a bridge and

ensures ground and signal pad are connected to each other Through structures are

realised by disconnecting both ground pad and leaving the signal pad to connect to each

other

To gain more accuracy this external effect must be removed by the implementation of

de-embedding structures on the same tile as the actual device Figure 55 shows the

fabricated de-embedding structures used in this study

Figure 55 Fabricated open short de-embedding structure and real device (zoomed) use for RF

calibration and measurements (Note Images are not to scale)

In summary the de-embedding which is used to extract out the parasitic elements

from entire single diode measurement is a very important step as normally on-wafer

measurement requires coplanar waveguide (CPW) to access the diode structure (active

region) The CPW will have some effects which will disturb the accuracy of the device

characteristics

157

55 S-Parameter Measurement Result and Analysis

This section will only present RF measurement results after all VNA setup and

calibration were performed The S-Parameter measurements were carried out on ASPAT

diodes at five different DC biases from -2 to 05 volt with a sweep frequency from

40MHz to 40GHz using a calibrated VNA and the input power was fixed at -30dBm

The measurement procedure as described in Chapter 2 was performed on the device

(on-wafer) equipped with the appropriate bond pads This is important to ensure the

results obtained are valid The reason for using different biases is to find at what voltage

the device capacitance is fully depleted This is also very important in determining the

cut- off frequency of the devices

In this research two phases of the experiment on the S-parameter measurements

were carried out The first phase is to qualify the process flow ie for manufacturability

and repeatability which can be obtained from the consistency of the result The S-

parameter measurements taken on the same wafer dies are repeated several times on

different GaAsAlAs ASPAT diodes There are three different tiles taken from 3

different wafers namely XMBE304A XMBE304B and XMBE304C carried out in

this experiment The repeatability tests are done mostly on the large devices (15times15microm2

up to 100times100 microm2) and the results are analysed based on the reflection coefficient (S11)

on Real and Imaginary measurements

In the second phase the measurement is toward producing devices that can

perform at high-frequencies This can be realised by utilising small emitter size devices

(4times4microm2 6times6 microm

2 and 10times10 microm

2) The measurement results of these devices will be

used to build the equivalent circuit models while both the intrinsic as well as the

parasitics of the device will be evaluated Hence all the values obtained from these S-

parameter measurements will be used to design the device that can be used in

millimetres-wave applications As can be seen in Figure 56 below the extracted S-

parameter measurement results comprise of a reflection coefficient (S11) for real

imaginary and Smith chart for XMBE304A While these measurement results are

extracted at zero bias voltage the other bias voltages will be used to extract the

capacitance This will be described in the final section of this chapter

158

551 Diode to diode uniformity

In order to study within tile uniformity and reproducibility statistics of the RF

performance five devices of different mesa sizes in the same tile (XMBE304A) were

measured at zero bias and represented in term of Real and Imaginary reflection

coefficient (S11) the uniformity check is carried out at three different frequencies step

under 15GHz since the cut-off frequency for these big devices is relatively low at about

~20GHz on average The variation of the reflection coefficient is taken from the

percentage of the (standard deviationmean values) for all five diodes from this run The

following figures show Real and Imaginary S11 of large mesa area ASPAT diode from

15times15microm2 up to 100times100microm

2 which are represented by lines graph in a few different

colours

Figure 56 Real S11 RF measurements for (XMBE304A) were done for four devices from

15times15microm2 to 100times100microm2 device sizes for within-wafer uniformity check

Figure 57 Imaginary S11 measurements (XMBE304A) to qualify the repeatability and

reproducibility were done on four devices from 15times15 microm2 to 100times100 microm2 at zero bias

159

The variance data extracted from the graph (Real S11) above for each device within-

wafer (device to device) uniformity study is summarize in Table 52 below

Table 52 Device to device uniformity check for large ASPAT diode

Device Size 100times100 microm2 50times50 microm

2 30times30 microm

2 20times20 microm

2 15times15 microm

2

Variation 5GHz 181 115 405 151 145

Variation 10GHz 106 133 522 424 293

Variation 15GHz 119 198 281 76 509

The majority of diodes show that the variations of S11 measurements are below

3 and only a few are below 8 These finding still can be considered as good for

manufacturing control since absolute I-V characteristics reported in [63] is set by

designerrsquos specification to be not more than plusmn10 variation Further extensive RF

measurements were carried out by the research collaboration with the University of

Cambridge on the same GaAsAlAs ASPAT diodes wafer [117] In their study they

focused on 50times50 microm2

mesa size 17 of diodes were chosen to be measured The study of

uniformity of RF characteristic only focused on frequencies below 20GHz The same

approach was used to get the variation of the reflection coefficient for all 17 diodes but

this work was carried at four different frequencies Table 53 shows the zero bias S11

result for four different frequencies and standard deviation of the devices

Table 53 The variation of reflection coefficient (S11) for GaAsAlAs ASPAT diode at four different

frequencies[117]

Frequency (GHz) 5 10 15 20

Variation () 197 243 26 276

From the results the variations of 50times50 microm2 mesa sizes measured in-house and

at the University of Cambridge are comparable with all variations showing good

uniformity ie recording variations below 3 This indicates that the RF performance

of the GaAs AlAs diode is valid and reproducible and is thus considered as a good

achievement for manufacturing Once the reproducibility and repeatability of the large

devices showed stable results the fabrication process then continued to obtain smaller

emitter size for work at high-frequencies

160

552 Wafer to wafer uniformity

Other RF measurements were conducted on sample XMBE304B which was

fabricated in-house using the same process steps but the only difference from

XMBE304A was the use of SiN3 as dielectric In this run three different mesa sizes

were measured (15times15mmicro2 20times20microm

2 and 30times30 microm

2) and Real and Imaginary S11

plotted against frequency The RF performances of both samples are gathered in one

graph as shown in Figure 58 below

Figure 58 Wafer to wafer uniformity test on 1st and 2

nd RF measurement(XMBE304B) to qualify

the repeatability and reproducibility were done for 30times30microm2 to 15times15 microm

2 device sizes (Real and

Imaginary) Note blue colour is XMBE304A and red colour is XMBE304B

For this wafer to wafer uniformity study four diodes with three different sizes as

specified previously were measured from sample XMBE304B and four diodes from

previous measurements of XMBE304A The blue line in Figure 58 represent

measurement result of real and imaginary for sample XMBE304A while the red line

161

represents XMBE304B The uniformity data is compared at three different frequencies

and gathered in the table below

Table 54 Wafer to wafer uniformity check on XMBE304A and XMBE304B

Device Size 30times30 microm2 20times20 microm

2 15times15 microm

2

Wafer A vs wafer B variation 5GHz 305 31 1 314

Wafer A vs wafer B Variation 10GHz 352 344 329

Wafer A vs wafer B Variation 15GHz 321 376 359

As can be seen in the Table 54 above the wafer to wafer uniformity is rather

large (30) on average The main reasons being that sample XMBE304A was

processed by utilizing S1805 as a dielectric layer while sample XMBE304B used

Si4N3 Although the process steps are similar for both wafer processing the use of

different dielectric layer will influence the diode parameters especially resistance and

capacitance as the dielectric constant for each materials is different Secondly the wafer

processing is not run concurrently at the same time thus the moisture and temperature in

the clean room might differ for both processing Although the wafer to wafer uniformity

test for this run might not be favourable for manufacturing tolerance at least the use of

different dielectric layer shows some significant result in term of capacitance resistance

effect to the GaAsAlAs ASPAT diode

553 Small devices RF measurements

The first objective of this study was to make smaller size mesa devices ie

1times2microm2 1times3 microm

2 2times2 microm

2 and 3times3 microm

2 However for GaAsAlAs ASPAT type this is

difficult to achieve in practise These issues were discussed in detail in Chapter 3

Hence the smallest emitter size that yields repeatable and reproducible results was

4times4um2 The final measurement which was done on sample XMBE304C focused on

small devices The measurements were done on four devices two with the diode bond

pads sitting on substrate (GaAs SI) and the other two sitting on dielectric layer (Si4N3)

Figure 59 below shows three measured results obtained from sample XMBE304C

using the 3rd

Gen Mask

162

Figure 59 RF measurement performed on device sizes of 10times10microm2 6times6mmicro

2 and 4times4microm

2 (Real and

Imaginary) Note that green red and blue colour represents 4times4microm2 6times6mmicro

2 and 10times10microm

2 diodes

respectively

Figure 510 RF measurement performed on device sizes of 10times10 microm2 6times6 microm2 and 4times4 microm2

(Smith Chart) Note that green red and blue colour represents 4times4microm2 6times6mmicro2 and 10times10microm2

diodes respectively

30MHz

40GHz

163

As can be seen in Figure 59 (Real Imaginary) and Figure 510 (Smith Chart) are

obtained from measurement of four diodes in the same tile The diode to diode

uniformity that is sitting on the same platform obtained at 35GHz frequency in this run

on average is ~15 25 and 1 for 4times4microm2 6times6mmicro

2 and 10times10microm

2 diodes

respectively On the other hand the uniformity between diode to diode sitting on

dielectric and substrate is quite high due to different capacitance value of devices on

average ~7 are attained from three different sizes of diode

From the measurement results above the Real S11 measurement of four different

sizes show the same trend for each frequency At low frequency resistances for each

diode is high as the S11 value is large At intermediate frequency the values drop

tremendously for big devices (30times30 mmicro2 and 20times20 mmicro

2) ie in Figure 57 At high

frequency all diode reach saturation limit as the value are constant Small devices

(4times4mmicro2 6times6mmicro

2 and 10times10microm

2) ie Figure 59 show S11 values that are higher than

those of large devices as smaller emitter diode have larger resistance value

The imaginary S11 value also shows the same trend as for big devices However

for small devices in this run (4times4mmicro2 6times6mmicro

2 and 10times10microm

2) the S11 value keep

dropping toward negative values at increasing frequencies This indicates that bigger

devices with positive value at high frequency are more capacitive than the smaller

devices It is worth mentioned that the capacitance and inductance values for device

sizes of 4times4microm2and 6times6microm

2 come from the CPW layouts and these are dominant while

for device sizes of 15times15 mmicro2 and above the device capacitance itself is dominant

The Smith Chart shows the reflection coefficient (S11) as a function of the

applied frequency (30MHz to 40GHz) All measurements from each mesa size follow

unique impedance circle which is that most of the lines are at the lower right outer ring

This means that the diode capacitance value is frequency dependent For the case of

10times10microm2

devices these impedance circles are mostly toward the outer ring meaning a

higher capacitance than the other two mesa dimensions All the device constantly follow

the outer ring without crossing any real axis at any frequency point meaning that the Cj

is not shorted at the maximum 40GHz measurement frequency (not reached cut-off

frequency) Therefore the entire small GaAsAlAs diodes in this run have capability to

work in the millimetre wave frequencies range

164

56 Extracting RF models of ASPAT at Zero Bias Voltage

The methodology used in the S-Parameters measurement for high-frequency

analysis must ensure that the derivation of the equivalent circuit corresponds to their port

characteristics In other words the component representing the ASPAT in the equivalent

circuit model must have physical significance otherwise the circuit will be meaningless

The fabricated ASPAT diodes as discussed in Chapter 3 have the cross section shown in

Figure 511

There are two main components that can be extracted from the fabricated

ASPAT depicted above ie intrinsic and parasitic The intrinsic refers to the main

structure of the diode itself and are represented by three bias dependent elements

namely Junction Resistance (Rj) Junction Capacitance (Cj) and diode Series Resistance

(Rs) The parasitic is the elements related to the bond pad of the anode and cathode as

well as interconnects They are represented by parasitic inductance (LP) resistance (RP)

and capacitance (CP)

The diode parameter extraction is different from the three terminal devices

(FETs) in the sense that FETs are a kind of direct extraction in which all the elements in

the transistor have linear functions to the port characteristics ie S-parameter Y-

Parameter Z-Parameter and can easily be solved by the matrix calculation method for

those particular parameters[118] The same extraction method cannot be applied to the

diodes because its elements will embroil with each other Therefore only one method is

used to extract the diode element which is optimisation by tuning the initial value toward

the measured S-Parameter values

Figure 511 Cross-sectional of fabricated ASPAT diode and the corresponding equivalent circuit

model

165

The strategy used to model the ASPAT is based on an initial fitting value of the

lumped elements to the extracted value from measurement on three S-parameter graphs

(real imaginary and Smith chart) for the reflection coefficient (S11) The refinement is

accomplished by optimisation and fine tuning of the values which result in minimum

error between extracted and modelled values Figure 517 (on page 153) shows fitted S-

parameter result for extracted and model numbers with each one fitted in a single line as

an example However to achieve this excellent fitting key prior steps have to be used

de-embedding fitting the intrinsic value and optimisation

561 Extraction of ASPAT parasitic element

Once the S-parameter measurements achieve stability repeatability and

reproducibility for each measurement in term of S11 results as mentioned above the

results of the de-embedding structure which had been measured prior to the device

structure are then extracted to form a well-defined equivalent circuit In order to build

and analyse the equivalent circuit firstly the measured data is imported into the ADS

software prior to any fitting This can be realised via the ldquoStart The Data File Toolrdquo

features provided by this particular software When successfully imported the data is

read by the function S2PMDIF (These files are a natural extension of two-port S-

parameter Touchstone files) as depicted in Figure 512 below

Figure 512 The S-parameter Touchstone file is used to read the measured files

166

For the open and short techniques after de-embedding the equivalent circuit

model which is represented by mainly a capacitor and an inductor is built The open

structure requires resistance and the capacitance values of 20KΩ and 26fF respectively

connected in parallel to be well fitted to the real imaginary and Smith chart (S11) output

On the other hand for the short structure the Real imaginary and Smith chart (S11) have

to satisfy the values of resistance and inductance elements of 1Ω and ~47pH respectively

connected in series These values strongly rely on the bond pad or CPW dimension and

length The Equivalent circuit models and fitted data as well as measurement can be

seen in Figure 513 and Figure 514 below

Figure 513 Equivalent Circuit Model of Open CPW or de-embedded structure

Figure 514 Equivalent circuit model for short de-embedded structure

To satisfy the equivalent circuit a self-consistence method introduced by

Ren[119] is utilised This approach accurately extracts the CPW capacitance (Cpad) and

inductance (Lpad) as well as intrinsic Junction capacitor (Cj) which is attained from the

one-port S-parameter measurements Therefore the pad capacitance introduced by the

self-consistence method for the open structure can be expressed by

167

119862119875 =

119868119898(11988411119874119901119890119899)

120596

(55)

Lpad which represents the short structure is given by

119871119875 =

1

120596(119868119898 (11988411119874119901119890119899minus 11988411119878ℎ119900119903119905

))

(56)

Here Y is the admittance parameter (Y-parameter) converted from the S-parameter

measurement data and ω is the angular frequency The extracted measurement data

represented in the equivalent circuit fits with the simulated data in three S11 graphs as

can be seen in Figure 515 below From the Smith Chart it can be clearly seen that both

open and short S11 results are on the circumference which means the resistance of the

short structure is very small while in the open it is very large Additionally the

calculated Cpad and Lpad using Equations 55 and 56 above produce results similar to

those obtained in the equivalent circuit model for the open and short structure The

values are ~25fF and ~45pH respectively These data completely verify and validate

both results

Figure 515 Smith chart representative S-parameter measurement for short (left) and open (right)

CPW The blue lines represent simulated data and the red is measured data

Short

Open

168

562 Extraction of ASPAT intrinsic elements

Once the parasitic elements are determined it is easy to build a complete ASPAT

equivalent circuit The ASPAT is not like other tunnelling diode which their equivalent

circuit models widely studied ie RTD [120] IMPATT and PDB The only literature

which reports ASPAT equivalent circuits can be found in [15] and other RT Symersquos

journal paper[16] Fortunately its equivalent circuit model is not much different

compared to other diode video detectors Thus other literature which is based on

Schottky diode equivalent circuit model used for detector application can be referred to

The simplest form of ASPAT equivalent circuit and other video detectors intrinsically

consist of junction capacitance (Cj) series resistance (RS) and junction resistance (Rj)

First and foremost to extract the equivalent circuit one must know the theory behind

each parameter that is developedbuilt as a spine to become a complete element This is

vital to ensure the equivalent circuit is correct In the case of the ASPAT Cj is predicted

from a simple fully depleted parallel plate capacitor approximation which was discussed

previously in Eq (230)[15] Additionally for the S-parameter measurements the Cj can

also be validated by the self-consistence method mentioned earlier and thus can be

expressed by

119862119895 =

[

(1

120596)

1

1

119868119898 (11988411119905119900119905119886119897minus 11988411119874119901119890119899

)+

1

119868119898 (11988411119874119901119890119899minus 11988411119878ℎ119900119903119905

)]

(57)

This approach helps to verify both the fully depleted parallel plate capacitor in S-

parameter measurements The basic component which is responsible for the ASPAT

series resistance RS was discussed in detail in Chapter 2 RS and Cj are key contributors

to the high-frequency operation as expressed by the device cut-off frequency Equation

(58) below

119891119888119906119905minus119900119891119891 =

1

2120587119877119904119862119895

(58)

The R and C parameters must be kept as low as possible in order to obtain high cut-off

frequencies for millimetre wave applications From the fabrication point of view Cj can

169

be reduced by making as small a diode emitter size as possible while for RS reducing

the D gap is of paramount importance as it dominates the series resistance The ASPAT

contact resistance in the electrodes (contacts between metal and semiconductor) can be

reduced by using high doping in the ohmic layers

The junction resistance (Rj) of the ASPAT is taken from the 1st derivative or

slope of the current-voltage characteristics Normally the value of Rj is very large

(several kilo Ω) compared to Rs The small signals ASPAT equivalent circuit built with

intrinsic and extrinsic components is shown in Figure 516 below while the fitting

results is shown in Figure 517 and Figure 518

Figure 517 Excellent fitting between measured (red colour) and simulated (blue colour) for S11

(Real and Imaginary) results for various small device designs

Rj

Cj

Cpad Rpad

Rs Lpad

Figure 516 Equivalent circuit of the ASPAT diode

170

Figure 518 Excellent fitting between measured (red colour) and simulated (blue colour) for S11

results (Smith Chart) for various small device designs

The equivalent circuit that was built for the ASPAT is taken from sample

XMBE304C with emitter dimensions of 4times4 microm2 6times6 microm

2 and 10times10 microm

2 These

devices are expected to work in the millimetre-wave region and have cut-off frequencies

(intrinsic) of ~650GHz ~200GHz and ~100GHz respectively

Table 55 Comparison between calculated (fully Depleted) and extracted (different biases) values

from equivalent circuit parameters for different ASPAT mesa sizes at zero bias voltage

Parameters 4times4microm2 6times6 microm

2 10times10 microm

2

Calculated

(fully

depleted)

Extracted Calculated

(fully

depleted)

Extracted Calculated

(fully

depleted)

Extracted

0V -05V -1V 0V -05V -1V 0V -05V -1V

Cj(fF) 879 23 7 6 198 55 148 139 549 171 486 46

Cpad(fF) - 15 15 15 - 15 152 152 - 15 15 15

Lpad(pH) - 45 43 42 - 50 473 473 - 51 51 46

Rj(KΩ) - 90 833 522 - 35 392 392 - 12 125 13

Rs(Ω) 99 11 11 11 67 95 8 7 41 95 45 37

fcut-off

Cj(GHz)

1828 629 2066 241

1

1208 192 1344 163

5

710 107 728 935

fcut-off

Cj+Cp(GHz)

- 380 658 688 - 151 663 781 - 98 556 705

171

The focus in this study is purposely to build ASPATs as zero bias detectors that are able

to work in the millimetre and sub-millimetre frequency range therefore all the

parameters which are obtained from equivalent circuit were extracted at zero bias

voltage Theoretically the calculations which are derived from both self-consistence amp

theory can only be solved for fully depleted device capacitance (using a parallel plate

configuration) Hence both extracted and calculated results are compiled in Table 55

Noticeably the calculation can only produce the intrinsic parameters of the

ASPAT for fully depleted capacitance On the other hand both intrinsic and extrinsic

parameters of GaAsAlAs ASPAT are obtainable from extraction and thus help to

determine at what bias the diode is start to deplete The junction and series resistance (Rj

and RS) of each dimensions shown in Table 55 above were achieved by fitting the

elements of the equivalent circuit with the three measured S11 graphs whereas the Cpad

and Lpad were extracted by utilising the self-consistent method from the S-parameter

measurements which is fitting the de-embedding structure Additionally the Cj values

are obtained via fitting the measured S11 data and employing the self-consistence

approach Results obtained from both techniques are identical

At zero bias all extracted junction capacitance from each device sizes are very

different from the calculated one while the extracted series resistance are closer to the

calculation This means that Cj is a highly voltage dependent parameter and Rs is

voltage independent but solely dependent on device structure and material used to

fabricate it The extraction at -05V and -1V shows that the values of junction resistance

is changing for most of the devices which means this parameter also rely on bias voltage

as discussed earlier in Section 52

The cut-off frequency for each calculated devices are near the THz range even

for the 100microm2 emitter area However with the introduction of parasitics elements ie

pad capacitances fcut-off is degraded tremendously Therefore it is important to make sure

all the intrinsic elements have optimum values so that the target operating frequency of

the ASPAT diode can be met Due to this it is advisable to operate the devices at no

more than 13 of fcut-off when designing detector systems

The parameters extraction at -05 and -1V also show that Cj values are closer the

calculated ones which means the ASPAT diode is reaching full depletion

172

563 Capacitances -Voltage (C-V) Extraction

Theoretically the junction capacitance of the ASPAT is calculated from the fully

depleted formula 119862119895 = 휀119900휀119903119860119889 which was also discussed in Equation (230) in Chapter

2 in page 43 Its value depends on the change of voltages to depletion at the emitter

contact[15] and make it one of the voltage dependent parameter for the diode[121]

Therefore the C-V characteristic of the GaAsAlAs must be precisely extracted

In practice the capacitance is difficult to measure due to the very low resistance at zero-

bias However alternatively it can be measured and extracted by applying different

voltage and identifying the point at which there is change which essentially represents

full-depletion Apart from these it can also be extracted from S-parameters measurement

which is then converted to Y-parameters A C-V characteristic of the GaAsAlAs

ASPAT from XMBE304C is extracted and plotted as depicted in Figure 518 below

Figure 519 Capacitance extraction showing full depletion at -025 volts (modelled capacitance vs

Voltage)

From the graph shown in Figure 518 the capacitance is extracted at eight

different biases for 4times4 microm2 6times6 microm

2 and 10times10 microm

2 The devicersquos junction

capacitance for each dimension increases and reaches a maximum value at 025V There

are additional quantum capacitance effect which comes from an increase in the negative

charges in the 2DEG region (when band bending happens creating an accumulation

0

50

100

150

200

250

-2 -15 -1 -05 0 05

Cap

acit

ance

(fF

)

Voltage (V)

Cj(4x4microm^2)

Cj(6x6microm^2)

Cj(10x10microm^2)

173

region at the barrier) This charge is imaged by the positive charge in the whole

depletion region Increasing the voltage toward positive values leads to a lowering of

the AlAs barrier and thus allowing thermionic emission to take place after certain bias

values leaving only the depletion capacitance and making the quantum capacitance

negligibly small

In the reverse bias case the device junction capacitance reaches a saturation

(fully depleted capacitance) at a voltage of -025V and remain constant up to -2V If the

reverse bias voltage is increased further the ASPAT may reach breakdown Therefore it

is important to know how far the diode can withstand applied reverse bias to ensure it

can still give full performance

57 Conclusions

In this chapter scalable DC characteristics of GaAsAlAs ASPAT diode derived

from three different emitter sizes of 4times4 microm2 6times6 microm

2 and 10times10 microm

2 was

demonstrated The current density obtained at zero bias is several microAmicrom2 These allow

1st order differential effect to exhibit high value of Junction resistance (Rj) at zero bias

However Rj is highly bias dependent The 2nd

order differential effect on IV

characteristics display a high value of curvature coefficient leading to high voltage

sensitivity when applied in millimeter wave detector applications These two parameters

are vital in the design of millimeter wave detectors and especially those operating at zero

bias

Subsequently RF measurement up to 40GHz of uniformity study for both within

wafer and wafer to wafer variance were undertaken An average uniformity below 7

was obtained for within wafer study on large device area ( 15times15 microm2 to 100times100 microm

2)

while for small device area ( 4times4 microm2 to 10times10 microm

2) a smaller 3 uniformity variance

was achieved in average However for wafer to wafer study the variant uniformity was

quite high at around 30 on average for relatively large device (15x15 microm2 to 30x30

microm2) This was mainly attributed to different dielectric layers used in the process flows

of the sample rather than fundamental MBE control of the AlAs barrier thickness

174

It was demonstrated in this chapter that careful on-wafer RF measurements of small

size GaAsAlAs ASPAT diodes allow accurate device parameter extraction of both

extrinsic and intrinsic parameters The extrinsic parameters are namely pad capacitance

and inductance with obtained values of 26fF and 47pH respectively These values were

obtained from de-embedding structure fabricated on the same tile as the real devices

The intrinsic parameters such as junction capacitor junction resistor and series

resistance had different values according to device dimensions The smallest zero bias

value of Cj obtained from 4times4 microm2 diodes was 23fF ensuring a high cut-off frequency of

380GHz and hence in the next chapter a 100GHz detector will be presented working at

slightly less than 13 of this cut off frequency

The C-V data extraction confirmed that the fully depleted capacitance started to

happen at around -025V The maximum junction resistance occurs at +025V largely

caused by the depletion region and an additional quantum capacitance effect CQ This

effect is strongly related to the size of a 2DEG which occurs under forward bias (01V to

025V) and can be reduced by having a smaller thickness AlAs barrier

175

6 MILLIMETRE WAVE GAASALAS ASPAT DETECTOR

DESIGN USING ADS

61 Introduction

The ASPAT diode having features of non-linear IV characteristics at zero voltage

make it useful for signals rectification ie detector and mixer for millimetre-wave

applications Additionally ASPAT diodes have a range of advantages such as large

dynamic range strong temperature insensitivity etc[15]over other rectifier diode This

makes ASPAT an appropriate candidate in RF detection applications Since 1940 the

only two terminals device that has been the workhorse for RF applications is the

Schottky Barrier Diode (SBD)[9] In its earliest form the SBD was built based on a

point-contact device which could not perform at high frequencies It was then developed

to work at higher frequencies by exploitation of epitaxial structures [10] and to date the

SBD remains the mainstay of two terminal devices that are able to work in the

millimetre and submillimetre-wave regions However as discussed in Chapter 4 the

performance of SBD is degraded at extremes of temperatures and these circuits

employing SBDs require temperature compensating circuitry Thus there is additional

complexity associated with technologies and applications related to SBDs

Before this work was carried out no model for the ASPAT diode as detector had

been available or developed especially using the empirical modelling ADS software

When the ASPAT diode was first introduced its function was conceptually explained it

was then built and tested to compare with other microwave detectors at X-band

frequency (95GHz) The comparisons were made in terms of detector parameters ie

sensitivity dynamic range temperature dependence etc[15] These early works lead by

RT Syme et al supplied the basic knowledge to model the ASPAT diode as a zero-bias

detector for mm-wave frequency gt100GHz For the SBD many models and equivalent

circuit approaches have been reported [122 123] The modelling of conventional SBD is

mostly implemented through fitting the S-parameter curves of the model to the

experimental one This approach is also carried out in this research since it is accurate to

predict the performance of the device under test [124]

176

This Chapter aims to introduce low cost reliable and sophisticated detector design

based on ASPAT diodes which is believed to be able to improvereplace SBD in

millimetre-wave applications especially in imaging The focus was on developing and

establishing an appropriate circuit design that suit the new ASPAT diode for such

applications The detector sensitivity as its key parameter ultimately limits the quality

and acquisition time of the detector In the subsequence section the theory of detection

including both direct and heterodyne will be discussed This is followed by definition of

detector characteristics of interest as well as noise consideration Section 65 present the

main focus of this chapter which is the development of 100GHz ASPAT detectors and

their result will be explained in term of all detector characteristic of interest

62 Detection Theory

Any incoming signal such as RF microwave or mm-wave in the form of envelope

function or single wave can be detected by rectification of the signal using a nonlinear

device ie transistor or diode The input and output signal (RF) signals are normally in

the form of amplitude as a function of time Typically the detector output is a low-

frequency signal known as the video signal which has amplitudes that are proportional to

the square of the input RF signalrsquos voltage amplitude

A complete receiver system as shown in Figure 61 below consists of receiver

antenna and a circuit designed to extract the signal and then amplify it The function of

the receiver system is the converse of the function of the transmitter side At the receiver

side the antenna is used to receive the signal it then conveys the signal to the extraction

circuit for detection as the information-bearing part of the signal (using the nonlinear

device as its heart) The signal is finally amplified to avoid any information strength

decay Additionally in a digital system the output signal which had been processed by

the detector circuit has to maintain an optimum input signal conveyed to in-phase and

quadrature (IQ) demodulators The output signal will go through a low pass filter then

to an analogue-to-digital converter (ADC) and thus a digital baseband output signals

will be produced

177

Figure 61 Block diagram represent a complete direct receiver system

There are two types of millimetre wave integrated circuit (MMIC) used for

detection purposes namely direct detectors and heterodyne detectors The direct detector

MMIC is the simplest circuit used for detector applications and has the simplest way of

extracting the RF information Due to its simplicity the direct detection method is

inexpensive and most attractive method used for measuring power in RF Laboratories

and Industries This detection scheme is also sometimes known as video detection[125]

The simplest way of explaining the detection process is that the incoming RF or

microwave signals depicted in Figure 62 below with an appropriate input power (Pin) is

rectified by using a diode and results in a corresponding output voltage (Vout) A detector

IC designed based on diodes is normally able to rectify very low levels of RF power (lt-

40dBm) then produces an output DC voltage that is proportional to the RF power A

rectifier diode can function at zero bias (which is very good for reducing noise) at very

small DC bias (003mA) and relatively high RF impedance which will produce around

600Ω This will affect the capacitance value and a low capacitance is needed to realise a

high detection sensitivity

Figure 62 The detection process of a single wave through a non-linear IV characteristic of a diode

Detector

Speakerdisplay unit Amplifier

Antenna

Vout

Pin

Tuner Amplifier

178

However this type of detector has a drawback which is itrsquos relatively low signal to

noise ratio Thus it will also rectify any incoming electrical noise at all frequencies and

up to the cut-off frequency (fC) The basics lumped components circuit as shown in

Figure 63 is used to build such detector which consists of Source impedance (Zo)

Rectifier diode (ASPAT or Schottky) wire or pad inductance and capacitance and Load

impedance (RL)

Figure 63 Lumped element illustration of microwave detector circuit

Another type of detector is the heterodyne method which mixes incoming RF or

microwave signal (fRF) with another constant signal produced by a secondary circuit

called the Local Oscillator (LO) The LO frequency (fLO) must be slightly higher than fRF

to enhance the RF signal This mixing between fRF and fLO happens in the nonlinear

device as depicted in the Figure 64 This will produce a signal at a different frequency

called the intermediate frequency (fIF) which can then be amplified and detected as

explained in the previous paragraph Theoretically a basic requirement of the mixer is to

have fIF as efficient as possible while practically the minimum conversion efficiency

obtained is around 20 The main reason for using a mixer is due to the fact that

selective amplifiers at RF frequencies are costly and hard to achieve Hence a mixer is a

good technique as it only convert the signal to a lower frequency in which good

selectivity and high gain can be more effortlessly realised[14] A good mixer diode is the

one that can produce a high cut-off frequency and reduce conversion losses (Lc) A

mixer and detector diode with a low driven input power result in reducing overall noise

figure and thus in the ideal case the fIF amplifier also should have a low noise figure for

better performance The advantage of the heterodyne method is that it has a higher

179

sensitivity compared to the direct detection method this is achieved by producing an fIF

which has a lower frequency than the incoming RF signal[14] Obviously a zero-bias

voltage diode is more favourable to be used in mixer and detector applications

Figure 64 The mixing process where the signals are processed by the non-linear I-V characteristic

to produce an intermediate frequency (fIF=fLO-fRF) when signals at fLO and fRF are applied to the

diode

To have good detection efficiency for both types of detectors the operating

frequency (fO) must be several times smaller than fC In the case of an incoming

maximum modulated signal fM the frequency that can be acquired is in the range of fO

plusmnfM and will normally come with noise The standard method that is used to reduce the

noise is using a filter of bandwidth about 2fM at the centre of fO with the condition that fO

must be smaller than fM (f0lefM) Otherwise it would be difficult to attain However in

most cases fM is smaller than fO and fO is smaller than fC thus this will make the video

impedance (RV) (or nonlinear impedance) very close to the differential resistance of the

diode (at fO) in the equivalent circuit[74]

63 Definition of Characteristics of Interest Detection Parameters in small signal

analysis

Theoretically a transfer function measurement is preferable prior to any empirical

modelling since no assumption can be made due to the detector non-linearity

Furthermore measuring voltage output at high frequency can be very low while

measuring the power incident on the detector is hard to achieve where the linearity of

180

the typical power meter is normally less than 3 over its operating range[126]

Therefore the modelling of detector output voltage vs input power (ie transfer

function) can help to determine both nonlinearity correction and appropriate operating

range for the detector itself

The performance of the diode that is often taken into account is the transfer function

(output voltage Vout versus incident power Pin) and the main parameters that is used to

characterise and determine the quality of any detector diode are the voltage sensitivity

(βV) tangential sensitivity (TSS) dynamic range (P1dB) ie under 1 dB roll off power and

variation of output voltage when examined in extreme temperature situation (ΔV(T))

The voltage sensitivity in small signal analysis can use the approach introduced by

Torrey and Whitmer [9] then βV can be expressed as

120573119881(119894119889119890119886119897) =

119877119895119877119871120581

2(119877119881 + 119877119871) (1 +119877119904

119877119895)

2

[1 +1205962119862119895

2119877119878119877119895

1 + 119877119904 119877119895frasl]

(61)

Where ω is the angular frequency (2πf) Cj is the junction capacitance of the diode

active region RL is the load resistance RV is the video impedance taken from the

expression of Rj + Rs and κ is the curvature efficiency that give small signal rectifying

action of the ASPAT diode which is given by the second order term and itrsquos expression

is

κ =

11988921198681198891198812frasl

119889119868119889119881frasl

(62)

The curvature coefficient or responsivity (κ) is translated directly from the non-linearity

of the IV characteristics of the ASPAT diode for detector application Both RV and κ are

the parameters that can be extracted directly from diode DC measurement as discussed

in Chapter 5

The voltage sensitivity is actually a quantitative relationship between input

power and detector response Meaning that it is a change in signal output over change in

input power Normally output power is measured in Volts and input signal is measured

in Watts Therefore the unit of responsivity is VW[127 128]

181

TSS is referred to the lowest or minimum signal that the detector could detect it is

determined by the diodersquos βV and total noise available in the system (from the diode and

any amplifiers in the detector circuit) For any diode with fO=10 GHz and low noise of

1MHz bandwidth amplifier the TSS is typically less than -55 dBm[74] The TSS equation

is given by

119879119878119878 =

radic[4119884119870119861119879119861(119865119886 + 119905 minus 1)]

119872

(63)

Where M is a figure of merits and is derived based on the expression M= 120573119881 radic119877119881frasl t

denotes the diode noise temperature Fa is the noise figure B T and Y are the amplifierrsquos

bandwidth temperature and power for signal-to-noise ratio respectively For a low-level

video detector ie lt10GHz the sensitivity mainly depends upon three factors firstly on

RF matching structure secondly on the rectification efficiency output impedance and

noise properties of the diode and finally the input impedance bandwidth and noise

properties of the video amplifier at the detector output The RF matching structure

controls the quantity of overall energy at the active junction for rectification The second

factor controls the reaction of the diode to incident microwave radiation and the last

factor will influence the detector sensitivity in general[109]

In practical the Tss is a direct measure of the signal-to-noise ratio of a detector and

is achieved by varying the amplitude of the input pulse (RF signal) until a point in which

the top of the noise level with no signal applied is at the same level of noise at the

bottom level of RF signal It is commonly measured on an oscilloscope as depicted in

Figure 65 below It is defined as the input power at which a signal to noise ratio of 251

is produced[109]

Figure 65 Measurement of Tangential Sensitivity[108 129]

182

The transfer function in many detector diodes is often divided into three sections

Firstly at low incident power secondly at higher input power and finally at very high

power static (continuous) In the first region the detector diode performs as a square-law

detector in which Vout is proportional to Pin This region normally is used to extract the

dynamic range of the diode detector In the second region Vout is approximately

proportional to Vin and this region is known as the linear regime Finally at higher Pin

still the transfer function or response rolls off and thus Vout ultimately become saturated

This roll-off point where Vout has dropped by 1dB below an extrapolation of the

dependence at low Pin is termed the ldquo1dB roll-off pointrdquo and this value is usually in the

range of -11 to 12 dBm[15] Therefore a dynamic range of the detector diode can be

obtained by taking the interval between TSS and 1dB roll-off point (in dBm)

Finally the temperature dependence of Vout for a detector is normally taken from

two extreme points of the temperature (-40C˚ to +80C˚) and thus can be determined

from

Δ119881(119879) = 10 11989711990011989210 |

119881(1198791)

119881(1198792)|

(64)

This Vout variation between -40C˚ and +80C˚ normally expressed in dB

64 Noise Consideration in a Detector diode

The existence of noise in a system limits the accuracy of device performance and

the precision of measurements In a detector system specifically using a diode the noise

which can reduce the sensitivity of signal encryption is called the Noise Equivalent

Power (NEP) By definition the NEP is a noise power density over the detection

sensitivity and it can be exploited to determine the overall noise performance of a

detector[130 131] In other words NEP is defined as the power from the input source

(Pin) that is required to supply a voltage output (Vout) equal to the root means square

noise at Vout [132] For an ideal lossless match and assuming only Johnson-Nyquist is

present the NEP of a zero-bias detector can be expressed as

119873119864119875119900119901119905 = radic4119896119879119861119877119881120573119900119901119905 (65)

183

Where βopt is responsivity with an optimum match which is given by1 2frasl 119877119895120581 This type

of noise appears when changing voltages across a diode and a noise voltage (Vn)

normally will arise Theoretically the NEP has units of Watts (as it is actually a power)

but it often normalized to 1Hz as it is independent of bandwidth and thus the unit

becomes WHz12

Additionally there are also several noise sources that contribute to Vn in a

semiconductor diode which are Johnson-Nyquist noise Flicker Noise and shot noise

Johnson-Nyquist noise [133 134] appears across any conductor or semiconductor at

thermal equilibrium this is due to the thermal agitation of the carriers or charges It can

be expressed in root mean square voltage as below

119881119869minus119873 = radic4119896119879119861119877119895

(66)

Where Rj is the differential intrinsic resistance B denotes the post-detection bandwidth

T is the device temperature and k is the Boltzmann constant[134]

The second noise that is taken into consideration when dealing with semiconductor

devices is Flicker noise more commonly known as 1f noise It is a group of known and

unknown noise sources that can be observed in the frequency spectrum and normally

display an opposite to the frequency power density curve[135] It comes from a variety

of different causes ie recombination effects at a defect in semiconductor mobility

fluctuation and flow of direct current as well as interface phenomena [136-138] In term

of voltage source Flicker noise can be expressed as[139]

1198811119891 = 119870119891119881119909119891119910 (67)

Where Kf denotes a device-specific constant V is the voltage and f is the frequency The

value of x and y typically used are 2 and -1 respectively This type of noise (1f noise)

will be neglected at frequencies high enough due to the fact that the NEP of the diode is

proportional to the thermal noise and resistance of the diode

Finally the noise that causes time-dependent fluctuation in a flow of electrical

current because of the carrier or electron charge crossing a potential barrier is called shot

noise Shot noise is due to the randomness in the diffusion and recombination of both

majority and minority carriers[140] The equation of shot noise term at random time is

given by[141]

184

119881119878ℎ119900119905 = 2119902119868119861 (68)

Where q is the electron charge I is the current and B is the bandwidth This type of noise

is not affected by changes in temperature or device parameters Therefore the total noise

voltage appearing in the semiconductor is found to be [141]

1198811198992 = 119881119869minus119873

2 + 11988111198912 + 119881119904ℎ119900119905

2 (69)

However a zero-bias device will greatly eliminate both shot and flicker noise

compared to a biased device This has been explained by Equation (66) and Equation

(67) above where both noises are significantly related to the current and voltage Thus

if V and I =0 in the nonappearance of incident power then Vn will also become zero For

a detection process with bias the diode will be self-biased by ΔV which causes both

flicker and shot noise to appear But the shot noise in practical situation is much smaller

and thus normally ignored [142-144] Usually the noise in a zero-bias detector is

estimated by considering the low power limit as good first order estimation in which the

presence of only Johnson-Nyquist noise and ΔV is arbitrarily low [11 145 146]

Additionally it has been reported that the noise in tunnelling type diodes displays very

low or no excess noise in the bias region of the current-voltage characteristic[147 148]

Therefore in general most of the noise specifically in tunnelling type of diode will be

neglected this is a great advantage compared to SBD or transistors

65 Modelling of a 100GHz Zero-biased ASPAT Detector

Once the DC and RF characteristics of the ASPAT diode had been accurately

obtained the next step is to model and design a detector circuit based on S-parameter

measurement results as was explained in the previous chapter The aim is to realise a

detector circuit design which can be operated at millimetre and sub-millimetre wave

regions from an accurate diode model prior to the circuit design A diode detector model

puts experimental observations into context and offers insight into future experiment

results Consequently an electrical model based on lumped element component is vital

for a deeper understanding of how and to what extent a new device like the ASPAT

diode can affect all the key detector parameters that were previously discussed The

prediction of the detector parameters mostly depends upon the ASPAT diode

185

geometrical emitter size and material parameters However for millimetre wave

operating frequency the accuracy of the model is more sensitive not only to diode size

but also to the diode periphery ie substrate as well as coplanar amp transmission line

adopted in the circuit Therefore both extracted intrinsic and parasitic element of such a

device must be taken into account

In this work an ASPAT diode with an emitter size of 4times4microm2 which is the

smallest size that could be fabricated so far was chosen to be exploited for detector

designs The important parameters related to the 4times4microm2 GaAsAlAs ASPAT which

works at 0V is summarize in table 61 below

Table 61 A summary of all the important parameters of the 4x4 microm2 diode

Device Rj(Ω) Rs(Ω) Cj(fF) Cp(fF) κ(V-1) Intrinsic_fcut-off(GHz)

4times4microm2 90K 11 21 15 23 629

The actual measured I-V characteristic is used to model the diode since the library

in the ADS simulation tool does not have an ASPAT diode model or any tunnelling

diode for that matter The procedure of realizing the diode model is by taking the I-V

characteristic obtained from the 4times4microm2 emitter size measurement results and

converting it into a10th

orders polynomial equation via MATLAB software to create a

virtual I-V characteristic Thereafter this equation is then defined as a two terminals

device namely Symbolically-defined Device (SDD1P) ie a component of the non-

linear equation provided by ADS (Figure 66) to represent the ASPAT Figure 67

shows the measured data and 10th

orders polynomial equation fit very well to each other

Hence this new component used to represent the whole ASPAT diode will be used in

this research for MMIC detector and Frequency Multiplier designs The device chosen to

be modelled (4times4microm2) has measured junction capacitance of 21fF (at 0V at 40GHz) The

detector circuit is designed to operate at 100GHz for a safe side due to the extrinsic

calculated fcut-off is around 380GHz Since there are a lot of advantages in using unbiased

detectors compared to biased one this work will discuss the performance of millimetre-

wave detector at zero-bias and their result will be compared to the current performance

of other diodes reported in the literature

186

Figure 66 The configuration circuit to verify 10th order polynomial equation extracted from

MATLAB to realize a virtual GaAsAlAs ASPAT diode

Figure 67 Verification of actual (blue measured) and virtual (red_10th order polynomial) I-V

characteristic of the 4times4 microm2 diode used in this study

To realize the ASPAT detector circuit a simple detector circuit topology as

depicted in Figure 68 was constructed Initial simulation was run to perform a

functionality check of the detector circuit utilizing the Harmonic Balance (HB)

simulation tool embedded in that particular software Such simulation tools will analyse

the detector performances in the frequency-domain as it is mostly beneficial and fully

compatible with microwave and millimetre wave problems The frequency domain is

also suitable for single and multi-tone power excitation The importance of harmonic

balance are described in [149]

-00005

0

00005

0001

00015

0002

00025

0003

-3 -2 -1 0 1 2 3

Cu

rre

nt

(A)

Voltage (V)

ADS

4x4

187

Figure 68 Direct detector circuit topology using an ASPAT diode

Initially the circuit topology that consists of P1_Tone power supply ASPAT

bypass capacitor and load resistance is simulated by setting up a fixed input frequency at

100GHz The ASPAT diode provides a DC output voltage proportional to the input

power strength depending on the absolute values of the DC terms associated with the

nonlinearity of the I-V characteristics The capacitance in the output part is a bypass

capacitor used to prevent millimetre-waves from leaking to the output The load

resistance is large enough to ensure the voltage divider between load impedance and

device impedance gives maximum voltage sensitivity This large load resistance is

achieved by creating an open circuit at the end of detector circuit terminals

Noted that this simulation was run using diode parameters that were extracted from on-

wafer one port S-Parameter measurements as described in chapter 5 To apply them in a

two port application ie detector circuit may or may not provide a very accurate

outcome it however worked adequately in the particular circuit described in this chapter

but may not work in other circuits in general Thus the one port extractions in this work

still provide adequate parameters to build and design specific MMIC detector circuits

but not in general applications

The main reason for these simulations and their results to be used in high frequency

applications is due to the fact that actual RF measurement were done up to 40GHz

Additionally the 100GHz operating frequency was obtained from extrapolation of each

ASPATrsquos component Since the on-wafer measurement that were carried out were

limited to one port characterization applying them to two port network applications may

188

have extra consequences which are unknown Therefore actual MMIC detectors are

needed to be built and test to validate this work

To find out what power the 4times4microm2 ASPAT diode can withstand the input power

is varied from -40dBm to 10dBm via control by the P1_tone As can be seen in Figure

69 the diode starts to saturate when the received input power is about -8 dBm Above

this power limit both output voltage and sensitivity drop dramatically

Figure 69 Output voltage and detector sensitivity over wide range of input power

This diode detector circuit can thus operate adequately at given input powers from -30

dBm to -8 dBm with a sensitivity of 950VW However for the best possible sensitivity

over a range of input frequencies only one optimized input power needs to be chosen

The parameters that directly influence the voltage sensitivity are the curvature

coefficients load resistance and video resistance as can be seen from Equation (61)

Therefore in the following simulation the values RL will be optimized according to the

diodes optimum input power with regards to the highest possible voltage sensitivity

Consequently five values of load resistance were chosen from few ohms to infinity

and with the same applied input power as depicted in Figure 610 For most load

resistors the sensitivity is constant at low input power and drop at the diode saturation

region (-8dBm and greater) However for an RL value of 100KΩ and below the loaded

voltage sensitivity shows a peak near 0dBm input power which corresponds to the

maximum slope of the ASPAT detector transfer function The highest voltage sensitivity

is obtained by using an Open circuit load impedance as shown in the graph The Open

circuit load impedance gives the highest voltage sensitivity due to the voltage divider

189

between source and load impedance therefore RL must be at least 5-10 times larger than

Rj to give a better sensitivity

Figure 610 Simulated dependence of Voltage Sensitivity on incident power and load resistance of

the ASPAT detector

Furthermore the value of voltage sensitivity (βV) depends on the junction

resistance of the diode and thus the large value of Rj of the ASPAT diode yields the high

βV observed Rj which was taken from the non-linear measured IV characteristic is a

voltage dependent parameter [142]and is inversely proportional to the forward bias

voltage as depicted in Figure 611 below

Figure 611 Junction resistance as a function of forward voltage

1

10

100

1000

-40 -30 -20 -10 0 10

Vo

lta

ge

Sen

siti

vit

y (

VW

)

Incident Power Pin (dBm)

Infinity

1KΩ

10KΩ

1MΩ

100Ω

100KΩ

0

20

40

60

80

100

0 001 002 003 004 005 006 007 008 009 01

Ju

nct

ion

Res

ista

nce

(k

Ω)

Bias Voltage (V)

190

In fact the expression of (120597119881120597119868frasl ) contribute to the video impedance expression via the

expression RV=Rj+RS (nonlinear resistance) As RS is very small compared to Rj thus it

was ignored when calculating RV Although the large value of Rj will increase the

voltage sensitivity it will also make matching difficult to achieve Therefore there will

be a compromise between the size of the matching circuit and voltage sensitivity to

attain the correct value of Rj Additionally a very high Rj (around ~1MΩ) will also

increase the detectorrsquos noise equivalent power (NEP in Eq(65)) Thus an average value

of RJ typically around 100kΩ is satisfactory[150] By having a large value of RJ one can

benefit from a low input power to drive the diode into the non-linear region and thus the

detector can work at very low RF input power The NEP for GaAsAlAs ASPAT diode

is then calculated based on parameters obtained from this simulation at room

temperature The values are compared to other diode detector available in the literature

as shown in Table 62 below

Table 62 Noise Equivalent Power (NEP) for zero bias detector diode

Device NEP (pWHz12) RJ(KΩ) Frequency(GHz)

ASPAT (4times4 microm2) 188 92 100

Tunnel Diode (08times08 microm2)[150] 370 26 220-330

Zero bias SBD[11] 15 3 150

Sb-Heterojunction Backward Diode[145] 024 32 94

VDI Zero bias SBD[151] 2 18 110

From the Table 62 above the NEP of the ASPAT is calculated based on RF input power

which is required to obtain an output signal-to-noise ratio of unity in a 1Hz at detector

output[142] and also the assumption of only Johnson-Nyquist (thermal noise) is

dominant for small incident power (-25dBm)[152 153] The prediction of NEP for the

ASPAT is comparable to the VDI Zero bias detector since the value of their junction

resistance is much lower than that of the ASPAT Therefore it is very important to

obtain a reasonable value of junction resistance From this it is clear that a trade-off of

high voltage sensitivity and low junction resistance is best to obtain low noise

191

Finally the curvature coefficient at specific operating voltages also influences

the voltage sensitivity of the detector Figure 612 shows the calculated curvature

coefficient of the 4times4microm2 ASPAT diode used in this work

Figure 612 Curvature coefficient for the measured I-V characteristic with a device size of 4times4μm2

The high zero-bias curvature (23V) is reached from the mutual effect of the intra-band

tunnelling in the GaAs-AlAs-GaAs and the highly doped GaAs at the anode and cathode

(Rs amp Rj) This was shown in the numerical simulation in[154] where the combined

effects of the optimum anode AlAs composition increases the curvature coefficients by

thinning the energy tunnelling window (intraband tunnelling process)

Other approaches that can lead to a large curvature is using smaller device area

with minimum series resistance [155] as was also discussed in Section 52 Having a

better curvature coefficient leads to increased voltage sensitivity as seen from Equation

(61) above and βV is also proportional to κ [111] The curvature coefficient calculated

using the above formula is nearly 23V at 1mV peak and high voltage sensitivity can be

achieved by having large value of curvature coefficient But one needs to remember that

this will also decrease with increasing input power because RJ which will also decrease

Therefore for a safe operating region the incident power that can be applied through the

diode is in between -30dBm to -8dBm for a 100GHz operating frequency

In this simulation -25dBm is chosen to simulate the GaAsAlAs ASPTAT diode

detector working at 100GHz input frequency By fixing the P1_Tone to this input power

-5

0

5

10

15

20

25

30

-001 001 003 005 007 009

Cu

rva

ture

Co

effi

cien

t(V

-1)

Voltage (V)

k(4x4 um^2)

192

the frequency is varied from 90GHz to 110GHz and the results are depicted in Figure

613 below

Figure 613 Voltage Sensitivity against frequency at -20dBm fixed input power

Noticeably the voltage sensitivity of the diode detector decreases linearly with

increasing input frequency At 100GHz a sensitivity of around 540VW is obtained

However the sensitivity in this case is roughly estimated from the Equation (61) above

and will not be sufficiently accurate because of the effect of other key factors such as

reflective power which was not included This parameter must be taken into the account

due the P_1Tone power source which provides a 50 Ω impedance source (Zin) which

does not match the load impedance (ZL) which consists of JωL 1JωC and Z and which

mainly comes from the ASPAT diode itself In order to determine the load impedance at

the diode a typical ohmrsquos law (Z=VI) equation must be used at the input side of the

diode with regards to the applied frequency (100GHz)

As a result the total load impedance obtained from the simulation is 11055-

j69057Ω which is clearly not matched to the 50Ω impedance source The mismatch

between source and load leads the available power from the source to be not fully

delivered to the load and hence there is loss of power leading to a lower detector

sensitivity Therefore actual calculation must take into the account the reflection impact

as expressed in the equation below

193

120573119881(119886119888119905119906119886119897) =

119877119869119877119871120581(1 minus |Γ|2)

2(119877119881 + 119877119871) (1 +119877119904

119877119895)2

[1 +1205962119862119895

2119877119878119877119895

1 + 119877119904 119877119895frasl]

(610)

Where the term (1-|Г|2) refers to the normalized power absorption by the ASPAT diode

and Г is the reflection coefficient due to discrepancy between 50Ω input impedance (Zin)

of the input port and the diode Consequently the calculation of reflection coefficient

(eq611) is carried out by using this expression and the result is shown in Figure 614

Γ =

119885119871 minus 119885119894119899

119885119871 + 119885119894119899

(611)

Figure 614 Reflection Coefficient versus operating frequency without matching circuitry

As can be seen in Figure 614 without the matching circuit the S11 is decrease

linearly with frequency but only very slowly This means that most of the RF power

transmitted from the source is reflected back (by that ratio) when it went through the

diode Therefore in order to resolve the mismatch a matching circuit is introduced in

between the source and load of the detector circuit as shown in Figure 615 (red

rectangular) This matching circuit works by transforming the load impedance into an

impedance that is identical to the source or input impedance Note that for any

impedance matching circuit the main purpose is usually to obtain maximum power

transfer to the load however in some cases (ie oscillators) the matching circuit is to

achieve a lower noise figure Hence in a broad sense the introduction of the matching

194

circuit in a detector circuit can be defined as a circuit that convert available impedance

into wanted impedance by obeying the maximum power transfer theorem [156]

Figure 615 Detector circuit with impedance matching circuit placed in between diode and source

There are many type of matching circuit that can be used to achieve both

objectives above such as circuits using lumped element transmission-line-impedance

matching circuit single and double-stub tuners as well as a quarter-wavelength In this

design the technique used to match source and load impedance is the single open and

short stub (in red rectangular) as it is simple convenient and very efficient in ADS

simulation The important parameter that needs to be tuned in both stubs is the electrical

length (E) at any designed frequency ie 100GHz in this case The E tuning is realized

by using the Smith Chart features available in the ADS simulation tools Figure 616

shows the reflection coefficient with matching circuit modelled over a broad frequency

band and it is clearly shown that at the desired operating frequency the reflection is very

low Note that it is difficult to obtain wide frequency band matching

Matching circuit

195

Figure 616 Reflection Coefficient over wide frequency band with matching

The simulation is continued to find the effect of the matching circuit placed in

the detector circuit on the voltage sensitivity The same input power (-25dBm) is applied

to the diode and the voltage sensitivity is plotted against frequency as depicted in Figure

617 Obviously at the desired operating frequency (100GHz) the sensitivity rises up to

a maximum value of 2100VW with this value obtained without any reflection and the

input port being completely matched with the ASPAT diode model used in this work

Figure 617 Simulated voltage sensitivity of GaAsAlAs ASPAT detector at W-band frequency

Once the matching circuit to be used for 100GHz operating frequency was confirmed

further simulations were made by applying a series of low input power to find the

tangential sensitivity (Tss) of the detector diode When determining the Tss it is very

important to include the matching circuit as it will minimize any power losses through

the diode thus a very small input power can be detected

196

Figure 618 Lowest detectable signal at 100GHz operating frequency

The transfer function depicted in Figure 618 shows incident power of -80 dBm

to -50 dBm applied and the lowest detectable signal that can be obtained with 4times4microm2

mesa size ASPAT diode is around 138microV at -68dBm Although a typical value of Tss is

normally not more than ~-55dBm as in ref [74] the lower value obtained in the

simulation is because the device is operated at zero bias operation and does not use any

amplifier therefore the noise and values related to amplifier as in Equation 63 have been

neglected Even though the TSS appeared very low it is most likely very dependant to

the 10th

order polynomial equation embedded into SDD in ADS software Therefore in

near future the TSS value has to be determined in real fabricated MMIC ASPAT detector

As discusses in Section 63 other important parameter that can be extracted from

the diode transfer function is the Dynamic Range of the diode It can be obtained in a

region called square law region which is a region where the Vout of the ASPAT diode is

proportional to the square law of the input power signal From Figure 619 the square

law region is in between -68dBm and -12dBm and the linear region or in this case

saturated region is above -12dBm Taking to the account the roll-off point where Vout

has dropped 1dB below the extrapolation of the dependence at low input power and

therefore the dynamic range of the detector diode can be obtained by taking the interval

between TSS and the 1dB roll-off point (in dBm) which is ~55dBm

197

Figure 619 The dynamic range of the ASPAT diode obtained from linear region of diode operation

The figure of Merit (M) of the detector (ie equation 63) is 2100 (90K) 12

where 2100 is the sensitivity and 90K is the value of RJ at zero bias which is equivalent

to 652 W The M value should be large however in this case due to RJ being very

large it has dropped tremendously when compared to the voltage sensitivity obtained in

this simulation The results obtained indicate a reasonably successful design of the

MMIC detector using the 4times4microm2

emitter size GaAsAlAs ASPAT diode The results

obtained lead to the design of other MMIC detector using the other fabricated diode

sizes (6times6microm2 and 10times10microm

2) A Similar procedure to the one used for the 4times4microm

2

diode was followed The only difference was the use a slight higher input power (-

20dBm) than in the 4times4microm2 design Hence the simulation results obtained are then

compiled and compared as depicted in Figure 620

Saturated

Region

198

Figure 620 Simulation result of diode sensitivity for three devices geometries obtained from the

fabricated ASPAT in this work

The graph plotted for each dimension was taken after matching circuits were included

As can be seen in Figure 620 the highest sensitivity is achieved using the smallest

device size as this has the highest cut off frequency Table 63 below summarises the

performances of the 100GHz ASPAT detectors obtained from the simulation in this

work

Table 63 Summary of characteristic of interest for GaAsAlAs ASPAT detector

Device Tss(dBm) fcut-

off(GHz)

Bv

(VW)

Dynamic range dB(dBm) M(W-12

)

4times4microm2 -68 380 2100 55 65

6times6microm2 -50 151 1445 48 45

10times10microm2 -40 98 247 40 21

From table 63 above it is clear that a lower cut-off frequency will affect the

voltage sensitivity The dynamic range between each ASPAT is different because larger

size area will allow more power to go through the diode as a result of high current that

such a device can handle before reaching the saturation their lowest detectable is higher

Small diode size will detect lowest voltage but cannot handle high power On the other

hand a large diode size is able to receive high power however can only offer lower

voltage output Therefore there is a trade-off between small diode size and receiving

input power which will directly affect Tss and 1dB roll off Once again all the

100

1000

10000

90 95 100 105 110

Vo

lta

ge

Sen

siti

vit

y (

VW

)

Frequency (GHz)

10times10microm^2

6times6microm^2

4times4microm^2

199

parameters obtained in this simulation are just estimation from the 10th

order polynomial

equation thus real ASPAT detector has to be fabricated for verification

The best device performance among all GaAsAlAs ASPAT diodes was obtained

with the 4times4microm2 mesa area size diode which was compared to other exiting millimetre

wave detector diode available in the literature Since the 100GHz is located in the W-

band spectrum frequency therefore the comparison will be performed in this frequency

band but with low input power The parameters for the-state-of-the-art zero bias

detectors are gathered in Table 64 below

Table 64 Comparison 4times4microm2 GaAsAlAs ASPAT diode to the-state-of-the-art zero bias detector

at W-band (75GHz-110GHz)

Device Size(microm2) Tss(dBm) βv (VW) Pin (dBm) f (GHz)

GaAsAlAs ASPAT 4times4 -68 2100 -25 100

GaAs SBD HSCH-

9161[157]

- -49 2200 94

HBD[158] 15x15 2540 -20 95

Planar SBD[159] - -68 2100 -25 100

Note that the ASPAT diode retains its favourable temperature stable characteristics

which are not the case for all the diodes used for comparison in Table 64

66 Conclusions

In this chapter all theory regarding RF detection using diodes ie parameters of

interest noise consideration etc have been discussed The aim to design and develop a

low cost reliable and sophisticated zero bias 100GHz detector circuit was achieved

through exploitation of a 4times4microm2 GaAsAlAs ASPAT diode The design was performed

with the aid of Keysight ADS modelling software utilizing harmonic balance simulation

The effect of load resistance junction resistor to the detector voltage sensitivity was also

discussed in details The 90KΩ Rj value and open circuit load resistance was chosen for

high sensitivity

200

A step by step design of a W-band ASPAT detector was presented The effect of

matching circuit was discussed in detail and where an unmatched sensitivity of 843VW

is obtained which then increases to 2100VW after matching Through RF

characterization simulation a detection at 100GHz (W-band) was successfully achieved

with a relatively large device mesa area (4times4microm2) at an input power of -25dBm

(8microWatt) leading to a 2100VW voltage sensitivity a -68dBm TSS and 55dBm dynamic

range All these values are comparable to others fabricated diodes in the literature

The zero bias ASPAT detectors based on the GaAsAlAs material system in this

work are still at an early stage of development a lot of work is still required to realize

high yielding integrated millimeter and sub-millimeter wave (MMIC) detector circuits

However as this work is on-going at Manchester it is expected that fabricated ASPAT

MMICs with even higher voltage sensitivity will be fabricated in the near future through

collaborating bodies involved in this research especially the University of Cambridge

and ICS Limited

201

7 FREQUENCY MULTIPLIERS (DOUBLER) DESIGN USING

GAASALAS ASPAT DIODES

71 Introduction

Originally the key application for the ASPAT diode was for use as microwave and

millimetre wave detectors[18] This is due to the fact that such diode demonstrates

strong non-linearity low noise and high cut-off frequency features as described in the

previous chapters However these features are not only beneficial for detection purposes

but also allow them to be used and designed as microwave and millimetre wave sources

The only way to generateenhance continuous wave (CW) power using a non-linear

device is through frequency multiplication techniques It is known that the frequency

multiplier is the alternative approach (to 3 terminal transistors) using non-linear devices

that are used to generate high frequency low phase noise signals Any high quality low

frequency signal that goes through a frequency multiplier circuit can be generated to any

desired high output frequency[160] Therefore the main objective of this chapter is to

demonstrate the feasibility of the ASPAT diode as a compact source of microwave and

millimeter-wave receiver for imaging applications[161]

The study of the ASPAT diode as a power source begins with a brief explanation

of the importance of a frequency source and the lack of compact device and technologies

at high-frequency signals The state-of-the-art for frequency multiplier will also be

discussed In the next section (Section 74) the fundamentals of the frequency multiplier

architecture ie the principle of operation and appropriate devices will be presented

Since this is the first attempt at using GaAsAlAs ASPAT diodes a simple multiplier

circuit design and topology was built This will be discussed in detail in the subsequent

sections where simulation results are discussed The focus of the discussion will be to

demonstrate the possibility of a GaAsAlAs ASPAT diode functioning as a frequency

multiplier and comparison with other state-of-the-art varistor mode frequency

multipliers

202

72 Motivation and Background

Typically continuous wave (CW) sources generating below 100 GHz can be

obtained through oscillators amplifiers and pin diode comb generator Below 10THz

the sources can be made from RTD IMPATT diodes and Gunn oscillators and above 10

THz it is commonly done by photonic mixing quantum cascade laser (QCL) and gas

lasers [162] Both types of sources and their performance are plotted in Figure 71

However these conventional ways of generating millimetre and sub-millimetre waves

have their own limitations ie high cost complexity and sometimes requirement for

cryonic cooling The most effective way to tackle the limitations of conventional mm-

wave and THz sources is by implementing frequency multiplication technique using

solid state nonlinear diodes [163-165] such as SBD and ASPAT diodes

Figure 71 performance of state-of the-art millimetre wave source [166]

Twenty years ago there were only two types of diodes (SBD and P-N junction

diodes) often used for frequency multiplication To date besides the SBD there are

many types of diode that have been used as frequency multipliers These include the

high electron mobility varactor (HEMV) single barrier varactor (SBV) [167] and hetero-

structure barrier varactor (HBV) (270GHz with 90mW input power and Conversion

Efficiency of 72)[168 169] Other variants that have developed to enhance the

frequency multiplier performance of the classic SBD [170] include the Barrier-intrinsic-

203

n+ (BIN)diode and Barrier N-layer N+ (BNN) diode [171] Other diodes for use in such

applications are the planar doped barrier diode (PDB) Resonant tunnelling diode (RTD)

amp it families ie Quantum well diode (QWD) and step recovery diode which is a

modification of the P-N junction diode

Although three terminal devices ie FET GaAs MESFET and HEMT had

shown better performance and are capable of achieving greater efficiency and

bandwidth as well as having additional conversion gain features [160] two terminal

devices (ie varactor diodes) which are passive multiplier are still preferred This is due

to their simplicity and most importantly their ability to generate very little noise Among

these types of device technologies the SBD is preferable as it is mature and has been

shown to be very suitable for high-frequency applications [172 173]

The ASPAT diode is exploited to investigate the possibility and the feasibility of

generating microwave and millimetre-wave power through well-known frequency

multiplication methods The utilisation of the ASPAT in frequency multiplication will

also aid in generating local oscillator sources which are critical components in

heterodyne receivers The ASPAT diode will work in resistive I-V mode (varistor mode)

and has features to work also at zero bias condition thus offering low power handling

than traditional high-efficiency varactor diode since the varactor diode requires a large

reverse bias supply of several tens of volts

73 Frequency Multiplier Architecture the Basics

In principle a frequency multiplier is an electronic circuit that gives an output

frequency that is a multiple integer of its input frequency signal pumped from a local

oscillator as depicted in Figure 72 The ability to generate any desired multiple output

signals is realised by a nonlinear device ie diode or transistor Such devices though

also can give distortion or cause sudden change to the input frequency Additionally

these devices generate multiples of the input frequency (fout) The distortion of the

sinusoidal signal refers to an abruptsudden change versus amplitude or time which thus

generates higher frequency with lower amplitudes of the input signal Usually a

frequency multiplier circuit will include a bandpass filter to select the desired harmonic

204

frequencies and deselect undesired harmonic frequencies especially fundamental

harmonic at the output for further processing

Any non-linear device either in symmetricalantisymmetric current-voltage or

capacitance-voltage can be utilised to realise a frequency multiplier source [168 174]

Figure 73 describes the method where a nonlinear resistance is utilised to convert a

harmonic input signal into periodic output signal containing components at multiples of

the input frequency Both non-linear resistance and reactance characteristics can be

extended into power series methods

Figure 73 Principle of operation for frequency multiplier utilising a non-linear resistance [10]

The operating principle of the frequency multiplier is shown in Figure 73 where

the I-V curve converts a harmonic frequency input into a periodic frequency output

including components at multiples of the input frequency The non-linear I-V

characteristic can be explained in term of a power series at the operating fixed point of

bias voltage (VB) [174]

Frequency

Multiplier Circuit

finput

foutput

= nfinput

Figure 72 Frequency multiplier has the property that foutput is an integer multiple of fin

205

119868(119881119861 + ∆119881) = 1198860 + 1198861∆119881 + 1198862∆1198812 + 1198863∆1198813 helliphellip (71)

For a given input voltage as below

∆119881 = 119881119904 cos120596119878119905 (72)

The input signal harmonics will become

119868(119905) = 1198680 + 1198681 cos120596119904119905 + 1198682 cos 2120596119904119905 + 1198683 cos 3120596119904119905 (73)

Where t and ω are the time and angular frequency respectively based on equation (73)

the output contains both signal source and harmonics Therefore a complete frequency

multiplier circuit has to have non-linear device and filter to allow the selection of any

frequency components needed

731 Types of frequency multipliers

Frequency multipliers can be classified into passive and active multipliers This

classification is based on the ability of the frequency multiplier to yield any conversion

gainlosses The passive multiplier is the one that only produces conversion losses In

other words it can be described as a multiplier that generates an output power level

lower than the excitation input power and it is mostly dominated by passive nonlinear

devices ie Diodes On the other hand the active multipliers refer to a device that would

produce an output signal with a power level that is greater than the input signal power

This conversion of power is termed as conversion gain These types of multipliers attract

much attention as they do not only increase the frequency at the output but also the

signal power

Passive frequency multiplier can be formed by using diodes that are classified as

being of the varistor (non-linear I-V) or varactor (non-linear C-V) type [160 174] The

varistor type will influence the frequency multiplication with a non-linear resistance or

conductance (resistive diodes) and this results in a very large potential bandwidth at the

output but poor conversion efficiency The varactor diode type where the frequency

multiplications are affected by the non-linear capacitance (reactive diode) as their

reactive element typical result is high conversion efficiency A diode that is used in this

206

application must have strong nonlinearity stable electrical characteristic repeatable and

has fast enough response to an applied frequency Therefore multipliers are classified

into Doubler Tripler quintuple and so forth depending on the highest power of output

harmonic signal

In general all varactor type diodes with such characteristics will produce high

power at odd-order harmonic oscillation if any microwave signal is pumped into them

The benefit of having odd-order in multiplier design is that it reduces the complexity of

the overall circuit ie it eliminates even-order idler frequencies [175] The varactor type

diode had been shown by Manley-Rowe to a get maximum 100 conversion efficiency

for generating an ideal harmonic [176] compared to the varistor type where the

maximum efficiency achievable ideally is 1 1198992frasl where n is the multiplication factor

(output harmonics number) [174] In the case of power handling (input excitation power)

for multipliers varactor mode diode required greater power (several milli Watt) than the

varistor mode due to the fact that reverse applied voltages are very large (many tens of

volts) Therefore these types of frequency multipliers may not be suitable for the case of

high input power excitation There no report in the literature of varistor based

multipliers working with high power excitations

74 Parameters of interest for Frequency Multipliers

The simplest way to describe an equivalent circuit for a complete frequency

multiplier is by setting a Source impedance (ZS) at the input side and load impedance

(ZL) at the output side as depicted in Figure 74 below This circuit usually has the same

properties as described in the previous chapter and most of the others two ports

networks However in this case the purpose is different and is the conversion of a sine

wave signal source (Vs) with angular frequency ωs to an output signal with frequency

nωs where n is the multiplication integer or the order of multiplication

207

Figure 74 A standard system for two port frequency multiplier circuit

Referring to Figure 74 above there are few sets of parameters for the frequency

multiplier to be taken out and compared Examples are the conversion loss maximum

input signal power Impedance at source and load Bandwidth multiplication factor or

harmonic amp subharmonic content and noise conversion properties The conversion loss

(CL) is described by the ratio of available power at source (Ps) to the output harmonic

power delivered to load resistance (PL) and is normally expressed in dB It occurs due to

the nature of passive semiconductor diodes and the electronic circuit itself that are lossy

and dissipate energy On the other hand the conversion efficiency (ηn) is a ratio of the

output power at load (PL) to the available power at the input (Ps) This is often expressed

in percentage () The conversion efficiency can be determined as

120578 =

119875119878

119875119871

(74)

while the conversion loss is expressed as[174]

119871119899[119889119861] = 10 log

119875119904

119875119871= 10 log (

|1198811199042|

4 119877119890 |119885119904||1198681198712|119877119890|119885119871|

) (75)

Where Vs is the input voltage and IL is the output current amplitude Besides this the

conversion efficiency is often referred to as the inverted value of the Ln In designing a

frequency multiplier it is crucial to minimise the conversion loss and maximise the

conversion efficiency value

To achieve a perfect multiplier with minimum conversion loss the impedance of

source and load must be at an optimum level This implies that the source impedance

Frequency

Multiplier Z

L

Zout

Z

in

V

Z

SWR Г

208

(Zs) must be very close to the complex conjugate of the multiplier input impedance

(Zin) hence minimum reflection loss will occur at the input side This can be realised by

introducing an impedance matching circuit between the diode and source The power

transfer between the source and the multiplier is quantitatively described by the value of

the multiplier input reflection coefficient (Ѓ) with source Zs assumed to represent a

reference impedance This specification also can be explained in the standing wave ratio

(SWR) Both relationships are described below respectively [174]

Γ =

119885119894119899 minus 119885119904lowast

119885119894119899 + 119885119904

(76)

119878119882119877 =

1 + |Γ|

1 minus |Γ|

(77)

Where the asterisk () represents the complex conjugate of the Zs impedance

On the other hand the situation of the load impedance is different when a standard

or an optimum value is provided by the designer This will either increase the conversion

loss or decrease the output power Thus one has to keep in mind that frequency

multipliers are non-linear devices and power transfer condition both at the input and the

output depend on each other and the input signal level[174]

75 20GHz40GHz Varistor Mode ASPAT Diode Frequency Doubler

In this work a similar ASPAT diode (4times4microm2) to that in designing millimetre

wave detector in the previous chapter is used The main objective of designing the

frequency multiplier circuit was first to investigate the performance of the ASPAT

diode as a microwave or millimetre wave signal source A design is deemed successful

when the diode physical parameters are optimised and the suitable impedance matching

network is produced for each desired harmonic as well as maximising the output power

These goals however are hard to achieve when a higher frequency operation is targeted

for use

There are many types of multiplier circuit topologies that can be implemented

using GaAsAlAs ASPAT diode in varistor mode to achieve high order of multiplication

209

Examples are single diode multiplier series or parallel connected diode multiplier anti-

parallel amp anti-series connection diode pair multiplier anti-parallel-series connected

diode multiplier and bridge frequency multiplier as well as nonlinear transmission line

frequency multiplier [174] Before designing a circuit there is one most important

consideration to make Prior to choosing any mentioned circuits to be used for frequency

doubler the design considerations are made based on the capability of ASPAT diode to

receive an optimum amount of input excitation RF power From the discussions in

Chapter 6 the ASPAT diode will reach saturation level (linear regime) at power ~

-10dBm for a device size of 4x4microm2 Once the optimum input power was confirmed the

circuit topology was carefully chosen to balance between the requirements of the

ASPAT to work at high frequencies ie low Rs and Cj amp high diode cut-off frequency

as well as the desired output signal frequency that needs to be produced

To realise the first attempt of an ASPAT diode as a signal source a simple circuit

topology of a frequency doubler was deployed as depicted in Figure 75 below The

frequency doubler circuit consists of a voltage source (can be power source) input

filtering with matching network ASPAT diode output filtering with matching network

and load impedance (ZL)

Figure 75 Block diagram of frequency doubler with series ASPAT mounted diode

In order to investigate the doubler performance the Keysight ADS simulation

tool and similar procedure to obtain accurate ASPAT model using a 10th

order

polynomial equation as in Section 65 was used The circuit in Figure 75 is translated

into ADS format as illustrated in Figure 76 Once the circuit was constructed the

analysis was performed using the Harmonics Balance (HB) simulator The circuit

requirements are matched terminations at the input and output frequencies open

Input Filtering

and matching network

Output Filtering

and matching network

Zs

Vs

ZL

Pin

fin

Pout

nfout

210

circuited terminations at the higher harmonics and optimum reactive terminations (an

inductance which resonates with the junction capacitance) at the output frequencies

Figure 76 Empirical modelling of ASPAT doubler utilising Keysight ADS tool

The circuit in Figure 76 is the simplest way of constructing a frequency doubler

circuit which consists of the signal source (P_1Tone) input matching circuit (Stubs)

filter (Short stub) ASPAT diode low pass filter output matching circuit(Stubs) and ZL

(load resistance) The utilisation of the stubs is an ideal case of simulation since in the

real fabrication stubs are normally formed in large sizes Therefore a proper design such

as using CPW instead of stubs is essential in real fabrication

Again this simulation works for this particular circuit in this chapter as all the diode

parameters were extracted from on-wafer one port S-Parameter measurement described

in chapter 5 To apply them in such two ports applications may not very accurate

however it still provides adequate parameters to build and design particular frequency

multipliers but not in general applications These simulations and results are adequate for

high frequency applications due to the fact that the actual RF measurements on the

diodes were carried out up to 40GHz and the target operating frequency in this multiplier

design does not exceed 40GHz

Since the on-wafer measurement that were carried out were limited to one port

characterization applying them to two port network applications may have extra

consequences which are unknown Therefore actual MMIC frequency multiplier is

needed to be built and test to validate this work

Input Matching

Output Matching

211

To find the optimum output power initial simulation without matching circuit

was performed This simulation was run by varying the input power from -35dBm to

20dBm but fixing the input frequency at 20GHz As can be seen in Figure 77 the lowest

point in the conversion loss (CL) and the highest point of the conversion efficiency (CE)

are obtained from an input power of -1dBm However this amount of input power is too

high for the ASPAT diode The lowest CL at -1dBm may not be accurate since it was

applied without matching Note that it is difficult to achieve a matching between source

impedance and load impedance when varying the input power Therefore a lower input

power of -10dBm is chosen for this frequency doubler operation Figure 77 shows the

Conversion Loss and Conversion Efficiency as a function of the available power of the

given input source

Figure 77 Conversion loss and conversion efficiency as a function of input power

The circuit in Figure 76 works with a -10dBm input power and 20GHz centre

frequency input signal is pumped from the power source (P1_tone) to the ASPAT diode

and distortionabrupt change of input waveform occurs at the fundamental frequency (f0

in this case 20GHz) Such abrupt change produces harmonics and these harmonics can

be classified into desired frequency component by placing two-quarter wavelength (λ4)

stubs (90˚) at both sides of the ASPAT diode At the input side of the diode short circuit

stubs are utilised to permit the f0 tone to reach the ASPAT diode and block the second

harmonics (2f0 in this case 40GHz) back to the input side and pushes it towards the

load resistance On the contrary at the output side of the diode the open circuit stubs are

used to ldquoopen circuitrdquo the 2f0 signals while ldquoshort circuitrdquo the f0 component Thus 2f0

212

signal will not be affected due to the open circuit stubs being half wave (λ2) long The

function of both stubs is basically to isolate the input and output signal from mixing each

other Therefore the design of input and output matching circuit can be achieved easily

The input matching circuit was designed based on the mentioned input frequency

(f0=20GHz) for an available input power (Ps=-10dBm) which is set up at the power

source by using two stubs with the same configuration as used in Chapter 6 Such

configuration is purposely deployed to increase the 50Ω coming from the P1_tone

source impedance to the conjugate thus reducing the reflection coefficient to the

ASPAT diode From the simulation without matching the input impedance to the diode

is 551Ω in magnitude for an available input power of -10dBm

On the output side of the diode output matching circuit is available to transform the 50Ω

port impedance in the optimum load impedance which provides minimum conversion

loss for the ASPAT diode The output matching circuit is designed based on expected

output frequency which is in this case 2f0 =40GHz Other than this optimum

impedance between load impedance and ASPAT will not be achieved thus resulting in

higher conversion loss

To ensure the proposed circuit is valid and suitable for the specific ASPAT diode

mesa size the response of conversion loss and efficiency are plotted as a function of

output frequency from 20GHz to 100GHz The results of both conversions are illustrated

in Figure 78

Figure 78 Conversion loss and Conversion efficiency as a function of output frequency

213

As can be seen in Figure 78 the conversion loss is obtained at the lowest point where

the output frequency is needed Meanwhile the conversion efficiency is maximum at the

same output frequency Therefore this indicates that the first attempt of an ASPAT

Doubler frequency source works well However the values obtained for Ln from this

simulation is 28dB which is rather high On the contrary the η achieved in this study is

very low with a value less than 1

Since the ASPAT is in varistor mode with no bias applied the conversion

efficiency is expected to be low due to resistive losses Another factor that may

contribute to lower η is the diode model itself as it is taken from a 10th

order polynomial

equation not from a diode model provided in the ADS software tools Thus some

properties of such tunnelling diode may not be included Hence it is necessary in due

course to fabricate and build such a compact frequency doubler in the future to verify

the simulation results

From the simulation point of view the less than 1 Conversion Efficiency

obtained is still good enough for a first attempt at a frequency doubler which utilises the

new ASPAT tunnelling diode The frequency doubler obtained from this work is suitable

for use in zero bias varistor modes for low power application The varistor mode doubler

performances from this simulation work are gathered and compared to other in the

literature as summarized in Table 71 below where fout is 40GHz η is 015 Ln is 28dB

and Pi is -10dBm

Table 71 Comparison parameter obtained from this simulation to the state-of-the-art multiplier

diode

Device fout (GHz) η() Ln(dB) Pi (dBm)

ASPAT 40 015 28 -10

SBD (Si)[177] 104 2 134 -10

SBD (GaAs)[178] 13 2 137 -16

214

The performance of the 2040GHz is compared to the literature based on their input

power below -8dBm since the ASPAT is only capable of working at low power To the

best of the author knowledge very few diodes operating in varistor mode at low power

excitation can be found in the field of research and industry Therefore this 2040GHz

ASPAT Doubler might a first for tunnel diodes if it can be fabricated and test at Ka band

and above

76 Conclusions

In this chapter another alternative application based on non-linear features of

GaAsAlAs ASPAT has been presented The simulation of a frequency multiplier

(doubler) was carried out utilizing the 4times4microm2 size ASPAT diode The theory and the

ability of the ASPAT diode to operate as a frequency source were explained in detail

A unique varistor mode frequency multiplier circuit topology for the 2040GHz

ASPAT doubler has been demonstrated and briefly discussed The details and step by

step simulation technique utilizing harmonic balance from Keysight ADS has been

presented Even though the conversion efficiency is very small at 015 and large

conversion loss of 28dB there is still space for improvement in term of design ie

different circuit topology optimized input and output matching circuit etc This design

can be a good reference for a doubler operating at very low power but produce high

frequencies in Ka band

215

8 CONCLUSION AND FUTURE WORK

81 Conclusion

The main focus of this research was the development of a new tunneling diode

namely the asymmetrical spacer layer tunnel (ASPAT) diode for process repeatability

manufacturability and reproducibility The broad study undertaken was to improve the

microwave performance technology by introducing a new type of tunneling diode

For years the asymmetrical spacer layer tunnel diode was unable to be manufactured

due to the high sensitivity of the tunneling current to the barrier thickness This changed

dramatically when the MBE method was carefully optimized to precisely control the

growth to sub-monolayer precisions When stability repeatability and reproducibility in

the epitaxial growth was achieved the next step was to qualify the fabrication process of

the diodes themselves thus ensuring high performance device can be delivered to the

market

For this purpose GaAsAlAs ASPAT diodes made of two different types of

substrates were grown The first batch was grown in the Riber V100HU SSMBE and

used doped substrates Samples XMBE307 and XMBE368 were successfully grown

and fabricated from that batch The DC characterization obtained from measurement

proved that this first batch had fully functional reproducible and manufacturable

devices Later a second batch using semi insulating substrates improvement in spacer

layer and doping concentration were grown This set of samples (9 x 2rdquo wafers grown

simultaneously) and denoted as XMBE304 also showed fully functional DC

characteristic and was used for RF characterization and detector integrated circuits

The conventional GaAsAlAs ASPAT diode structures grown on doped

substrates and developed previously in our lab were not suitable for high frequency RF

characterizations Therefore a major contribution of this work was to develop a new

fabrication technique for a new GaAsAlAs ASPAT structure using semi insulating

substrates to achieve repeatability manufacturability and reproducibility in term of

process flow DC characteristics and ultimately RF characteristics Apart from the

enhancement of the epitaxial layer the other important contribution of this research was

216

the optimization of the small 4times4microm2 emitter size by incorporating both vertical and

lateral structure based purely on low cost I-line optical lithography

To obtain a repeatable and manufacturable fabrication process of lateral

GaAsAlAs ASPAT structures the key issue was to solve the over etching of the

effective mesa area when qualifying the Air Bridge technique This issue caused all

semiconductors under the metal contact to be completely lost thus leaving the metal

contact hanging without connection to any bond pad area However the developments of

the Dielectric Bridge technique realized the true performance of the GaAsAlAs ASPAT

diode structures The samplersquos surface cleanliness as well as DC and RF performance

showed significant improvement when using Si3N4 as a dielectric layer Due to the

highly isotropic etching profile and thicker GaAs layer in XMBE304 samples and

although the smallest emitter size designed on the mask was 2times2microm2 only 4times4microm

2 were

reproducible and showed good uniformity in I-V characteristics

Upon successful optimization in the fabrication process flow of the small emitter

size diodes (4times4microm2 6times6microm

2 and 10times10microm

2) a good uniformity of better than 91

was obtained for DC measurement results within a tile containing over 1000 devices

This confirmed that the lateral GaAsAlAs ASPAT diode structure can only be realized

through the Dielectric Bridge technique These devices were further characterized with

S-parameter measurements and their intrinsic and extrinsic parameter values and

junction capacitances series resistances and junction resistances were extracted leading

to intrinsic cut-off frequencies of 600GHz 429GHz and 100GHz for the three device

sizes respectively

Temperature dependence measurements and simulations were also carried out in

order to confirm that the ASPAT diodersquos characteristics were temperature insensitive

showing less than 5 change in current at both extremes of temperatures 77K to 400K

By comparison the SBD I-V characteristics variations with temperature span orders of

magnitude Physical modelling agreed very well with measured data confirming good

and validated models that can also describe temperature effects

For the realization of the integrated ASPAT millimeter wave detector empirical

modelling using ADS simulation tools was carried out This was performed to predict

the detector performance at 100GHz to comply with the initial objective to develop a

217

millimeter wave detector circuit The simulations using the 4times4microm2 diode models led to

a successful 100GHz circuit design able to detect 100GHz incoming frequency with

2100VW voltage sensitivity

The first ever GaAsAlAs ASPAT diode frequency multiplier design was also

attempted A reasonably good result was obtained for a 20 to 40GHz frequency doubler

operating in varistor mode However the conversion efficiency obtained was less than

1 Further research on this is required to improve the efficiency by using other circuit

topology ie using a balun or other Co-planar waveguides Ultimately fabricating and

testing the actual multiplier circuits are essential to validate the simulation data

82 Future Work

This work has provided a foundation for a reproducible and repeatable GaAsAlAs

ASPAT (SI substrate) wafer fabrication process and recommendations for design and

simulation of ASPAT diode MMIC detectors and frequency source has also been

provided However the GaAsAlAs ASPAT diodes still remain immature and there are

many ways to improve its DC and RF performances both experimentally and in

simulations which will directly affect the detection performances

In term of fabrication process smaller size diodes ie submicron level can be

achieved using dry etching technique with proper calibration For wet etch technique the

etched profile still can be improved by thinning the doped layers so that etching time

will be reduced and hence dimensions down to 2times2microm2 or even 15times15microm

2 can be

reproducibly made

For simulations advanced AC and RF modelling utilizing physical device

simulation available software (SILVACO) must be include in future research hence

holistic study can be conducted to improve the understanding of the ASPAT

For MMIC detector and multiplier design it is vital to produce actual MMIC

devices so that the simulation results can be validated Ultimately tested devices with

good performances can be realized and manufactured

218

REFERENCES

1 Laeri F U Simon and M Wark Host-Guest-Systems Based on Nanoporous

Crystals 2006 John Wiley amp Sons

2 Łukasiak L and A Jakubowski History of semiconductors Journal of

Telecommunications and information technology 2010 p 3-9

3 Song H-J and T Nagatsuma Present and future of terahertz communications

IEEE Transactions on Terahertz Science and Technology 2011 1(1) p 256-

263

4 Hu B and M Nuss Imaging with terahertz waves Optics letters 1995 20(16)

p 1716-1718

5 Smith PR DH Auston and MC Nuss Subpicosecond photoconducting

dipole antennas IEEE Journal of Quantum Electronics 1988 24(2) p 255-260

6 Nagatsuma T Terahertz technologies present and future IEICE Electronics

Express 2011 8(14) p 1127-1142

7 Kumar S et al A 18-THz quantum cascade laser operating significantly above

the temperature of [planck][omega]kB Nature Physics 2011 7(2) p 166-171

8 Phillips T and D Woody Millimeter-and submillimeter-wave receivers Annual

Review of Astronomy and Astrophysics 1982 20(1) p 285-321

9 Whitmer HCTaCA Crystal Rectifiers McGraw-Hill book Company

London 1948

10 Young D and J Irvin Millimeter frequency conversion using Au-n-type GaAs

Schottky barrier epitaxial diodes with a novel contacting technique Proceedings

of the Ieee 1965 53(12) p 2130-2131

11 Liu L et al A broadband quasi-optical terahertz detector utilizing a zero bias

Schottky diode IEEE microwave and wireless components letters 2010 20(9) p

504-506

12 Sankaran S Schottky barrier diodes for millimeter wave detection in a foundry

CMOS process IEEE Electron Device Letters 2005 26(7) p 492-494

13 Chattopadhyay G Submillimeter-wave coherent and incoherent sensors for

space applications in Sensors 2008 Springer p 387-414

14 Anand Y and WJ Moroney Microwave mixer and detector diodes

Proceedings of the Ieee 1971 59(8) p 1182-1190

15 Syme RT Microwave Detection Using GaasAlas Tunnel Structures Gec

Journal of Research 1993 11(1) p 12-23

16 Syme RT et al Asymmetric superlattices for microwave detection in Physical

Concepts of Materials for Novel Optoelectronic Device Applications 1991

International Society for Optics and Photonics

17 Missous M MJ Kelly and J Sexton Extremely uniform tunnel barriers for

low-cost device manufacture IEEE Electron Device Letters 2015 36(6) p 543-

545

18 Syme RT et al Novel GaAsAlAs tunnel structures as microwave detectors in

Semiconductors 92 1992 International Society for Optics and Photonics

19 Schwierz F and JJ Liou Semiconductor devices for RF applications evolution

and current status Microelectronics Reliability 2001 41(2) p 145-168

219

20 HayasHi H Development of Compound Semiconductor DevicesmdashIn Search of

Immense Possibilitiesmdash SEI TECHNICAL REVIEW 2011(72) p 5

21 Mead C Schottky barrier gate field effect transistor Proceedings of the Ieee

1966 54(2) p 307-308

22 Hooper W and W Lehrer An epitaxial GaAs field-effect transistor Proceedings

of the Ieee 1967 55(7) p 1237-1238

23 Drangeid K R Sommerhalder and W Walter High-speed gallium-arsenide

Schottky-barrier field-effect transistors Electronics Letters 1970 6(8) p 228-

229

24 Pillarisetty R Academic and industry research progress in germanium

nanodevices Nature 2011 479(7373) p 324-328

25 Oxley TH 50 years development of the microwave mixer for heterodyne

reception IEEE transactions on microwave theory and techniques 2002 50(3)

p 867-876

26 Baca AG and CI Ashby Fabrication of GaAs devices 2005 IET

27 Cho AY and J Arthur Molecular beam epitaxy Progress in solid state

chemistry 1975 10 p 157-191

28 Cho A Growth of IIIndashV semiconductors by molecular beam epitaxy and their

properties Thin Solid Films 1983 100(4) p 291-317

29 Kiehl RA and TG Sollner High speed heterostructure devices 1994

Academic Press

30 Feiginov M et al Resonant-tunnelling-diode oscillators operating at

frequencies above 11 THz Applied Physics Letters 2011 99(23) p 233506

31 Chang LL L Esaki and R Tsu Resonant tunneling in semiconductor double

barriers Applied Physics Letters 1974 24(12) p 593-595

32 Kasjoo SR Novel Electronic Nanodevices Operating in the Terahertz Region

2012

33 Kanaya H et al Structure dependence of oscillation characteristics of

resonant-tunneling-diode terahertz oscillators associated with intrinsic and

extrinsic delay times Japanese Journal of Applied Physics 2015 54(9) p

094103

34 Chattopadhyay G Technology capabilities and performance of low power

terahertz sources IEEE Transactions on Terahertz Science and Technology

2011 1(1) p 33-53

35 Betz A and R Boreiko A practical Schottky mixer for 5 THz in Proceedings of

the 7th International Symposium on Space Terahertz Technology 1996

36 Yu D et al Ultra high-speed 025-spl mum emitter InP-InGaAs SHBTs with

fsub maxof 687 GHz in Electron Devices Meeting 2004 IEDM Technical

Digest IEEE International 2004 IEEE

37 Das MB Optoelectronic detectors and receivers speed and sensitivity limits in

Optoelectronic and Microelectronic Materials Devices 1998 Proceedings 1998

Conference on 1999 IEEE

38 Rodwell MJ et al Submicron scaling of HBTs IEEE Transactions on Electron

Devices 2001 48(11) p 2606-2624

220

39 Bouloukou A and S Missous Novel High-breakdown Low-noise InGaAs-

InA1As Transistors for Radio Astronomy Applications 2006 University of

Manchester

40 Bean J Materials and technologies 1990 John Wiley amp Sons New York p

13

41 Swaminathan V and A Macrander Materials aspects of GaAs and InP based

structures 1991 Prentice-Hall Inc

42 Vurgaftman I J Meyer and L Ram-Mohan Band parameters for IIIndashV

compound semiconductors and their alloys Journal of applied physics 2001

89(11) p 5815-5875

43 Dingle R W Wiegmann and CH Henry Quantum states of confined carriers

in very thin Al x Ga 1minus x As-GaAs-Al x Ga 1minus x As heterostructures Physical

Review Letters 1974 33(14) p 827

44 Sze SM and KK Ng Physics of semiconductor devices 2006 John wiley amp

sons

45 Tyagi MS Introduction to semiconductor materials and devices 2008 John

Wiley amp Sons

46 Schubert E Delta doping of IIIndashV compound semiconductors Fundamentals

and device applications Journal of Vacuum Science amp Technology A Vacuum

Surfaces and Films 1990 8(3) p 2980-2996

47 Rhoderick EH Metal-semiconductor contacts IEE Proceedings I-Solid-State

and Electron Devices 1982 129(1) p 1

48 Piotrowska A A Guivarch and G Pelous Ohmic contacts to IIIndashV compound

semiconductors A review of fabrication techniques Solid-State Electronics

1983 26(3) p 179-197

49 Rideout V A review of the theory and technology for ohmic contacts to group

IIIndashV compound semiconductors Solid-State Electronics 1975 18(6) p 541-

550

50 Baca A et al A survey of ohmic contacts to III-V compound semiconductors

Thin Solid Films 1997 308 p 599-606

51 Higman T et al Structural analysis of AundashNindashGe and AundashAgndashGe alloyed

ohmic contacts on modulation‐doped AlGaAsndashGaAs heterostructures Journal of

applied physics 1986 60(2) p 677-680

52 Chen KJ et al High-performance enhancement-mode InAlAsInGaAs HEMTs

using non-alloyed ohmic contact and Pt-based buried-gate in Indium Phosphide

and Related Materials 1995 Conference Proceedings Seventh International

Conference on 1995 IEEE

53 Berlin L The man behind the microchip Robert Noyce and the invention of

Silicon Valley 2005 Oxford University Press

54 Goodhue W et al Large room‐temperature effects from resonant tunneling

through AlAs barriers Applied Physics Letters 1986 49(17) p 1086-1088

55 Kerr A and Y Anand Schottky diode MM detectors with improved sensitivity

and dynamic range Microwave Journal 1981 24 p 67-71

56 Davies R Simulations of the current-voltage characteristics of semiconductor

tunnel structures Gec Journal of Research 1987 5(2) p 65-75

221

57 Kelly M Tunnel structures and devices over the coming decade Philosophical

Transactions of the Royal Society of London A Mathematical Physical and

Engineering Sciences 1996 354(1717) p 2291-2293

58 Wilkinson V and M Kelly Manufacturability of quantum semiconductor

devices in High Performance Electron Devices for Microwave and

Optoelectronic Applications 1995 EDMO IEEE 1995 Workshop on 1995

IEEE

59 Wilkinson V M Kelly and M Carr Tunnel devices are not yet

manufacturable Semiconductor Science and Technology 1997 12(1) p 91

60 Eaves L and MJ Kelly The current status of semiconductor tunnelling devices

Philos trans of the Roy soc of London Ser A Math phys and eng sciences

1996 354(1717)

61 Billen K V Wilkinson and M Kelly Manufacturability of heterojunction

tunnel devices further progress Semiconductor Science and Technology 1997

12(7) p 894

62 Kelly M The engineering of quantumndashdot devices Philosophical Transactions

of the Royal Society of London A Mathematical Physical and Engineering

Sciences 2003 361(1803) p 393-401

63 Kelly M New statistical analysis of tunnel diode barriers Semiconductor

Science and Technology 2000 15(1) p 79

64 Hayden R et al Ex situ re-calibration method for low-cost precision epitaxial

growth of heterostructure devices Semiconductor Science and Technology

2002 17(2) p 135

65 Dasmahapatra P et al Thickness control of molecular beam epitaxy grown

layers at the 001ndash01 monolayer level Semiconductor Science and Technology

2012 27(8) p 085007

66 Hayden R M Missous and M Kelly Precision growth for the manufacture of

semiconductor heterostructure devices Semiconductor Science and Technology

2001 16(8) p 676

67 Shao C et al Highly reproducible tunnel currents in MBE-grown

semiconductor multilayers Electronics Letters 2012 48(13) p 792-794

68 Abdullah MR et al GaAsAlAs tunnelling structure Temperature dependence

of ASPAT detectors in Millimeter Waves and THz Technology Workshop

(UCMMT) 2015 8th UK Europe China 2015 IEEE

69 Ariffin KZ et al Asymmetric Spacer Layer Tunnel In0 18Ga0 82AsAlAs

(ASPAT) Diode using double quantum wells for dual functions Detection and

oscillation in Millimeter Waves and THz Technology Workshop (UCMMT)

2015 8th UK Europe China 2015 IEEE

70 Liboff RL Introductory quantum mechanics 2003 Addison-Wesley

71 Esaki L Discovery of the tunnel diode IEEE Transactions on Electron Devices

1976 23(7) p 644-647

72 Landau LD LEM Quantum Mechanics Non-relativistic Theory Pergamon 3

73 Landau LD et al Quantum Mechanics Non‐Relativistic Theory Vol 3 of

Course of Theoretical Physics 1958 AIP

222

74 Syme R Tunnelling devices as microwave mixers and detectors Philosophical

Transactions of the Royal Society of London A Mathematical Physical and

Engineering Sciences 1996 354(1717) p 2351-2364

75 Syme R et al Tunnel diode with asymmetric spacer layers for use as

microwave detector Electronics Letters 1991 27(23) p 2192-2194

76 Brown E W Goodhue and T Sollner Fundamental oscillations up to 200 GHz

in resonant tunneling diodes and new estimates of their maximum oscillation

frequency from stationary‐state tunneling theory Journal of applied physics

1988 64(3) p 1519-1529

77 Reddy M Schottky-collector resonant tunnel diodes for sub-millimeter-wave

applications 1997 University of California Santa Barbara

78 Cox R and H Strack Ohmic contacts for GaAs devices Solid-State Electronics

1967 10(12) p 1213IN71215-1214IN81218

79 Valdes LB Resistivity measurements on germanium for transistors

Proceedings of the IRE 1954 42(2) p 420-427

80 Schroder DK Semiconductor material and device characterization 2006 John

Wiley amp Sons

81 Klootwijk J and C Timmering Merits and Limitations of Circular TLM

structures for contact resistance determination for novel 111-V HBTs Proc

fEEE 2004

82 Marlow GS and MB Das The effects of contact size and non-zero metal

resistance on the determination of specific contact resistance Solid-State

Electronics 1982 25(2) p 91-94

83 Murrmann H and D Widmann Current crowding on metal contacts to planar

devices IEEE Transactions on Electron Devices 1969 16(12) p 1022-1024

84 Berger H Models for contacts to planar devices Solid-State Electronics 1972

15(2) p 145-158

85 Reeves G and H Harrison Obtaining the specific contact resistance from

transmission line model measurements IEEE Electron Device Letters 1982

3(5) p 111-113

86 Shur MS GaAs devices and circuits 2013 Springer Science amp Business

Media

87 Popescu D and B Odbert The Advantages Of Remote Labs In Engineering

Education Educatorrsquos Corner-Agilent Technologies-application note 2011 p

11

88 DataSheet Karl Suss- PM5 Probe System Datasheet 2013

89 Keysight IC-CAP Device Modeling Software 2016 Available from

httpwwwkeysightcomenpc-1297149ic-cap-device-modeling-software-

measurement-control-and-parameter-extractioncc=USamplc=eng

90 DataSheet Anritsu 37369A Vector Network Analyzer Datasheet 2016 Available

from httpwwwtestequipmenthqcomdatasheetsANRITSU-37397D-

Datasheetpdf

91 Packard H HP 4142B Modular DC SourceManual Operation Manual 1992

Available from httpcpliteratureagilentcomlitwebpdf04142-90010pdf

223

92 Microtech C Cascade Microtech- Wincal High Performence RF calaibration

Software (Official Website) 2016 Available from

httpswwwcascademicrotechcom

93 Singh J Electronic and optoelectronic properties of semiconductor structures

2007 Cambridge University Press

94 Whyte W Cleanroom design 1999 Wiley Online Library

95 Vieu C et al Electron beam lithography resolution limits and applications

Applied Surface Science 2000 164(1) p 111-117

96 La Fontaine B Lasers and Moorersquos law SPIE Professional October 2010 p

20

97 Madou MJ Fundamentals of microfabrication the science of miniaturization

2002 CRC press

98 Serway R Physics for Scientists and Engineers 1996 Saunders Publ

Philadelphia

99 Jalali B and S Pearton InP HBTs growth processing and applications 1995

Artech House Publishers

100 Shih YC et al Effects of interfacial microstructure on uniformity and thermal

stability of AuNiGe ohmic contact to n‐type GaAs Journal of applied physics

1987 62(2) p 582-590

101 Zawawi M Advanced In0 8Ga0 2AsAlAs Resonant Tunneling Diodes

forApplications in Integrated mm-waves MMIC Oscillators 2015

102 Zawawi MAM et al Fabrication of Submicrometer InGaAsAlAs Resonant

Tunneling Diode Using a Trilayer Soft Reflow Technique With Excellent

Scalability IEEE Transactions on Electron Devices 2014 61(7) p 2338-2342

103 Silvaco I ATLAS Users Manual Device Simulation Software 2010 Santa

Clara CA

104 Kyono C et al Dependence of apparent barrier height on barrier thickness for

perpendicular transport in AlAsGaAs single‐barrier structures grown by

molecular beam epitaxy Applied Physics Letters 1989 54(6) p 549-551

105 Yang K JR East and GI Haddad Numerical modeling of abrupt

heterojunctions using a thermionic-field emission boundary condition Solid-

State Electronics 1993 36(3) p 321-330

106 Varshni YP Temperature dependence of the energy gap in semiconductors

Physica 1967 34(1) p 149-154

107 Handbook LLM Precision DC Current Voltage and Resistance

Measurements Keithley Instruments Inc[online] 6th revision Ohio 2004

108 Lipsky SE Microwave passive direction finding 2004 SciTech Publishing

109 Howell CM and SJ Parisi Principles Applications and Selection of Receiving

Diodes MACOM Semiconductor Products Division Application note AG314

110 Schulman J D Chow and D Jang InGaAs zero bias backward diodes for

millimeter wave direct detection IEEE Electron Device Letters 2001 22(5) p

200-202

111 Zhang Z et al Sub-Micron Area Heterojunction Backward Diode Millimeter-

Wave Detectors With 018$ rm pWHz^12 $ Noise Equivalent Power IEEE

microwave and wireless components letters 2011 21(5) p 267-269

224

112 Jin N et al High sensitivity Si-based backward diodes for zero-biased square-

law detection and the effect of post-growth annealing on performance IEEE

Electron Device Letters 2005 26(8) p 575-578

113 Shashkin VI et al Millimeter-wave detectors based on antenna-coupled low-

barrier Schottky diodes International journal of infrared and millimeter waves

2007 28(11) p 945-952

114 Zhao P et al GaN Heterostructure Barrier Diodes Exploiting Polarization-

Induced $delta $-Doping IEEE Electron Device Letters 2014 35(6) p 615-

617

115 Pozar DM Microwave engineering 2009 John Wiley amp Sons

116 Koolen M J Geelen and M Versleijen An improved de-embedding technique

for on-wafer high-frequency characterization in Bipolar Circuits and

Technology Meeting 1991 Proceedings of the 1991 1991 IEEE

117 Cao M et al RF characteristics uniformity of GaAsAlAs tunnel diodes in

Infrared Millimeter and Terahertz waves (IRMMW-THz) 2016 41st

International Conference on 2016 IEEE

118 Gao J RF and microwave modeling and measurement techniques for field effect

transistors 2010 SciTec

119 Ren T et al A 340-400 GHz Zero-Biased Waveguide Detector Using an Self-

Consistent Method to Extract the Parameters of Schottky Barrier Diode Applied

Computational Electromagnetics Society Journal 2015 30(12)

120 Fobelets K et al High‐frequency capacitances in resonant interband tunneling

diodes Applied Physics Letters 1994 64(19) p 2523-2525

121 Diebold S et al Modeling and Simulation of Terahertz Resonant Tunneling

Diode-Based Circuits IEEE Transactions on Terahertz Science and Technology

2016 6(5) p 716-723

122 Yong Z et al Design of a 220 GHz frequency tripler based on EM model of

Schottky diodes JOURNAL OF INFRARED AND MILLIMETER WAVES

2014 33(4) p 405-411

123 Louhi JT and AV Raisanen On the modeling and optimization of Schottky

varactor frequency multipliers at submillimeter wavelengths IEEE transactions

on microwave theory and techniques 1995 43(4) p 922-926

124 Guo J Z Zhang and C Qian Modeling of commercial millimeter wave

Schottky diodes in Microwave and Millimeter Wave Technology (ICMMT) 2016

IEEE International Conference on 2016 IEEE

125 Schneider M Metal-semiconductor junctions as frequency converters Infrared

and Millimeter Waves 1982 6 p 209

126 Muth C et al Advanced technology microwave sounder on NPOESS and NPP

in Geoscience and Remote Sensing Symposium 2004 IGARSS04 Proceedings

2004 IEEE International 2004 IEEE

127 Putley E Thermal detectors in Optical and Infrared Detectors 1977 Springer

p 71-100

128 Martin DH Spectroscopic techniques for far infra-red submillimetre and

millimetre waves in Spectroscopic Techniques for Far Infra-red Submillimetre

and Millimetre Waves 1967

225

129 Lucas W Tangential sensitivity of a detector video system with RF

preamplification in Proceedings of the Institution of Electrical Engineers 1966

IET

130 Balocco C et al Low-frequency noise of unipolar nanorectifiers Applied

Physics Letters 2011 99(11) p 113511

131 Benford D T Hunter and TG Phillips Noise equivalent power of background

limited thermal detectors at submillimeter wavelengths International journal of

infrared and millimeter waves 1998 19(7) p 931-938

132 Papoušek D Vibration-rotational Spectroscopy and Molecular Dynamics

Advances in Quantum Chemical and Spectroscopical Studies of Molecular

Structures and Dynamics Vol 9 1997 World Scientific

133 Nyquist H Thermal agitation of electric charge in conductors Physical review

1928 32(1) p 110

134 Turner CS Johnson-Nyquist Noise url httpwww claysturner

comdspJohnson-NyquistNoise pdf(Letzter Abruf Juli 2012)

135 Voss RF 1f (flicker) noise A brief review in 33rd Annual Symposium on

Frequency Control 1979 1979 IEEE

136 McWhorter AL 1f noise and related surface effects in germanium 1955

137 Hooge FN 1ƒ noise is no surface effect Physics letters A 1969 29(3) p 139-

140

138 Van der Ziel A Noise Sources characterization measurement Prentice-Hall

Information and System Sciences Series Englewood Cliffs Prentice-Hall 1970

1970

139 Hooge F 1f noise sources IEEE Transactions on Electron Devices 1994

41(11) p 1926-1935

140 Der Ziel A Theory of shot noise in junction diodes and junction transistors

Proceedings of the IRE 1955 43(11) p 1639-1646

141 Schottky W Small-shot effect and flicker effect Physical review 1926 28(1)

p 74

142 Cowley A and H Sorensen Quantitative comparison of solid-state microwave

detectors IEEE transactions on microwave theory and techniques 1966 14(12)

p 588-602

143 Schulman J et al 1$f $ Noise of Sb-Heterostructure Diodes for Pre-Amplified

Detection IEEE microwave and wireless components letters 2007 17(5) p

355-357

144 Lynch JJ et al Passive millimeter-wave imaging module with preamplified

zero-bias detection IEEE transactions on microwave theory and techniques

2008 56(7) p 1592-1600

145 Su N et al Sb-heterostructure millimeter-wave detectors with reduced

capacitance and noise equivalent power IEEE Electron Device Letters 2008

29(6) p 536-539

146 Westlund A Self-Switching Diodes for Zero-Bias Terahertz Detection 2015

Chalmers University of Technology

147 Yajima T and L Esaki Excess noise in narrow germanium pn junctions

Journal of the physical society of Japan 1958 13(11) p 1281-1287

226

148 Sommers H Tunnel diodes as high-frequency devices Proceedings of the IRE

1959 47(7) p 1201-1206

149 Miraftab V and A Abdipour Harmonic balance analysis of a microwave

balanced power amplifier in Electrical and Computer Engineering 2001

Canadian Conference on 2001 IEEE

150 Patrashin M et al GaAsSbInAlAsInGaAs Tunnel Diodes for Millimeter Wave

Detection in 220ndash330-GHz Band IEEE Transactions on Electron Devices 2015

62(3) p 1068-1071

151 Hesler JL and TW Crowe Responsivity and noise measurements of zero-bias

Schottky diode detectors Proc ISSTT 2007 p 89-92

152 Su N et al Temperature dependence of high frequency and noise performance

of Sb-heterostructure millimeter-wave detectors IEEE Electron Device Letters

2007 28(5) p 336-339

153 Lynch J et al Unamplified direct detection sensor for passive millimeter wave

imaging in Proc of SPIE Vol 2006

154 ZHANG Z et al A physics-based tunneling model for Sb-heterostructure

backward tunnel diode millimeter-wave detectors International Journal of High

Speed Electronics and Systems 2011 20(03) p 589-596

155 Bahl IJ and P Bhartia Microwave solid state circuit design 2003 John Wiley

amp Sons

156 Yeom K-W Microwave Circuit Design A Practical Approach Using ADS

2015 Prentice Hall Press

157 Xie L et al A W-band detector with high tangential signal sensitivity and

voltage sensitivity in Microwave and Millimeter Wave Technology (ICMMT)

2010 International Conference on 2010 IEEE

158 Fay P et al High-performance antimonide-based heterostructure backward

diodes for millimeter-wave detection IEEE Electron Device Letters 2002

23(10) p 585-587

159 Hrobak M et al Planar zero bias Schottky diode detector operating in the E-

and W-band in Microwave Conference (EuMC) 2013 European 2013 IEEE

160 Maas SA Nonlinear microwave and RF circuits 2003 Artech House

161 Appleby R and RN Anderton Millimeter-wave and submillimeter-wave

imaging for security and surveillance Proceedings of the Ieee 2007 95(8) p

1683-1690

162 Crowe TW et al Opening the terahertz window with integrated diode circuits

IEEE Journal of Solid-State Circuits 2005 40(10) p 2104-2110

163 Raisanen AV Frequency multipliers for millimeter and submillimeter

wavelengths Proceedings of the Ieee 1992 80(11) p 1842-1852

164 Erickson NR Diode frequency multipliers for terahertz local-oscillator

applications in Astronomical Telescopes amp Instrumentation 1998 International

Society for Optics and Photonics

165 Mehdi I et al Terahertz local oscillator sources performance and capabilities

in Astronomical Telescopes and Instrumentation 2003 International Society for

Optics and Photonics

166 Tonouchi M Cutting-edge terahertz technology Nature photonics 2007 1(2)

p 97-105

227

167 Nilsen SM et al Single barrier varactors for submillimeter wave power

generation IEEE transactions on microwave theory and techniques 1993 41(4)

p 572-580

168 Xiao Q et al A 270-GHz tuner-less heterostructure barrier varactor frequency

tripler IEEE microwave and wireless components letters 2007 17(4) p 241-

243

169 David T et al Monolithic integrated circuits incorporating InP-based

heterostructure barrier varactors IEEE microwave and wireless components

letters 2002 12(8) p 281-283

170 Lieneweg U B Hancock and J Maserjian Barrier-intrinsic-N+(BIN) diodes

for near-millimeter wave generation in Conference Digest for the Twelft

International Conference on Infrared and Millimeter Waves 1987

171 Lieneweg U et al Modeling of planar varactor frequency multiplier devices

with blocking barriers IEEE transactions on microwave theory and techniques

1992 40(5) p 839-845

172 Chattopadhyay G et al An all-solid-state broad-band frequency multiplier

chain at 1500 GHz IEEE transactions on microwave theory and techniques

2004 52(5) p 1538-1547

173 Maestrini A et al A 17-19 THz local oscillator source IEEE microwave and

wireless components letters 2004 14(6) p 253-255

174 Faber MT J Chramiec and ME Adamski Microwave and millimeter-wave

diode frequency multipliers 1995 Artech House Publishers

175 Frerking MA and JR East Novel heterojunction varactors Proceedings of the

Ieee 1992 80(11) p 1853-1860

176 Penfield P and RP Rafuse Varactor applications 1962

177 Palazzi V et al Low-power frequency doubler in cellulose-based materials for

harmonic RFID applications IEEE microwave and wireless components letters

2014 24(12) p 896-898

178 Presas SM Microwave frequency doubler integrated with miniaturized planar

antennas 2008

228

APPENDICES

Appendix I Doped substrate process details

Mask Stage Process Stage Process step Process detail Equipment

Mask 1 (Mesa

Etch)

Sample clean NMP 10 min 80˚C Beaker

Acetone 5 min Beaker

Isopropanol (IPA) 5 min Beaker

Apply Resist Prebake Bake for 5mins 150C Hotplate

1st Resist S1805 - Program 4 Laurell Spinner

Post bake 110C for 1 Min Hotplate

Photolithography Mask Align mask to wafer

MA4 Mask

aligner

Expose

1 min 09mW iline

(Compensation error set to 1)

MA4 Mask

aligner

Develop MIF 319 for 1 min Beaker

Post Exposure

Bake Oven or Bake 120C for 15

minutes Hotplate

Etch Etch Cal Cal Sample - etch for 2 minutes Beaker

Etch

H3PO4H2O2H2O 3150

time is determined by the etch cal Beaker

Resist Strip Acetone - 5 min and IPA - 5 min Beaker

Mask2 (Top

Contact)

Sample clean NMP ultrasonic for 15 Min Power 3 Beaker

Acetone ultrasonic for 5 Min Power 3 Beaker

Isopropanol (IPA) ultrasonic for 5 Min Power 3 Beaker

Apply Resist Prebake Bake for 5mins 150C Hotplate

1st Resist AznLoF - 2um grade - Program 6 Laurell Spinner

Post bake 110C for 1 Min Hotplate

Photolithography Mask Align mask to wafer

MA4 Mask

aligner

Expose

55 seconds 09mW i-line

(Compensation error set to 1)

MA4 Mask

aligner

Post Exposure bake 110C for 1 Min Hotplate

Develop MIF 326 for 1 mins Beaker

Clean O2 Plasma Etch

20sec 100W forward (340)

60mTorr Plasma Tech

Surface De-oxide HCLH2O 11 40 sec Beaker

Metallisation Evaporation Clean trays + metals Beaker

Load sample and metals

Edward

Evaporator

AuGe (100mg) deposits 55nm

Edward

Evaporator

Ni (1cm) deposits 15nm

Edward

Evaporator

Au (10cm) deposits 200nm

Edward

Evaporator

Lift-off

NMP1165 for 24 hours (25C) or

20 min 80C Beaker

Anneal 420˚C 2min Furnace

229

TLM

Measurement ICCAP

Measurement

Bench

Mask 3

(BottomBacks

ide Contact)

Sample clean Acetone Optional Beaker

Isopropanol (IPA) Optional Beaker

Apply Resist Prebake

Bake for 5mins 150C to dry

the sample Hotplate

Resist (top side) S1813 - Program 6 Laurell Spinner

Soft bake 110C for 1 Min Hotplate

Photolithography Mask Align mask to wafer

MA4 Mask

aligner

Exposure

10 seconds 09mW I-line

(Compensation error set to 1)

MA4 Mask

aligner

Post Exposure bake 110C for 1 Min Hotplate

Develop MIF 319 for 2 mins Beaker

De-scum O2 Plasma Etch

20sec 100W forward (340)

60mTorr Plasma Tech

De-oxidise

Surface De-oxide

HCLH2O 11 40

sec Beaker

metallisation Evaporation Clean trays + metals Beaker

Load sample and metals

Edward

Evaporator

AuGe (100mg) deposits 55nm

Edward

Evaporator

Ni (1cm) deposits 15nm

Edward

Evaporator

Au (10cm) deposits 200nm

Edward

Evaporator

Lift-off

NMP1165 for 24 hours (25C) or

20 min 80C Beaker

Clean Water 3min Beaker

TLM

Measurement ICCAP

230

Appendix II Four Mask step Process Flow

Mask Stage Process Stage Process step Process detail

Mask1 Top

Contact)

Sample clean NMP ultrasonic for 15 Min Power 3

Acetone ultrasonic for 5 Min Power 3

Isopropanol (IPA) ultrasonic for 5 Min Power 3

Apply Resist Prebake Bake for 5mins 150C

1st Resist AznLoF - 2um grade - Program 6

Hot Plate 110C for 1 Min

Photolithography Mask Align mask to wafer

Expose 55 seconds 09mW iline (Compensation

error set to 1)

Post Exposure bake 110C for 1 Min

Develop MIF 326 for 1 mins

Clean O2 Plasma Etch 20sec 100W forward (340) 60mTorr

Surface De-oxide HCLH2O 11 40 sec

Metalisation Evaporation Clean trays + metals

Load sample and metals

AuGe (100mg) deposits 55nm

Ni (1cm) deposits 15nm

Au (10cm) deposits 200nm

Lift-off NMP1165 for 24 hours (25C) or 20 mnt 80C

Mask 2 (Mesa

Etch)

Sample clean NMP Optional

Acetone Optional

Isopropanol (IPA) Optional

Apply Resist Prebake Bake for 5mins 150C

1st Resist S1805 - Program 4

HotPlate 110C for 1 Min

Photolithography Mask Align mask to wafer

Expose 1 min 09mW i-line (Compensation error set

to 1)

Develop MIF 319 for 1 min

Post Exposure Bake Oven or Bake 120C for 15 minutes

Etch Etch Cal Cal Sample - etch for 2 minutes

Etch H3PO4H2O2H2O 3150 time is

determined by the etch cal

Measure TLM

Resist Strip Acetone - 5 min and IPA - 5 min

Mask 3(Isolation) Sample clean Acetone Optional

Isopropanol (IPA) Optional

231

Apply Resist Prebake Bake for 5mins 150C

Resist S1828 - Program 4

Hot Plate 115C for 1 mins

Photolithography Mask Align mask to wafer

Expose 9 mins 09mW iline (Compensation error set

to 1)

Develop MF 319 3 mins

Post Bake Oven Bake 120C for 15mins

Etch Etch Cal Refer to Etch Cal instr tab

Sub-collector Etch H3PO4H2O2H2O 3150 time is

determined by the etch cal

Resist Strip Acetone 5mins + IPA 5 mins in ultrasonic bath

power 1

inspection Microscope

Sample clean Acetone Optional

Mask 4 (Bottom

Contact)

Isopropanol (IPA) Optional

Apply Resist Prebake Bake for 5mins 150C 1st

Resist AznLoF - 2um grade - Program 6

HotPlate 110C for 1 Min

Photolithography Mask Align mask to wafer

Expose 10 seconds 09mW i-line (Compensation

error set to 1)

Post Exposure bake 110C for 1 Min

Develop MIF 326 for 2 mins

Clean O2 Plasma Etch 20sec 100W forward (340) 60mTorr

Surface De-oxide

HCLH2O 11 40 sec

Metalisation Evaporation Clean trays + metals

Load sample and metals

AuGe (100mg) deposits 55nm

Ni (1cm) deposits 15nm

Au (10cm) deposits 200nm

Lift-off NMP1165 for 24 hours (25C) or 20 mnt 80C

232

Appendix III Epitaxial Layer XMBE277

TABLE I The epitaxial structure for sample XMBE277

Layer Thickness (nm) Doping Concentration (cm-3)

n+- In053Ga047As 45 200 x 1019

n- In053Ga047As 25 300 x 1018

In053Ga047As 20 undoped

AlAs 13 undoped

In08Ga02As 45 undoped

AlAs 13 undoped

In053Ga047As 20 undoped

n- In053Ga047As 25 300 x 1018

n+- In053Ga047As 400 100 x 1019

Semi-insulating InP

233

Appendix IV SilVaco (Atlas) Simulation Code

go atlas

---------------------------------------------------------

Structure parameter definition (Constants) values in um

---------------------------------------------------------

Thicknesses

set t_contact1=0

set t_ohmic1=03

set t_emitter=004

set t_spacer1=0005

set t_barrier=000283

set t_spacer2=02

set t_collector=004

set t_ohmic2=045

Doping concentrations

set d_ohmic1=4e18

set d_emitter=2e17

set d_collector=2e17

set d_ohmic2=4e18

set d_gap=2

set d_mesa=4

set d_device=10

set d_etch=008

Layers

set I=$t_contact1

set A=$I+$t_ohmic1

set B=$A+$t_emitter

set C=$B+$t_spacer1

set D=$C+$t_barrier

set E=$D+$t_spacer2

set F=$E+$t_collector

set G=$F+$t_ohmic2

-------------------------------------

Mesh generator

-------------------------------------

mesh diagflip width=45

xmesh location=0 s=1

xmesh location=1 s=1

xmesh location=2 s=1

xmesh location=4 s=1

xmesh location=5 s=1

xmesh location=6 s=1

xmesh location=7 s=1

xmesh location=8 s=1

xmesh location=$d_mesa s=1

xmesh location=$d_mesa+$d_gap s=1

xmesh location=$d_device s=1

Ohmic1

ymesh l=0000 s=005

ymesh l=$I s=005

234

ymesh l=$A s=0005

ymesh l=$B s=0005

ymesh l=$C s=00005

ymesh l=$D s=00005

ymesh l=$E s=0009

ymesh l=$F s=0005

ymesh l=$G s=0005

-----------------------------------

SECTION 2 Regions Structure definition

-----------------------------------

region num=1 name=contact1 material=Gold ymin=0 ymax=$I

region num=2 name=ohmic1 material=GaAs ymin=$I ymax=$A

region num=3 name=emitter material=GaAs ymin=$A ymax=$B

region num=4 name=spacer1 material=GaAs ymin=$B ymax=$C

region num=5 name=barrier material=AlAs ymin=$C ymax=$D xmin=0 xmax=$d_mesa

calcstrain qtregion=1

region num=6 name=spacer2 material=GaAs ymin=$D ymax=$E

region num=7 name=collector material=GaAs ymin=$E ymax=$F

region num=8 name=ohmic2 material=GaAs ymin=$F ymax=$G

region num=9

name=etch material=Air ymin=0 ymax=$F+$d_etch xmin=$d_mesa xmax=$d

_device

---------------------------------

Electrodes

---------------------------------

electrode num=1 name=anode xmin=0 xmax=$d_mesa ymin=0 ymax

=$I material=Gold

electrode num=2 name=cathode xmin=$d_mesa+$d_gap xmax=$d_device ymin=$F+

$d_etch ymax=$F+$d_etch material=Gold

--------------------------------

Doping

--------------------------------

doping uniform ntype conc=$d_ohmic1 Region=2 ymin=$I ymax=$A

doping uniform ntype conc=$d_emitter Region=3 ymin=$A ymax=$B

doping uniform ntype conc=$d_collector Region=7 ymin=$E ymax=$F

doping uniform ntype conc=$d_ohmic2 Region=8 ymin=$F ymax=$G

--------------------------

Contacts

--------------------------

interface sc region=1

interface ss region=2

interface ss region=3

interface si region=4

interface si region=5

interface ss region=6

interface ss region=7

interface sc ymin=$F ymax=$F xmin=$d_mesa+$d_gap xmax=$d_device

interface tunnel region=5 dytunnel=0001

contact name=cathode

contact name=anode

235

------------------------------------------

SECTION 3 Material amp Models Definitions

------------------------------------------

material material=AlAs

permittivity=10 eg300=28 mc=004 affinity=305 nc300=4e19 nimin=1e1

material material=GaAs permittivity=139 eg300=14 mc=0067

affinity=407 nc300=09e17 nimin=1e6

BAND DIAGRAM

output tquantum bandparam qfn qfp valband conband charge polarcharge flowlines

STRUCTURE GRAPHIC

solve init

save outf=XMBE304+real2str

tonyplot XMBE304+real2str

------------------------------------------

SECTION 4 ANALYSIS

------------------------------------------

trap acceptor structure=top elevel=03 density=48e17 degenfac=10 sign=284e-17 sigp=1e-17

ymin=$A ymax=$C xmin=0 xmax=$d_mesa

trap acceptor structure=BOTTOM elevel=035 density=47e17 degenfac=10 sign=284e-17 sigp=1e-17

ymin=$E ymax=$F xmin=0 xmax=$d_mesa

models sisel sisnlderivs qtregion=1 print

method climit=1e-4 itlimit=50 maxtraps=20

DC ANALYSIS

log outf=XMBE304log

solve init

solve vanode=0 name=anode vstep=001 vfinal=15

save outf=XMBE304str

log off

tonyplot XMBE304str

tonyplot XMBE304log

Page 4: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre

4

453 Dependence of current on Spacer II length l2 138

46 Temperature Dependent Simulation 140

47 Temperature Dependence characteristics of GaAsAlAs ASPAT Diodes 142

48 Conclusions 147

5 DC amp RF CHARACTERISTIC OF ASPAT DIODES 148

51 Introduction 148

52 General Analysis of Current-Voltage (I-V) characteristics of GaAsAlAs

ASPAT diodes 149

53 RF Test Fixture Theory and Experiment 153

531 On-Wafer Measurement and Small Signal One-Port Characterizations

154

54 Device Calibration 155

541 Open and Short De-Embedding Technique 155

55 S-Parameter Measurement Result and Analysis 157

551 Diode to diode uniformity 158

552 Wafer to wafer uniformity 160

553 Small devices RF measurements 161

56 Extracting RF models of ASPAT at Zero Bias Voltage 164

561 Extraction of ASPAT parasitic element 165

562 Extraction of ASPAT intrinsic elements 168

563 Capacitances -Voltage (C-V) Extraction 172

57 Conclusions 173

6 MILLIMETRE WAVE GAASALAS ASPAT DETECTOR DESIGN USING

ADS 175

61 Introduction 175

62 Detection Theory 176

63 Definition of Characteristics of Interest Detection Parameters in small signal

analysis 179

64 Noise Consideration in a Detector diode 182

65 Modelling of a 100GHz Zero-biased ASPAT Detector 184

66 Conclusions 199

7 FREQUENCY MULTIPLIERS (DOUBLER) DESIGN USING GAASALAS

ASPAT DIODES 201

71 Introduction 201

5

72 Motivation and Background 202

73 Frequency Multiplier Architecture the Basics 203

731 Types of frequency multipliers 205

74 Parameters of interest for Frequency Multipliers 206

75 20GHz40GHz Varistor Mode ASPAT Diode Frequency Doubler 208

76 Conclusions 214

8 CONCLUSION AND FUTURE WORK 215

81 Conclusion 215

82 Future Work 217

REFERENCES 218

APPENDICES 228

Word count including footnotes and endnotes 61500(approximately)

6

LIST OF TABLES

Table 11 Sample XMBE304 XMBE307 and XMBE368 GaAsAlAs materials

structure grown on GaAs Substrates by MBE 23

Table 12 Quantum wells sandwiching the quantum barrier for sample XMBE314

grown on a GaAs substrate by MBE 24

Table 13 Novel InGaAsAlAs ASPAT diode grown on InP substrate by MBE on InP

substrate 24

Table 21 Lattice constant (a) and band gap for group III-V binary and ternary

compound semiconductors a room temperature [41 42] 38

Table 22 epitaxial layer of sample XMBE304 which is the main focus of this work 56

Table 31 Chemical wet etch characteristics using Orthophosphoric-based and

Ammonia on GaAs and InGaAs materials 87

Table 32 Epitaxial layer of Doped substrate samples 93

Table 33 The epitaxial layers of sample XMBE304 with barrier 10ML~28nm 94

Table 34 Generic fabrication steps established by Dr Md Adzhar [101] 97

Table 35 Standard process flow for Air-Bridge design fabrication 101

Table 36 Standard fabrication process flow for Dielectric-Bridge design 107

Table 37 New arrangement of the mask number and step in Second Run 110

Table 38 New arrangement for the Third run using Dielectric-Bridge mask 113

Table 39 The outcome of the spreading resistance before and after using LOR

technique 115

Table 310 DC and RF characteristics for XMBE304 118

Table 311 3rd

Gen Mask process step 119

Table 312 Standard deviation at two different voltages 124

Table 41 The parameter values used in this simulation 134

Table 42 The calculated values of bandgap at different temperatures 140

Table 43 The calculated effective masses for each temperature used in this simulation

141

Table 44 Epitaxial layer profile for Schottky Barrier Diode (XMBE104) 145

Table 51 Performance of the ASPAT diode obtained from measured IV characteristics

in this work 153

Table 52 Device to device uniformity check for large ASPAT diode 159

Table 53 The variation of reflection coefficient (S11) for GaAsAlAs ASPAT diode at

four different frequencies[117] 159

7

Table 54 Wafer to wafer uniformity check on XMBE304A and XMBE304B 161

Table 55 Comparison between calculated (fully Depleted) and extracted (different

biases) values from equivalent circuit parameters for different ASPAT mesa sizes at zero

bias voltage 170

Table 61 A summary of all the important parameters of the 4x4 microm2 diode 185

Table 62 Noise Equivalent Power (NEP) for zero bias detector diode 190

Table 63 Summary of characteristic of interest for GaAsAlAs ASPAT detector 198

Table 64 Comparison 4times4microm2 GaAsAlAs ASPAT diode to the-state-of-the-art zero

bias detector at W-band (75GHz-110GHz) 199

Table 71 Comparison parameter obtained from this simulation to the state-of-the-art

multiplier diode 213

8

LIST OF FIGURES

Figure 21 III-V compound semiconductors mobility and band gap[24] 31 Figure 22 illustration of Homojunctions band structure material before (left) and after

(right) equilibrium 34 Figure 23 Energy band diagrams before (left) and after (after) interface combination at

equilibrium 35

Figure 24 Lattice Matching for both materials when aL=aS 37 Figure 25 Energy gap and lattice constant for direct and indirect band gap of compound

semiconductor at ambient temperature[40] 37 Figure 26 Lattice mismatched material 39

Figure 27 The formation of Pseudomorphic-layer in two situations (a) compressive and

(b) tensile strain [1] 40 Figure 28 The band diagram of an ideal quantum well formed by a narrow band gap

semiconductor eg InGaAs sandwiched between two large band gap semiconductors eg

GaAs (a) Structure (b) energy band diagram and (c) Conduction band diagram when

AlGaAs is n-doped[43] 41 Figure 29 Metal and semiconductor in two conditions (a) separation (b) in contact 43

Figure 210 Energy band diagram of Schottky contact on n-type material under (a)

reverse and (b) forward bias 45 Figure 211 Figure 10 N-type semiconductor Ohmic contact with different doping

concentration ND (a) Low (b) Intermediate and (c) high 47 Figure 212 Classical view of whether an electron is can surmount a barrier or not

Quantum mechanical view allows an electron to tunnel through a barrier The probability

(blue) is related to the barrier thickness 51

Figure 213 Rectangular tunnelling with incident reflected and transmitted wave

function[70] 52

Figure 214 Epitaxial layer structure of device XMBE 304 (lateral structure) used in

this study 55 Figure 215 The layer profile of ASPAT with 28nm barrier thickness [27] 57

Figure 216 Conduction band diagram showing band bending and 2DEG formation at

the L1 spacer 60 Figure 217 I-V characteristics of a fabricated ASPAT diode 63 Figure 218 Intrinsic Elements of the Asymmetric Spacer Tunnel Layer (ASPAT) Diode

63 Figure 219 Cross-sectional view of ASPAT vertical structure the dimensions of d b

and h are not drawn to scale 65 Figure 220 Cross-section view of the complete process ASPAT diode (XMBE304)

The dimensions are not drawn to scale 67 Figure 221 A simple TLM structure with effective length and sheet resistance

underneath 69 Figure 222 Top view of TLM ladder structure use in this work 71 Figure 223 Typical plot of resistance versus TLM spacing 71

Figure 224 The set of DC measurement apparatus arrangement to measure the TLM 73

9

Figure 225 The set of RF measurement for on-wafer DC and RF one-port measurement

75 Figure 226 Actual VNA system that was used for RF characterization 75 Figure 227 Block diagram of the ASPAT measurement step 77

Figure 31 3D illustration of Optical lithography process used in this research 85 Figure 32 Actual picture of thermal evaporator used in this study 89 Figure 33 Single layer lift-off process using negative photoresist 91 Figure 34 Current-Voltage characteristic of sample XMBE368 used in this study at

two different locations on the wafer tile 96

Figure 35 ASPAT mask design for (a) 100times100microm2 (b) 30times30microm

2 and (c) 15times15 microm

2

diode dimensions designed in the 1st Gen Mask 96

Figure 36 Air Bridge design (red circle) for large mesa area (100times100microm2) device with

option for doped substrates 100 Figure 37 The layout of 1

st design of Dielectric Bridge (green circle) mask design for

100 times 100microm2 emitter size with option for doped substrate processing 100

Figure 38 Dry Etching for the first run in this study 102 Figure 39 Severe undercut of 2times2 microm

2 and 6times6 microm

2 devices 103

Figure 310 Device with emitter area of 2times2 microm2 and 6times6 microm

2 after dry etch and wet

etched 104 Figure 311 SEM Images of the GaAs sample 104

Figure 312 The cross-sectional view of InGaAs (left) and GaAs (right) samples used in

this study 105

Figure 313 SEM images taken from the XMBE304 samples and a GaAs test sample

105

Figure 314 SEM images for InP and InGaAs taken from [56] 105 Figure 315 Short circuit behaviour on one of the fabricated device in this run 106

Figure 316 The surface of the sample after final processing 108 Figure 317 IV characteristics of sample XMBE304 for emitter sizes of 1000 microm2

2500 microm2 900 microm2 400 microm2 225 microm2 100 microm2 and 36 microm2 109

Figure 318 Optical images after MESA etching (a) 1 microm Tolerance (b) 05 microm

Tolerance 110

Figure 319 After lift-off processing 111 Figure 320 The current density of XMBE304 processed using Si3N4 as dielectric layer

112

Figure 321 side view of lateral ASPAT structure 113

Figure 322 The measured size of the emitter area and the length D (blue color marked)

114 Figure 323 Summary of LOR technique steps 115 Figure 324 IV characteristic in 3rd run (reduced d-Gap) for each diode size on the mask

116

Figure 325 3rd

Gen Mask layout showing actual devices de-embedding TLM and

alignment mark structures used in this study 118 Figure 326 Step by step device fabrication (in 3D and cross-sectional view) for semi-

insulating substrate device type used in this study 121 Figure 327 Example finished process device with bond pad using 3

rd Gen mask 121

Figure 328 XMBE304 TLM measurement for the top contact after annealing 122

10

Figure 329 XMBE304 TLM measurement for the bottom contact after annealing 122 Figure 330 XMBE304 GaAsAlAs ASPAT measured IV characteristics at room

temperature for 4times4microm2 mesa size 123

Figure 331 XMBE304 GaAsAlAs ASPAT measured IV characteristics at room

temperature for 6times6microm2 mesa size 124

Figure 332 XMBE304 GaAsAlAs ASPAT measured IV characteristics at room

temperature for 10times10microm2 mesa size 124

Figure 41 SILVACO Atlas simulation process flow 128 Figure 42 The GaAsAlAs ASPAT structures produced from simulation based on the

diode multilayer heterostructures on the right 131 Figure 43 The Energy-momentum diagram for AlAs heterojunction semiconductor 131 Figure 44 (a) The conduction band and the valence band of the ASPAT diode structure

(b) the energy band diagram of the ASPAT diode structure when under three different

biases 132 Figure 45 The current-voltage characteristic of an ASPAT diode (100times100microm

2) and

(4times4microm2) using SILVACO Atlas simulator for structure device XMBE304 showing

excellent agreement between simulated and experimental data 134

Figure 46 IV characteristics of the dependencies of current on AlAs barrier 136 Figure 47 Example of analysis at -1 and 1V to the current 136 Figure 48 I-V characteristic of the dependencies current to Spacer I layer 137

Figure 49 Current changes with layer thickness l1 138 Figure 410 IV characteristic of the dependencies current to Spacer 1 layer 139

Figure 411Current change with layer thickness l2 139

Figure 412 Measurement and simulation comparison result as a function of temperature

range from 100K to 398K 142 Figure 413 Log Current vs voltage as a function of temperature for ASPAT sample

XMBE304 143

Figure 414 The effective barrier high of the Schottky barrier diode and the ASPAT

Diode [3] 144

Figure 415 Log Current vs voltage as a function of temperature for SBD sample

XMBE104 145

Figure 416 Influence of temperature on IV characteristic for both ASPAT diode and

SBD 146

Figure 51 IV Characteristics of measured ASPAT diodes for emitter sizes of 4x4um2

6x6um2 and 10x10um

2 Note the good scalability 149

Figure 52 Junction resistance versus voltage 151

Figure 53 The rectifying action of the diode at zero voltage which gives the ASPAT

high sensitivity near zero bias detection 152

Figure 54 One port S-parameter measurements 155

Figure 55 Fabricated open short de-embedding structure and real device (zoomed) use

for RF calibration and measurements (Note Images are not to scale) 156

11

Figure 56 Real S11 RF measurements for (XMBE304A) were done for four devices

from 15times15microm2 to 100times100microm2 device sizes for within-wafer uniformity check 158

Figure 57 Imaginary S11 measurements (XMBE304A) to qualify the repeatability and

reproducibility were done on four devices from 15times15 microm2 to 100times100 microm2 at zero

bias 158

Figure 58 Wafer to wafer uniformity test on 1st and 2

nd RF measurement(XMBE304B)

to qualify the repeatability and reproducibility were done for 30times30microm2 to 15times15 microm

2

device sizes (Real and Imaginary) Note blue colour is XMBE304A and red colour is

XMBE304B 160

Figure 59 RF measurement performed on device sizes of 10times10microm2 6times6mmicro

2 and

4times4microm2

(Real and Imaginary) Note that green red and blue colour represents 4times4microm2

6times6mmicro2 and 10times10microm

2 diodes respectively 162

Figure 510 RF measurement performed on device sizes of 10times10 microm2 6times6 microm2 and

4times4 microm2 (Smith Chart) Note that green red and blue colour represents 4times4microm2

6times6mmicro2 and 10times10microm2 diodes respectively 162

Figure 511 Cross-sectional of fabricated ASPAT diode and the corresponding

equivalent circuit model 164

Figure 512 The S-parameter Touchstone file is used to read the measured files 165

Figure 513 Equivalent Circuit Model of Open CPW or de-embedded structure 166

Figure 514 Equivalent circuit model for short de-embedded structure 166

Figure 515 Smith chart representative S-parameter measurement for short (left) and

open (right) CPW The blue lines represent simulated data and the red is measured data

167

Figure 516 Equivalent circuit of the ASPAT diode 169

Figure 517 Excellent fitting between measured (red colour) and simulated (blue colour)

for S11 (Real and Imaginary) results for various small device designs 169

Figure 518 Excellent fitting between measured (red colour) and simulated (blue colour)

for S11 results (Smith Chart) for various small device designs 170

Figure 519 Capacitance extraction showing full depletion at -025 volts (modelled

capacitance vs Voltage) 172

Figure 61 Block diagram represent a complete direct receiver system 177

Figure 62 The detection process of a single wave through a non-linear IV characteristic

of a diode 177

Figure 63 Lumped element illustration of microwave detector circuit 178

Figure 64 The mixing process where the signals are processed by the non-linear I-V

characteristic to produce an intermediate frequency (fIF=fLO-fRF) when signals at fLO and

fRF are applied to the diode 179

Figure 65 Measurement of Tangential Sensitivity[108 129] 181

12

Figure 66 The configuration circuit to verify 10th order polynomial equation extracted

from MATLAB to realize a virtual GaAsAlAs ASPAT diode 186

Figure 67 Verification of actual (blue measured) and virtual (red_10th order

polynomial) I-V characteristic of the 4times4 microm2 diode used in this study 186

Figure 68 Direct detector circuit topology using an ASPAT diode 187

Figure 69 Output voltage and detector sensitivity over wide range of input power 188

Figure 610 Simulated dependence of Voltage Sensitivity on incident power and load

resistance of the ASPAT detector 189

Figure 611 Junction resistance as a function of forward voltage 189

Figure 612 Curvature coefficient for the measured I-V characteristic with a device size

of 4times4μm2 191

Figure 613 Voltage Sensitivity against frequency at -20dBm fixed input power 192

Figure 614 Reflection Coefficient versus operating frequency without matching

circuitry 193

Figure 615 Detector circuit with impedance matching circuit placed in between diode

and source 194

Figure 616 Reflection Coefficient over wide frequency band with matching 195

Figure 617 Simulated voltage sensitivity of GaAsAlAs ASPAT detector at W-band

frequency 195

Figure 618 Lowest detectable signal at 100GHz operating frequency 196

Figure 619 The dynamic range of the ASPAT diode obtained from linear region of

diode operation 197

Figure 620 Simulation result of diode sensitivity for three devices geometries obtained

from the fabricated ASPAT in this work 198

Figure 71 performance of state-of the-art millimetre wave source [166] 202

Figure 72 Frequency multiplier has the property that foutput is an integer multiple of fin

204

Figure 73 Principle of operation for frequency multiplier utilising a non-linear

resistance [10] 204

Figure 74 A standard system for two port frequency multiplier circuit 207

Figure 75 Block diagram of frequency doubler with series ASPAT mounted diode 209

Figure 76 Empirical modelling of ASPAT doubler utilising Keysight ADS tool 210

Figure 77 Conversion loss and conversion efficiency as a function of input power 211

Figure 78 Conversion loss and Conversion efficiency as a function of output frequency

212

13

LIST OF PUBLICATIONS CONFERENCE PRESENTATIONS

PUBLICATIONS

1 MRR Abdullah Y K Wang J Sexton M Missous and M J Kelly ldquoGaAsAlAs

Tunnelling Structure Temperature Dependence of ASPAT Detectorsrdquo 8th UK-Europe-

China Workshop on mm-waves and THz Technologies 2015 Cardiff University IEEE

proceedings DOI 101109UCMMT20157460591

2 Yuekun Wang Mohd Rashid Redza Abdullah James Sexton and M Missous

ldquoInGaAs-AlAs asymmetric space layer tunnel (ASPAT) diodes for THz electronicsrdquo 8th

UK-Europe-China Workshop on mm-waves and THz Technologies 2015 Cardiff

University IEEE proceedings DOI 101109UCMMT20157460589

3 K N Zainul Ariffin S G Muttlak M Abdullah M R R Abdullah Y Wang and M

Missous ldquoAsymmetric Spacer Layer Tunnel In018Ga082AsAlAs (ASPAT) Diode using

Double Quantum Wells for Dual Functions Detection and Oscillationrdquo 8th UK-

Europe-China Workshop on mm-waves and THz Technologies 2015 Cardiff

University IEEE proceedings Doi 101109UCMMT20157460599

4 K N Zainul Ariffin M R R Abdullah Y K Wang S G Muttlak O S

Abdulwahid J Sexton MJ Kelly and M Missous ldquoAsymmetric Spacer Layer Tunnel

Diode (ASPAT) Quantum Structure Design Linked to Current-Voltage Characteristics

A Physical Simulation Studyrdquo UK-China Millimetre Waves and Terahertz Technology

Workshop September 2017 Submitted 14 July 2017 Conference held on 11th -13th

September 2017 DOI 101109UCMMT20178068358

5 K N Zainul Ariffin Y Wang M R R Abdullah S G Muttlak Omar S

Abdulwahid J Sexton Ka Wa Ian Michael J Kelly and M Missous ldquoInvestigations of

Asymmetric Spacer Tunnel Layer (ASPAT) Diode for High-Frequency Applicationsrdquo

DOI 101109TED20172777803

6 Omar S Abdulwahid S G Muttlak M R R Abdullah K N Zainul Ariffin J

Sexton Ka Wa Ian Michael J Kelly and M Missous ldquoA 100GHz Zero-Biased

Quantum Tunnelling ASPAT Detectorrdquo Submitted to IEEE TED on DEC 2016 under

correctionamendment Pending fabrication data

14

CONFERENCE PRESENTATIONS

1 Mohd Rashid Redza Abdullah J Sexton Kawa Ian MJKelly and M Missousldquo

G2040GHz Frequency Doubler Varistor Mode using ASPAT diodesrdquo UK

Semiconductors 2017 2017 University of Sheffield Oral presentation

2 M R R Abdullah YueKun Wang J Sexton Kawa Ian and M Missousldquo

Microwave Performance of GaAsAlAs Asymmetric Spacer Layer Tunnel (ASPAT)

Diodesrdquo UK Semiconductors 2016 2016 University of Sheffield Oral presentation

3 M R R Abdullah J Sexton and M Missousldquo GaAsAlAs Tunnelling Structures

THz RTD oscillators and ASPAT detectorsrdquo UK Semiconductors 2015 2015

University of Sheffield Oral presentation

4 Yuekun Wang Mohd Rashid Redza Abdullah and M MissousldquoInGaAs-AlAs

asymmetric space layer tunnel (ASPAT) diodes for THz electronicsrdquo UK

Semiconductors 2015 2015 University of Sheffield Oral presentation

5 Mohd Rashid Redza Abdullah and M Missousldquo GaAsAlAs Tunnelling

Structure Temperature Dependence of ASPAT Detectorsrdquo PGR Conference2016

2016 University of Manchester Poster presentation

6 YueKun Wang KNZainul Ariffin Mohd Rashid Redza Abdullah J Sexton

Kawa Ian and M Missous ldquoPhysical Modelling and Experimental Studies of

InGaAsAlAs Asymmetric spacer Layer Tunnel Diodesrdquo UK Semiconductors 2016

2016 University of Sheffield Oral presentation

7 K N Zainul Ariffin S G Muttlak M R R Abdullah Y Wang Omar S

Abdulwahid M Missous ldquoExperimental and Physical Modelling of Temperature

Dependence of a Double Quantum Well In018Ga082AsAlAs ASPAT Dioderdquo UK

Semiconductor Conference 2016 Sheffield 6 ndash 7 July 2016 Oral Presentation

8 Omar S Abdulwahid Mohd Rashid Redza Abdullah S G Muttlak K N Zainul

Ariffin Mohamed Missous ldquoTunnelling Barrier Diode for Millimetre Wave

Mixingrdquo UK Semiconductor Conference 2016 Sheffield 6 ndash 7 July 2016 Oral

Presentation

9 M Abdullah K N Zainul Ariffin MRR Abdullah J Sexton M Missous and

MJ Kelly ldquoA Novel In18Ga82As-AlAs Asymmetric Spacer Layer Tunnel (ASPAT)

Diode with Double Quantum Wells for Microwave Detectionrdquo UK Semiconductor

Conference 2015 Sheffield 1 ndash 2 July 2015 Oral Presentation

15

ABSTRACT

Thesis Title GaAsAlAs ASPAT Diodes for Millimetre and Sub-Millimetre Wave

Applications

Institute School of Electrical and Electronic Engineering the University of Manchester

Candidate Mohd Rashid Redza bin Abdullah

Degree Doctor of Philosophy (PhD)

Date 3 October 2017

The Asymmetric Spacer layer Tunnel (ASPAT) diode is a new diode invented in

the early 90s as an alternative to the Schottky barrier diode (SBD) technology for

microwave detector applications due to its highly stable temperature characteristics The

ASPAT features a strong non-linear I-V characteristic as a result of tunnelling through a

thin barrier which enables RF detection at zero bias from microwaves up to

submillimetre wave frequencies In this work two heavily doped GaAs contact layer on

top and bottom layers adjacent to lightly doped GaAs intermediate layers enclose

undoped GaAs spacers with different lengths sandwiching an undoped AlAs layer that

acts as a tunnel barrier The ultimate ambition of this work was to develop a MMIC

detector as well as a frequency source based on optimized ASPAT diodes for millimetre

wave (100GHz) applications The effect of material parameter and dimensions on the

ASPAT source performances was described using an empirical model for the first time

Since this is a new device keys challenges in this work were to improve DC and

RF characteristic as well as to develop a repeatable reproducible and ultimately

manufacturable fabrication process flow This was investigated using two approaches

namely air-bridge and dielectric-bridge fabrication process flows Through this work it

was found that the GaAsAlAs heterostructures ASPAT diode are more amenable to the

dielectric-bridge technique as large-scale fabrication of mesa area up to 4times4microm2 with

device yields exceeding 80 routinely produced The fabrication of the ASPAT using i-

line optical lithography which has the capability to reduce emitter area to 4times4microm2 to

lower down the device capacitance for millimetre wave application has been made

feasible in this work The former challenge was extensively studied through materials

and structural characterisations by a SILVACO physical modelling and confirmed by

comparison with experimental data The I-V characteristic of the fabricated ASPAT

demonstrated outstanding scalability demonstrating robust processing A fair

comparison has been made between ASPAT and SBD fabricated in-house indicating

ASPAT is extremely stable to the temperature The RF characterisations were carried out

with the aid of Keysight ADS software

The DC characteristic from fabricated GaAsAlAs ASPAT diodes were absorbed

into an ADS simulation tool and utilized to demonstrate the performance of MMIC

100GHz detector as well as 20GHz40GHz signal generators Zero bias ASPAT with

mesa area of 4times4microm2 with video resistance of 90KΩ junction capacitance of 23fF and

curvature coefficient of 23V-1

has demonstrated detector voltage sensitivity above

2000VW while the signal source conversion loss and conversion efficiency are 28dB

and 03 respectively An estimate noise equivalent power (NEP) for this particular

device is 188pWHz12

16

DECLARATION AND COPYRIGHT STATEMENT

No portion of the work referred to in the dissertation has been submitted in support of an

application for another degree or qualification of this or any other university or other

institute of learning

COPYRIGHT STATEMENT

i The author of this thesis (including any appendices andor schedules to this thesis) owns

certain copyright or related rights in it (the ldquoCopyrightrdquo) and he has given The University of

Manchester certain rights to use such Copyright including for administrative purposes

ii Copies of this thesis either in full or in extracts and whether in hard or electronic copy

may be made only in accordance with the Copyright Designs and Patents Act 1988 (as

amended) and regulations issued under it or where appropriate in accordance with licensing

agreements which the University has from time to time This page must form part of any

such copies made

iii The ownership of certain Copyright patents designs trademarks and other intellectual

property (the ldquoIntellectual Propertyrdquo) and any reproductions of copyright works in the thesis

for example graphs and tables (ldquoReproductionsrdquo) which may be described in this thesis

may not be owned by the author and may be owned by third parties Such Intellectual

Property and Reproductions cannot and must not be made available for use without the prior

written permission of the owner(s) of the relevant Intellectual Property andor

Reproductions

iv Further information on the conditions under which disclosure publication and

commercialisation of this thesis the Copyright and any Intellectual Property andor

Reproductions described in it may take place is available in the University IP Policy

(httpdocumentsmanchesteracukDocuInfoaspxDocID=487) in any relevant Thesis

restriction declarations deposited in the University Library The University Libraryrsquos

regulations (httpwwwmanchesteracuklibraryaboutusregulations) and in The

Universityrsquos policy on Presentation of Theses

17

ACKNOWLEDGEMENTS

First and foremost all gratefulness and praise is to Allah swt for everything in my

life He is the one and the only one who granted me knowledge health patience and

ability to complete this thesis as well as colouring the whole journey of my PhD

I give my deepest and sincere gratitude to my PhD supervisor Professor Mohamed

Missous for his time support patience and guidance throughout the journey of this PhD

studies His encouragements valuable advice precious ideas and a wealth of knowledge

amp experiences have had a direct inspiration on this research Special thanks also to our

experimental officer Dr James Sexton for not only sharing his knowledge advice and

semiconductor fabrication skills but also his effort in maintaining our clean room

facilities to a great level My gratitude also extends to Mr Mallachi McGowan for his

help and assist in the lab-related issue

I am also obligated to Prof MJ Kelly from University of Cambridge and Dr Kawa

Ian from ICS limited for their measurement of the ASPAT samples on realizing the RF

characteristics This collaboration effort can hopefully last longer in designing and

implementing the ASPAT MMIC detectors

My deepest appreciation also goes to my PhD colleagues Khairul Nabilah Saad

GMuttlak Omar AbdulWahid and Yuekun Wang for their support as well as working

together with me to realize this exciting project directly and indirectly A sincere

thankfulness similarly to my seniors Dr Md Adzhar Zawawi and Dr Fauzi Packeer for

their support during the first and second year of my research For other friends and staff

members under Prof Missous and Dr M Migliorato I will always remember the strong

bond and friendship we made

I am really fortunate that I been blessed with my motherrsquos care who always make doarsquo

for my success every day during my studies As for my beloved wife Dr Nik Maryam

Anisah Nik Mursquotasim who had always encouraged me supported me and gave me

patience through all the hardship in this journey thank you very much

Finally I also would like to thank and acknowledge my sponsor Majlis Amanah

Rakyat (MARA) for financially supporting me during this studies I am greatly indebted

with your kind support which was vital to my study

18

DEDICATION

This thesis dedicated to

My respected and beloved parentshellip

My loving wife dearest siblings and in-lawshellip

19

1 INTRODUCTION

11 Background

It is an undeniable fact that semiconductors have changed the world much further

than anything people could have predicted in the last 60 or 70 years ie after the lsquocats

whisker and vacuum tube eras This field of research has been expanding from year to

year starting from the discovery of the first semiconductor (silver sulfide) in 1833 by

Michael Faraday [1 2] and it still remains very active to the present Semiconductors

have a large range of applications and are not just limited to use in communications they

can be found everywhere in other applications from Earth to space The widespread

usage and sheer number of applications have led to it growing very quickly and

contributing greatly to the growth of World Economics Over time the successful

development of semiconductor growth techniques such as Molecular Beam Epitaxy

(MBE) has enabled researchers to tailor and precisely control the semiconductor

material for new electronic devices with extra functionalities This has led to the

development of advanced devices such as high electron mobility transistors (HEMTs)

and Heterojunction bipolar transistors (HBTs) for use in wireless communication

technology Given this development today electronic devices such as computers

handheld smartphone tablets etc are no longer perceived as luxury and attractive items

but rather have become crucial in everyday life Such devices provide the means to allow

for people to remain connected to each other via the sending and receiving of

information electronically The huge demand for such types of devices has resulted in

competition in both the electronic market and technologies which only goes on to

advance the semiconductor industry

Nowadays the demand for electronic devices characterised by high speed high

efficiency ultra-low power and low manufacturing cost has increased exponentially To

fulfil this growth in demand high data rate systems are required in other words the

system must work at a higher frequency for both the transmitter and receiver The

frequency of interest for advanced wireless communication is in the Millimetre and sub-

20

millimetre wave region which is around 30-300GHz and 300 - 3000GHz respectively

The second frequency region is also sometime known as the terahertz (THz)

electromagnetic region This band lies between the microwave and infrared frequency

bands From the first time it was revealed in the late 80s[3-5] the THz region has gained

a lot of international attention due to its unique properties and since then the motivation

to develop these devices has increased significantly To date the THz frequencies region

has shown its ability to fulfil various applications such as high-resolution imaging in

medical security and surveillance field atmospheric monitoring and environment radio

astronomy as well as compact range radars[3] to name a few

However despite these developments not much effort has been made in exploring

alternative compact THz devices As a result electronic THz devices are still in the state

of immaturity as compared to microwave and photonics devices This is due to their high

cost and absence of compact amp solid-state THz sources (oscillators) and receivers

(detectors) that are capable of operating at both room and extreme temperatures[6 7] A

great deal of work still continues to fill up the lsquoTHz gaprsquo (between 300GHz and 3THz)

used for the most important part of a communication system namely the front-end

receiver or first stage Such a system is responsible for receiving detecting and

processing the received signal to be translated into useful information Furthermore THz

receivers systems still require the best-integrated components such as source mixer and

detector to reach their complete competencies[8] The detector which remains the

critical part of the receiver system requires devices or components that are able to fulfil

the THz gap requirement Studies conducted over a number of years have found out that

the key element in improving THz detection relies upon the use of passive devices ie

diodes Based on these findings many types of diode ie tunnelling diode point-contact

diode and Schottky barrier diode (SBD) have been proposed for detection applications

Amongst microwave and millimetre wave detector diode devices the Schottky

Barrier Diode (SBD) is the dominant detector that has been used since the 1940s[9] The

reason for this dominance is the ease of fabrication of a SBD (by either a point-contact

or evaporated semiconductor-metal structure) and its ability to produce a non-linear

current-voltage (I-V) characteristic which is necessary for rectifyingdetecting diodes [9

10] SBDs also have high cut-off frequency good dynamic range and are low cost To

21

date the SBD has been able to detect signals up to 100GHz [11] 1THz[12] and as high

as 10 THz[13] However the current transport mechanism in a SBD relies on thermionic

emission and therefore is strongly dependent on temperature and means that using them

in extreme conditions ie military and automotive applications is complex The SBD

also suffers from high noise figure[14] and is susceptible to burnout at a modest pulse

power level this will limit the use of ultra-high frequencies and low power signal

applications Other diodes that share the same characteristics are Planar Doped Barrier

(PDB) Germanium Backward Diode (GBD) ie a type of Esaki tunnel diode These

diodes are well known and are reliably used as millimetre wave detectors However it

still proves inefficient to substitute the SDB with any of the previously mentioned

diodes This is due to some drawbacks such as strong temperature dependence limited

dynamic range fabrication complications and hence poor reproducibility (ie GBD) and

other circuit complexities

Hence there is strong compulsion to study examine and produce new detector

diode structures that are able to solve the mentioned diodes limitations and which have

high sensitivity larger dynamic range low noise strong independence to temperature as

well as being able to work efficiently in the high-frequency band and at zero bias The

advantages of working at zero-bias relates very much to the need for a system with less

power consumption so that the device (ie mobile communication) is able to run off

small batteries for a reasonable length of time eliminating extra biasing circuit as well as

noise Therefore a new tunnelling device namely the Asymmetrical Spacer Layer

Tunnel diode (ASPAT) developed by RT Syme [15 16] and refined by Missous et

al[17] has been examined in this work The ASPAT which is in essence a

Semiconductor-insulator-semiconductor structure relies on tunnelling through a barrier

to provide current compared to conventional thermionic emission in SBDs The ASPAT

diode has many advantages a zero bias turn-on voltage very weak sensitivity to changes

in temperature (due to tunnelling) very low noise large dynamic range high resistance

to pulse burn-out [18] and as demonstrated recently can be reproducibly

manufactured[17] The growing interest in THz frequencies nowadays makes the

ASPAT an excellent choice to fulfil all requirements for ultra-high speed applications

22

ie communication (mobile computer networking) radar (military equipment) scalar

analyser and built-in test equipment

In this work an ASPAT diode based on group III-V elements of the periodic table

comprising compound semiconductors of large band gap material Aluminium Arsenide

(AlAs) sandwiched between two lower bandgap Gallium Arsenide (GaAs) are used and

intensively examined The AlAs semiconductor which is ten-monolayer thick has

almost the same lattice constant as GaAs but has a larger bandgap Consequently in the

conduction band a thin barrier of the AlAs is formed from the arrangement of such

structure The structure is made up of GaAs and AlAs both materials are grown on

GaAs substrate using Solid Source Molecular Beam Epitaxy (SSMBE) Therefore in

this study the ASPAT diode will be referred to as ldquoGaAsAlAs ASPATrdquo diode The

conventional GaAsAlAs ASPAT diode has been developed and successfully fabricated

in two different stages This work was the first carried out using facilities provided by

the University of Manchester The first stage of the work was to qualify the

reproducibility and repeatability of growth and fabrication technique which is mostly

performed on larger emitteranode size The second was to develop conventional

ASPAT diodes that can perform at Millimetre and sub-millimetre wave frequencies and

which are comprised of small emitter area

Prior to this work full physical modelling using SILVACO design software was

undertaken to generate models and to fully characterise and identify the fundamental

physical phenomenon of multi-junction ASPAT diode Therefore insight into and

performance based on diode structure and electron movement can be understood and

predicted which lead to the crucial idea in helping and advising iterations to epitaxial

growth as well as diode fabrication The verification of the physical models must be set

as a priority goal by comparing the results of statistically fabricated measured data The

advantage of physical modelling is that it can help reduce materials resources cost and

fabrication time

Further research into the field has led to the development of two other types of

ASPAT diodes that are used to compare with the conventional GaAsAlAs ASPAT

diode Their configuration involved the use of a more advanced semiconductor

technology which comprises InGaAsAlAs materials and GaAsAlAs with InGaAs

23

quantum wells The latter was a novel ASPAT diode and the former is identified as

advanced ASPAT diode However these two advanced ASPAT diodes have not been

extensively studied in this thesis as they will be covered by other co-workers at

Manchester Hence due to these some important parameters are compared to the

conventional one as it is the main focus of this work In the case of temperature

dependent studies the DC characteristic of conventional ASPAT is compared to in-

house fabrication AuGaAs SBD All the ASPATs epitaxial layer materials structures are

shown in the following tables

Table 11 Sample XMBE304 XMBE307 and XMBE368 GaAsAlAs materials structure grown

on GaAs Substrates by MBE

Thickness (Aring)

Layer Material Doping (cm-3

) XMBE368 XMBE304 XMBE307

Top Ohmic1 GaAs (Si) 4times1018

~3000 ~3000 ~3000

Buffer 1 GaAs(Si) 4times1017

~50 ~400 ~50

Spacer1 GaAs NID 50 50 50

Barrier AlAs NID 28 28 28

Spacer 2 GaAs NID 1000 2000 1000

Buffer GaAs(Si) 4times1017

50 400 50

Bottom Ohmic GaAs(Si) 5times1018

~3500 ~4500 ~3500

Substrate GaAs (Si) 50000 50000 50000

Note that sample XMBE368 and XMBE304 are grown on doped GaAs

substrates Sample XMBE368 was grown un-rotated to study the effect of barrier

thickness variation

24

Table 12 Quantum wells sandwiching the quantum barrier for sample XMBE314 grown on a

GaAs substrate by MBE

XMBE314

Layer Material Doping (119836119846minus120785) Thickness (Aring) Bandgap (eV)

Ohmic Layer GaAs (Si) 4times1018

3000 142

Emitter GaAs (Si) 1times1017

400 142

Spacer GaAs Undoped 50 142

Quantum Well In18Ga82As Undoped 60 116

Barrier AlAs Undoped 28 283

Quantum Well In18Ga82As Undoped 60 116

Spacer GaAs Undoped 2000 142

Collector GaAs (Si) 1times1017

400 142

Ohmic Layer GaAs (Si) 4times1018

4500 142

Substrate GaAs

Table 13 Novel InGaAsAlAs ASPAT diode grown on InP substrate by MBE on InP substrate

XMBE326

Layer Material Doping (cm-3

) Thickness (Aring) Bandgap (eV)

Top Ohmic1 In053Ga047As(Si) 5times1019

3000 075

Buffer 1 In053Ga047As(Si) 1times1017

350 075

Spacer1 In053Ga047As NID 50 073

Barrier AlAs NID 283 283

Spacer 2 In053Ga047As NID 2000 075

Buffer In053Ga047As(Si) 1times1017

350 075

Bottom Ohmic In053Ga047As(Si) 15times1019

4200 075

Substrate InP (Si) NID 620000

From the above tables it can be noted that XMBE307 is the first batch that was grown

in-house using a Riber V100HU MBE machine followed by XMBE368 XMBE 304

XMBE314 and finally XMBE326 The two earlier batches were grown on n+

substrate hence their fabrication process flow is simpler On the other hand the three

other batches were grown on semi insulating substrate thus requiring the development

of new repeatable reproducible and robust process flow which will be covered in this

thesis

25

In general the fabrication of the ASPAT diode is based on top-down processes this

is because the ASPAT is a vertical structure device and the junction capacitance of the

ASPAT is directly associated with the size of the anodeemitter area Further to these

the capacitance directly influences the diode cut-off frequency Therefore the simplest

way to reduce the capacitance is by reducing the lateral area of the device of the ASPAT

structure since it can be represented by a parallel plate capacitor where the capacitance

is inversely proportional to the area of the device In order to achieve high cut-off

frequencies minimising capacitance via small dimensions ie sub-micrometre level is

essential However this will also increase the series resistance of the diode As a result

the cut-off frequency will be degraded Thus there is a trade-off between small

dimension of device and high cut-off frequency to be achieved

Finally successful growth and fabrication for small area GaAsAlAs ASPAT diode

in this work has led to carefully extracted RF characteristics This becomes a stepping

stone to designing a millimetre wave integrated circuit (MMIC) detector using empirical

modelling in Keysight ADS tools Therefore a predicted performance for a 4times4 um2

fabricated ASPAT is that can operate at 100GHz ASPAT as a zero-bias detector with a

voltage sensitivity of over 2000VW Additionally the design of a millimetre wave

source using similar ASPAT diodes was also carried out The performance of a 2040

GHz doubler using GaAsAlAs ASPAT in varistor mode is demonstrated for the first

time with a conversion loss of 33dB and conversion efficiency of ~ 02

26

12 Aims and objectives

The aim of this study is to further improve the performance of microwave and

millimetre wave technology by incorporating the Asymmetrical Spacer Layer Tunnel

Diode (ASPAT) for ultimate operation near THz frequencies by designing a range of

low power high-speed devices enhancing the methods of Simulation layout and

materials amp structural characterisations with fabrication process optimization using the

facilities available at the University of Manchester

There are three main objectives in this research firstly to streamline the physical

device design and modelling using the GaAsAlAsGaAs materials systems in order to

produce a zero bias detector which is basically a rectifier of a microwave signal by

using the SILVACO Atlas simulation tools

Secondly to achieve reproducibility and manufacturability of the fabrication

process for new type of GaAsAlAs ASPAT structure (lateral structure) hence small size

ASPAT emitter by improving the device processing technique and maximising the

capability limit of the conventional i-line optical lithography that is available in Prof

Missousrsquos group laboratory

Thirdly to optimise DC parameters through electrical properties investigation as a

stepping stone to the next objective that is to characterise the RF performance of the

GaAsAlAs ASPAT detector circuit The detection properties of microwave and

millimetre wave diode will also be investigated with different ASPAT diode size at

100GHz Further to these the properties of microwave signal source will also be

developed by way of utilizing the non-linearity feature of the diode Therefore this new

type of tunnelling diode can be applied to both applications of signal detection and

signal source in the microwave and millimetre wave ranges

27

13 Outline of this Thesis

This thesis is organized into eight chapters The first chapter discusses the

contextual information that motivates the undertaking of the study An overview of the

work which includes the details of the studied samples the aim and objectives of the

whole research project are also outlined in this chapter

Chapter 2 deals with the literature review of the basic principles and concepts of

the group III-V compound semiconductors The historical background of such

semiconductors which is essential to the development of ternary structures etc and the

advancement of semiconductor materials engineering is presented The types of existing

tunneling diode as well as conventional microwave diodes are also discussed and

compared The fundamentals of ASPAT diode which includes structural parameters and

its operation are then explained in detail Finally discussions of the ASPAT key DC

characteristics which are important for detection purposes are presented

Chapter 3 focusses on the development of the experimental techniques which can

be divided into two stages In the first stage the development is towards repeatability

reproducibility and manufacturability of the ASPAT grown in-house by MBE and

fabricated by conventional i-line optical lithography The second stage involves

optimisation and fine tuning such fabrication method for GaAsAlAs ASPAT samples

that can operate at high frequency ie 100GHz detector For both stages of the

fabrication process all techniques including mask design generic and special process

flow are presented The chapter ends with discussions on issues related to sample

processing and improvements that are proposed to solve these issues

Chapter 4 dwells on the modelling of the GaAsAlAs ASPAT using the SILVACO

simulation package The discussions are expected to offer a better understanding or

insight into each layer that forms the ASPAT diode structure The chapter begins with

discussions of the operation of the SILVACO Atlas tool A validation of physical

modelling is essential and presented according to the fabricated mesa sizes of the diode

28

Thereafter towards the end of the chapter the analyses of the relationship between

device current-voltage (I-V) characteristics the structural parameter including various

temperatures dependent simulations with a comparison to an in-house fabricated SBD

are offered

Chapter 5 presents relevant DC results based on optimized fabrication process and

RF characterization which enable obtaining an intrinsic and extrinsic element of the

GaAsAlAs ASPAT diode The discussions also highlight the analysis of DC zero bias

equivalent circuit and de-embedding extraction using ADS The chapter ends with

discussions on the RF reproducibility performance which includes the performance as

well for millimeter-wave and sub-millimeter wave applications

Chapter 6 discusses the main applications of ASPAT diodes The chapter begins

with discussions on detection theory followed by the parameters of interest and ends

with circuit design as well as the performance of a 100GHz detector The circuit design

was conducted using Keysight ADS software via harmonics balance simulation tool The

performance in term of sensitivity depending on measured ASPAT emitter size is

demonstrated Finally a comparison with conventional Schottky diode is presented

towards the end of the chapter

Chapter 7 discusses a secondary application that can be applied to the ASPAT by

utilizing the nonlinearity feature of the diode to create a signal source namely a 20 to

40GHz frequency doubler in varistor mode The doubler performance of ASPAT will be

explored through circuit design constructed via Keysight ADS simulation software Each

key parameter is highlighted and discussed in detail

The final chapter of this thesis that is Chapter 8 discusses the conclusions of the

study with emphasis on the overall key research findings The chapter also highlights

suggestions for further research in this particular field of study

29

2 LITERATURE REVIEW

21 Introduction

Since 1940s the development in the technology of semiconductor electronics has

been expanding and now has led to the establishment one of the most astonishing

industries of the 3rd

-millennium era Leading this advancement is the integrated circuit

(IC) or chip which was driven mostly by silicon (Si) Overtime the IC has undergone

substantial revolution in term of power economics size and efficient energy

consumption Currently it covers every aspect of human life ie from desktop personal

computers in the office and house to the compact smartphone in the pocket and from a

gigantic satellite in space to small satellite navigation in cars In other words

semiconductor technology is crucial to human life Without developments in

semiconductor materials engineering and shrinking of device size such accomplishment

may not have been realised today Therefore this chapter presents a macro view of the

development in compound semiconductor technology especially in radio frequency (RF)

towards Millimetre and submillimetre wave applications with regard to the improvement

of material and device structures

The essentials of group III-V compound semiconductor will be emphasised for its

points of interest and application in this field (RF technology) This chapter comprises

five main sections The first section is an overview of the semiconductor history with

concentration on its advantages and applications in the RF field while the second and

third discuss the effects of III-V compounds when the interface occurs between

semiconductor-semiconductor and semiconductor-metal respectively which leads to a

basic understanding of hetero-structures device as well as contacts namely Schottky and

Ohmic The fourth section is predominantly concerned with high-speed devices ie

diodes and materials in this field which leads to the exploitation of the main researchrsquos

device Then the following section describes in detail the background works basic

principle and intrinsic amp extrinsic parameters of the Asymmetric spacer Tunnel Diode

(ASPAT) Finally the basic way of characterising the device will also include giving an

overview of how the device is measured and what parameters are needed

30

22 Historical review of III-V Compound Semiconductor for RF applications

The beginning of commercial electronic devices was marked with the first point-

contact semiconductor transistor developed in 1947 by William Shockley at Bell

Laboratories in New Jersey Shockley developed a device based on a Germanium

Bipolar Junction Transistor (Ge BJT) structure [19] with operating frequency above 1

GHz Since then and until early 1950s the development of Ge BJT was fast and it

became foremost in the market of semiconductor technology However the emergence

of Silicon (Si) challenged Ge in the market in the 1960s Si has the upper hand primarily

because it has better electron transport and low manufacturing costs compared to Ge

[19] By the 1970s almost all RF transistors were based on Si BJT Additionally the

development of Si which forms a new material from the formation of native oxide

namely Silicon Dioxide (SiO2) led to the invention of the Metal Oxide Semiconductor

Field Effect Transistor (MOSFET) [19] The future of digital electronic industries has

been ldquobrightrdquo ever since the MOSFET was ldquobornrdquo as it has become a fundamental

building block component in complex microprocessors and flash memories Despite this

development the exploitation of Si at RF frequencies did not last long since Si is not an

optimum semiconductor for RF electronic devices The emergence of GaAs has

improved RF applications for high-speed transistors

Ge and Si which are categorised as single element semiconductor are the earliest

materials used to build the first transistor devices These devices played a crucial role

towards the development of more advanced material such as GaAs of the compound

semiconductor type[20] A compound semiconductor is a semiconductor formed by the

ionic bond of different types of semiconductor material most widely known as the group

III-V compound semiconductors The main reason for the progression of the III-V

compound semiconductors is due to their better electron mobility compared to the single

element semiconductors The term ldquomobilityrdquo in the semiconductor industry refers to the

easiness of movement of charges in many directions inside a crystal In fact it is

determined by the access resistances values with saturated velocity under certain values

of electric fields (bias) the higher the electric field the faster is the carrier movement in

the crystal Figure 21 shows the electron mobility and band gap for the most common

31

group III-V compound semiconductors Besides higher mobility III-V compound

semiconductors also have light-emission capability and are suitable for bandgap-

engineering techniques

The work on III-V compound semiconductors mainly on GaAs FETs led to a new

change for the whole RF electronics industry For example in 1966 the first GaAs

MESFET was invented[21] and achieved a maximum operating frequency of operation

of 3GHz [22] Three years later the frequency increased to 30GHz [23]

Figure 21 III-V compound semiconductors mobility and band gap[24]

With better features in terms of having a higher electron mobility compared to Si

electronic devices based on III-V materials developed rapidly This attracted attention in

many aspects especially in military radar application electronic warfare system missile

guidance control electronic for smart warfare system and secure communication To be

specific those demands were fulfilled through the application of microwave mixer and

detectors [25] which were achieved based on Schottky barrier diodes and FETs

However these applications remained largely as niche markets for use only in military

and exotic scientific projects until 1980 In addition to the microwave industry two

important diodes that played a large role in very high-frequency power source namely

the Gunn diode and the Impact Avalanche and Transit Time (IMPATT) diode which

were discovered in the 1960s[26]

100

1000

10000

100000

0 05 1 15

Bu

lk M

ob

ility

(cm

2 V

-1 s

-1

BandGap (eV)

InS

b

InA

s

Ge

Ga

Sb

In

GaA

s

Si

GaA

s

InP

32

Furthermore the invention of Molecular Beam Epitaxy (MBE) growth technique at

the beginning of 1970s has enhanced the full potential of the III-V compound

semiconductors[27] This technique has led to the formation of a new class of materials

and heterojunction device with high-quality interfaces and accurate control of the

thickness during growth[28] The advancement of material engineering that tailored the

III-V compound semiconductor with MBE effect has been beneficial for both three-

terminal and two-terminal devices As a result of this more advanced devices in both

electronics and optics were developed such as quantum well (QW) laser Resonant

tunnelling diode (RTD) high electron mobility transistor (HEMT) and many more[29]

The aim was to achieve high-speed devices transporting data at high data rates and

robust devices These devices promised an excellent option to conventional transistor

(three terminal devices) in high-frequency systems especially in the terahertz (THz) or

Millimetre and sub-millimetre wave regions [30]

One of the promising diodes that received a lot of attention is the resonant

tunnelling diode (RTD) which was first described in 1974 by Chang [31] This device

which consists of a double barrier and one quantum well is the classical tunnelling diode

Due to its good symmetrical non-linearity in its current- voltage characteristic it can be

exploited for signal generation and detection However the main focus of RTD to date

has been in the generation of continuous wave (CW) ultra-high frequency and to a lesser

extent in detection Therefore other tunnelling based diodes were developed specifically

for detection purposes which are the main foci of this work The PDB and ASPAT

diodes are the workhorse candidates for detection purposes Most of these are built based

on group III-V compound semiconductors [32]

Unlike the Very Large Scale Integration (VLSI) market ie CMOS for personal

computer (PCs) the RF electronic device for civilian application reached the consumer

market only in the late 1980s through satellite television with operating frequencies

around 12GHz [19] Since then many RF application have been deployed on the mass

market depending on their operating frequency such as 09GHz ndash 25GHz for wireless

communication 20GHz to 30GHz for satellite communication 77GHz for car radar

systems and above 90GHz for different sensor applications Utilising GaAs as the main

material RF devices have become the key underpinning components for modern

33

communication systems As a result in 1998 the volume production of mobile phones

was greater than that of PCs for the first time in history Presently production is being

made for devices like smartphones cellular phones mobile internet access and new

communication services and tablets

The development of the RF field is never ending More and more improvements are

being made especially through the design and fabrication of oscillators and detectors

which are mainly built based on group III-V compound semiconductors When RF

devices were used by the military (in the 1970s to 1980s) cost was not a concern

However after getting into civil application market (ie in the 1990s) the most frequent

issues highlighted were performance and cost[19]

The ability to generate or receive high operating frequencies with high power large

bandwidth and high sensitivity is an indicator for a good performance of RF devices

(depending on specific applications) For example the highest room temperature based

oscillator of up to 186 THz was achieved in thin well AlAs-InGaAs RTD by Professor

Masahiro Asada from Tokyo Institute of Technology [33] An excellent review on THz

sources can be found in [34] For ultra-high frequency detector and mixer applications

the two terminals RF device that is mostly used is the SBD In 1996 the highest cut off

frequency achieved by a mixer utilizing the SBD was about 5THz[35] and this has kept

increasing ever since The factor that motivates the development of THz devices is the

requirement to have a compact coherent source in the THz range Undoubtedly in the

future there will be very exciting times for enthusiasts of terahertz sources and receiver

as new generations of compact broadband and tuneable solid source device based on

advanced compound semiconductor are developed

23 The Concept of Heterostructures

A III-V compound semiconductor is mostly grown on a single semiconductor

substrate forming a layer called epitaxial heterojunction layer It is a starting point and

the key feature that brings the idea of realising the most advanced semiconductor

devices currently being developed and manufactured by combining several epitaxial

semiconductors [36-38] Heterojunctions have the capability of manipulating carrier

transport ie electron and holes transport in crystal separately unlike homojunctions

34

This has resulted in the successful development of new devices for high-speed and high-

frequency applications as well as optical sources and detectors [37] This section will

discuss lattice matched material pseudomorphic material hetero-junction band

discontinuities and quantum wells

231 Homojunctions Heterojunctions and Band Discontinuities

The term homo-junction refers to the interface between identical semiconductor

materials that have different polarity ie p-type or n-type but similar in energy gap This

phenomenon is usually applied in forming p-n junction diodes and can be understood by

referring to Figure 22 below

Figure 22 illustration of Homojunctions band structure material before (left) and after (right)

equilibrium

The materials A and B which have similar bandgap (Eg) and different dopant

types ie p-type and n-type will have their Fermi levels (Ef) closer to the valence band

(EV) and conduction band (EC) respectively before ldquothermal equilibriumrdquo Once

equilibrium is achieved Ef of both p-type and n-type will be aligned causing band

bending of EC and EV As a result a built-in electric field is introduced (via diffusion of

carriers) for both holes and electrons and forcing them to move in one direction

On the contrary a heterojunction occurs when the interface between two

semiconductor materials with different bandgap energy are brought together (ie large

energy band gap material combined with a low band gap one eg wide band gap AlAs

and narrow bandgap GaAs) This results in a steep band bending which leads to the

formation of energy band discontinuities at the junction as shown in Figure 23 In a

semiconductor heterojunction the most important parameter is the band gap energy

EC

EV

Ef

Eg

p-type E

g

n-type

EC

EV

Ef

Material A

p-type

n-type

Material B Material A Material B

35

associated with each material in the structure where the degree of discontinuity can be

utilised in varying the carriers transport properties as well as the quality of the junction

depending on the interest of the designer This leads to flexibility in tailoring device

characteristics leading to vastly improved performance of the device

Figure 23 Energy band diagrams before (left) and after (after) interface combination at

equilibrium

Based on Figure 23 above Material A indicated with blue line is a large band

gap energy material and Material B highlighted with the red line is a low band gap

material EV represents the valence band EC the conduction band and Ef is the Fermi

level of the materials Alternatively the two materials band discontinuities are denoted

by ΔEC for the conduction band and ΔEV for the valence band χ and Eg represent the

electron affinity and band gap energy respectively

At some point where by the Fermi energy of both semiconductor materials are levelled

the structure would have reached its thermal equilibrium The band gap of materials A

and B have a discontinuity at the interface (ΔEg) of these two materials In general this

is given by

120549119864119892 = 119864119892119860 minus 119864119892

119861 (21)

Furthermore when thermal equilibrium is achieved ΔEg is then divided between

conduction band and valence band discontinuities (ΔEC and ΔEV respectively) at the

material A and B junction interfaces Their relationships can be expressed as

EC

A

EV

A

Ef

A

Eg

A

Material A Material B

χA

χ

B

ΔEC

ΔEV

Ef

B

E

g

B

EC

B

EV

B

EV

A

χA

ΔE

C

ΔEV

EC

A

Ef

EC

B

EV

B

χB

Material A Material B

Vacuum

Level Vacuum

Level

36

∆119864119862 = 120594119860 minus 120594119861 (22)

∆119864119881 = (119864119892119861 minus 119864119892

119860) minus (120594119860 minus 120594119861) (23)

∆119864119892 = 119864119892119860 minus 119864119892

119861 = ∆119864119862 + ∆119864119881 (24)

However these relationships which were introduced by Anderson can only offer an

approximation In practice the results are always different since dislocation and

interface strain occur at the junction Therefore precise control during epitaxial growth is

always required and growth technologies such as MBE are employed In due course the

band gap discontinuity can be further exploited by using different types of material

combination Examples are GaAsAl052Ga048As and In053Ga047As In052Al048As [39]

232 Lattice-Matched and Pseudomorphic Materials

As discussed earlier a heterojunction happens when any two different

semiconductor materials that have different bandgap are joined together At the atomic

level both materials often differ in lattice constant The easiest way to explain this is by

setting the formation of heterojunction which can be separated into two types lattice

matched and lattice mismatched (pseudomorphic)

2321 Lattice Matched Systems

To create discontinuities for use as a high-performance device the combination of

semiconductor materials is essential Selecting the appropriate materials that have

similar or very close lattice constants to combine is crucial to avoid disruption at the

atomic level heterojunction interface Figure 24 shows that the lattice constant of a

material A ie substrate (aS) and material B ie deposited over layer (aL) are identical

or very close and their surface atoms are perfectly matched This scenario is known as

lattice matching

37

Figure 24 Lattice Matching for both materials when aL=aS

As can be seen in Figure 25 while there are restricted binary materials available to

form good heterojunction interfaces it is possible to combine semiconductor materials in

binary ternary and quaternary forms to allow the formation of a variety of lattice-

matched heterojunction interfaces The examples of materials that have successfully

been alloyed are In053Ga047As In052Al048AsInP and GaAsAlxGa(1-x)As (x=0 to 1)

Even though the materials system hetero-junction of these materials has close lattice

constant value their band-gap will experience an abrupt variation

Figure 25 Energy gap and lattice constant for direct and indirect band gap of compound

semiconductor at ambient temperature[40]

Material A(layer)

Material B(substrate)

aL

aL

aS

aS

38

The existence of binary ternary and quaternary semiconductors formed by alloying

semiconductors has expanded the opportunity for heterojunction formation in devices

The alloy semiconductor which is produced by the combinations of two semiconductors

A and B has a lattice constant that obeys Vegardrsquos Law as follows

119886(119886119897119897119900119910) = 119909119886119860 + (1 minus 119909)119886(119861) (25)

For the alloy the band gap normally follows the virtual crystal approximation

119864119892(119886119897119897119900119910) = 119909119864119892(119860) + (1 minus 119909)119864119892(119861) (26)

Table 21 shows the list of the semiconductor alloy band gap and lattice constant for

common binary and ternary for group III-V compound semiconductors[41]

Table 21 Lattice constant (a) and band gap for group III-V binary and ternary compound

semiconductors a room temperature [41 42]

Alloy Lattice constant a(Aring) Band gap EgeV)

AlAs 5661 2239

AlSb 6136 1581

GaAs 5653 1424

GaN 3189 34

InAs 6058 0417

InP 5869 1344

Al052Ga048As 5657 2072

In053Ga047As 5868 0773

In052Al048As 5852 1543

39

2322 Pseudo-morphic Materials

The other scenario is when two different materials with different lattice constants

are brought into contact The observation can be made at the atomic level where the

atom will try to match each other as shown in Figure 26 below

Figure 26 Lattice mismatched material

In fact for both situations (ie lattice matched and lattice mismatched) the atom of

the material at the hetero-interface will change their position to maintain the geometry of

the lattice Due to distortion at this atomic level a strain is then induced at the hetero-

interface In order to form a good hetero-junction interface the strain must not exceed a

certain specific critical value which will cause crystal dislocations to occur The result of

crystal dislocation is generally bad as it will affect the carriers which will be

concentrated in the defect area thus degrading the carriersrsquo mobility This then makes

the overall function or performance of the device to become poor

Nowadays the Molecular Beam Epitaxial (MBE) technique is able to grow epitaxial

layers of mismatched semiconductor layers profile ie mismatched in their lattice

constant (aLneaS) The growth method works when the grown epitaxial layer assumes the

lattice parameters of the layer it is deposited on Nonetheless the layers must be kept

within a certain limit and the deposited layer must be thin enough to avoid defect or

dislocation formations This new layer known as a ldquopseudomorphicrdquo material will alter

its original crystal structure and physical properties

Material A(layer)

Material B(substrate)

aL

aL

aS

aS

40

Figure 27 The formation of Pseudomorphic-layer in two situations (a) compressive and (b) tensile

strain [1]

Figure 27 shows material A in which the pseudomorphic materials can be related to

two situations compressive and tensile strain The compressive strain occurs when the

deposited layer has a larger lattice constant than the substrate (aL gtaS) while tensile strain

happens when the deposited layer has a smaller lattice constant than the substrate

(aLltaS) These leads to aL either to compress or stretch to fit aS respectively Note that

the pseudomorphic layers can only be grown to a certain critical thickness hc From

Figure 27 the strain between the substrate and the deposited epitaxial layer is given by

휀 =119886119871 minus 119886119878

119886119878 (27)

Where Ɛ is strain between two layers aL is lattice constant of the deposited layer and

aS is lattice constant of the substrate layer The concern in deposition of the over layer is

to avoid dislocation occurring at the interface if there is too much strain at the junction

The strain is naturally influenced by the thickness of the deposited layer and thus the

thickness of growth must be controlled below the critical thickness hc which is

expresses as

(b)

41

ℎ119888 =119886119904

(28)

Moreover one needs to appreciate that even though the crystal structure and their

physical properties change the total energy within the unit cell is maintained This is

possible by distortion of the deposited layer in the direction perpendicular to the growth

direction while leading to lattice matching in the lateral plane Example of lattice

matched materials is GaAsAlAs and pseudomorphic material is In08Ga02AsInP

233 Quantum well and 2DEG

A typical application of heterojunction interface is one in which utilises ΔEC and ΔEV to

form barriers for electrons and holes One example of barriers that confines these

carriers is known as a Quantum Well (QW) A QW is a layered semiconductor usually

very thin ie about ~ 100 Aring thicknesses in which many quantum mechanical effects can

occur It is formed by a thin layer of a low bandgap energy semiconductor material eg

GaAs sandwiched between two similar large bandgap energy semiconductors eg AlAs

or AlGaAs The growth technique to achieve thin layers of QW is usually MBE The

benefit of this method is that it allows the formation of heterojunction with very thin

epitaxial layer

Figure 28 The band diagram of an ideal quantum well formed by a narrow band gap

semiconductor eg InGaAs sandwiched between two large band gap semiconductors eg GaAs (a)

Structure (b) energy band diagram and (c) Conduction band diagram when AlGaAs is n-doped[43]

42

The thickness of the layer that can be achieved can be as thin as the electron mean free

path (De Broglie wavelength) which is around 100 Aring to 300 Aring [44] The expression for

the De Broglie wavelength is given by

120582 = ℎ120588frasl (29)

Here h and ρ are Planckrsquos constant and momentum of the electron respectively Figure

28(b) illustrates a quantum well formation in abrupt semiconductor interfaces It can be

observed that the heterojunction boundary will experience discontinuities at the edges of

the conduction band and valence band with a quantum well generated for the carriers

(both electron and holes) The quantised energy sub-bands in the quantum well structure

in Figure 28(b) can be determined from [43]

119864 = 119864119899 + (

ℏ2

2119898lowast) (119896119909

2 + 1198961199102)

(210)

Where 119864119899 = (ℏ21205872

2119898lowast ) (119899

119871)2

and n is the energy level index that can be n=1 2 3hellip

The dopants in a semiconductor with large band gap layers may supply the

carrier to the quantum well and this occurs when the base or bottom of the quantum well

is lower than the Fermi Level and hence the high energy donors will go down to the

well therefore creating a Two-Dimensional Electron Gas (2DEG) In the 2DEG the

electrons and holes move freely in the quantum well in the plane perpendicular to the

growth direction however they are not capable of moving in the crystal growth

direction (confinement direction)[45 46] The 2DEG phenomena can be seen in Figure

28(c)

24 Metal-Semiconductor Contact

A semiconductor device is incomplete if there is no connection between the

semiconductors and the outside world A metal which is usually gold (Au) or gold

germanium (AuGe) is diffused into the semiconductor to allow for electrical connection

from the outside world to the semiconductor and vice versa The metal-semiconductor

contact can be either a Schottky contact or an Ohmic contact The Schottky contact is a

43

rectifying contact while the Ohmic contact is a contact that provides a low resistance

path between semiconductor and metal

241 Schottky Contact

The Schottky contact is basically a metal contact to the gate to enter a region or

channel in a transistor Figure 29 shows a schematic band diagram of a metal-

semiconductor contact before and after contact (Schottky-Mott concept)[47]

Figure 29 Metal and semiconductor in two conditions (a) separation (b) in contact

In Figure 29 the work function of the metal is represented by qm while the

semiconductor work function is qS The qχ is the energy difference of an electron

between the vacuum level and conduction band edge ie known as the electron affinity

and qVn is the difference between the conduction band and Fermi level in the

semiconductor EV EC Ef is the valence band energy conduction band energy and the

Fermi level respectively

The metal and semiconductor are brought together as showed in Figure 29(a)

both materials are at steady state However when the metal and semiconductors are in

contact as illustrated in Figure 29(b) the electrons that flow from the conduction band

in the semiconductor into the lower energy state of the metal will cause the Fermi level

to be aligned in thermal equilibrium Due to this process the positive charge donor is

trapped in the semiconductor interface hence forming a depletion region Xdep

Thereafter the upward bending of the energy in the semiconductor takes place On the

qχ(s)

Eg(s)

Vacuum Level

EV

EF(m)

E

F

EC

qϕ(m)

qϕ(s)

Metal Semiconductor

qVn

qϕB

qϕ(s)

qVbi

X

qχ(s)

Eg(s)

Vacuum Level

EV

EF

EC

qϕ(m)

Metal Semiconductor

Xdep

(a) (b)

44

other hand the negative charge (electron) will be accumulated within a narrow region in

the metal interface The existence of two different charges at the metal-semiconductor

boundary generates an electric field This leads to a potential barrier qB as seen by

electrons in the metal moving into the semiconductor and a built-in potential qVbi as

seen by electrons in the semiconductor trying to move into the metal

The built-in potential qVbi is defined as follows

119902119881119887119894 = 119902empty119861 minus 119902119881119899 (211)

The barrier height empty119861 in the ideal case is specified by the dissimilarity between a metal

work function empty119898 and electron affinity of the semiconductor

119902empty119861 = 119902empty119898 minus 119902120594 (212)

Referring to Eq (212) above the barrier height empty119861 rises linearly with the metalrsquos work

function empty119898 Nevertheless this is only in theory as the presence of localised surface

stated at the edges causes empty119861 to become unresponsive to the metal work function

Consequently Eq(211) is then reordered to match the difference in metal and

semiconductor work function Thus the new equation becomes

119902119881119887119894

= 119902(120601119898 minus 120601119904) (213)

A Schottky contact appears when a metal-semiconductor contact has a large

barrier height (B ge kT) and low doping concentration in the semiconductor (ND le NC) In

the case when the metal-semiconductor contact is under some bias eg reverse bias the

semiconductor will react to a positive bias according to the metal by a voltage V=-VR

This condition will affect the built-in potential and leads to increase from Vbi to

(Vbi+VR) thus increasing the barrier height empty119861 in the semiconductor as well

Consequently electrons are less able to flow from the semiconductor and cross into the

metal Therefore the current flow will be very small This phenomena is shown in

Figure 210(a)

45

Figure 210 Energy band diagram of Schottky contact on n-type material under (a) reverse and (b)

forward bias

As can be seen from Figure 210(b) when a forward bias is applied the semiconductor

is biased negatively with respect to the metal by a voltage V=Vf This will result in a

reduction in built-in potential from Vbi to Vbi-Vf The electrons in the semiconductor will

lower the barrier height and a lot of electrons will escape into the metal causing a large

current to flow Thus a large current flow in the forward direction compared to the

reverse direction Essentially this is the origin why the Schottky contact is named a

rectifying contact [48 49] For the metal side both forward and reverse biases applied

do not affect the barrier high empty119861 because there is no voltage drop there

In this system the electron and holes are transported by a phenomenon called

Thermionic Emission (TE) which happens when the semiconductor layer is lightly

doped Nd lt 1x1017119888119898minus3 The electron will only be thermionically emitted into the metal

when the energy is higher than the potential barrier[50] There is another phenomenon

called Thermionic Field Emission (TFE) which happens when the potential barrier

thickness is very thin (thin enough) to allow the electron to tunnel through the barrier

This will be discussed in the next section as this phenomenon leads to the formation of

an ohmic contact

(a) (b)

46

242 Ohmic Contact

Basically an ohmic contactrsquos purpose is to provide a low resistance path from

the semiconductor to the outside world It is different to a Schottky contact as it is a non-

rectifying contact and does not control the current flow the I-V characteristic of an

ohmic contact is linear in both forward and reverse directions (equality in current flow)

The ohmic contact also has a small voltage drop across it compared to the voltage drop

across the device

If a metal and semiconductor are bought together unavoidably a Schottky

contact will be formed Therefore to create an ohmic contact some techniques to reduce

barrier height and width of the depletion region must be used ie increase Nd In carrier

transport theory there are three mechanisms of carrier transport across the barrier

Firstly the Thermionic Emission (TE) which happens when the carries are excited to

overcome the barrier when the thermal energy is present Secondly the Thermionic

Field Emission (TFE) occurs when the electronholes have enough energy to tunnel

through an adequately thin barrier and some has overcome the low barrier at the top

Finally the Field Emission (FE) which results when carriers can tunnel through the

entire barrier The FE is the most favoured mechanism in the ohmic contact approach

[51]

From the three mechanisms above the current can be determined by the following

equations

1) Current in Thermionic emission (Figure 211(a))

exp (empty119861)

119896119879

(214)

2) Current in Thermionic field emission (Figure 211(b))

exp [

(empty119861)

11986400119888119900119905ℎ11986400

119896119879

] (215)

3) Current in Field Emission (Figure 211(c))

exp (empty119861)

11986400

(216)

47

Where k is the Boltzmann constant T is the temperature 11986400 is the tunnelling parameter

and is related to the doping concentration radic119873119863 The barrier height is denoted by empty119861

Figure 211 Figure 10 N-type semiconductor Ohmic contact with different doping concentration

ND (a) Low (b) Intermediate and (c) high

Figure 211 shows that the carrier transport mechanism is varied by the doping

concentration (ND) As can be seen from Figure 211(c) the doping concentration here is

the highest and influences the depletion region width Xdep to become smaller Therefore

Field Emission (FE) becomes dominant This FE method is the favourite method for

ohmic contact formation [51] and will be utilised in the fabrication carried out in this

work

In fabricating practical devices the ohmic contact is often split into two types

alloyed and Non-Alloyed The difference between the two is that the alloyed type is used

when the semiconductor is doped with a low doping concentration ie less than

1x1018119888119898minus3 while the Non-Alloyed is designed for heavily doped semiconductors with

more than1 times 1019119888119898minus3 doping

The alloyed ohmic contact requires thermal annealing to have a good performance

for electron transport In multi-layer metals one of the metals has the role of donor or

acceptor which is used to increase the doping concentration of the semiconductor If a

temperature anneals eg 420˚ Celsius is applied the metal will diffuse into the

semiconductor and carry the dopant into the semiconductor Therefore a heavily doped

region will be formed and the depletion width becomes narrow establishing the ohmic

contact The key example of this is the usage of the Gold-Germanium-Nickel (Au-Ge-

Ni) alloy where the Ge is the n-type dopant [52] which diffuses into the semiconductor

(a) (b) (c)

48

and perform atom replacement in the semiconductor ie in GaAs it replaces Ga On the

other hand the Non-Alloyed does not require any thermal annealing as it already has a

very high doping concentration and will automatically reduce the depletion region width

The Non-Alloyed ohmic contact has some advantages such as reproducible contact

reduced processing time and good uniformity [53]

25 Asymmetrical Spacer-layer Tunnel (ASPAT) diode Background work

In this section some historical background of the Asymmetrical Spacer-layer

Tunnel (ASPAT) diode is given Since this is the first thesis reporting about this new

device it is worth to mention some historical background about this tunnelling diode

The ASPAT was first proposed by a group of scientists from General Electrical

Company (GEC) in 1990[16] The works led by Richard T Syme and assisted by

Michael JKelly Angus Condie and Ian Dale initiated the idea of launching a new type

of tunnel diode The idea managed to attract the interest of many parties following the

development of resonant tunnelling diode (RTD) which earlier had shown a promising

weak temperature dependence [54] However the interest in RTD is mainly limited to

microwavesub-millimetre wave generation For THz detection the requirement is to

have a significantly asymmetric IV characteristic Given this the ASPAT which has

only a single energy barrier and most importantly weak temperature dependence and

large dynamic range would be a promising candidate for this application

The development of ASPAT is a kind of reverse engineering since it was built

purposely to replace the earliest receiver diode especially the Schottky Barrier Diode

(SBD) which has strong dependence on operating temperature [55] From the time when

it was first revealed a lot of works have been done to realise this most sophisticated

tunnel diode The first attempt which was reported in [56] was meant to gather some

insights into the device by using the well-known Schrodinger and Poissonrsquos equations

for simulation The second attempt on the other hand was directed to physically grow

and fabricate the device Here the real problem occurs At the first stage of qualifying

this device it was found not to be manufacturable Since then a new tunnel diode

structure based on GaAsAlAs materials system was built by both MBE and MOCVD

Its microwave performance was then tested at 94GHz [18] The same paper also

49

reported performance comparisons between ASPAT and another microwave diode ie

Germanium Backward Diode (GBD) PDB and SBD

Work on these devices stopped due to the inability to commercialise the ASPAT

and other tunnel based devices [57-60] The problems associated with low-cost

manufacture of tunnel diodes are due to firstly the thickness of the AlAs barrier layer

the dependence of tunnelling probability (electron) through a single barrier is

exponential and varies by a factor of more than 350 for one monolayer change in the

AlAs barrier thickness[61] The tunnelling of the electron through a barrier is

proportional to the current through a barrier as a function of a bias across the AlAs

barrier[62] Secondly the bandgap which is predominantly happens to be a ternary alloy

with relative composition ie AlxGa1-xAs Here the x can vary the bandgap in the

semiconductor layer For the ASPAT a 1 change in x results in a 30 change in the

current[62] To design an ASPAT for microwave and THz applications the designer

often allows at most plusmn10 variation of the absolute current through a specific diode at a

pre-identified bias This implies that within a wafer the uniformity that the ASPAT must

achieve is less than plusmn01 monolayers while between wafer to wafer the reproducibility

in barrier thickness in average must be identically controlled[63] This explains why at

the qualification stage of investigating ASPAT there was a need to focus on

GaAsAlAs-based material to diminish further errors because of the change in x This

type of work on ASPATs has been carried out by other co-workers at Manchester and

Cambridge

Thereafter the work then focused on repeatability and manufacturability tests

These result in many attempts being carried but failed with unacceptable between wafer

to wafer reproducibility [61 64] The development of reproducibility and repeatability of

the ASPAT was pursued for over 10 years until precise control of the growth of the

thickness AlAs layer was finally achieved using MBE[65] [66 67] This achievement is

confirmed by a current density produced which varied by less than plusmn30 indicating that

the reproducibility of AlAs barrier of the order of plusmn 02 monolayers A final step to

achieve the level control for the ASPATrsquos AlAs barrier thickness was carried out and

resulted in a 1 standard deviation of the IV characteristics for both within a wafer and

different wafer (2 inches wafer size)[17] In the early stage of this work some

50

repeatability and manufacturability test was also carried out Once this vital step is

accomplished further investigation on the ASPAT was made most recently and which

will be covered in this thesis ie temperature independence[68] and new ASPAT

types[69] characterisation to achieve smaller device RF measurement and development

of THz detectors The material systems that have been investigated so far are

GaAsAlAs and InGaAsAlAs both in the ManchesterCambridge group

Recently the ASPAT was commercialised by Linwave Technology as a wideband

zero-bias detector diode This was done in April 2016 Although it is now on the market

the ASPAT remains immature in term of research and development A lot of work is still

required to enhance the device ie working at the sub-millimetre wave using ternary

material etc

26 Asymmetrical Spacer-layer Tunnel (ASPAT) diode The Basics

The basic building block of the ASPAT diode is based on heterojunction of three

multilayer semiconductor structures which have two different band gaps The structure

comprises a thin layer of wider-gap semiconductor sandwiched between two

semiconductors with narrower-gap forming a tunnel barrier The basic principle of the

ASPAT device is based on the exploitation of quantum mechanics theory using

heterojunctions interface According to quantum mechanics theory moving particles

(electrons) with less energy than the barrier height have a probability of appearing on the

other side of the barrier by a tunneling through it This can be achieved in conditions

where the barrier must be very thin (~ 10 monolayer ) This is in contrast to classical

physics where a particle must have kinetic energy at least slightly greater than the

potential barrier height in order to overcome the barrier otherwise the probability of the

particle to appear on the side of the barrier is zero

Since the ASPAT diode operation is based on tunnelling through a barrier one

needs to know that the tunnelling mechanisms can be classified into two types[44]

intraband and interband The latter is described as tunnelling that occurs from

conduction band to valence band (electron) and valence band to conduction band (holes)

This normally happens in bipolar device ie p-n junction diode which has n-type and p-

type doped regions On the other hand intraband refers to tunnelling which occurs when

51

electron tunnel from the conduction band of a semiconductor to the conduction band of

its neighbouring semiconductor The same thing happens to the holes in the valence

band The device with this type of tunnelling is normally a unipolar device which is

either p-type or n-type doped The ASPAT diode can be considered as a device that is

based on intraband tunnelling mechanism Therefore the focus will be entirely based on

its principles

261 Principle of Quantum Tunneling

Generally all tunneling diodes obey the concept of quantum mechanical

tunneling Quantum mechanical tunneling is a phenomenon where a particle is able to go

through an energy barrier higher than the kinetic energy of the particle and if it is thin

enough compared to the de Broglie electron wavelength (λ) If the electron wave is

greater than the barrier the probability of the wave to occur at both side of the barrier is

higher

Figure 212 Classical view of whether an electron is can surmount a barrier or not Quantum

mechanical view allows an electron to tunnel through a barrier The probability (blue) is related to

the barrier thickness

For the case of classical physics (Figure 212(a)) the particles can be confined by

energy barriers of a semiconductor if their kinetic energy is less than the barrier energy

The particles thus require higher kinetic to escape to other states this phenomenon is

called thermal emission In quantum mechanics (Figure 212(b)) the particle is

described in two ways as a wave and as a particle If the particle moves like a wave it

will carry all the waversquos properties Therefore it will not brusquely end up at the

En

erg

y

(a) Classical view (b) Quantum mechanical view

En

erg

y

En

erg

y

52

boundary of the energy barrier Hence when the particles collide with the barrier

(incidence) there will be a probability of penetrating the barrier if the barrier is thin

enough and has a finite height For thicker barrier the probability of a wave that can be

found on the other side of the barrier is very small However the possibility of the

electron wave to appear on the other side of the barrier is increased by thinning the

barrier The potential barrier of semiconductor material technology can be determined by

using Homojunctions structures with different doping profile This will result in a

difference in band alignment and multilayer heterojunction structure (different

semiconductors have different band gap) which includes semiconductors insulators and

conductors (metals)

The easiest way to explain the phenomenon is by considering a potential barrier

Epot(x) with barrier height E0 energy bigger than the total energy E as shown in Figure

213 the potential energy occurs in a finite space between 0ltxlta and is 0 outside

Figure 213 Rectangular tunnelling with incident reflected and transmitted wave function[70]

The electron outside the region of the potential barrier (xlt0 and xgta) is free to

move The effective mass of the electron is different inside and outside of the barrier in

real tunneling devices when implemented using semiconductor heterostructures The

quantum mechanical equations predict the wave nature of matter which states that matter

unveils wavelike properties under some conditions and particle-like properties under

other conditions The wavelike properties as described by the Schrodinger Formula of

E0 Transmitted Ψ = 119862119890minus119894119896119909

Incident Ψ = 119860119890119894119896119909

Reflected Ψ = 119861119890minus119894119896119909 Ψ = 119863119890minus119894119896119909

0 a

E

x

V(x

)

53

quantum mechanics represent a particle penetrating through a potential barrier most

likely as an evanescent wave coupling of electromagnetic waves[44]

To start the calculation of the tunnelling probability the Schrodinger equation is given

by

119894ℏ

120597

120597119905Ψ(119903 119905) = ΨΗ(r t)

(217)

Where ℏ is Planckrsquos constant (662606957 times 10119890minus341198982119896119892119904

2120587frasl ) Ψ(119903 119905)is the wave

function at position r and time t Η is the Hamiltonian operator given by

Η = minus

ℏ2

2119898nabla2 + 119881(119903 119905)

(218)

Where 119881(119903 119905)is the potential energy which is dependent on space and time 119881(119903 119905) is

considered zero for a particle traveling in free space without any potentials The plane

wave with vector r and t is given by

Ψ(119903 119905) = 119890119894(119896119903minus120596119905) (219)

This equation satisfies Eq (217) above under the condition where the particle is

travelling in free space without potential k is the wave vector which is equivalent

to2120587120582frasl and the angular frequency 120596 is 2120587 multiplied by the frequency

In the case of tunneling through a potential barrier the method of separation of

variables is used to simplify the problem as in the equation below

Ψ(119903 119905) = 119877(119903)119879(119905) (220)

It is assumed that the problem above is divided into time-dependent and time-

independent parts 119877(119903) is the spatial component and 119879(119905) is the time-based component

of the wave function The time dependent problem as shown above in the Schroumldinger

Equation (1) can easily be solved by filling up all the finite parameters The solution of

54

the time-independent part gives the tunneling probability For the one-dimensional (1D)

time-independent Schroumldinger equation[44]

119864120595(119909) = minus

ℏ2

2119898

1198892

1198891199092120595(119909) + 119881(119909)120595(119909)

(221)

E is the total energy and 120595(119909)is the spatial component of the wave function along the x

axis The combination to the wave function is given by

120595(119903 119905) = 120595(119903)119890minus(

119894119864119905ℏ

)

(222)

The time-independent plane wave solution of 120595 = 119890119894119896119909 satisfies the equation (221) for

any constant potential V0 in space Plugging in the wave solution yields the condition

that

119896 = radic2119898lowast(119864 minus 1198810)

ℏ2

(223)

Referring to Figure 213 also the barrier with exact rectangular shape with height E0 and

width W the solution of the wave functions and tunneling probability can be extracted

by using the below equation [44]

119879119905 =

[119862119890minus119894119896119909]2

[119860119890119894119896119909]2= [1 +

11986402 sin ℎ2 (119896119882)

4119864(1198640 minus 119864)]minus1 asymp

16119864(1198640 minus 119864)

11986402 exp(minus2radic

2119898lowast(1198640 minus 119864)

ℎ2119882)

(224)

For more complex barrier shape Wentzel-Kramers-Brillouin has simplified the

Schrodinger equation for tunneling probability of carrier which becomes[44]

119879119905 =

[119862119890minus119894119896119909]2

[119860119890119894119896119909]2asymp 119890minus2int |119896(119886)|119889119886

1198860 asymp 119890

minus2int radic2119898lowast

ℎ2 [119880(119886)minus119864]1198891198861198860

(225)

55

This means that the incident electron has a finite probability T of tunneling

through the potential barrier and this leads to the concept of tunneling probability as well

as a tunnelling current Therefore this becomes the basis of tunneling phenomena and

thus all devices which are related to tunneling can be modelled and analysed based on

this basic example The tunneling phenomenon is a majority-carrier effect and the

tunneling time is set by the quantum transition probability per unit time (which is on the

order of picoseconds) rather than the transit time concept [44 71] This enables the

tunneling devices to work at a much higher switching speed They can also be used in

high-frequency applications such as microwave circuit and high-speed oscillators

262 ASPAT Structural Parameters of GaAsAlAs materials System

Figure 214 Epitaxial layer structure of device XMBE 304 (lateral structure) used in this study

Conduction band profile

56

The core materials that make up the ASPAT diode in this work is based on

heterostructure of group III-V compound semiconductors Such materials are chosen due

their mature excellent properties and their band gap which can be tailored to fit the

desired design as well as to improve the carrier mobility In this work the primary layers

that form an ASPAT diode are very thin pure Aluminium Arsenic (AlAs) of thickness

ten monolayers buried in between dissimilar thickness of pure Gallium Arsenic (GaAs)

layers as can be seen in the red circle in Figure 214 above These two GaAs layers are

known as spacer layer which normally have a ratio of 401 or 201 in thickness The

asymmetrical spacers layer and the thin barrier in such arrangement lead to an

asymmetric current-voltage characteristics as proposed firstly by Syme and Kelly[15]

To examine and investigate this GaAsAlAs ASPAT structure in term of electrical and

RF characteristics the device have been grown according to Table 22 below

Table 22 epitaxial layer of sample XMBE304 which is the main focus of this work

Material Doping (cmminus3) Thickness (Aring) Bandgap (eV)

Ohmic Layer GaAs (Si) 4times1018 3000 142

Emitter GaAs (Si) 1times1017 400 142

Spacer GaAs Undoped 50 142

Barrier AlAs Undoped 28 283

Spacer GaAs Undoped 2000 142

Collector GaAs (Si) 1times1017 400 142

Ohmic Layer GaAs (Si) 4times1018 4500 142

Substrate GaAs - 650 microm 142

The arrangement of the multi-layers that form a lattice matched GaAsAlAs ASPAT

diode can be transformed into band structure profile view for easy understanding The

conduction band profile at equilibrium is as sketched in Figure 214 As can be seen in

the Figure 214 the ASPAT diode is generally a heterojunction multilayer structure

tunnelling diode

57

Figure 215 The layer profile of ASPAT with 28nm barrier thickness [27]

The generic structure of ASPAT diode which is shown in Figure 215 above with a

schematic band structure comprises the following (starting from the top)

(1) A thick layer of heavily doped n++

about 4e+18cm-3

of GaAs with a thickness of

approximately 300nm

(2) An intermediate layer of lightly doped n-type about 1e+17cm-3

of GaAs with

thickness of approximately 40nm

(3) A spacer layer not intentionally doped (NID) GaAs with thickness of approximately

5nm

(4) An ultra-thin layer of NID AlAs with thickness of approximately 28nm

(5) A spacer layer of NID GaAs with thickness of approximately 200nm

(6) An intermediate layer of lightly doped n+ ~1e+17

GaAs with thickness of

approximately 40nm

(7) A thick layer of heavily doped n+ about 4e+18 of GaAs with a thickness of

approximately 750nm

Each layer has its own role For instance layer (1) and (7) are used as ohmic

contacts via connection to a AuGeNiAu metal stack This explains why they are

purposely heavily doped (gt 1018

cm-3

) for better low resistance ohmic contacts Two

intermediate layers (layers 2 and 6) are used to prevent the carrier in the contact layers

from diffusing into the undoped layers The two unequal length spacer layers with ratio

58

1198971 1198972 of about 401 are used as voltage arms to yield an asymmetric current-voltage

characteristic The asymmetry means that after a positive bias is applied from the long

spacer region an accumulation layer is formed and it is deeper than that formed by the

negative bias The thin layer positioned in the middle (Layer 4) is the tunneling barrier

The performance of a single barrier ASPAT diode can be optimised depending on

the applications by appropriate selection of the material system so that the band gap and

barrier height of such material can be modified Furthermore the mobility of electron

and doping concentration of the contacts region can be tuned The parameters to tune

during the growth for instant growth interrupt time and growth temperature will also

affect the performance of this diode The key layers that will affect the performance are

the barrier thickness and the two spacer layers enclosing it The study has shown that a

one monolayer change in thickness results in 300 change in in tunneling current for a

fixed voltage point[65]

The following discussions account for the effect of the main structure of the ASPAT

which is related to their performance

2621 Barrier Thickness and height

The probability of an electron tunnelling through a barrier depends exponentially on

the width and height of the barrier as well as the energy that is incident on the barrier[72

73] All these will affect the I-V characteristic of the ASPAT diode The tunnel current is

obtained by summing over all incidents electrons energies with tunnelling probabilities

through the barrier The tunnelling varies approximately as[74]

119879 prop 119890minus120581119889 (226)

Where d is the barrier thickness and κ is defined by the expression below

120581 =

radic2119898lowast(1198810 minus 119864)2

ℏfrasl

(227)

From textbook the tunnelling probability is given by

119879(119864) = 412058121198702

[(1205812 + 1198702)2119904119894119899ℎ2119870119897 + 412058121198702]frasl (228)

Where K is expressed as

59

119870 = radic2119898lowast119864

ℏfrasl (229)

Where ℏ represents the reduced Planckrsquos constant (h2π) E and m are the electron

energy and effective mass respectively Thus by inputting appropriate value into these

equations one finds that reducing the barrier thickness by one monolayer increases the

tunnelling probability by a factor of nearly three for every electron that tunnels through

the barrier As a result the current will also increase Further it indicates that the

tunnelling strongly depends upon barrier thickness and height By contrast the current

does not strongly depend on temperature

2622 Spacer Thickness

The reason for having two dissimilar undoped spacer lengths is mainly to avoid

diffusion of dopant to the barrier and subsequent layers during growth but in the case of

the ASPAT diode the spacer can also act as a voltage arm Varying the thick spacer

layer (1198971) results in the reverse current decreasing as the layer thickness increases and

varying the thin spacer layer (1198972) will affect the forward current which increases as the

thickness reduces To obtain appropriate asymmetrical I-V characteristics one needs to

maintain an adequate ratio between these two spacer thicknesses While a too thin

1198971 results in high leakage current at reverse bias a too thick 1198972 results in low forward

current One also needs to keep it thick enough to prevent carrier diffusion

These two spacers must be kept undoped or very low doped to allow the electron

moving in the electron mean free path region as it is clear from ionised donors Under

large forward bias an accumulation layer is formed between the spacer and barrier

segment and it is more noticeable compared to the accumulation layer that is formed if a

negative bias is applied

60

Figure 216 Conduction band diagram showing band bending and 2DEG formation at the L1

spacer

Consequently a triangular well is formed which creates an emitter 2D electron gas

(2DEG) population The electrons in this 2DEG occupy the quasi-bound states which

mean high excitation energy thus allowing the electron to tunnel through the barrier as

depicted in Figure 216

In term of RF performance it is important to highlight that a thicker spacer layer

will affect the depletion region which gets wider and thus will reduce the junction

capacitance of the device as per the following expression

119862 = 휀0휀119903

119860

119889

(230)

Where A is the area of the device d is the thickness of the main device structure which

consists of spacers barrier and well layers 휀0 is the permittivity of free space and the

relative permittivity of the spacer material is denoted by εr However the intrinsic delay

time will also increase and hence degrade the device high frequency performance

Therefore optimisation through spacer thickness requires trade-off between reducing

leakage current at reverse bias and degrading device junction capacitance

GaAs

GaAs

AlAs

2DEG

Γ

X

61

263 ASPAT Electrical Parameters

The classical approach in determining the current flow through an ASPAT diode is

by solving the Schrodinger and Poison equations Prior work had been done by Syme et

al in 1991 Due to the fact that AlAs barrier is very thin tunnelling is assumed to occur

at the Gamma valley ie AlAs bandgap= 283eV (rather than X valley)[18 75] and only

from accumulation layer (2DEG)[59] Here the DC characteristic of the ASPAT diode

can be calculated by solving Schroumldinger equation with the position vector represented

by z (in this case) Thus the equation is expressed as

minus

ћ2

2nabla

1

119898lowast(119911)nabla120569 + |119890|120593(119911)120569 = 119864120569(119911)

(231)

Supposing the current is uniform across x and y planes then this can be simplified to one

dimension Therefore the 1 D Schroumldinger equations becomes

minusћ2

2119898lowast

1198892

1198891199112120595 + |119890|(120595 minus ∆120595) = 119864119911120595

(232)

Where

120595 =

120569

exp (119894119896119911119911)

(233)

where ∆120595 is the correction term which reduces the effective barrier height The

Schroumldinger equation is solved using different values for Ez thus the quantum

mechanical current density in the z-direction is now expressed as[15]

119895119911 =

minus|119890|ћ

2119898lowast(120595lowast

119889120595

119889119911minus 120595

119889120595lowast

119889119911 )

(234)

It is different for a heavily doped contact which can describe as below the envelope

functions in the left and right contacts respectively can be described by plane wave

120595119897 = exp(119894119896119897119911) + 119877 exp(minus119894119896119897119911) (235)

120595119903 = 119879119890119909119901[119894119896119903(119911 minus 120580119873)] (236)

62

In this case the left contact covers the region zlt0 while the right contact covers the

regions zgt 120580119873 Equations (35 and 36) are then inserted into Eq (34) to form the

following expression

119895119911 =

|119890|ћ1198961

119898lowast(1 minus |119877|2) =

|119890|ћ1198961

119898lowast|119879|2

(237)

Where R (Ez) and T (Ez) are the complex reflection and transmission coefficients

respectively and they are solved by using the transfer matrix method This method has

been described in reference [56] The next step is to integrate the current in the z-

direction 119895119911 for all possible Ez values Thus the expression for the current density

becomes

119895119911 =|119890|119898lowast119896119861119879

2120587ћ2int(1 minus |119877|2)

infin

0

119897119899 |1 + exp (

119864119891 minus 119864119911

119896119861119879)

1 + exp (119864119891 minus 119881|119890| minus 119864119911

119896119861119879

|

(238)

The equations above are used to calculate the current density approximation from

the ASPAT main structure (two spacer layers and one barrier) only and based on

intraband tunnelling from the conduction band profile For real fabricated structure the

calculation must take into account both intrinsic and extrinsic elements of the diode

While the latter is mostly related to the pad and probe that is used to extract the I-V

characteristic the first element mostly comes from the epitaxial layer of the ASPAT

diode itself The ASPAT I-V characteristic is shown in Figure 217 which clearly

indicates nonlinear characteristics and thus can be used for detection applications

63

Figure 217 I-V characteristics of a fabricated ASPAT diode

2631 Intrinsic Elements of the ASPAT diode

In order to extract the intrinsic electrical characteristic of the ASPAT diode a

generic structure as shown in Figure 218 is essential Two main sources of contribution

to the electrical characteristics are the interfaces of each layer and properties of the

materials themselves

Figure 218 Intrinsic Elements of the Asymmetric Spacer Tunnel Layer (ASPAT) Diode

64

The electrical current flowing from the top contact to the bottom contact will go

through each epitaxial layer producing a close loop Each junction interface limits the

current flow and sum up the total resistance resulting in what is known as the diode

series resistance (Rs) In the ASPAT main structure there is a junction capacitance (Cj)

due to the undoped regions surrounded by heavily doped contacts thus acting as a

parallel plate capacitor The fully depleted capacitance (Cj) of the diode can be

expressed as in Eq (230)

2632 Series resistance of the ASPAT diode

The total series resistance (Rs) of an ASPAT diode can be calculated based on the

finished fabricated diode structure In general Rs depends on three contributors namely

the non-uniformities in the contact metallization the un-depleted epitaxial layer (total

thickness) on both side of the heterostructures and the resistance caused by the

spreading current from the Mesa into the much wider second contact layer ie doped

substrate or 2nd

ohmic layer[76] In fact the contribution toward building up the total Rs

solely depends on how the structure is designed In this work two types of structures

were deployed namely lateral structures and vertical structures These will be described

in the next subsection

For both types of structures the series resistance (Rs) of the ASPAT diode consists

of aspecific Ohmic contact resistance (ρcA) contact epitaxial layer resistance (Repi-

Layers) and spreading resistance (Rspr)[33 77] where Rspr is influenced by the type of

structure The specific contact resistance is obtained from Transmission Line

Measurements (TLM) of the sample The theory of the TLM will be discussed in detail

in the next section The expression for the specific contact resistance is

120588119888 = 119877119888119871119879119908sinh119889

119871119879frasl

cosh 119889119871119879

frasl

(239)

Here Rc is the contact resistance LT is the transfer length (effective length) w is the

contact pad width and d is the length of the contact pad

65

Repi-Layer is the sum of all doped layers that sandwich the main ASPAT device For

each doped layer the resistance is given by

119877119890119901119894minus119897119886119910119890119903 = 120588

119871

119860

(240)

Where ρ is the resistivity which is given by 120588 = 1

120583119899119902119873119863 L is the epitaxial layer thickness

in cm A is the device area (emitter size) micron denotes the mobility of the electron q is the

electron charge and ND is the donor concentration The spreading resistance depends on

the structure design of the device This will be elaborated in detail in the following

section

26321 Vertical structure (doped Substrate)

The XMBE307 structures were grown on n+ GaAs substrate to provide the

simplest fabrication process The cross section of the finished single diode can be seen in

Figure 219 below

Figure 219 Cross-sectional view of ASPAT vertical structure the dimensions of d b and h are not

drawn to scale

66

At low frequency and in a mesa that is etched into a doped substrate material the

spreading resistance can be approximated by[76]

119877119904119901119903 =120588119904

2119889

(241)

Where ρs is the substrate resistivity and d is the ASPAT diode mesa length

However the spreading resistance is increased at high enough frequencies as the skin

depth (δ) in the substrate is much lower than the effective mesa length (d) of the diode

A new spreading resistance is then calculated also based on the assumption that the skin

depth is much lower than the chip thickness (h) but much larger than the mesa length

Thus the spreading resistance at high frequency is given by

119877119904119901119903(119891) =

120588119904

120587120575[05 ln (

119887

119889) +

119887]

(242)

Where the skin depth (δ) is taken from standard planar formula and can be expressed as

120575 = [

2120588

(120583120596)]

12frasl

(243)

Where micro is the permeability and ω is the angular frequency During DC measurement

this type of structure requires having good suction on the stage for a good contact

However for small die (15mm times15mm) the suction sometimes is not strong enough to

provide a very good adhesion to the sample Therefore another type of ASPAT diode is

deployed which is based on the lateral structure by utilising semi-insulating substrate

and both contacts are connected to probes

67

26322 Lateral structure (Semi-insulating Substrate)

Figure 220 Cross-section view of the complete process ASPAT diode (XMBE304) The dimensions

are not drawn to scale

In order to obtain accurate measurement results so as to avoid contact errors to the

substrate between stage and Device Under Test (DUT) a lateral structure as shown in

Figure 220 above is deployed This type of design offers many advantages ie it

provides a path for on-wafer RF measurement However the proper design has to take

into account the increase in RS due to improper attention to the spreading resistance

This spreading resistance is different from the vertical structure that was discussed

above In this case it is mainly caused by a gap at the bottom contact The gap in the

horizontal direction between epitaxial layer and metal at bottom contact is denoted as D

gap in in Figure 220 Therefore Rspr for the lateral design can be expressed as [77]

119877119904119901119903 =

1

120587120590119889119866119886119860119904ln (

119886

119886119898119890119904119886)

(244)

Where σ is the conductivity between two coaxial half-cylindrical electrodes with inner

(amesa) and outer (a) rectangular length or bottom ohmic layer which is given by (σ =1

ρ) a d and amesa are the length and thickness indicated in Figure 220 above Noticeably

68

the D gap will have direct effect on the outer length (a) of the device which is also

proportional to Rspr For high-frequency operation where the skin depth is less than

d(GaAs) σ becomes

120590(120596) =

120590(0)

[1 + (120596120591119903119890119897)2]

(245)

Where τrel =microme and micro m as well as e are the mobility effective mass and electron

charge respectively It is recommended that for high-frequency applications the device

series resistance must be as low as possible

Hence for both type of structure the ASPAT series resistance is calculated based on

all the above-stated resistances and these are set by

119877119904 =120588119862

119860+ 119877119890119901119894minus119897119886119910119890119903119904 + 119877119904119901119903

(246)

The total RS can be decreased by increasing the emitter area of the diode However a

large device will not able to reach millimetre and submillimeter wave region (THz) as

the capacitance will also increase (Eq 230) Therefore both parameters will have a

trade-off between them to be able to work at ultra-high frequencies

27 Characterization of Ohmic Contacts

The semiconductor Ohmic contact can be characterised using techniques that will

be described in the following section First is the Cox-Strack technique which is

specially designed to characterise bulk type semiconductor (thick) contact resistance on

two opposite sides The detailed description of this technique can be found in [78]

Second is a technique called Four Point Probe This technique was developed in 1954 by

Valdes etel [79] to characterise semiconductor resistivity It can also be used to

characterise the contact resistance for planar type devices As this research does not

cover this method the details of the measurement can be referred to [80] Finally the

most common method which is also extensively used in this research is the standard

Transmission Line Measurement (TLM) The details of this method will be explained in

the next section There are simplified versions of the TLM method which require just

one lithography step but are nevertheless very powerful in characterising and optimising

the contact resistance known as Circular Transmission Line Measurement (CTLM)[81

69

82] However in this research this is not to be covered as the standard TLM is already

adequate for planar type devices

271 Transmission Line Measurement (TLM)

The formation of metal and semiconductor interfaces will create a contact that

becomes very important for the characterisations of any fabricated device Additionally

it enables the quality of certain process flow to be determined This interface must be

evaluated by a technique known as the Transmission Line Measurement (TLM) TLM

which was first introduced by Murrmann and Widmand [83] in 1969 underwent some

refinements by Berger [84] in 1971 The theory of the TLM can be described by

constructing a TLM structure which comprises a set of metals contact pads placed in

series on a highly doped semiconductor layer as depicted in Figure 221 The structure is

designed like a series of the islands to permit current flow in parallel to the contact pads

[80] which is a direction defined by etched patterns Each contact metal pad behaves

like a MESA which has a thickness (t) and width (W) The distance between each metal

contact pads is defined by d1 the gap between two neighbouring contact pads which are

beneath each contact pad is defined as effective length LT This will allow current flow

in and out of the subsequent neighbouring metal pad The resistance elements that will

be extracted are RA and RB which sit under the contact and in between two metal pads

These two elements represent the sheet resistance under the metal contact pad area and a

sheet resistance of the material between two metal pads

Figure 221 A simple TLM structure with effective length and sheet resistance underneath

t L

T

LT

RA R

A R

B

dn

Probe Probe

Metal Pad

GaAs

MESA

70

The basic relationship of resistance R with respect to the size of the metal contact

or in the standard transmission line can be expressed as [44]

119877 = 120588

119871

119860= 120588

119871

119905 times 119882=

120588

119905times

119871

119882= 119877119904ℎ

119871

119882

(247)

Where ρ is the materialrsquos resistivity L is the length t is the thickness W is the width

Rsh is the sheet resistance and A is the cross-sectional area of the transmission line The

unit for ρ and Rsh are Ωm and Ωm2 respectively

The total resistance RT of this structure can be taken from the sum of the two

neighbouring padrsquos resistance RA and RB In order to relate with Eq (247) above this

RT will be substituted into Eq(247) to become

119877119879 = 2119877119860 + 119877119861 = 2119877119904ℎ119860

119871119879

119882+ 119877119904ℎ119861

119889119899

119882

(248)

As suggested in [85] RshA and RshB are assumed to be identical Therefore Eq(248)

can be reorganised into specific contact resistance RC and semiconductor sheet

resistance Rsh above as 119877119862 = 119877119904ℎ119860119871119879

119882 the new equation is then

119877119879 = 2119877119862 + 119877119904ℎ119861

119889119899

119882

(249)

The common practice throughout this research is to design a TLM structure that

has a ladder structure consisting of 10 metal pads with each one measuring to a size of

100microm width and 50microm length and the space between the first and second metal pad

5microm The gap is increased after the second metal pad by a further 5microm until ten metal

pads are completed produce a separation between the ninth and tenth pad of 45microm The

TLM ladder structure as depicted in Figure 222 is supplied by a constant current of

1mA at the very left and right metal contacts by two probes This allows the extraction

of RC (Ωmm) and Rsh (Ω) instantly from such structure The potential difference

between the two adjacent metal pads is measured by another two probes and the reading

of a voltmeter is recorded The total resistance is obtained by using Ohm law where

voltage is divided by current (VI) Another voltage reading is taken for the next two

neighbouring metal pads until the largest gap is reached The readings of (conversion

71

VI) which result in the corresponding resistance RT are then plotted against spacing and

the result can be seen in the graph in Figure 223

Figure 222 Top view of TLM ladder structure use in this work

Additionally in the measurement the voltmeter used in work has a very high resistance

Otherwise there will be leakage of current occurring through the probes and cables

Therefore the parasitic resistance of the cable or connector and the probe contact can be

ignored The key parameters that can be extracted from the graph will be discussed in

the next paragraph

Figure 223 Typical plot of resistance versus TLM spacing

y = 02592x + 02375

0

2

4

6

8

10

12

14

0 10 20 30 40 50Res

ista

nce

Rn (

Oh

ms)

TLM Spacing dn (um)

LT

d1 d

2 d

3 d

n

I (1mA)

V

V

LT

ME

SA

Su

bst

rate

72

The straight-line graph plotted in Figure 223 can be referred to Eq (249) and

this must be done by assuming the sheet resistance Rsh of the material is constant If the

straight line of the graph is extended up to initial gap (d=0) the intercept on the y-axis

provides the 2RC value To extract the 2LT further extrapolation is made until the

interception at RT = 0 is reached Therefore an important parameter which is the specific

contact resistance ρC can be found from this expression[85 86]

120588119888 = 119877119862119871119879119882sinh

119897119871119879

cosh119897

119871119879

(250)

Where l represents the total conductive semiconductor thickness The final part that can

be extracted from the graph is the slope of the line This is obtained by dividing the sheet

resistance with the width of the metal pad ie represented by expression (119877119904ℎ

119882frasl )

28 Basic Characterization Techniques and procedures

281 Measuring tools and apparatuses

The success of every experiment is determined by the backend results that are

obtained from the measurements It is important to choose an appropriate instrument

which will provide the required data for a valid and detail analysis Thus this section

will give a brief explanation of the measuring instruments and methods that were

exploited in this research The measurement apparatus systems that are available and

have been utilised in completing this thesis are ldquoset of DCrdquo and ldquoSet of RFrdquo

measurements The DC set measurement consists of room temperature and variable

temperature system This system is built to perform process monitoring during device

fabrication and it comprises of five main components

The fundamental component in the ldquoDCrsquos set toolrdquo is the Agilent B1500A

Parameter Analyser [87] used to provide fixed currentbias during testing The other

component is a Karl Suss PM5 Cascade Prober [88] which is used to receive fixed

currentvoltage and to supply its to the semiconductor contact via probe tip The PM5

Prober has at least four probe arms and each of them is fitted with ldquoneedlerdquo called probe

73

tip The size of the tip that is normally used here is 2microm and in some cases the tip size

of 1microm is also utilised The currentbias supplied to the sample must go through the two

of Source Measurement Units (SMU) namely SMU1 and SMU2 to ensure no mismatch

occurs between parameter analyser and diode All the testing are controlled using a

software called Integrated Circuit Characterization and Analysis Program (IC-CAP

2009) brought from Keysight Technologies[89] The software is installed on a standard

Personal Computer (PC) and the PC is connected to a General Purpose Bus Interface

(GPIB) to link with the B1500A Parameter Analyser This system can be organised

based on the purpose of measurements ie IV characteristics TLM Schottky Diode

and transistor as it is very flexible to change the configuration For examples

Transmission Line Measurement (TLM) configuration requires the addition of a digital

multi-meter and four-point probe tip while diode measurement just needs two point

probe tip without a digital multi-meter Figure 224 illustrates the measurement system

for a set of DC to test the TLM structure

Figure 224 The set of DC measurement apparatus arrangement to measure the TLM

On the other hand the set of RF measurement consists of five block elements

configuration in the system namely the Vector Network Analyser (VNA) DC

sourcemonitor Cascade Microtech Prober SMU and Control PC via GPIB The RF set

Agilent B1500A

Parameter analyser

DUT

(TLM Diode

Capacitance)

Karl Suss

PM5 Cascade

Prober

PC

(MS windows 2000)

Digital Multi-meter

ICCAP 2014 Provide current

source

Measure and read the

voltage Stage and Probe

Signal

Current Source

SMU1 SMU2

General purpose bus interface

74

performs RF characterisation after the device fabrication is completed This system can

also perform DC characterisation as its basic instrument has this function too The VNA

machine used in this research is the Anritsu 37369A [90] which can perform the

Scattering Parameter (S-Parameter) measurement with a frequency range of 40MHz to

40GHz The DC sourceMonitor utilised in this experiment is the HP 4142B Modular

DC SourceMonitor [91] Both of these sub-systems are controlled by a standard PC

which exploits GPIB port to link them During operation the HP 4142B is connected to

the VNA by an internal bridge network and two SMUs to the Cascade Microtech Prober

which is then connected to the bond pads of the device It is identical to the set of DC

measurement where the SMU setting and data assembly are accomplished by IC-CAP

software package therefore the data that was obtained before and after completing the

fabrication can be compared This will enhance the validation of the data

However the stage and probes in both sets of measurement are different The

Cascade Microtech Prober has only two probe arms with each arm fitted with a 3-

fingers probe tip in the arrangement of Ground-Signal-Ground (GSG) as shown in

Figure 225 Each finger (pitch) is separated by 100microm thus to fit in with the pitch and

to reduce the mismatch in resistance the bond pad design must follow this separation

between each contact The GSG is configured by connecting the outer pads (Collector)

to the Ground probe tip while the inner pads (Emitter) are attached to the Signal probe

tip where the RF signal is sent and received through it Figure 226 shows the actual set-

up for RF measurement used in this research

75

Figure 225 The set of RF measurement for on-wafer DC and RF one-port measurement

Figure 226 Actual VNA system that was used for RF characterization

282 Measurement steps using a VNA

The RF measurement steps can be summarised in the following

- Step 1 Select or find the suitable VNA depending on applications

PC

(MS windows 2000)

ICCAP 2014

VNA

Anristru 32379A

40MHz to

40GHz

HP4142B

Modular DC

source

DC Bias Source

Cascade

Microtech

Probe Station

One 50microm pitch G-

S-G probe tips

Ground

SMU1

Cathode

Gen

eral

purp

ose

bus

inte

rfac

e

Lo

w P

ow

er H

igh P

ow

er

Cathode

Anode

SMU2

Ground

Ground

Signal

76

There are few factors that need to be well-thought-out before starting to use a VNA

especially for S-parameter measurements The factors are the availability of the VNA in

term of operational frequency and measurement port types air-filled metallic waveguide

or on-wafer Not all VNA can have an operational frequency for banded measurement

ie W-band Ku-band etc All these may require external signal sources to extend the

operational frequency

- Step 2 Properly setting up the VNA

The VNA can be set up depending on its application and the goal of measurement for an

instant number of point requires a desired frequency span IF bandwidth and the supplied

power level This very important to ensure the desired measurements are correct and

appropriate

- Step 3 Appropriate calibration system

In order to have a valid calibration appropriate calibration method has to be chosen

depending on the applications This will determine the accuracy and standards of

calibration As for on-wafer calibrations the de-embedding is normally used while for

the off-wafer the SOLR technique is more suitable to employ

- Step 4 Validation or verification of the calibration results

It is vital to validate the calibration results to ensure that the system has been properly

calibrated

- Step 5 Proper measurement

Proper alignment positioning and touching from probes tip to DUT is necessary to

guarantee a good repeatability and reproducibility of measurement results Normally

when positioning the probe an alignment marker is used as an aid By doing this similar

travel distance for the probes can be achieved The measuring plane will also be equally

well-defined

283 Measurement Practice and Flowchart

Essentially the device characterisation is performed in two stages ie during

fabrication as a process monitoring and after completion for data collection and

analysis Figure 227 shows the block diagram of a flow chart for testing a 15mm times

15mm wafer processing performance In the wafer processing after reaching the top and

77

bottom contactrsquos step the sample can be examined by measuring the TLM structure

according to the TLM procedure The measurement is conducted by exploiting a set of

DC measurement apparatus as mentioned above analogous to the TLM ladder structure

on the wafer surface Based on the TLM results the presence of any process issue during

the fabrication can be identified by examining the parameter such as contact resistance

and sheet resistance Thus a decision can be made either to proceed or to terminate the

fabrication should any issue is found early on As a result no materials will be wasted

further When a dielectric layer is involved Capacitance Dielectric measurement can be

tested This practise can be used to obtain the quality of the dielectric layer To connect

between a diode effective area and probe a bridge is requires It can be attached to the

diode emitter (to determine the diode size) to a bond pad for probe tip to touch This

bridge can either be left hanging in air or sticking to the isolated dielectric layer For the

GaAsAlAs material system the latter technique is preferable since it avoids issues with

the air-bridge which will be discussed in detail in Chapter 3 The opening area (via) for

metal connection can be checked by measuring the resistance on a special design pad

after a plasma dry etching step

Figure 227 Block diagram of the ASPAT measurement step

The next stage of characterising the device is the on-wafer diode measurement

which takes place after completion of all processing (including bond padsco-planar

pads) The work is carried out using the above set for RF measurement and employed

Ohmic Contact

Opening Via

TLM

Qualified

Device

Bad device

Dielectric

capacitance

Bonding pad DC and RF

Good

Good Good

Fail

Fail

Fail

78

purposely to access the DC and RF performance of each diode where the current-voltage

(I-V) characteristic and S-Parameter results are obtained In fact DC measurements are

first performed using a set of DC measurement and a rough IV characteristic can be

obtained to ensure the diode is working properly Usually the yield of any fabricated

device on 15mm times 15mm wafer in this research is between 70 to 90 In this research

the outcomes that will be discussed in the subsequent chapters are in term of the average

of measured values Thus it is very important to have a meaningful data to compare with

physical modelling and simulation in the future The diode IV characteristics are studied

by applying different DC bias at the emitter to collector terminals to extract its keys

parameters ie turn on voltage (supposedly zero bias) non-linear characteristic Rj RS

Cj etc

The device that has been measured by DC and having produced a valid result

will be marked for the next investigation ie the S-Parameter measurement This

measurement is executed using a similar system but with different probe types The

three fingers probe type is used and the device frequency response is measured via the

one-port network from the VNA The important parameters extracted from this

measurement are usually S11 (depending on how many ports are measured) Although

working devices are selected to measure accordingly there is a need to ensure the VNA

RF cable and probe tips are calibrated so that only valid data without errors will be

obtained A calibration technique called SHORT-OPEN-LOAD (SOL) is performed

prior to each daily measurement by exploiting a calibration sample with WinCal

software (Cascade Microtech)[92]

To avoid confusion it is worth mentioning that this technique used to calibrate

the device structure is different from what is used in SOL calibration to the equipment

Furthermore the de-embedding calibration is made on-wafer with the same substrate of

the actual device whereas the SOL is performed on a special calibration substrate

Normally the de-embedding results are not constantly automated with the VNA

equipment However the measurement is done separately starting with the special DUT

substrate then followed by the OPEN and SHORT de-embedding structures

79

3 RESULTS OF FABRICATION PROCESS OF ASPAT DIODES

31 Introduction

This chapter presents in detail the general fabrication techniques for a generic and

development work of micron scale ASPAT diodes The discussions will focus on the

semiconductor growth technique used in this work ie MBE and the fabrication process

steps which include sample cleaning photolithography etching and contact

metallization These techniques are ample to build and deliver commercially marketable

fabricated structure Hence all photolithography techniques used to complete this

project are based on conventional i-line optical lithography which is adaptable to

industry and commercial purposes

The fabrication process in this work can be fragmented into two major works

firstly the fabrication towards reproducibility repeatability and manufacturability in

term of device structure process flow and DC amp RF characteristics This work will

involve relatively larger emitter area which varies from 15times15microm2 to 100times100microm

2 The

larger area provides for ease and fast fabrication as well as DCamp RF measurements

Once repeatability and reproducibility of the process flow and performance is confirmed

the second part of this work which is concerned with applications in millimetre and sub-

millimetre-wave then took place In this work small emitter designs varying from

2times2microm2 up to 10times10microm

2 were considered with appropriate measurement pads The

successful fabrication of smaller diode geometries in the second part of the programme

naturally leads to a further investigations to understand its epitaxial layer structure and

extracting intrinsic components ie junction resistance junction capacitance and series

resistance which will determine the device performance in high-frequency applications

The I-V characteristic is obtained from DC measurement which is usually performed at

room temperature Its results are then compiled and compared with advanced simulation

It is worth mentioning that all samples that are investigated in this project are grown by

means of MBE and the activities related to the epitaxial layer growth using MBE in the

University of Manchester were done by the Materials Growth Team (Prof Missous) and

the authors has no responsibility for this particular task

80

32 Epitaxial Layer Growth Techniques

Before discussing the principle of the common lithography technique it is

important to discuss the wafer growth technique as it comes first before the fabrication

The growth technique that is extensively used in this study as well contributing a lot in

the electronic semiconductor industry is Molecular Beam Epitaxy (MBE) The following

section will discuss the basic operating principle of solid source MBE

321 Molecular Beam Epitaxy (MBE)

The MBE technique was developed in the early 1970s [27] and is purposely used for

growing high purity epitaxial layers of compound semiconductors Such sophisticated

growth technique provides significant functionality ie precise control of the thickness

(to one atomic layer) and contributes to the growth of various types of complex

semiconductor multilayers high quality and advanced materials This level of control is

vital for an assortment of heterostructures devices that are being utilized as part of the

development of the advanced electronics devices especially for the ASPAT diode which

require 01ML control over the AlAs barrier to attain acceptable variability in device

characteristics Additionally the accurate doping profile and excellent junction

abruptness also can be achieved by using this technique

Practically the MBE system used in this work is a solid source MBE which

utilises beams produced by heating up various sources The sources can be Si Al Ga

As and other group III-V compound semiconductors When the crucibles which contain

the sources are heated atoms or molecules of the various elements are evaporated and

travel in straight lines paths like beams directed toward a target (heated and rotating

substrate surface) The condition of the vacuum during evaporation is ultra-high vacuum

(UHV) ~10-11

torr in order to have high quality crystals The substrate is heated and

rotated to provide good growth uniformity over large areas ( up to 4times4rdquo wafers) [93]

The growth rate in typical MBE growth is ~1ML second and can be controlled

by the source temperature in the crucible The abruptness at the heterojunctions interface

and switching of the growth compositions can be obtained by precise control of shutters

that are placed in front of the crucibles Therefore an abrupt junction at GaAs and AlAs

81

interface can be formed to realize the barrier in the ASPAT diode In order to monitor

the quality of the growing crystal and measure the layer by layer growth mode

Reflection High Energy Electron diffraction (RHEED) technique is utilised This

technique works based on the diffraction of electrons from the crystal surface [93]

Additionally given that the ASPAT current density is very sensitive to the barrier

thickness a study has been made using different growth techniques namely MBE and

Metal-Organic Chemical Vapour Deposition (MOCVD)[59] From this investigation it

was concluded that the percentage local variability of current density produced by MBE

grown diodes is better than those grown by MOCVD Thus in this study to get benefit

from its performance all wafers are MBE grown

33 Basic Principles of Common Fabrication techniques

This section covers the generic fabrication process which underpins

reproducibility repeatability and process optimisations for high-frequency applications

331 Sample cleaning

Essentially semiconductor processing requires a ldquoclean environmentrdquo to produce

devices The clean environment is classified according to how many ldquounwantedrdquo

particles are present in a cubic meter There are four categories of clean room available

in the industry Class 10 Class 100 Class 1000 and Class 10000 [94]

The fabrication of all ASPAT diodes in this project was performed in a clean

room environment of Class 1000 equipped with laminar air flow and filter system to

give Class 100 or better during processing Although the sample was processed in a

highly controlled particle environment there is still a high chance for a sample to get

contaminated when handled by a human Besides this the source of particle which

contributes to the contamination can be from the apparatuses and processing equipment

used in the laboratory themselves Thus the process of cleaning the sample wafer

surface before the start of each step is vital

Generally in a clean room the standard solutions that are used to clean a sample are

N-Methyl Pyrrolidone (NMP) Acetone Propan-3-ol (Iso-Propane-ol) (IPA) and

82

deionized (DI) water The sample which is cut up into 15times15mm2 size tiles is cleaned

based on the following procedures

1 Hot NMP- The sample is dipped into the solution at 80˚C for 10 minutes This

solution acts as an organic type of nature pollutants removal

2 DI water- acts as NMP remover The sample is then washed by flowing DI water

throughout the samplersquos surface

3 Acetone- to ensure any remaining NMP is completely removed from the sample

The sample is dipped into Acetone for 5 minutes in a low power ultrasonic bath

at ambient temperature

4 IPA- is used to remove the Acetone from the samplersquos surface The sample is

then dipped into the IPA for 5 minutes in a low power ultrasonic bath at room

temperature The use of low power for the ultrasonic bath is to avoid the sample

cracking or breaking

5 Once done the sample is then blow-dried using nitrogen (N2) gun to remove any

moisture coming from the IPA Fortunately it is easy to remove the IPA

completely from the sample surface given its higher rate of evaporation

Once all these steps are accomplished a visual inspection using a high magnification

optical microscope is conducted to ensure the sample surface is clean The cleanliness of

the sample is determined by the (lack) of particles or other spots (liquid mark) that can

sometime be observed during the inspection Obviously the lower the number is the

better the sample is as it is impossible to totally remove dirt especially marks

Sometimes it is hard to remove the particles in one go There is always a need to repeat

each cleaning step for a few times However this will not affect the sample in term of

electrical performance as the cleaning solutions used are non-destructive

332 Photolithography

Lithography or sometimes called pattern transfer is the most important step in

realizing microelectronic devices The designed geometry and dimensions on a quartz

glass plate called a ldquophoto-maskrdquo must be done prior to the fabrication process While

the photo mask can be designed by using various software tools in this work the design

83

is done via the Advance Design System (ADS) by Keysight The details of the design

which includes three different mask designs will be covered in Section 34 The mask

consists of the desired patterns (master) that can be printed onto solid materialrsquos surface

by means of an electrochemically sensitive polymer (photoresist) using

photolithography In fact this type of optical lithography technique can be performed

with and without a mask due to its simplicity It is also easy and cheaper compared to

other techniques such as x-ray lithography or Electron-beam lithography (EBL)

Although the latter is expensive it is still worth to have since it provides a higher

resolution which is preferable when developing sub-micrometre technology

processing[95] The special feature about EBL is that it does not need a mask to pattern

samples but can be produced by the movement of an electron beam point source and

hence writing the patterns directly on the surface

The ultra-violet (UV) based light sources are more popular among researchers

and development workers because they are cheaper and have modest resolution Usually

photolithography operates at wavelengths (λ) from 193 nm to 436nm The source that

provides the UV light is a Mercury (Hg) arc lamp which uses narrowband filters to select

single emission lines First is the i-line at λ= 365nm then the h-line which is of lower

resolution and has λ of 405nm and thirdly the g-line with a λ= 436nm[96] The

conventional optical lithography used in this research uses the ldquoi-linerdquo at a wavelength of

365nm For shorter wavelengths than these excimer laser or krypton fluoride laser with

a λ= 248nm and argon-Fluoride with a λ = 193nm are also used in the industry The

higher power levels enable higher productivity (throughput) while narrower spectral

widths reduce chromatic aberration provide better resolution and larger depth-of-focus

In this research all the fabrication process are done by utilizing a conventional optical

lithography (i-line) using a Karl Suss MA4 mask aligner Before starting any UV

exposure it is important to check the UV light intensity as it will affect the resolution

and thus desired device dimensions Therefore every corner that is exposed to the UV

light is calibrated to be at 09mWatt power exposure

The complete set of photolithography components consists of photoresist

(photosensitive polymer) photo-mask (chromium) which is used to block the UV to

form a pattern mask-aligner and developer (chemical solutions) Standard fabrication

84

process usually practiced at the University of Manchester starts with sample cleaning as

mentioned earlier Then the sample needs to go through heat treatment to remove all the

moisture with a temperature set to be 150˚C and bake for 5-10 minutes After having

cooled down the samplersquos surface is coated with a thin layer of photoresist via a

technique called spin-coating using a Laurell CZ-650 series spinner The spinning speed

is set depending on the type of resist ie 3000rpm for negative photoresist and 5000rpm

for positive photoresist The rotating speed of the spinner does not have much effect on

the coated photoresist thickness but will have consequence on the uniformity distribution

over the sample surface The coated thickness photoresist however depends on the

concentration of the specific photoresist Once spin-coated is done another short heat

treatment (1 minute) is required to ensure the resist is hard enough to contact a

photomask and to remove any excess solvent The temperature is set on the hot plate to

about 115˚C and 110˚C for positive and negative photoresists respectively

The important segments contained in the photoresist are a polymer (base resin)

a sensitizer and a casting solvent [97] The polymer will react by changing its structure

when exposed to the radiation While sensitizers will govern the reaction of the

photochemical in the polymeric phase the casting solvent will permit the spin-coated

application on the wafer surface The photoresist consists of positive and negative

photoresist both of which were used in this research Their basic difference is with

respect to the area that is exposed by UV light ie whether it will remain on the

semiconductor surface or will be removed In the case of a positive resist the exposed

area will be removed by the developer and the covered area will remain In other words

whatever is displayed on the photomask goes onto the sample surface On the other

hand for the negative photoresist the area that is covered from UV exposure will be

dissolved by the developer Figure 31 shows a 3D picture of both processes used in this

research

85

Figure 31 3D illustration of Optical lithography process used in this research

To obtain the desired pattern on the surface after UV exposure for a certain

duration the sample is required to go through a development procedure by using a

developer The developer is used to expel the dissolvable part of the photoresist after

being exposed to UV The usefulness of the developer depends on the photoresist ie

MF319 developer is specifically for positive photoresist while MIF326 is suitable for

negative photoresist Both types of developers will not harm the devices as they are a

kind of metal free ion solution Thus no free ion will change the characteristic of the

device The common practice in the clean room at the University of Manchester is to use

positive photoresist namely Shipley Microposit S1800 supplied by The Dow Chemical

and negative resist AZnLOF2020 (AZ2microm) which is supplied by MicroChem For the

S1800 series the thickness of the photoresist is determined by the last two digits ie

13microm thick for S1813 and 05microm thick for S1805 Both positive photoresists are used

Resist

GaAs

Dielectric

Photo Mask

Negative Resist Positive Resist

After Etching

86

mostly as protective area during wet chemical etching and as sacrificial dielectric layer if

higher temperature is applied ie 190˚C On the contrary AZ2microm which normally has a

thickness of 2microm is useful for patterning small dimension which leaves small gaparea

for the metallization process to fill It has good aspect ratio and useful in single layer lift-

off (post metallization process) In fact this type of resist can be thinned by diluting into

an Edge Bead Removal (EBR) solution and smaller device feature size can be obtained

The final dimension of certain devices (mesa size) is governed by the exposure

time the distance between photomask and samplersquos surface and the development time

The appropriate UV exposure time is required to avoid over-exposure which will cause

the spreading of light into the purportedly dark-field area[98] However the effect of the

exposure time differs between the photoresist types smaller opening area than the mask

pattern for negative resist and larger opening area for positive resist[45] The gap

between the photomask and wafer surface must be reduced as much as possible to avoid

the UV light going through the unwanted area To obtain a good gap value an applied

pressure from stage to the mask is required The normal pressure used in this research is

between 04 to 1 Pascal depending on the type of photoresist (negative and positive) as

well as its thickness Lastly the development time also influences the dimension of the

device Appropriate development time is required because ie over-development will

cause the polymerized photoresist to etch laterally resulting in bad patterned geometries

On the other hand under-development will cause non-uniformity in the surface after wet

chemical etching (positive resist) as well as causing lift-off problem (negative resist)

333 Etching Process

The etching process is a process of removing unwanted semiconductor layers to

define device geometry and isolate each individual device in one sample There are two

types of etching technique used in this research ie wet chemical etching and plasma

dry etching The wet chemical etching is based on Orthophosphoric(H3PO4) solution

which is a selective etchant to materials like GaAs InGaAs and AlAs The selective

etchant is referred to a solution that can etch away a specific semiconductor with a

specific etch rate The etch rate depends on the mixture ratio and concentration of the

solutions ie the higher the concentration the higher the etching rate In practice the

87

temperature humidity and epitaxial layer doping level also have an impact on the etch

rate Hence to minimize variation in the etch rate both temperature and humidity in the

clean room are constantly monitored and regulated Ambient temperature between 18degC

to 19degC is usually suitable while humidity is kept within 30 to 40 The advantage of

the wet chemical etching is that it is inexpensive controllable and with high throughput

highly selective and simple The mixture solution that is used in this work is

Orthophosphoric (H3PO4) Hydrogen Peroxide (H2O2) and Di-ionised water (H2O) with

ratios of 3150 3110 and 212 The ratio of 3150 provides an etch rate of about

600Aminute and is good to define the area and opening the under-cut in air bridges for

InGaAs samples[99] The 3110 ratio results with highly anisotropic shape but is easy to

control as the etch rate is about 1500Aminute for GaAs material system However the

212 etchant solution will provide extremely anisotropic etch rate of about

1000Asecond and is quite difficult to control For GaAsAlAs ASPAT diode it can still

be controlled since the thickness that needs to be removed is about 7000A (refer to

section 342 for details on the fabrication of ASPAT) Table 31 summarizes the etch

rate with different ratios and different selective materials Common to all the ratios

mentioned above are the isotropic side walls with lateral and vertical etch rate of 11

Table 31 Chemical wet etch characteristics using Orthophosphoric-based and Ammonia on GaAs

and InGaAs materials

Material Etchant Ratio Etch Rate (Aringminute)

GaAs H3PO4H2O2H2O 3150 600

GaAs H3PO4H2O2H2O 3110 1500

GaAs H3PO4H2O2H2O 212 60000

GaAs NH4OHH2O2H2O 118 2000~3500

InGaAs H3PO4H2O2H2O 3150 850

The plasma dry etch is purposely run to obtain extremely high anisotropic etch

profile vertically and horizontally It can be done via mask with positive photoresist and

self-align mask which is metal contact as a mesa protector The precursors that are used

to etch away GaAs InGaAs and AlAs layer in this technique are Methane and

Hydrogen (CH4+H2) On the other hand Carbon Tetrafluoromethane (CF4) and H2 is

88

used in removing Si3N4 The etch rate is determined by how much power is applied to

the plasma to hit the sample surface the pressure inside the chamber and amount of the

precursor In this research plasma Technology is used for dry etching This machine

which is a conventional OXFORD INSTRUMENTS 1990 machine can produce an etch

rate on average of 100Aminute for an RF power of 100mWatt

334 Sputtering (dielectric deposition)

In this work a Kurt JLesker PVD 75 is used to deposit Si3N4 layer on the sample

surface The deposition rate depends on the RF power that is applied To avoid surface

damage on the sample surface a sacrificial layer formed by SiO is deposited using the

Bio-Rad Thermal evaporator before transferring to the PVD 75 to start deposition with a

low (75Watt) RF power Once the deposition time reaches 30 minutes the power is

increased up to 200Watt As a result good uniformity of the dielectric layer is obtained

335 Metallization Process Lift-off and Annealing

The metallization process is a process in which metal contacts on semiconductor

devices are created The purpose of this process is to make a proper interconnection

between the semiconductor devices to other parts of the circuit elements In other words

it is to connect the semiconductor device to the outside world This process will allow

the device to be examined electrically so that all electrical characteristics can be obtained

ie resistance I-V curve capacitance conductance etc The metal scheme used in this

process depends on the semiconductor material In the case of GaAsAlAs ASPAT

diode Gold-Germanium (AuGe) Nickel (Ni) and Gold (Au) are used However in the

case of InGaAsAlAs ASPAT diode the metal scheme used is Titanium (Ti) and Au

The technique used in this process is resistive thermal evaporation

The metallization process starts by cleaning all the metallic sources and boats

ie tungsten boat Au Ni and Ti metals by using Trichloroethylene Acetone and IPA

consecutively for 5 minutes each in a high power ultrasonic bath This step is very

important to reduce the risk of contamination during thermal evaporation Once done all

the metallic materials are dried using a high-pressure nitrogen gun and then dipped in

89

Hydrochloric acid (HCL) solution for 2 minutes to de-oxidize the metals so that it has

minimal effect on the series resistance of the ASPAT diode

Two types of thermal evaporators were used extensively in this study both

Edwards Auto 306 (one denoted as Junior Auto 306) The latter is used to deposit alloy

type of metals while the former is used for the non-alloyed type of metals In the case of

GaAsAlAs ASPAT alloy type metals are used while for InGaAsAlAs non-alloyed

metals are used The cleaned metals are then loaded in the thermal evaporators and

placed on a resistive tungsten boat Each metal is placed on its specific tungsten boat to

avoid unnecessary mixture during the evaporation

Figure 32 Actual picture of thermal evaporator used in this study

Prior to loading the sample into the evaporator a standard fabrication process for

ASPAT diodes takes place by patterning the samples with AZ2microm negative photoresist

At the end of this step an opening area is created for the metal contact to be filled and

connected to the ohmic contact of the semiconductor The sample is also deoxidised

using a mixture of HCL and water in the ratio of 11 prior to evaporation This has to be

done in a very short time in order to avoid re-development of the native oxide layer

Inside the chamber the sample is securely placed on a chuck upside down facing the

filled tungsten boat The distance between the sample and metallic source boat is about

90

40 cm Figure 32 illustrates the actual thermal evaporation system used in this study

The thermal evaporator chamber is pumped down to reach a minimum pressure below

1times10-5

mbar before vaporizing the metal It is important that the mean free path between

metal amp sample is created and each vaporized metal stick firmly on the samplersquos surface

The normal practice in this study is to keep the vacuum pressure under 1 times10-6

bar so

that better device performance can be obtained The amount of current required to melt

down the metal is between 4 Amps to 6 Amp This amount of current is forced through

the tungsten boat and generates very high heat melting down the metallic source and

vaporizing it towards the sample surface The deposition rate for each metal can be

monitored by using a built-in film thickness monitoring (FTM) on the thermal

evaporator which proportionally depends on the amount of materials deposited The

GaAsAlAs sample is started with deposit of 55nm AuGe 13nm Ni and 500nm of Au

The reason of depositing AuGe first is due to the fact that the ohmic layer of the

ASPAT only can only be doped with a maximum doping of 4 times1018

cm-3

which is not

high enough for good conductivity Thus here Ge atoms will diffuse into the GaAs and

replaces Ga atoms during annealing process leading to higher doping levels (gt1 times1019

cm-3

) and hence improved conductivity

3351 Lift-off

The use of AZ2microm allows for the exact patterning of metals without the need for etching

using a single layer lift-off technique The negative photoresist also provides an undercut

profile which will create disjointedness between the desired metal pattern (on the

semiconductor) and undesired metal (on photoresist) The process of getting rid of the

unwanted metal from a sample surface is called lift-off The process starts once the

evaporation process is accomplished The sample is soaked into 80˚C N-Methyl-2-

Pyrrolidone (NMP) solution for usually 20 minutes (fast lift-off process) In most

instances the sample is in NMP for more than 12 hours in ambient temperature (slow

lift-off process) as the NMP solution is not destructive to the sample In this solution

the negative resist will be softened and the metal part which sticks on it will also be

eliminated from the sample As depicted in Figure 33 the lift-off process for a single

device shows the usual photoresist undercut profile observed To ensure that NMP

91

residues on the sample surface are completely removed DI water is used to rinse the

sample which is then blown dry with a nitrogen gun

Figure 33 Single layer lift-off process using negative photoresist

3352 Annealing

The alloyed (AuGeNiAu) metal stack requires a thermal treatment called annealing to

improve the ohmic contact between metal and semiconductor The sample which has

deposited top and bottom contacts is loaded into an annealing furnace at a temperature of

420˚C for 2 minutes In the case of GaAs during thermal annealing Ge atoms penetrates

into the GaAs crystal for approximately 70nm-250nm depending on evaporated

thickness of the metal layers annealing temperature as well as time [100] In this work

the total metal thickness evaporated for each contact layer is around 500nm This

thermal annealing treatment will also melt down the Au if it is too thick and is subjected

to too long a heat treatment Therefore it is not advisable to do annealing after the

sample is coated with bond pad metals as it can result in short circuited devices

sometimes

Az2micro Negative

photoresist profile

after UV exposure

and development

Evaporation to

form metal layer

(AuGe Ni Au)

Desired metal

contact

Metallisation

process

Lift-off undesired metal

92

34 GaAsAlAs ASPAT Process Optimization

As mentioned earlier this section present details two major process flows of the

fabrication process for the ASPAT diodes which utilised three stages of development of

photomasks design namely a ldquoFirst generation mask designrdquo (1st Gen) a ldquoSecond

Generation mask (2nd

Gen)rdquo and a ldquoThird Generation mask (3rd

Gen)rdquo design The 1st

and 2nd

generations mask designs are the designs that were produced in the first stage of

this work to develop the fabrication process know-how and to get familiarized with the

actual fabrication techniques in the cleanroom The difference between the 1st Gen and

2nd

Gen masks is in term of the development towards realizing Air Bridges and

Dielectric Bridges which were mostly covered by the 2nd

Gen mask design The

analyses in term of reproducibility repeatability and manufacturability for process

control and current-voltage characteristic as well as ultimately RF measurement are

obtained on relatively large size devices via these two mask designs The large area

emitter dimensions range in size from 15times15 microm2 to 100times100microm

2

The 3rd

Gen mask design was designed based on the optimization of the 2nd

Gen

mask which was to realize ASPAT diodes that are able to work at very high operating

frequencies Therefore in such design the ASPAT devices have to have a minimal

amount of capacitance and low series resistance this is can be achieved by shrinking the

emitter size of the diode to the smallest area possible as well as optimising the

fabrication process ton reduce parasitics The smallest fabricated devices designed on the

new mask has an 2times2 microm2 MESA area which also includes Ground Signal Ground bond

pads for both device and de-embedding structures (open and Short) for RF

measurements

Before the commencement of any fabrications and designing any layouts on eg

Si GaAs InP etc the epitaxial layer must be grown first to a desired design In the

University of Manchester the epitaxial layer structures are grown using one of the two

Molecular Beam Epitaxy (MBE) machines which are either the RIBER V90H or the

V100HU system Both systems are managed by Professor Missous Epitaxial wafers or

sample grown by each system are identified by a prefix and numbers that are prefix

VMBE for the V90H system and XMBE for the V100 system The epitaxial layers are

93

grown on four-inch wafer diameter The maximum diameter that can be grown on using

the V100H is 8 inches but generally single 4rdquo or 4x4rdquo wafers are used The wafers are

then diced and cut using a diamond scriber into 15mm times 15mm tiles for easy handling

and fabrication in the D12 cleanroom lab in the University of Manchester

In this section the structures fabrication and performance of the various ASPAT

diodes for both repeatability and high frequency applications will be discussed further

The ASPAT is manufactured on wafer sample XMBE304 which is a GaAsAlAsGaAs

lattice matched to a GaAs semi insulating substrate is the main focus The ground works

on this ASPAT such as the initial design and fabrication process flow optimization had

been conducted by fellow PhD colleagues in the group led by Professor Missous at the

University of Manchester

341 ASPAT Devices used in Fabrication

3411 XMBE368 and XMBE307

Table 32 Epitaxial layer of Doped substrate samples

Thickness (Aring)

Layer Material Doping (cm-3

) XMBE368 XMBE307

Top Ohmic1 GaAs (Si) 4times1018

~3000 ~3000

Buffer 1 GaAs(Si) 4times1017

~50 ~50

Spacer1 GaAs NID 50 50

Barrier AlAs NID 28 28

Spacer 2 GaAs NID 1000 1000

Buffer GaAs(Si) 4times1017

50 50

Bottom Ohmic GaAs(Si) 5times1018

~3500 ~3500

Substrate GaAs (Si) Doped 50000 50000

These two samples were the first batch of diode structures used in this work and

were grown using the RIBER V100 MBE machine A great deal of work was expanded

to ensure that it is able to produce appropriate non-linear I-V characteristics The work

carried out including finding suitable fabrication process steps mask designs process

control limitations ie etching rates etc The results obtained from processing these

94

samples mostly on large area anode and cathode sizes and their analysis included both

growth profiles and fabrication process flows These samples were grown on doped

GaAs substrates As such the finished diodes were vertical structures and the fabrication

process has marked differences compared to undoped substrate samples While

XMBE368 had similar epitaxial layer profile to XMBE307 during growth the AlAs

layer was set to be stagnant (ie no-rotation of the substrate during growth) to

investigate the effects of slight variations in barrier thickness

3412 XMBE304

The next sets of samples were all grown on semi-insulating GaAs substrates

Table 33 details the epitaxial layer profile of sample XMBE304 the main work horse

of this research work The growth of this structure was performed on a multi-wafer

platen and consisted of 9 x 2rdquo wafers (from XMBE304A to XMBE304I)

Table 33 The epitaxial layers of sample XMBE304 with barrier 10ML~28nm

Epitaxial layer Material Doping(cm-3

) Thickness(Aring)

Emitter GaAs(Si) 4e+18

3000

Emitter 2 GaAs(Si) 1e+17

400

Spacer GaAs NID 50

Barrier AlAs NID 28

Spacer 2 GaAs NID 2000

Collector GaAs(Si) 1e+17

400

Collector 2 GaAs(Si) 4e+18

4500

Substrate GaAs(SI) Semi-Insulating

For a typical ASPAT structure the emitter is essentially highly doped to

4times10+18

cm-3

to provide accumulation of electron in the emitter contact region It was

purposely highly doped to also achieve low ohmic contact with the metal The spacers

are used to avoid diffusion of dopants to the subsequent layers The ASPAT structure as

mentioned earlier has a single AlAs barrier with a very small thickness sandwiched

between two different length GaAs spacer layers

Batch XMBE304 is the main focus in these studies All activities required for

repeatability reproducibility process flow and devices as well as qualifying new or

95

optimization fabrication technique for small devices and RF performance which were

then used for high frequency were based on the set of wafers grow in this batch

342 Fabrication process of GaAsAlAs ASPAT diode toward repeatability

reproducibility and manufacturability

As the new wafer structures have to be tested and evaluated to gauge the

performance of the ASPAT diodes their uniformity also needs to be tested so as to

ensure it exhibited fully functional diode with zero bias detection in minimal variation in

IV characteristics between diodes As described previously 4rdquo wafers are always diced

up into 15mm times15 mm size for ease of handling in the cleanroom and for masks cost

purposes

3421 Doped n+ Substrate Wafers

For wafers grown on n+ substrates (XMBE368 and XMBE307) a two-level mask

set is sufficient to complete the fabrication process In this case the devices are designed

with mesa structures and top metal contact on the upper surface of the wafer and a

bottom contact on the backside of the wafer One mask plate is used for defining the

mesa structure (wet etch) and the other mask plate is used for defining the metal pad

Since the mesa is relatively large (30times30microm2

to 100times100microm2) there is no requirement to

define bond pads for measurement purposes A prior RTD mask designed by fellow

colleague Dr Md Adzhar for his PhD work was used for processing the ASPAT diodes

as well in the first instance The detail fabrication process is summarized in appendix I

for the doped substrate wafers Figure 34 shows typical IV characteristics for 30x30microm2

emitter size diodes obtained from sample XMBE368

96

The I-V measurement was taken from two different tiles (15mm times15 mm) located

on top and bottom of a 4rdquo XMBE368 wafer The location 1 marked as a blue line in

figure 34 refers to a device on a tile taken from the top of the 4rdquo wafer while location

2 (red) refers to a device on a tile taken from the bottom of 4rdquo wafer At 07V the

separation difference in current is ~228 for both tiles This shows that the device in

tile 1 is less resistive than the sample in tile 2 implying that the AlAs layer for tile

located on the top corner of the wafer is thinner and thus allows more electrons to tunnel

through it at a given bias

3422 First Generation Mask Design (1st Gen)

-0015

0005

0025

0045

0065

0085

-15 -1 -05 0 05 1 15

I C

urr

en

t (A

mp

)

V Voltage (Volt)

Current vs Voltage (XMBE368_1) for 30x30microm2

MidLeft2

MidRight3

Figure 34 Current-Voltage characteristic of sample XMBE368 used

in this study at two different locations on the wafer tile

(a) (b) (c)

Figure 35 ASPAT mask design for (a) 100times100microm2 (b) 30times30microm

2 and (c) 15times15 microm

2 diode dimensions

designed in the 1st Gen Mask

Location 2

Location 1

97

The fabrication of the GaAsAlAs ASPAT conducted in this 1st Gen mask is

followed the established generic process flow for InGaAs material which was done by

former co-worker Dr Md Adzhar Md Zawawi[101] The generic process flow is fairly

simple consisting of four mask steps as shown in Table 34 below The fabrication

process starts with the sample being cleaned in a NMP solution at a temperature of 80˚C

with Acetone and Propan-2-ol (IPA) The purpose of cleaning using NMP and Acetone

is to remove any organic material The IPA is used to remove Acetone residues

Table 34 Generic fabrication steps established by Dr Md Adzhar [101]

Step number Process

1 Top Contact

2 Mesa Etch

3 Isolation

4 Bottom Contact

In the 1st Gen mask the first step in fabricating the GaAsAlAs ASPAT diodes is

to use the first mask to define the emitter contact area The emitter has three different

sizes 100times100microm2 30times30microm

2 and 15times15microm

2 The lithography technique uses the

negative photo-resists AZ2micron with 55 second UV-photolithography to pattern the

top contact Then it is developed using MC319 developer to clear and define the

exposed area for the metals to stick to The sample is then subjected to plasma etching to

remove all organic residues and contaminants Then it is dipped into a mixture of

diluted Hydrochloric acid and water HCL H2O with a concentration of 11 for de-

oxidation This must be done in a short time right before the evaporation in order to

ensure good contact between metal and semiconductor surface with very low oxide

formation in between The ASPAT device performance depends on this step as contact

to the channel is by means of current flowing through the anode to the cathode terminals

In our lab the approach taken to achieve the Ohmic contact is by evaporation of Gold

Germanium (AuGe) Nickel (Ni) and Gold (Au) metals layers on top of the cap layer

Subsequently the metal is defined via a lift-off process using NMP

98

The next critical step is to define the MESA or island The mesa etch mask is

designed with two options with 05 microm or 10 microm tolerance The different mesa

tolerances are introduced to act as a safeguard for the emitter from producing excessive

undercut caused by lateral etching This step is to isolate and eliminate the unwanted

GaAs which will electrically link the active layers as many devices will be fabricated at

the same time on the same wafer tile (15mm times 15mm size) The lithography process in

this step uses positive resist and is developed using MC326 developer This step is

achieved using a wet etch process where a non-selective etchants mixture of

H3PO4H2O2H2O etches down the epitaxial layers until it slightly exceeds the AlAs

barrier with an etch rate of about 600Aring to 900Aring per minute The outcome of the MESA

or island step is the formation MESA active layers which are surrounded by inactive

layers of semi-insulating substrate (when using semi-insulating substrates)

The isolation mask is a step that is basically the same as the MESA etch step

eg using similar etchants mixture same lithography process but a different mask This

step is used to etch down until the GaAs substrate is reached which means that the

etching time is longer than that in the MESA etching step The purpose of this step is to

fully isolate the device from other neighboring devices hence ensuring no electrical

connection exists between each device within one sample Since these ASPAT diodes

employ an air bridge structure of size 1times5 microm the two minutes mesa etches will

simultaneously provide an initial undercut through lateral etching for the air bridge

formation

Based on the results so far obtained using the first-generation mask which

provided large mesa areas the current voltage characteristics of the ASPAT as far as the

non-linear zero bias is concerned did work very well However there was a need to

reduce the ASPAT mesa area down to very small dimensions to achieve mm-wave or

THz detection frequencies It is certainly a general rule for semiconductor device which

operate at very high frequency to have extremely small lateral dimension which

minimises the capacitance within the ASPAT device Furthermore for wider adoption of

the technology it is also important to develop a simple reproducible and low-cost

fabrication method for ASPAT diodes Details of this fabrication process are attached in

appendix II

99

3423 Second Generation (2nd

Gen) mask (ASPAT-GSG)

This work intended to enhance the 1st Gen-Large Area ASPAT photomask by

adding many features including 2 times 2microm2 mesa areas ground-signal-ground (GSG) bond

pads to enable RF measurement and three options device processing eg Air Bridge

Dielectric Bridge for semi-insulating doped substrate and dry etch-mesa Other reasons

for designing this mask was also to qualify process steps when deploying thin (1microm

width) bridge to connect small mesa area ASPAT diodes to the co-planar GSG bond pad

for DC and ultimately RF measurements

The mask was designed to fulfil the basic rules of fabricating various types of

tunnelling diode for instance RTD and ASPAT The diodes layouts were designed using

the commercial software Advanced Design Software (ADS) from Keysight

Technologies Ltd Once the design was completed it was ready to be sent to Compu-

Graphics Company for printing and patterned on a special chrome coated glass plate

The new 2nd

Gen mask design was termed ASPAT with Ground-Signal-Ground

(ASPAT-GSG) and consisted of two main designs ldquoAir Bridge and Dielectric Bridgerdquo

where each contained eight diodes with different emitter sizes (100times100μm2 50times50μm

2

30times30μm2 20times20μm

2 15times15μm

2 10times10μm

2 6times6μm

2 and 2times2μm

2) The Air Bridge

design is comprised of Design 1 (doped substrate) and Design 2 (undoped substrate)

This can be selected by changing the order of each individual layer of the mask steps

The same options are applied for Dielectric Bridge design which included processing for

doped and undoped substrates The details of the masks will be explained in the

following paragraph

(1) Air Bridge

Design 1 Contains five steps or layers mask of size 15mm times 15mm

suitable for air-bridge for undoped semi-insulating substrate

Design 2 Consist of seven layers steps mask of 15mm times 15mm size

There are 413 die chips in about 6mm times 6mm sizes in this design Figure 36 shows both

type of design for Air Bridge mask processing

(i) Mask 1- Top Contact

(ii) Mask 2- MESA

100

(iii) Mask 3-Isolation

(iv) Mask 4- Bottom Contact

(v) Mask 5- Collector Bond pad with GSG

(vi) Mask 2A-Dielectric

(vii) Mask 6A-Collector Bond Pad with GSG

(2) Dielectric Bridge

Design 1 Consists of seven identical steplayers masks with lateral

lengths of 15mm times 15mm each suitable for fabrication on semi-insulating

substrate

Design 2 Also has seven steps layers masked with dimension of 15mm

times15mm each suitable for doped substrate processing

There are 357 die chips in this type of mask design with an estimated size of 6mm times

6mm separately The smallest emitter size that is connected with a dielectric as the

bridge is 6times6microm2 The smallest 2times2microm

2 diodes was designed with the air-bridge

connected to the emitter bond pad due to the difficulties of opening viascavity for less

than 2 microm2 devices Figure 37 show both options for Dielectric Bridge mask processing

(i) Mask 1- Top Contact

(ii)Mask 2- MESA

(iii)Mask 3-Isolation

(iv) Mask 4- Bottom Contact

(v) Mask 5-GSG bond pad

(vi) Mask 6- Via Dielectric

(vii) Mask 7- Dielectric Bridge

(vii)Mask 5A- Via Dielectric

(ix) Mask 6A-GSG bond Pad

(x) Dielectric Bridge

Figure 36 Air Bridge design (red circle) for large mesa area (100times100microm2) device with

option for doped substrates

Figure 37 The layout of 1st design of Dielectric Bridge (green circle) mask design for 100 times

100microm2 emitter size with option for doped substrate processing

101

34231 Fabrication Process of the Air-Bridge Design

The fabrication using 1st Gen mask as mentioned in Section 3422 is less

complex however the fabrication process for Air-Bridge mask design contains a few

additional steps which are to add bond pads for the Ground-Signal-Ground radio

frequency (RF) layouts If the sample is on a doped substrate another layer needs to be

added leading to a total of six steps all together

Table 35 Standard process flow for Air-Bridge design fabrication

Mask Number Step number Process

1 1 Top Contact

22A 2 Mesa EtchDielectric

3 3 Isolation

4

5

4

5

Bottom Contact

Bond Pad

In this work the samples that have been processed so far are XMBE304

XMBE314 (GaAsAlAs base material system) and XMBE326 (InGaAsAlAs base

material system) All of them are built up on semi-insulating substrate thus the normal

process flow was followed Step 1 to step 4 follow exactly the same route as the 1st Gen-

ASPAT Large Area mask formerly discussed in Section 3422 This process is then

continued by spin coating AZ2microm on the surface Mask 5 is used to pattern the bond

pads The metallisation scheme used for bond pads is TiAu and the thickness must be at

least 1 micrometre thick This is to minimise series resistance at the pads and ensure a

robust surface is created when used for DC and RF measurements

342311 Air-Bridge Process Optimization

Since the Air Bridge design approach is focused more on developing air-bridges

which have a width of 1 microm and as the smallest device is 2 times2 microm2 the negative resist

(Az2microm) which was used in the 1st Gen design type was changed to Az1microm After spin

coating an Edge Bead Remover (EBR) is required to remove the beads at the edge of the

sample this is to ensure no gap is created between samples and mask when ldquohard-

102

contactrdquo is applied during the photolithography step The use of Az1microm and EBR are

critical to enable the fabrication of small air-bridges and emitter sizes

The exposure time during the photolithography technique also needs to be

increased to achieve a 1microm size air bridge A longer time than normal is required thus

95 seconds is used for exposure under UV light Once finished appropriate

development time must be applied to ensure the line for the air bridge is perfectly

opened for the metal to fill up To get impeccable result in this step both combination of

exposure and development have to be tuned naturally leading to trade off in both Too

long exposure and developing time will break the bridge whereas for too short exposure

time and developing the air bridge will not open

There are two types of etching method used in this fabrication wet

(H3PO4H2O2H2O) and dry (NH4H2) etching Both have their own advantages and

disadvantages Wet etching is faster than dry etching However the etching profile is

highly isotropic and causes serious undercut The smaller devices which have

dimensional sizes of 6times6microm2 and below will ldquoshrinkrdquo the effective area under the top

contact metal Even though dry etching can have highly anisotropic etching profile

which can prevent excessive undercuts it requires a lot of time for etching the

semiconductor layers The NH4 and H2 plasma that are used in this technique do not only

etch the GaAs and AlAs layers but also etch the metal contact at the same rate as the

semiconductors Thus the metal area must be covered with a photoresist or thicker metal

(~1microm) must be used to counter this issue Figure 38 shows that the emitter bond pad

which is not covered by the resist will eventually be etched away by the dry etching

Figure 38 Dry Etching for the first run in this study

As mentioned in section 333 the advantage of H3PO4H2O2H2O wet etching is

that it can be made using two solutions fast and slow The fast solution is based on a

concentration ratio of 212 while the slow solution is based on a 3150 mixture The

103

212 fast solution will produce good anisotropic etch profile but is tricky to control due

to the rather high etch rate of 1000Aring per second Both fast and slow wet etching

solutions being carried out in this experiment are replicated from successful recipes that

were developed by other co-worker from Prof Missousrsquos group

342312 Issue of Over Etch under Top Metal (Wet Etching for Air-Bridge

Design)

The first run using the solutions mentioned above started with the deposited

metal as a top contact (Top contact must be defined first to make a bridge) then Mesa

etching to remove ~7000Aring GaAs using the fast etch solution The next process to be

followed was to employ the slow etch solution to etch down the epitaxial layer to the

substrate Doing this ensured that air-bridge was open and the devices isolated

individually Unfortunately unexpected results were observed where large undercuts

still appeared for both 2times2 microm2 and 6times6microm

2 devices Figure 39 shows severe undercut

under the big device that can clearly be seen in the digital scope Most probably this

problem occurs due to excessive time used for the wet etching (10 seconds for fast

solution and 10 minutes for the slow etch solution)

Figure 39 Severe undercut of 2times2 microm2 and 6times6 microm

2 devices

The second run was employed to reduce the wet etching time in total thus dry

etching was needed As discussed above CH4 and H2 are used to etch away the GaAs

and AlAs semiconductor material while O2 is used to remove polymer residues To

reduce redundant generated polymer during dry etching the process must be separated

into several goes (runs) In this run at the mesa etch step the metal area is covered by

S1805 resist and then the plasma etches down until the heavily doped Ohmic layer is

reached For the isolation step the slow wet etching solution is used to ensure the

semiconductor under the bridge is removed The etching time is still long ie over 5

Zoomed Zoomed

104

minutes Hence this will still generate appreciable undercuts Although this run

managed to reduce the wet etching time it still did not solve the undercut Figure 310

shows undercuts under the effective mesa area still occurring in this run

Figure 310 Device with emitter area of 2times2 microm2 and 6times6 microm

2 after dry etch and wet etched

Further investigation has been made to find the root cause of this issue Scanning

Electron Microscopy (SEM) picture were taken to see deep into the completed devices to

investigate what is actually happening Figure 311 shows clearly that the semiconductor

under the bridge and effective area under the 2times2 microm2 device went missing The same

goes to for the large device where about half the size of the designed effective area under

to metal was also unintentionally removed by the solution

Figure 311 SEM Images of the GaAs sample

Since this work was also run together with an InGaAs based sample

(XMBE326) new knowledge regarding etching profile was acquired The lesson learnt

from this run is that the wet etch profile of the Gallium Arsenic (GaAs) material is

totally different from that of InGaAs Figure 312 shows the cross section of GaAs and

InGaAs materials after wet etching The evidences of these issues are also indicated by

the SEM images in Figure 313 and Figure 314 This more or less explained the reason

why the semiconductor under the top metal contact is always missing when wet

chemical etching is applied

Zoomed Zoomed

105

Figure 312 The cross-sectional view of InGaAs (left) and GaAs (right) samples used in this study

Figure 313 SEM images taken from the XMBE304 samples and a GaAs test sample

Figure 314 SEM images for InP and InGaAs taken from [56]

Furthermore previous work done by co-worker Dr Md Adzhar Md Zawawi has

optimised the process for RTD samples based on InGaAs and InAlAs heterojuction

semiconductor materials In this work Dr Zawawi found out that achieving submicron

dimensions is possible when using the soft reflow technique on InGaAs[102] However

it does not apply to GaAs when using the same technique There are few reasons to

106

believe that the work that was carried out by the previous student is not repeatable to the

XMBE304 GaAsAlAs material The main reason is due to the much thicker Ohmic

layers used in the GaAs of ASPAT structures which requires longer etch time The

sample that was used in the previous submicron work namely XMBE277 (RTD) [101]

and corresponding epitaxial layer structure is attached in appendix III The thickness

needed to be removed for the MESA step is 1421Aring compared to XMBE304 (ASPAT)

which is ~6900Aring

The fabrication results shown from the Air-Bridge design does not seem

favourable to the GaAsAlAs heterojunction sample These include unreproducible IV

characteristics and excessive undercut under the metal Therefore the GaAsAlAs

ASPAT sample cannot be processed using this type of mask (Air-Bridge approach) The

fabrication efforts were then moved to the Dielectric Bridge design described below

342313 DC measurement

Fully functional ASPAT I-V characteristics were not obtainable in this run due to

severe damages caused by the undercuts that happened underneath the top metal contact

As can be seen from Figure 315 the behaviour of the current response suggests a very

leaky diode This confirms that the air bridge approach does not work with the

GaAsAlAs ASPAT diode

Figure 315 Short circuit behaviour on one of the fabricated device in this run

-0015

-001

-0005

0

0005

001

0015

002

0025

003

0035

-3 -2 -1 0 1 2 3

Cu

rren

t (A

mp

)

Voltage (Volt)

30x30BP

107

34232 Fabrication Process of the Dielectric-Bridge Design

Since the Air-Bridge design was not successful an alternative Dielectric-bridge

process flow was then designed to solve this issue The first run in this Dielectric-bridge

process was performed to ensure that when the fabrication was completed it was able to

produce the correct and reproducible ASPAT current-voltage characteristics The

process flow in this run is to follow the initial design which identifies the steps according

to the mask number Step 1 to step 4 in the process are exactly the same as in the 1st Gen

mask design Step 5 which is the bond pad process is then continued with spin coating

the AZ2microm After exposure and development a TiAu metal scheme with a minimum

thickness of approximately 1 microm is deposited

Table 36 Standard fabrication process flow for Dielectric-Bridge design

Mask Number Step number Process

1 1 Top Contact

22A 2 Mesa Etch

3 3 Isolation

4

5

6

7

4

5

6

7

Bottom Contact

Bond Pad

Via Etch

Bridge

Step 6 is quite complicated compared to the other steps where the introduction of

S1805 resist is used as a dielectric layer to prevent short circuit between top and bottom

contacts This includes opening the smallest via (holes within the dielectric layer) of 2times2

microm2 and 6times6 microm

2 emitter sizes This step started by spin-coating the S1805 resist and

was then followed by baking it at 150˚C for 30 minutes The longer baking time is to

ensure it is hard enough to be deployed as a dielectric layer Once the S1805 had

hardened as dielectric layer the sample was then spin-coated again with S1813 (thicker

resist) on the dielectric layer to act as a mask for vias opening (to cover the dielectric

layer from via etching damages) After the S813 was developed O2 plasma etches was

108

applied to remove the S1805 that covered the top metal and bond pads The hole on the

metal was then exposed This required only two minutersquos plasma etches time

As soon as the holes were created within the dielectric layer a quick clean to

remove S1813 was performed using Acetone and IPA This has to be fast enough to

remove the S1813 without removing the S1805 dielectric layer Mask 7 then defined the

area for the metal that will fill up into the open vias This metal connected the top

contact metal of the device to the bond pad (GSG pad) The same metal scheme as

mentioned in Section 3422 was used as the bond pad

The fabrication using the dielectric bridge approach in the 2nd

Gen mask design

appeared to work successfully when initial I-V characteristics were taken and it also

showed that the ASPAT diode was fully functional However using hardened S1805 as

a dielectric layer is not a good practise for manufacturing device since there were left

over residue on the sample surface as depicted in Figure 316 This residue comes from

the non-uniformity of S1805 resit that was formed during heat treatment Therefore the

next run was to avoid using S1805 but replacing it with a standard dielectric layer based

on Silicon nitrite (Si3N4)

Figure 316 The surface of the sample after final processing

342321 DC measurements

Given that the process of qualifying mask steps for GaAsAlAs ASPAT diodes

looked promising using the dielectric bridge approaches initial I-V characteristics

measurements were carried out for XMBE304A Figure 317 shows that the I-V

characteristics are comparable to those of the other runs ie in the 1st Gen mask

Noticeably the IV characteristic (Figure 317) between each device size does not scale

109

on a single line This is attributed to the unintentional variation in the size of the emitter

during wet etching within each sample

Figure 317 IV characteristics of sample XMBE304 for emitter sizes of 1000 microm2 2500 microm2 900

microm2 400 microm2 225 microm2 100 microm2 and 36 microm2

The current densities shown in Figure 317 are calculated based on a reduced

effective area of the devices by a factor of 08 This assumption is made as actual area

(without metal) is un-measurable Theoretically devices with small areas will have

higher resistances compared to larger devices However the opposite occurred in these

devices This problem is still under investigation It could be due to the spreading

resistance producing different values according to device sizes (smaller devices have

longer D while bigger devices have shorter D)

34233 Dielectric-Bridge Process Optimization (Si3N4)

Due to poor surface roughness of the sample when using S1805 as dielectric

layer this run was employed to improve the surface quality by using Si3N4 In order to

get the actual size of the emitter the processing needed to start by defining the

semiconductor area using either top contact mask or mesa etch mask and not to deposit

the metal first Smaller effective area will be obtained if the top contact mask is used to

define the area compared to the MESA mask since there are 05 microm and 1 microm tolerances

designed in the MESA mask

-000001

0

000001

000002

000003

000004

000005

-2 -1 0 1 2

Cu

rren

t D

en

sity

(A

mp

microm

2 )

Voltage (Volt)

Current Density vs Voltage

density36

density100

density225

density400

density900

density2500

density10000

110

Table 37 New arrangement of the mask number and step in Second Run

Mask Number Step number Process

3 1 Isolation

22A 2 Mesa Etch

4 3 Bottom Contact

6

1

7

5

4

5

6

7

Via Etch

Top Contact

Bridge

Bond Pad

The fabrication was initiated by isolating each device using Mask 3 to define

them individually Then MESA mask (Mask 2) was used to cover the area from wet

etching optical measurement were used to obtain the actual size of the emitter without

metal Figure 318 shows the process results after H3P04H2O2H2O etching for a ratio of

3150

Figure 318 Optical images after MESA etching (a) 1 microm Tolerance (b) 05 microm Tolerance

Although the mask was designed with smaller size of 6times6microm2 using the slow etchant

solution the effective emitter size can still shrink to ~3times3 microm2 and ~2times2 microm

2 as can be

seen in Figure 318 The next step was to define the bottom contact using a metal scheme

that is suitable for GaAs ohmic layer ie AuGeNiAu After lift-off the samples were

then cleaned and spin-coated with AZ2microm Then the area was defined by using Mask 6

allowing the exposed area to be filled by Si3N4 which is the dielectric layer used in the

second run However a tricky issue happened here as the Si3N4 was difficult to lift-off

especially on the smallest emitter area in this mask In this run the lift-off process was

successfully done after three days Step five which uses Mask 1 was to define the top

111

contact The same metal scheme will fill up the tiny holes within the dielectric layer to

attach on the emitter semiconductor Figure 319 below shows an optical image of a

6times6microm2

device after the top contact lift-off process Once lifted-off annealing took place

to ensure Ge diffuse into the GaAs contact layer to lower the resistance

Figure 319 After lift-off processing

Step six uses Mask 7 to define the area of metal connection between active areas

to the bond pad In this step and step seven a different metal scheme from the top and

bottom contact was used to connect metal to metal Here a TiAu scheme was used The

final step (step 7) was to define the bond pad area In this context mask 5 is used The

bond pad requires thicker metal thickness to reduce the series resistance and to increase

the robustness of the metal surface when probed with needles during measurements

This second run of 2nd

Gen mask dielectric approach went well with better samplersquos

surface when using Si3N4 as the dielectric layer compared to the previous run using

S1805 Once the bond pad were defined for each device as done in previous run the

preliminary current-voltage characteristics of the Si3N4 run were obtained This is

described in the next section

342331 DC measurements

As mentioned in the previous section the Si3N4 run started with a defined emitter

area This allows actual ASPAT dimension to be measured thus the current density

versus voltage that are plotted in the following figures are based on actual measured

sizes

112

Figure 320 The current density of XMBE304 processed using Si3N4 as dielectric layer

Although the actual area of each device is obtainable from measurement the current

density of this sample does not scale well either Figure 320 show the current density is

not scalable from 03V to 1V This probably happened due to the same issue related to

the spreading resistance for each diode varying To investigate this issue new runs were

required with intention of reducing Rsprd and were expected to lead to more scalable

devices Therefore a study of reducing the spreading resistance (Rsprd ) due to

contribution of the large D-gap was carried out using the 2nd

Gen mask with the Si3N4

dielectric bridge approach The following section discusses in detail the steps used in

reducing the gap between the top and bottom contact for the GaAsAlAs ASPAT diode

34234 Dielectric-Bridge (Si3N4) Process Optimization by varying the D-Gap

The series resistance as discussed in detail in Section 26322 is generally due to

the total contribution of specific contact resistance in particular the diode size the sum

of all doped layer resistances that sandwich the main ASPAT layer and the spreading

resistance which comes from the lateral structure diode design[33] Thus in order to

acquire better performance at high frequency these contributors must be controlled One

parameter that can be controlled through this fabrication process is to reduce the

spreading resistance by controlling the separation between top and bottom contact ie

the D-gap as indicated in Figure 321

-000001

0

000001

000002

000003

000004

000005

-1 -05 0 05 1

Cu

rren

t D

ensi

ty

(Am

pm

2)

Voltage (Volt)

Current Density vs Voltage

D36

D100

D225

D400

D900

D2500

D10000

113

Figure 321 side view of lateral ASPAT structure

Without designing a new mask at this stage the same 2nd

Gen Dielectric-Bridge

mask was used but the process flows did not follow the sequence order of mask

numbers The reason for deciding not to design a new mask to reduce the D-Gap was

because the fabrication process flow had not yet confirmed it repeatability and

reproducibility Thus the available photo mask at that moment was fully utilised

Furthermore the feature of the 2nd

Gen mask that included the mesa tolerance can be

exploited in this study Hence the initial idea was to reduce the length of D by using the

tolerance that was designed in with Mask 2 (MESA etch mask)

Table 38 New arrangement for the Third run using Dielectric-Bridge mask

Mask Number Step number Process

1 1 Mesa Etch

3 2 Isolation

3

2

3

4

Bottom Contact

Mesa Cover

6

1

7

5

5

6

7

8

Via Etch

Top Contact

Bridge

Bond Pad

The process therefore was initiated by cleaning the sample and was then

followed by spin-coating it with S1805 to cover the emitter contact and define the actual

D-gap

114

size of the diode This first step used Mask 1 with wet chemical etching using the slow

solution (H3P04H2O2H2O etch with a ratio of 3150) Once the resist was striped

optical measurements were performed to obtain the effective area of the emitter Figure

322 below shows the smallest emitter area obtained after etching

Figure 322 The measured size of the emitter area and the length D (blue color marked)

Step two was to isolate the devices individually by using Mask 3 This was

performed using the fast etch solution (H3P04H2O2H2O etch with ratio of 212) This

took about 10 seconds to remove about 10000Aring of material

Step three which involved two masks was the most complicated in this process

Firstly Mask 3 was used to define the bottom contact by covering the sample with

AZ2microm and then the sample was hard- baked at 190 ˚C for 4- 5 minutes to ensure it had

totally dried before applying Polydimethylglutarimide (PMGI) The PMGI type used in

this process is Lift-off Resist (LOR) SF11 After spin-coating the SF11 at 7000RPM for

45 seconds it was post-baked at 190 ˚C for 4 minutes This was done to ensure that the

LOR SF11 was hard enough for the S1805 to stick on it This was followed by spin-

coating S1805 at 4000RPM for 30 seconds and exposing it under i-line UV for 20

seconds and developed using Micro Dev mix with DI water (11 ratio) for one minute

When the correct shape of S1805 was formed the sample was exposed under flood UV

for 15 minutes The SF11 was developed with 101A developer for 1 minute Figure 323

below summarizes the whole process in step three The step three processes concluded

by depositing the bottom contact with alloyed metal scheme and once lift-off had taken

place the next step continues as usual

115

Figure 323 Summary of LOR technique steps

Step four as well as the subsequent steps were completed in this run by copying

from the previous run (ie second run) which was to deposit the Si3N4 as the dielectric

layer and so on All the devices that underwent fabrication using this technique were

measured under optical microscope MC60 assisted by the Walsall software tool in the

lab From the measured value obtained it was expected that reducing the length of D in

this technique would improve the spreading resistance by up to 70 from the original

2nd

Gen mask design Table 39 summarizes the calculation of original spreading

resistance and improvement using this technique The calculation method and equation

are taken from Section 26322

Table 39 The outcome of the spreading resistance before and after using LOR technique

Device Size(microm2) D-original(microm) RSprd(Ω) D-new(microm) RSprd(Ω)

100x100 4 056 163 023

50x50 14 327 194 054

30x30 19 602 163 097

20x20 21 832 212 135

15x15 22 1007 202 18

10x10 24 1293 205 252

6x6 25 1643 230 418

116

Technically from the experiment in this run it was observed that the development time

also controlled the final length of D longer development time produced shorter D

lengths while shorter time yielded larger D gap lengths Therefore in this run the

development time was kept constant because of the desired D length from the mesa

tolerance in general is only ~2microm

342341 DC measurement

As usual when wafer processing was completed early DC measurement took

place to check the diodes performance Each diode size was measured and compiled

into one graph as shown in Figure 324 It is clear that the performance was better than

that in the previous run in term of scalability and current conductivity

Figure 324 IV characteristic in 3rd run (reduced d-Gap) for each diode size on the mask

To compare the current conductivity from this run and the previous run current

density at 05V was measured for both 2nd

and 3rd

runs Assuming the contact resistances

(TLM) for both runs were constant for 50times50microm2 device size the improvement of

current in the 3rd

run is about 92 This showed that such approach to reduce the D-gap

-500E-06

000E+00

500E-06

100E-05

150E-05

200E-05

-2 -15 -1 -05 0 05 1 15 2Cu

rre

nt

De

nsi

ty (

Amicro

m2)

Voltage (Volt)

Den_100microm2

Den_225microm2

Den_400microm2

Den_900microm2

Den_2500microm2

Den_10000microm2

Den2_36microm2

117

was successful in this run A new mask design was ready to take the challenge for

processing toward millimetre and sub-millimetre wave application

343 Fabrication process of GaAsAlAs ASPAT diode toward High frequency

Applications

So far the information that can be gathered from previous processing is that the

optimum process flow is achieved through dielectric approach design The effort in

reducing the series resistance by lowering its biggest contribution was attained through

lowering the D-Gap in the structure Once everything had been optimized ie the

process flow series resistance junction capacitance etc it was time to develop a new

mask design which only focused on the development of small ASPAT devices for use in

the millimetre and sub-millimetre wave regions

3431 Third Generation (3rd

Gen) mask design

The 3rd

Gen mask design was developed by taking into the account every aspect

of parameters that can contribute to the device by means of robust devices that are able

to function properly at ultra-high frequency The device cut-off frequency is given by

119891119888119906119905minus119900119891119891 =

1

2120587119877119904119862119895

(31)

Here RS is the series resistance and Cj is the junction capacitance To obtain high cut-off

frequencies Rs and Cj must be kept as low as possible From the fabrication point of

view two parameters that can be directly and easily controlled are the device area (A)

and the D-gap (which contributes to Rs)

Therefore calculations were made to find the best option Table 310 shows the

calculated value of capacitance (eq230 in page 44) and cut-off frequencies (eq31

above) for the ASPAT diodes studied These two equations extract the cut-off frequency

of the ASPAT assuming no external effects and for fully depleted devices The

XMBE304 ASPAT sample is expected to be suitable for millimetre wave applications

if small devices are successfully made In real devices sub-millimetrewave operation

can be hard to achieve due to increased series resistance and other process related

parasitics

118

Table 310 DC and RF characteristics for XMBE304

Device Size (microm2) Fully depleted

Capacitance(fF)

Calculated Series

Resistance (Ω)

Fully depleted Cut-Off

Frequency(GHz)

10000 5490 04 72

900 490 15 216

36 198 7 1148

16 879 11 1646

4 22 29 2500

Therefore in the 3rd

Gen mask design the smallest device that can possibly be obtained

in the GaAs based material fabricated using i-line lithography which is available at the

University of Manchester is 2times2microm2 and the gap between top and bottom is 15microm at

least The connecting bridge technique applied only utilised dielectric bridge method for

GSG features Figure 325 shows the layout of actual 3rd

Gen mask used in this study

There are 344 die chips on this type of mask design It also includes the Ground-signal

ground pad with 50um pitch for each chip and six de-embedding test structures as well

as eight TLM structures

Figure 325 3rd

Gen Mask layout showing actual devices de-embedding TLM and alignment mark

structures used in this study

119

This mask is designed generally from optimizations from 2nd

Gen Mask which

deploys a dielectric bridge for connection between the devices to the bond pads

Consequently the processing steps are not being much different but mostly follow what

is shown in Table 311 below The difference only applies to the much smaller mesa

size Other features included in this mask are de-embedding structures for RF

measurements via-hole test structure TLM structures and parallel plate capacitance test

structures

Table 311 3rd

Gen Mask process step

Mask Number Step number Process

1 1 Mesa Etch

2 2 Isolation

3

4

3

4

Bottom Contact

Mesa Cover

5

6

7

5

6

7

Via Etch

Top Contact

Bond Pad

The high frequency fabrication process flow is summarized in the following section

which shows illustrations in three dimensional and cross sectional (Figure 326) views

for easy understanding

34311 Step by Step Processing

0 Wafer preparation

Cleaning using NMP and DI

water or

Acetone and IPA

120

1 Mesa Etch

2 Isolation

3 Bottom Contact

4 Mesa Cover (Dielectric Deposition)

5 Via Etch

Etching using etchant solution

H3P04H2O2H2O etch with ratio

of 3110

Etching using etchant solution

H3P04H2O2H2O etch with ratio

of 212

Metal deposition with metal

scheme AuGeNiAu for

~500nm thick

Dielectric deposition 500nm

thick Si3N4

Via etch to open holes for metal

contact using reactive ion

etching using CF4

121

6 Top Contact

7 Bridge and Bond Pad

Figure 326 Step by step device fabrication (in 3D and cross-sectional view) for semi-insulating

substrate device type used in this study

Figure 327 Example finished process device with bond pad using 3rd

Gen mask

3432 TLM measurements

Transmission Line model (TLM) measurements for this run were carried out

after annealing (ie at 420˚C for 2 minutes) for process control monitoring by extracting

contact resistance (Rc) values These values are a measure of the quality of ohmic metal

contacts for a given process As discussed in Chapter 2 the TLM technique used four-

Metal deposition with metal

scheme AuGeNiAu for ~500nm

thickness and thermal annealing

420˚C for 2 minutes

Metal deposition with metal

scheme TiAu (1microm thick)

122

point measurement on TLM test structures located around the 15times15mm2 tiles as shown

in Figure 222 (in page 55) Normally five TLM structures are measured across the tile

for both top and bottom contacts Figures 328 and 329 display graphical TLM results

for top and bottom contact respectively As can be seen both graphs exhibit excellent

uniformity

Figure 328 XMBE304 TLM measurement for the top contact after annealing

Figure 329 XMBE304 TLM measurement for the bottom contact after annealing

Based on the graphs above for the top metal contact the average contact

resistance (RC) value using the metal scheme of AuGeNiAu is found to be ~005Ωmicrom

and the sheet resistance (RSH) is 22Ω However for the bottom contact the average

value for contact resistance is 012Ωmicrom and sheet resistance obtained is 26Ω Both

values are in a very good agreement with the known doping of both ohmic contact

0

2

4

6

8

10

0 10 20 30 40 50

Res

ista

nce

(O

hm

s)

Spacing (microm)

After 420C for 2mins top contact

y = 02592x + 02375

0

2

4

6

8

10

12

14

0 10 20 30 40 50

Res

ista

nce

(O

hm

s)

Spacing (microm)

After 420C for 2mins bottom contact

y = 02157x + 01053

123

layers Therefore the specific contact resistance (Eq 250) that contributes to the total

series resistance can be calculated and the value obtained is 15Ωmicrom2 and 54Ωmicrom

2 for

top and bottom TLM structures respectively

3433 DC characteristic measurements

Again once the wafer processing is completed room temperature DC

characteristics are taken using an HP 414B or HP500B parameter analyser and its actual

setup is as described in Chapter 2 This initial measurement was to ensure the

functionality of the diodes For this run using the 3rd

Gen mask the I-V measurements

for mesa active area of 4times4microm2 6times6 microm

2 and 10times10 microm

2 were taken and are depicted in

Figure 330 Figure 331 and Figure 332 respectively Nine diodes were measured for

each three device sizes to check their uniformity Thus the average current and standard

deviation are taken at two voltage steps of 1V and 15V Table 312 summarizes the

standard deviation for each measured data obtained from this final run

Figure 330 XMBE304 GaAsAlAs ASPAT measured IV characteristics at room temperature for

4times4microm2 mesa size

-00005

0

00005

0001

00015

0002

00025

0003

00035

-3 -2 -1 0 1 2 3

Cu

rren

t (A

)

Voltage (V)

C01_4um_1P_SI

C22_4um_1P_SI

C40_4um_1P_SI

W01_4um_1P_SI

W22_4um_1P_SI

W40_4um_1P_SI

AI01_4um_1P_SI

AI22_4um_1P_SI

AI40_4um_1P_SI

124

Figure 331 XMBE304 GaAsAlAs ASPAT measured IV characteristics at room temperature for

6times6microm2 mesa size

Figure 332 XMBE304 GaAsAlAs ASPAT measured IV characteristics at room temperature

for 10times10microm2 mesa size

Table 312 Standard deviation at two different voltages

Device size (um2) Standard Deviation 1V () Standard Deviation 15V ()

4times4 14 14

6times6 65 72

10times10 44 39

Noticeably the standard deviation of the device increases for the smaller size

devices This trend happened probably because the active mesa area is not uniform

causing different series resistances Since the smallest mesa active area which achieved

-0001

0

0001

0002

0003

0004

0005

0006

0007

0008

-3 -2 -1 0 1 2 3

Cu

rren

t (A

)

Voltage (V)

C02_6um_1P_SI

C23_6um_1P_SI

C41_6um_1P_SI

W02_6um_1P_SI

W23_6um_1P_SI

W41_6um_1P_SI

AI02_6um_1P_SI

AI23_6um_1P_SI

AI41_6um_1P_SI

-0005

0

0005

001

0015

002

0025

-3 -2 -1 0 1 2 3

Cu

rren

t (A

)

Voltage (V)

C03_10um_1P_SI

C24_10um_1P_SI

C42_10um_1P_SI

W03_10um_1P_SI

W24_10um_1P_SI

W42_10um_1P_SI

AI03_10um_1P_SI

AI24_10um_1P_SI

AI42_10um_1P_SI

125

a good I-V characteristic is 4times4microm2

in this run these devices were used for the next step

of characterisation which is the S-parameter measurement to extract their behaviour at

different frequencies

35 Conclusions

In this chapter basic fabrication techniques of GaAsAlAs ASPAT on both doped

and semi insulating substrates using standard I-line lithography as well as step by step

descriptions to achieve reproducibility in the process fabrication flow has been

demonstrated with relevant initial measured results Two major outcomes have been

demonstrated firstly related to repeatability reproducibility and manufacturability

mostly done on large device areas (15times15microm2 to 100times100 microm

2) Secondly a successful

process flow for small emitter size devices (2times2 microm2 to 10times10 microm

2) has been

developed

Subsequently two types of designs during process optimization were developed

namely Air-Bridge and Dielectric-Bridge approaches The latter approach seems

favourable for GaAsAlAs materials but was only successful when reducing the series

resistance of the device This problem was addressed by optimising the D-gap between

top and bottom contact which resulted in good scalability of each ASPAT dimensions

improving current conductivity by 92 and achieving a reproducible process On the

other hand in the former approach issues were encountered with over etching underneath

the diode effective area and thus it was hard to achieve reproducible devices

Repeatability reproducibility and manufacturability fabrication processes based on

dielectric bridge method were successfully developed This new process provides a

highly efficient and economical solution for the fabrication of GaAsAlAs ASPAT

diode Emitter sizes down to 4times4 microm2 dimensions are routinely and reproducibly

achieved in this process Series resistance which is an important parameter in

determining high frequency application are greatly reduced by changing the gap between

top and bottom contact

With all the new improvements implemented the possibility of the proposed Dielectric

Bridge method fabrication process was successfully applied for the fabrication of

ASPAT diodes

126

4 PHYSICAL MODELLING OF THE GAASALAS ASPAT

DIODE USING SILVACO

41 Introduction

Fabricating any device or circuit requires a lot of time resources cost etc

especially elements related to the production of semiconductor devices Furthermore in

realising such a device a clean room is required Hence fabrication and processing in

wafer fab need full attention to get it done with a higher rate of success reproducibility

and manufacturability One solution that can be highlighted which will be able to reduce

all the resources that are mentioned above is by using computer simulation approaches

or to be precise physical modelling For epitaxial layer based devices physical

modelling is a good choice as it would give a better understanding and insight into each

layer and how electrical characteristics are derived A software that is most suitable for

epitaxial layer physically modelled is SILVACO This software is a very comprehensive

tool to simulate epitaxial based devices and to predict their behaviour Such software

covers many aspects starting from the first principles of physic epitaxial layer

definition as well as device layout thus making it the most powerful virtual wafer fab

tool in the market

In this chapter the SILVACO packages are discussed The discussions include the

method of defining a new material defining models and constructing the AC or DC

supply to obtain the output characteristics of the virtual device Apart from these the

focus will also be on the ASPAT diode modelling simulation and analyses of the results

which will include the ASPAT structure suitable models and DC current-voltage

characteristics The dependencies of individual structure on the I-V curves will also be

highlighted Finally the discussion of results and analyses involving a range of

operational temperatures dependency as well as from comparison made to conventional

SBD used in this study will be examined

127

42 SILVACO modelling Tools

SILVACO is a modelling software introduced in 1984 by Dr Ivan Pesic It is

purposely created for electronicsrsquo devices physical modelling and characterization This

software company has become the major supplier for most of the Electronic Design

Automation (EDA) for circuit simulation amp design of analogue Mixed-Signal and RF

circuit market This Technology Computer Aided Design (TCAD) software can predict

the simulated device performances starting from first principles It has a package which

can provide Virtual Wafer Fabrication (VWF) simulation to the device designer and

which has the capability to perform two or three dimensional physical device modelling

by using the ATLAS simulator [103] SILVACO allows all parameters such as

electrical thermal and optical characteristics of a device to be simulated under desired

bias conditions It offers cost effectiveness as well as quick prediction of results for

many semiconductor devices compared to real experiments Some hands-on experiment

may not always perform hence SILVACO can be used as an alternative

The core of SILVACO is Atlas itself which provides a platform to perform DC AC

and transient analysis for such dimensional device structure regardless of the

heterojunction material type ie binary ternary quaternary etc As the brain of

SILVACO Atlas which receives input command files containing instruction text for

execution from a runtime environment known as Decbuild will process the instruction

text and display progression error and warning via Runtime Output All the calculations

of the resultsoutcomes of the simulation are plotted via a tool called TonyPlot which is a

tool to visualise the output Figure 41 below shows how precisely the physical

modelling takes place the process of building the structure how its parameters and

variables are defined how an appropriate model statement is selected how performance

is analysed and lastly how the outcomes are displayed

128

Figure 41 SILVACO Atlas simulation process flow

As can be seen in Figure 41 the structure specification statement is used to define

any desired structure by setting the command in Deckbuid of the following parameters

a) Mesh where the structure can be defined either in 2D or 3D Cartesian grids The

unit of the coordinates used is in microns and the spacing parameters which

define the netting size can be used to improve the accuracy of the analysis at any

given position The density of netting size in this statement determines the

processing time

b) Region where the multi-layers in a structure are defined and this statement has

to outline each layer that represents a separated region independently The mesh

must be assigned to a region and the sequence of the region is arranged from low

to high

-Mesh -Region -Electrode

-Contact -Material

Structure Specification

-Model -Interface

Model Specification

Method

Numerical Method

-Log -Solve -Load -Save

Solution Specification

-Tony Plot

Display Results

129

c) Electrode are used to define the location of bias point for a designed structure

when performing the electrical analysis In this work or for the case of a diode

two electrodes are allocated as an anode and a cathode In a vertical device the

latter electrode is placed at the bottom of the device while the anode is at the top

d) Doping this statement refers to doping concentration injected into the desired

region and it normally depends on the material types

e) Materials since SILVACO was developed specifically for Silicon-based devices

default parameters are set up for Silicon properties However the use of

materials statement allows SILVACO to run for different material ie GaAs

InGaAs etc In order to make it work this parameter is defined first and then

followed by the material name and its properties such as bandgap permittivity

conductionamp valence band discontinuities mobility etc

The most crucial part is to determine whether the simulated structure of a particular

device is correct or incorrect This is done by properly choosing the specific model

statement The Specific model statement is employed to express the physic equations

that are used during the device analysis The models statement depends solely on the

structure definition Examples are device structure with double barrier use Non-

equilibrium Green function (NEGF) model and single barrier uses Semiconductor-

insulator-Semiconductor (SIS) model for effective and accurate analysis process For the

case of a single barrier in SILVACO there are many model statements that can be used

for such analysis Therefore it is recommended to check model by model in order to

ensure that all the needed parameters are defined in the material statements and results

produced are valid

SILVACO is able to calculate such models by using different numerical methods

which means semiconductor device problematic is computed to make successive

solutions by random discretisation There are three different numerical methods that are

regularly used by SILVACO to perform its calculations which are Newton Gummel and

Block Basically this is solved by using a non-linear iteration procedure which begins

from an initial guess and which then uses an iterative process to find the predicted

solution The detail of these can be found in reference [103]

130

In order to turn on the problem solution the solution specification is defined This

include log solve save and load statement All these work together to provide data for

analysis by other functions The log statement is a file type that saves in memory and can

be loaded by Atlas Any solved device will be stored in the log file Therefore it is

necessary to define the LOG before SOLVE statement and close it after the calculations

While the save statement is used to store all data point to a node in the output file the

load statement is utilised to recall all the saved data to be read by Atlas

Finally all these files can be displayed or plotted on TonyPlot it is recommended to

make SET files at plotting point for a better visualisation The plotted or displayed files

in TonyPlot can be manipulated for scaling graphs overlaying different curves and most

importantly to export the data to other files formats

43 SILVACO Implementation GaAs AlAs ASPAT Modelling

The structures play an important role in determining the terminal output

characteristics Therefore to start simulating the ASPAT device the first thing that must

be specified is its structure As mentioned earlier in Section 3412 the ASPAT diode is

a top down multilayer structure This structure which is adopted from that of Section

3412 will be used as a basis to perform the ASPAT simulation Figure 42 shows the

structure of the ASPAT in this real simulation which is exactly the same as been

discussed in Section 26322 The ASPAT diode consists of two heavily doped (up to

5x1018119888119898minus3) GaAs contact layers on top and bottom slices adjacent to lightly doped (up

to 3x1017119888119898minus3) GaAs intermediate layers In between these layers is a sandwiched

structure consisting of two different lengths of undoped GaAs spacer layers and a thin

layer of AlAs that act as a tunnel barrier In this work the simulation result will be

compared with the fabricated measurement result depending on the size of the diode

The actual device that will be used to validate the simulation is the main device used in

this study (XMBE304) which is based on a lattice matched GaAsAlAs grown on a

semi-insulating substrate Therefore the design structures that are proposed in the

fabrication are compatible with the fabricated devices and are based on lateral structures

as can be seen in Figure 42

131

Figure 42 The GaAsAlAs ASPAT structures produced from simulation based on the diode

multilayer heterostructures on the right

In the consequent simulations a key observation regarding the AlAsGaAs

heterojunction is that there are two types of tunnelling processes direct tunnelling and

indirect tunnelling Figure 43 shows the Energy-Momentum (Ek) diagram depicting the

three valleys for the AlAs conduction band namely L Γ and X points In normal

circumstances the transition of electrons happens to the X-point which is the lowest

energy in the conduction band In the case of very thin barrier the tunnelling process

occurs at the Γ point in both AlAs and GaAs materials which is the direct tunnelling

process[18] It has been reported that in the case of an ASPAT diode tunnelling which

occurs at the Γ point will be the dominant component in the tunnelling current

Therefore the actual band gap will be different from the one at the X-point which is

which 216eV[15] By contrast the energy band gap at the Γ-point is around

283eV[104] Thus this simulation uses this band gap value

Figure 43 The Energy-momentum diagram for AlAs heterojunction semiconductor

216eV

Γ

X

E

K

L

AlAs

CB

VB

283e

132

The simulation code as attached in Appendix IV and the output of the simulation with all

the input mentioned above are shown in Figure 44 (a) and Figure 44 (b) Figure 44 (a)

is the band diagram at equilibrium and Figure 44 (b) is the band diagram when a bias is

applied

Figure 44 (a) The conduction band and the valence band of the ASPAT diode structure (b) the

energy band diagram of the ASPAT diode structure when under three different biases

44 Simulation Result and Analysis

Basically the SILVACOrsquos Atlas simulation package is used to calculate the I-V

characteristics from multilayers structures In the case of GaAsAlAs heterojunctions all

details of the structure as shown in Figure 44 above are calculated based on solving the

Schrodinger time-independent equation in each layer taking into account the variation of

effective mass and conduction band offset between GaAs and AlAs

For thermionic emission and tunnelling mechanism across an abrupt heterojunction

interface the general method used in SILVACO is taken from K Y Yang work [105]

The tunnelling current of the ASPAT diode uses the equation below[74]

119869 = sum2119898lowast119864119894(1 minus 119877)119896119879

1205872ħ3

119873

119894=1ln [

1 + exp (119864119865 minus 119864119894

119896119879)

1 + exp (119864119865 minus 119881 minus 119864119894

119896119879)]

(41)

Where m is the electron effective mass E denotes the energy of the electron R is the

total resistance k represent Boltzmann constant ħ is the reduced Planck constant EF is

the fermi level V is applied voltage and Ei is the electron energy perpendicular to the

-2

-15

-1

-05

0

05

1

15

1

82

16

3

24

4

32

5

40

6

48

7

56

8

64

9

73

0

81

1

89

2

97

3

Ene

rgy(

eV

)

Thickness (microm)

VB

CB

133

barrier The important parameters that enable SILVACO Atlas to perform correct

calculations and analysis are the choice of appropriate models The suitable model that is

available for evaluating the GaAsAlAs ASPAT is based on the Semiconductor-

insulator-semiconductor (SIS) model

Thus in this run Non-local Quantum Barrier Tunnelling Model (SISEL and

SISHO) are utilized specifically semiconductor-insulator-semiconductor mode This

model enables the tunnelling current between two semiconducting regions separated by a

quantum barrier to be calculated [103] It is assumed that the charge tunnels across the

whole barrier with the source or sinks at the interface with the semiconductor regions

Under the Non-Local Quantum Barrier tunnelling model another model that can be used

is semiconductor-semiconductor-semiconductor (SS) tunnelling model if the materials

are specified By correctly inserting all parameters with the right model an excellent DC

IV characteristic match between simulation and measurement can be produced as shown

in Figure 45

441 DC Current-Voltage Characteristic

In this simulation the current at each bias step and each mesh point can be set up by the

user however the detailed calculation such as formula usage methodology and

approach that is adopted by SILVACO Atlas is unknown As mentioned above in order

to produce the energy band diagram of the ASPAT the DC characteristic of the structure

can be solved by using the Schroumldinger and Poisson equation self-consistently

To ensure that these simulations are valid one fabrication was performed on an

ASPAT diode sample XMBE304 For this sample the structure parameters are the

same as has been set in this simulation The result of the measurement and simulation

are then compared Figure 45 shows that the simulation result is in excellent agreement

with the measured data for this sample

134

Figure 45 The current-voltage characteristic of an ASPAT diode (100times100microm2) and (4times4microm

2)

using SILVACO Atlas simulator for structure device XMBE304 showing excellent agreement

between simulated and experimental data

These fitted results were performed on both a large 100times100microm2 device and the

smallest obtainable from fabrication (4times4microm2) which was to be used for the repeatability

amp reproducibility studies as well as for high-frequency applications study respectively at

room temperature In order to get a good fit a few parameters had been modified in the

SILVACO software via the Deck-built tool for example energy band effective masses

and bandgap discontinuity of GaAs spacer and AlAs barrier (The mentioned parameters

values are summarized in Table 41) The percentages of bandgap discontinuity in

SILVACO using the ALIGN parameter is given by[103]

119860119897119894119892119899 =

Δ119864119862

Δ119864119862 + Δ119864119881

(42)

Where ΔEc and ΔEv are the conduction band discontinuity and valence band

discontinuity respectively The m0 in the tables denotes the electron rest mass Once all

agreement between measurement and simulation has been met the simulation is then

carried out with structure analysis at room temperature and different temperatures

simulation

Table 41 The parameter values used in this simulation

Material Bandgap(eV) ΔEg(eV) Effective mass(kg)

GaAs 1424 03 0067m0

AlAs 2835 071 0126m0

-00002

0

00002

00004

00006

00008

0001

00012

-2 -1 0 1 2

Cu

rre

nt

I (A

mp

)

Voltage V (Volt)

Current vs Voltage

Measurment

Simulation

135

45 Structure Analysis of ASPAT Diode

Once the device structure was modelled and having successfully produced a

precise band diagram as well as validated the simulation results with experimental I-V

characteristics the next step is to further analyse the relationship between basic device

structure and its I-V characteristics This approach is used to predict what would happen

to the DC output if some of the parameters were varied especially with regards to the

AlAs barrier thickness In the subsequence simulations the thickness of each main

ASPAT (unequal spacers and barrier) layer will be studied independently as a variable in

order to determine how each parameter affects the I-V characteristic in both magnitude

and curvature The analysis will also include manufacturing tolerance where the

structure parameter which will result in a 10 difference in their I-V characteristics is

examined[59 75] This will provide an overview of how precisely to manufacture each

layer of the device The following simulation is based on the XMBE304 structure with

emitter size of 4x4microm2

451 Dependencies of current on AlAs Barrier thickness

Since the AlAs barrier is what limits the transportation of electron flow and

hence the current (which depends exponentially on the tunnelling barrier thickness)

therefore the first analysis to run on the simulation is the variation on barrier thickness

In the simulation the barrier thickness is measured in term of the monolayer Generally

one monolayer can be calculated by dividing the lattice constant of the material by two

In the case of AlAs one monolayer is calculated as follows

119900119899119890 119898119900119899119900119897119886119910119890119903(1119872119871) =

119860119897119860119904 119897119886119905119905119894119888119890 119898119886119905119888ℎ119890119889 (5666Å)

2

(43)

= 283Å

The nominal value of the AlAs barrier thickness for sample XMBE304 is 283nm

ie ten monolayers In the first simulation test the current change due to the barrier

thickness variation from 9ML to 11ML was examined first followed by the amount of

barrier thickness change that would produce a 10 change in current The simulation is

setup by fixing all other parameters and varying the AlAs barrier thickness as mentioned

136

above with a step of 02ML In order to determine what fraction of a ML would yield a

10 difference in the current the barrier thickness is slightly changed to fit the curve for

both 5 above and 5 below the original curve

Figure 46 IV characteristics of the dependencies of current on AlAs barrier

The current-voltage characteristics of the ASPAT diode do change dramatically with

barrier thickness in forward bias but not that much in reverse bias (Figure 46) The

current decreases as the barrier thickness increases For a 1ML change in layer thickness

(from 9ML to 11ML) the current changes by over ~300 at 05V

Figure 47 Example of analysis at -1 and 1V to the current

-00002

0

00002

00004

00006

00008

0001

00012

00014

-15 -1 -05 0 05 1 15

Cu

rre

nt

(A)

Voltage(V)

9ML

92ML

94ML

96ML

98ML

10ML

102ML

104ML

106ML

108ML

11ML

973E-06

0

0000005

000001

0000015

000002

0000025

000003

0000035

85 95 105 115

Cu

rren

t a

t 1V

B

ias

(Am

p)

Barrier Thickness (ML)

Current change with tunnel barrier thickness

Forward Current

5

-5

137

From the simulation result shown in Figure 46 a 9945ML barrier thickness will

give a 5 higher current and a 10056ML barrier will give a 5 lower current (Figure

47) Therefore in total 01 ML difference yields 10 current difference These indicate

that in order to control the current within 10 barrier thickness difference the growth

precision in the barrier must be precise to better than 01ML Extensive studies have

also shown that the I-V characteristic of a GaAsAlAsGaAs diode is very sensitive to

the thickness of AlAs barrier This work has been reported elsewhere[17]

452 Dependence of current on Spacer I length l1

For the longer spacer length (l1) five different values are chosen from 01microm to

03microm The lengths are changed in the order of 005microm Therefore the arrangement of

length is as follows l1=01microm l1=015microm l1=02microm l1=0 25microm and l1=03microm

respectively The results are plotted from -15V to 15V anode voltage in Figure 48 The

I-V characteristic of the ASPAT diode does not change much with the length in the

forward bias region but in the reverse bias region the current decreases as the layer

thickness increases Here the l1 layer acts as a voltage arm and a small size device

cannot sustain big changes in spacer length Changing the length at the forward region

will also change the energy states on the anode side as well changing the states

distribution on the cathode side in reverse bias

Figure 48 I-V characteristic of the dependencies current to Spacer I layer

-00002

0

00002

00004

00006

00008

0001

-2 -1 0 1 2

Cu

rren

t (A

)

Voltage (V)

L=01

L=015

L=02

L=025

L=03

138

By fixing the current at -1V and 1V the current and the layer thickness relationship is

illustrated in Figure 49

Figure 49 Current changes with layer thickness l1

It is noticeable that there is a dramatic change in reverse current from 01 microm to

05 microm layer thicknesses However the forward current only falls slightly from 01microm to

015microm and is stable afterwards Hence for a small size device a large change in spacer

layer at the cathode will allow more current to pass

453 Dependence of current on Spacer II length l2

Finally is the variation in the spacer II Similar to spacer I above five values are

chosen for the shorter undoped GaAs layer length l2 the thickness is varied from

00025microm to 00075 microm (steps are l2=00025microm l2=000375microm l2=0005microm

l2=000625microm and l2=00075microm respectively The results are plotted from -15V to

15V anode voltage as shown in Figure 410 In this case a slight change in I-V

characteristic in the forward bias can be seen clearly which means the I-V

characteristic depends on the length of the shorter undoped layer Therefore the l2 layer

also acts as another voltage arm due to the asymmetrical length The effect is quite

similar to the spacer l1 but this times the forward current only slight changes The reason

for the small change in current is that the length change is small and it linearly affect the

states distribution

-000002

-0000018

-0000016

-0000014

-0000012

-000001

-0000008

-0000006

-0000004

-0000002

0

0

000005

00001

000015

00002

000025

00003

000035

01 015 02 025 03

Rev

erse

Cu

rren

t (A

)

Forw

ard

Cu

rren

t (A

)

Spacer Lenght (microm)

Forward Current (1V)Reverse Current(-1V)

139

Figure 410 IV characteristic of the dependencies current to Spacer 1 layer

Fixing the current at -1V and 1V the current versus layer thickness relationship is

illustrated in Figure 411

Figure 411Current change with layer thickness l2

The I-V curve depends on the length of the shorter undoped spacer layer quite linearly

The forward current changes in increase to the layer compared to the backward current

The layer thickness l2 should be small as long as it prevents carrier diffusion Therefore

all these three layers must be kept within limit to ensure that the high performance of the

ASPAT diode can be fully utilised

-00002

0

00002

00004

00006

00008

0001

-2 -1 0 1 2

Cu

rren

t (A

)

Voltage (V)

L=75n

L=25n

L=625n

L=5n

L=375n

-2E-07

-18E-07

-16E-07

-14E-07

-12E-07

-1E-07

-8E-08

-6E-08

-4E-08

-2E-08

0

0

00002

00004

00006

00008

0001

00012

0002 0004 0006 0008

Rev

erse

Cu

rren

t (A

)

Forw

ard

Cu

rren

t (A

)

Spacer Lenght (microm)

Forward Current (1V)

Reverse Current(-1V)

140

46 Temperature Dependent Simulation

This section will discuss in detail how changes in operating temperatures influence

the IV characteristics of the ASPAT diode The same codes as in the previous

simulation with fitted results are used for this temperature dependence study but a few

parameters were changed for different temperatures

Theoretically the material parameters that are influenced by the change of

temperature are band gap electron effective mass the density of state (NC NV) light

hole mass heavy hole mass permittivity and electron amp hole mobilities However not

all mentioned parameters will have a large impact on the IV characteristic in the

SILVACO Atlas simulation The most significant factors that give appreciable impact on

the DC output current-voltage were the energy bandgap and the effective mass The

GaAs bandgap as a function of temperature is given by the equation below [106]

119864119892 = 1198641198920 minus

120572 1198791198712

120573 + 119879119871

(44)

Here Eg is the bandgap Eg0 denotes the bandgap at 0K TL is the Temperature α=

Constant (Varshni Parameter) AlAs6e-4 GaAs5405e-4 β= Constant (Varshni

Parameter) AlAs408 and GaAs204 The calculated parameters that are used in this

simulation are shown in Table 42

Table 42 The calculated values of bandgap at different temperatures

Temperature (K) GaAs Eg(eV) AlAs Eg(eV)

77 1506 2903

100 1500 2899

125 149 2893

150 1486 2887

175 1478 2879

200 1470 2871

225 1461 2863

250 1452 2854

275 1443 2844

300 1424 2835

325 1419 2824

350 1414 2814

375 1404 2803

398 1394 2793

141

The effective mass of the materials used in this simulation can be expressed by

119898119899 = 1198980119899 + 11989810 (

119879119871

300119870)

119898119901 = 1198980119901 + 1198981119901 (119879119871

300119870) + 1198982119901 (

119879119871

300119870)2

(45)

Where mn is the effective electron mass mp represents the effective hole mass m1n and

m1p are constant number for the basic GaAs material m0n

for 119898119883119898119871 119886119903119890 119892119894119907119890119899 119887119910 119905ℎ119890 119890119902119906119886119905119894119900119899 (1198981198991199052 lowast 119898119899119897)

13 while m0p is based on the

expression 1198980119901 = (11989811990111989732

+ 119898119901ℎ32

)23 The calculated parameters are shown in Table

43

Table 43 The calculated effective masses for each temperature used in this simulation

Temperatures (K) GaAs Effective Mass (kg) AlAs Effective Mass (kg)

77 00660m0 03790 m0

100 00658 m0 03788 m0

125 00655 m0 03785 m0

150 00652 m0 03782 m0

175 00649 m0 03779 m0

200 00646 m0 03776 m0

225 00643 m0 03773 m0

250 0064 m0 0377 m0

275 00637 m0 03767 m0

300 00634 m0 03764 m0

325 00631 m0 03761 m0

350 00628 m0 03758 m0

375 00625 m0 03755 m0

398 00622 m0 03752 m0

142

Figure 412 Measurement and simulation comparison result as a function of temperature range

from 100K to 398K

Figure 412 above shows excellent agreement between simulation and

measurement results at various temperatures The IV characteristics correspond to a

device of size 100times100microm2 as presented in Chapter 3

47 Temperature Dependence characteristics of GaAsAlAs ASPAT Diodes

As mentioned earlier in chapter one the tunnelling diode has many advantages

over conventional Schottky barrier diodes some of which are a large dynamic range

low power consumption and very weak temperature dependence This section will

discuss the effect of variable temperature applied to the GaAsAlAs ASPAT diode and a

similarly processed TiAu Schottky diode Two samples were fabricated together for

these studies (XMBE304 and XMBE104 representing an ASPAT and a SBD

respectively) The fabrication technique is exactly the same as has been discussed in

Chapter 3 In order to make it fair for direct comparisons as well as easy probing both

diodes were fabricated with the same emitter size (100times100microm2) The DC measurements

at different temperature were carried out using a Lakeshore Cryogenic probe station over

the range of 77K to 398K in 25K step interval

-002

-001

0

001

002

003

004

005

006

-2 -15 -1 -05 0 05 1 15 2

Cu

rren

t I

(Am

p)

Voltage V (Volt)

T=100K_Simu

T=100K_Meas

T=398K_Simu

T=398K_Meas

T=200K_Simu

T=200K_Meas

T=300K_Simu

T=300K_Meas

143

471 GaAsAlAs ASPAT diode vs TiAu SBD

Once fabrication and measurement were completed both DC outputs of the diodes

were characterised and analysed Figure 413 shows a semi-logarithmic plot for

measured current versus voltage as a function of temperature for ASPAT sample

XMBE304 In forwards bias the current changes for different temperatures from 77K

to 398K are less than 5 percent This confirms the very weak temperature dependence of

current transport as it is dominated by tunnelling through the barrier On the other hand

the backward bias shows the current changes at different temperature are slightly bigger

than in forward this due to band bending occurring faster (making the effective barrier

lower) and allowing thermionic emission to significantly contribute to transport of the

current The only other study of temperature dependence for the ASPAT was made by et

el RT Syme[15] but details were not stated in their report

Figure 413 Log Current vs voltage as a function of temperature for ASPAT sample XMBE304

The effective barrier height for the GaAsAlAs ASPAT diode is higher than that of

the SBD (See Figure 414) therefore there is an expectation of more limited thermionic

current flow in the ASPAT than the SBD As mentioned earlier the conventional

Schottky Barrier diode that is used in this study consists of a Gold Titanium and GaAs

(AuTiGaAs) interface which is the baseline for the temperature dependence study

0000001

000001

00001

0001

001

-15 -1 -05 0 05 1 15

Log

Cu

rren

t (A

)

Voltage (V)

T=77K

T=100K

T=125K

T=150K

T=175K

T=200K

T=225K

T=250K

T=275K

T=300K

T=325K

T=350K

T=375K

T=398K

144

Figure 414 The effective barrier high of the Schottky barrier diode and the ASPAT Diode [3]

The SBD epitaxial layers profile is as shown in Table 44 below Theoretically the SBD

obey thermionic emission transport[44] and its I-V characteristic is given by

119868 = 1198680[exp (

119902119881

119899119896119879) minus 1]

(46)

Where q is the electron charge V is the applied voltage across the diode n denotes the

diode ideality factor k is the Boltzmann Constant T is the absolute temperature in

Kelvin and I0 is the diode saturation current which is given by the expression

1198680 = 119860119860lowast1198792 exp (minus

119902empty1198870

119899119896119879) exp (minus120572120594119890

12120575)

(47)

here A is the area of the diode A denotes the effective Richardson constant Oslashb0 is the

barrier height at zero bias δ represents the thickness of interfacial insulator layer χ

denotes the mean tunnelling barrier and α = radic(4120587

ℎ)(2119898lowast) is a constant value The

ideality factor n is taken from the slope of the SBD current-voltage characteristic and in

this study its value varies from 1 to 2 (depending on temperature) In the case of the

ASPAT diode thermionic emission can also happen if a thicker barrier is used (~ 100Aring

or thicker) as shown by CS Kyono et el [104] who concluded that when a thicker

barrier of AlAs barrier is used the current transport is dominated by thermionic emission

145

Table 44 Epitaxial layer profile for Schottky Barrier Diode (XMBE104)

Layer Material Doping(cm-3

) Thickness(Aring) Bandgap (eV)

Schottky GaAs(Si) 500E+15 7500 14

Semiconductor GaAs(Si) 300E+16 7500 14

Semiconductor GaAs(Si) 100E+17 7500 14

Ohmic GaAs(Si) 500E+17 7500 14

Buffer GaAs(Si) 300E+18 7500 14

Substrate GaAs(Si) N+ 3000 14

The fabricated SBD was also measured and its I-V characteristic is plotted as a

function of temperature in Figure 415 Unlike the ASPAT diode the current at forward

bias for the SBD change enormously with temperature from 77K to 398K and at all

biases For the ASPAT diode the slight change in current only started after 08V bias as

the current starts to have some component of thermionic emission over barrier

Figure 415 Log Current vs voltage as a function of temperature for SBD sample XMBE104

In order to see clearly how much the current is changing in forward bias for both

ASPAT and SBDs diode a log current at different voltages versus 1000temperature is

plotted as shown in Figure 416

1E-08

00000001

0000001

000001

00001

0001

001

01

-2 -1 0 1 2

Log

Cu

rren

t (A

)

Voltage (V)

T=398K

T=375K

T=350K

T=325K

T=300K

T=275K

T=250K

T=225K

T=200K

T=175K

T=150K

T=125K

T=100K

T=77K

146

Figure 416 Influence of temperature on IV characteristic for both ASPAT diode and SBD

Semi-logarithmic plots of current (at V= 05 06 07 08V) versus inverse

temperature for both SBD and ASPAT are shown in Figure 416 When the temperature

is increased the current also increases in the SBD as a result of thermionic emission

over the barrier for sample XMBE104 This is in contrast to the temperature-

independent tunnelling through the thin AlAs barrier of sample XMBE304 ASPAT

diode where when the temperature is increased the current is almost constant

At low and high temperatures the ASPAT shows excellent temperature

independence with a constant current flow It exhibits a tunnelling current in excess of

values expected by the elastic tunnelling current calculation equation suggested by RT

Syme [16 18] above (Eq 1) using a Oslash value of 105eV (ΓGaAs to Γ AlAs tunnelling) By

contrast for the SBD at low temperature (77K-275K) the changes of currents were very

high and for every 02V there is an exponential change of more than 40 This

temperature dependent study was also reported in[68]

147

48 Conclusions

This chapter demonstrated the establishment of an excellent physical model and

comparison of room temperature I-V characteristics of GaAsAlAs ASPAT diodes for

different emitter sizes their scalability as well as an investigation of their characteristics

at different temperatures from 77K to 398K Simulation are validated on well-

characterized experimental data and excellent fitting which had been achieved in this

work permit the designer to extract all related parameters of heterojunctionmultilayer

ASPAT structures thus creating modification for future growth specification in order to

achieved precise designs

It is clear that the work which had been carried out in this chapter is able to

achieve with adequate accuracy a claim of reverse engineering capability The ability of

the GaAsAlAs ASPAT to act as a zero-bias detector has been analysed and compared

with the SBD It is clear that the temperature stability which is shown by the

GaAsAlAs ASPAT is much better than that of the SBD thus demonstrating that the

tunnelling current is dominant over the thermionic emission in ASPAT diodes

148

5 DC amp RF CHARACTERISTIC OF ASPAT DIODES

51 Introduction

To assist in circuit designs for any type of high-frequency circuits such as

millimetre wave detectors frequency multipliers and mixer circuits which are built

based on non-linear devices (diodes) an equivalent-circuit model for the diode is

required This is among the simplest and most effective method for analysing

semiconductor devices which work at high frequency where the electrical characteristics

measured obtained from the devices are extracted and presented in a circuit consisting of

lumped elements components (resistor inductor capacitor etc) However accurate DC

and RF measurement data is essential to extract the equivalent-circuit elements quickly

and correctly The extracted parameters values from the circuit that are taken into

account usually depends on bias and frequency associated with the device physically

which is also interrelated to the semiconductor material parameters device structure as

well as fabrication process flows

In this work the DC and RF data were derived from DC and S-parameter

measurements respectively These measurements were carried out both in-house and at

the University of Cambridge by a collaborator partner (Prof MJ Kelly) The I-V

characteristics of the diode obtained from DC measurements were measured from -2V to

2V while the S-parameters were carried out over a wide frequency range from 40MHz to

40GHz with eight different biases In this chapter the DC measurements for various

sizes of the diodes with analysis of their IV characteristics will be discussed The one-

port on-wafer ASPAT measurement setup as well as the de-embedding method will

also be explained Thereafter the equivalent circuit models with all lumped element

effect will be discussed This work is carried out with the help of the VNA which

principle has been described in Chapter 2 and Keysight ADS simulation tool All

technical details regarding the equivalent circuit models will be explained together with

the method used for the ASPAT diode evaluation The equivalent circuit model also will

cover the diode intrinsic elements such as Cj Rj and Rs and extrinsic elements ie CP

149

and RP Finally an equivalent circuit model with the small signal characterization of the

fabricated ASPAT diodes will be presented

52 General Analysis of Current-Voltage (I-V) characteristics of GaAsAlAs

ASPAT diodes

The recent development of state-of-art for DC measurement apparatus has led to

capabilities for high-level accuracy of measuring voltages to a few nano Volts and

current signals in the femto Amp range[107] This can easily be obtained by exploitation

of proper connections and high-quality cables connecting the equipment to the Device

under Test (DUT) In this work the DC measurements were carried out using an Agilent

B1500A Parameter analyser whose description was covered in Chapter 2

As was discussed in Chapter 3 the GaAsAlAs ASPAT diodes have been

fabricated with different mesa areas between 2times2microm2 to 100times100microm

2 but the smallest

size obtained with good I-Vs was 4times4microm2 In this section the focus will be on how the

extracted data can be expanded further for empirical modelling Figure 51 shows typical

results for measured ASPAT diodes with various dimensions to check for their

uniformity According to our standard procedure the DC measurement has to be

conducted prior to the RF to ensure the diode is in fully working order as this will later

save a lot of time during RF characterization

Figure 51 IV Characteristics of measured ASPAT diodes for emitter sizes of 4x4um2 6x6um

2 and

10x10um2 Note the good scalability

1E-10

1E-09

1E-08

00000001

0000001

000001

00001

0001-2 -1 0 1 2

Cu

rre

nt

De

nsi

ti (

Amicro

m2)

Voltage (V)

4x4microm^2

6x6microm^2

10x10microm^2

150

Figure 51 above demonstrates the IV Characteristics of measured GaAsAlAs

ASPAT diodes (XMBE304) for emitter sizes of 4times4um2 6times6um

2 and 10times10um

2 This

sample was processed using the dielectric bridge technique developed in this work It

can be observed that current per unit area for each dimension fits and scales to each

other The scalability of each diodes measurement is very important to ensure no process

related issues are hampering the devicersquos proper operations This also confirms that the

diodes are completely functional and can be used for the next stage of measurements

The advantage of having an excellent scalability of those diode sizes is that a prediction

of smaller emitter size can be made

This type of IV characteristic shows asymmetric behaviour which results from

the unequal spacer lengths of the device This behaviour is very useful for detection

application as it obeys a square law model The square law predicts that the current is

proportional to the square of the applied bias

119868 = 1198861198812 119908ℎ119890119899 119881 gt 0

119868 = 0 119908ℎ119890119899 119881 lt 0

(51)

To extract the first order effects of ASPAT diodes DC measurements which

result in asymmetric I-V characteristics are analysed The slope of the non-linear region

is used to determine the junction resistance (Rj) which is obtained from the first

derivative of voltage versus current (dVdI) The expression of Rj is given by

119877119895 =

120597119881

120597119868

(52)

In order to understand the relationship between Rj and diode sizes of the ASPAT the IV

characteristic for each diode displayed in Figure 51 is used to extract the Rj This has

been done by using the expression in equation (52) above and their response is plotted

against bias as displayed in Figure 52

151

Figure 52 Junction resistance versus voltage

As can be seen in the Figure 52 above the Rj for each device decreases strongly

when the voltage increases At zero bias the 4times4um2 devices show the highest Rj value

followed by 6times6um2 and 10times10um

2 devices The junction resistance at zero bias obtained

from the 4times4um2 diode is around 86KΩ while reducing by a third for the 6times6um

2 and

10times10um2 diodes with Rj of 27KΩ and 10KΩ respectively A diode with a smaller

forward current under the same applied voltage will exhibit a larger Rj For a good

millimetre wave detector a device with a large value of Rj is desirable since it will

provide high detection sensitivity

The slope at the IV characteristic contributes to an important parameter that is

commonly used by electronic manufacturers to describe diode specification namely the

video impedance (RV) which is also known as the non-linear resistance The RV which is

extracted from the real part of the diode small signal impedance is highly dependent on

the DC bias current and only weakly depends on the series resistance of the diode (RS)

Therefore the video impedance is given by

119877119881 = 119877119895 + 119877119878 (53)

Where RS is the series resistance of the diode whose value is normally very small and

does not contribute much to the whole slope and hence RV is dominated by Rj The RV

changes in behaviour if any DC current is flowing through the diode Practically small

DC current in the range of 1 to 10 microAmp or total zero bias is used to maintain the

appropriate RV value (1-2KΩ to several MΩ) RV will also determine the voltage

-10

10

30

50

70

90

110

130

150

-01 0 01 02 03

Rj(

)

Voltage (V)

6x6um^2

10x10um^2

4x4um^2

152

sensitivity of the whole detector circuit This will be explained further in the next

chapter In the case of a detector with an amplifier RV of the diode acts as the RF

impedance which needs to be matched with the video amplifier ( impedance looking into

the diode from the amplifier)[108 109]

The quotient of the second order derivative to the first derivative

((d2IdV

2)dIdV)) when calculated from the whole I-V characteristic translates directly

into a curvature coefficient (k) This is the most commonly used figure-of-merit to

quantify diode nonlinearity at zero bias Figure 53 below shows the variation of k with

bias and more importantly the zero bias rectifying action for device sizes of 4times4 um2

6times6 um2 10times10 um

2 This parameter which represents the small-signal rectifying

action of the diode will affect the performance of the detector (voltage sensitivity)

Detailed discussions on how this parameter effect the detector performance will also be

discussed in the next chapter

Figure 53 The rectifying action of the diode at zero voltage which gives the ASPAT high sensitivity

near zero bias detection

Figure 53 above shows calculated curvature coefficient of the measured I-V

characteristics from the same diodes shown in Figure 51 The highest k value is

obtained from the diodes with size of 4times4um2 followed by 6times6um

2 then 10times10um

2

The curvature coefficient decreases sharply as the bias increases for each diode This can

be attributed to a significantly increasing number of electrons that tunnel through the

thin barrier which were accumulated in the 2DEG formed in the intrinsic spacer region

-5

0

5

10

15

20

25

30

-001 004 009

Cu

rvat

ure

Co

effi

cien

t(V

-1)

Voltage (V)

k(10x10 um^2)

k(6x6 um^2)

k(4x4 um^2)

153

An ASPAT diode with a smaller size will have a larger Rj with a corresponding smaller

current under the same bias condition and hence will demonstrate a larger k value In

this calculation the curvature coefficient at zero bias obtained from 4times4um2 6times6um

2

and 10times10um2 diode is 23V

-1 17V

-1 and 16V

-1 respectively

A summary of the ASPAT diodes parameters obtained from measured I-V

characteristics that have been translated into first and second order differentials are

gathered in Table 51 below and compared to other diodes in the literature

Table 51 Performance of the ASPAT diode obtained from measured IV characteristics in this work

Sample Rj(Ω) k(1volt)

ASPAT 10times10 microm2 10K 16

ASPAT 6times6 microm2 27K 17

ASPAT 4times4 microm2 86K 23

Ge Backward diode 182K[110] 159[110]

InGaAs Backward diode 154[110] 23[110]

Sb Backward diode 5K[111] 47[111]

Si-Backward diode 135K[112] 31[112]

PDB 15K[8] -

AlGaAs SBD 20-100K[113] 34-38[113]

GaN HBD - 16[114]

From Table 51 above it is clearly that the ASPAT diode has a comparable value of Rj

and k to existing detector diode in the research community and in the commercial

market Based on literature of each diodes stated in the table the key to obtaining a high

value of k at zero bias is to minimize any forward tunnelling current Furthermore the

largest ASPAT diode used here (10times10microm2) has very close performances to that of a

commercial diode ie discrete Ge backward diode (ref[110]) where both Rj and k value

are close to each other

53 RF Test Fixture Theory and Experiment

RF measurements differ from DC measurement as they are more complicated

and it is necessary to comprehend the basic measurement principles to achieve

meaningful data This is obligatory especially for on-wafer RF characterization and

154

analysis to attain precise results Most of the electronics component measurements

which have input and output for instance antenna amplifier cables etc are based on a

two-port network configuration The characteristics which can be extracted from these

components are usually used to define their impact on a more complicated system

The performance of the two-port network can be described by a few parameters

ie scattering (S-Parameter) admittance (Y-Parameter) Impedance (Z-parameter) and

Hybrid (H-Parameter) However the S-parameter approach is favoured for high-

frequency measurements as it is relatively easier to characterize the microwave

performance and is able to convert to other parameters when necessary The advantage

of S-parameters is that they can straightforwardlydirectly convert into other two-port

parameters as mentioned above in term of currents and voltages[115] In fact to obtain

the device capacitance the appropriate S-parameters needs to be transformed into Y-

parameters using specific equations Furthermore the devicersquos cut-off frequency can

also be obtained when S-parameter measurements are performed over a wide frequency

range

531 On-Wafer Measurement and Small Signal One-Port Characterizations

In this work the arrangement of the RF measurement setup is assumed to be a

linear system as small voltage amplitude signals are used this means that the signals

have only a linear effect on the network without any gain compression or attenuation

The assumption is still acceptable even though the typical ASPAT is characterised as

non-linear in nature because it is a passive device which will act linearly at any input

power level

Generally the S-Parameter measurements on a diode can be adequately and

suitably performed using a one port measurement The technique used to characterise the

output is similar to the two ports technique but only incident and reflected waves are

used to characterise the input and output ports of the device Essentially this is because

the ASPAT has only two terminals and it is a passive device like other diodes

Therefore the analysis will revolve around the S11 parameter Figure 54 below show the

S11 is a ratio of reflected wave to the incident wave

155

11987811 =

119877119890119891119897119890119888119905119890119889

119868119899119888119894119889119890119899119905=

1198871

1198861 119908ℎ119890119899 1198862 = 0

(54)

A VNA as described in Chapter 2 is used to measure the ASPAT diodes This

powerful equipment is able to measure S-parameters up to 40 GHz To conduct accurate

S-parameter measurement at the diode the measurement setup must be calibrated prior

to the actual measurements taking place

54 Device Calibration

541 Open and Short De-Embedding Technique

Further calibration to be made involves anything related or attached to the

device The co-planar waveguide (CPW) bond pad and interconnect line that are

attached to the intrinsic diode are the main contributors of the errors also are required to

be calibrated In general the bonds pad could generate a capacitance (parasitic) in

parallel with the intrinsic diode and its contribution depends on the size of the bonds pad

as well as the operating frequency Meanwhile the CPW and interconnect line may

cause a parasitic inductance in series with the diode

The method that is used to get rid of this parasitic is called de-embedding and the

most common technique to realise it is by introducing OPEN-SHORT structure[116]

This method is based on a lumped-elements model Parasitic elements of the diode

De-Embedding Structure

Incident wave

Reflected wave

One-port device

a1

b1

Figure 54 One port S-parameter measurements

156

equivalent circuit correlate directly to the access section of the CPW hence can be

derived from de-embedding structures The aim of the de-embedding technique is to

represent these parasitic elements so that the one-port characteristic of the actual diode

can be determined

The two types of de-embedding structure OPEN and SHORT are conventional

techniques that are widely used in this study The design of all structure must be

identical (in size) to the device to avoid any discrepancy It is very simple to design all

these three structures for example open structures are obtained by eliminating the diode

layout and keeping the bond pad layer only The short structure just adds a bridge and

ensures ground and signal pad are connected to each other Through structures are

realised by disconnecting both ground pad and leaving the signal pad to connect to each

other

To gain more accuracy this external effect must be removed by the implementation of

de-embedding structures on the same tile as the actual device Figure 55 shows the

fabricated de-embedding structures used in this study

Figure 55 Fabricated open short de-embedding structure and real device (zoomed) use for RF

calibration and measurements (Note Images are not to scale)

In summary the de-embedding which is used to extract out the parasitic elements

from entire single diode measurement is a very important step as normally on-wafer

measurement requires coplanar waveguide (CPW) to access the diode structure (active

region) The CPW will have some effects which will disturb the accuracy of the device

characteristics

157

55 S-Parameter Measurement Result and Analysis

This section will only present RF measurement results after all VNA setup and

calibration were performed The S-Parameter measurements were carried out on ASPAT

diodes at five different DC biases from -2 to 05 volt with a sweep frequency from

40MHz to 40GHz using a calibrated VNA and the input power was fixed at -30dBm

The measurement procedure as described in Chapter 2 was performed on the device

(on-wafer) equipped with the appropriate bond pads This is important to ensure the

results obtained are valid The reason for using different biases is to find at what voltage

the device capacitance is fully depleted This is also very important in determining the

cut- off frequency of the devices

In this research two phases of the experiment on the S-parameter measurements

were carried out The first phase is to qualify the process flow ie for manufacturability

and repeatability which can be obtained from the consistency of the result The S-

parameter measurements taken on the same wafer dies are repeated several times on

different GaAsAlAs ASPAT diodes There are three different tiles taken from 3

different wafers namely XMBE304A XMBE304B and XMBE304C carried out in

this experiment The repeatability tests are done mostly on the large devices (15times15microm2

up to 100times100 microm2) and the results are analysed based on the reflection coefficient (S11)

on Real and Imaginary measurements

In the second phase the measurement is toward producing devices that can

perform at high-frequencies This can be realised by utilising small emitter size devices

(4times4microm2 6times6 microm

2 and 10times10 microm

2) The measurement results of these devices will be

used to build the equivalent circuit models while both the intrinsic as well as the

parasitics of the device will be evaluated Hence all the values obtained from these S-

parameter measurements will be used to design the device that can be used in

millimetres-wave applications As can be seen in Figure 56 below the extracted S-

parameter measurement results comprise of a reflection coefficient (S11) for real

imaginary and Smith chart for XMBE304A While these measurement results are

extracted at zero bias voltage the other bias voltages will be used to extract the

capacitance This will be described in the final section of this chapter

158

551 Diode to diode uniformity

In order to study within tile uniformity and reproducibility statistics of the RF

performance five devices of different mesa sizes in the same tile (XMBE304A) were

measured at zero bias and represented in term of Real and Imaginary reflection

coefficient (S11) the uniformity check is carried out at three different frequencies step

under 15GHz since the cut-off frequency for these big devices is relatively low at about

~20GHz on average The variation of the reflection coefficient is taken from the

percentage of the (standard deviationmean values) for all five diodes from this run The

following figures show Real and Imaginary S11 of large mesa area ASPAT diode from

15times15microm2 up to 100times100microm

2 which are represented by lines graph in a few different

colours

Figure 56 Real S11 RF measurements for (XMBE304A) were done for four devices from

15times15microm2 to 100times100microm2 device sizes for within-wafer uniformity check

Figure 57 Imaginary S11 measurements (XMBE304A) to qualify the repeatability and

reproducibility were done on four devices from 15times15 microm2 to 100times100 microm2 at zero bias

159

The variance data extracted from the graph (Real S11) above for each device within-

wafer (device to device) uniformity study is summarize in Table 52 below

Table 52 Device to device uniformity check for large ASPAT diode

Device Size 100times100 microm2 50times50 microm

2 30times30 microm

2 20times20 microm

2 15times15 microm

2

Variation 5GHz 181 115 405 151 145

Variation 10GHz 106 133 522 424 293

Variation 15GHz 119 198 281 76 509

The majority of diodes show that the variations of S11 measurements are below

3 and only a few are below 8 These finding still can be considered as good for

manufacturing control since absolute I-V characteristics reported in [63] is set by

designerrsquos specification to be not more than plusmn10 variation Further extensive RF

measurements were carried out by the research collaboration with the University of

Cambridge on the same GaAsAlAs ASPAT diodes wafer [117] In their study they

focused on 50times50 microm2

mesa size 17 of diodes were chosen to be measured The study of

uniformity of RF characteristic only focused on frequencies below 20GHz The same

approach was used to get the variation of the reflection coefficient for all 17 diodes but

this work was carried at four different frequencies Table 53 shows the zero bias S11

result for four different frequencies and standard deviation of the devices

Table 53 The variation of reflection coefficient (S11) for GaAsAlAs ASPAT diode at four different

frequencies[117]

Frequency (GHz) 5 10 15 20

Variation () 197 243 26 276

From the results the variations of 50times50 microm2 mesa sizes measured in-house and

at the University of Cambridge are comparable with all variations showing good

uniformity ie recording variations below 3 This indicates that the RF performance

of the GaAs AlAs diode is valid and reproducible and is thus considered as a good

achievement for manufacturing Once the reproducibility and repeatability of the large

devices showed stable results the fabrication process then continued to obtain smaller

emitter size for work at high-frequencies

160

552 Wafer to wafer uniformity

Other RF measurements were conducted on sample XMBE304B which was

fabricated in-house using the same process steps but the only difference from

XMBE304A was the use of SiN3 as dielectric In this run three different mesa sizes

were measured (15times15mmicro2 20times20microm

2 and 30times30 microm

2) and Real and Imaginary S11

plotted against frequency The RF performances of both samples are gathered in one

graph as shown in Figure 58 below

Figure 58 Wafer to wafer uniformity test on 1st and 2

nd RF measurement(XMBE304B) to qualify

the repeatability and reproducibility were done for 30times30microm2 to 15times15 microm

2 device sizes (Real and

Imaginary) Note blue colour is XMBE304A and red colour is XMBE304B

For this wafer to wafer uniformity study four diodes with three different sizes as

specified previously were measured from sample XMBE304B and four diodes from

previous measurements of XMBE304A The blue line in Figure 58 represent

measurement result of real and imaginary for sample XMBE304A while the red line

161

represents XMBE304B The uniformity data is compared at three different frequencies

and gathered in the table below

Table 54 Wafer to wafer uniformity check on XMBE304A and XMBE304B

Device Size 30times30 microm2 20times20 microm

2 15times15 microm

2

Wafer A vs wafer B variation 5GHz 305 31 1 314

Wafer A vs wafer B Variation 10GHz 352 344 329

Wafer A vs wafer B Variation 15GHz 321 376 359

As can be seen in the Table 54 above the wafer to wafer uniformity is rather

large (30) on average The main reasons being that sample XMBE304A was

processed by utilizing S1805 as a dielectric layer while sample XMBE304B used

Si4N3 Although the process steps are similar for both wafer processing the use of

different dielectric layer will influence the diode parameters especially resistance and

capacitance as the dielectric constant for each materials is different Secondly the wafer

processing is not run concurrently at the same time thus the moisture and temperature in

the clean room might differ for both processing Although the wafer to wafer uniformity

test for this run might not be favourable for manufacturing tolerance at least the use of

different dielectric layer shows some significant result in term of capacitance resistance

effect to the GaAsAlAs ASPAT diode

553 Small devices RF measurements

The first objective of this study was to make smaller size mesa devices ie

1times2microm2 1times3 microm

2 2times2 microm

2 and 3times3 microm

2 However for GaAsAlAs ASPAT type this is

difficult to achieve in practise These issues were discussed in detail in Chapter 3

Hence the smallest emitter size that yields repeatable and reproducible results was

4times4um2 The final measurement which was done on sample XMBE304C focused on

small devices The measurements were done on four devices two with the diode bond

pads sitting on substrate (GaAs SI) and the other two sitting on dielectric layer (Si4N3)

Figure 59 below shows three measured results obtained from sample XMBE304C

using the 3rd

Gen Mask

162

Figure 59 RF measurement performed on device sizes of 10times10microm2 6times6mmicro

2 and 4times4microm

2 (Real and

Imaginary) Note that green red and blue colour represents 4times4microm2 6times6mmicro

2 and 10times10microm

2 diodes

respectively

Figure 510 RF measurement performed on device sizes of 10times10 microm2 6times6 microm2 and 4times4 microm2

(Smith Chart) Note that green red and blue colour represents 4times4microm2 6times6mmicro2 and 10times10microm2

diodes respectively

30MHz

40GHz

163

As can be seen in Figure 59 (Real Imaginary) and Figure 510 (Smith Chart) are

obtained from measurement of four diodes in the same tile The diode to diode

uniformity that is sitting on the same platform obtained at 35GHz frequency in this run

on average is ~15 25 and 1 for 4times4microm2 6times6mmicro

2 and 10times10microm

2 diodes

respectively On the other hand the uniformity between diode to diode sitting on

dielectric and substrate is quite high due to different capacitance value of devices on

average ~7 are attained from three different sizes of diode

From the measurement results above the Real S11 measurement of four different

sizes show the same trend for each frequency At low frequency resistances for each

diode is high as the S11 value is large At intermediate frequency the values drop

tremendously for big devices (30times30 mmicro2 and 20times20 mmicro

2) ie in Figure 57 At high

frequency all diode reach saturation limit as the value are constant Small devices

(4times4mmicro2 6times6mmicro

2 and 10times10microm

2) ie Figure 59 show S11 values that are higher than

those of large devices as smaller emitter diode have larger resistance value

The imaginary S11 value also shows the same trend as for big devices However

for small devices in this run (4times4mmicro2 6times6mmicro

2 and 10times10microm

2) the S11 value keep

dropping toward negative values at increasing frequencies This indicates that bigger

devices with positive value at high frequency are more capacitive than the smaller

devices It is worth mentioned that the capacitance and inductance values for device

sizes of 4times4microm2and 6times6microm

2 come from the CPW layouts and these are dominant while

for device sizes of 15times15 mmicro2 and above the device capacitance itself is dominant

The Smith Chart shows the reflection coefficient (S11) as a function of the

applied frequency (30MHz to 40GHz) All measurements from each mesa size follow

unique impedance circle which is that most of the lines are at the lower right outer ring

This means that the diode capacitance value is frequency dependent For the case of

10times10microm2

devices these impedance circles are mostly toward the outer ring meaning a

higher capacitance than the other two mesa dimensions All the device constantly follow

the outer ring without crossing any real axis at any frequency point meaning that the Cj

is not shorted at the maximum 40GHz measurement frequency (not reached cut-off

frequency) Therefore the entire small GaAsAlAs diodes in this run have capability to

work in the millimetre wave frequencies range

164

56 Extracting RF models of ASPAT at Zero Bias Voltage

The methodology used in the S-Parameters measurement for high-frequency

analysis must ensure that the derivation of the equivalent circuit corresponds to their port

characteristics In other words the component representing the ASPAT in the equivalent

circuit model must have physical significance otherwise the circuit will be meaningless

The fabricated ASPAT diodes as discussed in Chapter 3 have the cross section shown in

Figure 511

There are two main components that can be extracted from the fabricated

ASPAT depicted above ie intrinsic and parasitic The intrinsic refers to the main

structure of the diode itself and are represented by three bias dependent elements

namely Junction Resistance (Rj) Junction Capacitance (Cj) and diode Series Resistance

(Rs) The parasitic is the elements related to the bond pad of the anode and cathode as

well as interconnects They are represented by parasitic inductance (LP) resistance (RP)

and capacitance (CP)

The diode parameter extraction is different from the three terminal devices

(FETs) in the sense that FETs are a kind of direct extraction in which all the elements in

the transistor have linear functions to the port characteristics ie S-parameter Y-

Parameter Z-Parameter and can easily be solved by the matrix calculation method for

those particular parameters[118] The same extraction method cannot be applied to the

diodes because its elements will embroil with each other Therefore only one method is

used to extract the diode element which is optimisation by tuning the initial value toward

the measured S-Parameter values

Figure 511 Cross-sectional of fabricated ASPAT diode and the corresponding equivalent circuit

model

165

The strategy used to model the ASPAT is based on an initial fitting value of the

lumped elements to the extracted value from measurement on three S-parameter graphs

(real imaginary and Smith chart) for the reflection coefficient (S11) The refinement is

accomplished by optimisation and fine tuning of the values which result in minimum

error between extracted and modelled values Figure 517 (on page 153) shows fitted S-

parameter result for extracted and model numbers with each one fitted in a single line as

an example However to achieve this excellent fitting key prior steps have to be used

de-embedding fitting the intrinsic value and optimisation

561 Extraction of ASPAT parasitic element

Once the S-parameter measurements achieve stability repeatability and

reproducibility for each measurement in term of S11 results as mentioned above the

results of the de-embedding structure which had been measured prior to the device

structure are then extracted to form a well-defined equivalent circuit In order to build

and analyse the equivalent circuit firstly the measured data is imported into the ADS

software prior to any fitting This can be realised via the ldquoStart The Data File Toolrdquo

features provided by this particular software When successfully imported the data is

read by the function S2PMDIF (These files are a natural extension of two-port S-

parameter Touchstone files) as depicted in Figure 512 below

Figure 512 The S-parameter Touchstone file is used to read the measured files

166

For the open and short techniques after de-embedding the equivalent circuit

model which is represented by mainly a capacitor and an inductor is built The open

structure requires resistance and the capacitance values of 20KΩ and 26fF respectively

connected in parallel to be well fitted to the real imaginary and Smith chart (S11) output

On the other hand for the short structure the Real imaginary and Smith chart (S11) have

to satisfy the values of resistance and inductance elements of 1Ω and ~47pH respectively

connected in series These values strongly rely on the bond pad or CPW dimension and

length The Equivalent circuit models and fitted data as well as measurement can be

seen in Figure 513 and Figure 514 below

Figure 513 Equivalent Circuit Model of Open CPW or de-embedded structure

Figure 514 Equivalent circuit model for short de-embedded structure

To satisfy the equivalent circuit a self-consistence method introduced by

Ren[119] is utilised This approach accurately extracts the CPW capacitance (Cpad) and

inductance (Lpad) as well as intrinsic Junction capacitor (Cj) which is attained from the

one-port S-parameter measurements Therefore the pad capacitance introduced by the

self-consistence method for the open structure can be expressed by

167

119862119875 =

119868119898(11988411119874119901119890119899)

120596

(55)

Lpad which represents the short structure is given by

119871119875 =

1

120596(119868119898 (11988411119874119901119890119899minus 11988411119878ℎ119900119903119905

))

(56)

Here Y is the admittance parameter (Y-parameter) converted from the S-parameter

measurement data and ω is the angular frequency The extracted measurement data

represented in the equivalent circuit fits with the simulated data in three S11 graphs as

can be seen in Figure 515 below From the Smith Chart it can be clearly seen that both

open and short S11 results are on the circumference which means the resistance of the

short structure is very small while in the open it is very large Additionally the

calculated Cpad and Lpad using Equations 55 and 56 above produce results similar to

those obtained in the equivalent circuit model for the open and short structure The

values are ~25fF and ~45pH respectively These data completely verify and validate

both results

Figure 515 Smith chart representative S-parameter measurement for short (left) and open (right)

CPW The blue lines represent simulated data and the red is measured data

Short

Open

168

562 Extraction of ASPAT intrinsic elements

Once the parasitic elements are determined it is easy to build a complete ASPAT

equivalent circuit The ASPAT is not like other tunnelling diode which their equivalent

circuit models widely studied ie RTD [120] IMPATT and PDB The only literature

which reports ASPAT equivalent circuits can be found in [15] and other RT Symersquos

journal paper[16] Fortunately its equivalent circuit model is not much different

compared to other diode video detectors Thus other literature which is based on

Schottky diode equivalent circuit model used for detector application can be referred to

The simplest form of ASPAT equivalent circuit and other video detectors intrinsically

consist of junction capacitance (Cj) series resistance (RS) and junction resistance (Rj)

First and foremost to extract the equivalent circuit one must know the theory behind

each parameter that is developedbuilt as a spine to become a complete element This is

vital to ensure the equivalent circuit is correct In the case of the ASPAT Cj is predicted

from a simple fully depleted parallel plate capacitor approximation which was discussed

previously in Eq (230)[15] Additionally for the S-parameter measurements the Cj can

also be validated by the self-consistence method mentioned earlier and thus can be

expressed by

119862119895 =

[

(1

120596)

1

1

119868119898 (11988411119905119900119905119886119897minus 11988411119874119901119890119899

)+

1

119868119898 (11988411119874119901119890119899minus 11988411119878ℎ119900119903119905

)]

(57)

This approach helps to verify both the fully depleted parallel plate capacitor in S-

parameter measurements The basic component which is responsible for the ASPAT

series resistance RS was discussed in detail in Chapter 2 RS and Cj are key contributors

to the high-frequency operation as expressed by the device cut-off frequency Equation

(58) below

119891119888119906119905minus119900119891119891 =

1

2120587119877119904119862119895

(58)

The R and C parameters must be kept as low as possible in order to obtain high cut-off

frequencies for millimetre wave applications From the fabrication point of view Cj can

169

be reduced by making as small a diode emitter size as possible while for RS reducing

the D gap is of paramount importance as it dominates the series resistance The ASPAT

contact resistance in the electrodes (contacts between metal and semiconductor) can be

reduced by using high doping in the ohmic layers

The junction resistance (Rj) of the ASPAT is taken from the 1st derivative or

slope of the current-voltage characteristics Normally the value of Rj is very large

(several kilo Ω) compared to Rs The small signals ASPAT equivalent circuit built with

intrinsic and extrinsic components is shown in Figure 516 below while the fitting

results is shown in Figure 517 and Figure 518

Figure 517 Excellent fitting between measured (red colour) and simulated (blue colour) for S11

(Real and Imaginary) results for various small device designs

Rj

Cj

Cpad Rpad

Rs Lpad

Figure 516 Equivalent circuit of the ASPAT diode

170

Figure 518 Excellent fitting between measured (red colour) and simulated (blue colour) for S11

results (Smith Chart) for various small device designs

The equivalent circuit that was built for the ASPAT is taken from sample

XMBE304C with emitter dimensions of 4times4 microm2 6times6 microm

2 and 10times10 microm

2 These

devices are expected to work in the millimetre-wave region and have cut-off frequencies

(intrinsic) of ~650GHz ~200GHz and ~100GHz respectively

Table 55 Comparison between calculated (fully Depleted) and extracted (different biases) values

from equivalent circuit parameters for different ASPAT mesa sizes at zero bias voltage

Parameters 4times4microm2 6times6 microm

2 10times10 microm

2

Calculated

(fully

depleted)

Extracted Calculated

(fully

depleted)

Extracted Calculated

(fully

depleted)

Extracted

0V -05V -1V 0V -05V -1V 0V -05V -1V

Cj(fF) 879 23 7 6 198 55 148 139 549 171 486 46

Cpad(fF) - 15 15 15 - 15 152 152 - 15 15 15

Lpad(pH) - 45 43 42 - 50 473 473 - 51 51 46

Rj(KΩ) - 90 833 522 - 35 392 392 - 12 125 13

Rs(Ω) 99 11 11 11 67 95 8 7 41 95 45 37

fcut-off

Cj(GHz)

1828 629 2066 241

1

1208 192 1344 163

5

710 107 728 935

fcut-off

Cj+Cp(GHz)

- 380 658 688 - 151 663 781 - 98 556 705

171

The focus in this study is purposely to build ASPATs as zero bias detectors that are able

to work in the millimetre and sub-millimetre frequency range therefore all the

parameters which are obtained from equivalent circuit were extracted at zero bias

voltage Theoretically the calculations which are derived from both self-consistence amp

theory can only be solved for fully depleted device capacitance (using a parallel plate

configuration) Hence both extracted and calculated results are compiled in Table 55

Noticeably the calculation can only produce the intrinsic parameters of the

ASPAT for fully depleted capacitance On the other hand both intrinsic and extrinsic

parameters of GaAsAlAs ASPAT are obtainable from extraction and thus help to

determine at what bias the diode is start to deplete The junction and series resistance (Rj

and RS) of each dimensions shown in Table 55 above were achieved by fitting the

elements of the equivalent circuit with the three measured S11 graphs whereas the Cpad

and Lpad were extracted by utilising the self-consistent method from the S-parameter

measurements which is fitting the de-embedding structure Additionally the Cj values

are obtained via fitting the measured S11 data and employing the self-consistence

approach Results obtained from both techniques are identical

At zero bias all extracted junction capacitance from each device sizes are very

different from the calculated one while the extracted series resistance are closer to the

calculation This means that Cj is a highly voltage dependent parameter and Rs is

voltage independent but solely dependent on device structure and material used to

fabricate it The extraction at -05V and -1V shows that the values of junction resistance

is changing for most of the devices which means this parameter also rely on bias voltage

as discussed earlier in Section 52

The cut-off frequency for each calculated devices are near the THz range even

for the 100microm2 emitter area However with the introduction of parasitics elements ie

pad capacitances fcut-off is degraded tremendously Therefore it is important to make sure

all the intrinsic elements have optimum values so that the target operating frequency of

the ASPAT diode can be met Due to this it is advisable to operate the devices at no

more than 13 of fcut-off when designing detector systems

The parameters extraction at -05 and -1V also show that Cj values are closer the

calculated ones which means the ASPAT diode is reaching full depletion

172

563 Capacitances -Voltage (C-V) Extraction

Theoretically the junction capacitance of the ASPAT is calculated from the fully

depleted formula 119862119895 = 휀119900휀119903119860119889 which was also discussed in Equation (230) in Chapter

2 in page 43 Its value depends on the change of voltages to depletion at the emitter

contact[15] and make it one of the voltage dependent parameter for the diode[121]

Therefore the C-V characteristic of the GaAsAlAs must be precisely extracted

In practice the capacitance is difficult to measure due to the very low resistance at zero-

bias However alternatively it can be measured and extracted by applying different

voltage and identifying the point at which there is change which essentially represents

full-depletion Apart from these it can also be extracted from S-parameters measurement

which is then converted to Y-parameters A C-V characteristic of the GaAsAlAs

ASPAT from XMBE304C is extracted and plotted as depicted in Figure 518 below

Figure 519 Capacitance extraction showing full depletion at -025 volts (modelled capacitance vs

Voltage)

From the graph shown in Figure 518 the capacitance is extracted at eight

different biases for 4times4 microm2 6times6 microm

2 and 10times10 microm

2 The devicersquos junction

capacitance for each dimension increases and reaches a maximum value at 025V There

are additional quantum capacitance effect which comes from an increase in the negative

charges in the 2DEG region (when band bending happens creating an accumulation

0

50

100

150

200

250

-2 -15 -1 -05 0 05

Cap

acit

ance

(fF

)

Voltage (V)

Cj(4x4microm^2)

Cj(6x6microm^2)

Cj(10x10microm^2)

173

region at the barrier) This charge is imaged by the positive charge in the whole

depletion region Increasing the voltage toward positive values leads to a lowering of

the AlAs barrier and thus allowing thermionic emission to take place after certain bias

values leaving only the depletion capacitance and making the quantum capacitance

negligibly small

In the reverse bias case the device junction capacitance reaches a saturation

(fully depleted capacitance) at a voltage of -025V and remain constant up to -2V If the

reverse bias voltage is increased further the ASPAT may reach breakdown Therefore it

is important to know how far the diode can withstand applied reverse bias to ensure it

can still give full performance

57 Conclusions

In this chapter scalable DC characteristics of GaAsAlAs ASPAT diode derived

from three different emitter sizes of 4times4 microm2 6times6 microm

2 and 10times10 microm

2 was

demonstrated The current density obtained at zero bias is several microAmicrom2 These allow

1st order differential effect to exhibit high value of Junction resistance (Rj) at zero bias

However Rj is highly bias dependent The 2nd

order differential effect on IV

characteristics display a high value of curvature coefficient leading to high voltage

sensitivity when applied in millimeter wave detector applications These two parameters

are vital in the design of millimeter wave detectors and especially those operating at zero

bias

Subsequently RF measurement up to 40GHz of uniformity study for both within

wafer and wafer to wafer variance were undertaken An average uniformity below 7

was obtained for within wafer study on large device area ( 15times15 microm2 to 100times100 microm

2)

while for small device area ( 4times4 microm2 to 10times10 microm

2) a smaller 3 uniformity variance

was achieved in average However for wafer to wafer study the variant uniformity was

quite high at around 30 on average for relatively large device (15x15 microm2 to 30x30

microm2) This was mainly attributed to different dielectric layers used in the process flows

of the sample rather than fundamental MBE control of the AlAs barrier thickness

174

It was demonstrated in this chapter that careful on-wafer RF measurements of small

size GaAsAlAs ASPAT diodes allow accurate device parameter extraction of both

extrinsic and intrinsic parameters The extrinsic parameters are namely pad capacitance

and inductance with obtained values of 26fF and 47pH respectively These values were

obtained from de-embedding structure fabricated on the same tile as the real devices

The intrinsic parameters such as junction capacitor junction resistor and series

resistance had different values according to device dimensions The smallest zero bias

value of Cj obtained from 4times4 microm2 diodes was 23fF ensuring a high cut-off frequency of

380GHz and hence in the next chapter a 100GHz detector will be presented working at

slightly less than 13 of this cut off frequency

The C-V data extraction confirmed that the fully depleted capacitance started to

happen at around -025V The maximum junction resistance occurs at +025V largely

caused by the depletion region and an additional quantum capacitance effect CQ This

effect is strongly related to the size of a 2DEG which occurs under forward bias (01V to

025V) and can be reduced by having a smaller thickness AlAs barrier

175

6 MILLIMETRE WAVE GAASALAS ASPAT DETECTOR

DESIGN USING ADS

61 Introduction

The ASPAT diode having features of non-linear IV characteristics at zero voltage

make it useful for signals rectification ie detector and mixer for millimetre-wave

applications Additionally ASPAT diodes have a range of advantages such as large

dynamic range strong temperature insensitivity etc[15]over other rectifier diode This

makes ASPAT an appropriate candidate in RF detection applications Since 1940 the

only two terminals device that has been the workhorse for RF applications is the

Schottky Barrier Diode (SBD)[9] In its earliest form the SBD was built based on a

point-contact device which could not perform at high frequencies It was then developed

to work at higher frequencies by exploitation of epitaxial structures [10] and to date the

SBD remains the mainstay of two terminal devices that are able to work in the

millimetre and submillimetre-wave regions However as discussed in Chapter 4 the

performance of SBD is degraded at extremes of temperatures and these circuits

employing SBDs require temperature compensating circuitry Thus there is additional

complexity associated with technologies and applications related to SBDs

Before this work was carried out no model for the ASPAT diode as detector had

been available or developed especially using the empirical modelling ADS software

When the ASPAT diode was first introduced its function was conceptually explained it

was then built and tested to compare with other microwave detectors at X-band

frequency (95GHz) The comparisons were made in terms of detector parameters ie

sensitivity dynamic range temperature dependence etc[15] These early works lead by

RT Syme et al supplied the basic knowledge to model the ASPAT diode as a zero-bias

detector for mm-wave frequency gt100GHz For the SBD many models and equivalent

circuit approaches have been reported [122 123] The modelling of conventional SBD is

mostly implemented through fitting the S-parameter curves of the model to the

experimental one This approach is also carried out in this research since it is accurate to

predict the performance of the device under test [124]

176

This Chapter aims to introduce low cost reliable and sophisticated detector design

based on ASPAT diodes which is believed to be able to improvereplace SBD in

millimetre-wave applications especially in imaging The focus was on developing and

establishing an appropriate circuit design that suit the new ASPAT diode for such

applications The detector sensitivity as its key parameter ultimately limits the quality

and acquisition time of the detector In the subsequence section the theory of detection

including both direct and heterodyne will be discussed This is followed by definition of

detector characteristics of interest as well as noise consideration Section 65 present the

main focus of this chapter which is the development of 100GHz ASPAT detectors and

their result will be explained in term of all detector characteristic of interest

62 Detection Theory

Any incoming signal such as RF microwave or mm-wave in the form of envelope

function or single wave can be detected by rectification of the signal using a nonlinear

device ie transistor or diode The input and output signal (RF) signals are normally in

the form of amplitude as a function of time Typically the detector output is a low-

frequency signal known as the video signal which has amplitudes that are proportional to

the square of the input RF signalrsquos voltage amplitude

A complete receiver system as shown in Figure 61 below consists of receiver

antenna and a circuit designed to extract the signal and then amplify it The function of

the receiver system is the converse of the function of the transmitter side At the receiver

side the antenna is used to receive the signal it then conveys the signal to the extraction

circuit for detection as the information-bearing part of the signal (using the nonlinear

device as its heart) The signal is finally amplified to avoid any information strength

decay Additionally in a digital system the output signal which had been processed by

the detector circuit has to maintain an optimum input signal conveyed to in-phase and

quadrature (IQ) demodulators The output signal will go through a low pass filter then

to an analogue-to-digital converter (ADC) and thus a digital baseband output signals

will be produced

177

Figure 61 Block diagram represent a complete direct receiver system

There are two types of millimetre wave integrated circuit (MMIC) used for

detection purposes namely direct detectors and heterodyne detectors The direct detector

MMIC is the simplest circuit used for detector applications and has the simplest way of

extracting the RF information Due to its simplicity the direct detection method is

inexpensive and most attractive method used for measuring power in RF Laboratories

and Industries This detection scheme is also sometimes known as video detection[125]

The simplest way of explaining the detection process is that the incoming RF or

microwave signals depicted in Figure 62 below with an appropriate input power (Pin) is

rectified by using a diode and results in a corresponding output voltage (Vout) A detector

IC designed based on diodes is normally able to rectify very low levels of RF power (lt-

40dBm) then produces an output DC voltage that is proportional to the RF power A

rectifier diode can function at zero bias (which is very good for reducing noise) at very

small DC bias (003mA) and relatively high RF impedance which will produce around

600Ω This will affect the capacitance value and a low capacitance is needed to realise a

high detection sensitivity

Figure 62 The detection process of a single wave through a non-linear IV characteristic of a diode

Detector

Speakerdisplay unit Amplifier

Antenna

Vout

Pin

Tuner Amplifier

178

However this type of detector has a drawback which is itrsquos relatively low signal to

noise ratio Thus it will also rectify any incoming electrical noise at all frequencies and

up to the cut-off frequency (fC) The basics lumped components circuit as shown in

Figure 63 is used to build such detector which consists of Source impedance (Zo)

Rectifier diode (ASPAT or Schottky) wire or pad inductance and capacitance and Load

impedance (RL)

Figure 63 Lumped element illustration of microwave detector circuit

Another type of detector is the heterodyne method which mixes incoming RF or

microwave signal (fRF) with another constant signal produced by a secondary circuit

called the Local Oscillator (LO) The LO frequency (fLO) must be slightly higher than fRF

to enhance the RF signal This mixing between fRF and fLO happens in the nonlinear

device as depicted in the Figure 64 This will produce a signal at a different frequency

called the intermediate frequency (fIF) which can then be amplified and detected as

explained in the previous paragraph Theoretically a basic requirement of the mixer is to

have fIF as efficient as possible while practically the minimum conversion efficiency

obtained is around 20 The main reason for using a mixer is due to the fact that

selective amplifiers at RF frequencies are costly and hard to achieve Hence a mixer is a

good technique as it only convert the signal to a lower frequency in which good

selectivity and high gain can be more effortlessly realised[14] A good mixer diode is the

one that can produce a high cut-off frequency and reduce conversion losses (Lc) A

mixer and detector diode with a low driven input power result in reducing overall noise

figure and thus in the ideal case the fIF amplifier also should have a low noise figure for

better performance The advantage of the heterodyne method is that it has a higher

179

sensitivity compared to the direct detection method this is achieved by producing an fIF

which has a lower frequency than the incoming RF signal[14] Obviously a zero-bias

voltage diode is more favourable to be used in mixer and detector applications

Figure 64 The mixing process where the signals are processed by the non-linear I-V characteristic

to produce an intermediate frequency (fIF=fLO-fRF) when signals at fLO and fRF are applied to the

diode

To have good detection efficiency for both types of detectors the operating

frequency (fO) must be several times smaller than fC In the case of an incoming

maximum modulated signal fM the frequency that can be acquired is in the range of fO

plusmnfM and will normally come with noise The standard method that is used to reduce the

noise is using a filter of bandwidth about 2fM at the centre of fO with the condition that fO

must be smaller than fM (f0lefM) Otherwise it would be difficult to attain However in

most cases fM is smaller than fO and fO is smaller than fC thus this will make the video

impedance (RV) (or nonlinear impedance) very close to the differential resistance of the

diode (at fO) in the equivalent circuit[74]

63 Definition of Characteristics of Interest Detection Parameters in small signal

analysis

Theoretically a transfer function measurement is preferable prior to any empirical

modelling since no assumption can be made due to the detector non-linearity

Furthermore measuring voltage output at high frequency can be very low while

measuring the power incident on the detector is hard to achieve where the linearity of

180

the typical power meter is normally less than 3 over its operating range[126]

Therefore the modelling of detector output voltage vs input power (ie transfer

function) can help to determine both nonlinearity correction and appropriate operating

range for the detector itself

The performance of the diode that is often taken into account is the transfer function

(output voltage Vout versus incident power Pin) and the main parameters that is used to

characterise and determine the quality of any detector diode are the voltage sensitivity

(βV) tangential sensitivity (TSS) dynamic range (P1dB) ie under 1 dB roll off power and

variation of output voltage when examined in extreme temperature situation (ΔV(T))

The voltage sensitivity in small signal analysis can use the approach introduced by

Torrey and Whitmer [9] then βV can be expressed as

120573119881(119894119889119890119886119897) =

119877119895119877119871120581

2(119877119881 + 119877119871) (1 +119877119904

119877119895)

2

[1 +1205962119862119895

2119877119878119877119895

1 + 119877119904 119877119895frasl]

(61)

Where ω is the angular frequency (2πf) Cj is the junction capacitance of the diode

active region RL is the load resistance RV is the video impedance taken from the

expression of Rj + Rs and κ is the curvature efficiency that give small signal rectifying

action of the ASPAT diode which is given by the second order term and itrsquos expression

is

κ =

11988921198681198891198812frasl

119889119868119889119881frasl

(62)

The curvature coefficient or responsivity (κ) is translated directly from the non-linearity

of the IV characteristics of the ASPAT diode for detector application Both RV and κ are

the parameters that can be extracted directly from diode DC measurement as discussed

in Chapter 5

The voltage sensitivity is actually a quantitative relationship between input

power and detector response Meaning that it is a change in signal output over change in

input power Normally output power is measured in Volts and input signal is measured

in Watts Therefore the unit of responsivity is VW[127 128]

181

TSS is referred to the lowest or minimum signal that the detector could detect it is

determined by the diodersquos βV and total noise available in the system (from the diode and

any amplifiers in the detector circuit) For any diode with fO=10 GHz and low noise of

1MHz bandwidth amplifier the TSS is typically less than -55 dBm[74] The TSS equation

is given by

119879119878119878 =

radic[4119884119870119861119879119861(119865119886 + 119905 minus 1)]

119872

(63)

Where M is a figure of merits and is derived based on the expression M= 120573119881 radic119877119881frasl t

denotes the diode noise temperature Fa is the noise figure B T and Y are the amplifierrsquos

bandwidth temperature and power for signal-to-noise ratio respectively For a low-level

video detector ie lt10GHz the sensitivity mainly depends upon three factors firstly on

RF matching structure secondly on the rectification efficiency output impedance and

noise properties of the diode and finally the input impedance bandwidth and noise

properties of the video amplifier at the detector output The RF matching structure

controls the quantity of overall energy at the active junction for rectification The second

factor controls the reaction of the diode to incident microwave radiation and the last

factor will influence the detector sensitivity in general[109]

In practical the Tss is a direct measure of the signal-to-noise ratio of a detector and

is achieved by varying the amplitude of the input pulse (RF signal) until a point in which

the top of the noise level with no signal applied is at the same level of noise at the

bottom level of RF signal It is commonly measured on an oscilloscope as depicted in

Figure 65 below It is defined as the input power at which a signal to noise ratio of 251

is produced[109]

Figure 65 Measurement of Tangential Sensitivity[108 129]

182

The transfer function in many detector diodes is often divided into three sections

Firstly at low incident power secondly at higher input power and finally at very high

power static (continuous) In the first region the detector diode performs as a square-law

detector in which Vout is proportional to Pin This region normally is used to extract the

dynamic range of the diode detector In the second region Vout is approximately

proportional to Vin and this region is known as the linear regime Finally at higher Pin

still the transfer function or response rolls off and thus Vout ultimately become saturated

This roll-off point where Vout has dropped by 1dB below an extrapolation of the

dependence at low Pin is termed the ldquo1dB roll-off pointrdquo and this value is usually in the

range of -11 to 12 dBm[15] Therefore a dynamic range of the detector diode can be

obtained by taking the interval between TSS and 1dB roll-off point (in dBm)

Finally the temperature dependence of Vout for a detector is normally taken from

two extreme points of the temperature (-40C˚ to +80C˚) and thus can be determined

from

Δ119881(119879) = 10 11989711990011989210 |

119881(1198791)

119881(1198792)|

(64)

This Vout variation between -40C˚ and +80C˚ normally expressed in dB

64 Noise Consideration in a Detector diode

The existence of noise in a system limits the accuracy of device performance and

the precision of measurements In a detector system specifically using a diode the noise

which can reduce the sensitivity of signal encryption is called the Noise Equivalent

Power (NEP) By definition the NEP is a noise power density over the detection

sensitivity and it can be exploited to determine the overall noise performance of a

detector[130 131] In other words NEP is defined as the power from the input source

(Pin) that is required to supply a voltage output (Vout) equal to the root means square

noise at Vout [132] For an ideal lossless match and assuming only Johnson-Nyquist is

present the NEP of a zero-bias detector can be expressed as

119873119864119875119900119901119905 = radic4119896119879119861119877119881120573119900119901119905 (65)

183

Where βopt is responsivity with an optimum match which is given by1 2frasl 119877119895120581 This type

of noise appears when changing voltages across a diode and a noise voltage (Vn)

normally will arise Theoretically the NEP has units of Watts (as it is actually a power)

but it often normalized to 1Hz as it is independent of bandwidth and thus the unit

becomes WHz12

Additionally there are also several noise sources that contribute to Vn in a

semiconductor diode which are Johnson-Nyquist noise Flicker Noise and shot noise

Johnson-Nyquist noise [133 134] appears across any conductor or semiconductor at

thermal equilibrium this is due to the thermal agitation of the carriers or charges It can

be expressed in root mean square voltage as below

119881119869minus119873 = radic4119896119879119861119877119895

(66)

Where Rj is the differential intrinsic resistance B denotes the post-detection bandwidth

T is the device temperature and k is the Boltzmann constant[134]

The second noise that is taken into consideration when dealing with semiconductor

devices is Flicker noise more commonly known as 1f noise It is a group of known and

unknown noise sources that can be observed in the frequency spectrum and normally

display an opposite to the frequency power density curve[135] It comes from a variety

of different causes ie recombination effects at a defect in semiconductor mobility

fluctuation and flow of direct current as well as interface phenomena [136-138] In term

of voltage source Flicker noise can be expressed as[139]

1198811119891 = 119870119891119881119909119891119910 (67)

Where Kf denotes a device-specific constant V is the voltage and f is the frequency The

value of x and y typically used are 2 and -1 respectively This type of noise (1f noise)

will be neglected at frequencies high enough due to the fact that the NEP of the diode is

proportional to the thermal noise and resistance of the diode

Finally the noise that causes time-dependent fluctuation in a flow of electrical

current because of the carrier or electron charge crossing a potential barrier is called shot

noise Shot noise is due to the randomness in the diffusion and recombination of both

majority and minority carriers[140] The equation of shot noise term at random time is

given by[141]

184

119881119878ℎ119900119905 = 2119902119868119861 (68)

Where q is the electron charge I is the current and B is the bandwidth This type of noise

is not affected by changes in temperature or device parameters Therefore the total noise

voltage appearing in the semiconductor is found to be [141]

1198811198992 = 119881119869minus119873

2 + 11988111198912 + 119881119904ℎ119900119905

2 (69)

However a zero-bias device will greatly eliminate both shot and flicker noise

compared to a biased device This has been explained by Equation (66) and Equation

(67) above where both noises are significantly related to the current and voltage Thus

if V and I =0 in the nonappearance of incident power then Vn will also become zero For

a detection process with bias the diode will be self-biased by ΔV which causes both

flicker and shot noise to appear But the shot noise in practical situation is much smaller

and thus normally ignored [142-144] Usually the noise in a zero-bias detector is

estimated by considering the low power limit as good first order estimation in which the

presence of only Johnson-Nyquist noise and ΔV is arbitrarily low [11 145 146]

Additionally it has been reported that the noise in tunnelling type diodes displays very

low or no excess noise in the bias region of the current-voltage characteristic[147 148]

Therefore in general most of the noise specifically in tunnelling type of diode will be

neglected this is a great advantage compared to SBD or transistors

65 Modelling of a 100GHz Zero-biased ASPAT Detector

Once the DC and RF characteristics of the ASPAT diode had been accurately

obtained the next step is to model and design a detector circuit based on S-parameter

measurement results as was explained in the previous chapter The aim is to realise a

detector circuit design which can be operated at millimetre and sub-millimetre wave

regions from an accurate diode model prior to the circuit design A diode detector model

puts experimental observations into context and offers insight into future experiment

results Consequently an electrical model based on lumped element component is vital

for a deeper understanding of how and to what extent a new device like the ASPAT

diode can affect all the key detector parameters that were previously discussed The

prediction of the detector parameters mostly depends upon the ASPAT diode

185

geometrical emitter size and material parameters However for millimetre wave

operating frequency the accuracy of the model is more sensitive not only to diode size

but also to the diode periphery ie substrate as well as coplanar amp transmission line

adopted in the circuit Therefore both extracted intrinsic and parasitic element of such a

device must be taken into account

In this work an ASPAT diode with an emitter size of 4times4microm2 which is the

smallest size that could be fabricated so far was chosen to be exploited for detector

designs The important parameters related to the 4times4microm2 GaAsAlAs ASPAT which

works at 0V is summarize in table 61 below

Table 61 A summary of all the important parameters of the 4x4 microm2 diode

Device Rj(Ω) Rs(Ω) Cj(fF) Cp(fF) κ(V-1) Intrinsic_fcut-off(GHz)

4times4microm2 90K 11 21 15 23 629

The actual measured I-V characteristic is used to model the diode since the library

in the ADS simulation tool does not have an ASPAT diode model or any tunnelling

diode for that matter The procedure of realizing the diode model is by taking the I-V

characteristic obtained from the 4times4microm2 emitter size measurement results and

converting it into a10th

orders polynomial equation via MATLAB software to create a

virtual I-V characteristic Thereafter this equation is then defined as a two terminals

device namely Symbolically-defined Device (SDD1P) ie a component of the non-

linear equation provided by ADS (Figure 66) to represent the ASPAT Figure 67

shows the measured data and 10th

orders polynomial equation fit very well to each other

Hence this new component used to represent the whole ASPAT diode will be used in

this research for MMIC detector and Frequency Multiplier designs The device chosen to

be modelled (4times4microm2) has measured junction capacitance of 21fF (at 0V at 40GHz) The

detector circuit is designed to operate at 100GHz for a safe side due to the extrinsic

calculated fcut-off is around 380GHz Since there are a lot of advantages in using unbiased

detectors compared to biased one this work will discuss the performance of millimetre-

wave detector at zero-bias and their result will be compared to the current performance

of other diodes reported in the literature

186

Figure 66 The configuration circuit to verify 10th order polynomial equation extracted from

MATLAB to realize a virtual GaAsAlAs ASPAT diode

Figure 67 Verification of actual (blue measured) and virtual (red_10th order polynomial) I-V

characteristic of the 4times4 microm2 diode used in this study

To realize the ASPAT detector circuit a simple detector circuit topology as

depicted in Figure 68 was constructed Initial simulation was run to perform a

functionality check of the detector circuit utilizing the Harmonic Balance (HB)

simulation tool embedded in that particular software Such simulation tools will analyse

the detector performances in the frequency-domain as it is mostly beneficial and fully

compatible with microwave and millimetre wave problems The frequency domain is

also suitable for single and multi-tone power excitation The importance of harmonic

balance are described in [149]

-00005

0

00005

0001

00015

0002

00025

0003

-3 -2 -1 0 1 2 3

Cu

rre

nt

(A)

Voltage (V)

ADS

4x4

187

Figure 68 Direct detector circuit topology using an ASPAT diode

Initially the circuit topology that consists of P1_Tone power supply ASPAT

bypass capacitor and load resistance is simulated by setting up a fixed input frequency at

100GHz The ASPAT diode provides a DC output voltage proportional to the input

power strength depending on the absolute values of the DC terms associated with the

nonlinearity of the I-V characteristics The capacitance in the output part is a bypass

capacitor used to prevent millimetre-waves from leaking to the output The load

resistance is large enough to ensure the voltage divider between load impedance and

device impedance gives maximum voltage sensitivity This large load resistance is

achieved by creating an open circuit at the end of detector circuit terminals

Noted that this simulation was run using diode parameters that were extracted from on-

wafer one port S-Parameter measurements as described in chapter 5 To apply them in a

two port application ie detector circuit may or may not provide a very accurate

outcome it however worked adequately in the particular circuit described in this chapter

but may not work in other circuits in general Thus the one port extractions in this work

still provide adequate parameters to build and design specific MMIC detector circuits

but not in general applications

The main reason for these simulations and their results to be used in high frequency

applications is due to the fact that actual RF measurement were done up to 40GHz

Additionally the 100GHz operating frequency was obtained from extrapolation of each

ASPATrsquos component Since the on-wafer measurement that were carried out were

limited to one port characterization applying them to two port network applications may

188

have extra consequences which are unknown Therefore actual MMIC detectors are

needed to be built and test to validate this work

To find out what power the 4times4microm2 ASPAT diode can withstand the input power

is varied from -40dBm to 10dBm via control by the P1_tone As can be seen in Figure

69 the diode starts to saturate when the received input power is about -8 dBm Above

this power limit both output voltage and sensitivity drop dramatically

Figure 69 Output voltage and detector sensitivity over wide range of input power

This diode detector circuit can thus operate adequately at given input powers from -30

dBm to -8 dBm with a sensitivity of 950VW However for the best possible sensitivity

over a range of input frequencies only one optimized input power needs to be chosen

The parameters that directly influence the voltage sensitivity are the curvature

coefficients load resistance and video resistance as can be seen from Equation (61)

Therefore in the following simulation the values RL will be optimized according to the

diodes optimum input power with regards to the highest possible voltage sensitivity

Consequently five values of load resistance were chosen from few ohms to infinity

and with the same applied input power as depicted in Figure 610 For most load

resistors the sensitivity is constant at low input power and drop at the diode saturation

region (-8dBm and greater) However for an RL value of 100KΩ and below the loaded

voltage sensitivity shows a peak near 0dBm input power which corresponds to the

maximum slope of the ASPAT detector transfer function The highest voltage sensitivity

is obtained by using an Open circuit load impedance as shown in the graph The Open

circuit load impedance gives the highest voltage sensitivity due to the voltage divider

189

between source and load impedance therefore RL must be at least 5-10 times larger than

Rj to give a better sensitivity

Figure 610 Simulated dependence of Voltage Sensitivity on incident power and load resistance of

the ASPAT detector

Furthermore the value of voltage sensitivity (βV) depends on the junction

resistance of the diode and thus the large value of Rj of the ASPAT diode yields the high

βV observed Rj which was taken from the non-linear measured IV characteristic is a

voltage dependent parameter [142]and is inversely proportional to the forward bias

voltage as depicted in Figure 611 below

Figure 611 Junction resistance as a function of forward voltage

1

10

100

1000

-40 -30 -20 -10 0 10

Vo

lta

ge

Sen

siti

vit

y (

VW

)

Incident Power Pin (dBm)

Infinity

1KΩ

10KΩ

1MΩ

100Ω

100KΩ

0

20

40

60

80

100

0 001 002 003 004 005 006 007 008 009 01

Ju

nct

ion

Res

ista

nce

(k

Ω)

Bias Voltage (V)

190

In fact the expression of (120597119881120597119868frasl ) contribute to the video impedance expression via the

expression RV=Rj+RS (nonlinear resistance) As RS is very small compared to Rj thus it

was ignored when calculating RV Although the large value of Rj will increase the

voltage sensitivity it will also make matching difficult to achieve Therefore there will

be a compromise between the size of the matching circuit and voltage sensitivity to

attain the correct value of Rj Additionally a very high Rj (around ~1MΩ) will also

increase the detectorrsquos noise equivalent power (NEP in Eq(65)) Thus an average value

of RJ typically around 100kΩ is satisfactory[150] By having a large value of RJ one can

benefit from a low input power to drive the diode into the non-linear region and thus the

detector can work at very low RF input power The NEP for GaAsAlAs ASPAT diode

is then calculated based on parameters obtained from this simulation at room

temperature The values are compared to other diode detector available in the literature

as shown in Table 62 below

Table 62 Noise Equivalent Power (NEP) for zero bias detector diode

Device NEP (pWHz12) RJ(KΩ) Frequency(GHz)

ASPAT (4times4 microm2) 188 92 100

Tunnel Diode (08times08 microm2)[150] 370 26 220-330

Zero bias SBD[11] 15 3 150

Sb-Heterojunction Backward Diode[145] 024 32 94

VDI Zero bias SBD[151] 2 18 110

From the Table 62 above the NEP of the ASPAT is calculated based on RF input power

which is required to obtain an output signal-to-noise ratio of unity in a 1Hz at detector

output[142] and also the assumption of only Johnson-Nyquist (thermal noise) is

dominant for small incident power (-25dBm)[152 153] The prediction of NEP for the

ASPAT is comparable to the VDI Zero bias detector since the value of their junction

resistance is much lower than that of the ASPAT Therefore it is very important to

obtain a reasonable value of junction resistance From this it is clear that a trade-off of

high voltage sensitivity and low junction resistance is best to obtain low noise

191

Finally the curvature coefficient at specific operating voltages also influences

the voltage sensitivity of the detector Figure 612 shows the calculated curvature

coefficient of the 4times4microm2 ASPAT diode used in this work

Figure 612 Curvature coefficient for the measured I-V characteristic with a device size of 4times4μm2

The high zero-bias curvature (23V) is reached from the mutual effect of the intra-band

tunnelling in the GaAs-AlAs-GaAs and the highly doped GaAs at the anode and cathode

(Rs amp Rj) This was shown in the numerical simulation in[154] where the combined

effects of the optimum anode AlAs composition increases the curvature coefficients by

thinning the energy tunnelling window (intraband tunnelling process)

Other approaches that can lead to a large curvature is using smaller device area

with minimum series resistance [155] as was also discussed in Section 52 Having a

better curvature coefficient leads to increased voltage sensitivity as seen from Equation

(61) above and βV is also proportional to κ [111] The curvature coefficient calculated

using the above formula is nearly 23V at 1mV peak and high voltage sensitivity can be

achieved by having large value of curvature coefficient But one needs to remember that

this will also decrease with increasing input power because RJ which will also decrease

Therefore for a safe operating region the incident power that can be applied through the

diode is in between -30dBm to -8dBm for a 100GHz operating frequency

In this simulation -25dBm is chosen to simulate the GaAsAlAs ASPTAT diode

detector working at 100GHz input frequency By fixing the P1_Tone to this input power

-5

0

5

10

15

20

25

30

-001 001 003 005 007 009

Cu

rva

ture

Co

effi

cien

t(V

-1)

Voltage (V)

k(4x4 um^2)

192

the frequency is varied from 90GHz to 110GHz and the results are depicted in Figure

613 below

Figure 613 Voltage Sensitivity against frequency at -20dBm fixed input power

Noticeably the voltage sensitivity of the diode detector decreases linearly with

increasing input frequency At 100GHz a sensitivity of around 540VW is obtained

However the sensitivity in this case is roughly estimated from the Equation (61) above

and will not be sufficiently accurate because of the effect of other key factors such as

reflective power which was not included This parameter must be taken into the account

due the P_1Tone power source which provides a 50 Ω impedance source (Zin) which

does not match the load impedance (ZL) which consists of JωL 1JωC and Z and which

mainly comes from the ASPAT diode itself In order to determine the load impedance at

the diode a typical ohmrsquos law (Z=VI) equation must be used at the input side of the

diode with regards to the applied frequency (100GHz)

As a result the total load impedance obtained from the simulation is 11055-

j69057Ω which is clearly not matched to the 50Ω impedance source The mismatch

between source and load leads the available power from the source to be not fully

delivered to the load and hence there is loss of power leading to a lower detector

sensitivity Therefore actual calculation must take into the account the reflection impact

as expressed in the equation below

193

120573119881(119886119888119905119906119886119897) =

119877119869119877119871120581(1 minus |Γ|2)

2(119877119881 + 119877119871) (1 +119877119904

119877119895)2

[1 +1205962119862119895

2119877119878119877119895

1 + 119877119904 119877119895frasl]

(610)

Where the term (1-|Г|2) refers to the normalized power absorption by the ASPAT diode

and Г is the reflection coefficient due to discrepancy between 50Ω input impedance (Zin)

of the input port and the diode Consequently the calculation of reflection coefficient

(eq611) is carried out by using this expression and the result is shown in Figure 614

Γ =

119885119871 minus 119885119894119899

119885119871 + 119885119894119899

(611)

Figure 614 Reflection Coefficient versus operating frequency without matching circuitry

As can be seen in Figure 614 without the matching circuit the S11 is decrease

linearly with frequency but only very slowly This means that most of the RF power

transmitted from the source is reflected back (by that ratio) when it went through the

diode Therefore in order to resolve the mismatch a matching circuit is introduced in

between the source and load of the detector circuit as shown in Figure 615 (red

rectangular) This matching circuit works by transforming the load impedance into an

impedance that is identical to the source or input impedance Note that for any

impedance matching circuit the main purpose is usually to obtain maximum power

transfer to the load however in some cases (ie oscillators) the matching circuit is to

achieve a lower noise figure Hence in a broad sense the introduction of the matching

194

circuit in a detector circuit can be defined as a circuit that convert available impedance

into wanted impedance by obeying the maximum power transfer theorem [156]

Figure 615 Detector circuit with impedance matching circuit placed in between diode and source

There are many type of matching circuit that can be used to achieve both

objectives above such as circuits using lumped element transmission-line-impedance

matching circuit single and double-stub tuners as well as a quarter-wavelength In this

design the technique used to match source and load impedance is the single open and

short stub (in red rectangular) as it is simple convenient and very efficient in ADS

simulation The important parameter that needs to be tuned in both stubs is the electrical

length (E) at any designed frequency ie 100GHz in this case The E tuning is realized

by using the Smith Chart features available in the ADS simulation tools Figure 616

shows the reflection coefficient with matching circuit modelled over a broad frequency

band and it is clearly shown that at the desired operating frequency the reflection is very

low Note that it is difficult to obtain wide frequency band matching

Matching circuit

195

Figure 616 Reflection Coefficient over wide frequency band with matching

The simulation is continued to find the effect of the matching circuit placed in

the detector circuit on the voltage sensitivity The same input power (-25dBm) is applied

to the diode and the voltage sensitivity is plotted against frequency as depicted in Figure

617 Obviously at the desired operating frequency (100GHz) the sensitivity rises up to

a maximum value of 2100VW with this value obtained without any reflection and the

input port being completely matched with the ASPAT diode model used in this work

Figure 617 Simulated voltage sensitivity of GaAsAlAs ASPAT detector at W-band frequency

Once the matching circuit to be used for 100GHz operating frequency was confirmed

further simulations were made by applying a series of low input power to find the

tangential sensitivity (Tss) of the detector diode When determining the Tss it is very

important to include the matching circuit as it will minimize any power losses through

the diode thus a very small input power can be detected

196

Figure 618 Lowest detectable signal at 100GHz operating frequency

The transfer function depicted in Figure 618 shows incident power of -80 dBm

to -50 dBm applied and the lowest detectable signal that can be obtained with 4times4microm2

mesa size ASPAT diode is around 138microV at -68dBm Although a typical value of Tss is

normally not more than ~-55dBm as in ref [74] the lower value obtained in the

simulation is because the device is operated at zero bias operation and does not use any

amplifier therefore the noise and values related to amplifier as in Equation 63 have been

neglected Even though the TSS appeared very low it is most likely very dependant to

the 10th

order polynomial equation embedded into SDD in ADS software Therefore in

near future the TSS value has to be determined in real fabricated MMIC ASPAT detector

As discusses in Section 63 other important parameter that can be extracted from

the diode transfer function is the Dynamic Range of the diode It can be obtained in a

region called square law region which is a region where the Vout of the ASPAT diode is

proportional to the square law of the input power signal From Figure 619 the square

law region is in between -68dBm and -12dBm and the linear region or in this case

saturated region is above -12dBm Taking to the account the roll-off point where Vout

has dropped 1dB below the extrapolation of the dependence at low input power and

therefore the dynamic range of the detector diode can be obtained by taking the interval

between TSS and the 1dB roll-off point (in dBm) which is ~55dBm

197

Figure 619 The dynamic range of the ASPAT diode obtained from linear region of diode operation

The figure of Merit (M) of the detector (ie equation 63) is 2100 (90K) 12

where 2100 is the sensitivity and 90K is the value of RJ at zero bias which is equivalent

to 652 W The M value should be large however in this case due to RJ being very

large it has dropped tremendously when compared to the voltage sensitivity obtained in

this simulation The results obtained indicate a reasonably successful design of the

MMIC detector using the 4times4microm2

emitter size GaAsAlAs ASPAT diode The results

obtained lead to the design of other MMIC detector using the other fabricated diode

sizes (6times6microm2 and 10times10microm

2) A Similar procedure to the one used for the 4times4microm

2

diode was followed The only difference was the use a slight higher input power (-

20dBm) than in the 4times4microm2 design Hence the simulation results obtained are then

compiled and compared as depicted in Figure 620

Saturated

Region

198

Figure 620 Simulation result of diode sensitivity for three devices geometries obtained from the

fabricated ASPAT in this work

The graph plotted for each dimension was taken after matching circuits were included

As can be seen in Figure 620 the highest sensitivity is achieved using the smallest

device size as this has the highest cut off frequency Table 63 below summarises the

performances of the 100GHz ASPAT detectors obtained from the simulation in this

work

Table 63 Summary of characteristic of interest for GaAsAlAs ASPAT detector

Device Tss(dBm) fcut-

off(GHz)

Bv

(VW)

Dynamic range dB(dBm) M(W-12

)

4times4microm2 -68 380 2100 55 65

6times6microm2 -50 151 1445 48 45

10times10microm2 -40 98 247 40 21

From table 63 above it is clear that a lower cut-off frequency will affect the

voltage sensitivity The dynamic range between each ASPAT is different because larger

size area will allow more power to go through the diode as a result of high current that

such a device can handle before reaching the saturation their lowest detectable is higher

Small diode size will detect lowest voltage but cannot handle high power On the other

hand a large diode size is able to receive high power however can only offer lower

voltage output Therefore there is a trade-off between small diode size and receiving

input power which will directly affect Tss and 1dB roll off Once again all the

100

1000

10000

90 95 100 105 110

Vo

lta

ge

Sen

siti

vit

y (

VW

)

Frequency (GHz)

10times10microm^2

6times6microm^2

4times4microm^2

199

parameters obtained in this simulation are just estimation from the 10th

order polynomial

equation thus real ASPAT detector has to be fabricated for verification

The best device performance among all GaAsAlAs ASPAT diodes was obtained

with the 4times4microm2 mesa area size diode which was compared to other exiting millimetre

wave detector diode available in the literature Since the 100GHz is located in the W-

band spectrum frequency therefore the comparison will be performed in this frequency

band but with low input power The parameters for the-state-of-the-art zero bias

detectors are gathered in Table 64 below

Table 64 Comparison 4times4microm2 GaAsAlAs ASPAT diode to the-state-of-the-art zero bias detector

at W-band (75GHz-110GHz)

Device Size(microm2) Tss(dBm) βv (VW) Pin (dBm) f (GHz)

GaAsAlAs ASPAT 4times4 -68 2100 -25 100

GaAs SBD HSCH-

9161[157]

- -49 2200 94

HBD[158] 15x15 2540 -20 95

Planar SBD[159] - -68 2100 -25 100

Note that the ASPAT diode retains its favourable temperature stable characteristics

which are not the case for all the diodes used for comparison in Table 64

66 Conclusions

In this chapter all theory regarding RF detection using diodes ie parameters of

interest noise consideration etc have been discussed The aim to design and develop a

low cost reliable and sophisticated zero bias 100GHz detector circuit was achieved

through exploitation of a 4times4microm2 GaAsAlAs ASPAT diode The design was performed

with the aid of Keysight ADS modelling software utilizing harmonic balance simulation

The effect of load resistance junction resistor to the detector voltage sensitivity was also

discussed in details The 90KΩ Rj value and open circuit load resistance was chosen for

high sensitivity

200

A step by step design of a W-band ASPAT detector was presented The effect of

matching circuit was discussed in detail and where an unmatched sensitivity of 843VW

is obtained which then increases to 2100VW after matching Through RF

characterization simulation a detection at 100GHz (W-band) was successfully achieved

with a relatively large device mesa area (4times4microm2) at an input power of -25dBm

(8microWatt) leading to a 2100VW voltage sensitivity a -68dBm TSS and 55dBm dynamic

range All these values are comparable to others fabricated diodes in the literature

The zero bias ASPAT detectors based on the GaAsAlAs material system in this

work are still at an early stage of development a lot of work is still required to realize

high yielding integrated millimeter and sub-millimeter wave (MMIC) detector circuits

However as this work is on-going at Manchester it is expected that fabricated ASPAT

MMICs with even higher voltage sensitivity will be fabricated in the near future through

collaborating bodies involved in this research especially the University of Cambridge

and ICS Limited

201

7 FREQUENCY MULTIPLIERS (DOUBLER) DESIGN USING

GAASALAS ASPAT DIODES

71 Introduction

Originally the key application for the ASPAT diode was for use as microwave and

millimetre wave detectors[18] This is due to the fact that such diode demonstrates

strong non-linearity low noise and high cut-off frequency features as described in the

previous chapters However these features are not only beneficial for detection purposes

but also allow them to be used and designed as microwave and millimetre wave sources

The only way to generateenhance continuous wave (CW) power using a non-linear

device is through frequency multiplication techniques It is known that the frequency

multiplier is the alternative approach (to 3 terminal transistors) using non-linear devices

that are used to generate high frequency low phase noise signals Any high quality low

frequency signal that goes through a frequency multiplier circuit can be generated to any

desired high output frequency[160] Therefore the main objective of this chapter is to

demonstrate the feasibility of the ASPAT diode as a compact source of microwave and

millimeter-wave receiver for imaging applications[161]

The study of the ASPAT diode as a power source begins with a brief explanation

of the importance of a frequency source and the lack of compact device and technologies

at high-frequency signals The state-of-the-art for frequency multiplier will also be

discussed In the next section (Section 74) the fundamentals of the frequency multiplier

architecture ie the principle of operation and appropriate devices will be presented

Since this is the first attempt at using GaAsAlAs ASPAT diodes a simple multiplier

circuit design and topology was built This will be discussed in detail in the subsequent

sections where simulation results are discussed The focus of the discussion will be to

demonstrate the possibility of a GaAsAlAs ASPAT diode functioning as a frequency

multiplier and comparison with other state-of-the-art varistor mode frequency

multipliers

202

72 Motivation and Background

Typically continuous wave (CW) sources generating below 100 GHz can be

obtained through oscillators amplifiers and pin diode comb generator Below 10THz

the sources can be made from RTD IMPATT diodes and Gunn oscillators and above 10

THz it is commonly done by photonic mixing quantum cascade laser (QCL) and gas

lasers [162] Both types of sources and their performance are plotted in Figure 71

However these conventional ways of generating millimetre and sub-millimetre waves

have their own limitations ie high cost complexity and sometimes requirement for

cryonic cooling The most effective way to tackle the limitations of conventional mm-

wave and THz sources is by implementing frequency multiplication technique using

solid state nonlinear diodes [163-165] such as SBD and ASPAT diodes

Figure 71 performance of state-of the-art millimetre wave source [166]

Twenty years ago there were only two types of diodes (SBD and P-N junction

diodes) often used for frequency multiplication To date besides the SBD there are

many types of diode that have been used as frequency multipliers These include the

high electron mobility varactor (HEMV) single barrier varactor (SBV) [167] and hetero-

structure barrier varactor (HBV) (270GHz with 90mW input power and Conversion

Efficiency of 72)[168 169] Other variants that have developed to enhance the

frequency multiplier performance of the classic SBD [170] include the Barrier-intrinsic-

203

n+ (BIN)diode and Barrier N-layer N+ (BNN) diode [171] Other diodes for use in such

applications are the planar doped barrier diode (PDB) Resonant tunnelling diode (RTD)

amp it families ie Quantum well diode (QWD) and step recovery diode which is a

modification of the P-N junction diode

Although three terminal devices ie FET GaAs MESFET and HEMT had

shown better performance and are capable of achieving greater efficiency and

bandwidth as well as having additional conversion gain features [160] two terminal

devices (ie varactor diodes) which are passive multiplier are still preferred This is due

to their simplicity and most importantly their ability to generate very little noise Among

these types of device technologies the SBD is preferable as it is mature and has been

shown to be very suitable for high-frequency applications [172 173]

The ASPAT diode is exploited to investigate the possibility and the feasibility of

generating microwave and millimetre-wave power through well-known frequency

multiplication methods The utilisation of the ASPAT in frequency multiplication will

also aid in generating local oscillator sources which are critical components in

heterodyne receivers The ASPAT diode will work in resistive I-V mode (varistor mode)

and has features to work also at zero bias condition thus offering low power handling

than traditional high-efficiency varactor diode since the varactor diode requires a large

reverse bias supply of several tens of volts

73 Frequency Multiplier Architecture the Basics

In principle a frequency multiplier is an electronic circuit that gives an output

frequency that is a multiple integer of its input frequency signal pumped from a local

oscillator as depicted in Figure 72 The ability to generate any desired multiple output

signals is realised by a nonlinear device ie diode or transistor Such devices though

also can give distortion or cause sudden change to the input frequency Additionally

these devices generate multiples of the input frequency (fout) The distortion of the

sinusoidal signal refers to an abruptsudden change versus amplitude or time which thus

generates higher frequency with lower amplitudes of the input signal Usually a

frequency multiplier circuit will include a bandpass filter to select the desired harmonic

204

frequencies and deselect undesired harmonic frequencies especially fundamental

harmonic at the output for further processing

Any non-linear device either in symmetricalantisymmetric current-voltage or

capacitance-voltage can be utilised to realise a frequency multiplier source [168 174]

Figure 73 describes the method where a nonlinear resistance is utilised to convert a

harmonic input signal into periodic output signal containing components at multiples of

the input frequency Both non-linear resistance and reactance characteristics can be

extended into power series methods

Figure 73 Principle of operation for frequency multiplier utilising a non-linear resistance [10]

The operating principle of the frequency multiplier is shown in Figure 73 where

the I-V curve converts a harmonic frequency input into a periodic frequency output

including components at multiples of the input frequency The non-linear I-V

characteristic can be explained in term of a power series at the operating fixed point of

bias voltage (VB) [174]

Frequency

Multiplier Circuit

finput

foutput

= nfinput

Figure 72 Frequency multiplier has the property that foutput is an integer multiple of fin

205

119868(119881119861 + ∆119881) = 1198860 + 1198861∆119881 + 1198862∆1198812 + 1198863∆1198813 helliphellip (71)

For a given input voltage as below

∆119881 = 119881119904 cos120596119878119905 (72)

The input signal harmonics will become

119868(119905) = 1198680 + 1198681 cos120596119904119905 + 1198682 cos 2120596119904119905 + 1198683 cos 3120596119904119905 (73)

Where t and ω are the time and angular frequency respectively based on equation (73)

the output contains both signal source and harmonics Therefore a complete frequency

multiplier circuit has to have non-linear device and filter to allow the selection of any

frequency components needed

731 Types of frequency multipliers

Frequency multipliers can be classified into passive and active multipliers This

classification is based on the ability of the frequency multiplier to yield any conversion

gainlosses The passive multiplier is the one that only produces conversion losses In

other words it can be described as a multiplier that generates an output power level

lower than the excitation input power and it is mostly dominated by passive nonlinear

devices ie Diodes On the other hand the active multipliers refer to a device that would

produce an output signal with a power level that is greater than the input signal power

This conversion of power is termed as conversion gain These types of multipliers attract

much attention as they do not only increase the frequency at the output but also the

signal power

Passive frequency multiplier can be formed by using diodes that are classified as

being of the varistor (non-linear I-V) or varactor (non-linear C-V) type [160 174] The

varistor type will influence the frequency multiplication with a non-linear resistance or

conductance (resistive diodes) and this results in a very large potential bandwidth at the

output but poor conversion efficiency The varactor diode type where the frequency

multiplications are affected by the non-linear capacitance (reactive diode) as their

reactive element typical result is high conversion efficiency A diode that is used in this

206

application must have strong nonlinearity stable electrical characteristic repeatable and

has fast enough response to an applied frequency Therefore multipliers are classified

into Doubler Tripler quintuple and so forth depending on the highest power of output

harmonic signal

In general all varactor type diodes with such characteristics will produce high

power at odd-order harmonic oscillation if any microwave signal is pumped into them

The benefit of having odd-order in multiplier design is that it reduces the complexity of

the overall circuit ie it eliminates even-order idler frequencies [175] The varactor type

diode had been shown by Manley-Rowe to a get maximum 100 conversion efficiency

for generating an ideal harmonic [176] compared to the varistor type where the

maximum efficiency achievable ideally is 1 1198992frasl where n is the multiplication factor

(output harmonics number) [174] In the case of power handling (input excitation power)

for multipliers varactor mode diode required greater power (several milli Watt) than the

varistor mode due to the fact that reverse applied voltages are very large (many tens of

volts) Therefore these types of frequency multipliers may not be suitable for the case of

high input power excitation There no report in the literature of varistor based

multipliers working with high power excitations

74 Parameters of interest for Frequency Multipliers

The simplest way to describe an equivalent circuit for a complete frequency

multiplier is by setting a Source impedance (ZS) at the input side and load impedance

(ZL) at the output side as depicted in Figure 74 below This circuit usually has the same

properties as described in the previous chapter and most of the others two ports

networks However in this case the purpose is different and is the conversion of a sine

wave signal source (Vs) with angular frequency ωs to an output signal with frequency

nωs where n is the multiplication integer or the order of multiplication

207

Figure 74 A standard system for two port frequency multiplier circuit

Referring to Figure 74 above there are few sets of parameters for the frequency

multiplier to be taken out and compared Examples are the conversion loss maximum

input signal power Impedance at source and load Bandwidth multiplication factor or

harmonic amp subharmonic content and noise conversion properties The conversion loss

(CL) is described by the ratio of available power at source (Ps) to the output harmonic

power delivered to load resistance (PL) and is normally expressed in dB It occurs due to

the nature of passive semiconductor diodes and the electronic circuit itself that are lossy

and dissipate energy On the other hand the conversion efficiency (ηn) is a ratio of the

output power at load (PL) to the available power at the input (Ps) This is often expressed

in percentage () The conversion efficiency can be determined as

120578 =

119875119878

119875119871

(74)

while the conversion loss is expressed as[174]

119871119899[119889119861] = 10 log

119875119904

119875119871= 10 log (

|1198811199042|

4 119877119890 |119885119904||1198681198712|119877119890|119885119871|

) (75)

Where Vs is the input voltage and IL is the output current amplitude Besides this the

conversion efficiency is often referred to as the inverted value of the Ln In designing a

frequency multiplier it is crucial to minimise the conversion loss and maximise the

conversion efficiency value

To achieve a perfect multiplier with minimum conversion loss the impedance of

source and load must be at an optimum level This implies that the source impedance

Frequency

Multiplier Z

L

Zout

Z

in

V

Z

SWR Г

208

(Zs) must be very close to the complex conjugate of the multiplier input impedance

(Zin) hence minimum reflection loss will occur at the input side This can be realised by

introducing an impedance matching circuit between the diode and source The power

transfer between the source and the multiplier is quantitatively described by the value of

the multiplier input reflection coefficient (Ѓ) with source Zs assumed to represent a

reference impedance This specification also can be explained in the standing wave ratio

(SWR) Both relationships are described below respectively [174]

Γ =

119885119894119899 minus 119885119904lowast

119885119894119899 + 119885119904

(76)

119878119882119877 =

1 + |Γ|

1 minus |Γ|

(77)

Where the asterisk () represents the complex conjugate of the Zs impedance

On the other hand the situation of the load impedance is different when a standard

or an optimum value is provided by the designer This will either increase the conversion

loss or decrease the output power Thus one has to keep in mind that frequency

multipliers are non-linear devices and power transfer condition both at the input and the

output depend on each other and the input signal level[174]

75 20GHz40GHz Varistor Mode ASPAT Diode Frequency Doubler

In this work a similar ASPAT diode (4times4microm2) to that in designing millimetre

wave detector in the previous chapter is used The main objective of designing the

frequency multiplier circuit was first to investigate the performance of the ASPAT

diode as a microwave or millimetre wave signal source A design is deemed successful

when the diode physical parameters are optimised and the suitable impedance matching

network is produced for each desired harmonic as well as maximising the output power

These goals however are hard to achieve when a higher frequency operation is targeted

for use

There are many types of multiplier circuit topologies that can be implemented

using GaAsAlAs ASPAT diode in varistor mode to achieve high order of multiplication

209

Examples are single diode multiplier series or parallel connected diode multiplier anti-

parallel amp anti-series connection diode pair multiplier anti-parallel-series connected

diode multiplier and bridge frequency multiplier as well as nonlinear transmission line

frequency multiplier [174] Before designing a circuit there is one most important

consideration to make Prior to choosing any mentioned circuits to be used for frequency

doubler the design considerations are made based on the capability of ASPAT diode to

receive an optimum amount of input excitation RF power From the discussions in

Chapter 6 the ASPAT diode will reach saturation level (linear regime) at power ~

-10dBm for a device size of 4x4microm2 Once the optimum input power was confirmed the

circuit topology was carefully chosen to balance between the requirements of the

ASPAT to work at high frequencies ie low Rs and Cj amp high diode cut-off frequency

as well as the desired output signal frequency that needs to be produced

To realise the first attempt of an ASPAT diode as a signal source a simple circuit

topology of a frequency doubler was deployed as depicted in Figure 75 below The

frequency doubler circuit consists of a voltage source (can be power source) input

filtering with matching network ASPAT diode output filtering with matching network

and load impedance (ZL)

Figure 75 Block diagram of frequency doubler with series ASPAT mounted diode

In order to investigate the doubler performance the Keysight ADS simulation

tool and similar procedure to obtain accurate ASPAT model using a 10th

order

polynomial equation as in Section 65 was used The circuit in Figure 75 is translated

into ADS format as illustrated in Figure 76 Once the circuit was constructed the

analysis was performed using the Harmonics Balance (HB) simulator The circuit

requirements are matched terminations at the input and output frequencies open

Input Filtering

and matching network

Output Filtering

and matching network

Zs

Vs

ZL

Pin

fin

Pout

nfout

210

circuited terminations at the higher harmonics and optimum reactive terminations (an

inductance which resonates with the junction capacitance) at the output frequencies

Figure 76 Empirical modelling of ASPAT doubler utilising Keysight ADS tool

The circuit in Figure 76 is the simplest way of constructing a frequency doubler

circuit which consists of the signal source (P_1Tone) input matching circuit (Stubs)

filter (Short stub) ASPAT diode low pass filter output matching circuit(Stubs) and ZL

(load resistance) The utilisation of the stubs is an ideal case of simulation since in the

real fabrication stubs are normally formed in large sizes Therefore a proper design such

as using CPW instead of stubs is essential in real fabrication

Again this simulation works for this particular circuit in this chapter as all the diode

parameters were extracted from on-wafer one port S-Parameter measurement described

in chapter 5 To apply them in such two ports applications may not very accurate

however it still provides adequate parameters to build and design particular frequency

multipliers but not in general applications These simulations and results are adequate for

high frequency applications due to the fact that the actual RF measurements on the

diodes were carried out up to 40GHz and the target operating frequency in this multiplier

design does not exceed 40GHz

Since the on-wafer measurement that were carried out were limited to one port

characterization applying them to two port network applications may have extra

consequences which are unknown Therefore actual MMIC frequency multiplier is

needed to be built and test to validate this work

Input Matching

Output Matching

211

To find the optimum output power initial simulation without matching circuit

was performed This simulation was run by varying the input power from -35dBm to

20dBm but fixing the input frequency at 20GHz As can be seen in Figure 77 the lowest

point in the conversion loss (CL) and the highest point of the conversion efficiency (CE)

are obtained from an input power of -1dBm However this amount of input power is too

high for the ASPAT diode The lowest CL at -1dBm may not be accurate since it was

applied without matching Note that it is difficult to achieve a matching between source

impedance and load impedance when varying the input power Therefore a lower input

power of -10dBm is chosen for this frequency doubler operation Figure 77 shows the

Conversion Loss and Conversion Efficiency as a function of the available power of the

given input source

Figure 77 Conversion loss and conversion efficiency as a function of input power

The circuit in Figure 76 works with a -10dBm input power and 20GHz centre

frequency input signal is pumped from the power source (P1_tone) to the ASPAT diode

and distortionabrupt change of input waveform occurs at the fundamental frequency (f0

in this case 20GHz) Such abrupt change produces harmonics and these harmonics can

be classified into desired frequency component by placing two-quarter wavelength (λ4)

stubs (90˚) at both sides of the ASPAT diode At the input side of the diode short circuit

stubs are utilised to permit the f0 tone to reach the ASPAT diode and block the second

harmonics (2f0 in this case 40GHz) back to the input side and pushes it towards the

load resistance On the contrary at the output side of the diode the open circuit stubs are

used to ldquoopen circuitrdquo the 2f0 signals while ldquoshort circuitrdquo the f0 component Thus 2f0

212

signal will not be affected due to the open circuit stubs being half wave (λ2) long The

function of both stubs is basically to isolate the input and output signal from mixing each

other Therefore the design of input and output matching circuit can be achieved easily

The input matching circuit was designed based on the mentioned input frequency

(f0=20GHz) for an available input power (Ps=-10dBm) which is set up at the power

source by using two stubs with the same configuration as used in Chapter 6 Such

configuration is purposely deployed to increase the 50Ω coming from the P1_tone

source impedance to the conjugate thus reducing the reflection coefficient to the

ASPAT diode From the simulation without matching the input impedance to the diode

is 551Ω in magnitude for an available input power of -10dBm

On the output side of the diode output matching circuit is available to transform the 50Ω

port impedance in the optimum load impedance which provides minimum conversion

loss for the ASPAT diode The output matching circuit is designed based on expected

output frequency which is in this case 2f0 =40GHz Other than this optimum

impedance between load impedance and ASPAT will not be achieved thus resulting in

higher conversion loss

To ensure the proposed circuit is valid and suitable for the specific ASPAT diode

mesa size the response of conversion loss and efficiency are plotted as a function of

output frequency from 20GHz to 100GHz The results of both conversions are illustrated

in Figure 78

Figure 78 Conversion loss and Conversion efficiency as a function of output frequency

213

As can be seen in Figure 78 the conversion loss is obtained at the lowest point where

the output frequency is needed Meanwhile the conversion efficiency is maximum at the

same output frequency Therefore this indicates that the first attempt of an ASPAT

Doubler frequency source works well However the values obtained for Ln from this

simulation is 28dB which is rather high On the contrary the η achieved in this study is

very low with a value less than 1

Since the ASPAT is in varistor mode with no bias applied the conversion

efficiency is expected to be low due to resistive losses Another factor that may

contribute to lower η is the diode model itself as it is taken from a 10th

order polynomial

equation not from a diode model provided in the ADS software tools Thus some

properties of such tunnelling diode may not be included Hence it is necessary in due

course to fabricate and build such a compact frequency doubler in the future to verify

the simulation results

From the simulation point of view the less than 1 Conversion Efficiency

obtained is still good enough for a first attempt at a frequency doubler which utilises the

new ASPAT tunnelling diode The frequency doubler obtained from this work is suitable

for use in zero bias varistor modes for low power application The varistor mode doubler

performances from this simulation work are gathered and compared to other in the

literature as summarized in Table 71 below where fout is 40GHz η is 015 Ln is 28dB

and Pi is -10dBm

Table 71 Comparison parameter obtained from this simulation to the state-of-the-art multiplier

diode

Device fout (GHz) η() Ln(dB) Pi (dBm)

ASPAT 40 015 28 -10

SBD (Si)[177] 104 2 134 -10

SBD (GaAs)[178] 13 2 137 -16

214

The performance of the 2040GHz is compared to the literature based on their input

power below -8dBm since the ASPAT is only capable of working at low power To the

best of the author knowledge very few diodes operating in varistor mode at low power

excitation can be found in the field of research and industry Therefore this 2040GHz

ASPAT Doubler might a first for tunnel diodes if it can be fabricated and test at Ka band

and above

76 Conclusions

In this chapter another alternative application based on non-linear features of

GaAsAlAs ASPAT has been presented The simulation of a frequency multiplier

(doubler) was carried out utilizing the 4times4microm2 size ASPAT diode The theory and the

ability of the ASPAT diode to operate as a frequency source were explained in detail

A unique varistor mode frequency multiplier circuit topology for the 2040GHz

ASPAT doubler has been demonstrated and briefly discussed The details and step by

step simulation technique utilizing harmonic balance from Keysight ADS has been

presented Even though the conversion efficiency is very small at 015 and large

conversion loss of 28dB there is still space for improvement in term of design ie

different circuit topology optimized input and output matching circuit etc This design

can be a good reference for a doubler operating at very low power but produce high

frequencies in Ka band

215

8 CONCLUSION AND FUTURE WORK

81 Conclusion

The main focus of this research was the development of a new tunneling diode

namely the asymmetrical spacer layer tunnel (ASPAT) diode for process repeatability

manufacturability and reproducibility The broad study undertaken was to improve the

microwave performance technology by introducing a new type of tunneling diode

For years the asymmetrical spacer layer tunnel diode was unable to be manufactured

due to the high sensitivity of the tunneling current to the barrier thickness This changed

dramatically when the MBE method was carefully optimized to precisely control the

growth to sub-monolayer precisions When stability repeatability and reproducibility in

the epitaxial growth was achieved the next step was to qualify the fabrication process of

the diodes themselves thus ensuring high performance device can be delivered to the

market

For this purpose GaAsAlAs ASPAT diodes made of two different types of

substrates were grown The first batch was grown in the Riber V100HU SSMBE and

used doped substrates Samples XMBE307 and XMBE368 were successfully grown

and fabricated from that batch The DC characterization obtained from measurement

proved that this first batch had fully functional reproducible and manufacturable

devices Later a second batch using semi insulating substrates improvement in spacer

layer and doping concentration were grown This set of samples (9 x 2rdquo wafers grown

simultaneously) and denoted as XMBE304 also showed fully functional DC

characteristic and was used for RF characterization and detector integrated circuits

The conventional GaAsAlAs ASPAT diode structures grown on doped

substrates and developed previously in our lab were not suitable for high frequency RF

characterizations Therefore a major contribution of this work was to develop a new

fabrication technique for a new GaAsAlAs ASPAT structure using semi insulating

substrates to achieve repeatability manufacturability and reproducibility in term of

process flow DC characteristics and ultimately RF characteristics Apart from the

enhancement of the epitaxial layer the other important contribution of this research was

216

the optimization of the small 4times4microm2 emitter size by incorporating both vertical and

lateral structure based purely on low cost I-line optical lithography

To obtain a repeatable and manufacturable fabrication process of lateral

GaAsAlAs ASPAT structures the key issue was to solve the over etching of the

effective mesa area when qualifying the Air Bridge technique This issue caused all

semiconductors under the metal contact to be completely lost thus leaving the metal

contact hanging without connection to any bond pad area However the developments of

the Dielectric Bridge technique realized the true performance of the GaAsAlAs ASPAT

diode structures The samplersquos surface cleanliness as well as DC and RF performance

showed significant improvement when using Si3N4 as a dielectric layer Due to the

highly isotropic etching profile and thicker GaAs layer in XMBE304 samples and

although the smallest emitter size designed on the mask was 2times2microm2 only 4times4microm

2 were

reproducible and showed good uniformity in I-V characteristics

Upon successful optimization in the fabrication process flow of the small emitter

size diodes (4times4microm2 6times6microm

2 and 10times10microm

2) a good uniformity of better than 91

was obtained for DC measurement results within a tile containing over 1000 devices

This confirmed that the lateral GaAsAlAs ASPAT diode structure can only be realized

through the Dielectric Bridge technique These devices were further characterized with

S-parameter measurements and their intrinsic and extrinsic parameter values and

junction capacitances series resistances and junction resistances were extracted leading

to intrinsic cut-off frequencies of 600GHz 429GHz and 100GHz for the three device

sizes respectively

Temperature dependence measurements and simulations were also carried out in

order to confirm that the ASPAT diodersquos characteristics were temperature insensitive

showing less than 5 change in current at both extremes of temperatures 77K to 400K

By comparison the SBD I-V characteristics variations with temperature span orders of

magnitude Physical modelling agreed very well with measured data confirming good

and validated models that can also describe temperature effects

For the realization of the integrated ASPAT millimeter wave detector empirical

modelling using ADS simulation tools was carried out This was performed to predict

the detector performance at 100GHz to comply with the initial objective to develop a

217

millimeter wave detector circuit The simulations using the 4times4microm2 diode models led to

a successful 100GHz circuit design able to detect 100GHz incoming frequency with

2100VW voltage sensitivity

The first ever GaAsAlAs ASPAT diode frequency multiplier design was also

attempted A reasonably good result was obtained for a 20 to 40GHz frequency doubler

operating in varistor mode However the conversion efficiency obtained was less than

1 Further research on this is required to improve the efficiency by using other circuit

topology ie using a balun or other Co-planar waveguides Ultimately fabricating and

testing the actual multiplier circuits are essential to validate the simulation data

82 Future Work

This work has provided a foundation for a reproducible and repeatable GaAsAlAs

ASPAT (SI substrate) wafer fabrication process and recommendations for design and

simulation of ASPAT diode MMIC detectors and frequency source has also been

provided However the GaAsAlAs ASPAT diodes still remain immature and there are

many ways to improve its DC and RF performances both experimentally and in

simulations which will directly affect the detection performances

In term of fabrication process smaller size diodes ie submicron level can be

achieved using dry etching technique with proper calibration For wet etch technique the

etched profile still can be improved by thinning the doped layers so that etching time

will be reduced and hence dimensions down to 2times2microm2 or even 15times15microm

2 can be

reproducibly made

For simulations advanced AC and RF modelling utilizing physical device

simulation available software (SILVACO) must be include in future research hence

holistic study can be conducted to improve the understanding of the ASPAT

For MMIC detector and multiplier design it is vital to produce actual MMIC

devices so that the simulation results can be validated Ultimately tested devices with

good performances can be realized and manufactured

218

REFERENCES

1 Laeri F U Simon and M Wark Host-Guest-Systems Based on Nanoporous

Crystals 2006 John Wiley amp Sons

2 Łukasiak L and A Jakubowski History of semiconductors Journal of

Telecommunications and information technology 2010 p 3-9

3 Song H-J and T Nagatsuma Present and future of terahertz communications

IEEE Transactions on Terahertz Science and Technology 2011 1(1) p 256-

263

4 Hu B and M Nuss Imaging with terahertz waves Optics letters 1995 20(16)

p 1716-1718

5 Smith PR DH Auston and MC Nuss Subpicosecond photoconducting

dipole antennas IEEE Journal of Quantum Electronics 1988 24(2) p 255-260

6 Nagatsuma T Terahertz technologies present and future IEICE Electronics

Express 2011 8(14) p 1127-1142

7 Kumar S et al A 18-THz quantum cascade laser operating significantly above

the temperature of [planck][omega]kB Nature Physics 2011 7(2) p 166-171

8 Phillips T and D Woody Millimeter-and submillimeter-wave receivers Annual

Review of Astronomy and Astrophysics 1982 20(1) p 285-321

9 Whitmer HCTaCA Crystal Rectifiers McGraw-Hill book Company

London 1948

10 Young D and J Irvin Millimeter frequency conversion using Au-n-type GaAs

Schottky barrier epitaxial diodes with a novel contacting technique Proceedings

of the Ieee 1965 53(12) p 2130-2131

11 Liu L et al A broadband quasi-optical terahertz detector utilizing a zero bias

Schottky diode IEEE microwave and wireless components letters 2010 20(9) p

504-506

12 Sankaran S Schottky barrier diodes for millimeter wave detection in a foundry

CMOS process IEEE Electron Device Letters 2005 26(7) p 492-494

13 Chattopadhyay G Submillimeter-wave coherent and incoherent sensors for

space applications in Sensors 2008 Springer p 387-414

14 Anand Y and WJ Moroney Microwave mixer and detector diodes

Proceedings of the Ieee 1971 59(8) p 1182-1190

15 Syme RT Microwave Detection Using GaasAlas Tunnel Structures Gec

Journal of Research 1993 11(1) p 12-23

16 Syme RT et al Asymmetric superlattices for microwave detection in Physical

Concepts of Materials for Novel Optoelectronic Device Applications 1991

International Society for Optics and Photonics

17 Missous M MJ Kelly and J Sexton Extremely uniform tunnel barriers for

low-cost device manufacture IEEE Electron Device Letters 2015 36(6) p 543-

545

18 Syme RT et al Novel GaAsAlAs tunnel structures as microwave detectors in

Semiconductors 92 1992 International Society for Optics and Photonics

19 Schwierz F and JJ Liou Semiconductor devices for RF applications evolution

and current status Microelectronics Reliability 2001 41(2) p 145-168

219

20 HayasHi H Development of Compound Semiconductor DevicesmdashIn Search of

Immense Possibilitiesmdash SEI TECHNICAL REVIEW 2011(72) p 5

21 Mead C Schottky barrier gate field effect transistor Proceedings of the Ieee

1966 54(2) p 307-308

22 Hooper W and W Lehrer An epitaxial GaAs field-effect transistor Proceedings

of the Ieee 1967 55(7) p 1237-1238

23 Drangeid K R Sommerhalder and W Walter High-speed gallium-arsenide

Schottky-barrier field-effect transistors Electronics Letters 1970 6(8) p 228-

229

24 Pillarisetty R Academic and industry research progress in germanium

nanodevices Nature 2011 479(7373) p 324-328

25 Oxley TH 50 years development of the microwave mixer for heterodyne

reception IEEE transactions on microwave theory and techniques 2002 50(3)

p 867-876

26 Baca AG and CI Ashby Fabrication of GaAs devices 2005 IET

27 Cho AY and J Arthur Molecular beam epitaxy Progress in solid state

chemistry 1975 10 p 157-191

28 Cho A Growth of IIIndashV semiconductors by molecular beam epitaxy and their

properties Thin Solid Films 1983 100(4) p 291-317

29 Kiehl RA and TG Sollner High speed heterostructure devices 1994

Academic Press

30 Feiginov M et al Resonant-tunnelling-diode oscillators operating at

frequencies above 11 THz Applied Physics Letters 2011 99(23) p 233506

31 Chang LL L Esaki and R Tsu Resonant tunneling in semiconductor double

barriers Applied Physics Letters 1974 24(12) p 593-595

32 Kasjoo SR Novel Electronic Nanodevices Operating in the Terahertz Region

2012

33 Kanaya H et al Structure dependence of oscillation characteristics of

resonant-tunneling-diode terahertz oscillators associated with intrinsic and

extrinsic delay times Japanese Journal of Applied Physics 2015 54(9) p

094103

34 Chattopadhyay G Technology capabilities and performance of low power

terahertz sources IEEE Transactions on Terahertz Science and Technology

2011 1(1) p 33-53

35 Betz A and R Boreiko A practical Schottky mixer for 5 THz in Proceedings of

the 7th International Symposium on Space Terahertz Technology 1996

36 Yu D et al Ultra high-speed 025-spl mum emitter InP-InGaAs SHBTs with

fsub maxof 687 GHz in Electron Devices Meeting 2004 IEDM Technical

Digest IEEE International 2004 IEEE

37 Das MB Optoelectronic detectors and receivers speed and sensitivity limits in

Optoelectronic and Microelectronic Materials Devices 1998 Proceedings 1998

Conference on 1999 IEEE

38 Rodwell MJ et al Submicron scaling of HBTs IEEE Transactions on Electron

Devices 2001 48(11) p 2606-2624

220

39 Bouloukou A and S Missous Novel High-breakdown Low-noise InGaAs-

InA1As Transistors for Radio Astronomy Applications 2006 University of

Manchester

40 Bean J Materials and technologies 1990 John Wiley amp Sons New York p

13

41 Swaminathan V and A Macrander Materials aspects of GaAs and InP based

structures 1991 Prentice-Hall Inc

42 Vurgaftman I J Meyer and L Ram-Mohan Band parameters for IIIndashV

compound semiconductors and their alloys Journal of applied physics 2001

89(11) p 5815-5875

43 Dingle R W Wiegmann and CH Henry Quantum states of confined carriers

in very thin Al x Ga 1minus x As-GaAs-Al x Ga 1minus x As heterostructures Physical

Review Letters 1974 33(14) p 827

44 Sze SM and KK Ng Physics of semiconductor devices 2006 John wiley amp

sons

45 Tyagi MS Introduction to semiconductor materials and devices 2008 John

Wiley amp Sons

46 Schubert E Delta doping of IIIndashV compound semiconductors Fundamentals

and device applications Journal of Vacuum Science amp Technology A Vacuum

Surfaces and Films 1990 8(3) p 2980-2996

47 Rhoderick EH Metal-semiconductor contacts IEE Proceedings I-Solid-State

and Electron Devices 1982 129(1) p 1

48 Piotrowska A A Guivarch and G Pelous Ohmic contacts to IIIndashV compound

semiconductors A review of fabrication techniques Solid-State Electronics

1983 26(3) p 179-197

49 Rideout V A review of the theory and technology for ohmic contacts to group

IIIndashV compound semiconductors Solid-State Electronics 1975 18(6) p 541-

550

50 Baca A et al A survey of ohmic contacts to III-V compound semiconductors

Thin Solid Films 1997 308 p 599-606

51 Higman T et al Structural analysis of AundashNindashGe and AundashAgndashGe alloyed

ohmic contacts on modulation‐doped AlGaAsndashGaAs heterostructures Journal of

applied physics 1986 60(2) p 677-680

52 Chen KJ et al High-performance enhancement-mode InAlAsInGaAs HEMTs

using non-alloyed ohmic contact and Pt-based buried-gate in Indium Phosphide

and Related Materials 1995 Conference Proceedings Seventh International

Conference on 1995 IEEE

53 Berlin L The man behind the microchip Robert Noyce and the invention of

Silicon Valley 2005 Oxford University Press

54 Goodhue W et al Large room‐temperature effects from resonant tunneling

through AlAs barriers Applied Physics Letters 1986 49(17) p 1086-1088

55 Kerr A and Y Anand Schottky diode MM detectors with improved sensitivity

and dynamic range Microwave Journal 1981 24 p 67-71

56 Davies R Simulations of the current-voltage characteristics of semiconductor

tunnel structures Gec Journal of Research 1987 5(2) p 65-75

221

57 Kelly M Tunnel structures and devices over the coming decade Philosophical

Transactions of the Royal Society of London A Mathematical Physical and

Engineering Sciences 1996 354(1717) p 2291-2293

58 Wilkinson V and M Kelly Manufacturability of quantum semiconductor

devices in High Performance Electron Devices for Microwave and

Optoelectronic Applications 1995 EDMO IEEE 1995 Workshop on 1995

IEEE

59 Wilkinson V M Kelly and M Carr Tunnel devices are not yet

manufacturable Semiconductor Science and Technology 1997 12(1) p 91

60 Eaves L and MJ Kelly The current status of semiconductor tunnelling devices

Philos trans of the Roy soc of London Ser A Math phys and eng sciences

1996 354(1717)

61 Billen K V Wilkinson and M Kelly Manufacturability of heterojunction

tunnel devices further progress Semiconductor Science and Technology 1997

12(7) p 894

62 Kelly M The engineering of quantumndashdot devices Philosophical Transactions

of the Royal Society of London A Mathematical Physical and Engineering

Sciences 2003 361(1803) p 393-401

63 Kelly M New statistical analysis of tunnel diode barriers Semiconductor

Science and Technology 2000 15(1) p 79

64 Hayden R et al Ex situ re-calibration method for low-cost precision epitaxial

growth of heterostructure devices Semiconductor Science and Technology

2002 17(2) p 135

65 Dasmahapatra P et al Thickness control of molecular beam epitaxy grown

layers at the 001ndash01 monolayer level Semiconductor Science and Technology

2012 27(8) p 085007

66 Hayden R M Missous and M Kelly Precision growth for the manufacture of

semiconductor heterostructure devices Semiconductor Science and Technology

2001 16(8) p 676

67 Shao C et al Highly reproducible tunnel currents in MBE-grown

semiconductor multilayers Electronics Letters 2012 48(13) p 792-794

68 Abdullah MR et al GaAsAlAs tunnelling structure Temperature dependence

of ASPAT detectors in Millimeter Waves and THz Technology Workshop

(UCMMT) 2015 8th UK Europe China 2015 IEEE

69 Ariffin KZ et al Asymmetric Spacer Layer Tunnel In0 18Ga0 82AsAlAs

(ASPAT) Diode using double quantum wells for dual functions Detection and

oscillation in Millimeter Waves and THz Technology Workshop (UCMMT)

2015 8th UK Europe China 2015 IEEE

70 Liboff RL Introductory quantum mechanics 2003 Addison-Wesley

71 Esaki L Discovery of the tunnel diode IEEE Transactions on Electron Devices

1976 23(7) p 644-647

72 Landau LD LEM Quantum Mechanics Non-relativistic Theory Pergamon 3

73 Landau LD et al Quantum Mechanics Non‐Relativistic Theory Vol 3 of

Course of Theoretical Physics 1958 AIP

222

74 Syme R Tunnelling devices as microwave mixers and detectors Philosophical

Transactions of the Royal Society of London A Mathematical Physical and

Engineering Sciences 1996 354(1717) p 2351-2364

75 Syme R et al Tunnel diode with asymmetric spacer layers for use as

microwave detector Electronics Letters 1991 27(23) p 2192-2194

76 Brown E W Goodhue and T Sollner Fundamental oscillations up to 200 GHz

in resonant tunneling diodes and new estimates of their maximum oscillation

frequency from stationary‐state tunneling theory Journal of applied physics

1988 64(3) p 1519-1529

77 Reddy M Schottky-collector resonant tunnel diodes for sub-millimeter-wave

applications 1997 University of California Santa Barbara

78 Cox R and H Strack Ohmic contacts for GaAs devices Solid-State Electronics

1967 10(12) p 1213IN71215-1214IN81218

79 Valdes LB Resistivity measurements on germanium for transistors

Proceedings of the IRE 1954 42(2) p 420-427

80 Schroder DK Semiconductor material and device characterization 2006 John

Wiley amp Sons

81 Klootwijk J and C Timmering Merits and Limitations of Circular TLM

structures for contact resistance determination for novel 111-V HBTs Proc

fEEE 2004

82 Marlow GS and MB Das The effects of contact size and non-zero metal

resistance on the determination of specific contact resistance Solid-State

Electronics 1982 25(2) p 91-94

83 Murrmann H and D Widmann Current crowding on metal contacts to planar

devices IEEE Transactions on Electron Devices 1969 16(12) p 1022-1024

84 Berger H Models for contacts to planar devices Solid-State Electronics 1972

15(2) p 145-158

85 Reeves G and H Harrison Obtaining the specific contact resistance from

transmission line model measurements IEEE Electron Device Letters 1982

3(5) p 111-113

86 Shur MS GaAs devices and circuits 2013 Springer Science amp Business

Media

87 Popescu D and B Odbert The Advantages Of Remote Labs In Engineering

Education Educatorrsquos Corner-Agilent Technologies-application note 2011 p

11

88 DataSheet Karl Suss- PM5 Probe System Datasheet 2013

89 Keysight IC-CAP Device Modeling Software 2016 Available from

httpwwwkeysightcomenpc-1297149ic-cap-device-modeling-software-

measurement-control-and-parameter-extractioncc=USamplc=eng

90 DataSheet Anritsu 37369A Vector Network Analyzer Datasheet 2016 Available

from httpwwwtestequipmenthqcomdatasheetsANRITSU-37397D-

Datasheetpdf

91 Packard H HP 4142B Modular DC SourceManual Operation Manual 1992

Available from httpcpliteratureagilentcomlitwebpdf04142-90010pdf

223

92 Microtech C Cascade Microtech- Wincal High Performence RF calaibration

Software (Official Website) 2016 Available from

httpswwwcascademicrotechcom

93 Singh J Electronic and optoelectronic properties of semiconductor structures

2007 Cambridge University Press

94 Whyte W Cleanroom design 1999 Wiley Online Library

95 Vieu C et al Electron beam lithography resolution limits and applications

Applied Surface Science 2000 164(1) p 111-117

96 La Fontaine B Lasers and Moorersquos law SPIE Professional October 2010 p

20

97 Madou MJ Fundamentals of microfabrication the science of miniaturization

2002 CRC press

98 Serway R Physics for Scientists and Engineers 1996 Saunders Publ

Philadelphia

99 Jalali B and S Pearton InP HBTs growth processing and applications 1995

Artech House Publishers

100 Shih YC et al Effects of interfacial microstructure on uniformity and thermal

stability of AuNiGe ohmic contact to n‐type GaAs Journal of applied physics

1987 62(2) p 582-590

101 Zawawi M Advanced In0 8Ga0 2AsAlAs Resonant Tunneling Diodes

forApplications in Integrated mm-waves MMIC Oscillators 2015

102 Zawawi MAM et al Fabrication of Submicrometer InGaAsAlAs Resonant

Tunneling Diode Using a Trilayer Soft Reflow Technique With Excellent

Scalability IEEE Transactions on Electron Devices 2014 61(7) p 2338-2342

103 Silvaco I ATLAS Users Manual Device Simulation Software 2010 Santa

Clara CA

104 Kyono C et al Dependence of apparent barrier height on barrier thickness for

perpendicular transport in AlAsGaAs single‐barrier structures grown by

molecular beam epitaxy Applied Physics Letters 1989 54(6) p 549-551

105 Yang K JR East and GI Haddad Numerical modeling of abrupt

heterojunctions using a thermionic-field emission boundary condition Solid-

State Electronics 1993 36(3) p 321-330

106 Varshni YP Temperature dependence of the energy gap in semiconductors

Physica 1967 34(1) p 149-154

107 Handbook LLM Precision DC Current Voltage and Resistance

Measurements Keithley Instruments Inc[online] 6th revision Ohio 2004

108 Lipsky SE Microwave passive direction finding 2004 SciTech Publishing

109 Howell CM and SJ Parisi Principles Applications and Selection of Receiving

Diodes MACOM Semiconductor Products Division Application note AG314

110 Schulman J D Chow and D Jang InGaAs zero bias backward diodes for

millimeter wave direct detection IEEE Electron Device Letters 2001 22(5) p

200-202

111 Zhang Z et al Sub-Micron Area Heterojunction Backward Diode Millimeter-

Wave Detectors With 018$ rm pWHz^12 $ Noise Equivalent Power IEEE

microwave and wireless components letters 2011 21(5) p 267-269

224

112 Jin N et al High sensitivity Si-based backward diodes for zero-biased square-

law detection and the effect of post-growth annealing on performance IEEE

Electron Device Letters 2005 26(8) p 575-578

113 Shashkin VI et al Millimeter-wave detectors based on antenna-coupled low-

barrier Schottky diodes International journal of infrared and millimeter waves

2007 28(11) p 945-952

114 Zhao P et al GaN Heterostructure Barrier Diodes Exploiting Polarization-

Induced $delta $-Doping IEEE Electron Device Letters 2014 35(6) p 615-

617

115 Pozar DM Microwave engineering 2009 John Wiley amp Sons

116 Koolen M J Geelen and M Versleijen An improved de-embedding technique

for on-wafer high-frequency characterization in Bipolar Circuits and

Technology Meeting 1991 Proceedings of the 1991 1991 IEEE

117 Cao M et al RF characteristics uniformity of GaAsAlAs tunnel diodes in

Infrared Millimeter and Terahertz waves (IRMMW-THz) 2016 41st

International Conference on 2016 IEEE

118 Gao J RF and microwave modeling and measurement techniques for field effect

transistors 2010 SciTec

119 Ren T et al A 340-400 GHz Zero-Biased Waveguide Detector Using an Self-

Consistent Method to Extract the Parameters of Schottky Barrier Diode Applied

Computational Electromagnetics Society Journal 2015 30(12)

120 Fobelets K et al High‐frequency capacitances in resonant interband tunneling

diodes Applied Physics Letters 1994 64(19) p 2523-2525

121 Diebold S et al Modeling and Simulation of Terahertz Resonant Tunneling

Diode-Based Circuits IEEE Transactions on Terahertz Science and Technology

2016 6(5) p 716-723

122 Yong Z et al Design of a 220 GHz frequency tripler based on EM model of

Schottky diodes JOURNAL OF INFRARED AND MILLIMETER WAVES

2014 33(4) p 405-411

123 Louhi JT and AV Raisanen On the modeling and optimization of Schottky

varactor frequency multipliers at submillimeter wavelengths IEEE transactions

on microwave theory and techniques 1995 43(4) p 922-926

124 Guo J Z Zhang and C Qian Modeling of commercial millimeter wave

Schottky diodes in Microwave and Millimeter Wave Technology (ICMMT) 2016

IEEE International Conference on 2016 IEEE

125 Schneider M Metal-semiconductor junctions as frequency converters Infrared

and Millimeter Waves 1982 6 p 209

126 Muth C et al Advanced technology microwave sounder on NPOESS and NPP

in Geoscience and Remote Sensing Symposium 2004 IGARSS04 Proceedings

2004 IEEE International 2004 IEEE

127 Putley E Thermal detectors in Optical and Infrared Detectors 1977 Springer

p 71-100

128 Martin DH Spectroscopic techniques for far infra-red submillimetre and

millimetre waves in Spectroscopic Techniques for Far Infra-red Submillimetre

and Millimetre Waves 1967

225

129 Lucas W Tangential sensitivity of a detector video system with RF

preamplification in Proceedings of the Institution of Electrical Engineers 1966

IET

130 Balocco C et al Low-frequency noise of unipolar nanorectifiers Applied

Physics Letters 2011 99(11) p 113511

131 Benford D T Hunter and TG Phillips Noise equivalent power of background

limited thermal detectors at submillimeter wavelengths International journal of

infrared and millimeter waves 1998 19(7) p 931-938

132 Papoušek D Vibration-rotational Spectroscopy and Molecular Dynamics

Advances in Quantum Chemical and Spectroscopical Studies of Molecular

Structures and Dynamics Vol 9 1997 World Scientific

133 Nyquist H Thermal agitation of electric charge in conductors Physical review

1928 32(1) p 110

134 Turner CS Johnson-Nyquist Noise url httpwww claysturner

comdspJohnson-NyquistNoise pdf(Letzter Abruf Juli 2012)

135 Voss RF 1f (flicker) noise A brief review in 33rd Annual Symposium on

Frequency Control 1979 1979 IEEE

136 McWhorter AL 1f noise and related surface effects in germanium 1955

137 Hooge FN 1ƒ noise is no surface effect Physics letters A 1969 29(3) p 139-

140

138 Van der Ziel A Noise Sources characterization measurement Prentice-Hall

Information and System Sciences Series Englewood Cliffs Prentice-Hall 1970

1970

139 Hooge F 1f noise sources IEEE Transactions on Electron Devices 1994

41(11) p 1926-1935

140 Der Ziel A Theory of shot noise in junction diodes and junction transistors

Proceedings of the IRE 1955 43(11) p 1639-1646

141 Schottky W Small-shot effect and flicker effect Physical review 1926 28(1)

p 74

142 Cowley A and H Sorensen Quantitative comparison of solid-state microwave

detectors IEEE transactions on microwave theory and techniques 1966 14(12)

p 588-602

143 Schulman J et al 1$f $ Noise of Sb-Heterostructure Diodes for Pre-Amplified

Detection IEEE microwave and wireless components letters 2007 17(5) p

355-357

144 Lynch JJ et al Passive millimeter-wave imaging module with preamplified

zero-bias detection IEEE transactions on microwave theory and techniques

2008 56(7) p 1592-1600

145 Su N et al Sb-heterostructure millimeter-wave detectors with reduced

capacitance and noise equivalent power IEEE Electron Device Letters 2008

29(6) p 536-539

146 Westlund A Self-Switching Diodes for Zero-Bias Terahertz Detection 2015

Chalmers University of Technology

147 Yajima T and L Esaki Excess noise in narrow germanium pn junctions

Journal of the physical society of Japan 1958 13(11) p 1281-1287

226

148 Sommers H Tunnel diodes as high-frequency devices Proceedings of the IRE

1959 47(7) p 1201-1206

149 Miraftab V and A Abdipour Harmonic balance analysis of a microwave

balanced power amplifier in Electrical and Computer Engineering 2001

Canadian Conference on 2001 IEEE

150 Patrashin M et al GaAsSbInAlAsInGaAs Tunnel Diodes for Millimeter Wave

Detection in 220ndash330-GHz Band IEEE Transactions on Electron Devices 2015

62(3) p 1068-1071

151 Hesler JL and TW Crowe Responsivity and noise measurements of zero-bias

Schottky diode detectors Proc ISSTT 2007 p 89-92

152 Su N et al Temperature dependence of high frequency and noise performance

of Sb-heterostructure millimeter-wave detectors IEEE Electron Device Letters

2007 28(5) p 336-339

153 Lynch J et al Unamplified direct detection sensor for passive millimeter wave

imaging in Proc of SPIE Vol 2006

154 ZHANG Z et al A physics-based tunneling model for Sb-heterostructure

backward tunnel diode millimeter-wave detectors International Journal of High

Speed Electronics and Systems 2011 20(03) p 589-596

155 Bahl IJ and P Bhartia Microwave solid state circuit design 2003 John Wiley

amp Sons

156 Yeom K-W Microwave Circuit Design A Practical Approach Using ADS

2015 Prentice Hall Press

157 Xie L et al A W-band detector with high tangential signal sensitivity and

voltage sensitivity in Microwave and Millimeter Wave Technology (ICMMT)

2010 International Conference on 2010 IEEE

158 Fay P et al High-performance antimonide-based heterostructure backward

diodes for millimeter-wave detection IEEE Electron Device Letters 2002

23(10) p 585-587

159 Hrobak M et al Planar zero bias Schottky diode detector operating in the E-

and W-band in Microwave Conference (EuMC) 2013 European 2013 IEEE

160 Maas SA Nonlinear microwave and RF circuits 2003 Artech House

161 Appleby R and RN Anderton Millimeter-wave and submillimeter-wave

imaging for security and surveillance Proceedings of the Ieee 2007 95(8) p

1683-1690

162 Crowe TW et al Opening the terahertz window with integrated diode circuits

IEEE Journal of Solid-State Circuits 2005 40(10) p 2104-2110

163 Raisanen AV Frequency multipliers for millimeter and submillimeter

wavelengths Proceedings of the Ieee 1992 80(11) p 1842-1852

164 Erickson NR Diode frequency multipliers for terahertz local-oscillator

applications in Astronomical Telescopes amp Instrumentation 1998 International

Society for Optics and Photonics

165 Mehdi I et al Terahertz local oscillator sources performance and capabilities

in Astronomical Telescopes and Instrumentation 2003 International Society for

Optics and Photonics

166 Tonouchi M Cutting-edge terahertz technology Nature photonics 2007 1(2)

p 97-105

227

167 Nilsen SM et al Single barrier varactors for submillimeter wave power

generation IEEE transactions on microwave theory and techniques 1993 41(4)

p 572-580

168 Xiao Q et al A 270-GHz tuner-less heterostructure barrier varactor frequency

tripler IEEE microwave and wireless components letters 2007 17(4) p 241-

243

169 David T et al Monolithic integrated circuits incorporating InP-based

heterostructure barrier varactors IEEE microwave and wireless components

letters 2002 12(8) p 281-283

170 Lieneweg U B Hancock and J Maserjian Barrier-intrinsic-N+(BIN) diodes

for near-millimeter wave generation in Conference Digest for the Twelft

International Conference on Infrared and Millimeter Waves 1987

171 Lieneweg U et al Modeling of planar varactor frequency multiplier devices

with blocking barriers IEEE transactions on microwave theory and techniques

1992 40(5) p 839-845

172 Chattopadhyay G et al An all-solid-state broad-band frequency multiplier

chain at 1500 GHz IEEE transactions on microwave theory and techniques

2004 52(5) p 1538-1547

173 Maestrini A et al A 17-19 THz local oscillator source IEEE microwave and

wireless components letters 2004 14(6) p 253-255

174 Faber MT J Chramiec and ME Adamski Microwave and millimeter-wave

diode frequency multipliers 1995 Artech House Publishers

175 Frerking MA and JR East Novel heterojunction varactors Proceedings of the

Ieee 1992 80(11) p 1853-1860

176 Penfield P and RP Rafuse Varactor applications 1962

177 Palazzi V et al Low-power frequency doubler in cellulose-based materials for

harmonic RFID applications IEEE microwave and wireless components letters

2014 24(12) p 896-898

178 Presas SM Microwave frequency doubler integrated with miniaturized planar

antennas 2008

228

APPENDICES

Appendix I Doped substrate process details

Mask Stage Process Stage Process step Process detail Equipment

Mask 1 (Mesa

Etch)

Sample clean NMP 10 min 80˚C Beaker

Acetone 5 min Beaker

Isopropanol (IPA) 5 min Beaker

Apply Resist Prebake Bake for 5mins 150C Hotplate

1st Resist S1805 - Program 4 Laurell Spinner

Post bake 110C for 1 Min Hotplate

Photolithography Mask Align mask to wafer

MA4 Mask

aligner

Expose

1 min 09mW iline

(Compensation error set to 1)

MA4 Mask

aligner

Develop MIF 319 for 1 min Beaker

Post Exposure

Bake Oven or Bake 120C for 15

minutes Hotplate

Etch Etch Cal Cal Sample - etch for 2 minutes Beaker

Etch

H3PO4H2O2H2O 3150

time is determined by the etch cal Beaker

Resist Strip Acetone - 5 min and IPA - 5 min Beaker

Mask2 (Top

Contact)

Sample clean NMP ultrasonic for 15 Min Power 3 Beaker

Acetone ultrasonic for 5 Min Power 3 Beaker

Isopropanol (IPA) ultrasonic for 5 Min Power 3 Beaker

Apply Resist Prebake Bake for 5mins 150C Hotplate

1st Resist AznLoF - 2um grade - Program 6 Laurell Spinner

Post bake 110C for 1 Min Hotplate

Photolithography Mask Align mask to wafer

MA4 Mask

aligner

Expose

55 seconds 09mW i-line

(Compensation error set to 1)

MA4 Mask

aligner

Post Exposure bake 110C for 1 Min Hotplate

Develop MIF 326 for 1 mins Beaker

Clean O2 Plasma Etch

20sec 100W forward (340)

60mTorr Plasma Tech

Surface De-oxide HCLH2O 11 40 sec Beaker

Metallisation Evaporation Clean trays + metals Beaker

Load sample and metals

Edward

Evaporator

AuGe (100mg) deposits 55nm

Edward

Evaporator

Ni (1cm) deposits 15nm

Edward

Evaporator

Au (10cm) deposits 200nm

Edward

Evaporator

Lift-off

NMP1165 for 24 hours (25C) or

20 min 80C Beaker

Anneal 420˚C 2min Furnace

229

TLM

Measurement ICCAP

Measurement

Bench

Mask 3

(BottomBacks

ide Contact)

Sample clean Acetone Optional Beaker

Isopropanol (IPA) Optional Beaker

Apply Resist Prebake

Bake for 5mins 150C to dry

the sample Hotplate

Resist (top side) S1813 - Program 6 Laurell Spinner

Soft bake 110C for 1 Min Hotplate

Photolithography Mask Align mask to wafer

MA4 Mask

aligner

Exposure

10 seconds 09mW I-line

(Compensation error set to 1)

MA4 Mask

aligner

Post Exposure bake 110C for 1 Min Hotplate

Develop MIF 319 for 2 mins Beaker

De-scum O2 Plasma Etch

20sec 100W forward (340)

60mTorr Plasma Tech

De-oxidise

Surface De-oxide

HCLH2O 11 40

sec Beaker

metallisation Evaporation Clean trays + metals Beaker

Load sample and metals

Edward

Evaporator

AuGe (100mg) deposits 55nm

Edward

Evaporator

Ni (1cm) deposits 15nm

Edward

Evaporator

Au (10cm) deposits 200nm

Edward

Evaporator

Lift-off

NMP1165 for 24 hours (25C) or

20 min 80C Beaker

Clean Water 3min Beaker

TLM

Measurement ICCAP

230

Appendix II Four Mask step Process Flow

Mask Stage Process Stage Process step Process detail

Mask1 Top

Contact)

Sample clean NMP ultrasonic for 15 Min Power 3

Acetone ultrasonic for 5 Min Power 3

Isopropanol (IPA) ultrasonic for 5 Min Power 3

Apply Resist Prebake Bake for 5mins 150C

1st Resist AznLoF - 2um grade - Program 6

Hot Plate 110C for 1 Min

Photolithography Mask Align mask to wafer

Expose 55 seconds 09mW iline (Compensation

error set to 1)

Post Exposure bake 110C for 1 Min

Develop MIF 326 for 1 mins

Clean O2 Plasma Etch 20sec 100W forward (340) 60mTorr

Surface De-oxide HCLH2O 11 40 sec

Metalisation Evaporation Clean trays + metals

Load sample and metals

AuGe (100mg) deposits 55nm

Ni (1cm) deposits 15nm

Au (10cm) deposits 200nm

Lift-off NMP1165 for 24 hours (25C) or 20 mnt 80C

Mask 2 (Mesa

Etch)

Sample clean NMP Optional

Acetone Optional

Isopropanol (IPA) Optional

Apply Resist Prebake Bake for 5mins 150C

1st Resist S1805 - Program 4

HotPlate 110C for 1 Min

Photolithography Mask Align mask to wafer

Expose 1 min 09mW i-line (Compensation error set

to 1)

Develop MIF 319 for 1 min

Post Exposure Bake Oven or Bake 120C for 15 minutes

Etch Etch Cal Cal Sample - etch for 2 minutes

Etch H3PO4H2O2H2O 3150 time is

determined by the etch cal

Measure TLM

Resist Strip Acetone - 5 min and IPA - 5 min

Mask 3(Isolation) Sample clean Acetone Optional

Isopropanol (IPA) Optional

231

Apply Resist Prebake Bake for 5mins 150C

Resist S1828 - Program 4

Hot Plate 115C for 1 mins

Photolithography Mask Align mask to wafer

Expose 9 mins 09mW iline (Compensation error set

to 1)

Develop MF 319 3 mins

Post Bake Oven Bake 120C for 15mins

Etch Etch Cal Refer to Etch Cal instr tab

Sub-collector Etch H3PO4H2O2H2O 3150 time is

determined by the etch cal

Resist Strip Acetone 5mins + IPA 5 mins in ultrasonic bath

power 1

inspection Microscope

Sample clean Acetone Optional

Mask 4 (Bottom

Contact)

Isopropanol (IPA) Optional

Apply Resist Prebake Bake for 5mins 150C 1st

Resist AznLoF - 2um grade - Program 6

HotPlate 110C for 1 Min

Photolithography Mask Align mask to wafer

Expose 10 seconds 09mW i-line (Compensation

error set to 1)

Post Exposure bake 110C for 1 Min

Develop MIF 326 for 2 mins

Clean O2 Plasma Etch 20sec 100W forward (340) 60mTorr

Surface De-oxide

HCLH2O 11 40 sec

Metalisation Evaporation Clean trays + metals

Load sample and metals

AuGe (100mg) deposits 55nm

Ni (1cm) deposits 15nm

Au (10cm) deposits 200nm

Lift-off NMP1165 for 24 hours (25C) or 20 mnt 80C

232

Appendix III Epitaxial Layer XMBE277

TABLE I The epitaxial structure for sample XMBE277

Layer Thickness (nm) Doping Concentration (cm-3)

n+- In053Ga047As 45 200 x 1019

n- In053Ga047As 25 300 x 1018

In053Ga047As 20 undoped

AlAs 13 undoped

In08Ga02As 45 undoped

AlAs 13 undoped

In053Ga047As 20 undoped

n- In053Ga047As 25 300 x 1018

n+- In053Ga047As 400 100 x 1019

Semi-insulating InP

233

Appendix IV SilVaco (Atlas) Simulation Code

go atlas

---------------------------------------------------------

Structure parameter definition (Constants) values in um

---------------------------------------------------------

Thicknesses

set t_contact1=0

set t_ohmic1=03

set t_emitter=004

set t_spacer1=0005

set t_barrier=000283

set t_spacer2=02

set t_collector=004

set t_ohmic2=045

Doping concentrations

set d_ohmic1=4e18

set d_emitter=2e17

set d_collector=2e17

set d_ohmic2=4e18

set d_gap=2

set d_mesa=4

set d_device=10

set d_etch=008

Layers

set I=$t_contact1

set A=$I+$t_ohmic1

set B=$A+$t_emitter

set C=$B+$t_spacer1

set D=$C+$t_barrier

set E=$D+$t_spacer2

set F=$E+$t_collector

set G=$F+$t_ohmic2

-------------------------------------

Mesh generator

-------------------------------------

mesh diagflip width=45

xmesh location=0 s=1

xmesh location=1 s=1

xmesh location=2 s=1

xmesh location=4 s=1

xmesh location=5 s=1

xmesh location=6 s=1

xmesh location=7 s=1

xmesh location=8 s=1

xmesh location=$d_mesa s=1

xmesh location=$d_mesa+$d_gap s=1

xmesh location=$d_device s=1

Ohmic1

ymesh l=0000 s=005

ymesh l=$I s=005

234

ymesh l=$A s=0005

ymesh l=$B s=0005

ymesh l=$C s=00005

ymesh l=$D s=00005

ymesh l=$E s=0009

ymesh l=$F s=0005

ymesh l=$G s=0005

-----------------------------------

SECTION 2 Regions Structure definition

-----------------------------------

region num=1 name=contact1 material=Gold ymin=0 ymax=$I

region num=2 name=ohmic1 material=GaAs ymin=$I ymax=$A

region num=3 name=emitter material=GaAs ymin=$A ymax=$B

region num=4 name=spacer1 material=GaAs ymin=$B ymax=$C

region num=5 name=barrier material=AlAs ymin=$C ymax=$D xmin=0 xmax=$d_mesa

calcstrain qtregion=1

region num=6 name=spacer2 material=GaAs ymin=$D ymax=$E

region num=7 name=collector material=GaAs ymin=$E ymax=$F

region num=8 name=ohmic2 material=GaAs ymin=$F ymax=$G

region num=9

name=etch material=Air ymin=0 ymax=$F+$d_etch xmin=$d_mesa xmax=$d

_device

---------------------------------

Electrodes

---------------------------------

electrode num=1 name=anode xmin=0 xmax=$d_mesa ymin=0 ymax

=$I material=Gold

electrode num=2 name=cathode xmin=$d_mesa+$d_gap xmax=$d_device ymin=$F+

$d_etch ymax=$F+$d_etch material=Gold

--------------------------------

Doping

--------------------------------

doping uniform ntype conc=$d_ohmic1 Region=2 ymin=$I ymax=$A

doping uniform ntype conc=$d_emitter Region=3 ymin=$A ymax=$B

doping uniform ntype conc=$d_collector Region=7 ymin=$E ymax=$F

doping uniform ntype conc=$d_ohmic2 Region=8 ymin=$F ymax=$G

--------------------------

Contacts

--------------------------

interface sc region=1

interface ss region=2

interface ss region=3

interface si region=4

interface si region=5

interface ss region=6

interface ss region=7

interface sc ymin=$F ymax=$F xmin=$d_mesa+$d_gap xmax=$d_device

interface tunnel region=5 dytunnel=0001

contact name=cathode

contact name=anode

235

------------------------------------------

SECTION 3 Material amp Models Definitions

------------------------------------------

material material=AlAs

permittivity=10 eg300=28 mc=004 affinity=305 nc300=4e19 nimin=1e1

material material=GaAs permittivity=139 eg300=14 mc=0067

affinity=407 nc300=09e17 nimin=1e6

BAND DIAGRAM

output tquantum bandparam qfn qfp valband conband charge polarcharge flowlines

STRUCTURE GRAPHIC

solve init

save outf=XMBE304+real2str

tonyplot XMBE304+real2str

------------------------------------------

SECTION 4 ANALYSIS

------------------------------------------

trap acceptor structure=top elevel=03 density=48e17 degenfac=10 sign=284e-17 sigp=1e-17

ymin=$A ymax=$C xmin=0 xmax=$d_mesa

trap acceptor structure=BOTTOM elevel=035 density=47e17 degenfac=10 sign=284e-17 sigp=1e-17

ymin=$E ymax=$F xmin=0 xmax=$d_mesa

models sisel sisnlderivs qtregion=1 print

method climit=1e-4 itlimit=50 maxtraps=20

DC ANALYSIS

log outf=XMBE304log

solve init

solve vanode=0 name=anode vstep=001 vfinal=15

save outf=XMBE304str

log off

tonyplot XMBE304str

tonyplot XMBE304log

Page 5: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre

5

72 Motivation and Background 202

73 Frequency Multiplier Architecture the Basics 203

731 Types of frequency multipliers 205

74 Parameters of interest for Frequency Multipliers 206

75 20GHz40GHz Varistor Mode ASPAT Diode Frequency Doubler 208

76 Conclusions 214

8 CONCLUSION AND FUTURE WORK 215

81 Conclusion 215

82 Future Work 217

REFERENCES 218

APPENDICES 228

Word count including footnotes and endnotes 61500(approximately)

6

LIST OF TABLES

Table 11 Sample XMBE304 XMBE307 and XMBE368 GaAsAlAs materials

structure grown on GaAs Substrates by MBE 23

Table 12 Quantum wells sandwiching the quantum barrier for sample XMBE314

grown on a GaAs substrate by MBE 24

Table 13 Novel InGaAsAlAs ASPAT diode grown on InP substrate by MBE on InP

substrate 24

Table 21 Lattice constant (a) and band gap for group III-V binary and ternary

compound semiconductors a room temperature [41 42] 38

Table 22 epitaxial layer of sample XMBE304 which is the main focus of this work 56

Table 31 Chemical wet etch characteristics using Orthophosphoric-based and

Ammonia on GaAs and InGaAs materials 87

Table 32 Epitaxial layer of Doped substrate samples 93

Table 33 The epitaxial layers of sample XMBE304 with barrier 10ML~28nm 94

Table 34 Generic fabrication steps established by Dr Md Adzhar [101] 97

Table 35 Standard process flow for Air-Bridge design fabrication 101

Table 36 Standard fabrication process flow for Dielectric-Bridge design 107

Table 37 New arrangement of the mask number and step in Second Run 110

Table 38 New arrangement for the Third run using Dielectric-Bridge mask 113

Table 39 The outcome of the spreading resistance before and after using LOR

technique 115

Table 310 DC and RF characteristics for XMBE304 118

Table 311 3rd

Gen Mask process step 119

Table 312 Standard deviation at two different voltages 124

Table 41 The parameter values used in this simulation 134

Table 42 The calculated values of bandgap at different temperatures 140

Table 43 The calculated effective masses for each temperature used in this simulation

141

Table 44 Epitaxial layer profile for Schottky Barrier Diode (XMBE104) 145

Table 51 Performance of the ASPAT diode obtained from measured IV characteristics

in this work 153

Table 52 Device to device uniformity check for large ASPAT diode 159

Table 53 The variation of reflection coefficient (S11) for GaAsAlAs ASPAT diode at

four different frequencies[117] 159

7

Table 54 Wafer to wafer uniformity check on XMBE304A and XMBE304B 161

Table 55 Comparison between calculated (fully Depleted) and extracted (different

biases) values from equivalent circuit parameters for different ASPAT mesa sizes at zero

bias voltage 170

Table 61 A summary of all the important parameters of the 4x4 microm2 diode 185

Table 62 Noise Equivalent Power (NEP) for zero bias detector diode 190

Table 63 Summary of characteristic of interest for GaAsAlAs ASPAT detector 198

Table 64 Comparison 4times4microm2 GaAsAlAs ASPAT diode to the-state-of-the-art zero

bias detector at W-band (75GHz-110GHz) 199

Table 71 Comparison parameter obtained from this simulation to the state-of-the-art

multiplier diode 213

8

LIST OF FIGURES

Figure 21 III-V compound semiconductors mobility and band gap[24] 31 Figure 22 illustration of Homojunctions band structure material before (left) and after

(right) equilibrium 34 Figure 23 Energy band diagrams before (left) and after (after) interface combination at

equilibrium 35

Figure 24 Lattice Matching for both materials when aL=aS 37 Figure 25 Energy gap and lattice constant for direct and indirect band gap of compound

semiconductor at ambient temperature[40] 37 Figure 26 Lattice mismatched material 39

Figure 27 The formation of Pseudomorphic-layer in two situations (a) compressive and

(b) tensile strain [1] 40 Figure 28 The band diagram of an ideal quantum well formed by a narrow band gap

semiconductor eg InGaAs sandwiched between two large band gap semiconductors eg

GaAs (a) Structure (b) energy band diagram and (c) Conduction band diagram when

AlGaAs is n-doped[43] 41 Figure 29 Metal and semiconductor in two conditions (a) separation (b) in contact 43

Figure 210 Energy band diagram of Schottky contact on n-type material under (a)

reverse and (b) forward bias 45 Figure 211 Figure 10 N-type semiconductor Ohmic contact with different doping

concentration ND (a) Low (b) Intermediate and (c) high 47 Figure 212 Classical view of whether an electron is can surmount a barrier or not

Quantum mechanical view allows an electron to tunnel through a barrier The probability

(blue) is related to the barrier thickness 51

Figure 213 Rectangular tunnelling with incident reflected and transmitted wave

function[70] 52

Figure 214 Epitaxial layer structure of device XMBE 304 (lateral structure) used in

this study 55 Figure 215 The layer profile of ASPAT with 28nm barrier thickness [27] 57

Figure 216 Conduction band diagram showing band bending and 2DEG formation at

the L1 spacer 60 Figure 217 I-V characteristics of a fabricated ASPAT diode 63 Figure 218 Intrinsic Elements of the Asymmetric Spacer Tunnel Layer (ASPAT) Diode

63 Figure 219 Cross-sectional view of ASPAT vertical structure the dimensions of d b

and h are not drawn to scale 65 Figure 220 Cross-section view of the complete process ASPAT diode (XMBE304)

The dimensions are not drawn to scale 67 Figure 221 A simple TLM structure with effective length and sheet resistance

underneath 69 Figure 222 Top view of TLM ladder structure use in this work 71 Figure 223 Typical plot of resistance versus TLM spacing 71

Figure 224 The set of DC measurement apparatus arrangement to measure the TLM 73

9

Figure 225 The set of RF measurement for on-wafer DC and RF one-port measurement

75 Figure 226 Actual VNA system that was used for RF characterization 75 Figure 227 Block diagram of the ASPAT measurement step 77

Figure 31 3D illustration of Optical lithography process used in this research 85 Figure 32 Actual picture of thermal evaporator used in this study 89 Figure 33 Single layer lift-off process using negative photoresist 91 Figure 34 Current-Voltage characteristic of sample XMBE368 used in this study at

two different locations on the wafer tile 96

Figure 35 ASPAT mask design for (a) 100times100microm2 (b) 30times30microm

2 and (c) 15times15 microm

2

diode dimensions designed in the 1st Gen Mask 96

Figure 36 Air Bridge design (red circle) for large mesa area (100times100microm2) device with

option for doped substrates 100 Figure 37 The layout of 1

st design of Dielectric Bridge (green circle) mask design for

100 times 100microm2 emitter size with option for doped substrate processing 100

Figure 38 Dry Etching for the first run in this study 102 Figure 39 Severe undercut of 2times2 microm

2 and 6times6 microm

2 devices 103

Figure 310 Device with emitter area of 2times2 microm2 and 6times6 microm

2 after dry etch and wet

etched 104 Figure 311 SEM Images of the GaAs sample 104

Figure 312 The cross-sectional view of InGaAs (left) and GaAs (right) samples used in

this study 105

Figure 313 SEM images taken from the XMBE304 samples and a GaAs test sample

105

Figure 314 SEM images for InP and InGaAs taken from [56] 105 Figure 315 Short circuit behaviour on one of the fabricated device in this run 106

Figure 316 The surface of the sample after final processing 108 Figure 317 IV characteristics of sample XMBE304 for emitter sizes of 1000 microm2

2500 microm2 900 microm2 400 microm2 225 microm2 100 microm2 and 36 microm2 109

Figure 318 Optical images after MESA etching (a) 1 microm Tolerance (b) 05 microm

Tolerance 110

Figure 319 After lift-off processing 111 Figure 320 The current density of XMBE304 processed using Si3N4 as dielectric layer

112

Figure 321 side view of lateral ASPAT structure 113

Figure 322 The measured size of the emitter area and the length D (blue color marked)

114 Figure 323 Summary of LOR technique steps 115 Figure 324 IV characteristic in 3rd run (reduced d-Gap) for each diode size on the mask

116

Figure 325 3rd

Gen Mask layout showing actual devices de-embedding TLM and

alignment mark structures used in this study 118 Figure 326 Step by step device fabrication (in 3D and cross-sectional view) for semi-

insulating substrate device type used in this study 121 Figure 327 Example finished process device with bond pad using 3

rd Gen mask 121

Figure 328 XMBE304 TLM measurement for the top contact after annealing 122

10

Figure 329 XMBE304 TLM measurement for the bottom contact after annealing 122 Figure 330 XMBE304 GaAsAlAs ASPAT measured IV characteristics at room

temperature for 4times4microm2 mesa size 123

Figure 331 XMBE304 GaAsAlAs ASPAT measured IV characteristics at room

temperature for 6times6microm2 mesa size 124

Figure 332 XMBE304 GaAsAlAs ASPAT measured IV characteristics at room

temperature for 10times10microm2 mesa size 124

Figure 41 SILVACO Atlas simulation process flow 128 Figure 42 The GaAsAlAs ASPAT structures produced from simulation based on the

diode multilayer heterostructures on the right 131 Figure 43 The Energy-momentum diagram for AlAs heterojunction semiconductor 131 Figure 44 (a) The conduction band and the valence band of the ASPAT diode structure

(b) the energy band diagram of the ASPAT diode structure when under three different

biases 132 Figure 45 The current-voltage characteristic of an ASPAT diode (100times100microm

2) and

(4times4microm2) using SILVACO Atlas simulator for structure device XMBE304 showing

excellent agreement between simulated and experimental data 134

Figure 46 IV characteristics of the dependencies of current on AlAs barrier 136 Figure 47 Example of analysis at -1 and 1V to the current 136 Figure 48 I-V characteristic of the dependencies current to Spacer I layer 137

Figure 49 Current changes with layer thickness l1 138 Figure 410 IV characteristic of the dependencies current to Spacer 1 layer 139

Figure 411Current change with layer thickness l2 139

Figure 412 Measurement and simulation comparison result as a function of temperature

range from 100K to 398K 142 Figure 413 Log Current vs voltage as a function of temperature for ASPAT sample

XMBE304 143

Figure 414 The effective barrier high of the Schottky barrier diode and the ASPAT

Diode [3] 144

Figure 415 Log Current vs voltage as a function of temperature for SBD sample

XMBE104 145

Figure 416 Influence of temperature on IV characteristic for both ASPAT diode and

SBD 146

Figure 51 IV Characteristics of measured ASPAT diodes for emitter sizes of 4x4um2

6x6um2 and 10x10um

2 Note the good scalability 149

Figure 52 Junction resistance versus voltage 151

Figure 53 The rectifying action of the diode at zero voltage which gives the ASPAT

high sensitivity near zero bias detection 152

Figure 54 One port S-parameter measurements 155

Figure 55 Fabricated open short de-embedding structure and real device (zoomed) use

for RF calibration and measurements (Note Images are not to scale) 156

11

Figure 56 Real S11 RF measurements for (XMBE304A) were done for four devices

from 15times15microm2 to 100times100microm2 device sizes for within-wafer uniformity check 158

Figure 57 Imaginary S11 measurements (XMBE304A) to qualify the repeatability and

reproducibility were done on four devices from 15times15 microm2 to 100times100 microm2 at zero

bias 158

Figure 58 Wafer to wafer uniformity test on 1st and 2

nd RF measurement(XMBE304B)

to qualify the repeatability and reproducibility were done for 30times30microm2 to 15times15 microm

2

device sizes (Real and Imaginary) Note blue colour is XMBE304A and red colour is

XMBE304B 160

Figure 59 RF measurement performed on device sizes of 10times10microm2 6times6mmicro

2 and

4times4microm2

(Real and Imaginary) Note that green red and blue colour represents 4times4microm2

6times6mmicro2 and 10times10microm

2 diodes respectively 162

Figure 510 RF measurement performed on device sizes of 10times10 microm2 6times6 microm2 and

4times4 microm2 (Smith Chart) Note that green red and blue colour represents 4times4microm2

6times6mmicro2 and 10times10microm2 diodes respectively 162

Figure 511 Cross-sectional of fabricated ASPAT diode and the corresponding

equivalent circuit model 164

Figure 512 The S-parameter Touchstone file is used to read the measured files 165

Figure 513 Equivalent Circuit Model of Open CPW or de-embedded structure 166

Figure 514 Equivalent circuit model for short de-embedded structure 166

Figure 515 Smith chart representative S-parameter measurement for short (left) and

open (right) CPW The blue lines represent simulated data and the red is measured data

167

Figure 516 Equivalent circuit of the ASPAT diode 169

Figure 517 Excellent fitting between measured (red colour) and simulated (blue colour)

for S11 (Real and Imaginary) results for various small device designs 169

Figure 518 Excellent fitting between measured (red colour) and simulated (blue colour)

for S11 results (Smith Chart) for various small device designs 170

Figure 519 Capacitance extraction showing full depletion at -025 volts (modelled

capacitance vs Voltage) 172

Figure 61 Block diagram represent a complete direct receiver system 177

Figure 62 The detection process of a single wave through a non-linear IV characteristic

of a diode 177

Figure 63 Lumped element illustration of microwave detector circuit 178

Figure 64 The mixing process where the signals are processed by the non-linear I-V

characteristic to produce an intermediate frequency (fIF=fLO-fRF) when signals at fLO and

fRF are applied to the diode 179

Figure 65 Measurement of Tangential Sensitivity[108 129] 181

12

Figure 66 The configuration circuit to verify 10th order polynomial equation extracted

from MATLAB to realize a virtual GaAsAlAs ASPAT diode 186

Figure 67 Verification of actual (blue measured) and virtual (red_10th order

polynomial) I-V characteristic of the 4times4 microm2 diode used in this study 186

Figure 68 Direct detector circuit topology using an ASPAT diode 187

Figure 69 Output voltage and detector sensitivity over wide range of input power 188

Figure 610 Simulated dependence of Voltage Sensitivity on incident power and load

resistance of the ASPAT detector 189

Figure 611 Junction resistance as a function of forward voltage 189

Figure 612 Curvature coefficient for the measured I-V characteristic with a device size

of 4times4μm2 191

Figure 613 Voltage Sensitivity against frequency at -20dBm fixed input power 192

Figure 614 Reflection Coefficient versus operating frequency without matching

circuitry 193

Figure 615 Detector circuit with impedance matching circuit placed in between diode

and source 194

Figure 616 Reflection Coefficient over wide frequency band with matching 195

Figure 617 Simulated voltage sensitivity of GaAsAlAs ASPAT detector at W-band

frequency 195

Figure 618 Lowest detectable signal at 100GHz operating frequency 196

Figure 619 The dynamic range of the ASPAT diode obtained from linear region of

diode operation 197

Figure 620 Simulation result of diode sensitivity for three devices geometries obtained

from the fabricated ASPAT in this work 198

Figure 71 performance of state-of the-art millimetre wave source [166] 202

Figure 72 Frequency multiplier has the property that foutput is an integer multiple of fin

204

Figure 73 Principle of operation for frequency multiplier utilising a non-linear

resistance [10] 204

Figure 74 A standard system for two port frequency multiplier circuit 207

Figure 75 Block diagram of frequency doubler with series ASPAT mounted diode 209

Figure 76 Empirical modelling of ASPAT doubler utilising Keysight ADS tool 210

Figure 77 Conversion loss and conversion efficiency as a function of input power 211

Figure 78 Conversion loss and Conversion efficiency as a function of output frequency

212

13

LIST OF PUBLICATIONS CONFERENCE PRESENTATIONS

PUBLICATIONS

1 MRR Abdullah Y K Wang J Sexton M Missous and M J Kelly ldquoGaAsAlAs

Tunnelling Structure Temperature Dependence of ASPAT Detectorsrdquo 8th UK-Europe-

China Workshop on mm-waves and THz Technologies 2015 Cardiff University IEEE

proceedings DOI 101109UCMMT20157460591

2 Yuekun Wang Mohd Rashid Redza Abdullah James Sexton and M Missous

ldquoInGaAs-AlAs asymmetric space layer tunnel (ASPAT) diodes for THz electronicsrdquo 8th

UK-Europe-China Workshop on mm-waves and THz Technologies 2015 Cardiff

University IEEE proceedings DOI 101109UCMMT20157460589

3 K N Zainul Ariffin S G Muttlak M Abdullah M R R Abdullah Y Wang and M

Missous ldquoAsymmetric Spacer Layer Tunnel In018Ga082AsAlAs (ASPAT) Diode using

Double Quantum Wells for Dual Functions Detection and Oscillationrdquo 8th UK-

Europe-China Workshop on mm-waves and THz Technologies 2015 Cardiff

University IEEE proceedings Doi 101109UCMMT20157460599

4 K N Zainul Ariffin M R R Abdullah Y K Wang S G Muttlak O S

Abdulwahid J Sexton MJ Kelly and M Missous ldquoAsymmetric Spacer Layer Tunnel

Diode (ASPAT) Quantum Structure Design Linked to Current-Voltage Characteristics

A Physical Simulation Studyrdquo UK-China Millimetre Waves and Terahertz Technology

Workshop September 2017 Submitted 14 July 2017 Conference held on 11th -13th

September 2017 DOI 101109UCMMT20178068358

5 K N Zainul Ariffin Y Wang M R R Abdullah S G Muttlak Omar S

Abdulwahid J Sexton Ka Wa Ian Michael J Kelly and M Missous ldquoInvestigations of

Asymmetric Spacer Tunnel Layer (ASPAT) Diode for High-Frequency Applicationsrdquo

DOI 101109TED20172777803

6 Omar S Abdulwahid S G Muttlak M R R Abdullah K N Zainul Ariffin J

Sexton Ka Wa Ian Michael J Kelly and M Missous ldquoA 100GHz Zero-Biased

Quantum Tunnelling ASPAT Detectorrdquo Submitted to IEEE TED on DEC 2016 under

correctionamendment Pending fabrication data

14

CONFERENCE PRESENTATIONS

1 Mohd Rashid Redza Abdullah J Sexton Kawa Ian MJKelly and M Missousldquo

G2040GHz Frequency Doubler Varistor Mode using ASPAT diodesrdquo UK

Semiconductors 2017 2017 University of Sheffield Oral presentation

2 M R R Abdullah YueKun Wang J Sexton Kawa Ian and M Missousldquo

Microwave Performance of GaAsAlAs Asymmetric Spacer Layer Tunnel (ASPAT)

Diodesrdquo UK Semiconductors 2016 2016 University of Sheffield Oral presentation

3 M R R Abdullah J Sexton and M Missousldquo GaAsAlAs Tunnelling Structures

THz RTD oscillators and ASPAT detectorsrdquo UK Semiconductors 2015 2015

University of Sheffield Oral presentation

4 Yuekun Wang Mohd Rashid Redza Abdullah and M MissousldquoInGaAs-AlAs

asymmetric space layer tunnel (ASPAT) diodes for THz electronicsrdquo UK

Semiconductors 2015 2015 University of Sheffield Oral presentation

5 Mohd Rashid Redza Abdullah and M Missousldquo GaAsAlAs Tunnelling

Structure Temperature Dependence of ASPAT Detectorsrdquo PGR Conference2016

2016 University of Manchester Poster presentation

6 YueKun Wang KNZainul Ariffin Mohd Rashid Redza Abdullah J Sexton

Kawa Ian and M Missous ldquoPhysical Modelling and Experimental Studies of

InGaAsAlAs Asymmetric spacer Layer Tunnel Diodesrdquo UK Semiconductors 2016

2016 University of Sheffield Oral presentation

7 K N Zainul Ariffin S G Muttlak M R R Abdullah Y Wang Omar S

Abdulwahid M Missous ldquoExperimental and Physical Modelling of Temperature

Dependence of a Double Quantum Well In018Ga082AsAlAs ASPAT Dioderdquo UK

Semiconductor Conference 2016 Sheffield 6 ndash 7 July 2016 Oral Presentation

8 Omar S Abdulwahid Mohd Rashid Redza Abdullah S G Muttlak K N Zainul

Ariffin Mohamed Missous ldquoTunnelling Barrier Diode for Millimetre Wave

Mixingrdquo UK Semiconductor Conference 2016 Sheffield 6 ndash 7 July 2016 Oral

Presentation

9 M Abdullah K N Zainul Ariffin MRR Abdullah J Sexton M Missous and

MJ Kelly ldquoA Novel In18Ga82As-AlAs Asymmetric Spacer Layer Tunnel (ASPAT)

Diode with Double Quantum Wells for Microwave Detectionrdquo UK Semiconductor

Conference 2015 Sheffield 1 ndash 2 July 2015 Oral Presentation

15

ABSTRACT

Thesis Title GaAsAlAs ASPAT Diodes for Millimetre and Sub-Millimetre Wave

Applications

Institute School of Electrical and Electronic Engineering the University of Manchester

Candidate Mohd Rashid Redza bin Abdullah

Degree Doctor of Philosophy (PhD)

Date 3 October 2017

The Asymmetric Spacer layer Tunnel (ASPAT) diode is a new diode invented in

the early 90s as an alternative to the Schottky barrier diode (SBD) technology for

microwave detector applications due to its highly stable temperature characteristics The

ASPAT features a strong non-linear I-V characteristic as a result of tunnelling through a

thin barrier which enables RF detection at zero bias from microwaves up to

submillimetre wave frequencies In this work two heavily doped GaAs contact layer on

top and bottom layers adjacent to lightly doped GaAs intermediate layers enclose

undoped GaAs spacers with different lengths sandwiching an undoped AlAs layer that

acts as a tunnel barrier The ultimate ambition of this work was to develop a MMIC

detector as well as a frequency source based on optimized ASPAT diodes for millimetre

wave (100GHz) applications The effect of material parameter and dimensions on the

ASPAT source performances was described using an empirical model for the first time

Since this is a new device keys challenges in this work were to improve DC and

RF characteristic as well as to develop a repeatable reproducible and ultimately

manufacturable fabrication process flow This was investigated using two approaches

namely air-bridge and dielectric-bridge fabrication process flows Through this work it

was found that the GaAsAlAs heterostructures ASPAT diode are more amenable to the

dielectric-bridge technique as large-scale fabrication of mesa area up to 4times4microm2 with

device yields exceeding 80 routinely produced The fabrication of the ASPAT using i-

line optical lithography which has the capability to reduce emitter area to 4times4microm2 to

lower down the device capacitance for millimetre wave application has been made

feasible in this work The former challenge was extensively studied through materials

and structural characterisations by a SILVACO physical modelling and confirmed by

comparison with experimental data The I-V characteristic of the fabricated ASPAT

demonstrated outstanding scalability demonstrating robust processing A fair

comparison has been made between ASPAT and SBD fabricated in-house indicating

ASPAT is extremely stable to the temperature The RF characterisations were carried out

with the aid of Keysight ADS software

The DC characteristic from fabricated GaAsAlAs ASPAT diodes were absorbed

into an ADS simulation tool and utilized to demonstrate the performance of MMIC

100GHz detector as well as 20GHz40GHz signal generators Zero bias ASPAT with

mesa area of 4times4microm2 with video resistance of 90KΩ junction capacitance of 23fF and

curvature coefficient of 23V-1

has demonstrated detector voltage sensitivity above

2000VW while the signal source conversion loss and conversion efficiency are 28dB

and 03 respectively An estimate noise equivalent power (NEP) for this particular

device is 188pWHz12

16

DECLARATION AND COPYRIGHT STATEMENT

No portion of the work referred to in the dissertation has been submitted in support of an

application for another degree or qualification of this or any other university or other

institute of learning

COPYRIGHT STATEMENT

i The author of this thesis (including any appendices andor schedules to this thesis) owns

certain copyright or related rights in it (the ldquoCopyrightrdquo) and he has given The University of

Manchester certain rights to use such Copyright including for administrative purposes

ii Copies of this thesis either in full or in extracts and whether in hard or electronic copy

may be made only in accordance with the Copyright Designs and Patents Act 1988 (as

amended) and regulations issued under it or where appropriate in accordance with licensing

agreements which the University has from time to time This page must form part of any

such copies made

iii The ownership of certain Copyright patents designs trademarks and other intellectual

property (the ldquoIntellectual Propertyrdquo) and any reproductions of copyright works in the thesis

for example graphs and tables (ldquoReproductionsrdquo) which may be described in this thesis

may not be owned by the author and may be owned by third parties Such Intellectual

Property and Reproductions cannot and must not be made available for use without the prior

written permission of the owner(s) of the relevant Intellectual Property andor

Reproductions

iv Further information on the conditions under which disclosure publication and

commercialisation of this thesis the Copyright and any Intellectual Property andor

Reproductions described in it may take place is available in the University IP Policy

(httpdocumentsmanchesteracukDocuInfoaspxDocID=487) in any relevant Thesis

restriction declarations deposited in the University Library The University Libraryrsquos

regulations (httpwwwmanchesteracuklibraryaboutusregulations) and in The

Universityrsquos policy on Presentation of Theses

17

ACKNOWLEDGEMENTS

First and foremost all gratefulness and praise is to Allah swt for everything in my

life He is the one and the only one who granted me knowledge health patience and

ability to complete this thesis as well as colouring the whole journey of my PhD

I give my deepest and sincere gratitude to my PhD supervisor Professor Mohamed

Missous for his time support patience and guidance throughout the journey of this PhD

studies His encouragements valuable advice precious ideas and a wealth of knowledge

amp experiences have had a direct inspiration on this research Special thanks also to our

experimental officer Dr James Sexton for not only sharing his knowledge advice and

semiconductor fabrication skills but also his effort in maintaining our clean room

facilities to a great level My gratitude also extends to Mr Mallachi McGowan for his

help and assist in the lab-related issue

I am also obligated to Prof MJ Kelly from University of Cambridge and Dr Kawa

Ian from ICS limited for their measurement of the ASPAT samples on realizing the RF

characteristics This collaboration effort can hopefully last longer in designing and

implementing the ASPAT MMIC detectors

My deepest appreciation also goes to my PhD colleagues Khairul Nabilah Saad

GMuttlak Omar AbdulWahid and Yuekun Wang for their support as well as working

together with me to realize this exciting project directly and indirectly A sincere

thankfulness similarly to my seniors Dr Md Adzhar Zawawi and Dr Fauzi Packeer for

their support during the first and second year of my research For other friends and staff

members under Prof Missous and Dr M Migliorato I will always remember the strong

bond and friendship we made

I am really fortunate that I been blessed with my motherrsquos care who always make doarsquo

for my success every day during my studies As for my beloved wife Dr Nik Maryam

Anisah Nik Mursquotasim who had always encouraged me supported me and gave me

patience through all the hardship in this journey thank you very much

Finally I also would like to thank and acknowledge my sponsor Majlis Amanah

Rakyat (MARA) for financially supporting me during this studies I am greatly indebted

with your kind support which was vital to my study

18

DEDICATION

This thesis dedicated to

My respected and beloved parentshellip

My loving wife dearest siblings and in-lawshellip

19

1 INTRODUCTION

11 Background

It is an undeniable fact that semiconductors have changed the world much further

than anything people could have predicted in the last 60 or 70 years ie after the lsquocats

whisker and vacuum tube eras This field of research has been expanding from year to

year starting from the discovery of the first semiconductor (silver sulfide) in 1833 by

Michael Faraday [1 2] and it still remains very active to the present Semiconductors

have a large range of applications and are not just limited to use in communications they

can be found everywhere in other applications from Earth to space The widespread

usage and sheer number of applications have led to it growing very quickly and

contributing greatly to the growth of World Economics Over time the successful

development of semiconductor growth techniques such as Molecular Beam Epitaxy

(MBE) has enabled researchers to tailor and precisely control the semiconductor

material for new electronic devices with extra functionalities This has led to the

development of advanced devices such as high electron mobility transistors (HEMTs)

and Heterojunction bipolar transistors (HBTs) for use in wireless communication

technology Given this development today electronic devices such as computers

handheld smartphone tablets etc are no longer perceived as luxury and attractive items

but rather have become crucial in everyday life Such devices provide the means to allow

for people to remain connected to each other via the sending and receiving of

information electronically The huge demand for such types of devices has resulted in

competition in both the electronic market and technologies which only goes on to

advance the semiconductor industry

Nowadays the demand for electronic devices characterised by high speed high

efficiency ultra-low power and low manufacturing cost has increased exponentially To

fulfil this growth in demand high data rate systems are required in other words the

system must work at a higher frequency for both the transmitter and receiver The

frequency of interest for advanced wireless communication is in the Millimetre and sub-

20

millimetre wave region which is around 30-300GHz and 300 - 3000GHz respectively

The second frequency region is also sometime known as the terahertz (THz)

electromagnetic region This band lies between the microwave and infrared frequency

bands From the first time it was revealed in the late 80s[3-5] the THz region has gained

a lot of international attention due to its unique properties and since then the motivation

to develop these devices has increased significantly To date the THz frequencies region

has shown its ability to fulfil various applications such as high-resolution imaging in

medical security and surveillance field atmospheric monitoring and environment radio

astronomy as well as compact range radars[3] to name a few

However despite these developments not much effort has been made in exploring

alternative compact THz devices As a result electronic THz devices are still in the state

of immaturity as compared to microwave and photonics devices This is due to their high

cost and absence of compact amp solid-state THz sources (oscillators) and receivers

(detectors) that are capable of operating at both room and extreme temperatures[6 7] A

great deal of work still continues to fill up the lsquoTHz gaprsquo (between 300GHz and 3THz)

used for the most important part of a communication system namely the front-end

receiver or first stage Such a system is responsible for receiving detecting and

processing the received signal to be translated into useful information Furthermore THz

receivers systems still require the best-integrated components such as source mixer and

detector to reach their complete competencies[8] The detector which remains the

critical part of the receiver system requires devices or components that are able to fulfil

the THz gap requirement Studies conducted over a number of years have found out that

the key element in improving THz detection relies upon the use of passive devices ie

diodes Based on these findings many types of diode ie tunnelling diode point-contact

diode and Schottky barrier diode (SBD) have been proposed for detection applications

Amongst microwave and millimetre wave detector diode devices the Schottky

Barrier Diode (SBD) is the dominant detector that has been used since the 1940s[9] The

reason for this dominance is the ease of fabrication of a SBD (by either a point-contact

or evaporated semiconductor-metal structure) and its ability to produce a non-linear

current-voltage (I-V) characteristic which is necessary for rectifyingdetecting diodes [9

10] SBDs also have high cut-off frequency good dynamic range and are low cost To

21

date the SBD has been able to detect signals up to 100GHz [11] 1THz[12] and as high

as 10 THz[13] However the current transport mechanism in a SBD relies on thermionic

emission and therefore is strongly dependent on temperature and means that using them

in extreme conditions ie military and automotive applications is complex The SBD

also suffers from high noise figure[14] and is susceptible to burnout at a modest pulse

power level this will limit the use of ultra-high frequencies and low power signal

applications Other diodes that share the same characteristics are Planar Doped Barrier

(PDB) Germanium Backward Diode (GBD) ie a type of Esaki tunnel diode These

diodes are well known and are reliably used as millimetre wave detectors However it

still proves inefficient to substitute the SDB with any of the previously mentioned

diodes This is due to some drawbacks such as strong temperature dependence limited

dynamic range fabrication complications and hence poor reproducibility (ie GBD) and

other circuit complexities

Hence there is strong compulsion to study examine and produce new detector

diode structures that are able to solve the mentioned diodes limitations and which have

high sensitivity larger dynamic range low noise strong independence to temperature as

well as being able to work efficiently in the high-frequency band and at zero bias The

advantages of working at zero-bias relates very much to the need for a system with less

power consumption so that the device (ie mobile communication) is able to run off

small batteries for a reasonable length of time eliminating extra biasing circuit as well as

noise Therefore a new tunnelling device namely the Asymmetrical Spacer Layer

Tunnel diode (ASPAT) developed by RT Syme [15 16] and refined by Missous et

al[17] has been examined in this work The ASPAT which is in essence a

Semiconductor-insulator-semiconductor structure relies on tunnelling through a barrier

to provide current compared to conventional thermionic emission in SBDs The ASPAT

diode has many advantages a zero bias turn-on voltage very weak sensitivity to changes

in temperature (due to tunnelling) very low noise large dynamic range high resistance

to pulse burn-out [18] and as demonstrated recently can be reproducibly

manufactured[17] The growing interest in THz frequencies nowadays makes the

ASPAT an excellent choice to fulfil all requirements for ultra-high speed applications

22

ie communication (mobile computer networking) radar (military equipment) scalar

analyser and built-in test equipment

In this work an ASPAT diode based on group III-V elements of the periodic table

comprising compound semiconductors of large band gap material Aluminium Arsenide

(AlAs) sandwiched between two lower bandgap Gallium Arsenide (GaAs) are used and

intensively examined The AlAs semiconductor which is ten-monolayer thick has

almost the same lattice constant as GaAs but has a larger bandgap Consequently in the

conduction band a thin barrier of the AlAs is formed from the arrangement of such

structure The structure is made up of GaAs and AlAs both materials are grown on

GaAs substrate using Solid Source Molecular Beam Epitaxy (SSMBE) Therefore in

this study the ASPAT diode will be referred to as ldquoGaAsAlAs ASPATrdquo diode The

conventional GaAsAlAs ASPAT diode has been developed and successfully fabricated

in two different stages This work was the first carried out using facilities provided by

the University of Manchester The first stage of the work was to qualify the

reproducibility and repeatability of growth and fabrication technique which is mostly

performed on larger emitteranode size The second was to develop conventional

ASPAT diodes that can perform at Millimetre and sub-millimetre wave frequencies and

which are comprised of small emitter area

Prior to this work full physical modelling using SILVACO design software was

undertaken to generate models and to fully characterise and identify the fundamental

physical phenomenon of multi-junction ASPAT diode Therefore insight into and

performance based on diode structure and electron movement can be understood and

predicted which lead to the crucial idea in helping and advising iterations to epitaxial

growth as well as diode fabrication The verification of the physical models must be set

as a priority goal by comparing the results of statistically fabricated measured data The

advantage of physical modelling is that it can help reduce materials resources cost and

fabrication time

Further research into the field has led to the development of two other types of

ASPAT diodes that are used to compare with the conventional GaAsAlAs ASPAT

diode Their configuration involved the use of a more advanced semiconductor

technology which comprises InGaAsAlAs materials and GaAsAlAs with InGaAs

23

quantum wells The latter was a novel ASPAT diode and the former is identified as

advanced ASPAT diode However these two advanced ASPAT diodes have not been

extensively studied in this thesis as they will be covered by other co-workers at

Manchester Hence due to these some important parameters are compared to the

conventional one as it is the main focus of this work In the case of temperature

dependent studies the DC characteristic of conventional ASPAT is compared to in-

house fabrication AuGaAs SBD All the ASPATs epitaxial layer materials structures are

shown in the following tables

Table 11 Sample XMBE304 XMBE307 and XMBE368 GaAsAlAs materials structure grown

on GaAs Substrates by MBE

Thickness (Aring)

Layer Material Doping (cm-3

) XMBE368 XMBE304 XMBE307

Top Ohmic1 GaAs (Si) 4times1018

~3000 ~3000 ~3000

Buffer 1 GaAs(Si) 4times1017

~50 ~400 ~50

Spacer1 GaAs NID 50 50 50

Barrier AlAs NID 28 28 28

Spacer 2 GaAs NID 1000 2000 1000

Buffer GaAs(Si) 4times1017

50 400 50

Bottom Ohmic GaAs(Si) 5times1018

~3500 ~4500 ~3500

Substrate GaAs (Si) 50000 50000 50000

Note that sample XMBE368 and XMBE304 are grown on doped GaAs

substrates Sample XMBE368 was grown un-rotated to study the effect of barrier

thickness variation

24

Table 12 Quantum wells sandwiching the quantum barrier for sample XMBE314 grown on a

GaAs substrate by MBE

XMBE314

Layer Material Doping (119836119846minus120785) Thickness (Aring) Bandgap (eV)

Ohmic Layer GaAs (Si) 4times1018

3000 142

Emitter GaAs (Si) 1times1017

400 142

Spacer GaAs Undoped 50 142

Quantum Well In18Ga82As Undoped 60 116

Barrier AlAs Undoped 28 283

Quantum Well In18Ga82As Undoped 60 116

Spacer GaAs Undoped 2000 142

Collector GaAs (Si) 1times1017

400 142

Ohmic Layer GaAs (Si) 4times1018

4500 142

Substrate GaAs

Table 13 Novel InGaAsAlAs ASPAT diode grown on InP substrate by MBE on InP substrate

XMBE326

Layer Material Doping (cm-3

) Thickness (Aring) Bandgap (eV)

Top Ohmic1 In053Ga047As(Si) 5times1019

3000 075

Buffer 1 In053Ga047As(Si) 1times1017

350 075

Spacer1 In053Ga047As NID 50 073

Barrier AlAs NID 283 283

Spacer 2 In053Ga047As NID 2000 075

Buffer In053Ga047As(Si) 1times1017

350 075

Bottom Ohmic In053Ga047As(Si) 15times1019

4200 075

Substrate InP (Si) NID 620000

From the above tables it can be noted that XMBE307 is the first batch that was grown

in-house using a Riber V100HU MBE machine followed by XMBE368 XMBE 304

XMBE314 and finally XMBE326 The two earlier batches were grown on n+

substrate hence their fabrication process flow is simpler On the other hand the three

other batches were grown on semi insulating substrate thus requiring the development

of new repeatable reproducible and robust process flow which will be covered in this

thesis

25

In general the fabrication of the ASPAT diode is based on top-down processes this

is because the ASPAT is a vertical structure device and the junction capacitance of the

ASPAT is directly associated with the size of the anodeemitter area Further to these

the capacitance directly influences the diode cut-off frequency Therefore the simplest

way to reduce the capacitance is by reducing the lateral area of the device of the ASPAT

structure since it can be represented by a parallel plate capacitor where the capacitance

is inversely proportional to the area of the device In order to achieve high cut-off

frequencies minimising capacitance via small dimensions ie sub-micrometre level is

essential However this will also increase the series resistance of the diode As a result

the cut-off frequency will be degraded Thus there is a trade-off between small

dimension of device and high cut-off frequency to be achieved

Finally successful growth and fabrication for small area GaAsAlAs ASPAT diode

in this work has led to carefully extracted RF characteristics This becomes a stepping

stone to designing a millimetre wave integrated circuit (MMIC) detector using empirical

modelling in Keysight ADS tools Therefore a predicted performance for a 4times4 um2

fabricated ASPAT is that can operate at 100GHz ASPAT as a zero-bias detector with a

voltage sensitivity of over 2000VW Additionally the design of a millimetre wave

source using similar ASPAT diodes was also carried out The performance of a 2040

GHz doubler using GaAsAlAs ASPAT in varistor mode is demonstrated for the first

time with a conversion loss of 33dB and conversion efficiency of ~ 02

26

12 Aims and objectives

The aim of this study is to further improve the performance of microwave and

millimetre wave technology by incorporating the Asymmetrical Spacer Layer Tunnel

Diode (ASPAT) for ultimate operation near THz frequencies by designing a range of

low power high-speed devices enhancing the methods of Simulation layout and

materials amp structural characterisations with fabrication process optimization using the

facilities available at the University of Manchester

There are three main objectives in this research firstly to streamline the physical

device design and modelling using the GaAsAlAsGaAs materials systems in order to

produce a zero bias detector which is basically a rectifier of a microwave signal by

using the SILVACO Atlas simulation tools

Secondly to achieve reproducibility and manufacturability of the fabrication

process for new type of GaAsAlAs ASPAT structure (lateral structure) hence small size

ASPAT emitter by improving the device processing technique and maximising the

capability limit of the conventional i-line optical lithography that is available in Prof

Missousrsquos group laboratory

Thirdly to optimise DC parameters through electrical properties investigation as a

stepping stone to the next objective that is to characterise the RF performance of the

GaAsAlAs ASPAT detector circuit The detection properties of microwave and

millimetre wave diode will also be investigated with different ASPAT diode size at

100GHz Further to these the properties of microwave signal source will also be

developed by way of utilizing the non-linearity feature of the diode Therefore this new

type of tunnelling diode can be applied to both applications of signal detection and

signal source in the microwave and millimetre wave ranges

27

13 Outline of this Thesis

This thesis is organized into eight chapters The first chapter discusses the

contextual information that motivates the undertaking of the study An overview of the

work which includes the details of the studied samples the aim and objectives of the

whole research project are also outlined in this chapter

Chapter 2 deals with the literature review of the basic principles and concepts of

the group III-V compound semiconductors The historical background of such

semiconductors which is essential to the development of ternary structures etc and the

advancement of semiconductor materials engineering is presented The types of existing

tunneling diode as well as conventional microwave diodes are also discussed and

compared The fundamentals of ASPAT diode which includes structural parameters and

its operation are then explained in detail Finally discussions of the ASPAT key DC

characteristics which are important for detection purposes are presented

Chapter 3 focusses on the development of the experimental techniques which can

be divided into two stages In the first stage the development is towards repeatability

reproducibility and manufacturability of the ASPAT grown in-house by MBE and

fabricated by conventional i-line optical lithography The second stage involves

optimisation and fine tuning such fabrication method for GaAsAlAs ASPAT samples

that can operate at high frequency ie 100GHz detector For both stages of the

fabrication process all techniques including mask design generic and special process

flow are presented The chapter ends with discussions on issues related to sample

processing and improvements that are proposed to solve these issues

Chapter 4 dwells on the modelling of the GaAsAlAs ASPAT using the SILVACO

simulation package The discussions are expected to offer a better understanding or

insight into each layer that forms the ASPAT diode structure The chapter begins with

discussions of the operation of the SILVACO Atlas tool A validation of physical

modelling is essential and presented according to the fabricated mesa sizes of the diode

28

Thereafter towards the end of the chapter the analyses of the relationship between

device current-voltage (I-V) characteristics the structural parameter including various

temperatures dependent simulations with a comparison to an in-house fabricated SBD

are offered

Chapter 5 presents relevant DC results based on optimized fabrication process and

RF characterization which enable obtaining an intrinsic and extrinsic element of the

GaAsAlAs ASPAT diode The discussions also highlight the analysis of DC zero bias

equivalent circuit and de-embedding extraction using ADS The chapter ends with

discussions on the RF reproducibility performance which includes the performance as

well for millimeter-wave and sub-millimeter wave applications

Chapter 6 discusses the main applications of ASPAT diodes The chapter begins

with discussions on detection theory followed by the parameters of interest and ends

with circuit design as well as the performance of a 100GHz detector The circuit design

was conducted using Keysight ADS software via harmonics balance simulation tool The

performance in term of sensitivity depending on measured ASPAT emitter size is

demonstrated Finally a comparison with conventional Schottky diode is presented

towards the end of the chapter

Chapter 7 discusses a secondary application that can be applied to the ASPAT by

utilizing the nonlinearity feature of the diode to create a signal source namely a 20 to

40GHz frequency doubler in varistor mode The doubler performance of ASPAT will be

explored through circuit design constructed via Keysight ADS simulation software Each

key parameter is highlighted and discussed in detail

The final chapter of this thesis that is Chapter 8 discusses the conclusions of the

study with emphasis on the overall key research findings The chapter also highlights

suggestions for further research in this particular field of study

29

2 LITERATURE REVIEW

21 Introduction

Since 1940s the development in the technology of semiconductor electronics has

been expanding and now has led to the establishment one of the most astonishing

industries of the 3rd

-millennium era Leading this advancement is the integrated circuit

(IC) or chip which was driven mostly by silicon (Si) Overtime the IC has undergone

substantial revolution in term of power economics size and efficient energy

consumption Currently it covers every aspect of human life ie from desktop personal

computers in the office and house to the compact smartphone in the pocket and from a

gigantic satellite in space to small satellite navigation in cars In other words

semiconductor technology is crucial to human life Without developments in

semiconductor materials engineering and shrinking of device size such accomplishment

may not have been realised today Therefore this chapter presents a macro view of the

development in compound semiconductor technology especially in radio frequency (RF)

towards Millimetre and submillimetre wave applications with regard to the improvement

of material and device structures

The essentials of group III-V compound semiconductor will be emphasised for its

points of interest and application in this field (RF technology) This chapter comprises

five main sections The first section is an overview of the semiconductor history with

concentration on its advantages and applications in the RF field while the second and

third discuss the effects of III-V compounds when the interface occurs between

semiconductor-semiconductor and semiconductor-metal respectively which leads to a

basic understanding of hetero-structures device as well as contacts namely Schottky and

Ohmic The fourth section is predominantly concerned with high-speed devices ie

diodes and materials in this field which leads to the exploitation of the main researchrsquos

device Then the following section describes in detail the background works basic

principle and intrinsic amp extrinsic parameters of the Asymmetric spacer Tunnel Diode

(ASPAT) Finally the basic way of characterising the device will also include giving an

overview of how the device is measured and what parameters are needed

30

22 Historical review of III-V Compound Semiconductor for RF applications

The beginning of commercial electronic devices was marked with the first point-

contact semiconductor transistor developed in 1947 by William Shockley at Bell

Laboratories in New Jersey Shockley developed a device based on a Germanium

Bipolar Junction Transistor (Ge BJT) structure [19] with operating frequency above 1

GHz Since then and until early 1950s the development of Ge BJT was fast and it

became foremost in the market of semiconductor technology However the emergence

of Silicon (Si) challenged Ge in the market in the 1960s Si has the upper hand primarily

because it has better electron transport and low manufacturing costs compared to Ge

[19] By the 1970s almost all RF transistors were based on Si BJT Additionally the

development of Si which forms a new material from the formation of native oxide

namely Silicon Dioxide (SiO2) led to the invention of the Metal Oxide Semiconductor

Field Effect Transistor (MOSFET) [19] The future of digital electronic industries has

been ldquobrightrdquo ever since the MOSFET was ldquobornrdquo as it has become a fundamental

building block component in complex microprocessors and flash memories Despite this

development the exploitation of Si at RF frequencies did not last long since Si is not an

optimum semiconductor for RF electronic devices The emergence of GaAs has

improved RF applications for high-speed transistors

Ge and Si which are categorised as single element semiconductor are the earliest

materials used to build the first transistor devices These devices played a crucial role

towards the development of more advanced material such as GaAs of the compound

semiconductor type[20] A compound semiconductor is a semiconductor formed by the

ionic bond of different types of semiconductor material most widely known as the group

III-V compound semiconductors The main reason for the progression of the III-V

compound semiconductors is due to their better electron mobility compared to the single

element semiconductors The term ldquomobilityrdquo in the semiconductor industry refers to the

easiness of movement of charges in many directions inside a crystal In fact it is

determined by the access resistances values with saturated velocity under certain values

of electric fields (bias) the higher the electric field the faster is the carrier movement in

the crystal Figure 21 shows the electron mobility and band gap for the most common

31

group III-V compound semiconductors Besides higher mobility III-V compound

semiconductors also have light-emission capability and are suitable for bandgap-

engineering techniques

The work on III-V compound semiconductors mainly on GaAs FETs led to a new

change for the whole RF electronics industry For example in 1966 the first GaAs

MESFET was invented[21] and achieved a maximum operating frequency of operation

of 3GHz [22] Three years later the frequency increased to 30GHz [23]

Figure 21 III-V compound semiconductors mobility and band gap[24]

With better features in terms of having a higher electron mobility compared to Si

electronic devices based on III-V materials developed rapidly This attracted attention in

many aspects especially in military radar application electronic warfare system missile

guidance control electronic for smart warfare system and secure communication To be

specific those demands were fulfilled through the application of microwave mixer and

detectors [25] which were achieved based on Schottky barrier diodes and FETs

However these applications remained largely as niche markets for use only in military

and exotic scientific projects until 1980 In addition to the microwave industry two

important diodes that played a large role in very high-frequency power source namely

the Gunn diode and the Impact Avalanche and Transit Time (IMPATT) diode which

were discovered in the 1960s[26]

100

1000

10000

100000

0 05 1 15

Bu

lk M

ob

ility

(cm

2 V

-1 s

-1

BandGap (eV)

InS

b

InA

s

Ge

Ga

Sb

In

GaA

s

Si

GaA

s

InP

32

Furthermore the invention of Molecular Beam Epitaxy (MBE) growth technique at

the beginning of 1970s has enhanced the full potential of the III-V compound

semiconductors[27] This technique has led to the formation of a new class of materials

and heterojunction device with high-quality interfaces and accurate control of the

thickness during growth[28] The advancement of material engineering that tailored the

III-V compound semiconductor with MBE effect has been beneficial for both three-

terminal and two-terminal devices As a result of this more advanced devices in both

electronics and optics were developed such as quantum well (QW) laser Resonant

tunnelling diode (RTD) high electron mobility transistor (HEMT) and many more[29]

The aim was to achieve high-speed devices transporting data at high data rates and

robust devices These devices promised an excellent option to conventional transistor

(three terminal devices) in high-frequency systems especially in the terahertz (THz) or

Millimetre and sub-millimetre wave regions [30]

One of the promising diodes that received a lot of attention is the resonant

tunnelling diode (RTD) which was first described in 1974 by Chang [31] This device

which consists of a double barrier and one quantum well is the classical tunnelling diode

Due to its good symmetrical non-linearity in its current- voltage characteristic it can be

exploited for signal generation and detection However the main focus of RTD to date

has been in the generation of continuous wave (CW) ultra-high frequency and to a lesser

extent in detection Therefore other tunnelling based diodes were developed specifically

for detection purposes which are the main foci of this work The PDB and ASPAT

diodes are the workhorse candidates for detection purposes Most of these are built based

on group III-V compound semiconductors [32]

Unlike the Very Large Scale Integration (VLSI) market ie CMOS for personal

computer (PCs) the RF electronic device for civilian application reached the consumer

market only in the late 1980s through satellite television with operating frequencies

around 12GHz [19] Since then many RF application have been deployed on the mass

market depending on their operating frequency such as 09GHz ndash 25GHz for wireless

communication 20GHz to 30GHz for satellite communication 77GHz for car radar

systems and above 90GHz for different sensor applications Utilising GaAs as the main

material RF devices have become the key underpinning components for modern

33

communication systems As a result in 1998 the volume production of mobile phones

was greater than that of PCs for the first time in history Presently production is being

made for devices like smartphones cellular phones mobile internet access and new

communication services and tablets

The development of the RF field is never ending More and more improvements are

being made especially through the design and fabrication of oscillators and detectors

which are mainly built based on group III-V compound semiconductors When RF

devices were used by the military (in the 1970s to 1980s) cost was not a concern

However after getting into civil application market (ie in the 1990s) the most frequent

issues highlighted were performance and cost[19]

The ability to generate or receive high operating frequencies with high power large

bandwidth and high sensitivity is an indicator for a good performance of RF devices

(depending on specific applications) For example the highest room temperature based

oscillator of up to 186 THz was achieved in thin well AlAs-InGaAs RTD by Professor

Masahiro Asada from Tokyo Institute of Technology [33] An excellent review on THz

sources can be found in [34] For ultra-high frequency detector and mixer applications

the two terminals RF device that is mostly used is the SBD In 1996 the highest cut off

frequency achieved by a mixer utilizing the SBD was about 5THz[35] and this has kept

increasing ever since The factor that motivates the development of THz devices is the

requirement to have a compact coherent source in the THz range Undoubtedly in the

future there will be very exciting times for enthusiasts of terahertz sources and receiver

as new generations of compact broadband and tuneable solid source device based on

advanced compound semiconductor are developed

23 The Concept of Heterostructures

A III-V compound semiconductor is mostly grown on a single semiconductor

substrate forming a layer called epitaxial heterojunction layer It is a starting point and

the key feature that brings the idea of realising the most advanced semiconductor

devices currently being developed and manufactured by combining several epitaxial

semiconductors [36-38] Heterojunctions have the capability of manipulating carrier

transport ie electron and holes transport in crystal separately unlike homojunctions

34

This has resulted in the successful development of new devices for high-speed and high-

frequency applications as well as optical sources and detectors [37] This section will

discuss lattice matched material pseudomorphic material hetero-junction band

discontinuities and quantum wells

231 Homojunctions Heterojunctions and Band Discontinuities

The term homo-junction refers to the interface between identical semiconductor

materials that have different polarity ie p-type or n-type but similar in energy gap This

phenomenon is usually applied in forming p-n junction diodes and can be understood by

referring to Figure 22 below

Figure 22 illustration of Homojunctions band structure material before (left) and after (right)

equilibrium

The materials A and B which have similar bandgap (Eg) and different dopant

types ie p-type and n-type will have their Fermi levels (Ef) closer to the valence band

(EV) and conduction band (EC) respectively before ldquothermal equilibriumrdquo Once

equilibrium is achieved Ef of both p-type and n-type will be aligned causing band

bending of EC and EV As a result a built-in electric field is introduced (via diffusion of

carriers) for both holes and electrons and forcing them to move in one direction

On the contrary a heterojunction occurs when the interface between two

semiconductor materials with different bandgap energy are brought together (ie large

energy band gap material combined with a low band gap one eg wide band gap AlAs

and narrow bandgap GaAs) This results in a steep band bending which leads to the

formation of energy band discontinuities at the junction as shown in Figure 23 In a

semiconductor heterojunction the most important parameter is the band gap energy

EC

EV

Ef

Eg

p-type E

g

n-type

EC

EV

Ef

Material A

p-type

n-type

Material B Material A Material B

35

associated with each material in the structure where the degree of discontinuity can be

utilised in varying the carriers transport properties as well as the quality of the junction

depending on the interest of the designer This leads to flexibility in tailoring device

characteristics leading to vastly improved performance of the device

Figure 23 Energy band diagrams before (left) and after (after) interface combination at

equilibrium

Based on Figure 23 above Material A indicated with blue line is a large band

gap energy material and Material B highlighted with the red line is a low band gap

material EV represents the valence band EC the conduction band and Ef is the Fermi

level of the materials Alternatively the two materials band discontinuities are denoted

by ΔEC for the conduction band and ΔEV for the valence band χ and Eg represent the

electron affinity and band gap energy respectively

At some point where by the Fermi energy of both semiconductor materials are levelled

the structure would have reached its thermal equilibrium The band gap of materials A

and B have a discontinuity at the interface (ΔEg) of these two materials In general this

is given by

120549119864119892 = 119864119892119860 minus 119864119892

119861 (21)

Furthermore when thermal equilibrium is achieved ΔEg is then divided between

conduction band and valence band discontinuities (ΔEC and ΔEV respectively) at the

material A and B junction interfaces Their relationships can be expressed as

EC

A

EV

A

Ef

A

Eg

A

Material A Material B

χA

χ

B

ΔEC

ΔEV

Ef

B

E

g

B

EC

B

EV

B

EV

A

χA

ΔE

C

ΔEV

EC

A

Ef

EC

B

EV

B

χB

Material A Material B

Vacuum

Level Vacuum

Level

36

∆119864119862 = 120594119860 minus 120594119861 (22)

∆119864119881 = (119864119892119861 minus 119864119892

119860) minus (120594119860 minus 120594119861) (23)

∆119864119892 = 119864119892119860 minus 119864119892

119861 = ∆119864119862 + ∆119864119881 (24)

However these relationships which were introduced by Anderson can only offer an

approximation In practice the results are always different since dislocation and

interface strain occur at the junction Therefore precise control during epitaxial growth is

always required and growth technologies such as MBE are employed In due course the

band gap discontinuity can be further exploited by using different types of material

combination Examples are GaAsAl052Ga048As and In053Ga047As In052Al048As [39]

232 Lattice-Matched and Pseudomorphic Materials

As discussed earlier a heterojunction happens when any two different

semiconductor materials that have different bandgap are joined together At the atomic

level both materials often differ in lattice constant The easiest way to explain this is by

setting the formation of heterojunction which can be separated into two types lattice

matched and lattice mismatched (pseudomorphic)

2321 Lattice Matched Systems

To create discontinuities for use as a high-performance device the combination of

semiconductor materials is essential Selecting the appropriate materials that have

similar or very close lattice constants to combine is crucial to avoid disruption at the

atomic level heterojunction interface Figure 24 shows that the lattice constant of a

material A ie substrate (aS) and material B ie deposited over layer (aL) are identical

or very close and their surface atoms are perfectly matched This scenario is known as

lattice matching

37

Figure 24 Lattice Matching for both materials when aL=aS

As can be seen in Figure 25 while there are restricted binary materials available to

form good heterojunction interfaces it is possible to combine semiconductor materials in

binary ternary and quaternary forms to allow the formation of a variety of lattice-

matched heterojunction interfaces The examples of materials that have successfully

been alloyed are In053Ga047As In052Al048AsInP and GaAsAlxGa(1-x)As (x=0 to 1)

Even though the materials system hetero-junction of these materials has close lattice

constant value their band-gap will experience an abrupt variation

Figure 25 Energy gap and lattice constant for direct and indirect band gap of compound

semiconductor at ambient temperature[40]

Material A(layer)

Material B(substrate)

aL

aL

aS

aS

38

The existence of binary ternary and quaternary semiconductors formed by alloying

semiconductors has expanded the opportunity for heterojunction formation in devices

The alloy semiconductor which is produced by the combinations of two semiconductors

A and B has a lattice constant that obeys Vegardrsquos Law as follows

119886(119886119897119897119900119910) = 119909119886119860 + (1 minus 119909)119886(119861) (25)

For the alloy the band gap normally follows the virtual crystal approximation

119864119892(119886119897119897119900119910) = 119909119864119892(119860) + (1 minus 119909)119864119892(119861) (26)

Table 21 shows the list of the semiconductor alloy band gap and lattice constant for

common binary and ternary for group III-V compound semiconductors[41]

Table 21 Lattice constant (a) and band gap for group III-V binary and ternary compound

semiconductors a room temperature [41 42]

Alloy Lattice constant a(Aring) Band gap EgeV)

AlAs 5661 2239

AlSb 6136 1581

GaAs 5653 1424

GaN 3189 34

InAs 6058 0417

InP 5869 1344

Al052Ga048As 5657 2072

In053Ga047As 5868 0773

In052Al048As 5852 1543

39

2322 Pseudo-morphic Materials

The other scenario is when two different materials with different lattice constants

are brought into contact The observation can be made at the atomic level where the

atom will try to match each other as shown in Figure 26 below

Figure 26 Lattice mismatched material

In fact for both situations (ie lattice matched and lattice mismatched) the atom of

the material at the hetero-interface will change their position to maintain the geometry of

the lattice Due to distortion at this atomic level a strain is then induced at the hetero-

interface In order to form a good hetero-junction interface the strain must not exceed a

certain specific critical value which will cause crystal dislocations to occur The result of

crystal dislocation is generally bad as it will affect the carriers which will be

concentrated in the defect area thus degrading the carriersrsquo mobility This then makes

the overall function or performance of the device to become poor

Nowadays the Molecular Beam Epitaxial (MBE) technique is able to grow epitaxial

layers of mismatched semiconductor layers profile ie mismatched in their lattice

constant (aLneaS) The growth method works when the grown epitaxial layer assumes the

lattice parameters of the layer it is deposited on Nonetheless the layers must be kept

within a certain limit and the deposited layer must be thin enough to avoid defect or

dislocation formations This new layer known as a ldquopseudomorphicrdquo material will alter

its original crystal structure and physical properties

Material A(layer)

Material B(substrate)

aL

aL

aS

aS

40

Figure 27 The formation of Pseudomorphic-layer in two situations (a) compressive and (b) tensile

strain [1]

Figure 27 shows material A in which the pseudomorphic materials can be related to

two situations compressive and tensile strain The compressive strain occurs when the

deposited layer has a larger lattice constant than the substrate (aL gtaS) while tensile strain

happens when the deposited layer has a smaller lattice constant than the substrate

(aLltaS) These leads to aL either to compress or stretch to fit aS respectively Note that

the pseudomorphic layers can only be grown to a certain critical thickness hc From

Figure 27 the strain between the substrate and the deposited epitaxial layer is given by

휀 =119886119871 minus 119886119878

119886119878 (27)

Where Ɛ is strain between two layers aL is lattice constant of the deposited layer and

aS is lattice constant of the substrate layer The concern in deposition of the over layer is

to avoid dislocation occurring at the interface if there is too much strain at the junction

The strain is naturally influenced by the thickness of the deposited layer and thus the

thickness of growth must be controlled below the critical thickness hc which is

expresses as

(b)

41

ℎ119888 =119886119904

(28)

Moreover one needs to appreciate that even though the crystal structure and their

physical properties change the total energy within the unit cell is maintained This is

possible by distortion of the deposited layer in the direction perpendicular to the growth

direction while leading to lattice matching in the lateral plane Example of lattice

matched materials is GaAsAlAs and pseudomorphic material is In08Ga02AsInP

233 Quantum well and 2DEG

A typical application of heterojunction interface is one in which utilises ΔEC and ΔEV to

form barriers for electrons and holes One example of barriers that confines these

carriers is known as a Quantum Well (QW) A QW is a layered semiconductor usually

very thin ie about ~ 100 Aring thicknesses in which many quantum mechanical effects can

occur It is formed by a thin layer of a low bandgap energy semiconductor material eg

GaAs sandwiched between two similar large bandgap energy semiconductors eg AlAs

or AlGaAs The growth technique to achieve thin layers of QW is usually MBE The

benefit of this method is that it allows the formation of heterojunction with very thin

epitaxial layer

Figure 28 The band diagram of an ideal quantum well formed by a narrow band gap

semiconductor eg InGaAs sandwiched between two large band gap semiconductors eg GaAs (a)

Structure (b) energy band diagram and (c) Conduction band diagram when AlGaAs is n-doped[43]

42

The thickness of the layer that can be achieved can be as thin as the electron mean free

path (De Broglie wavelength) which is around 100 Aring to 300 Aring [44] The expression for

the De Broglie wavelength is given by

120582 = ℎ120588frasl (29)

Here h and ρ are Planckrsquos constant and momentum of the electron respectively Figure

28(b) illustrates a quantum well formation in abrupt semiconductor interfaces It can be

observed that the heterojunction boundary will experience discontinuities at the edges of

the conduction band and valence band with a quantum well generated for the carriers

(both electron and holes) The quantised energy sub-bands in the quantum well structure

in Figure 28(b) can be determined from [43]

119864 = 119864119899 + (

ℏ2

2119898lowast) (119896119909

2 + 1198961199102)

(210)

Where 119864119899 = (ℏ21205872

2119898lowast ) (119899

119871)2

and n is the energy level index that can be n=1 2 3hellip

The dopants in a semiconductor with large band gap layers may supply the

carrier to the quantum well and this occurs when the base or bottom of the quantum well

is lower than the Fermi Level and hence the high energy donors will go down to the

well therefore creating a Two-Dimensional Electron Gas (2DEG) In the 2DEG the

electrons and holes move freely in the quantum well in the plane perpendicular to the

growth direction however they are not capable of moving in the crystal growth

direction (confinement direction)[45 46] The 2DEG phenomena can be seen in Figure

28(c)

24 Metal-Semiconductor Contact

A semiconductor device is incomplete if there is no connection between the

semiconductors and the outside world A metal which is usually gold (Au) or gold

germanium (AuGe) is diffused into the semiconductor to allow for electrical connection

from the outside world to the semiconductor and vice versa The metal-semiconductor

contact can be either a Schottky contact or an Ohmic contact The Schottky contact is a

43

rectifying contact while the Ohmic contact is a contact that provides a low resistance

path between semiconductor and metal

241 Schottky Contact

The Schottky contact is basically a metal contact to the gate to enter a region or

channel in a transistor Figure 29 shows a schematic band diagram of a metal-

semiconductor contact before and after contact (Schottky-Mott concept)[47]

Figure 29 Metal and semiconductor in two conditions (a) separation (b) in contact

In Figure 29 the work function of the metal is represented by qm while the

semiconductor work function is qS The qχ is the energy difference of an electron

between the vacuum level and conduction band edge ie known as the electron affinity

and qVn is the difference between the conduction band and Fermi level in the

semiconductor EV EC Ef is the valence band energy conduction band energy and the

Fermi level respectively

The metal and semiconductor are brought together as showed in Figure 29(a)

both materials are at steady state However when the metal and semiconductors are in

contact as illustrated in Figure 29(b) the electrons that flow from the conduction band

in the semiconductor into the lower energy state of the metal will cause the Fermi level

to be aligned in thermal equilibrium Due to this process the positive charge donor is

trapped in the semiconductor interface hence forming a depletion region Xdep

Thereafter the upward bending of the energy in the semiconductor takes place On the

qχ(s)

Eg(s)

Vacuum Level

EV

EF(m)

E

F

EC

qϕ(m)

qϕ(s)

Metal Semiconductor

qVn

qϕB

qϕ(s)

qVbi

X

qχ(s)

Eg(s)

Vacuum Level

EV

EF

EC

qϕ(m)

Metal Semiconductor

Xdep

(a) (b)

44

other hand the negative charge (electron) will be accumulated within a narrow region in

the metal interface The existence of two different charges at the metal-semiconductor

boundary generates an electric field This leads to a potential barrier qB as seen by

electrons in the metal moving into the semiconductor and a built-in potential qVbi as

seen by electrons in the semiconductor trying to move into the metal

The built-in potential qVbi is defined as follows

119902119881119887119894 = 119902empty119861 minus 119902119881119899 (211)

The barrier height empty119861 in the ideal case is specified by the dissimilarity between a metal

work function empty119898 and electron affinity of the semiconductor

119902empty119861 = 119902empty119898 minus 119902120594 (212)

Referring to Eq (212) above the barrier height empty119861 rises linearly with the metalrsquos work

function empty119898 Nevertheless this is only in theory as the presence of localised surface

stated at the edges causes empty119861 to become unresponsive to the metal work function

Consequently Eq(211) is then reordered to match the difference in metal and

semiconductor work function Thus the new equation becomes

119902119881119887119894

= 119902(120601119898 minus 120601119904) (213)

A Schottky contact appears when a metal-semiconductor contact has a large

barrier height (B ge kT) and low doping concentration in the semiconductor (ND le NC) In

the case when the metal-semiconductor contact is under some bias eg reverse bias the

semiconductor will react to a positive bias according to the metal by a voltage V=-VR

This condition will affect the built-in potential and leads to increase from Vbi to

(Vbi+VR) thus increasing the barrier height empty119861 in the semiconductor as well

Consequently electrons are less able to flow from the semiconductor and cross into the

metal Therefore the current flow will be very small This phenomena is shown in

Figure 210(a)

45

Figure 210 Energy band diagram of Schottky contact on n-type material under (a) reverse and (b)

forward bias

As can be seen from Figure 210(b) when a forward bias is applied the semiconductor

is biased negatively with respect to the metal by a voltage V=Vf This will result in a

reduction in built-in potential from Vbi to Vbi-Vf The electrons in the semiconductor will

lower the barrier height and a lot of electrons will escape into the metal causing a large

current to flow Thus a large current flow in the forward direction compared to the

reverse direction Essentially this is the origin why the Schottky contact is named a

rectifying contact [48 49] For the metal side both forward and reverse biases applied

do not affect the barrier high empty119861 because there is no voltage drop there

In this system the electron and holes are transported by a phenomenon called

Thermionic Emission (TE) which happens when the semiconductor layer is lightly

doped Nd lt 1x1017119888119898minus3 The electron will only be thermionically emitted into the metal

when the energy is higher than the potential barrier[50] There is another phenomenon

called Thermionic Field Emission (TFE) which happens when the potential barrier

thickness is very thin (thin enough) to allow the electron to tunnel through the barrier

This will be discussed in the next section as this phenomenon leads to the formation of

an ohmic contact

(a) (b)

46

242 Ohmic Contact

Basically an ohmic contactrsquos purpose is to provide a low resistance path from

the semiconductor to the outside world It is different to a Schottky contact as it is a non-

rectifying contact and does not control the current flow the I-V characteristic of an

ohmic contact is linear in both forward and reverse directions (equality in current flow)

The ohmic contact also has a small voltage drop across it compared to the voltage drop

across the device

If a metal and semiconductor are bought together unavoidably a Schottky

contact will be formed Therefore to create an ohmic contact some techniques to reduce

barrier height and width of the depletion region must be used ie increase Nd In carrier

transport theory there are three mechanisms of carrier transport across the barrier

Firstly the Thermionic Emission (TE) which happens when the carries are excited to

overcome the barrier when the thermal energy is present Secondly the Thermionic

Field Emission (TFE) occurs when the electronholes have enough energy to tunnel

through an adequately thin barrier and some has overcome the low barrier at the top

Finally the Field Emission (FE) which results when carriers can tunnel through the

entire barrier The FE is the most favoured mechanism in the ohmic contact approach

[51]

From the three mechanisms above the current can be determined by the following

equations

1) Current in Thermionic emission (Figure 211(a))

exp (empty119861)

119896119879

(214)

2) Current in Thermionic field emission (Figure 211(b))

exp [

(empty119861)

11986400119888119900119905ℎ11986400

119896119879

] (215)

3) Current in Field Emission (Figure 211(c))

exp (empty119861)

11986400

(216)

47

Where k is the Boltzmann constant T is the temperature 11986400 is the tunnelling parameter

and is related to the doping concentration radic119873119863 The barrier height is denoted by empty119861

Figure 211 Figure 10 N-type semiconductor Ohmic contact with different doping concentration

ND (a) Low (b) Intermediate and (c) high

Figure 211 shows that the carrier transport mechanism is varied by the doping

concentration (ND) As can be seen from Figure 211(c) the doping concentration here is

the highest and influences the depletion region width Xdep to become smaller Therefore

Field Emission (FE) becomes dominant This FE method is the favourite method for

ohmic contact formation [51] and will be utilised in the fabrication carried out in this

work

In fabricating practical devices the ohmic contact is often split into two types

alloyed and Non-Alloyed The difference between the two is that the alloyed type is used

when the semiconductor is doped with a low doping concentration ie less than

1x1018119888119898minus3 while the Non-Alloyed is designed for heavily doped semiconductors with

more than1 times 1019119888119898minus3 doping

The alloyed ohmic contact requires thermal annealing to have a good performance

for electron transport In multi-layer metals one of the metals has the role of donor or

acceptor which is used to increase the doping concentration of the semiconductor If a

temperature anneals eg 420˚ Celsius is applied the metal will diffuse into the

semiconductor and carry the dopant into the semiconductor Therefore a heavily doped

region will be formed and the depletion width becomes narrow establishing the ohmic

contact The key example of this is the usage of the Gold-Germanium-Nickel (Au-Ge-

Ni) alloy where the Ge is the n-type dopant [52] which diffuses into the semiconductor

(a) (b) (c)

48

and perform atom replacement in the semiconductor ie in GaAs it replaces Ga On the

other hand the Non-Alloyed does not require any thermal annealing as it already has a

very high doping concentration and will automatically reduce the depletion region width

The Non-Alloyed ohmic contact has some advantages such as reproducible contact

reduced processing time and good uniformity [53]

25 Asymmetrical Spacer-layer Tunnel (ASPAT) diode Background work

In this section some historical background of the Asymmetrical Spacer-layer

Tunnel (ASPAT) diode is given Since this is the first thesis reporting about this new

device it is worth to mention some historical background about this tunnelling diode

The ASPAT was first proposed by a group of scientists from General Electrical

Company (GEC) in 1990[16] The works led by Richard T Syme and assisted by

Michael JKelly Angus Condie and Ian Dale initiated the idea of launching a new type

of tunnel diode The idea managed to attract the interest of many parties following the

development of resonant tunnelling diode (RTD) which earlier had shown a promising

weak temperature dependence [54] However the interest in RTD is mainly limited to

microwavesub-millimetre wave generation For THz detection the requirement is to

have a significantly asymmetric IV characteristic Given this the ASPAT which has

only a single energy barrier and most importantly weak temperature dependence and

large dynamic range would be a promising candidate for this application

The development of ASPAT is a kind of reverse engineering since it was built

purposely to replace the earliest receiver diode especially the Schottky Barrier Diode

(SBD) which has strong dependence on operating temperature [55] From the time when

it was first revealed a lot of works have been done to realise this most sophisticated

tunnel diode The first attempt which was reported in [56] was meant to gather some

insights into the device by using the well-known Schrodinger and Poissonrsquos equations

for simulation The second attempt on the other hand was directed to physically grow

and fabricate the device Here the real problem occurs At the first stage of qualifying

this device it was found not to be manufacturable Since then a new tunnel diode

structure based on GaAsAlAs materials system was built by both MBE and MOCVD

Its microwave performance was then tested at 94GHz [18] The same paper also

49

reported performance comparisons between ASPAT and another microwave diode ie

Germanium Backward Diode (GBD) PDB and SBD

Work on these devices stopped due to the inability to commercialise the ASPAT

and other tunnel based devices [57-60] The problems associated with low-cost

manufacture of tunnel diodes are due to firstly the thickness of the AlAs barrier layer

the dependence of tunnelling probability (electron) through a single barrier is

exponential and varies by a factor of more than 350 for one monolayer change in the

AlAs barrier thickness[61] The tunnelling of the electron through a barrier is

proportional to the current through a barrier as a function of a bias across the AlAs

barrier[62] Secondly the bandgap which is predominantly happens to be a ternary alloy

with relative composition ie AlxGa1-xAs Here the x can vary the bandgap in the

semiconductor layer For the ASPAT a 1 change in x results in a 30 change in the

current[62] To design an ASPAT for microwave and THz applications the designer

often allows at most plusmn10 variation of the absolute current through a specific diode at a

pre-identified bias This implies that within a wafer the uniformity that the ASPAT must

achieve is less than plusmn01 monolayers while between wafer to wafer the reproducibility

in barrier thickness in average must be identically controlled[63] This explains why at

the qualification stage of investigating ASPAT there was a need to focus on

GaAsAlAs-based material to diminish further errors because of the change in x This

type of work on ASPATs has been carried out by other co-workers at Manchester and

Cambridge

Thereafter the work then focused on repeatability and manufacturability tests

These result in many attempts being carried but failed with unacceptable between wafer

to wafer reproducibility [61 64] The development of reproducibility and repeatability of

the ASPAT was pursued for over 10 years until precise control of the growth of the

thickness AlAs layer was finally achieved using MBE[65] [66 67] This achievement is

confirmed by a current density produced which varied by less than plusmn30 indicating that

the reproducibility of AlAs barrier of the order of plusmn 02 monolayers A final step to

achieve the level control for the ASPATrsquos AlAs barrier thickness was carried out and

resulted in a 1 standard deviation of the IV characteristics for both within a wafer and

different wafer (2 inches wafer size)[17] In the early stage of this work some

50

repeatability and manufacturability test was also carried out Once this vital step is

accomplished further investigation on the ASPAT was made most recently and which

will be covered in this thesis ie temperature independence[68] and new ASPAT

types[69] characterisation to achieve smaller device RF measurement and development

of THz detectors The material systems that have been investigated so far are

GaAsAlAs and InGaAsAlAs both in the ManchesterCambridge group

Recently the ASPAT was commercialised by Linwave Technology as a wideband

zero-bias detector diode This was done in April 2016 Although it is now on the market

the ASPAT remains immature in term of research and development A lot of work is still

required to enhance the device ie working at the sub-millimetre wave using ternary

material etc

26 Asymmetrical Spacer-layer Tunnel (ASPAT) diode The Basics

The basic building block of the ASPAT diode is based on heterojunction of three

multilayer semiconductor structures which have two different band gaps The structure

comprises a thin layer of wider-gap semiconductor sandwiched between two

semiconductors with narrower-gap forming a tunnel barrier The basic principle of the

ASPAT device is based on the exploitation of quantum mechanics theory using

heterojunctions interface According to quantum mechanics theory moving particles

(electrons) with less energy than the barrier height have a probability of appearing on the

other side of the barrier by a tunneling through it This can be achieved in conditions

where the barrier must be very thin (~ 10 monolayer ) This is in contrast to classical

physics where a particle must have kinetic energy at least slightly greater than the

potential barrier height in order to overcome the barrier otherwise the probability of the

particle to appear on the side of the barrier is zero

Since the ASPAT diode operation is based on tunnelling through a barrier one

needs to know that the tunnelling mechanisms can be classified into two types[44]

intraband and interband The latter is described as tunnelling that occurs from

conduction band to valence band (electron) and valence band to conduction band (holes)

This normally happens in bipolar device ie p-n junction diode which has n-type and p-

type doped regions On the other hand intraband refers to tunnelling which occurs when

51

electron tunnel from the conduction band of a semiconductor to the conduction band of

its neighbouring semiconductor The same thing happens to the holes in the valence

band The device with this type of tunnelling is normally a unipolar device which is

either p-type or n-type doped The ASPAT diode can be considered as a device that is

based on intraband tunnelling mechanism Therefore the focus will be entirely based on

its principles

261 Principle of Quantum Tunneling

Generally all tunneling diodes obey the concept of quantum mechanical

tunneling Quantum mechanical tunneling is a phenomenon where a particle is able to go

through an energy barrier higher than the kinetic energy of the particle and if it is thin

enough compared to the de Broglie electron wavelength (λ) If the electron wave is

greater than the barrier the probability of the wave to occur at both side of the barrier is

higher

Figure 212 Classical view of whether an electron is can surmount a barrier or not Quantum

mechanical view allows an electron to tunnel through a barrier The probability (blue) is related to

the barrier thickness

For the case of classical physics (Figure 212(a)) the particles can be confined by

energy barriers of a semiconductor if their kinetic energy is less than the barrier energy

The particles thus require higher kinetic to escape to other states this phenomenon is

called thermal emission In quantum mechanics (Figure 212(b)) the particle is

described in two ways as a wave and as a particle If the particle moves like a wave it

will carry all the waversquos properties Therefore it will not brusquely end up at the

En

erg

y

(a) Classical view (b) Quantum mechanical view

En

erg

y

En

erg

y

52

boundary of the energy barrier Hence when the particles collide with the barrier

(incidence) there will be a probability of penetrating the barrier if the barrier is thin

enough and has a finite height For thicker barrier the probability of a wave that can be

found on the other side of the barrier is very small However the possibility of the

electron wave to appear on the other side of the barrier is increased by thinning the

barrier The potential barrier of semiconductor material technology can be determined by

using Homojunctions structures with different doping profile This will result in a

difference in band alignment and multilayer heterojunction structure (different

semiconductors have different band gap) which includes semiconductors insulators and

conductors (metals)

The easiest way to explain the phenomenon is by considering a potential barrier

Epot(x) with barrier height E0 energy bigger than the total energy E as shown in Figure

213 the potential energy occurs in a finite space between 0ltxlta and is 0 outside

Figure 213 Rectangular tunnelling with incident reflected and transmitted wave function[70]

The electron outside the region of the potential barrier (xlt0 and xgta) is free to

move The effective mass of the electron is different inside and outside of the barrier in

real tunneling devices when implemented using semiconductor heterostructures The

quantum mechanical equations predict the wave nature of matter which states that matter

unveils wavelike properties under some conditions and particle-like properties under

other conditions The wavelike properties as described by the Schrodinger Formula of

E0 Transmitted Ψ = 119862119890minus119894119896119909

Incident Ψ = 119860119890119894119896119909

Reflected Ψ = 119861119890minus119894119896119909 Ψ = 119863119890minus119894119896119909

0 a

E

x

V(x

)

53

quantum mechanics represent a particle penetrating through a potential barrier most

likely as an evanescent wave coupling of electromagnetic waves[44]

To start the calculation of the tunnelling probability the Schrodinger equation is given

by

119894ℏ

120597

120597119905Ψ(119903 119905) = ΨΗ(r t)

(217)

Where ℏ is Planckrsquos constant (662606957 times 10119890minus341198982119896119892119904

2120587frasl ) Ψ(119903 119905)is the wave

function at position r and time t Η is the Hamiltonian operator given by

Η = minus

ℏ2

2119898nabla2 + 119881(119903 119905)

(218)

Where 119881(119903 119905)is the potential energy which is dependent on space and time 119881(119903 119905) is

considered zero for a particle traveling in free space without any potentials The plane

wave with vector r and t is given by

Ψ(119903 119905) = 119890119894(119896119903minus120596119905) (219)

This equation satisfies Eq (217) above under the condition where the particle is

travelling in free space without potential k is the wave vector which is equivalent

to2120587120582frasl and the angular frequency 120596 is 2120587 multiplied by the frequency

In the case of tunneling through a potential barrier the method of separation of

variables is used to simplify the problem as in the equation below

Ψ(119903 119905) = 119877(119903)119879(119905) (220)

It is assumed that the problem above is divided into time-dependent and time-

independent parts 119877(119903) is the spatial component and 119879(119905) is the time-based component

of the wave function The time dependent problem as shown above in the Schroumldinger

Equation (1) can easily be solved by filling up all the finite parameters The solution of

54

the time-independent part gives the tunneling probability For the one-dimensional (1D)

time-independent Schroumldinger equation[44]

119864120595(119909) = minus

ℏ2

2119898

1198892

1198891199092120595(119909) + 119881(119909)120595(119909)

(221)

E is the total energy and 120595(119909)is the spatial component of the wave function along the x

axis The combination to the wave function is given by

120595(119903 119905) = 120595(119903)119890minus(

119894119864119905ℏ

)

(222)

The time-independent plane wave solution of 120595 = 119890119894119896119909 satisfies the equation (221) for

any constant potential V0 in space Plugging in the wave solution yields the condition

that

119896 = radic2119898lowast(119864 minus 1198810)

ℏ2

(223)

Referring to Figure 213 also the barrier with exact rectangular shape with height E0 and

width W the solution of the wave functions and tunneling probability can be extracted

by using the below equation [44]

119879119905 =

[119862119890minus119894119896119909]2

[119860119890119894119896119909]2= [1 +

11986402 sin ℎ2 (119896119882)

4119864(1198640 minus 119864)]minus1 asymp

16119864(1198640 minus 119864)

11986402 exp(minus2radic

2119898lowast(1198640 minus 119864)

ℎ2119882)

(224)

For more complex barrier shape Wentzel-Kramers-Brillouin has simplified the

Schrodinger equation for tunneling probability of carrier which becomes[44]

119879119905 =

[119862119890minus119894119896119909]2

[119860119890119894119896119909]2asymp 119890minus2int |119896(119886)|119889119886

1198860 asymp 119890

minus2int radic2119898lowast

ℎ2 [119880(119886)minus119864]1198891198861198860

(225)

55

This means that the incident electron has a finite probability T of tunneling

through the potential barrier and this leads to the concept of tunneling probability as well

as a tunnelling current Therefore this becomes the basis of tunneling phenomena and

thus all devices which are related to tunneling can be modelled and analysed based on

this basic example The tunneling phenomenon is a majority-carrier effect and the

tunneling time is set by the quantum transition probability per unit time (which is on the

order of picoseconds) rather than the transit time concept [44 71] This enables the

tunneling devices to work at a much higher switching speed They can also be used in

high-frequency applications such as microwave circuit and high-speed oscillators

262 ASPAT Structural Parameters of GaAsAlAs materials System

Figure 214 Epitaxial layer structure of device XMBE 304 (lateral structure) used in this study

Conduction band profile

56

The core materials that make up the ASPAT diode in this work is based on

heterostructure of group III-V compound semiconductors Such materials are chosen due

their mature excellent properties and their band gap which can be tailored to fit the

desired design as well as to improve the carrier mobility In this work the primary layers

that form an ASPAT diode are very thin pure Aluminium Arsenic (AlAs) of thickness

ten monolayers buried in between dissimilar thickness of pure Gallium Arsenic (GaAs)

layers as can be seen in the red circle in Figure 214 above These two GaAs layers are

known as spacer layer which normally have a ratio of 401 or 201 in thickness The

asymmetrical spacers layer and the thin barrier in such arrangement lead to an

asymmetric current-voltage characteristics as proposed firstly by Syme and Kelly[15]

To examine and investigate this GaAsAlAs ASPAT structure in term of electrical and

RF characteristics the device have been grown according to Table 22 below

Table 22 epitaxial layer of sample XMBE304 which is the main focus of this work

Material Doping (cmminus3) Thickness (Aring) Bandgap (eV)

Ohmic Layer GaAs (Si) 4times1018 3000 142

Emitter GaAs (Si) 1times1017 400 142

Spacer GaAs Undoped 50 142

Barrier AlAs Undoped 28 283

Spacer GaAs Undoped 2000 142

Collector GaAs (Si) 1times1017 400 142

Ohmic Layer GaAs (Si) 4times1018 4500 142

Substrate GaAs - 650 microm 142

The arrangement of the multi-layers that form a lattice matched GaAsAlAs ASPAT

diode can be transformed into band structure profile view for easy understanding The

conduction band profile at equilibrium is as sketched in Figure 214 As can be seen in

the Figure 214 the ASPAT diode is generally a heterojunction multilayer structure

tunnelling diode

57

Figure 215 The layer profile of ASPAT with 28nm barrier thickness [27]

The generic structure of ASPAT diode which is shown in Figure 215 above with a

schematic band structure comprises the following (starting from the top)

(1) A thick layer of heavily doped n++

about 4e+18cm-3

of GaAs with a thickness of

approximately 300nm

(2) An intermediate layer of lightly doped n-type about 1e+17cm-3

of GaAs with

thickness of approximately 40nm

(3) A spacer layer not intentionally doped (NID) GaAs with thickness of approximately

5nm

(4) An ultra-thin layer of NID AlAs with thickness of approximately 28nm

(5) A spacer layer of NID GaAs with thickness of approximately 200nm

(6) An intermediate layer of lightly doped n+ ~1e+17

GaAs with thickness of

approximately 40nm

(7) A thick layer of heavily doped n+ about 4e+18 of GaAs with a thickness of

approximately 750nm

Each layer has its own role For instance layer (1) and (7) are used as ohmic

contacts via connection to a AuGeNiAu metal stack This explains why they are

purposely heavily doped (gt 1018

cm-3

) for better low resistance ohmic contacts Two

intermediate layers (layers 2 and 6) are used to prevent the carrier in the contact layers

from diffusing into the undoped layers The two unequal length spacer layers with ratio

58

1198971 1198972 of about 401 are used as voltage arms to yield an asymmetric current-voltage

characteristic The asymmetry means that after a positive bias is applied from the long

spacer region an accumulation layer is formed and it is deeper than that formed by the

negative bias The thin layer positioned in the middle (Layer 4) is the tunneling barrier

The performance of a single barrier ASPAT diode can be optimised depending on

the applications by appropriate selection of the material system so that the band gap and

barrier height of such material can be modified Furthermore the mobility of electron

and doping concentration of the contacts region can be tuned The parameters to tune

during the growth for instant growth interrupt time and growth temperature will also

affect the performance of this diode The key layers that will affect the performance are

the barrier thickness and the two spacer layers enclosing it The study has shown that a

one monolayer change in thickness results in 300 change in in tunneling current for a

fixed voltage point[65]

The following discussions account for the effect of the main structure of the ASPAT

which is related to their performance

2621 Barrier Thickness and height

The probability of an electron tunnelling through a barrier depends exponentially on

the width and height of the barrier as well as the energy that is incident on the barrier[72

73] All these will affect the I-V characteristic of the ASPAT diode The tunnel current is

obtained by summing over all incidents electrons energies with tunnelling probabilities

through the barrier The tunnelling varies approximately as[74]

119879 prop 119890minus120581119889 (226)

Where d is the barrier thickness and κ is defined by the expression below

120581 =

radic2119898lowast(1198810 minus 119864)2

ℏfrasl

(227)

From textbook the tunnelling probability is given by

119879(119864) = 412058121198702

[(1205812 + 1198702)2119904119894119899ℎ2119870119897 + 412058121198702]frasl (228)

Where K is expressed as

59

119870 = radic2119898lowast119864

ℏfrasl (229)

Where ℏ represents the reduced Planckrsquos constant (h2π) E and m are the electron

energy and effective mass respectively Thus by inputting appropriate value into these

equations one finds that reducing the barrier thickness by one monolayer increases the

tunnelling probability by a factor of nearly three for every electron that tunnels through

the barrier As a result the current will also increase Further it indicates that the

tunnelling strongly depends upon barrier thickness and height By contrast the current

does not strongly depend on temperature

2622 Spacer Thickness

The reason for having two dissimilar undoped spacer lengths is mainly to avoid

diffusion of dopant to the barrier and subsequent layers during growth but in the case of

the ASPAT diode the spacer can also act as a voltage arm Varying the thick spacer

layer (1198971) results in the reverse current decreasing as the layer thickness increases and

varying the thin spacer layer (1198972) will affect the forward current which increases as the

thickness reduces To obtain appropriate asymmetrical I-V characteristics one needs to

maintain an adequate ratio between these two spacer thicknesses While a too thin

1198971 results in high leakage current at reverse bias a too thick 1198972 results in low forward

current One also needs to keep it thick enough to prevent carrier diffusion

These two spacers must be kept undoped or very low doped to allow the electron

moving in the electron mean free path region as it is clear from ionised donors Under

large forward bias an accumulation layer is formed between the spacer and barrier

segment and it is more noticeable compared to the accumulation layer that is formed if a

negative bias is applied

60

Figure 216 Conduction band diagram showing band bending and 2DEG formation at the L1

spacer

Consequently a triangular well is formed which creates an emitter 2D electron gas

(2DEG) population The electrons in this 2DEG occupy the quasi-bound states which

mean high excitation energy thus allowing the electron to tunnel through the barrier as

depicted in Figure 216

In term of RF performance it is important to highlight that a thicker spacer layer

will affect the depletion region which gets wider and thus will reduce the junction

capacitance of the device as per the following expression

119862 = 휀0휀119903

119860

119889

(230)

Where A is the area of the device d is the thickness of the main device structure which

consists of spacers barrier and well layers 휀0 is the permittivity of free space and the

relative permittivity of the spacer material is denoted by εr However the intrinsic delay

time will also increase and hence degrade the device high frequency performance

Therefore optimisation through spacer thickness requires trade-off between reducing

leakage current at reverse bias and degrading device junction capacitance

GaAs

GaAs

AlAs

2DEG

Γ

X

61

263 ASPAT Electrical Parameters

The classical approach in determining the current flow through an ASPAT diode is

by solving the Schrodinger and Poison equations Prior work had been done by Syme et

al in 1991 Due to the fact that AlAs barrier is very thin tunnelling is assumed to occur

at the Gamma valley ie AlAs bandgap= 283eV (rather than X valley)[18 75] and only

from accumulation layer (2DEG)[59] Here the DC characteristic of the ASPAT diode

can be calculated by solving Schroumldinger equation with the position vector represented

by z (in this case) Thus the equation is expressed as

minus

ћ2

2nabla

1

119898lowast(119911)nabla120569 + |119890|120593(119911)120569 = 119864120569(119911)

(231)

Supposing the current is uniform across x and y planes then this can be simplified to one

dimension Therefore the 1 D Schroumldinger equations becomes

minusћ2

2119898lowast

1198892

1198891199112120595 + |119890|(120595 minus ∆120595) = 119864119911120595

(232)

Where

120595 =

120569

exp (119894119896119911119911)

(233)

where ∆120595 is the correction term which reduces the effective barrier height The

Schroumldinger equation is solved using different values for Ez thus the quantum

mechanical current density in the z-direction is now expressed as[15]

119895119911 =

minus|119890|ћ

2119898lowast(120595lowast

119889120595

119889119911minus 120595

119889120595lowast

119889119911 )

(234)

It is different for a heavily doped contact which can describe as below the envelope

functions in the left and right contacts respectively can be described by plane wave

120595119897 = exp(119894119896119897119911) + 119877 exp(minus119894119896119897119911) (235)

120595119903 = 119879119890119909119901[119894119896119903(119911 minus 120580119873)] (236)

62

In this case the left contact covers the region zlt0 while the right contact covers the

regions zgt 120580119873 Equations (35 and 36) are then inserted into Eq (34) to form the

following expression

119895119911 =

|119890|ћ1198961

119898lowast(1 minus |119877|2) =

|119890|ћ1198961

119898lowast|119879|2

(237)

Where R (Ez) and T (Ez) are the complex reflection and transmission coefficients

respectively and they are solved by using the transfer matrix method This method has

been described in reference [56] The next step is to integrate the current in the z-

direction 119895119911 for all possible Ez values Thus the expression for the current density

becomes

119895119911 =|119890|119898lowast119896119861119879

2120587ћ2int(1 minus |119877|2)

infin

0

119897119899 |1 + exp (

119864119891 minus 119864119911

119896119861119879)

1 + exp (119864119891 minus 119881|119890| minus 119864119911

119896119861119879

|

(238)

The equations above are used to calculate the current density approximation from

the ASPAT main structure (two spacer layers and one barrier) only and based on

intraband tunnelling from the conduction band profile For real fabricated structure the

calculation must take into account both intrinsic and extrinsic elements of the diode

While the latter is mostly related to the pad and probe that is used to extract the I-V

characteristic the first element mostly comes from the epitaxial layer of the ASPAT

diode itself The ASPAT I-V characteristic is shown in Figure 217 which clearly

indicates nonlinear characteristics and thus can be used for detection applications

63

Figure 217 I-V characteristics of a fabricated ASPAT diode

2631 Intrinsic Elements of the ASPAT diode

In order to extract the intrinsic electrical characteristic of the ASPAT diode a

generic structure as shown in Figure 218 is essential Two main sources of contribution

to the electrical characteristics are the interfaces of each layer and properties of the

materials themselves

Figure 218 Intrinsic Elements of the Asymmetric Spacer Tunnel Layer (ASPAT) Diode

64

The electrical current flowing from the top contact to the bottom contact will go

through each epitaxial layer producing a close loop Each junction interface limits the

current flow and sum up the total resistance resulting in what is known as the diode

series resistance (Rs) In the ASPAT main structure there is a junction capacitance (Cj)

due to the undoped regions surrounded by heavily doped contacts thus acting as a

parallel plate capacitor The fully depleted capacitance (Cj) of the diode can be

expressed as in Eq (230)

2632 Series resistance of the ASPAT diode

The total series resistance (Rs) of an ASPAT diode can be calculated based on the

finished fabricated diode structure In general Rs depends on three contributors namely

the non-uniformities in the contact metallization the un-depleted epitaxial layer (total

thickness) on both side of the heterostructures and the resistance caused by the

spreading current from the Mesa into the much wider second contact layer ie doped

substrate or 2nd

ohmic layer[76] In fact the contribution toward building up the total Rs

solely depends on how the structure is designed In this work two types of structures

were deployed namely lateral structures and vertical structures These will be described

in the next subsection

For both types of structures the series resistance (Rs) of the ASPAT diode consists

of aspecific Ohmic contact resistance (ρcA) contact epitaxial layer resistance (Repi-

Layers) and spreading resistance (Rspr)[33 77] where Rspr is influenced by the type of

structure The specific contact resistance is obtained from Transmission Line

Measurements (TLM) of the sample The theory of the TLM will be discussed in detail

in the next section The expression for the specific contact resistance is

120588119888 = 119877119888119871119879119908sinh119889

119871119879frasl

cosh 119889119871119879

frasl

(239)

Here Rc is the contact resistance LT is the transfer length (effective length) w is the

contact pad width and d is the length of the contact pad

65

Repi-Layer is the sum of all doped layers that sandwich the main ASPAT device For

each doped layer the resistance is given by

119877119890119901119894minus119897119886119910119890119903 = 120588

119871

119860

(240)

Where ρ is the resistivity which is given by 120588 = 1

120583119899119902119873119863 L is the epitaxial layer thickness

in cm A is the device area (emitter size) micron denotes the mobility of the electron q is the

electron charge and ND is the donor concentration The spreading resistance depends on

the structure design of the device This will be elaborated in detail in the following

section

26321 Vertical structure (doped Substrate)

The XMBE307 structures were grown on n+ GaAs substrate to provide the

simplest fabrication process The cross section of the finished single diode can be seen in

Figure 219 below

Figure 219 Cross-sectional view of ASPAT vertical structure the dimensions of d b and h are not

drawn to scale

66

At low frequency and in a mesa that is etched into a doped substrate material the

spreading resistance can be approximated by[76]

119877119904119901119903 =120588119904

2119889

(241)

Where ρs is the substrate resistivity and d is the ASPAT diode mesa length

However the spreading resistance is increased at high enough frequencies as the skin

depth (δ) in the substrate is much lower than the effective mesa length (d) of the diode

A new spreading resistance is then calculated also based on the assumption that the skin

depth is much lower than the chip thickness (h) but much larger than the mesa length

Thus the spreading resistance at high frequency is given by

119877119904119901119903(119891) =

120588119904

120587120575[05 ln (

119887

119889) +

119887]

(242)

Where the skin depth (δ) is taken from standard planar formula and can be expressed as

120575 = [

2120588

(120583120596)]

12frasl

(243)

Where micro is the permeability and ω is the angular frequency During DC measurement

this type of structure requires having good suction on the stage for a good contact

However for small die (15mm times15mm) the suction sometimes is not strong enough to

provide a very good adhesion to the sample Therefore another type of ASPAT diode is

deployed which is based on the lateral structure by utilising semi-insulating substrate

and both contacts are connected to probes

67

26322 Lateral structure (Semi-insulating Substrate)

Figure 220 Cross-section view of the complete process ASPAT diode (XMBE304) The dimensions

are not drawn to scale

In order to obtain accurate measurement results so as to avoid contact errors to the

substrate between stage and Device Under Test (DUT) a lateral structure as shown in

Figure 220 above is deployed This type of design offers many advantages ie it

provides a path for on-wafer RF measurement However the proper design has to take

into account the increase in RS due to improper attention to the spreading resistance

This spreading resistance is different from the vertical structure that was discussed

above In this case it is mainly caused by a gap at the bottom contact The gap in the

horizontal direction between epitaxial layer and metal at bottom contact is denoted as D

gap in in Figure 220 Therefore Rspr for the lateral design can be expressed as [77]

119877119904119901119903 =

1

120587120590119889119866119886119860119904ln (

119886

119886119898119890119904119886)

(244)

Where σ is the conductivity between two coaxial half-cylindrical electrodes with inner

(amesa) and outer (a) rectangular length or bottom ohmic layer which is given by (σ =1

ρ) a d and amesa are the length and thickness indicated in Figure 220 above Noticeably

68

the D gap will have direct effect on the outer length (a) of the device which is also

proportional to Rspr For high-frequency operation where the skin depth is less than

d(GaAs) σ becomes

120590(120596) =

120590(0)

[1 + (120596120591119903119890119897)2]

(245)

Where τrel =microme and micro m as well as e are the mobility effective mass and electron

charge respectively It is recommended that for high-frequency applications the device

series resistance must be as low as possible

Hence for both type of structure the ASPAT series resistance is calculated based on

all the above-stated resistances and these are set by

119877119904 =120588119862

119860+ 119877119890119901119894minus119897119886119910119890119903119904 + 119877119904119901119903

(246)

The total RS can be decreased by increasing the emitter area of the diode However a

large device will not able to reach millimetre and submillimeter wave region (THz) as

the capacitance will also increase (Eq 230) Therefore both parameters will have a

trade-off between them to be able to work at ultra-high frequencies

27 Characterization of Ohmic Contacts

The semiconductor Ohmic contact can be characterised using techniques that will

be described in the following section First is the Cox-Strack technique which is

specially designed to characterise bulk type semiconductor (thick) contact resistance on

two opposite sides The detailed description of this technique can be found in [78]

Second is a technique called Four Point Probe This technique was developed in 1954 by

Valdes etel [79] to characterise semiconductor resistivity It can also be used to

characterise the contact resistance for planar type devices As this research does not

cover this method the details of the measurement can be referred to [80] Finally the

most common method which is also extensively used in this research is the standard

Transmission Line Measurement (TLM) The details of this method will be explained in

the next section There are simplified versions of the TLM method which require just

one lithography step but are nevertheless very powerful in characterising and optimising

the contact resistance known as Circular Transmission Line Measurement (CTLM)[81

69

82] However in this research this is not to be covered as the standard TLM is already

adequate for planar type devices

271 Transmission Line Measurement (TLM)

The formation of metal and semiconductor interfaces will create a contact that

becomes very important for the characterisations of any fabricated device Additionally

it enables the quality of certain process flow to be determined This interface must be

evaluated by a technique known as the Transmission Line Measurement (TLM) TLM

which was first introduced by Murrmann and Widmand [83] in 1969 underwent some

refinements by Berger [84] in 1971 The theory of the TLM can be described by

constructing a TLM structure which comprises a set of metals contact pads placed in

series on a highly doped semiconductor layer as depicted in Figure 221 The structure is

designed like a series of the islands to permit current flow in parallel to the contact pads

[80] which is a direction defined by etched patterns Each contact metal pad behaves

like a MESA which has a thickness (t) and width (W) The distance between each metal

contact pads is defined by d1 the gap between two neighbouring contact pads which are

beneath each contact pad is defined as effective length LT This will allow current flow

in and out of the subsequent neighbouring metal pad The resistance elements that will

be extracted are RA and RB which sit under the contact and in between two metal pads

These two elements represent the sheet resistance under the metal contact pad area and a

sheet resistance of the material between two metal pads

Figure 221 A simple TLM structure with effective length and sheet resistance underneath

t L

T

LT

RA R

A R

B

dn

Probe Probe

Metal Pad

GaAs

MESA

70

The basic relationship of resistance R with respect to the size of the metal contact

or in the standard transmission line can be expressed as [44]

119877 = 120588

119871

119860= 120588

119871

119905 times 119882=

120588

119905times

119871

119882= 119877119904ℎ

119871

119882

(247)

Where ρ is the materialrsquos resistivity L is the length t is the thickness W is the width

Rsh is the sheet resistance and A is the cross-sectional area of the transmission line The

unit for ρ and Rsh are Ωm and Ωm2 respectively

The total resistance RT of this structure can be taken from the sum of the two

neighbouring padrsquos resistance RA and RB In order to relate with Eq (247) above this

RT will be substituted into Eq(247) to become

119877119879 = 2119877119860 + 119877119861 = 2119877119904ℎ119860

119871119879

119882+ 119877119904ℎ119861

119889119899

119882

(248)

As suggested in [85] RshA and RshB are assumed to be identical Therefore Eq(248)

can be reorganised into specific contact resistance RC and semiconductor sheet

resistance Rsh above as 119877119862 = 119877119904ℎ119860119871119879

119882 the new equation is then

119877119879 = 2119877119862 + 119877119904ℎ119861

119889119899

119882

(249)

The common practice throughout this research is to design a TLM structure that

has a ladder structure consisting of 10 metal pads with each one measuring to a size of

100microm width and 50microm length and the space between the first and second metal pad

5microm The gap is increased after the second metal pad by a further 5microm until ten metal

pads are completed produce a separation between the ninth and tenth pad of 45microm The

TLM ladder structure as depicted in Figure 222 is supplied by a constant current of

1mA at the very left and right metal contacts by two probes This allows the extraction

of RC (Ωmm) and Rsh (Ω) instantly from such structure The potential difference

between the two adjacent metal pads is measured by another two probes and the reading

of a voltmeter is recorded The total resistance is obtained by using Ohm law where

voltage is divided by current (VI) Another voltage reading is taken for the next two

neighbouring metal pads until the largest gap is reached The readings of (conversion

71

VI) which result in the corresponding resistance RT are then plotted against spacing and

the result can be seen in the graph in Figure 223

Figure 222 Top view of TLM ladder structure use in this work

Additionally in the measurement the voltmeter used in work has a very high resistance

Otherwise there will be leakage of current occurring through the probes and cables

Therefore the parasitic resistance of the cable or connector and the probe contact can be

ignored The key parameters that can be extracted from the graph will be discussed in

the next paragraph

Figure 223 Typical plot of resistance versus TLM spacing

y = 02592x + 02375

0

2

4

6

8

10

12

14

0 10 20 30 40 50Res

ista

nce

Rn (

Oh

ms)

TLM Spacing dn (um)

LT

d1 d

2 d

3 d

n

I (1mA)

V

V

LT

ME

SA

Su

bst

rate

72

The straight-line graph plotted in Figure 223 can be referred to Eq (249) and

this must be done by assuming the sheet resistance Rsh of the material is constant If the

straight line of the graph is extended up to initial gap (d=0) the intercept on the y-axis

provides the 2RC value To extract the 2LT further extrapolation is made until the

interception at RT = 0 is reached Therefore an important parameter which is the specific

contact resistance ρC can be found from this expression[85 86]

120588119888 = 119877119862119871119879119882sinh

119897119871119879

cosh119897

119871119879

(250)

Where l represents the total conductive semiconductor thickness The final part that can

be extracted from the graph is the slope of the line This is obtained by dividing the sheet

resistance with the width of the metal pad ie represented by expression (119877119904ℎ

119882frasl )

28 Basic Characterization Techniques and procedures

281 Measuring tools and apparatuses

The success of every experiment is determined by the backend results that are

obtained from the measurements It is important to choose an appropriate instrument

which will provide the required data for a valid and detail analysis Thus this section

will give a brief explanation of the measuring instruments and methods that were

exploited in this research The measurement apparatus systems that are available and

have been utilised in completing this thesis are ldquoset of DCrdquo and ldquoSet of RFrdquo

measurements The DC set measurement consists of room temperature and variable

temperature system This system is built to perform process monitoring during device

fabrication and it comprises of five main components

The fundamental component in the ldquoDCrsquos set toolrdquo is the Agilent B1500A

Parameter Analyser [87] used to provide fixed currentbias during testing The other

component is a Karl Suss PM5 Cascade Prober [88] which is used to receive fixed

currentvoltage and to supply its to the semiconductor contact via probe tip The PM5

Prober has at least four probe arms and each of them is fitted with ldquoneedlerdquo called probe

73

tip The size of the tip that is normally used here is 2microm and in some cases the tip size

of 1microm is also utilised The currentbias supplied to the sample must go through the two

of Source Measurement Units (SMU) namely SMU1 and SMU2 to ensure no mismatch

occurs between parameter analyser and diode All the testing are controlled using a

software called Integrated Circuit Characterization and Analysis Program (IC-CAP

2009) brought from Keysight Technologies[89] The software is installed on a standard

Personal Computer (PC) and the PC is connected to a General Purpose Bus Interface

(GPIB) to link with the B1500A Parameter Analyser This system can be organised

based on the purpose of measurements ie IV characteristics TLM Schottky Diode

and transistor as it is very flexible to change the configuration For examples

Transmission Line Measurement (TLM) configuration requires the addition of a digital

multi-meter and four-point probe tip while diode measurement just needs two point

probe tip without a digital multi-meter Figure 224 illustrates the measurement system

for a set of DC to test the TLM structure

Figure 224 The set of DC measurement apparatus arrangement to measure the TLM

On the other hand the set of RF measurement consists of five block elements

configuration in the system namely the Vector Network Analyser (VNA) DC

sourcemonitor Cascade Microtech Prober SMU and Control PC via GPIB The RF set

Agilent B1500A

Parameter analyser

DUT

(TLM Diode

Capacitance)

Karl Suss

PM5 Cascade

Prober

PC

(MS windows 2000)

Digital Multi-meter

ICCAP 2014 Provide current

source

Measure and read the

voltage Stage and Probe

Signal

Current Source

SMU1 SMU2

General purpose bus interface

74

performs RF characterisation after the device fabrication is completed This system can

also perform DC characterisation as its basic instrument has this function too The VNA

machine used in this research is the Anritsu 37369A [90] which can perform the

Scattering Parameter (S-Parameter) measurement with a frequency range of 40MHz to

40GHz The DC sourceMonitor utilised in this experiment is the HP 4142B Modular

DC SourceMonitor [91] Both of these sub-systems are controlled by a standard PC

which exploits GPIB port to link them During operation the HP 4142B is connected to

the VNA by an internal bridge network and two SMUs to the Cascade Microtech Prober

which is then connected to the bond pads of the device It is identical to the set of DC

measurement where the SMU setting and data assembly are accomplished by IC-CAP

software package therefore the data that was obtained before and after completing the

fabrication can be compared This will enhance the validation of the data

However the stage and probes in both sets of measurement are different The

Cascade Microtech Prober has only two probe arms with each arm fitted with a 3-

fingers probe tip in the arrangement of Ground-Signal-Ground (GSG) as shown in

Figure 225 Each finger (pitch) is separated by 100microm thus to fit in with the pitch and

to reduce the mismatch in resistance the bond pad design must follow this separation

between each contact The GSG is configured by connecting the outer pads (Collector)

to the Ground probe tip while the inner pads (Emitter) are attached to the Signal probe

tip where the RF signal is sent and received through it Figure 226 shows the actual set-

up for RF measurement used in this research

75

Figure 225 The set of RF measurement for on-wafer DC and RF one-port measurement

Figure 226 Actual VNA system that was used for RF characterization

282 Measurement steps using a VNA

The RF measurement steps can be summarised in the following

- Step 1 Select or find the suitable VNA depending on applications

PC

(MS windows 2000)

ICCAP 2014

VNA

Anristru 32379A

40MHz to

40GHz

HP4142B

Modular DC

source

DC Bias Source

Cascade

Microtech

Probe Station

One 50microm pitch G-

S-G probe tips

Ground

SMU1

Cathode

Gen

eral

purp

ose

bus

inte

rfac

e

Lo

w P

ow

er H

igh P

ow

er

Cathode

Anode

SMU2

Ground

Ground

Signal

76

There are few factors that need to be well-thought-out before starting to use a VNA

especially for S-parameter measurements The factors are the availability of the VNA in

term of operational frequency and measurement port types air-filled metallic waveguide

or on-wafer Not all VNA can have an operational frequency for banded measurement

ie W-band Ku-band etc All these may require external signal sources to extend the

operational frequency

- Step 2 Properly setting up the VNA

The VNA can be set up depending on its application and the goal of measurement for an

instant number of point requires a desired frequency span IF bandwidth and the supplied

power level This very important to ensure the desired measurements are correct and

appropriate

- Step 3 Appropriate calibration system

In order to have a valid calibration appropriate calibration method has to be chosen

depending on the applications This will determine the accuracy and standards of

calibration As for on-wafer calibrations the de-embedding is normally used while for

the off-wafer the SOLR technique is more suitable to employ

- Step 4 Validation or verification of the calibration results

It is vital to validate the calibration results to ensure that the system has been properly

calibrated

- Step 5 Proper measurement

Proper alignment positioning and touching from probes tip to DUT is necessary to

guarantee a good repeatability and reproducibility of measurement results Normally

when positioning the probe an alignment marker is used as an aid By doing this similar

travel distance for the probes can be achieved The measuring plane will also be equally

well-defined

283 Measurement Practice and Flowchart

Essentially the device characterisation is performed in two stages ie during

fabrication as a process monitoring and after completion for data collection and

analysis Figure 227 shows the block diagram of a flow chart for testing a 15mm times

15mm wafer processing performance In the wafer processing after reaching the top and

77

bottom contactrsquos step the sample can be examined by measuring the TLM structure

according to the TLM procedure The measurement is conducted by exploiting a set of

DC measurement apparatus as mentioned above analogous to the TLM ladder structure

on the wafer surface Based on the TLM results the presence of any process issue during

the fabrication can be identified by examining the parameter such as contact resistance

and sheet resistance Thus a decision can be made either to proceed or to terminate the

fabrication should any issue is found early on As a result no materials will be wasted

further When a dielectric layer is involved Capacitance Dielectric measurement can be

tested This practise can be used to obtain the quality of the dielectric layer To connect

between a diode effective area and probe a bridge is requires It can be attached to the

diode emitter (to determine the diode size) to a bond pad for probe tip to touch This

bridge can either be left hanging in air or sticking to the isolated dielectric layer For the

GaAsAlAs material system the latter technique is preferable since it avoids issues with

the air-bridge which will be discussed in detail in Chapter 3 The opening area (via) for

metal connection can be checked by measuring the resistance on a special design pad

after a plasma dry etching step

Figure 227 Block diagram of the ASPAT measurement step

The next stage of characterising the device is the on-wafer diode measurement

which takes place after completion of all processing (including bond padsco-planar

pads) The work is carried out using the above set for RF measurement and employed

Ohmic Contact

Opening Via

TLM

Qualified

Device

Bad device

Dielectric

capacitance

Bonding pad DC and RF

Good

Good Good

Fail

Fail

Fail

78

purposely to access the DC and RF performance of each diode where the current-voltage

(I-V) characteristic and S-Parameter results are obtained In fact DC measurements are

first performed using a set of DC measurement and a rough IV characteristic can be

obtained to ensure the diode is working properly Usually the yield of any fabricated

device on 15mm times 15mm wafer in this research is between 70 to 90 In this research

the outcomes that will be discussed in the subsequent chapters are in term of the average

of measured values Thus it is very important to have a meaningful data to compare with

physical modelling and simulation in the future The diode IV characteristics are studied

by applying different DC bias at the emitter to collector terminals to extract its keys

parameters ie turn on voltage (supposedly zero bias) non-linear characteristic Rj RS

Cj etc

The device that has been measured by DC and having produced a valid result

will be marked for the next investigation ie the S-Parameter measurement This

measurement is executed using a similar system but with different probe types The

three fingers probe type is used and the device frequency response is measured via the

one-port network from the VNA The important parameters extracted from this

measurement are usually S11 (depending on how many ports are measured) Although

working devices are selected to measure accordingly there is a need to ensure the VNA

RF cable and probe tips are calibrated so that only valid data without errors will be

obtained A calibration technique called SHORT-OPEN-LOAD (SOL) is performed

prior to each daily measurement by exploiting a calibration sample with WinCal

software (Cascade Microtech)[92]

To avoid confusion it is worth mentioning that this technique used to calibrate

the device structure is different from what is used in SOL calibration to the equipment

Furthermore the de-embedding calibration is made on-wafer with the same substrate of

the actual device whereas the SOL is performed on a special calibration substrate

Normally the de-embedding results are not constantly automated with the VNA

equipment However the measurement is done separately starting with the special DUT

substrate then followed by the OPEN and SHORT de-embedding structures

79

3 RESULTS OF FABRICATION PROCESS OF ASPAT DIODES

31 Introduction

This chapter presents in detail the general fabrication techniques for a generic and

development work of micron scale ASPAT diodes The discussions will focus on the

semiconductor growth technique used in this work ie MBE and the fabrication process

steps which include sample cleaning photolithography etching and contact

metallization These techniques are ample to build and deliver commercially marketable

fabricated structure Hence all photolithography techniques used to complete this

project are based on conventional i-line optical lithography which is adaptable to

industry and commercial purposes

The fabrication process in this work can be fragmented into two major works

firstly the fabrication towards reproducibility repeatability and manufacturability in

term of device structure process flow and DC amp RF characteristics This work will

involve relatively larger emitter area which varies from 15times15microm2 to 100times100microm

2 The

larger area provides for ease and fast fabrication as well as DCamp RF measurements

Once repeatability and reproducibility of the process flow and performance is confirmed

the second part of this work which is concerned with applications in millimetre and sub-

millimetre-wave then took place In this work small emitter designs varying from

2times2microm2 up to 10times10microm

2 were considered with appropriate measurement pads The

successful fabrication of smaller diode geometries in the second part of the programme

naturally leads to a further investigations to understand its epitaxial layer structure and

extracting intrinsic components ie junction resistance junction capacitance and series

resistance which will determine the device performance in high-frequency applications

The I-V characteristic is obtained from DC measurement which is usually performed at

room temperature Its results are then compiled and compared with advanced simulation

It is worth mentioning that all samples that are investigated in this project are grown by

means of MBE and the activities related to the epitaxial layer growth using MBE in the

University of Manchester were done by the Materials Growth Team (Prof Missous) and

the authors has no responsibility for this particular task

80

32 Epitaxial Layer Growth Techniques

Before discussing the principle of the common lithography technique it is

important to discuss the wafer growth technique as it comes first before the fabrication

The growth technique that is extensively used in this study as well contributing a lot in

the electronic semiconductor industry is Molecular Beam Epitaxy (MBE) The following

section will discuss the basic operating principle of solid source MBE

321 Molecular Beam Epitaxy (MBE)

The MBE technique was developed in the early 1970s [27] and is purposely used for

growing high purity epitaxial layers of compound semiconductors Such sophisticated

growth technique provides significant functionality ie precise control of the thickness

(to one atomic layer) and contributes to the growth of various types of complex

semiconductor multilayers high quality and advanced materials This level of control is

vital for an assortment of heterostructures devices that are being utilized as part of the

development of the advanced electronics devices especially for the ASPAT diode which

require 01ML control over the AlAs barrier to attain acceptable variability in device

characteristics Additionally the accurate doping profile and excellent junction

abruptness also can be achieved by using this technique

Practically the MBE system used in this work is a solid source MBE which

utilises beams produced by heating up various sources The sources can be Si Al Ga

As and other group III-V compound semiconductors When the crucibles which contain

the sources are heated atoms or molecules of the various elements are evaporated and

travel in straight lines paths like beams directed toward a target (heated and rotating

substrate surface) The condition of the vacuum during evaporation is ultra-high vacuum

(UHV) ~10-11

torr in order to have high quality crystals The substrate is heated and

rotated to provide good growth uniformity over large areas ( up to 4times4rdquo wafers) [93]

The growth rate in typical MBE growth is ~1ML second and can be controlled

by the source temperature in the crucible The abruptness at the heterojunctions interface

and switching of the growth compositions can be obtained by precise control of shutters

that are placed in front of the crucibles Therefore an abrupt junction at GaAs and AlAs

81

interface can be formed to realize the barrier in the ASPAT diode In order to monitor

the quality of the growing crystal and measure the layer by layer growth mode

Reflection High Energy Electron diffraction (RHEED) technique is utilised This

technique works based on the diffraction of electrons from the crystal surface [93]

Additionally given that the ASPAT current density is very sensitive to the barrier

thickness a study has been made using different growth techniques namely MBE and

Metal-Organic Chemical Vapour Deposition (MOCVD)[59] From this investigation it

was concluded that the percentage local variability of current density produced by MBE

grown diodes is better than those grown by MOCVD Thus in this study to get benefit

from its performance all wafers are MBE grown

33 Basic Principles of Common Fabrication techniques

This section covers the generic fabrication process which underpins

reproducibility repeatability and process optimisations for high-frequency applications

331 Sample cleaning

Essentially semiconductor processing requires a ldquoclean environmentrdquo to produce

devices The clean environment is classified according to how many ldquounwantedrdquo

particles are present in a cubic meter There are four categories of clean room available

in the industry Class 10 Class 100 Class 1000 and Class 10000 [94]

The fabrication of all ASPAT diodes in this project was performed in a clean

room environment of Class 1000 equipped with laminar air flow and filter system to

give Class 100 or better during processing Although the sample was processed in a

highly controlled particle environment there is still a high chance for a sample to get

contaminated when handled by a human Besides this the source of particle which

contributes to the contamination can be from the apparatuses and processing equipment

used in the laboratory themselves Thus the process of cleaning the sample wafer

surface before the start of each step is vital

Generally in a clean room the standard solutions that are used to clean a sample are

N-Methyl Pyrrolidone (NMP) Acetone Propan-3-ol (Iso-Propane-ol) (IPA) and

82

deionized (DI) water The sample which is cut up into 15times15mm2 size tiles is cleaned

based on the following procedures

1 Hot NMP- The sample is dipped into the solution at 80˚C for 10 minutes This

solution acts as an organic type of nature pollutants removal

2 DI water- acts as NMP remover The sample is then washed by flowing DI water

throughout the samplersquos surface

3 Acetone- to ensure any remaining NMP is completely removed from the sample

The sample is dipped into Acetone for 5 minutes in a low power ultrasonic bath

at ambient temperature

4 IPA- is used to remove the Acetone from the samplersquos surface The sample is

then dipped into the IPA for 5 minutes in a low power ultrasonic bath at room

temperature The use of low power for the ultrasonic bath is to avoid the sample

cracking or breaking

5 Once done the sample is then blow-dried using nitrogen (N2) gun to remove any

moisture coming from the IPA Fortunately it is easy to remove the IPA

completely from the sample surface given its higher rate of evaporation

Once all these steps are accomplished a visual inspection using a high magnification

optical microscope is conducted to ensure the sample surface is clean The cleanliness of

the sample is determined by the (lack) of particles or other spots (liquid mark) that can

sometime be observed during the inspection Obviously the lower the number is the

better the sample is as it is impossible to totally remove dirt especially marks

Sometimes it is hard to remove the particles in one go There is always a need to repeat

each cleaning step for a few times However this will not affect the sample in term of

electrical performance as the cleaning solutions used are non-destructive

332 Photolithography

Lithography or sometimes called pattern transfer is the most important step in

realizing microelectronic devices The designed geometry and dimensions on a quartz

glass plate called a ldquophoto-maskrdquo must be done prior to the fabrication process While

the photo mask can be designed by using various software tools in this work the design

83

is done via the Advance Design System (ADS) by Keysight The details of the design

which includes three different mask designs will be covered in Section 34 The mask

consists of the desired patterns (master) that can be printed onto solid materialrsquos surface

by means of an electrochemically sensitive polymer (photoresist) using

photolithography In fact this type of optical lithography technique can be performed

with and without a mask due to its simplicity It is also easy and cheaper compared to

other techniques such as x-ray lithography or Electron-beam lithography (EBL)

Although the latter is expensive it is still worth to have since it provides a higher

resolution which is preferable when developing sub-micrometre technology

processing[95] The special feature about EBL is that it does not need a mask to pattern

samples but can be produced by the movement of an electron beam point source and

hence writing the patterns directly on the surface

The ultra-violet (UV) based light sources are more popular among researchers

and development workers because they are cheaper and have modest resolution Usually

photolithography operates at wavelengths (λ) from 193 nm to 436nm The source that

provides the UV light is a Mercury (Hg) arc lamp which uses narrowband filters to select

single emission lines First is the i-line at λ= 365nm then the h-line which is of lower

resolution and has λ of 405nm and thirdly the g-line with a λ= 436nm[96] The

conventional optical lithography used in this research uses the ldquoi-linerdquo at a wavelength of

365nm For shorter wavelengths than these excimer laser or krypton fluoride laser with

a λ= 248nm and argon-Fluoride with a λ = 193nm are also used in the industry The

higher power levels enable higher productivity (throughput) while narrower spectral

widths reduce chromatic aberration provide better resolution and larger depth-of-focus

In this research all the fabrication process are done by utilizing a conventional optical

lithography (i-line) using a Karl Suss MA4 mask aligner Before starting any UV

exposure it is important to check the UV light intensity as it will affect the resolution

and thus desired device dimensions Therefore every corner that is exposed to the UV

light is calibrated to be at 09mWatt power exposure

The complete set of photolithography components consists of photoresist

(photosensitive polymer) photo-mask (chromium) which is used to block the UV to

form a pattern mask-aligner and developer (chemical solutions) Standard fabrication

84

process usually practiced at the University of Manchester starts with sample cleaning as

mentioned earlier Then the sample needs to go through heat treatment to remove all the

moisture with a temperature set to be 150˚C and bake for 5-10 minutes After having

cooled down the samplersquos surface is coated with a thin layer of photoresist via a

technique called spin-coating using a Laurell CZ-650 series spinner The spinning speed

is set depending on the type of resist ie 3000rpm for negative photoresist and 5000rpm

for positive photoresist The rotating speed of the spinner does not have much effect on

the coated photoresist thickness but will have consequence on the uniformity distribution

over the sample surface The coated thickness photoresist however depends on the

concentration of the specific photoresist Once spin-coated is done another short heat

treatment (1 minute) is required to ensure the resist is hard enough to contact a

photomask and to remove any excess solvent The temperature is set on the hot plate to

about 115˚C and 110˚C for positive and negative photoresists respectively

The important segments contained in the photoresist are a polymer (base resin)

a sensitizer and a casting solvent [97] The polymer will react by changing its structure

when exposed to the radiation While sensitizers will govern the reaction of the

photochemical in the polymeric phase the casting solvent will permit the spin-coated

application on the wafer surface The photoresist consists of positive and negative

photoresist both of which were used in this research Their basic difference is with

respect to the area that is exposed by UV light ie whether it will remain on the

semiconductor surface or will be removed In the case of a positive resist the exposed

area will be removed by the developer and the covered area will remain In other words

whatever is displayed on the photomask goes onto the sample surface On the other

hand for the negative photoresist the area that is covered from UV exposure will be

dissolved by the developer Figure 31 shows a 3D picture of both processes used in this

research

85

Figure 31 3D illustration of Optical lithography process used in this research

To obtain the desired pattern on the surface after UV exposure for a certain

duration the sample is required to go through a development procedure by using a

developer The developer is used to expel the dissolvable part of the photoresist after

being exposed to UV The usefulness of the developer depends on the photoresist ie

MF319 developer is specifically for positive photoresist while MIF326 is suitable for

negative photoresist Both types of developers will not harm the devices as they are a

kind of metal free ion solution Thus no free ion will change the characteristic of the

device The common practice in the clean room at the University of Manchester is to use

positive photoresist namely Shipley Microposit S1800 supplied by The Dow Chemical

and negative resist AZnLOF2020 (AZ2microm) which is supplied by MicroChem For the

S1800 series the thickness of the photoresist is determined by the last two digits ie

13microm thick for S1813 and 05microm thick for S1805 Both positive photoresists are used

Resist

GaAs

Dielectric

Photo Mask

Negative Resist Positive Resist

After Etching

86

mostly as protective area during wet chemical etching and as sacrificial dielectric layer if

higher temperature is applied ie 190˚C On the contrary AZ2microm which normally has a

thickness of 2microm is useful for patterning small dimension which leaves small gaparea

for the metallization process to fill It has good aspect ratio and useful in single layer lift-

off (post metallization process) In fact this type of resist can be thinned by diluting into

an Edge Bead Removal (EBR) solution and smaller device feature size can be obtained

The final dimension of certain devices (mesa size) is governed by the exposure

time the distance between photomask and samplersquos surface and the development time

The appropriate UV exposure time is required to avoid over-exposure which will cause

the spreading of light into the purportedly dark-field area[98] However the effect of the

exposure time differs between the photoresist types smaller opening area than the mask

pattern for negative resist and larger opening area for positive resist[45] The gap

between the photomask and wafer surface must be reduced as much as possible to avoid

the UV light going through the unwanted area To obtain a good gap value an applied

pressure from stage to the mask is required The normal pressure used in this research is

between 04 to 1 Pascal depending on the type of photoresist (negative and positive) as

well as its thickness Lastly the development time also influences the dimension of the

device Appropriate development time is required because ie over-development will

cause the polymerized photoresist to etch laterally resulting in bad patterned geometries

On the other hand under-development will cause non-uniformity in the surface after wet

chemical etching (positive resist) as well as causing lift-off problem (negative resist)

333 Etching Process

The etching process is a process of removing unwanted semiconductor layers to

define device geometry and isolate each individual device in one sample There are two

types of etching technique used in this research ie wet chemical etching and plasma

dry etching The wet chemical etching is based on Orthophosphoric(H3PO4) solution

which is a selective etchant to materials like GaAs InGaAs and AlAs The selective

etchant is referred to a solution that can etch away a specific semiconductor with a

specific etch rate The etch rate depends on the mixture ratio and concentration of the

solutions ie the higher the concentration the higher the etching rate In practice the

87

temperature humidity and epitaxial layer doping level also have an impact on the etch

rate Hence to minimize variation in the etch rate both temperature and humidity in the

clean room are constantly monitored and regulated Ambient temperature between 18degC

to 19degC is usually suitable while humidity is kept within 30 to 40 The advantage of

the wet chemical etching is that it is inexpensive controllable and with high throughput

highly selective and simple The mixture solution that is used in this work is

Orthophosphoric (H3PO4) Hydrogen Peroxide (H2O2) and Di-ionised water (H2O) with

ratios of 3150 3110 and 212 The ratio of 3150 provides an etch rate of about

600Aminute and is good to define the area and opening the under-cut in air bridges for

InGaAs samples[99] The 3110 ratio results with highly anisotropic shape but is easy to

control as the etch rate is about 1500Aminute for GaAs material system However the

212 etchant solution will provide extremely anisotropic etch rate of about

1000Asecond and is quite difficult to control For GaAsAlAs ASPAT diode it can still

be controlled since the thickness that needs to be removed is about 7000A (refer to

section 342 for details on the fabrication of ASPAT) Table 31 summarizes the etch

rate with different ratios and different selective materials Common to all the ratios

mentioned above are the isotropic side walls with lateral and vertical etch rate of 11

Table 31 Chemical wet etch characteristics using Orthophosphoric-based and Ammonia on GaAs

and InGaAs materials

Material Etchant Ratio Etch Rate (Aringminute)

GaAs H3PO4H2O2H2O 3150 600

GaAs H3PO4H2O2H2O 3110 1500

GaAs H3PO4H2O2H2O 212 60000

GaAs NH4OHH2O2H2O 118 2000~3500

InGaAs H3PO4H2O2H2O 3150 850

The plasma dry etch is purposely run to obtain extremely high anisotropic etch

profile vertically and horizontally It can be done via mask with positive photoresist and

self-align mask which is metal contact as a mesa protector The precursors that are used

to etch away GaAs InGaAs and AlAs layer in this technique are Methane and

Hydrogen (CH4+H2) On the other hand Carbon Tetrafluoromethane (CF4) and H2 is

88

used in removing Si3N4 The etch rate is determined by how much power is applied to

the plasma to hit the sample surface the pressure inside the chamber and amount of the

precursor In this research plasma Technology is used for dry etching This machine

which is a conventional OXFORD INSTRUMENTS 1990 machine can produce an etch

rate on average of 100Aminute for an RF power of 100mWatt

334 Sputtering (dielectric deposition)

In this work a Kurt JLesker PVD 75 is used to deposit Si3N4 layer on the sample

surface The deposition rate depends on the RF power that is applied To avoid surface

damage on the sample surface a sacrificial layer formed by SiO is deposited using the

Bio-Rad Thermal evaporator before transferring to the PVD 75 to start deposition with a

low (75Watt) RF power Once the deposition time reaches 30 minutes the power is

increased up to 200Watt As a result good uniformity of the dielectric layer is obtained

335 Metallization Process Lift-off and Annealing

The metallization process is a process in which metal contacts on semiconductor

devices are created The purpose of this process is to make a proper interconnection

between the semiconductor devices to other parts of the circuit elements In other words

it is to connect the semiconductor device to the outside world This process will allow

the device to be examined electrically so that all electrical characteristics can be obtained

ie resistance I-V curve capacitance conductance etc The metal scheme used in this

process depends on the semiconductor material In the case of GaAsAlAs ASPAT

diode Gold-Germanium (AuGe) Nickel (Ni) and Gold (Au) are used However in the

case of InGaAsAlAs ASPAT diode the metal scheme used is Titanium (Ti) and Au

The technique used in this process is resistive thermal evaporation

The metallization process starts by cleaning all the metallic sources and boats

ie tungsten boat Au Ni and Ti metals by using Trichloroethylene Acetone and IPA

consecutively for 5 minutes each in a high power ultrasonic bath This step is very

important to reduce the risk of contamination during thermal evaporation Once done all

the metallic materials are dried using a high-pressure nitrogen gun and then dipped in

89

Hydrochloric acid (HCL) solution for 2 minutes to de-oxidize the metals so that it has

minimal effect on the series resistance of the ASPAT diode

Two types of thermal evaporators were used extensively in this study both

Edwards Auto 306 (one denoted as Junior Auto 306) The latter is used to deposit alloy

type of metals while the former is used for the non-alloyed type of metals In the case of

GaAsAlAs ASPAT alloy type metals are used while for InGaAsAlAs non-alloyed

metals are used The cleaned metals are then loaded in the thermal evaporators and

placed on a resistive tungsten boat Each metal is placed on its specific tungsten boat to

avoid unnecessary mixture during the evaporation

Figure 32 Actual picture of thermal evaporator used in this study

Prior to loading the sample into the evaporator a standard fabrication process for

ASPAT diodes takes place by patterning the samples with AZ2microm negative photoresist

At the end of this step an opening area is created for the metal contact to be filled and

connected to the ohmic contact of the semiconductor The sample is also deoxidised

using a mixture of HCL and water in the ratio of 11 prior to evaporation This has to be

done in a very short time in order to avoid re-development of the native oxide layer

Inside the chamber the sample is securely placed on a chuck upside down facing the

filled tungsten boat The distance between the sample and metallic source boat is about

90

40 cm Figure 32 illustrates the actual thermal evaporation system used in this study

The thermal evaporator chamber is pumped down to reach a minimum pressure below

1times10-5

mbar before vaporizing the metal It is important that the mean free path between

metal amp sample is created and each vaporized metal stick firmly on the samplersquos surface

The normal practice in this study is to keep the vacuum pressure under 1 times10-6

bar so

that better device performance can be obtained The amount of current required to melt

down the metal is between 4 Amps to 6 Amp This amount of current is forced through

the tungsten boat and generates very high heat melting down the metallic source and

vaporizing it towards the sample surface The deposition rate for each metal can be

monitored by using a built-in film thickness monitoring (FTM) on the thermal

evaporator which proportionally depends on the amount of materials deposited The

GaAsAlAs sample is started with deposit of 55nm AuGe 13nm Ni and 500nm of Au

The reason of depositing AuGe first is due to the fact that the ohmic layer of the

ASPAT only can only be doped with a maximum doping of 4 times1018

cm-3

which is not

high enough for good conductivity Thus here Ge atoms will diffuse into the GaAs and

replaces Ga atoms during annealing process leading to higher doping levels (gt1 times1019

cm-3

) and hence improved conductivity

3351 Lift-off

The use of AZ2microm allows for the exact patterning of metals without the need for etching

using a single layer lift-off technique The negative photoresist also provides an undercut

profile which will create disjointedness between the desired metal pattern (on the

semiconductor) and undesired metal (on photoresist) The process of getting rid of the

unwanted metal from a sample surface is called lift-off The process starts once the

evaporation process is accomplished The sample is soaked into 80˚C N-Methyl-2-

Pyrrolidone (NMP) solution for usually 20 minutes (fast lift-off process) In most

instances the sample is in NMP for more than 12 hours in ambient temperature (slow

lift-off process) as the NMP solution is not destructive to the sample In this solution

the negative resist will be softened and the metal part which sticks on it will also be

eliminated from the sample As depicted in Figure 33 the lift-off process for a single

device shows the usual photoresist undercut profile observed To ensure that NMP

91

residues on the sample surface are completely removed DI water is used to rinse the

sample which is then blown dry with a nitrogen gun

Figure 33 Single layer lift-off process using negative photoresist

3352 Annealing

The alloyed (AuGeNiAu) metal stack requires a thermal treatment called annealing to

improve the ohmic contact between metal and semiconductor The sample which has

deposited top and bottom contacts is loaded into an annealing furnace at a temperature of

420˚C for 2 minutes In the case of GaAs during thermal annealing Ge atoms penetrates

into the GaAs crystal for approximately 70nm-250nm depending on evaporated

thickness of the metal layers annealing temperature as well as time [100] In this work

the total metal thickness evaporated for each contact layer is around 500nm This

thermal annealing treatment will also melt down the Au if it is too thick and is subjected

to too long a heat treatment Therefore it is not advisable to do annealing after the

sample is coated with bond pad metals as it can result in short circuited devices

sometimes

Az2micro Negative

photoresist profile

after UV exposure

and development

Evaporation to

form metal layer

(AuGe Ni Au)

Desired metal

contact

Metallisation

process

Lift-off undesired metal

92

34 GaAsAlAs ASPAT Process Optimization

As mentioned earlier this section present details two major process flows of the

fabrication process for the ASPAT diodes which utilised three stages of development of

photomasks design namely a ldquoFirst generation mask designrdquo (1st Gen) a ldquoSecond

Generation mask (2nd

Gen)rdquo and a ldquoThird Generation mask (3rd

Gen)rdquo design The 1st

and 2nd

generations mask designs are the designs that were produced in the first stage of

this work to develop the fabrication process know-how and to get familiarized with the

actual fabrication techniques in the cleanroom The difference between the 1st Gen and

2nd

Gen masks is in term of the development towards realizing Air Bridges and

Dielectric Bridges which were mostly covered by the 2nd

Gen mask design The

analyses in term of reproducibility repeatability and manufacturability for process

control and current-voltage characteristic as well as ultimately RF measurement are

obtained on relatively large size devices via these two mask designs The large area

emitter dimensions range in size from 15times15 microm2 to 100times100microm

2

The 3rd

Gen mask design was designed based on the optimization of the 2nd

Gen

mask which was to realize ASPAT diodes that are able to work at very high operating

frequencies Therefore in such design the ASPAT devices have to have a minimal

amount of capacitance and low series resistance this is can be achieved by shrinking the

emitter size of the diode to the smallest area possible as well as optimising the

fabrication process ton reduce parasitics The smallest fabricated devices designed on the

new mask has an 2times2 microm2 MESA area which also includes Ground Signal Ground bond

pads for both device and de-embedding structures (open and Short) for RF

measurements

Before the commencement of any fabrications and designing any layouts on eg

Si GaAs InP etc the epitaxial layer must be grown first to a desired design In the

University of Manchester the epitaxial layer structures are grown using one of the two

Molecular Beam Epitaxy (MBE) machines which are either the RIBER V90H or the

V100HU system Both systems are managed by Professor Missous Epitaxial wafers or

sample grown by each system are identified by a prefix and numbers that are prefix

VMBE for the V90H system and XMBE for the V100 system The epitaxial layers are

93

grown on four-inch wafer diameter The maximum diameter that can be grown on using

the V100H is 8 inches but generally single 4rdquo or 4x4rdquo wafers are used The wafers are

then diced and cut using a diamond scriber into 15mm times 15mm tiles for easy handling

and fabrication in the D12 cleanroom lab in the University of Manchester

In this section the structures fabrication and performance of the various ASPAT

diodes for both repeatability and high frequency applications will be discussed further

The ASPAT is manufactured on wafer sample XMBE304 which is a GaAsAlAsGaAs

lattice matched to a GaAs semi insulating substrate is the main focus The ground works

on this ASPAT such as the initial design and fabrication process flow optimization had

been conducted by fellow PhD colleagues in the group led by Professor Missous at the

University of Manchester

341 ASPAT Devices used in Fabrication

3411 XMBE368 and XMBE307

Table 32 Epitaxial layer of Doped substrate samples

Thickness (Aring)

Layer Material Doping (cm-3

) XMBE368 XMBE307

Top Ohmic1 GaAs (Si) 4times1018

~3000 ~3000

Buffer 1 GaAs(Si) 4times1017

~50 ~50

Spacer1 GaAs NID 50 50

Barrier AlAs NID 28 28

Spacer 2 GaAs NID 1000 1000

Buffer GaAs(Si) 4times1017

50 50

Bottom Ohmic GaAs(Si) 5times1018

~3500 ~3500

Substrate GaAs (Si) Doped 50000 50000

These two samples were the first batch of diode structures used in this work and

were grown using the RIBER V100 MBE machine A great deal of work was expanded

to ensure that it is able to produce appropriate non-linear I-V characteristics The work

carried out including finding suitable fabrication process steps mask designs process

control limitations ie etching rates etc The results obtained from processing these

94

samples mostly on large area anode and cathode sizes and their analysis included both

growth profiles and fabrication process flows These samples were grown on doped

GaAs substrates As such the finished diodes were vertical structures and the fabrication

process has marked differences compared to undoped substrate samples While

XMBE368 had similar epitaxial layer profile to XMBE307 during growth the AlAs

layer was set to be stagnant (ie no-rotation of the substrate during growth) to

investigate the effects of slight variations in barrier thickness

3412 XMBE304

The next sets of samples were all grown on semi-insulating GaAs substrates

Table 33 details the epitaxial layer profile of sample XMBE304 the main work horse

of this research work The growth of this structure was performed on a multi-wafer

platen and consisted of 9 x 2rdquo wafers (from XMBE304A to XMBE304I)

Table 33 The epitaxial layers of sample XMBE304 with barrier 10ML~28nm

Epitaxial layer Material Doping(cm-3

) Thickness(Aring)

Emitter GaAs(Si) 4e+18

3000

Emitter 2 GaAs(Si) 1e+17

400

Spacer GaAs NID 50

Barrier AlAs NID 28

Spacer 2 GaAs NID 2000

Collector GaAs(Si) 1e+17

400

Collector 2 GaAs(Si) 4e+18

4500

Substrate GaAs(SI) Semi-Insulating

For a typical ASPAT structure the emitter is essentially highly doped to

4times10+18

cm-3

to provide accumulation of electron in the emitter contact region It was

purposely highly doped to also achieve low ohmic contact with the metal The spacers

are used to avoid diffusion of dopants to the subsequent layers The ASPAT structure as

mentioned earlier has a single AlAs barrier with a very small thickness sandwiched

between two different length GaAs spacer layers

Batch XMBE304 is the main focus in these studies All activities required for

repeatability reproducibility process flow and devices as well as qualifying new or

95

optimization fabrication technique for small devices and RF performance which were

then used for high frequency were based on the set of wafers grow in this batch

342 Fabrication process of GaAsAlAs ASPAT diode toward repeatability

reproducibility and manufacturability

As the new wafer structures have to be tested and evaluated to gauge the

performance of the ASPAT diodes their uniformity also needs to be tested so as to

ensure it exhibited fully functional diode with zero bias detection in minimal variation in

IV characteristics between diodes As described previously 4rdquo wafers are always diced

up into 15mm times15 mm size for ease of handling in the cleanroom and for masks cost

purposes

3421 Doped n+ Substrate Wafers

For wafers grown on n+ substrates (XMBE368 and XMBE307) a two-level mask

set is sufficient to complete the fabrication process In this case the devices are designed

with mesa structures and top metal contact on the upper surface of the wafer and a

bottom contact on the backside of the wafer One mask plate is used for defining the

mesa structure (wet etch) and the other mask plate is used for defining the metal pad

Since the mesa is relatively large (30times30microm2

to 100times100microm2) there is no requirement to

define bond pads for measurement purposes A prior RTD mask designed by fellow

colleague Dr Md Adzhar for his PhD work was used for processing the ASPAT diodes

as well in the first instance The detail fabrication process is summarized in appendix I

for the doped substrate wafers Figure 34 shows typical IV characteristics for 30x30microm2

emitter size diodes obtained from sample XMBE368

96

The I-V measurement was taken from two different tiles (15mm times15 mm) located

on top and bottom of a 4rdquo XMBE368 wafer The location 1 marked as a blue line in

figure 34 refers to a device on a tile taken from the top of the 4rdquo wafer while location

2 (red) refers to a device on a tile taken from the bottom of 4rdquo wafer At 07V the

separation difference in current is ~228 for both tiles This shows that the device in

tile 1 is less resistive than the sample in tile 2 implying that the AlAs layer for tile

located on the top corner of the wafer is thinner and thus allows more electrons to tunnel

through it at a given bias

3422 First Generation Mask Design (1st Gen)

-0015

0005

0025

0045

0065

0085

-15 -1 -05 0 05 1 15

I C

urr

en

t (A

mp

)

V Voltage (Volt)

Current vs Voltage (XMBE368_1) for 30x30microm2

MidLeft2

MidRight3

Figure 34 Current-Voltage characteristic of sample XMBE368 used

in this study at two different locations on the wafer tile

(a) (b) (c)

Figure 35 ASPAT mask design for (a) 100times100microm2 (b) 30times30microm

2 and (c) 15times15 microm

2 diode dimensions

designed in the 1st Gen Mask

Location 2

Location 1

97

The fabrication of the GaAsAlAs ASPAT conducted in this 1st Gen mask is

followed the established generic process flow for InGaAs material which was done by

former co-worker Dr Md Adzhar Md Zawawi[101] The generic process flow is fairly

simple consisting of four mask steps as shown in Table 34 below The fabrication

process starts with the sample being cleaned in a NMP solution at a temperature of 80˚C

with Acetone and Propan-2-ol (IPA) The purpose of cleaning using NMP and Acetone

is to remove any organic material The IPA is used to remove Acetone residues

Table 34 Generic fabrication steps established by Dr Md Adzhar [101]

Step number Process

1 Top Contact

2 Mesa Etch

3 Isolation

4 Bottom Contact

In the 1st Gen mask the first step in fabricating the GaAsAlAs ASPAT diodes is

to use the first mask to define the emitter contact area The emitter has three different

sizes 100times100microm2 30times30microm

2 and 15times15microm

2 The lithography technique uses the

negative photo-resists AZ2micron with 55 second UV-photolithography to pattern the

top contact Then it is developed using MC319 developer to clear and define the

exposed area for the metals to stick to The sample is then subjected to plasma etching to

remove all organic residues and contaminants Then it is dipped into a mixture of

diluted Hydrochloric acid and water HCL H2O with a concentration of 11 for de-

oxidation This must be done in a short time right before the evaporation in order to

ensure good contact between metal and semiconductor surface with very low oxide

formation in between The ASPAT device performance depends on this step as contact

to the channel is by means of current flowing through the anode to the cathode terminals

In our lab the approach taken to achieve the Ohmic contact is by evaporation of Gold

Germanium (AuGe) Nickel (Ni) and Gold (Au) metals layers on top of the cap layer

Subsequently the metal is defined via a lift-off process using NMP

98

The next critical step is to define the MESA or island The mesa etch mask is

designed with two options with 05 microm or 10 microm tolerance The different mesa

tolerances are introduced to act as a safeguard for the emitter from producing excessive

undercut caused by lateral etching This step is to isolate and eliminate the unwanted

GaAs which will electrically link the active layers as many devices will be fabricated at

the same time on the same wafer tile (15mm times 15mm size) The lithography process in

this step uses positive resist and is developed using MC326 developer This step is

achieved using a wet etch process where a non-selective etchants mixture of

H3PO4H2O2H2O etches down the epitaxial layers until it slightly exceeds the AlAs

barrier with an etch rate of about 600Aring to 900Aring per minute The outcome of the MESA

or island step is the formation MESA active layers which are surrounded by inactive

layers of semi-insulating substrate (when using semi-insulating substrates)

The isolation mask is a step that is basically the same as the MESA etch step

eg using similar etchants mixture same lithography process but a different mask This

step is used to etch down until the GaAs substrate is reached which means that the

etching time is longer than that in the MESA etching step The purpose of this step is to

fully isolate the device from other neighboring devices hence ensuring no electrical

connection exists between each device within one sample Since these ASPAT diodes

employ an air bridge structure of size 1times5 microm the two minutes mesa etches will

simultaneously provide an initial undercut through lateral etching for the air bridge

formation

Based on the results so far obtained using the first-generation mask which

provided large mesa areas the current voltage characteristics of the ASPAT as far as the

non-linear zero bias is concerned did work very well However there was a need to

reduce the ASPAT mesa area down to very small dimensions to achieve mm-wave or

THz detection frequencies It is certainly a general rule for semiconductor device which

operate at very high frequency to have extremely small lateral dimension which

minimises the capacitance within the ASPAT device Furthermore for wider adoption of

the technology it is also important to develop a simple reproducible and low-cost

fabrication method for ASPAT diodes Details of this fabrication process are attached in

appendix II

99

3423 Second Generation (2nd

Gen) mask (ASPAT-GSG)

This work intended to enhance the 1st Gen-Large Area ASPAT photomask by

adding many features including 2 times 2microm2 mesa areas ground-signal-ground (GSG) bond

pads to enable RF measurement and three options device processing eg Air Bridge

Dielectric Bridge for semi-insulating doped substrate and dry etch-mesa Other reasons

for designing this mask was also to qualify process steps when deploying thin (1microm

width) bridge to connect small mesa area ASPAT diodes to the co-planar GSG bond pad

for DC and ultimately RF measurements

The mask was designed to fulfil the basic rules of fabricating various types of

tunnelling diode for instance RTD and ASPAT The diodes layouts were designed using

the commercial software Advanced Design Software (ADS) from Keysight

Technologies Ltd Once the design was completed it was ready to be sent to Compu-

Graphics Company for printing and patterned on a special chrome coated glass plate

The new 2nd

Gen mask design was termed ASPAT with Ground-Signal-Ground

(ASPAT-GSG) and consisted of two main designs ldquoAir Bridge and Dielectric Bridgerdquo

where each contained eight diodes with different emitter sizes (100times100μm2 50times50μm

2

30times30μm2 20times20μm

2 15times15μm

2 10times10μm

2 6times6μm

2 and 2times2μm

2) The Air Bridge

design is comprised of Design 1 (doped substrate) and Design 2 (undoped substrate)

This can be selected by changing the order of each individual layer of the mask steps

The same options are applied for Dielectric Bridge design which included processing for

doped and undoped substrates The details of the masks will be explained in the

following paragraph

(1) Air Bridge

Design 1 Contains five steps or layers mask of size 15mm times 15mm

suitable for air-bridge for undoped semi-insulating substrate

Design 2 Consist of seven layers steps mask of 15mm times 15mm size

There are 413 die chips in about 6mm times 6mm sizes in this design Figure 36 shows both

type of design for Air Bridge mask processing

(i) Mask 1- Top Contact

(ii) Mask 2- MESA

100

(iii) Mask 3-Isolation

(iv) Mask 4- Bottom Contact

(v) Mask 5- Collector Bond pad with GSG

(vi) Mask 2A-Dielectric

(vii) Mask 6A-Collector Bond Pad with GSG

(2) Dielectric Bridge

Design 1 Consists of seven identical steplayers masks with lateral

lengths of 15mm times 15mm each suitable for fabrication on semi-insulating

substrate

Design 2 Also has seven steps layers masked with dimension of 15mm

times15mm each suitable for doped substrate processing

There are 357 die chips in this type of mask design with an estimated size of 6mm times

6mm separately The smallest emitter size that is connected with a dielectric as the

bridge is 6times6microm2 The smallest 2times2microm

2 diodes was designed with the air-bridge

connected to the emitter bond pad due to the difficulties of opening viascavity for less

than 2 microm2 devices Figure 37 show both options for Dielectric Bridge mask processing

(i) Mask 1- Top Contact

(ii)Mask 2- MESA

(iii)Mask 3-Isolation

(iv) Mask 4- Bottom Contact

(v) Mask 5-GSG bond pad

(vi) Mask 6- Via Dielectric

(vii) Mask 7- Dielectric Bridge

(vii)Mask 5A- Via Dielectric

(ix) Mask 6A-GSG bond Pad

(x) Dielectric Bridge

Figure 36 Air Bridge design (red circle) for large mesa area (100times100microm2) device with

option for doped substrates

Figure 37 The layout of 1st design of Dielectric Bridge (green circle) mask design for 100 times

100microm2 emitter size with option for doped substrate processing

101

34231 Fabrication Process of the Air-Bridge Design

The fabrication using 1st Gen mask as mentioned in Section 3422 is less

complex however the fabrication process for Air-Bridge mask design contains a few

additional steps which are to add bond pads for the Ground-Signal-Ground radio

frequency (RF) layouts If the sample is on a doped substrate another layer needs to be

added leading to a total of six steps all together

Table 35 Standard process flow for Air-Bridge design fabrication

Mask Number Step number Process

1 1 Top Contact

22A 2 Mesa EtchDielectric

3 3 Isolation

4

5

4

5

Bottom Contact

Bond Pad

In this work the samples that have been processed so far are XMBE304

XMBE314 (GaAsAlAs base material system) and XMBE326 (InGaAsAlAs base

material system) All of them are built up on semi-insulating substrate thus the normal

process flow was followed Step 1 to step 4 follow exactly the same route as the 1st Gen-

ASPAT Large Area mask formerly discussed in Section 3422 This process is then

continued by spin coating AZ2microm on the surface Mask 5 is used to pattern the bond

pads The metallisation scheme used for bond pads is TiAu and the thickness must be at

least 1 micrometre thick This is to minimise series resistance at the pads and ensure a

robust surface is created when used for DC and RF measurements

342311 Air-Bridge Process Optimization

Since the Air Bridge design approach is focused more on developing air-bridges

which have a width of 1 microm and as the smallest device is 2 times2 microm2 the negative resist

(Az2microm) which was used in the 1st Gen design type was changed to Az1microm After spin

coating an Edge Bead Remover (EBR) is required to remove the beads at the edge of the

sample this is to ensure no gap is created between samples and mask when ldquohard-

102

contactrdquo is applied during the photolithography step The use of Az1microm and EBR are

critical to enable the fabrication of small air-bridges and emitter sizes

The exposure time during the photolithography technique also needs to be

increased to achieve a 1microm size air bridge A longer time than normal is required thus

95 seconds is used for exposure under UV light Once finished appropriate

development time must be applied to ensure the line for the air bridge is perfectly

opened for the metal to fill up To get impeccable result in this step both combination of

exposure and development have to be tuned naturally leading to trade off in both Too

long exposure and developing time will break the bridge whereas for too short exposure

time and developing the air bridge will not open

There are two types of etching method used in this fabrication wet

(H3PO4H2O2H2O) and dry (NH4H2) etching Both have their own advantages and

disadvantages Wet etching is faster than dry etching However the etching profile is

highly isotropic and causes serious undercut The smaller devices which have

dimensional sizes of 6times6microm2 and below will ldquoshrinkrdquo the effective area under the top

contact metal Even though dry etching can have highly anisotropic etching profile

which can prevent excessive undercuts it requires a lot of time for etching the

semiconductor layers The NH4 and H2 plasma that are used in this technique do not only

etch the GaAs and AlAs layers but also etch the metal contact at the same rate as the

semiconductors Thus the metal area must be covered with a photoresist or thicker metal

(~1microm) must be used to counter this issue Figure 38 shows that the emitter bond pad

which is not covered by the resist will eventually be etched away by the dry etching

Figure 38 Dry Etching for the first run in this study

As mentioned in section 333 the advantage of H3PO4H2O2H2O wet etching is

that it can be made using two solutions fast and slow The fast solution is based on a

concentration ratio of 212 while the slow solution is based on a 3150 mixture The

103

212 fast solution will produce good anisotropic etch profile but is tricky to control due

to the rather high etch rate of 1000Aring per second Both fast and slow wet etching

solutions being carried out in this experiment are replicated from successful recipes that

were developed by other co-worker from Prof Missousrsquos group

342312 Issue of Over Etch under Top Metal (Wet Etching for Air-Bridge

Design)

The first run using the solutions mentioned above started with the deposited

metal as a top contact (Top contact must be defined first to make a bridge) then Mesa

etching to remove ~7000Aring GaAs using the fast etch solution The next process to be

followed was to employ the slow etch solution to etch down the epitaxial layer to the

substrate Doing this ensured that air-bridge was open and the devices isolated

individually Unfortunately unexpected results were observed where large undercuts

still appeared for both 2times2 microm2 and 6times6microm

2 devices Figure 39 shows severe undercut

under the big device that can clearly be seen in the digital scope Most probably this

problem occurs due to excessive time used for the wet etching (10 seconds for fast

solution and 10 minutes for the slow etch solution)

Figure 39 Severe undercut of 2times2 microm2 and 6times6 microm

2 devices

The second run was employed to reduce the wet etching time in total thus dry

etching was needed As discussed above CH4 and H2 are used to etch away the GaAs

and AlAs semiconductor material while O2 is used to remove polymer residues To

reduce redundant generated polymer during dry etching the process must be separated

into several goes (runs) In this run at the mesa etch step the metal area is covered by

S1805 resist and then the plasma etches down until the heavily doped Ohmic layer is

reached For the isolation step the slow wet etching solution is used to ensure the

semiconductor under the bridge is removed The etching time is still long ie over 5

Zoomed Zoomed

104

minutes Hence this will still generate appreciable undercuts Although this run

managed to reduce the wet etching time it still did not solve the undercut Figure 310

shows undercuts under the effective mesa area still occurring in this run

Figure 310 Device with emitter area of 2times2 microm2 and 6times6 microm

2 after dry etch and wet etched

Further investigation has been made to find the root cause of this issue Scanning

Electron Microscopy (SEM) picture were taken to see deep into the completed devices to

investigate what is actually happening Figure 311 shows clearly that the semiconductor

under the bridge and effective area under the 2times2 microm2 device went missing The same

goes to for the large device where about half the size of the designed effective area under

to metal was also unintentionally removed by the solution

Figure 311 SEM Images of the GaAs sample

Since this work was also run together with an InGaAs based sample

(XMBE326) new knowledge regarding etching profile was acquired The lesson learnt

from this run is that the wet etch profile of the Gallium Arsenic (GaAs) material is

totally different from that of InGaAs Figure 312 shows the cross section of GaAs and

InGaAs materials after wet etching The evidences of these issues are also indicated by

the SEM images in Figure 313 and Figure 314 This more or less explained the reason

why the semiconductor under the top metal contact is always missing when wet

chemical etching is applied

Zoomed Zoomed

105

Figure 312 The cross-sectional view of InGaAs (left) and GaAs (right) samples used in this study

Figure 313 SEM images taken from the XMBE304 samples and a GaAs test sample

Figure 314 SEM images for InP and InGaAs taken from [56]

Furthermore previous work done by co-worker Dr Md Adzhar Md Zawawi has

optimised the process for RTD samples based on InGaAs and InAlAs heterojuction

semiconductor materials In this work Dr Zawawi found out that achieving submicron

dimensions is possible when using the soft reflow technique on InGaAs[102] However

it does not apply to GaAs when using the same technique There are few reasons to

106

believe that the work that was carried out by the previous student is not repeatable to the

XMBE304 GaAsAlAs material The main reason is due to the much thicker Ohmic

layers used in the GaAs of ASPAT structures which requires longer etch time The

sample that was used in the previous submicron work namely XMBE277 (RTD) [101]

and corresponding epitaxial layer structure is attached in appendix III The thickness

needed to be removed for the MESA step is 1421Aring compared to XMBE304 (ASPAT)

which is ~6900Aring

The fabrication results shown from the Air-Bridge design does not seem

favourable to the GaAsAlAs heterojunction sample These include unreproducible IV

characteristics and excessive undercut under the metal Therefore the GaAsAlAs

ASPAT sample cannot be processed using this type of mask (Air-Bridge approach) The

fabrication efforts were then moved to the Dielectric Bridge design described below

342313 DC measurement

Fully functional ASPAT I-V characteristics were not obtainable in this run due to

severe damages caused by the undercuts that happened underneath the top metal contact

As can be seen from Figure 315 the behaviour of the current response suggests a very

leaky diode This confirms that the air bridge approach does not work with the

GaAsAlAs ASPAT diode

Figure 315 Short circuit behaviour on one of the fabricated device in this run

-0015

-001

-0005

0

0005

001

0015

002

0025

003

0035

-3 -2 -1 0 1 2 3

Cu

rren

t (A

mp

)

Voltage (Volt)

30x30BP

107

34232 Fabrication Process of the Dielectric-Bridge Design

Since the Air-Bridge design was not successful an alternative Dielectric-bridge

process flow was then designed to solve this issue The first run in this Dielectric-bridge

process was performed to ensure that when the fabrication was completed it was able to

produce the correct and reproducible ASPAT current-voltage characteristics The

process flow in this run is to follow the initial design which identifies the steps according

to the mask number Step 1 to step 4 in the process are exactly the same as in the 1st Gen

mask design Step 5 which is the bond pad process is then continued with spin coating

the AZ2microm After exposure and development a TiAu metal scheme with a minimum

thickness of approximately 1 microm is deposited

Table 36 Standard fabrication process flow for Dielectric-Bridge design

Mask Number Step number Process

1 1 Top Contact

22A 2 Mesa Etch

3 3 Isolation

4

5

6

7

4

5

6

7

Bottom Contact

Bond Pad

Via Etch

Bridge

Step 6 is quite complicated compared to the other steps where the introduction of

S1805 resist is used as a dielectric layer to prevent short circuit between top and bottom

contacts This includes opening the smallest via (holes within the dielectric layer) of 2times2

microm2 and 6times6 microm

2 emitter sizes This step started by spin-coating the S1805 resist and

was then followed by baking it at 150˚C for 30 minutes The longer baking time is to

ensure it is hard enough to be deployed as a dielectric layer Once the S1805 had

hardened as dielectric layer the sample was then spin-coated again with S1813 (thicker

resist) on the dielectric layer to act as a mask for vias opening (to cover the dielectric

layer from via etching damages) After the S813 was developed O2 plasma etches was

108

applied to remove the S1805 that covered the top metal and bond pads The hole on the

metal was then exposed This required only two minutersquos plasma etches time

As soon as the holes were created within the dielectric layer a quick clean to

remove S1813 was performed using Acetone and IPA This has to be fast enough to

remove the S1813 without removing the S1805 dielectric layer Mask 7 then defined the

area for the metal that will fill up into the open vias This metal connected the top

contact metal of the device to the bond pad (GSG pad) The same metal scheme as

mentioned in Section 3422 was used as the bond pad

The fabrication using the dielectric bridge approach in the 2nd

Gen mask design

appeared to work successfully when initial I-V characteristics were taken and it also

showed that the ASPAT diode was fully functional However using hardened S1805 as

a dielectric layer is not a good practise for manufacturing device since there were left

over residue on the sample surface as depicted in Figure 316 This residue comes from

the non-uniformity of S1805 resit that was formed during heat treatment Therefore the

next run was to avoid using S1805 but replacing it with a standard dielectric layer based

on Silicon nitrite (Si3N4)

Figure 316 The surface of the sample after final processing

342321 DC measurements

Given that the process of qualifying mask steps for GaAsAlAs ASPAT diodes

looked promising using the dielectric bridge approaches initial I-V characteristics

measurements were carried out for XMBE304A Figure 317 shows that the I-V

characteristics are comparable to those of the other runs ie in the 1st Gen mask

Noticeably the IV characteristic (Figure 317) between each device size does not scale

109

on a single line This is attributed to the unintentional variation in the size of the emitter

during wet etching within each sample

Figure 317 IV characteristics of sample XMBE304 for emitter sizes of 1000 microm2 2500 microm2 900

microm2 400 microm2 225 microm2 100 microm2 and 36 microm2

The current densities shown in Figure 317 are calculated based on a reduced

effective area of the devices by a factor of 08 This assumption is made as actual area

(without metal) is un-measurable Theoretically devices with small areas will have

higher resistances compared to larger devices However the opposite occurred in these

devices This problem is still under investigation It could be due to the spreading

resistance producing different values according to device sizes (smaller devices have

longer D while bigger devices have shorter D)

34233 Dielectric-Bridge Process Optimization (Si3N4)

Due to poor surface roughness of the sample when using S1805 as dielectric

layer this run was employed to improve the surface quality by using Si3N4 In order to

get the actual size of the emitter the processing needed to start by defining the

semiconductor area using either top contact mask or mesa etch mask and not to deposit

the metal first Smaller effective area will be obtained if the top contact mask is used to

define the area compared to the MESA mask since there are 05 microm and 1 microm tolerances

designed in the MESA mask

-000001

0

000001

000002

000003

000004

000005

-2 -1 0 1 2

Cu

rren

t D

en

sity

(A

mp

microm

2 )

Voltage (Volt)

Current Density vs Voltage

density36

density100

density225

density400

density900

density2500

density10000

110

Table 37 New arrangement of the mask number and step in Second Run

Mask Number Step number Process

3 1 Isolation

22A 2 Mesa Etch

4 3 Bottom Contact

6

1

7

5

4

5

6

7

Via Etch

Top Contact

Bridge

Bond Pad

The fabrication was initiated by isolating each device using Mask 3 to define

them individually Then MESA mask (Mask 2) was used to cover the area from wet

etching optical measurement were used to obtain the actual size of the emitter without

metal Figure 318 shows the process results after H3P04H2O2H2O etching for a ratio of

3150

Figure 318 Optical images after MESA etching (a) 1 microm Tolerance (b) 05 microm Tolerance

Although the mask was designed with smaller size of 6times6microm2 using the slow etchant

solution the effective emitter size can still shrink to ~3times3 microm2 and ~2times2 microm

2 as can be

seen in Figure 318 The next step was to define the bottom contact using a metal scheme

that is suitable for GaAs ohmic layer ie AuGeNiAu After lift-off the samples were

then cleaned and spin-coated with AZ2microm Then the area was defined by using Mask 6

allowing the exposed area to be filled by Si3N4 which is the dielectric layer used in the

second run However a tricky issue happened here as the Si3N4 was difficult to lift-off

especially on the smallest emitter area in this mask In this run the lift-off process was

successfully done after three days Step five which uses Mask 1 was to define the top

111

contact The same metal scheme will fill up the tiny holes within the dielectric layer to

attach on the emitter semiconductor Figure 319 below shows an optical image of a

6times6microm2

device after the top contact lift-off process Once lifted-off annealing took place

to ensure Ge diffuse into the GaAs contact layer to lower the resistance

Figure 319 After lift-off processing

Step six uses Mask 7 to define the area of metal connection between active areas

to the bond pad In this step and step seven a different metal scheme from the top and

bottom contact was used to connect metal to metal Here a TiAu scheme was used The

final step (step 7) was to define the bond pad area In this context mask 5 is used The

bond pad requires thicker metal thickness to reduce the series resistance and to increase

the robustness of the metal surface when probed with needles during measurements

This second run of 2nd

Gen mask dielectric approach went well with better samplersquos

surface when using Si3N4 as the dielectric layer compared to the previous run using

S1805 Once the bond pad were defined for each device as done in previous run the

preliminary current-voltage characteristics of the Si3N4 run were obtained This is

described in the next section

342331 DC measurements

As mentioned in the previous section the Si3N4 run started with a defined emitter

area This allows actual ASPAT dimension to be measured thus the current density

versus voltage that are plotted in the following figures are based on actual measured

sizes

112

Figure 320 The current density of XMBE304 processed using Si3N4 as dielectric layer

Although the actual area of each device is obtainable from measurement the current

density of this sample does not scale well either Figure 320 show the current density is

not scalable from 03V to 1V This probably happened due to the same issue related to

the spreading resistance for each diode varying To investigate this issue new runs were

required with intention of reducing Rsprd and were expected to lead to more scalable

devices Therefore a study of reducing the spreading resistance (Rsprd ) due to

contribution of the large D-gap was carried out using the 2nd

Gen mask with the Si3N4

dielectric bridge approach The following section discusses in detail the steps used in

reducing the gap between the top and bottom contact for the GaAsAlAs ASPAT diode

34234 Dielectric-Bridge (Si3N4) Process Optimization by varying the D-Gap

The series resistance as discussed in detail in Section 26322 is generally due to

the total contribution of specific contact resistance in particular the diode size the sum

of all doped layer resistances that sandwich the main ASPAT layer and the spreading

resistance which comes from the lateral structure diode design[33] Thus in order to

acquire better performance at high frequency these contributors must be controlled One

parameter that can be controlled through this fabrication process is to reduce the

spreading resistance by controlling the separation between top and bottom contact ie

the D-gap as indicated in Figure 321

-000001

0

000001

000002

000003

000004

000005

-1 -05 0 05 1

Cu

rren

t D

ensi

ty

(Am

pm

2)

Voltage (Volt)

Current Density vs Voltage

D36

D100

D225

D400

D900

D2500

D10000

113

Figure 321 side view of lateral ASPAT structure

Without designing a new mask at this stage the same 2nd

Gen Dielectric-Bridge

mask was used but the process flows did not follow the sequence order of mask

numbers The reason for deciding not to design a new mask to reduce the D-Gap was

because the fabrication process flow had not yet confirmed it repeatability and

reproducibility Thus the available photo mask at that moment was fully utilised

Furthermore the feature of the 2nd

Gen mask that included the mesa tolerance can be

exploited in this study Hence the initial idea was to reduce the length of D by using the

tolerance that was designed in with Mask 2 (MESA etch mask)

Table 38 New arrangement for the Third run using Dielectric-Bridge mask

Mask Number Step number Process

1 1 Mesa Etch

3 2 Isolation

3

2

3

4

Bottom Contact

Mesa Cover

6

1

7

5

5

6

7

8

Via Etch

Top Contact

Bridge

Bond Pad

The process therefore was initiated by cleaning the sample and was then

followed by spin-coating it with S1805 to cover the emitter contact and define the actual

D-gap

114

size of the diode This first step used Mask 1 with wet chemical etching using the slow

solution (H3P04H2O2H2O etch with a ratio of 3150) Once the resist was striped

optical measurements were performed to obtain the effective area of the emitter Figure

322 below shows the smallest emitter area obtained after etching

Figure 322 The measured size of the emitter area and the length D (blue color marked)

Step two was to isolate the devices individually by using Mask 3 This was

performed using the fast etch solution (H3P04H2O2H2O etch with ratio of 212) This

took about 10 seconds to remove about 10000Aring of material

Step three which involved two masks was the most complicated in this process

Firstly Mask 3 was used to define the bottom contact by covering the sample with

AZ2microm and then the sample was hard- baked at 190 ˚C for 4- 5 minutes to ensure it had

totally dried before applying Polydimethylglutarimide (PMGI) The PMGI type used in

this process is Lift-off Resist (LOR) SF11 After spin-coating the SF11 at 7000RPM for

45 seconds it was post-baked at 190 ˚C for 4 minutes This was done to ensure that the

LOR SF11 was hard enough for the S1805 to stick on it This was followed by spin-

coating S1805 at 4000RPM for 30 seconds and exposing it under i-line UV for 20

seconds and developed using Micro Dev mix with DI water (11 ratio) for one minute

When the correct shape of S1805 was formed the sample was exposed under flood UV

for 15 minutes The SF11 was developed with 101A developer for 1 minute Figure 323

below summarizes the whole process in step three The step three processes concluded

by depositing the bottom contact with alloyed metal scheme and once lift-off had taken

place the next step continues as usual

115

Figure 323 Summary of LOR technique steps

Step four as well as the subsequent steps were completed in this run by copying

from the previous run (ie second run) which was to deposit the Si3N4 as the dielectric

layer and so on All the devices that underwent fabrication using this technique were

measured under optical microscope MC60 assisted by the Walsall software tool in the

lab From the measured value obtained it was expected that reducing the length of D in

this technique would improve the spreading resistance by up to 70 from the original

2nd

Gen mask design Table 39 summarizes the calculation of original spreading

resistance and improvement using this technique The calculation method and equation

are taken from Section 26322

Table 39 The outcome of the spreading resistance before and after using LOR technique

Device Size(microm2) D-original(microm) RSprd(Ω) D-new(microm) RSprd(Ω)

100x100 4 056 163 023

50x50 14 327 194 054

30x30 19 602 163 097

20x20 21 832 212 135

15x15 22 1007 202 18

10x10 24 1293 205 252

6x6 25 1643 230 418

116

Technically from the experiment in this run it was observed that the development time

also controlled the final length of D longer development time produced shorter D

lengths while shorter time yielded larger D gap lengths Therefore in this run the

development time was kept constant because of the desired D length from the mesa

tolerance in general is only ~2microm

342341 DC measurement

As usual when wafer processing was completed early DC measurement took

place to check the diodes performance Each diode size was measured and compiled

into one graph as shown in Figure 324 It is clear that the performance was better than

that in the previous run in term of scalability and current conductivity

Figure 324 IV characteristic in 3rd run (reduced d-Gap) for each diode size on the mask

To compare the current conductivity from this run and the previous run current

density at 05V was measured for both 2nd

and 3rd

runs Assuming the contact resistances

(TLM) for both runs were constant for 50times50microm2 device size the improvement of

current in the 3rd

run is about 92 This showed that such approach to reduce the D-gap

-500E-06

000E+00

500E-06

100E-05

150E-05

200E-05

-2 -15 -1 -05 0 05 1 15 2Cu

rre

nt

De

nsi

ty (

Amicro

m2)

Voltage (Volt)

Den_100microm2

Den_225microm2

Den_400microm2

Den_900microm2

Den_2500microm2

Den_10000microm2

Den2_36microm2

117

was successful in this run A new mask design was ready to take the challenge for

processing toward millimetre and sub-millimetre wave application

343 Fabrication process of GaAsAlAs ASPAT diode toward High frequency

Applications

So far the information that can be gathered from previous processing is that the

optimum process flow is achieved through dielectric approach design The effort in

reducing the series resistance by lowering its biggest contribution was attained through

lowering the D-Gap in the structure Once everything had been optimized ie the

process flow series resistance junction capacitance etc it was time to develop a new

mask design which only focused on the development of small ASPAT devices for use in

the millimetre and sub-millimetre wave regions

3431 Third Generation (3rd

Gen) mask design

The 3rd

Gen mask design was developed by taking into the account every aspect

of parameters that can contribute to the device by means of robust devices that are able

to function properly at ultra-high frequency The device cut-off frequency is given by

119891119888119906119905minus119900119891119891 =

1

2120587119877119904119862119895

(31)

Here RS is the series resistance and Cj is the junction capacitance To obtain high cut-off

frequencies Rs and Cj must be kept as low as possible From the fabrication point of

view two parameters that can be directly and easily controlled are the device area (A)

and the D-gap (which contributes to Rs)

Therefore calculations were made to find the best option Table 310 shows the

calculated value of capacitance (eq230 in page 44) and cut-off frequencies (eq31

above) for the ASPAT diodes studied These two equations extract the cut-off frequency

of the ASPAT assuming no external effects and for fully depleted devices The

XMBE304 ASPAT sample is expected to be suitable for millimetre wave applications

if small devices are successfully made In real devices sub-millimetrewave operation

can be hard to achieve due to increased series resistance and other process related

parasitics

118

Table 310 DC and RF characteristics for XMBE304

Device Size (microm2) Fully depleted

Capacitance(fF)

Calculated Series

Resistance (Ω)

Fully depleted Cut-Off

Frequency(GHz)

10000 5490 04 72

900 490 15 216

36 198 7 1148

16 879 11 1646

4 22 29 2500

Therefore in the 3rd

Gen mask design the smallest device that can possibly be obtained

in the GaAs based material fabricated using i-line lithography which is available at the

University of Manchester is 2times2microm2 and the gap between top and bottom is 15microm at

least The connecting bridge technique applied only utilised dielectric bridge method for

GSG features Figure 325 shows the layout of actual 3rd

Gen mask used in this study

There are 344 die chips on this type of mask design It also includes the Ground-signal

ground pad with 50um pitch for each chip and six de-embedding test structures as well

as eight TLM structures

Figure 325 3rd

Gen Mask layout showing actual devices de-embedding TLM and alignment mark

structures used in this study

119

This mask is designed generally from optimizations from 2nd

Gen Mask which

deploys a dielectric bridge for connection between the devices to the bond pads

Consequently the processing steps are not being much different but mostly follow what

is shown in Table 311 below The difference only applies to the much smaller mesa

size Other features included in this mask are de-embedding structures for RF

measurements via-hole test structure TLM structures and parallel plate capacitance test

structures

Table 311 3rd

Gen Mask process step

Mask Number Step number Process

1 1 Mesa Etch

2 2 Isolation

3

4

3

4

Bottom Contact

Mesa Cover

5

6

7

5

6

7

Via Etch

Top Contact

Bond Pad

The high frequency fabrication process flow is summarized in the following section

which shows illustrations in three dimensional and cross sectional (Figure 326) views

for easy understanding

34311 Step by Step Processing

0 Wafer preparation

Cleaning using NMP and DI

water or

Acetone and IPA

120

1 Mesa Etch

2 Isolation

3 Bottom Contact

4 Mesa Cover (Dielectric Deposition)

5 Via Etch

Etching using etchant solution

H3P04H2O2H2O etch with ratio

of 3110

Etching using etchant solution

H3P04H2O2H2O etch with ratio

of 212

Metal deposition with metal

scheme AuGeNiAu for

~500nm thick

Dielectric deposition 500nm

thick Si3N4

Via etch to open holes for metal

contact using reactive ion

etching using CF4

121

6 Top Contact

7 Bridge and Bond Pad

Figure 326 Step by step device fabrication (in 3D and cross-sectional view) for semi-insulating

substrate device type used in this study

Figure 327 Example finished process device with bond pad using 3rd

Gen mask

3432 TLM measurements

Transmission Line model (TLM) measurements for this run were carried out

after annealing (ie at 420˚C for 2 minutes) for process control monitoring by extracting

contact resistance (Rc) values These values are a measure of the quality of ohmic metal

contacts for a given process As discussed in Chapter 2 the TLM technique used four-

Metal deposition with metal

scheme AuGeNiAu for ~500nm

thickness and thermal annealing

420˚C for 2 minutes

Metal deposition with metal

scheme TiAu (1microm thick)

122

point measurement on TLM test structures located around the 15times15mm2 tiles as shown

in Figure 222 (in page 55) Normally five TLM structures are measured across the tile

for both top and bottom contacts Figures 328 and 329 display graphical TLM results

for top and bottom contact respectively As can be seen both graphs exhibit excellent

uniformity

Figure 328 XMBE304 TLM measurement for the top contact after annealing

Figure 329 XMBE304 TLM measurement for the bottom contact after annealing

Based on the graphs above for the top metal contact the average contact

resistance (RC) value using the metal scheme of AuGeNiAu is found to be ~005Ωmicrom

and the sheet resistance (RSH) is 22Ω However for the bottom contact the average

value for contact resistance is 012Ωmicrom and sheet resistance obtained is 26Ω Both

values are in a very good agreement with the known doping of both ohmic contact

0

2

4

6

8

10

0 10 20 30 40 50

Res

ista

nce

(O

hm

s)

Spacing (microm)

After 420C for 2mins top contact

y = 02592x + 02375

0

2

4

6

8

10

12

14

0 10 20 30 40 50

Res

ista

nce

(O

hm

s)

Spacing (microm)

After 420C for 2mins bottom contact

y = 02157x + 01053

123

layers Therefore the specific contact resistance (Eq 250) that contributes to the total

series resistance can be calculated and the value obtained is 15Ωmicrom2 and 54Ωmicrom

2 for

top and bottom TLM structures respectively

3433 DC characteristic measurements

Again once the wafer processing is completed room temperature DC

characteristics are taken using an HP 414B or HP500B parameter analyser and its actual

setup is as described in Chapter 2 This initial measurement was to ensure the

functionality of the diodes For this run using the 3rd

Gen mask the I-V measurements

for mesa active area of 4times4microm2 6times6 microm

2 and 10times10 microm

2 were taken and are depicted in

Figure 330 Figure 331 and Figure 332 respectively Nine diodes were measured for

each three device sizes to check their uniformity Thus the average current and standard

deviation are taken at two voltage steps of 1V and 15V Table 312 summarizes the

standard deviation for each measured data obtained from this final run

Figure 330 XMBE304 GaAsAlAs ASPAT measured IV characteristics at room temperature for

4times4microm2 mesa size

-00005

0

00005

0001

00015

0002

00025

0003

00035

-3 -2 -1 0 1 2 3

Cu

rren

t (A

)

Voltage (V)

C01_4um_1P_SI

C22_4um_1P_SI

C40_4um_1P_SI

W01_4um_1P_SI

W22_4um_1P_SI

W40_4um_1P_SI

AI01_4um_1P_SI

AI22_4um_1P_SI

AI40_4um_1P_SI

124

Figure 331 XMBE304 GaAsAlAs ASPAT measured IV characteristics at room temperature for

6times6microm2 mesa size

Figure 332 XMBE304 GaAsAlAs ASPAT measured IV characteristics at room temperature

for 10times10microm2 mesa size

Table 312 Standard deviation at two different voltages

Device size (um2) Standard Deviation 1V () Standard Deviation 15V ()

4times4 14 14

6times6 65 72

10times10 44 39

Noticeably the standard deviation of the device increases for the smaller size

devices This trend happened probably because the active mesa area is not uniform

causing different series resistances Since the smallest mesa active area which achieved

-0001

0

0001

0002

0003

0004

0005

0006

0007

0008

-3 -2 -1 0 1 2 3

Cu

rren

t (A

)

Voltage (V)

C02_6um_1P_SI

C23_6um_1P_SI

C41_6um_1P_SI

W02_6um_1P_SI

W23_6um_1P_SI

W41_6um_1P_SI

AI02_6um_1P_SI

AI23_6um_1P_SI

AI41_6um_1P_SI

-0005

0

0005

001

0015

002

0025

-3 -2 -1 0 1 2 3

Cu

rren

t (A

)

Voltage (V)

C03_10um_1P_SI

C24_10um_1P_SI

C42_10um_1P_SI

W03_10um_1P_SI

W24_10um_1P_SI

W42_10um_1P_SI

AI03_10um_1P_SI

AI24_10um_1P_SI

AI42_10um_1P_SI

125

a good I-V characteristic is 4times4microm2

in this run these devices were used for the next step

of characterisation which is the S-parameter measurement to extract their behaviour at

different frequencies

35 Conclusions

In this chapter basic fabrication techniques of GaAsAlAs ASPAT on both doped

and semi insulating substrates using standard I-line lithography as well as step by step

descriptions to achieve reproducibility in the process fabrication flow has been

demonstrated with relevant initial measured results Two major outcomes have been

demonstrated firstly related to repeatability reproducibility and manufacturability

mostly done on large device areas (15times15microm2 to 100times100 microm

2) Secondly a successful

process flow for small emitter size devices (2times2 microm2 to 10times10 microm

2) has been

developed

Subsequently two types of designs during process optimization were developed

namely Air-Bridge and Dielectric-Bridge approaches The latter approach seems

favourable for GaAsAlAs materials but was only successful when reducing the series

resistance of the device This problem was addressed by optimising the D-gap between

top and bottom contact which resulted in good scalability of each ASPAT dimensions

improving current conductivity by 92 and achieving a reproducible process On the

other hand in the former approach issues were encountered with over etching underneath

the diode effective area and thus it was hard to achieve reproducible devices

Repeatability reproducibility and manufacturability fabrication processes based on

dielectric bridge method were successfully developed This new process provides a

highly efficient and economical solution for the fabrication of GaAsAlAs ASPAT

diode Emitter sizes down to 4times4 microm2 dimensions are routinely and reproducibly

achieved in this process Series resistance which is an important parameter in

determining high frequency application are greatly reduced by changing the gap between

top and bottom contact

With all the new improvements implemented the possibility of the proposed Dielectric

Bridge method fabrication process was successfully applied for the fabrication of

ASPAT diodes

126

4 PHYSICAL MODELLING OF THE GAASALAS ASPAT

DIODE USING SILVACO

41 Introduction

Fabricating any device or circuit requires a lot of time resources cost etc

especially elements related to the production of semiconductor devices Furthermore in

realising such a device a clean room is required Hence fabrication and processing in

wafer fab need full attention to get it done with a higher rate of success reproducibility

and manufacturability One solution that can be highlighted which will be able to reduce

all the resources that are mentioned above is by using computer simulation approaches

or to be precise physical modelling For epitaxial layer based devices physical

modelling is a good choice as it would give a better understanding and insight into each

layer and how electrical characteristics are derived A software that is most suitable for

epitaxial layer physically modelled is SILVACO This software is a very comprehensive

tool to simulate epitaxial based devices and to predict their behaviour Such software

covers many aspects starting from the first principles of physic epitaxial layer

definition as well as device layout thus making it the most powerful virtual wafer fab

tool in the market

In this chapter the SILVACO packages are discussed The discussions include the

method of defining a new material defining models and constructing the AC or DC

supply to obtain the output characteristics of the virtual device Apart from these the

focus will also be on the ASPAT diode modelling simulation and analyses of the results

which will include the ASPAT structure suitable models and DC current-voltage

characteristics The dependencies of individual structure on the I-V curves will also be

highlighted Finally the discussion of results and analyses involving a range of

operational temperatures dependency as well as from comparison made to conventional

SBD used in this study will be examined

127

42 SILVACO modelling Tools

SILVACO is a modelling software introduced in 1984 by Dr Ivan Pesic It is

purposely created for electronicsrsquo devices physical modelling and characterization This

software company has become the major supplier for most of the Electronic Design

Automation (EDA) for circuit simulation amp design of analogue Mixed-Signal and RF

circuit market This Technology Computer Aided Design (TCAD) software can predict

the simulated device performances starting from first principles It has a package which

can provide Virtual Wafer Fabrication (VWF) simulation to the device designer and

which has the capability to perform two or three dimensional physical device modelling

by using the ATLAS simulator [103] SILVACO allows all parameters such as

electrical thermal and optical characteristics of a device to be simulated under desired

bias conditions It offers cost effectiveness as well as quick prediction of results for

many semiconductor devices compared to real experiments Some hands-on experiment

may not always perform hence SILVACO can be used as an alternative

The core of SILVACO is Atlas itself which provides a platform to perform DC AC

and transient analysis for such dimensional device structure regardless of the

heterojunction material type ie binary ternary quaternary etc As the brain of

SILVACO Atlas which receives input command files containing instruction text for

execution from a runtime environment known as Decbuild will process the instruction

text and display progression error and warning via Runtime Output All the calculations

of the resultsoutcomes of the simulation are plotted via a tool called TonyPlot which is a

tool to visualise the output Figure 41 below shows how precisely the physical

modelling takes place the process of building the structure how its parameters and

variables are defined how an appropriate model statement is selected how performance

is analysed and lastly how the outcomes are displayed

128

Figure 41 SILVACO Atlas simulation process flow

As can be seen in Figure 41 the structure specification statement is used to define

any desired structure by setting the command in Deckbuid of the following parameters

a) Mesh where the structure can be defined either in 2D or 3D Cartesian grids The

unit of the coordinates used is in microns and the spacing parameters which

define the netting size can be used to improve the accuracy of the analysis at any

given position The density of netting size in this statement determines the

processing time

b) Region where the multi-layers in a structure are defined and this statement has

to outline each layer that represents a separated region independently The mesh

must be assigned to a region and the sequence of the region is arranged from low

to high

-Mesh -Region -Electrode

-Contact -Material

Structure Specification

-Model -Interface

Model Specification

Method

Numerical Method

-Log -Solve -Load -Save

Solution Specification

-Tony Plot

Display Results

129

c) Electrode are used to define the location of bias point for a designed structure

when performing the electrical analysis In this work or for the case of a diode

two electrodes are allocated as an anode and a cathode In a vertical device the

latter electrode is placed at the bottom of the device while the anode is at the top

d) Doping this statement refers to doping concentration injected into the desired

region and it normally depends on the material types

e) Materials since SILVACO was developed specifically for Silicon-based devices

default parameters are set up for Silicon properties However the use of

materials statement allows SILVACO to run for different material ie GaAs

InGaAs etc In order to make it work this parameter is defined first and then

followed by the material name and its properties such as bandgap permittivity

conductionamp valence band discontinuities mobility etc

The most crucial part is to determine whether the simulated structure of a particular

device is correct or incorrect This is done by properly choosing the specific model

statement The Specific model statement is employed to express the physic equations

that are used during the device analysis The models statement depends solely on the

structure definition Examples are device structure with double barrier use Non-

equilibrium Green function (NEGF) model and single barrier uses Semiconductor-

insulator-Semiconductor (SIS) model for effective and accurate analysis process For the

case of a single barrier in SILVACO there are many model statements that can be used

for such analysis Therefore it is recommended to check model by model in order to

ensure that all the needed parameters are defined in the material statements and results

produced are valid

SILVACO is able to calculate such models by using different numerical methods

which means semiconductor device problematic is computed to make successive

solutions by random discretisation There are three different numerical methods that are

regularly used by SILVACO to perform its calculations which are Newton Gummel and

Block Basically this is solved by using a non-linear iteration procedure which begins

from an initial guess and which then uses an iterative process to find the predicted

solution The detail of these can be found in reference [103]

130

In order to turn on the problem solution the solution specification is defined This

include log solve save and load statement All these work together to provide data for

analysis by other functions The log statement is a file type that saves in memory and can

be loaded by Atlas Any solved device will be stored in the log file Therefore it is

necessary to define the LOG before SOLVE statement and close it after the calculations

While the save statement is used to store all data point to a node in the output file the

load statement is utilised to recall all the saved data to be read by Atlas

Finally all these files can be displayed or plotted on TonyPlot it is recommended to

make SET files at plotting point for a better visualisation The plotted or displayed files

in TonyPlot can be manipulated for scaling graphs overlaying different curves and most

importantly to export the data to other files formats

43 SILVACO Implementation GaAs AlAs ASPAT Modelling

The structures play an important role in determining the terminal output

characteristics Therefore to start simulating the ASPAT device the first thing that must

be specified is its structure As mentioned earlier in Section 3412 the ASPAT diode is

a top down multilayer structure This structure which is adopted from that of Section

3412 will be used as a basis to perform the ASPAT simulation Figure 42 shows the

structure of the ASPAT in this real simulation which is exactly the same as been

discussed in Section 26322 The ASPAT diode consists of two heavily doped (up to

5x1018119888119898minus3) GaAs contact layers on top and bottom slices adjacent to lightly doped (up

to 3x1017119888119898minus3) GaAs intermediate layers In between these layers is a sandwiched

structure consisting of two different lengths of undoped GaAs spacer layers and a thin

layer of AlAs that act as a tunnel barrier In this work the simulation result will be

compared with the fabricated measurement result depending on the size of the diode

The actual device that will be used to validate the simulation is the main device used in

this study (XMBE304) which is based on a lattice matched GaAsAlAs grown on a

semi-insulating substrate Therefore the design structures that are proposed in the

fabrication are compatible with the fabricated devices and are based on lateral structures

as can be seen in Figure 42

131

Figure 42 The GaAsAlAs ASPAT structures produced from simulation based on the diode

multilayer heterostructures on the right

In the consequent simulations a key observation regarding the AlAsGaAs

heterojunction is that there are two types of tunnelling processes direct tunnelling and

indirect tunnelling Figure 43 shows the Energy-Momentum (Ek) diagram depicting the

three valleys for the AlAs conduction band namely L Γ and X points In normal

circumstances the transition of electrons happens to the X-point which is the lowest

energy in the conduction band In the case of very thin barrier the tunnelling process

occurs at the Γ point in both AlAs and GaAs materials which is the direct tunnelling

process[18] It has been reported that in the case of an ASPAT diode tunnelling which

occurs at the Γ point will be the dominant component in the tunnelling current

Therefore the actual band gap will be different from the one at the X-point which is

which 216eV[15] By contrast the energy band gap at the Γ-point is around

283eV[104] Thus this simulation uses this band gap value

Figure 43 The Energy-momentum diagram for AlAs heterojunction semiconductor

216eV

Γ

X

E

K

L

AlAs

CB

VB

283e

132

The simulation code as attached in Appendix IV and the output of the simulation with all

the input mentioned above are shown in Figure 44 (a) and Figure 44 (b) Figure 44 (a)

is the band diagram at equilibrium and Figure 44 (b) is the band diagram when a bias is

applied

Figure 44 (a) The conduction band and the valence band of the ASPAT diode structure (b) the

energy band diagram of the ASPAT diode structure when under three different biases

44 Simulation Result and Analysis

Basically the SILVACOrsquos Atlas simulation package is used to calculate the I-V

characteristics from multilayers structures In the case of GaAsAlAs heterojunctions all

details of the structure as shown in Figure 44 above are calculated based on solving the

Schrodinger time-independent equation in each layer taking into account the variation of

effective mass and conduction band offset between GaAs and AlAs

For thermionic emission and tunnelling mechanism across an abrupt heterojunction

interface the general method used in SILVACO is taken from K Y Yang work [105]

The tunnelling current of the ASPAT diode uses the equation below[74]

119869 = sum2119898lowast119864119894(1 minus 119877)119896119879

1205872ħ3

119873

119894=1ln [

1 + exp (119864119865 minus 119864119894

119896119879)

1 + exp (119864119865 minus 119881 minus 119864119894

119896119879)]

(41)

Where m is the electron effective mass E denotes the energy of the electron R is the

total resistance k represent Boltzmann constant ħ is the reduced Planck constant EF is

the fermi level V is applied voltage and Ei is the electron energy perpendicular to the

-2

-15

-1

-05

0

05

1

15

1

82

16

3

24

4

32

5

40

6

48

7

56

8

64

9

73

0

81

1

89

2

97

3

Ene

rgy(

eV

)

Thickness (microm)

VB

CB

133

barrier The important parameters that enable SILVACO Atlas to perform correct

calculations and analysis are the choice of appropriate models The suitable model that is

available for evaluating the GaAsAlAs ASPAT is based on the Semiconductor-

insulator-semiconductor (SIS) model

Thus in this run Non-local Quantum Barrier Tunnelling Model (SISEL and

SISHO) are utilized specifically semiconductor-insulator-semiconductor mode This

model enables the tunnelling current between two semiconducting regions separated by a

quantum barrier to be calculated [103] It is assumed that the charge tunnels across the

whole barrier with the source or sinks at the interface with the semiconductor regions

Under the Non-Local Quantum Barrier tunnelling model another model that can be used

is semiconductor-semiconductor-semiconductor (SS) tunnelling model if the materials

are specified By correctly inserting all parameters with the right model an excellent DC

IV characteristic match between simulation and measurement can be produced as shown

in Figure 45

441 DC Current-Voltage Characteristic

In this simulation the current at each bias step and each mesh point can be set up by the

user however the detailed calculation such as formula usage methodology and

approach that is adopted by SILVACO Atlas is unknown As mentioned above in order

to produce the energy band diagram of the ASPAT the DC characteristic of the structure

can be solved by using the Schroumldinger and Poisson equation self-consistently

To ensure that these simulations are valid one fabrication was performed on an

ASPAT diode sample XMBE304 For this sample the structure parameters are the

same as has been set in this simulation The result of the measurement and simulation

are then compared Figure 45 shows that the simulation result is in excellent agreement

with the measured data for this sample

134

Figure 45 The current-voltage characteristic of an ASPAT diode (100times100microm2) and (4times4microm

2)

using SILVACO Atlas simulator for structure device XMBE304 showing excellent agreement

between simulated and experimental data

These fitted results were performed on both a large 100times100microm2 device and the

smallest obtainable from fabrication (4times4microm2) which was to be used for the repeatability

amp reproducibility studies as well as for high-frequency applications study respectively at

room temperature In order to get a good fit a few parameters had been modified in the

SILVACO software via the Deck-built tool for example energy band effective masses

and bandgap discontinuity of GaAs spacer and AlAs barrier (The mentioned parameters

values are summarized in Table 41) The percentages of bandgap discontinuity in

SILVACO using the ALIGN parameter is given by[103]

119860119897119894119892119899 =

Δ119864119862

Δ119864119862 + Δ119864119881

(42)

Where ΔEc and ΔEv are the conduction band discontinuity and valence band

discontinuity respectively The m0 in the tables denotes the electron rest mass Once all

agreement between measurement and simulation has been met the simulation is then

carried out with structure analysis at room temperature and different temperatures

simulation

Table 41 The parameter values used in this simulation

Material Bandgap(eV) ΔEg(eV) Effective mass(kg)

GaAs 1424 03 0067m0

AlAs 2835 071 0126m0

-00002

0

00002

00004

00006

00008

0001

00012

-2 -1 0 1 2

Cu

rre

nt

I (A

mp

)

Voltage V (Volt)

Current vs Voltage

Measurment

Simulation

135

45 Structure Analysis of ASPAT Diode

Once the device structure was modelled and having successfully produced a

precise band diagram as well as validated the simulation results with experimental I-V

characteristics the next step is to further analyse the relationship between basic device

structure and its I-V characteristics This approach is used to predict what would happen

to the DC output if some of the parameters were varied especially with regards to the

AlAs barrier thickness In the subsequence simulations the thickness of each main

ASPAT (unequal spacers and barrier) layer will be studied independently as a variable in

order to determine how each parameter affects the I-V characteristic in both magnitude

and curvature The analysis will also include manufacturing tolerance where the

structure parameter which will result in a 10 difference in their I-V characteristics is

examined[59 75] This will provide an overview of how precisely to manufacture each

layer of the device The following simulation is based on the XMBE304 structure with

emitter size of 4x4microm2

451 Dependencies of current on AlAs Barrier thickness

Since the AlAs barrier is what limits the transportation of electron flow and

hence the current (which depends exponentially on the tunnelling barrier thickness)

therefore the first analysis to run on the simulation is the variation on barrier thickness

In the simulation the barrier thickness is measured in term of the monolayer Generally

one monolayer can be calculated by dividing the lattice constant of the material by two

In the case of AlAs one monolayer is calculated as follows

119900119899119890 119898119900119899119900119897119886119910119890119903(1119872119871) =

119860119897119860119904 119897119886119905119905119894119888119890 119898119886119905119888ℎ119890119889 (5666Å)

2

(43)

= 283Å

The nominal value of the AlAs barrier thickness for sample XMBE304 is 283nm

ie ten monolayers In the first simulation test the current change due to the barrier

thickness variation from 9ML to 11ML was examined first followed by the amount of

barrier thickness change that would produce a 10 change in current The simulation is

setup by fixing all other parameters and varying the AlAs barrier thickness as mentioned

136

above with a step of 02ML In order to determine what fraction of a ML would yield a

10 difference in the current the barrier thickness is slightly changed to fit the curve for

both 5 above and 5 below the original curve

Figure 46 IV characteristics of the dependencies of current on AlAs barrier

The current-voltage characteristics of the ASPAT diode do change dramatically with

barrier thickness in forward bias but not that much in reverse bias (Figure 46) The

current decreases as the barrier thickness increases For a 1ML change in layer thickness

(from 9ML to 11ML) the current changes by over ~300 at 05V

Figure 47 Example of analysis at -1 and 1V to the current

-00002

0

00002

00004

00006

00008

0001

00012

00014

-15 -1 -05 0 05 1 15

Cu

rre

nt

(A)

Voltage(V)

9ML

92ML

94ML

96ML

98ML

10ML

102ML

104ML

106ML

108ML

11ML

973E-06

0

0000005

000001

0000015

000002

0000025

000003

0000035

85 95 105 115

Cu

rren

t a

t 1V

B

ias

(Am

p)

Barrier Thickness (ML)

Current change with tunnel barrier thickness

Forward Current

5

-5

137

From the simulation result shown in Figure 46 a 9945ML barrier thickness will

give a 5 higher current and a 10056ML barrier will give a 5 lower current (Figure

47) Therefore in total 01 ML difference yields 10 current difference These indicate

that in order to control the current within 10 barrier thickness difference the growth

precision in the barrier must be precise to better than 01ML Extensive studies have

also shown that the I-V characteristic of a GaAsAlAsGaAs diode is very sensitive to

the thickness of AlAs barrier This work has been reported elsewhere[17]

452 Dependence of current on Spacer I length l1

For the longer spacer length (l1) five different values are chosen from 01microm to

03microm The lengths are changed in the order of 005microm Therefore the arrangement of

length is as follows l1=01microm l1=015microm l1=02microm l1=0 25microm and l1=03microm

respectively The results are plotted from -15V to 15V anode voltage in Figure 48 The

I-V characteristic of the ASPAT diode does not change much with the length in the

forward bias region but in the reverse bias region the current decreases as the layer

thickness increases Here the l1 layer acts as a voltage arm and a small size device

cannot sustain big changes in spacer length Changing the length at the forward region

will also change the energy states on the anode side as well changing the states

distribution on the cathode side in reverse bias

Figure 48 I-V characteristic of the dependencies current to Spacer I layer

-00002

0

00002

00004

00006

00008

0001

-2 -1 0 1 2

Cu

rren

t (A

)

Voltage (V)

L=01

L=015

L=02

L=025

L=03

138

By fixing the current at -1V and 1V the current and the layer thickness relationship is

illustrated in Figure 49

Figure 49 Current changes with layer thickness l1

It is noticeable that there is a dramatic change in reverse current from 01 microm to

05 microm layer thicknesses However the forward current only falls slightly from 01microm to

015microm and is stable afterwards Hence for a small size device a large change in spacer

layer at the cathode will allow more current to pass

453 Dependence of current on Spacer II length l2

Finally is the variation in the spacer II Similar to spacer I above five values are

chosen for the shorter undoped GaAs layer length l2 the thickness is varied from

00025microm to 00075 microm (steps are l2=00025microm l2=000375microm l2=0005microm

l2=000625microm and l2=00075microm respectively The results are plotted from -15V to

15V anode voltage as shown in Figure 410 In this case a slight change in I-V

characteristic in the forward bias can be seen clearly which means the I-V

characteristic depends on the length of the shorter undoped layer Therefore the l2 layer

also acts as another voltage arm due to the asymmetrical length The effect is quite

similar to the spacer l1 but this times the forward current only slight changes The reason

for the small change in current is that the length change is small and it linearly affect the

states distribution

-000002

-0000018

-0000016

-0000014

-0000012

-000001

-0000008

-0000006

-0000004

-0000002

0

0

000005

00001

000015

00002

000025

00003

000035

01 015 02 025 03

Rev

erse

Cu

rren

t (A

)

Forw

ard

Cu

rren

t (A

)

Spacer Lenght (microm)

Forward Current (1V)Reverse Current(-1V)

139

Figure 410 IV characteristic of the dependencies current to Spacer 1 layer

Fixing the current at -1V and 1V the current versus layer thickness relationship is

illustrated in Figure 411

Figure 411Current change with layer thickness l2

The I-V curve depends on the length of the shorter undoped spacer layer quite linearly

The forward current changes in increase to the layer compared to the backward current

The layer thickness l2 should be small as long as it prevents carrier diffusion Therefore

all these three layers must be kept within limit to ensure that the high performance of the

ASPAT diode can be fully utilised

-00002

0

00002

00004

00006

00008

0001

-2 -1 0 1 2

Cu

rren

t (A

)

Voltage (V)

L=75n

L=25n

L=625n

L=5n

L=375n

-2E-07

-18E-07

-16E-07

-14E-07

-12E-07

-1E-07

-8E-08

-6E-08

-4E-08

-2E-08

0

0

00002

00004

00006

00008

0001

00012

0002 0004 0006 0008

Rev

erse

Cu

rren

t (A

)

Forw

ard

Cu

rren

t (A

)

Spacer Lenght (microm)

Forward Current (1V)

Reverse Current(-1V)

140

46 Temperature Dependent Simulation

This section will discuss in detail how changes in operating temperatures influence

the IV characteristics of the ASPAT diode The same codes as in the previous

simulation with fitted results are used for this temperature dependence study but a few

parameters were changed for different temperatures

Theoretically the material parameters that are influenced by the change of

temperature are band gap electron effective mass the density of state (NC NV) light

hole mass heavy hole mass permittivity and electron amp hole mobilities However not

all mentioned parameters will have a large impact on the IV characteristic in the

SILVACO Atlas simulation The most significant factors that give appreciable impact on

the DC output current-voltage were the energy bandgap and the effective mass The

GaAs bandgap as a function of temperature is given by the equation below [106]

119864119892 = 1198641198920 minus

120572 1198791198712

120573 + 119879119871

(44)

Here Eg is the bandgap Eg0 denotes the bandgap at 0K TL is the Temperature α=

Constant (Varshni Parameter) AlAs6e-4 GaAs5405e-4 β= Constant (Varshni

Parameter) AlAs408 and GaAs204 The calculated parameters that are used in this

simulation are shown in Table 42

Table 42 The calculated values of bandgap at different temperatures

Temperature (K) GaAs Eg(eV) AlAs Eg(eV)

77 1506 2903

100 1500 2899

125 149 2893

150 1486 2887

175 1478 2879

200 1470 2871

225 1461 2863

250 1452 2854

275 1443 2844

300 1424 2835

325 1419 2824

350 1414 2814

375 1404 2803

398 1394 2793

141

The effective mass of the materials used in this simulation can be expressed by

119898119899 = 1198980119899 + 11989810 (

119879119871

300119870)

119898119901 = 1198980119901 + 1198981119901 (119879119871

300119870) + 1198982119901 (

119879119871

300119870)2

(45)

Where mn is the effective electron mass mp represents the effective hole mass m1n and

m1p are constant number for the basic GaAs material m0n

for 119898119883119898119871 119886119903119890 119892119894119907119890119899 119887119910 119905ℎ119890 119890119902119906119886119905119894119900119899 (1198981198991199052 lowast 119898119899119897)

13 while m0p is based on the

expression 1198980119901 = (11989811990111989732

+ 119898119901ℎ32

)23 The calculated parameters are shown in Table

43

Table 43 The calculated effective masses for each temperature used in this simulation

Temperatures (K) GaAs Effective Mass (kg) AlAs Effective Mass (kg)

77 00660m0 03790 m0

100 00658 m0 03788 m0

125 00655 m0 03785 m0

150 00652 m0 03782 m0

175 00649 m0 03779 m0

200 00646 m0 03776 m0

225 00643 m0 03773 m0

250 0064 m0 0377 m0

275 00637 m0 03767 m0

300 00634 m0 03764 m0

325 00631 m0 03761 m0

350 00628 m0 03758 m0

375 00625 m0 03755 m0

398 00622 m0 03752 m0

142

Figure 412 Measurement and simulation comparison result as a function of temperature range

from 100K to 398K

Figure 412 above shows excellent agreement between simulation and

measurement results at various temperatures The IV characteristics correspond to a

device of size 100times100microm2 as presented in Chapter 3

47 Temperature Dependence characteristics of GaAsAlAs ASPAT Diodes

As mentioned earlier in chapter one the tunnelling diode has many advantages

over conventional Schottky barrier diodes some of which are a large dynamic range

low power consumption and very weak temperature dependence This section will

discuss the effect of variable temperature applied to the GaAsAlAs ASPAT diode and a

similarly processed TiAu Schottky diode Two samples were fabricated together for

these studies (XMBE304 and XMBE104 representing an ASPAT and a SBD

respectively) The fabrication technique is exactly the same as has been discussed in

Chapter 3 In order to make it fair for direct comparisons as well as easy probing both

diodes were fabricated with the same emitter size (100times100microm2) The DC measurements

at different temperature were carried out using a Lakeshore Cryogenic probe station over

the range of 77K to 398K in 25K step interval

-002

-001

0

001

002

003

004

005

006

-2 -15 -1 -05 0 05 1 15 2

Cu

rren

t I

(Am

p)

Voltage V (Volt)

T=100K_Simu

T=100K_Meas

T=398K_Simu

T=398K_Meas

T=200K_Simu

T=200K_Meas

T=300K_Simu

T=300K_Meas

143

471 GaAsAlAs ASPAT diode vs TiAu SBD

Once fabrication and measurement were completed both DC outputs of the diodes

were characterised and analysed Figure 413 shows a semi-logarithmic plot for

measured current versus voltage as a function of temperature for ASPAT sample

XMBE304 In forwards bias the current changes for different temperatures from 77K

to 398K are less than 5 percent This confirms the very weak temperature dependence of

current transport as it is dominated by tunnelling through the barrier On the other hand

the backward bias shows the current changes at different temperature are slightly bigger

than in forward this due to band bending occurring faster (making the effective barrier

lower) and allowing thermionic emission to significantly contribute to transport of the

current The only other study of temperature dependence for the ASPAT was made by et

el RT Syme[15] but details were not stated in their report

Figure 413 Log Current vs voltage as a function of temperature for ASPAT sample XMBE304

The effective barrier height for the GaAsAlAs ASPAT diode is higher than that of

the SBD (See Figure 414) therefore there is an expectation of more limited thermionic

current flow in the ASPAT than the SBD As mentioned earlier the conventional

Schottky Barrier diode that is used in this study consists of a Gold Titanium and GaAs

(AuTiGaAs) interface which is the baseline for the temperature dependence study

0000001

000001

00001

0001

001

-15 -1 -05 0 05 1 15

Log

Cu

rren

t (A

)

Voltage (V)

T=77K

T=100K

T=125K

T=150K

T=175K

T=200K

T=225K

T=250K

T=275K

T=300K

T=325K

T=350K

T=375K

T=398K

144

Figure 414 The effective barrier high of the Schottky barrier diode and the ASPAT Diode [3]

The SBD epitaxial layers profile is as shown in Table 44 below Theoretically the SBD

obey thermionic emission transport[44] and its I-V characteristic is given by

119868 = 1198680[exp (

119902119881

119899119896119879) minus 1]

(46)

Where q is the electron charge V is the applied voltage across the diode n denotes the

diode ideality factor k is the Boltzmann Constant T is the absolute temperature in

Kelvin and I0 is the diode saturation current which is given by the expression

1198680 = 119860119860lowast1198792 exp (minus

119902empty1198870

119899119896119879) exp (minus120572120594119890

12120575)

(47)

here A is the area of the diode A denotes the effective Richardson constant Oslashb0 is the

barrier height at zero bias δ represents the thickness of interfacial insulator layer χ

denotes the mean tunnelling barrier and α = radic(4120587

ℎ)(2119898lowast) is a constant value The

ideality factor n is taken from the slope of the SBD current-voltage characteristic and in

this study its value varies from 1 to 2 (depending on temperature) In the case of the

ASPAT diode thermionic emission can also happen if a thicker barrier is used (~ 100Aring

or thicker) as shown by CS Kyono et el [104] who concluded that when a thicker

barrier of AlAs barrier is used the current transport is dominated by thermionic emission

145

Table 44 Epitaxial layer profile for Schottky Barrier Diode (XMBE104)

Layer Material Doping(cm-3

) Thickness(Aring) Bandgap (eV)

Schottky GaAs(Si) 500E+15 7500 14

Semiconductor GaAs(Si) 300E+16 7500 14

Semiconductor GaAs(Si) 100E+17 7500 14

Ohmic GaAs(Si) 500E+17 7500 14

Buffer GaAs(Si) 300E+18 7500 14

Substrate GaAs(Si) N+ 3000 14

The fabricated SBD was also measured and its I-V characteristic is plotted as a

function of temperature in Figure 415 Unlike the ASPAT diode the current at forward

bias for the SBD change enormously with temperature from 77K to 398K and at all

biases For the ASPAT diode the slight change in current only started after 08V bias as

the current starts to have some component of thermionic emission over barrier

Figure 415 Log Current vs voltage as a function of temperature for SBD sample XMBE104

In order to see clearly how much the current is changing in forward bias for both

ASPAT and SBDs diode a log current at different voltages versus 1000temperature is

plotted as shown in Figure 416

1E-08

00000001

0000001

000001

00001

0001

001

01

-2 -1 0 1 2

Log

Cu

rren

t (A

)

Voltage (V)

T=398K

T=375K

T=350K

T=325K

T=300K

T=275K

T=250K

T=225K

T=200K

T=175K

T=150K

T=125K

T=100K

T=77K

146

Figure 416 Influence of temperature on IV characteristic for both ASPAT diode and SBD

Semi-logarithmic plots of current (at V= 05 06 07 08V) versus inverse

temperature for both SBD and ASPAT are shown in Figure 416 When the temperature

is increased the current also increases in the SBD as a result of thermionic emission

over the barrier for sample XMBE104 This is in contrast to the temperature-

independent tunnelling through the thin AlAs barrier of sample XMBE304 ASPAT

diode where when the temperature is increased the current is almost constant

At low and high temperatures the ASPAT shows excellent temperature

independence with a constant current flow It exhibits a tunnelling current in excess of

values expected by the elastic tunnelling current calculation equation suggested by RT

Syme [16 18] above (Eq 1) using a Oslash value of 105eV (ΓGaAs to Γ AlAs tunnelling) By

contrast for the SBD at low temperature (77K-275K) the changes of currents were very

high and for every 02V there is an exponential change of more than 40 This

temperature dependent study was also reported in[68]

147

48 Conclusions

This chapter demonstrated the establishment of an excellent physical model and

comparison of room temperature I-V characteristics of GaAsAlAs ASPAT diodes for

different emitter sizes their scalability as well as an investigation of their characteristics

at different temperatures from 77K to 398K Simulation are validated on well-

characterized experimental data and excellent fitting which had been achieved in this

work permit the designer to extract all related parameters of heterojunctionmultilayer

ASPAT structures thus creating modification for future growth specification in order to

achieved precise designs

It is clear that the work which had been carried out in this chapter is able to

achieve with adequate accuracy a claim of reverse engineering capability The ability of

the GaAsAlAs ASPAT to act as a zero-bias detector has been analysed and compared

with the SBD It is clear that the temperature stability which is shown by the

GaAsAlAs ASPAT is much better than that of the SBD thus demonstrating that the

tunnelling current is dominant over the thermionic emission in ASPAT diodes

148

5 DC amp RF CHARACTERISTIC OF ASPAT DIODES

51 Introduction

To assist in circuit designs for any type of high-frequency circuits such as

millimetre wave detectors frequency multipliers and mixer circuits which are built

based on non-linear devices (diodes) an equivalent-circuit model for the diode is

required This is among the simplest and most effective method for analysing

semiconductor devices which work at high frequency where the electrical characteristics

measured obtained from the devices are extracted and presented in a circuit consisting of

lumped elements components (resistor inductor capacitor etc) However accurate DC

and RF measurement data is essential to extract the equivalent-circuit elements quickly

and correctly The extracted parameters values from the circuit that are taken into

account usually depends on bias and frequency associated with the device physically

which is also interrelated to the semiconductor material parameters device structure as

well as fabrication process flows

In this work the DC and RF data were derived from DC and S-parameter

measurements respectively These measurements were carried out both in-house and at

the University of Cambridge by a collaborator partner (Prof MJ Kelly) The I-V

characteristics of the diode obtained from DC measurements were measured from -2V to

2V while the S-parameters were carried out over a wide frequency range from 40MHz to

40GHz with eight different biases In this chapter the DC measurements for various

sizes of the diodes with analysis of their IV characteristics will be discussed The one-

port on-wafer ASPAT measurement setup as well as the de-embedding method will

also be explained Thereafter the equivalent circuit models with all lumped element

effect will be discussed This work is carried out with the help of the VNA which

principle has been described in Chapter 2 and Keysight ADS simulation tool All

technical details regarding the equivalent circuit models will be explained together with

the method used for the ASPAT diode evaluation The equivalent circuit model also will

cover the diode intrinsic elements such as Cj Rj and Rs and extrinsic elements ie CP

149

and RP Finally an equivalent circuit model with the small signal characterization of the

fabricated ASPAT diodes will be presented

52 General Analysis of Current-Voltage (I-V) characteristics of GaAsAlAs

ASPAT diodes

The recent development of state-of-art for DC measurement apparatus has led to

capabilities for high-level accuracy of measuring voltages to a few nano Volts and

current signals in the femto Amp range[107] This can easily be obtained by exploitation

of proper connections and high-quality cables connecting the equipment to the Device

under Test (DUT) In this work the DC measurements were carried out using an Agilent

B1500A Parameter analyser whose description was covered in Chapter 2

As was discussed in Chapter 3 the GaAsAlAs ASPAT diodes have been

fabricated with different mesa areas between 2times2microm2 to 100times100microm

2 but the smallest

size obtained with good I-Vs was 4times4microm2 In this section the focus will be on how the

extracted data can be expanded further for empirical modelling Figure 51 shows typical

results for measured ASPAT diodes with various dimensions to check for their

uniformity According to our standard procedure the DC measurement has to be

conducted prior to the RF to ensure the diode is in fully working order as this will later

save a lot of time during RF characterization

Figure 51 IV Characteristics of measured ASPAT diodes for emitter sizes of 4x4um2 6x6um

2 and

10x10um2 Note the good scalability

1E-10

1E-09

1E-08

00000001

0000001

000001

00001

0001-2 -1 0 1 2

Cu

rre

nt

De

nsi

ti (

Amicro

m2)

Voltage (V)

4x4microm^2

6x6microm^2

10x10microm^2

150

Figure 51 above demonstrates the IV Characteristics of measured GaAsAlAs

ASPAT diodes (XMBE304) for emitter sizes of 4times4um2 6times6um

2 and 10times10um

2 This

sample was processed using the dielectric bridge technique developed in this work It

can be observed that current per unit area for each dimension fits and scales to each

other The scalability of each diodes measurement is very important to ensure no process

related issues are hampering the devicersquos proper operations This also confirms that the

diodes are completely functional and can be used for the next stage of measurements

The advantage of having an excellent scalability of those diode sizes is that a prediction

of smaller emitter size can be made

This type of IV characteristic shows asymmetric behaviour which results from

the unequal spacer lengths of the device This behaviour is very useful for detection

application as it obeys a square law model The square law predicts that the current is

proportional to the square of the applied bias

119868 = 1198861198812 119908ℎ119890119899 119881 gt 0

119868 = 0 119908ℎ119890119899 119881 lt 0

(51)

To extract the first order effects of ASPAT diodes DC measurements which

result in asymmetric I-V characteristics are analysed The slope of the non-linear region

is used to determine the junction resistance (Rj) which is obtained from the first

derivative of voltage versus current (dVdI) The expression of Rj is given by

119877119895 =

120597119881

120597119868

(52)

In order to understand the relationship between Rj and diode sizes of the ASPAT the IV

characteristic for each diode displayed in Figure 51 is used to extract the Rj This has

been done by using the expression in equation (52) above and their response is plotted

against bias as displayed in Figure 52

151

Figure 52 Junction resistance versus voltage

As can be seen in the Figure 52 above the Rj for each device decreases strongly

when the voltage increases At zero bias the 4times4um2 devices show the highest Rj value

followed by 6times6um2 and 10times10um

2 devices The junction resistance at zero bias obtained

from the 4times4um2 diode is around 86KΩ while reducing by a third for the 6times6um

2 and

10times10um2 diodes with Rj of 27KΩ and 10KΩ respectively A diode with a smaller

forward current under the same applied voltage will exhibit a larger Rj For a good

millimetre wave detector a device with a large value of Rj is desirable since it will

provide high detection sensitivity

The slope at the IV characteristic contributes to an important parameter that is

commonly used by electronic manufacturers to describe diode specification namely the

video impedance (RV) which is also known as the non-linear resistance The RV which is

extracted from the real part of the diode small signal impedance is highly dependent on

the DC bias current and only weakly depends on the series resistance of the diode (RS)

Therefore the video impedance is given by

119877119881 = 119877119895 + 119877119878 (53)

Where RS is the series resistance of the diode whose value is normally very small and

does not contribute much to the whole slope and hence RV is dominated by Rj The RV

changes in behaviour if any DC current is flowing through the diode Practically small

DC current in the range of 1 to 10 microAmp or total zero bias is used to maintain the

appropriate RV value (1-2KΩ to several MΩ) RV will also determine the voltage

-10

10

30

50

70

90

110

130

150

-01 0 01 02 03

Rj(

)

Voltage (V)

6x6um^2

10x10um^2

4x4um^2

152

sensitivity of the whole detector circuit This will be explained further in the next

chapter In the case of a detector with an amplifier RV of the diode acts as the RF

impedance which needs to be matched with the video amplifier ( impedance looking into

the diode from the amplifier)[108 109]

The quotient of the second order derivative to the first derivative

((d2IdV

2)dIdV)) when calculated from the whole I-V characteristic translates directly

into a curvature coefficient (k) This is the most commonly used figure-of-merit to

quantify diode nonlinearity at zero bias Figure 53 below shows the variation of k with

bias and more importantly the zero bias rectifying action for device sizes of 4times4 um2

6times6 um2 10times10 um

2 This parameter which represents the small-signal rectifying

action of the diode will affect the performance of the detector (voltage sensitivity)

Detailed discussions on how this parameter effect the detector performance will also be

discussed in the next chapter

Figure 53 The rectifying action of the diode at zero voltage which gives the ASPAT high sensitivity

near zero bias detection

Figure 53 above shows calculated curvature coefficient of the measured I-V

characteristics from the same diodes shown in Figure 51 The highest k value is

obtained from the diodes with size of 4times4um2 followed by 6times6um

2 then 10times10um

2

The curvature coefficient decreases sharply as the bias increases for each diode This can

be attributed to a significantly increasing number of electrons that tunnel through the

thin barrier which were accumulated in the 2DEG formed in the intrinsic spacer region

-5

0

5

10

15

20

25

30

-001 004 009

Cu

rvat

ure

Co

effi

cien

t(V

-1)

Voltage (V)

k(10x10 um^2)

k(6x6 um^2)

k(4x4 um^2)

153

An ASPAT diode with a smaller size will have a larger Rj with a corresponding smaller

current under the same bias condition and hence will demonstrate a larger k value In

this calculation the curvature coefficient at zero bias obtained from 4times4um2 6times6um

2

and 10times10um2 diode is 23V

-1 17V

-1 and 16V

-1 respectively

A summary of the ASPAT diodes parameters obtained from measured I-V

characteristics that have been translated into first and second order differentials are

gathered in Table 51 below and compared to other diodes in the literature

Table 51 Performance of the ASPAT diode obtained from measured IV characteristics in this work

Sample Rj(Ω) k(1volt)

ASPAT 10times10 microm2 10K 16

ASPAT 6times6 microm2 27K 17

ASPAT 4times4 microm2 86K 23

Ge Backward diode 182K[110] 159[110]

InGaAs Backward diode 154[110] 23[110]

Sb Backward diode 5K[111] 47[111]

Si-Backward diode 135K[112] 31[112]

PDB 15K[8] -

AlGaAs SBD 20-100K[113] 34-38[113]

GaN HBD - 16[114]

From Table 51 above it is clearly that the ASPAT diode has a comparable value of Rj

and k to existing detector diode in the research community and in the commercial

market Based on literature of each diodes stated in the table the key to obtaining a high

value of k at zero bias is to minimize any forward tunnelling current Furthermore the

largest ASPAT diode used here (10times10microm2) has very close performances to that of a

commercial diode ie discrete Ge backward diode (ref[110]) where both Rj and k value

are close to each other

53 RF Test Fixture Theory and Experiment

RF measurements differ from DC measurement as they are more complicated

and it is necessary to comprehend the basic measurement principles to achieve

meaningful data This is obligatory especially for on-wafer RF characterization and

154

analysis to attain precise results Most of the electronics component measurements

which have input and output for instance antenna amplifier cables etc are based on a

two-port network configuration The characteristics which can be extracted from these

components are usually used to define their impact on a more complicated system

The performance of the two-port network can be described by a few parameters

ie scattering (S-Parameter) admittance (Y-Parameter) Impedance (Z-parameter) and

Hybrid (H-Parameter) However the S-parameter approach is favoured for high-

frequency measurements as it is relatively easier to characterize the microwave

performance and is able to convert to other parameters when necessary The advantage

of S-parameters is that they can straightforwardlydirectly convert into other two-port

parameters as mentioned above in term of currents and voltages[115] In fact to obtain

the device capacitance the appropriate S-parameters needs to be transformed into Y-

parameters using specific equations Furthermore the devicersquos cut-off frequency can

also be obtained when S-parameter measurements are performed over a wide frequency

range

531 On-Wafer Measurement and Small Signal One-Port Characterizations

In this work the arrangement of the RF measurement setup is assumed to be a

linear system as small voltage amplitude signals are used this means that the signals

have only a linear effect on the network without any gain compression or attenuation

The assumption is still acceptable even though the typical ASPAT is characterised as

non-linear in nature because it is a passive device which will act linearly at any input

power level

Generally the S-Parameter measurements on a diode can be adequately and

suitably performed using a one port measurement The technique used to characterise the

output is similar to the two ports technique but only incident and reflected waves are

used to characterise the input and output ports of the device Essentially this is because

the ASPAT has only two terminals and it is a passive device like other diodes

Therefore the analysis will revolve around the S11 parameter Figure 54 below show the

S11 is a ratio of reflected wave to the incident wave

155

11987811 =

119877119890119891119897119890119888119905119890119889

119868119899119888119894119889119890119899119905=

1198871

1198861 119908ℎ119890119899 1198862 = 0

(54)

A VNA as described in Chapter 2 is used to measure the ASPAT diodes This

powerful equipment is able to measure S-parameters up to 40 GHz To conduct accurate

S-parameter measurement at the diode the measurement setup must be calibrated prior

to the actual measurements taking place

54 Device Calibration

541 Open and Short De-Embedding Technique

Further calibration to be made involves anything related or attached to the

device The co-planar waveguide (CPW) bond pad and interconnect line that are

attached to the intrinsic diode are the main contributors of the errors also are required to

be calibrated In general the bonds pad could generate a capacitance (parasitic) in

parallel with the intrinsic diode and its contribution depends on the size of the bonds pad

as well as the operating frequency Meanwhile the CPW and interconnect line may

cause a parasitic inductance in series with the diode

The method that is used to get rid of this parasitic is called de-embedding and the

most common technique to realise it is by introducing OPEN-SHORT structure[116]

This method is based on a lumped-elements model Parasitic elements of the diode

De-Embedding Structure

Incident wave

Reflected wave

One-port device

a1

b1

Figure 54 One port S-parameter measurements

156

equivalent circuit correlate directly to the access section of the CPW hence can be

derived from de-embedding structures The aim of the de-embedding technique is to

represent these parasitic elements so that the one-port characteristic of the actual diode

can be determined

The two types of de-embedding structure OPEN and SHORT are conventional

techniques that are widely used in this study The design of all structure must be

identical (in size) to the device to avoid any discrepancy It is very simple to design all

these three structures for example open structures are obtained by eliminating the diode

layout and keeping the bond pad layer only The short structure just adds a bridge and

ensures ground and signal pad are connected to each other Through structures are

realised by disconnecting both ground pad and leaving the signal pad to connect to each

other

To gain more accuracy this external effect must be removed by the implementation of

de-embedding structures on the same tile as the actual device Figure 55 shows the

fabricated de-embedding structures used in this study

Figure 55 Fabricated open short de-embedding structure and real device (zoomed) use for RF

calibration and measurements (Note Images are not to scale)

In summary the de-embedding which is used to extract out the parasitic elements

from entire single diode measurement is a very important step as normally on-wafer

measurement requires coplanar waveguide (CPW) to access the diode structure (active

region) The CPW will have some effects which will disturb the accuracy of the device

characteristics

157

55 S-Parameter Measurement Result and Analysis

This section will only present RF measurement results after all VNA setup and

calibration were performed The S-Parameter measurements were carried out on ASPAT

diodes at five different DC biases from -2 to 05 volt with a sweep frequency from

40MHz to 40GHz using a calibrated VNA and the input power was fixed at -30dBm

The measurement procedure as described in Chapter 2 was performed on the device

(on-wafer) equipped with the appropriate bond pads This is important to ensure the

results obtained are valid The reason for using different biases is to find at what voltage

the device capacitance is fully depleted This is also very important in determining the

cut- off frequency of the devices

In this research two phases of the experiment on the S-parameter measurements

were carried out The first phase is to qualify the process flow ie for manufacturability

and repeatability which can be obtained from the consistency of the result The S-

parameter measurements taken on the same wafer dies are repeated several times on

different GaAsAlAs ASPAT diodes There are three different tiles taken from 3

different wafers namely XMBE304A XMBE304B and XMBE304C carried out in

this experiment The repeatability tests are done mostly on the large devices (15times15microm2

up to 100times100 microm2) and the results are analysed based on the reflection coefficient (S11)

on Real and Imaginary measurements

In the second phase the measurement is toward producing devices that can

perform at high-frequencies This can be realised by utilising small emitter size devices

(4times4microm2 6times6 microm

2 and 10times10 microm

2) The measurement results of these devices will be

used to build the equivalent circuit models while both the intrinsic as well as the

parasitics of the device will be evaluated Hence all the values obtained from these S-

parameter measurements will be used to design the device that can be used in

millimetres-wave applications As can be seen in Figure 56 below the extracted S-

parameter measurement results comprise of a reflection coefficient (S11) for real

imaginary and Smith chart for XMBE304A While these measurement results are

extracted at zero bias voltage the other bias voltages will be used to extract the

capacitance This will be described in the final section of this chapter

158

551 Diode to diode uniformity

In order to study within tile uniformity and reproducibility statistics of the RF

performance five devices of different mesa sizes in the same tile (XMBE304A) were

measured at zero bias and represented in term of Real and Imaginary reflection

coefficient (S11) the uniformity check is carried out at three different frequencies step

under 15GHz since the cut-off frequency for these big devices is relatively low at about

~20GHz on average The variation of the reflection coefficient is taken from the

percentage of the (standard deviationmean values) for all five diodes from this run The

following figures show Real and Imaginary S11 of large mesa area ASPAT diode from

15times15microm2 up to 100times100microm

2 which are represented by lines graph in a few different

colours

Figure 56 Real S11 RF measurements for (XMBE304A) were done for four devices from

15times15microm2 to 100times100microm2 device sizes for within-wafer uniformity check

Figure 57 Imaginary S11 measurements (XMBE304A) to qualify the repeatability and

reproducibility were done on four devices from 15times15 microm2 to 100times100 microm2 at zero bias

159

The variance data extracted from the graph (Real S11) above for each device within-

wafer (device to device) uniformity study is summarize in Table 52 below

Table 52 Device to device uniformity check for large ASPAT diode

Device Size 100times100 microm2 50times50 microm

2 30times30 microm

2 20times20 microm

2 15times15 microm

2

Variation 5GHz 181 115 405 151 145

Variation 10GHz 106 133 522 424 293

Variation 15GHz 119 198 281 76 509

The majority of diodes show that the variations of S11 measurements are below

3 and only a few are below 8 These finding still can be considered as good for

manufacturing control since absolute I-V characteristics reported in [63] is set by

designerrsquos specification to be not more than plusmn10 variation Further extensive RF

measurements were carried out by the research collaboration with the University of

Cambridge on the same GaAsAlAs ASPAT diodes wafer [117] In their study they

focused on 50times50 microm2

mesa size 17 of diodes were chosen to be measured The study of

uniformity of RF characteristic only focused on frequencies below 20GHz The same

approach was used to get the variation of the reflection coefficient for all 17 diodes but

this work was carried at four different frequencies Table 53 shows the zero bias S11

result for four different frequencies and standard deviation of the devices

Table 53 The variation of reflection coefficient (S11) for GaAsAlAs ASPAT diode at four different

frequencies[117]

Frequency (GHz) 5 10 15 20

Variation () 197 243 26 276

From the results the variations of 50times50 microm2 mesa sizes measured in-house and

at the University of Cambridge are comparable with all variations showing good

uniformity ie recording variations below 3 This indicates that the RF performance

of the GaAs AlAs diode is valid and reproducible and is thus considered as a good

achievement for manufacturing Once the reproducibility and repeatability of the large

devices showed stable results the fabrication process then continued to obtain smaller

emitter size for work at high-frequencies

160

552 Wafer to wafer uniformity

Other RF measurements were conducted on sample XMBE304B which was

fabricated in-house using the same process steps but the only difference from

XMBE304A was the use of SiN3 as dielectric In this run three different mesa sizes

were measured (15times15mmicro2 20times20microm

2 and 30times30 microm

2) and Real and Imaginary S11

plotted against frequency The RF performances of both samples are gathered in one

graph as shown in Figure 58 below

Figure 58 Wafer to wafer uniformity test on 1st and 2

nd RF measurement(XMBE304B) to qualify

the repeatability and reproducibility were done for 30times30microm2 to 15times15 microm

2 device sizes (Real and

Imaginary) Note blue colour is XMBE304A and red colour is XMBE304B

For this wafer to wafer uniformity study four diodes with three different sizes as

specified previously were measured from sample XMBE304B and four diodes from

previous measurements of XMBE304A The blue line in Figure 58 represent

measurement result of real and imaginary for sample XMBE304A while the red line

161

represents XMBE304B The uniformity data is compared at three different frequencies

and gathered in the table below

Table 54 Wafer to wafer uniformity check on XMBE304A and XMBE304B

Device Size 30times30 microm2 20times20 microm

2 15times15 microm

2

Wafer A vs wafer B variation 5GHz 305 31 1 314

Wafer A vs wafer B Variation 10GHz 352 344 329

Wafer A vs wafer B Variation 15GHz 321 376 359

As can be seen in the Table 54 above the wafer to wafer uniformity is rather

large (30) on average The main reasons being that sample XMBE304A was

processed by utilizing S1805 as a dielectric layer while sample XMBE304B used

Si4N3 Although the process steps are similar for both wafer processing the use of

different dielectric layer will influence the diode parameters especially resistance and

capacitance as the dielectric constant for each materials is different Secondly the wafer

processing is not run concurrently at the same time thus the moisture and temperature in

the clean room might differ for both processing Although the wafer to wafer uniformity

test for this run might not be favourable for manufacturing tolerance at least the use of

different dielectric layer shows some significant result in term of capacitance resistance

effect to the GaAsAlAs ASPAT diode

553 Small devices RF measurements

The first objective of this study was to make smaller size mesa devices ie

1times2microm2 1times3 microm

2 2times2 microm

2 and 3times3 microm

2 However for GaAsAlAs ASPAT type this is

difficult to achieve in practise These issues were discussed in detail in Chapter 3

Hence the smallest emitter size that yields repeatable and reproducible results was

4times4um2 The final measurement which was done on sample XMBE304C focused on

small devices The measurements were done on four devices two with the diode bond

pads sitting on substrate (GaAs SI) and the other two sitting on dielectric layer (Si4N3)

Figure 59 below shows three measured results obtained from sample XMBE304C

using the 3rd

Gen Mask

162

Figure 59 RF measurement performed on device sizes of 10times10microm2 6times6mmicro

2 and 4times4microm

2 (Real and

Imaginary) Note that green red and blue colour represents 4times4microm2 6times6mmicro

2 and 10times10microm

2 diodes

respectively

Figure 510 RF measurement performed on device sizes of 10times10 microm2 6times6 microm2 and 4times4 microm2

(Smith Chart) Note that green red and blue colour represents 4times4microm2 6times6mmicro2 and 10times10microm2

diodes respectively

30MHz

40GHz

163

As can be seen in Figure 59 (Real Imaginary) and Figure 510 (Smith Chart) are

obtained from measurement of four diodes in the same tile The diode to diode

uniformity that is sitting on the same platform obtained at 35GHz frequency in this run

on average is ~15 25 and 1 for 4times4microm2 6times6mmicro

2 and 10times10microm

2 diodes

respectively On the other hand the uniformity between diode to diode sitting on

dielectric and substrate is quite high due to different capacitance value of devices on

average ~7 are attained from three different sizes of diode

From the measurement results above the Real S11 measurement of four different

sizes show the same trend for each frequency At low frequency resistances for each

diode is high as the S11 value is large At intermediate frequency the values drop

tremendously for big devices (30times30 mmicro2 and 20times20 mmicro

2) ie in Figure 57 At high

frequency all diode reach saturation limit as the value are constant Small devices

(4times4mmicro2 6times6mmicro

2 and 10times10microm

2) ie Figure 59 show S11 values that are higher than

those of large devices as smaller emitter diode have larger resistance value

The imaginary S11 value also shows the same trend as for big devices However

for small devices in this run (4times4mmicro2 6times6mmicro

2 and 10times10microm

2) the S11 value keep

dropping toward negative values at increasing frequencies This indicates that bigger

devices with positive value at high frequency are more capacitive than the smaller

devices It is worth mentioned that the capacitance and inductance values for device

sizes of 4times4microm2and 6times6microm

2 come from the CPW layouts and these are dominant while

for device sizes of 15times15 mmicro2 and above the device capacitance itself is dominant

The Smith Chart shows the reflection coefficient (S11) as a function of the

applied frequency (30MHz to 40GHz) All measurements from each mesa size follow

unique impedance circle which is that most of the lines are at the lower right outer ring

This means that the diode capacitance value is frequency dependent For the case of

10times10microm2

devices these impedance circles are mostly toward the outer ring meaning a

higher capacitance than the other two mesa dimensions All the device constantly follow

the outer ring without crossing any real axis at any frequency point meaning that the Cj

is not shorted at the maximum 40GHz measurement frequency (not reached cut-off

frequency) Therefore the entire small GaAsAlAs diodes in this run have capability to

work in the millimetre wave frequencies range

164

56 Extracting RF models of ASPAT at Zero Bias Voltage

The methodology used in the S-Parameters measurement for high-frequency

analysis must ensure that the derivation of the equivalent circuit corresponds to their port

characteristics In other words the component representing the ASPAT in the equivalent

circuit model must have physical significance otherwise the circuit will be meaningless

The fabricated ASPAT diodes as discussed in Chapter 3 have the cross section shown in

Figure 511

There are two main components that can be extracted from the fabricated

ASPAT depicted above ie intrinsic and parasitic The intrinsic refers to the main

structure of the diode itself and are represented by three bias dependent elements

namely Junction Resistance (Rj) Junction Capacitance (Cj) and diode Series Resistance

(Rs) The parasitic is the elements related to the bond pad of the anode and cathode as

well as interconnects They are represented by parasitic inductance (LP) resistance (RP)

and capacitance (CP)

The diode parameter extraction is different from the three terminal devices

(FETs) in the sense that FETs are a kind of direct extraction in which all the elements in

the transistor have linear functions to the port characteristics ie S-parameter Y-

Parameter Z-Parameter and can easily be solved by the matrix calculation method for

those particular parameters[118] The same extraction method cannot be applied to the

diodes because its elements will embroil with each other Therefore only one method is

used to extract the diode element which is optimisation by tuning the initial value toward

the measured S-Parameter values

Figure 511 Cross-sectional of fabricated ASPAT diode and the corresponding equivalent circuit

model

165

The strategy used to model the ASPAT is based on an initial fitting value of the

lumped elements to the extracted value from measurement on three S-parameter graphs

(real imaginary and Smith chart) for the reflection coefficient (S11) The refinement is

accomplished by optimisation and fine tuning of the values which result in minimum

error between extracted and modelled values Figure 517 (on page 153) shows fitted S-

parameter result for extracted and model numbers with each one fitted in a single line as

an example However to achieve this excellent fitting key prior steps have to be used

de-embedding fitting the intrinsic value and optimisation

561 Extraction of ASPAT parasitic element

Once the S-parameter measurements achieve stability repeatability and

reproducibility for each measurement in term of S11 results as mentioned above the

results of the de-embedding structure which had been measured prior to the device

structure are then extracted to form a well-defined equivalent circuit In order to build

and analyse the equivalent circuit firstly the measured data is imported into the ADS

software prior to any fitting This can be realised via the ldquoStart The Data File Toolrdquo

features provided by this particular software When successfully imported the data is

read by the function S2PMDIF (These files are a natural extension of two-port S-

parameter Touchstone files) as depicted in Figure 512 below

Figure 512 The S-parameter Touchstone file is used to read the measured files

166

For the open and short techniques after de-embedding the equivalent circuit

model which is represented by mainly a capacitor and an inductor is built The open

structure requires resistance and the capacitance values of 20KΩ and 26fF respectively

connected in parallel to be well fitted to the real imaginary and Smith chart (S11) output

On the other hand for the short structure the Real imaginary and Smith chart (S11) have

to satisfy the values of resistance and inductance elements of 1Ω and ~47pH respectively

connected in series These values strongly rely on the bond pad or CPW dimension and

length The Equivalent circuit models and fitted data as well as measurement can be

seen in Figure 513 and Figure 514 below

Figure 513 Equivalent Circuit Model of Open CPW or de-embedded structure

Figure 514 Equivalent circuit model for short de-embedded structure

To satisfy the equivalent circuit a self-consistence method introduced by

Ren[119] is utilised This approach accurately extracts the CPW capacitance (Cpad) and

inductance (Lpad) as well as intrinsic Junction capacitor (Cj) which is attained from the

one-port S-parameter measurements Therefore the pad capacitance introduced by the

self-consistence method for the open structure can be expressed by

167

119862119875 =

119868119898(11988411119874119901119890119899)

120596

(55)

Lpad which represents the short structure is given by

119871119875 =

1

120596(119868119898 (11988411119874119901119890119899minus 11988411119878ℎ119900119903119905

))

(56)

Here Y is the admittance parameter (Y-parameter) converted from the S-parameter

measurement data and ω is the angular frequency The extracted measurement data

represented in the equivalent circuit fits with the simulated data in three S11 graphs as

can be seen in Figure 515 below From the Smith Chart it can be clearly seen that both

open and short S11 results are on the circumference which means the resistance of the

short structure is very small while in the open it is very large Additionally the

calculated Cpad and Lpad using Equations 55 and 56 above produce results similar to

those obtained in the equivalent circuit model for the open and short structure The

values are ~25fF and ~45pH respectively These data completely verify and validate

both results

Figure 515 Smith chart representative S-parameter measurement for short (left) and open (right)

CPW The blue lines represent simulated data and the red is measured data

Short

Open

168

562 Extraction of ASPAT intrinsic elements

Once the parasitic elements are determined it is easy to build a complete ASPAT

equivalent circuit The ASPAT is not like other tunnelling diode which their equivalent

circuit models widely studied ie RTD [120] IMPATT and PDB The only literature

which reports ASPAT equivalent circuits can be found in [15] and other RT Symersquos

journal paper[16] Fortunately its equivalent circuit model is not much different

compared to other diode video detectors Thus other literature which is based on

Schottky diode equivalent circuit model used for detector application can be referred to

The simplest form of ASPAT equivalent circuit and other video detectors intrinsically

consist of junction capacitance (Cj) series resistance (RS) and junction resistance (Rj)

First and foremost to extract the equivalent circuit one must know the theory behind

each parameter that is developedbuilt as a spine to become a complete element This is

vital to ensure the equivalent circuit is correct In the case of the ASPAT Cj is predicted

from a simple fully depleted parallel plate capacitor approximation which was discussed

previously in Eq (230)[15] Additionally for the S-parameter measurements the Cj can

also be validated by the self-consistence method mentioned earlier and thus can be

expressed by

119862119895 =

[

(1

120596)

1

1

119868119898 (11988411119905119900119905119886119897minus 11988411119874119901119890119899

)+

1

119868119898 (11988411119874119901119890119899minus 11988411119878ℎ119900119903119905

)]

(57)

This approach helps to verify both the fully depleted parallel plate capacitor in S-

parameter measurements The basic component which is responsible for the ASPAT

series resistance RS was discussed in detail in Chapter 2 RS and Cj are key contributors

to the high-frequency operation as expressed by the device cut-off frequency Equation

(58) below

119891119888119906119905minus119900119891119891 =

1

2120587119877119904119862119895

(58)

The R and C parameters must be kept as low as possible in order to obtain high cut-off

frequencies for millimetre wave applications From the fabrication point of view Cj can

169

be reduced by making as small a diode emitter size as possible while for RS reducing

the D gap is of paramount importance as it dominates the series resistance The ASPAT

contact resistance in the electrodes (contacts between metal and semiconductor) can be

reduced by using high doping in the ohmic layers

The junction resistance (Rj) of the ASPAT is taken from the 1st derivative or

slope of the current-voltage characteristics Normally the value of Rj is very large

(several kilo Ω) compared to Rs The small signals ASPAT equivalent circuit built with

intrinsic and extrinsic components is shown in Figure 516 below while the fitting

results is shown in Figure 517 and Figure 518

Figure 517 Excellent fitting between measured (red colour) and simulated (blue colour) for S11

(Real and Imaginary) results for various small device designs

Rj

Cj

Cpad Rpad

Rs Lpad

Figure 516 Equivalent circuit of the ASPAT diode

170

Figure 518 Excellent fitting between measured (red colour) and simulated (blue colour) for S11

results (Smith Chart) for various small device designs

The equivalent circuit that was built for the ASPAT is taken from sample

XMBE304C with emitter dimensions of 4times4 microm2 6times6 microm

2 and 10times10 microm

2 These

devices are expected to work in the millimetre-wave region and have cut-off frequencies

(intrinsic) of ~650GHz ~200GHz and ~100GHz respectively

Table 55 Comparison between calculated (fully Depleted) and extracted (different biases) values

from equivalent circuit parameters for different ASPAT mesa sizes at zero bias voltage

Parameters 4times4microm2 6times6 microm

2 10times10 microm

2

Calculated

(fully

depleted)

Extracted Calculated

(fully

depleted)

Extracted Calculated

(fully

depleted)

Extracted

0V -05V -1V 0V -05V -1V 0V -05V -1V

Cj(fF) 879 23 7 6 198 55 148 139 549 171 486 46

Cpad(fF) - 15 15 15 - 15 152 152 - 15 15 15

Lpad(pH) - 45 43 42 - 50 473 473 - 51 51 46

Rj(KΩ) - 90 833 522 - 35 392 392 - 12 125 13

Rs(Ω) 99 11 11 11 67 95 8 7 41 95 45 37

fcut-off

Cj(GHz)

1828 629 2066 241

1

1208 192 1344 163

5

710 107 728 935

fcut-off

Cj+Cp(GHz)

- 380 658 688 - 151 663 781 - 98 556 705

171

The focus in this study is purposely to build ASPATs as zero bias detectors that are able

to work in the millimetre and sub-millimetre frequency range therefore all the

parameters which are obtained from equivalent circuit were extracted at zero bias

voltage Theoretically the calculations which are derived from both self-consistence amp

theory can only be solved for fully depleted device capacitance (using a parallel plate

configuration) Hence both extracted and calculated results are compiled in Table 55

Noticeably the calculation can only produce the intrinsic parameters of the

ASPAT for fully depleted capacitance On the other hand both intrinsic and extrinsic

parameters of GaAsAlAs ASPAT are obtainable from extraction and thus help to

determine at what bias the diode is start to deplete The junction and series resistance (Rj

and RS) of each dimensions shown in Table 55 above were achieved by fitting the

elements of the equivalent circuit with the three measured S11 graphs whereas the Cpad

and Lpad were extracted by utilising the self-consistent method from the S-parameter

measurements which is fitting the de-embedding structure Additionally the Cj values

are obtained via fitting the measured S11 data and employing the self-consistence

approach Results obtained from both techniques are identical

At zero bias all extracted junction capacitance from each device sizes are very

different from the calculated one while the extracted series resistance are closer to the

calculation This means that Cj is a highly voltage dependent parameter and Rs is

voltage independent but solely dependent on device structure and material used to

fabricate it The extraction at -05V and -1V shows that the values of junction resistance

is changing for most of the devices which means this parameter also rely on bias voltage

as discussed earlier in Section 52

The cut-off frequency for each calculated devices are near the THz range even

for the 100microm2 emitter area However with the introduction of parasitics elements ie

pad capacitances fcut-off is degraded tremendously Therefore it is important to make sure

all the intrinsic elements have optimum values so that the target operating frequency of

the ASPAT diode can be met Due to this it is advisable to operate the devices at no

more than 13 of fcut-off when designing detector systems

The parameters extraction at -05 and -1V also show that Cj values are closer the

calculated ones which means the ASPAT diode is reaching full depletion

172

563 Capacitances -Voltage (C-V) Extraction

Theoretically the junction capacitance of the ASPAT is calculated from the fully

depleted formula 119862119895 = 휀119900휀119903119860119889 which was also discussed in Equation (230) in Chapter

2 in page 43 Its value depends on the change of voltages to depletion at the emitter

contact[15] and make it one of the voltage dependent parameter for the diode[121]

Therefore the C-V characteristic of the GaAsAlAs must be precisely extracted

In practice the capacitance is difficult to measure due to the very low resistance at zero-

bias However alternatively it can be measured and extracted by applying different

voltage and identifying the point at which there is change which essentially represents

full-depletion Apart from these it can also be extracted from S-parameters measurement

which is then converted to Y-parameters A C-V characteristic of the GaAsAlAs

ASPAT from XMBE304C is extracted and plotted as depicted in Figure 518 below

Figure 519 Capacitance extraction showing full depletion at -025 volts (modelled capacitance vs

Voltage)

From the graph shown in Figure 518 the capacitance is extracted at eight

different biases for 4times4 microm2 6times6 microm

2 and 10times10 microm

2 The devicersquos junction

capacitance for each dimension increases and reaches a maximum value at 025V There

are additional quantum capacitance effect which comes from an increase in the negative

charges in the 2DEG region (when band bending happens creating an accumulation

0

50

100

150

200

250

-2 -15 -1 -05 0 05

Cap

acit

ance

(fF

)

Voltage (V)

Cj(4x4microm^2)

Cj(6x6microm^2)

Cj(10x10microm^2)

173

region at the barrier) This charge is imaged by the positive charge in the whole

depletion region Increasing the voltage toward positive values leads to a lowering of

the AlAs barrier and thus allowing thermionic emission to take place after certain bias

values leaving only the depletion capacitance and making the quantum capacitance

negligibly small

In the reverse bias case the device junction capacitance reaches a saturation

(fully depleted capacitance) at a voltage of -025V and remain constant up to -2V If the

reverse bias voltage is increased further the ASPAT may reach breakdown Therefore it

is important to know how far the diode can withstand applied reverse bias to ensure it

can still give full performance

57 Conclusions

In this chapter scalable DC characteristics of GaAsAlAs ASPAT diode derived

from three different emitter sizes of 4times4 microm2 6times6 microm

2 and 10times10 microm

2 was

demonstrated The current density obtained at zero bias is several microAmicrom2 These allow

1st order differential effect to exhibit high value of Junction resistance (Rj) at zero bias

However Rj is highly bias dependent The 2nd

order differential effect on IV

characteristics display a high value of curvature coefficient leading to high voltage

sensitivity when applied in millimeter wave detector applications These two parameters

are vital in the design of millimeter wave detectors and especially those operating at zero

bias

Subsequently RF measurement up to 40GHz of uniformity study for both within

wafer and wafer to wafer variance were undertaken An average uniformity below 7

was obtained for within wafer study on large device area ( 15times15 microm2 to 100times100 microm

2)

while for small device area ( 4times4 microm2 to 10times10 microm

2) a smaller 3 uniformity variance

was achieved in average However for wafer to wafer study the variant uniformity was

quite high at around 30 on average for relatively large device (15x15 microm2 to 30x30

microm2) This was mainly attributed to different dielectric layers used in the process flows

of the sample rather than fundamental MBE control of the AlAs barrier thickness

174

It was demonstrated in this chapter that careful on-wafer RF measurements of small

size GaAsAlAs ASPAT diodes allow accurate device parameter extraction of both

extrinsic and intrinsic parameters The extrinsic parameters are namely pad capacitance

and inductance with obtained values of 26fF and 47pH respectively These values were

obtained from de-embedding structure fabricated on the same tile as the real devices

The intrinsic parameters such as junction capacitor junction resistor and series

resistance had different values according to device dimensions The smallest zero bias

value of Cj obtained from 4times4 microm2 diodes was 23fF ensuring a high cut-off frequency of

380GHz and hence in the next chapter a 100GHz detector will be presented working at

slightly less than 13 of this cut off frequency

The C-V data extraction confirmed that the fully depleted capacitance started to

happen at around -025V The maximum junction resistance occurs at +025V largely

caused by the depletion region and an additional quantum capacitance effect CQ This

effect is strongly related to the size of a 2DEG which occurs under forward bias (01V to

025V) and can be reduced by having a smaller thickness AlAs barrier

175

6 MILLIMETRE WAVE GAASALAS ASPAT DETECTOR

DESIGN USING ADS

61 Introduction

The ASPAT diode having features of non-linear IV characteristics at zero voltage

make it useful for signals rectification ie detector and mixer for millimetre-wave

applications Additionally ASPAT diodes have a range of advantages such as large

dynamic range strong temperature insensitivity etc[15]over other rectifier diode This

makes ASPAT an appropriate candidate in RF detection applications Since 1940 the

only two terminals device that has been the workhorse for RF applications is the

Schottky Barrier Diode (SBD)[9] In its earliest form the SBD was built based on a

point-contact device which could not perform at high frequencies It was then developed

to work at higher frequencies by exploitation of epitaxial structures [10] and to date the

SBD remains the mainstay of two terminal devices that are able to work in the

millimetre and submillimetre-wave regions However as discussed in Chapter 4 the

performance of SBD is degraded at extremes of temperatures and these circuits

employing SBDs require temperature compensating circuitry Thus there is additional

complexity associated with technologies and applications related to SBDs

Before this work was carried out no model for the ASPAT diode as detector had

been available or developed especially using the empirical modelling ADS software

When the ASPAT diode was first introduced its function was conceptually explained it

was then built and tested to compare with other microwave detectors at X-band

frequency (95GHz) The comparisons were made in terms of detector parameters ie

sensitivity dynamic range temperature dependence etc[15] These early works lead by

RT Syme et al supplied the basic knowledge to model the ASPAT diode as a zero-bias

detector for mm-wave frequency gt100GHz For the SBD many models and equivalent

circuit approaches have been reported [122 123] The modelling of conventional SBD is

mostly implemented through fitting the S-parameter curves of the model to the

experimental one This approach is also carried out in this research since it is accurate to

predict the performance of the device under test [124]

176

This Chapter aims to introduce low cost reliable and sophisticated detector design

based on ASPAT diodes which is believed to be able to improvereplace SBD in

millimetre-wave applications especially in imaging The focus was on developing and

establishing an appropriate circuit design that suit the new ASPAT diode for such

applications The detector sensitivity as its key parameter ultimately limits the quality

and acquisition time of the detector In the subsequence section the theory of detection

including both direct and heterodyne will be discussed This is followed by definition of

detector characteristics of interest as well as noise consideration Section 65 present the

main focus of this chapter which is the development of 100GHz ASPAT detectors and

their result will be explained in term of all detector characteristic of interest

62 Detection Theory

Any incoming signal such as RF microwave or mm-wave in the form of envelope

function or single wave can be detected by rectification of the signal using a nonlinear

device ie transistor or diode The input and output signal (RF) signals are normally in

the form of amplitude as a function of time Typically the detector output is a low-

frequency signal known as the video signal which has amplitudes that are proportional to

the square of the input RF signalrsquos voltage amplitude

A complete receiver system as shown in Figure 61 below consists of receiver

antenna and a circuit designed to extract the signal and then amplify it The function of

the receiver system is the converse of the function of the transmitter side At the receiver

side the antenna is used to receive the signal it then conveys the signal to the extraction

circuit for detection as the information-bearing part of the signal (using the nonlinear

device as its heart) The signal is finally amplified to avoid any information strength

decay Additionally in a digital system the output signal which had been processed by

the detector circuit has to maintain an optimum input signal conveyed to in-phase and

quadrature (IQ) demodulators The output signal will go through a low pass filter then

to an analogue-to-digital converter (ADC) and thus a digital baseband output signals

will be produced

177

Figure 61 Block diagram represent a complete direct receiver system

There are two types of millimetre wave integrated circuit (MMIC) used for

detection purposes namely direct detectors and heterodyne detectors The direct detector

MMIC is the simplest circuit used for detector applications and has the simplest way of

extracting the RF information Due to its simplicity the direct detection method is

inexpensive and most attractive method used for measuring power in RF Laboratories

and Industries This detection scheme is also sometimes known as video detection[125]

The simplest way of explaining the detection process is that the incoming RF or

microwave signals depicted in Figure 62 below with an appropriate input power (Pin) is

rectified by using a diode and results in a corresponding output voltage (Vout) A detector

IC designed based on diodes is normally able to rectify very low levels of RF power (lt-

40dBm) then produces an output DC voltage that is proportional to the RF power A

rectifier diode can function at zero bias (which is very good for reducing noise) at very

small DC bias (003mA) and relatively high RF impedance which will produce around

600Ω This will affect the capacitance value and a low capacitance is needed to realise a

high detection sensitivity

Figure 62 The detection process of a single wave through a non-linear IV characteristic of a diode

Detector

Speakerdisplay unit Amplifier

Antenna

Vout

Pin

Tuner Amplifier

178

However this type of detector has a drawback which is itrsquos relatively low signal to

noise ratio Thus it will also rectify any incoming electrical noise at all frequencies and

up to the cut-off frequency (fC) The basics lumped components circuit as shown in

Figure 63 is used to build such detector which consists of Source impedance (Zo)

Rectifier diode (ASPAT or Schottky) wire or pad inductance and capacitance and Load

impedance (RL)

Figure 63 Lumped element illustration of microwave detector circuit

Another type of detector is the heterodyne method which mixes incoming RF or

microwave signal (fRF) with another constant signal produced by a secondary circuit

called the Local Oscillator (LO) The LO frequency (fLO) must be slightly higher than fRF

to enhance the RF signal This mixing between fRF and fLO happens in the nonlinear

device as depicted in the Figure 64 This will produce a signal at a different frequency

called the intermediate frequency (fIF) which can then be amplified and detected as

explained in the previous paragraph Theoretically a basic requirement of the mixer is to

have fIF as efficient as possible while practically the minimum conversion efficiency

obtained is around 20 The main reason for using a mixer is due to the fact that

selective amplifiers at RF frequencies are costly and hard to achieve Hence a mixer is a

good technique as it only convert the signal to a lower frequency in which good

selectivity and high gain can be more effortlessly realised[14] A good mixer diode is the

one that can produce a high cut-off frequency and reduce conversion losses (Lc) A

mixer and detector diode with a low driven input power result in reducing overall noise

figure and thus in the ideal case the fIF amplifier also should have a low noise figure for

better performance The advantage of the heterodyne method is that it has a higher

179

sensitivity compared to the direct detection method this is achieved by producing an fIF

which has a lower frequency than the incoming RF signal[14] Obviously a zero-bias

voltage diode is more favourable to be used in mixer and detector applications

Figure 64 The mixing process where the signals are processed by the non-linear I-V characteristic

to produce an intermediate frequency (fIF=fLO-fRF) when signals at fLO and fRF are applied to the

diode

To have good detection efficiency for both types of detectors the operating

frequency (fO) must be several times smaller than fC In the case of an incoming

maximum modulated signal fM the frequency that can be acquired is in the range of fO

plusmnfM and will normally come with noise The standard method that is used to reduce the

noise is using a filter of bandwidth about 2fM at the centre of fO with the condition that fO

must be smaller than fM (f0lefM) Otherwise it would be difficult to attain However in

most cases fM is smaller than fO and fO is smaller than fC thus this will make the video

impedance (RV) (or nonlinear impedance) very close to the differential resistance of the

diode (at fO) in the equivalent circuit[74]

63 Definition of Characteristics of Interest Detection Parameters in small signal

analysis

Theoretically a transfer function measurement is preferable prior to any empirical

modelling since no assumption can be made due to the detector non-linearity

Furthermore measuring voltage output at high frequency can be very low while

measuring the power incident on the detector is hard to achieve where the linearity of

180

the typical power meter is normally less than 3 over its operating range[126]

Therefore the modelling of detector output voltage vs input power (ie transfer

function) can help to determine both nonlinearity correction and appropriate operating

range for the detector itself

The performance of the diode that is often taken into account is the transfer function

(output voltage Vout versus incident power Pin) and the main parameters that is used to

characterise and determine the quality of any detector diode are the voltage sensitivity

(βV) tangential sensitivity (TSS) dynamic range (P1dB) ie under 1 dB roll off power and

variation of output voltage when examined in extreme temperature situation (ΔV(T))

The voltage sensitivity in small signal analysis can use the approach introduced by

Torrey and Whitmer [9] then βV can be expressed as

120573119881(119894119889119890119886119897) =

119877119895119877119871120581

2(119877119881 + 119877119871) (1 +119877119904

119877119895)

2

[1 +1205962119862119895

2119877119878119877119895

1 + 119877119904 119877119895frasl]

(61)

Where ω is the angular frequency (2πf) Cj is the junction capacitance of the diode

active region RL is the load resistance RV is the video impedance taken from the

expression of Rj + Rs and κ is the curvature efficiency that give small signal rectifying

action of the ASPAT diode which is given by the second order term and itrsquos expression

is

κ =

11988921198681198891198812frasl

119889119868119889119881frasl

(62)

The curvature coefficient or responsivity (κ) is translated directly from the non-linearity

of the IV characteristics of the ASPAT diode for detector application Both RV and κ are

the parameters that can be extracted directly from diode DC measurement as discussed

in Chapter 5

The voltage sensitivity is actually a quantitative relationship between input

power and detector response Meaning that it is a change in signal output over change in

input power Normally output power is measured in Volts and input signal is measured

in Watts Therefore the unit of responsivity is VW[127 128]

181

TSS is referred to the lowest or minimum signal that the detector could detect it is

determined by the diodersquos βV and total noise available in the system (from the diode and

any amplifiers in the detector circuit) For any diode with fO=10 GHz and low noise of

1MHz bandwidth amplifier the TSS is typically less than -55 dBm[74] The TSS equation

is given by

119879119878119878 =

radic[4119884119870119861119879119861(119865119886 + 119905 minus 1)]

119872

(63)

Where M is a figure of merits and is derived based on the expression M= 120573119881 radic119877119881frasl t

denotes the diode noise temperature Fa is the noise figure B T and Y are the amplifierrsquos

bandwidth temperature and power for signal-to-noise ratio respectively For a low-level

video detector ie lt10GHz the sensitivity mainly depends upon three factors firstly on

RF matching structure secondly on the rectification efficiency output impedance and

noise properties of the diode and finally the input impedance bandwidth and noise

properties of the video amplifier at the detector output The RF matching structure

controls the quantity of overall energy at the active junction for rectification The second

factor controls the reaction of the diode to incident microwave radiation and the last

factor will influence the detector sensitivity in general[109]

In practical the Tss is a direct measure of the signal-to-noise ratio of a detector and

is achieved by varying the amplitude of the input pulse (RF signal) until a point in which

the top of the noise level with no signal applied is at the same level of noise at the

bottom level of RF signal It is commonly measured on an oscilloscope as depicted in

Figure 65 below It is defined as the input power at which a signal to noise ratio of 251

is produced[109]

Figure 65 Measurement of Tangential Sensitivity[108 129]

182

The transfer function in many detector diodes is often divided into three sections

Firstly at low incident power secondly at higher input power and finally at very high

power static (continuous) In the first region the detector diode performs as a square-law

detector in which Vout is proportional to Pin This region normally is used to extract the

dynamic range of the diode detector In the second region Vout is approximately

proportional to Vin and this region is known as the linear regime Finally at higher Pin

still the transfer function or response rolls off and thus Vout ultimately become saturated

This roll-off point where Vout has dropped by 1dB below an extrapolation of the

dependence at low Pin is termed the ldquo1dB roll-off pointrdquo and this value is usually in the

range of -11 to 12 dBm[15] Therefore a dynamic range of the detector diode can be

obtained by taking the interval between TSS and 1dB roll-off point (in dBm)

Finally the temperature dependence of Vout for a detector is normally taken from

two extreme points of the temperature (-40C˚ to +80C˚) and thus can be determined

from

Δ119881(119879) = 10 11989711990011989210 |

119881(1198791)

119881(1198792)|

(64)

This Vout variation between -40C˚ and +80C˚ normally expressed in dB

64 Noise Consideration in a Detector diode

The existence of noise in a system limits the accuracy of device performance and

the precision of measurements In a detector system specifically using a diode the noise

which can reduce the sensitivity of signal encryption is called the Noise Equivalent

Power (NEP) By definition the NEP is a noise power density over the detection

sensitivity and it can be exploited to determine the overall noise performance of a

detector[130 131] In other words NEP is defined as the power from the input source

(Pin) that is required to supply a voltage output (Vout) equal to the root means square

noise at Vout [132] For an ideal lossless match and assuming only Johnson-Nyquist is

present the NEP of a zero-bias detector can be expressed as

119873119864119875119900119901119905 = radic4119896119879119861119877119881120573119900119901119905 (65)

183

Where βopt is responsivity with an optimum match which is given by1 2frasl 119877119895120581 This type

of noise appears when changing voltages across a diode and a noise voltage (Vn)

normally will arise Theoretically the NEP has units of Watts (as it is actually a power)

but it often normalized to 1Hz as it is independent of bandwidth and thus the unit

becomes WHz12

Additionally there are also several noise sources that contribute to Vn in a

semiconductor diode which are Johnson-Nyquist noise Flicker Noise and shot noise

Johnson-Nyquist noise [133 134] appears across any conductor or semiconductor at

thermal equilibrium this is due to the thermal agitation of the carriers or charges It can

be expressed in root mean square voltage as below

119881119869minus119873 = radic4119896119879119861119877119895

(66)

Where Rj is the differential intrinsic resistance B denotes the post-detection bandwidth

T is the device temperature and k is the Boltzmann constant[134]

The second noise that is taken into consideration when dealing with semiconductor

devices is Flicker noise more commonly known as 1f noise It is a group of known and

unknown noise sources that can be observed in the frequency spectrum and normally

display an opposite to the frequency power density curve[135] It comes from a variety

of different causes ie recombination effects at a defect in semiconductor mobility

fluctuation and flow of direct current as well as interface phenomena [136-138] In term

of voltage source Flicker noise can be expressed as[139]

1198811119891 = 119870119891119881119909119891119910 (67)

Where Kf denotes a device-specific constant V is the voltage and f is the frequency The

value of x and y typically used are 2 and -1 respectively This type of noise (1f noise)

will be neglected at frequencies high enough due to the fact that the NEP of the diode is

proportional to the thermal noise and resistance of the diode

Finally the noise that causes time-dependent fluctuation in a flow of electrical

current because of the carrier or electron charge crossing a potential barrier is called shot

noise Shot noise is due to the randomness in the diffusion and recombination of both

majority and minority carriers[140] The equation of shot noise term at random time is

given by[141]

184

119881119878ℎ119900119905 = 2119902119868119861 (68)

Where q is the electron charge I is the current and B is the bandwidth This type of noise

is not affected by changes in temperature or device parameters Therefore the total noise

voltage appearing in the semiconductor is found to be [141]

1198811198992 = 119881119869minus119873

2 + 11988111198912 + 119881119904ℎ119900119905

2 (69)

However a zero-bias device will greatly eliminate both shot and flicker noise

compared to a biased device This has been explained by Equation (66) and Equation

(67) above where both noises are significantly related to the current and voltage Thus

if V and I =0 in the nonappearance of incident power then Vn will also become zero For

a detection process with bias the diode will be self-biased by ΔV which causes both

flicker and shot noise to appear But the shot noise in practical situation is much smaller

and thus normally ignored [142-144] Usually the noise in a zero-bias detector is

estimated by considering the low power limit as good first order estimation in which the

presence of only Johnson-Nyquist noise and ΔV is arbitrarily low [11 145 146]

Additionally it has been reported that the noise in tunnelling type diodes displays very

low or no excess noise in the bias region of the current-voltage characteristic[147 148]

Therefore in general most of the noise specifically in tunnelling type of diode will be

neglected this is a great advantage compared to SBD or transistors

65 Modelling of a 100GHz Zero-biased ASPAT Detector

Once the DC and RF characteristics of the ASPAT diode had been accurately

obtained the next step is to model and design a detector circuit based on S-parameter

measurement results as was explained in the previous chapter The aim is to realise a

detector circuit design which can be operated at millimetre and sub-millimetre wave

regions from an accurate diode model prior to the circuit design A diode detector model

puts experimental observations into context and offers insight into future experiment

results Consequently an electrical model based on lumped element component is vital

for a deeper understanding of how and to what extent a new device like the ASPAT

diode can affect all the key detector parameters that were previously discussed The

prediction of the detector parameters mostly depends upon the ASPAT diode

185

geometrical emitter size and material parameters However for millimetre wave

operating frequency the accuracy of the model is more sensitive not only to diode size

but also to the diode periphery ie substrate as well as coplanar amp transmission line

adopted in the circuit Therefore both extracted intrinsic and parasitic element of such a

device must be taken into account

In this work an ASPAT diode with an emitter size of 4times4microm2 which is the

smallest size that could be fabricated so far was chosen to be exploited for detector

designs The important parameters related to the 4times4microm2 GaAsAlAs ASPAT which

works at 0V is summarize in table 61 below

Table 61 A summary of all the important parameters of the 4x4 microm2 diode

Device Rj(Ω) Rs(Ω) Cj(fF) Cp(fF) κ(V-1) Intrinsic_fcut-off(GHz)

4times4microm2 90K 11 21 15 23 629

The actual measured I-V characteristic is used to model the diode since the library

in the ADS simulation tool does not have an ASPAT diode model or any tunnelling

diode for that matter The procedure of realizing the diode model is by taking the I-V

characteristic obtained from the 4times4microm2 emitter size measurement results and

converting it into a10th

orders polynomial equation via MATLAB software to create a

virtual I-V characteristic Thereafter this equation is then defined as a two terminals

device namely Symbolically-defined Device (SDD1P) ie a component of the non-

linear equation provided by ADS (Figure 66) to represent the ASPAT Figure 67

shows the measured data and 10th

orders polynomial equation fit very well to each other

Hence this new component used to represent the whole ASPAT diode will be used in

this research for MMIC detector and Frequency Multiplier designs The device chosen to

be modelled (4times4microm2) has measured junction capacitance of 21fF (at 0V at 40GHz) The

detector circuit is designed to operate at 100GHz for a safe side due to the extrinsic

calculated fcut-off is around 380GHz Since there are a lot of advantages in using unbiased

detectors compared to biased one this work will discuss the performance of millimetre-

wave detector at zero-bias and their result will be compared to the current performance

of other diodes reported in the literature

186

Figure 66 The configuration circuit to verify 10th order polynomial equation extracted from

MATLAB to realize a virtual GaAsAlAs ASPAT diode

Figure 67 Verification of actual (blue measured) and virtual (red_10th order polynomial) I-V

characteristic of the 4times4 microm2 diode used in this study

To realize the ASPAT detector circuit a simple detector circuit topology as

depicted in Figure 68 was constructed Initial simulation was run to perform a

functionality check of the detector circuit utilizing the Harmonic Balance (HB)

simulation tool embedded in that particular software Such simulation tools will analyse

the detector performances in the frequency-domain as it is mostly beneficial and fully

compatible with microwave and millimetre wave problems The frequency domain is

also suitable for single and multi-tone power excitation The importance of harmonic

balance are described in [149]

-00005

0

00005

0001

00015

0002

00025

0003

-3 -2 -1 0 1 2 3

Cu

rre

nt

(A)

Voltage (V)

ADS

4x4

187

Figure 68 Direct detector circuit topology using an ASPAT diode

Initially the circuit topology that consists of P1_Tone power supply ASPAT

bypass capacitor and load resistance is simulated by setting up a fixed input frequency at

100GHz The ASPAT diode provides a DC output voltage proportional to the input

power strength depending on the absolute values of the DC terms associated with the

nonlinearity of the I-V characteristics The capacitance in the output part is a bypass

capacitor used to prevent millimetre-waves from leaking to the output The load

resistance is large enough to ensure the voltage divider between load impedance and

device impedance gives maximum voltage sensitivity This large load resistance is

achieved by creating an open circuit at the end of detector circuit terminals

Noted that this simulation was run using diode parameters that were extracted from on-

wafer one port S-Parameter measurements as described in chapter 5 To apply them in a

two port application ie detector circuit may or may not provide a very accurate

outcome it however worked adequately in the particular circuit described in this chapter

but may not work in other circuits in general Thus the one port extractions in this work

still provide adequate parameters to build and design specific MMIC detector circuits

but not in general applications

The main reason for these simulations and their results to be used in high frequency

applications is due to the fact that actual RF measurement were done up to 40GHz

Additionally the 100GHz operating frequency was obtained from extrapolation of each

ASPATrsquos component Since the on-wafer measurement that were carried out were

limited to one port characterization applying them to two port network applications may

188

have extra consequences which are unknown Therefore actual MMIC detectors are

needed to be built and test to validate this work

To find out what power the 4times4microm2 ASPAT diode can withstand the input power

is varied from -40dBm to 10dBm via control by the P1_tone As can be seen in Figure

69 the diode starts to saturate when the received input power is about -8 dBm Above

this power limit both output voltage and sensitivity drop dramatically

Figure 69 Output voltage and detector sensitivity over wide range of input power

This diode detector circuit can thus operate adequately at given input powers from -30

dBm to -8 dBm with a sensitivity of 950VW However for the best possible sensitivity

over a range of input frequencies only one optimized input power needs to be chosen

The parameters that directly influence the voltage sensitivity are the curvature

coefficients load resistance and video resistance as can be seen from Equation (61)

Therefore in the following simulation the values RL will be optimized according to the

diodes optimum input power with regards to the highest possible voltage sensitivity

Consequently five values of load resistance were chosen from few ohms to infinity

and with the same applied input power as depicted in Figure 610 For most load

resistors the sensitivity is constant at low input power and drop at the diode saturation

region (-8dBm and greater) However for an RL value of 100KΩ and below the loaded

voltage sensitivity shows a peak near 0dBm input power which corresponds to the

maximum slope of the ASPAT detector transfer function The highest voltage sensitivity

is obtained by using an Open circuit load impedance as shown in the graph The Open

circuit load impedance gives the highest voltage sensitivity due to the voltage divider

189

between source and load impedance therefore RL must be at least 5-10 times larger than

Rj to give a better sensitivity

Figure 610 Simulated dependence of Voltage Sensitivity on incident power and load resistance of

the ASPAT detector

Furthermore the value of voltage sensitivity (βV) depends on the junction

resistance of the diode and thus the large value of Rj of the ASPAT diode yields the high

βV observed Rj which was taken from the non-linear measured IV characteristic is a

voltage dependent parameter [142]and is inversely proportional to the forward bias

voltage as depicted in Figure 611 below

Figure 611 Junction resistance as a function of forward voltage

1

10

100

1000

-40 -30 -20 -10 0 10

Vo

lta

ge

Sen

siti

vit

y (

VW

)

Incident Power Pin (dBm)

Infinity

1KΩ

10KΩ

1MΩ

100Ω

100KΩ

0

20

40

60

80

100

0 001 002 003 004 005 006 007 008 009 01

Ju

nct

ion

Res

ista

nce

(k

Ω)

Bias Voltage (V)

190

In fact the expression of (120597119881120597119868frasl ) contribute to the video impedance expression via the

expression RV=Rj+RS (nonlinear resistance) As RS is very small compared to Rj thus it

was ignored when calculating RV Although the large value of Rj will increase the

voltage sensitivity it will also make matching difficult to achieve Therefore there will

be a compromise between the size of the matching circuit and voltage sensitivity to

attain the correct value of Rj Additionally a very high Rj (around ~1MΩ) will also

increase the detectorrsquos noise equivalent power (NEP in Eq(65)) Thus an average value

of RJ typically around 100kΩ is satisfactory[150] By having a large value of RJ one can

benefit from a low input power to drive the diode into the non-linear region and thus the

detector can work at very low RF input power The NEP for GaAsAlAs ASPAT diode

is then calculated based on parameters obtained from this simulation at room

temperature The values are compared to other diode detector available in the literature

as shown in Table 62 below

Table 62 Noise Equivalent Power (NEP) for zero bias detector diode

Device NEP (pWHz12) RJ(KΩ) Frequency(GHz)

ASPAT (4times4 microm2) 188 92 100

Tunnel Diode (08times08 microm2)[150] 370 26 220-330

Zero bias SBD[11] 15 3 150

Sb-Heterojunction Backward Diode[145] 024 32 94

VDI Zero bias SBD[151] 2 18 110

From the Table 62 above the NEP of the ASPAT is calculated based on RF input power

which is required to obtain an output signal-to-noise ratio of unity in a 1Hz at detector

output[142] and also the assumption of only Johnson-Nyquist (thermal noise) is

dominant for small incident power (-25dBm)[152 153] The prediction of NEP for the

ASPAT is comparable to the VDI Zero bias detector since the value of their junction

resistance is much lower than that of the ASPAT Therefore it is very important to

obtain a reasonable value of junction resistance From this it is clear that a trade-off of

high voltage sensitivity and low junction resistance is best to obtain low noise

191

Finally the curvature coefficient at specific operating voltages also influences

the voltage sensitivity of the detector Figure 612 shows the calculated curvature

coefficient of the 4times4microm2 ASPAT diode used in this work

Figure 612 Curvature coefficient for the measured I-V characteristic with a device size of 4times4μm2

The high zero-bias curvature (23V) is reached from the mutual effect of the intra-band

tunnelling in the GaAs-AlAs-GaAs and the highly doped GaAs at the anode and cathode

(Rs amp Rj) This was shown in the numerical simulation in[154] where the combined

effects of the optimum anode AlAs composition increases the curvature coefficients by

thinning the energy tunnelling window (intraband tunnelling process)

Other approaches that can lead to a large curvature is using smaller device area

with minimum series resistance [155] as was also discussed in Section 52 Having a

better curvature coefficient leads to increased voltage sensitivity as seen from Equation

(61) above and βV is also proportional to κ [111] The curvature coefficient calculated

using the above formula is nearly 23V at 1mV peak and high voltage sensitivity can be

achieved by having large value of curvature coefficient But one needs to remember that

this will also decrease with increasing input power because RJ which will also decrease

Therefore for a safe operating region the incident power that can be applied through the

diode is in between -30dBm to -8dBm for a 100GHz operating frequency

In this simulation -25dBm is chosen to simulate the GaAsAlAs ASPTAT diode

detector working at 100GHz input frequency By fixing the P1_Tone to this input power

-5

0

5

10

15

20

25

30

-001 001 003 005 007 009

Cu

rva

ture

Co

effi

cien

t(V

-1)

Voltage (V)

k(4x4 um^2)

192

the frequency is varied from 90GHz to 110GHz and the results are depicted in Figure

613 below

Figure 613 Voltage Sensitivity against frequency at -20dBm fixed input power

Noticeably the voltage sensitivity of the diode detector decreases linearly with

increasing input frequency At 100GHz a sensitivity of around 540VW is obtained

However the sensitivity in this case is roughly estimated from the Equation (61) above

and will not be sufficiently accurate because of the effect of other key factors such as

reflective power which was not included This parameter must be taken into the account

due the P_1Tone power source which provides a 50 Ω impedance source (Zin) which

does not match the load impedance (ZL) which consists of JωL 1JωC and Z and which

mainly comes from the ASPAT diode itself In order to determine the load impedance at

the diode a typical ohmrsquos law (Z=VI) equation must be used at the input side of the

diode with regards to the applied frequency (100GHz)

As a result the total load impedance obtained from the simulation is 11055-

j69057Ω which is clearly not matched to the 50Ω impedance source The mismatch

between source and load leads the available power from the source to be not fully

delivered to the load and hence there is loss of power leading to a lower detector

sensitivity Therefore actual calculation must take into the account the reflection impact

as expressed in the equation below

193

120573119881(119886119888119905119906119886119897) =

119877119869119877119871120581(1 minus |Γ|2)

2(119877119881 + 119877119871) (1 +119877119904

119877119895)2

[1 +1205962119862119895

2119877119878119877119895

1 + 119877119904 119877119895frasl]

(610)

Where the term (1-|Г|2) refers to the normalized power absorption by the ASPAT diode

and Г is the reflection coefficient due to discrepancy between 50Ω input impedance (Zin)

of the input port and the diode Consequently the calculation of reflection coefficient

(eq611) is carried out by using this expression and the result is shown in Figure 614

Γ =

119885119871 minus 119885119894119899

119885119871 + 119885119894119899

(611)

Figure 614 Reflection Coefficient versus operating frequency without matching circuitry

As can be seen in Figure 614 without the matching circuit the S11 is decrease

linearly with frequency but only very slowly This means that most of the RF power

transmitted from the source is reflected back (by that ratio) when it went through the

diode Therefore in order to resolve the mismatch a matching circuit is introduced in

between the source and load of the detector circuit as shown in Figure 615 (red

rectangular) This matching circuit works by transforming the load impedance into an

impedance that is identical to the source or input impedance Note that for any

impedance matching circuit the main purpose is usually to obtain maximum power

transfer to the load however in some cases (ie oscillators) the matching circuit is to

achieve a lower noise figure Hence in a broad sense the introduction of the matching

194

circuit in a detector circuit can be defined as a circuit that convert available impedance

into wanted impedance by obeying the maximum power transfer theorem [156]

Figure 615 Detector circuit with impedance matching circuit placed in between diode and source

There are many type of matching circuit that can be used to achieve both

objectives above such as circuits using lumped element transmission-line-impedance

matching circuit single and double-stub tuners as well as a quarter-wavelength In this

design the technique used to match source and load impedance is the single open and

short stub (in red rectangular) as it is simple convenient and very efficient in ADS

simulation The important parameter that needs to be tuned in both stubs is the electrical

length (E) at any designed frequency ie 100GHz in this case The E tuning is realized

by using the Smith Chart features available in the ADS simulation tools Figure 616

shows the reflection coefficient with matching circuit modelled over a broad frequency

band and it is clearly shown that at the desired operating frequency the reflection is very

low Note that it is difficult to obtain wide frequency band matching

Matching circuit

195

Figure 616 Reflection Coefficient over wide frequency band with matching

The simulation is continued to find the effect of the matching circuit placed in

the detector circuit on the voltage sensitivity The same input power (-25dBm) is applied

to the diode and the voltage sensitivity is plotted against frequency as depicted in Figure

617 Obviously at the desired operating frequency (100GHz) the sensitivity rises up to

a maximum value of 2100VW with this value obtained without any reflection and the

input port being completely matched with the ASPAT diode model used in this work

Figure 617 Simulated voltage sensitivity of GaAsAlAs ASPAT detector at W-band frequency

Once the matching circuit to be used for 100GHz operating frequency was confirmed

further simulations were made by applying a series of low input power to find the

tangential sensitivity (Tss) of the detector diode When determining the Tss it is very

important to include the matching circuit as it will minimize any power losses through

the diode thus a very small input power can be detected

196

Figure 618 Lowest detectable signal at 100GHz operating frequency

The transfer function depicted in Figure 618 shows incident power of -80 dBm

to -50 dBm applied and the lowest detectable signal that can be obtained with 4times4microm2

mesa size ASPAT diode is around 138microV at -68dBm Although a typical value of Tss is

normally not more than ~-55dBm as in ref [74] the lower value obtained in the

simulation is because the device is operated at zero bias operation and does not use any

amplifier therefore the noise and values related to amplifier as in Equation 63 have been

neglected Even though the TSS appeared very low it is most likely very dependant to

the 10th

order polynomial equation embedded into SDD in ADS software Therefore in

near future the TSS value has to be determined in real fabricated MMIC ASPAT detector

As discusses in Section 63 other important parameter that can be extracted from

the diode transfer function is the Dynamic Range of the diode It can be obtained in a

region called square law region which is a region where the Vout of the ASPAT diode is

proportional to the square law of the input power signal From Figure 619 the square

law region is in between -68dBm and -12dBm and the linear region or in this case

saturated region is above -12dBm Taking to the account the roll-off point where Vout

has dropped 1dB below the extrapolation of the dependence at low input power and

therefore the dynamic range of the detector diode can be obtained by taking the interval

between TSS and the 1dB roll-off point (in dBm) which is ~55dBm

197

Figure 619 The dynamic range of the ASPAT diode obtained from linear region of diode operation

The figure of Merit (M) of the detector (ie equation 63) is 2100 (90K) 12

where 2100 is the sensitivity and 90K is the value of RJ at zero bias which is equivalent

to 652 W The M value should be large however in this case due to RJ being very

large it has dropped tremendously when compared to the voltage sensitivity obtained in

this simulation The results obtained indicate a reasonably successful design of the

MMIC detector using the 4times4microm2

emitter size GaAsAlAs ASPAT diode The results

obtained lead to the design of other MMIC detector using the other fabricated diode

sizes (6times6microm2 and 10times10microm

2) A Similar procedure to the one used for the 4times4microm

2

diode was followed The only difference was the use a slight higher input power (-

20dBm) than in the 4times4microm2 design Hence the simulation results obtained are then

compiled and compared as depicted in Figure 620

Saturated

Region

198

Figure 620 Simulation result of diode sensitivity for three devices geometries obtained from the

fabricated ASPAT in this work

The graph plotted for each dimension was taken after matching circuits were included

As can be seen in Figure 620 the highest sensitivity is achieved using the smallest

device size as this has the highest cut off frequency Table 63 below summarises the

performances of the 100GHz ASPAT detectors obtained from the simulation in this

work

Table 63 Summary of characteristic of interest for GaAsAlAs ASPAT detector

Device Tss(dBm) fcut-

off(GHz)

Bv

(VW)

Dynamic range dB(dBm) M(W-12

)

4times4microm2 -68 380 2100 55 65

6times6microm2 -50 151 1445 48 45

10times10microm2 -40 98 247 40 21

From table 63 above it is clear that a lower cut-off frequency will affect the

voltage sensitivity The dynamic range between each ASPAT is different because larger

size area will allow more power to go through the diode as a result of high current that

such a device can handle before reaching the saturation their lowest detectable is higher

Small diode size will detect lowest voltage but cannot handle high power On the other

hand a large diode size is able to receive high power however can only offer lower

voltage output Therefore there is a trade-off between small diode size and receiving

input power which will directly affect Tss and 1dB roll off Once again all the

100

1000

10000

90 95 100 105 110

Vo

lta

ge

Sen

siti

vit

y (

VW

)

Frequency (GHz)

10times10microm^2

6times6microm^2

4times4microm^2

199

parameters obtained in this simulation are just estimation from the 10th

order polynomial

equation thus real ASPAT detector has to be fabricated for verification

The best device performance among all GaAsAlAs ASPAT diodes was obtained

with the 4times4microm2 mesa area size diode which was compared to other exiting millimetre

wave detector diode available in the literature Since the 100GHz is located in the W-

band spectrum frequency therefore the comparison will be performed in this frequency

band but with low input power The parameters for the-state-of-the-art zero bias

detectors are gathered in Table 64 below

Table 64 Comparison 4times4microm2 GaAsAlAs ASPAT diode to the-state-of-the-art zero bias detector

at W-band (75GHz-110GHz)

Device Size(microm2) Tss(dBm) βv (VW) Pin (dBm) f (GHz)

GaAsAlAs ASPAT 4times4 -68 2100 -25 100

GaAs SBD HSCH-

9161[157]

- -49 2200 94

HBD[158] 15x15 2540 -20 95

Planar SBD[159] - -68 2100 -25 100

Note that the ASPAT diode retains its favourable temperature stable characteristics

which are not the case for all the diodes used for comparison in Table 64

66 Conclusions

In this chapter all theory regarding RF detection using diodes ie parameters of

interest noise consideration etc have been discussed The aim to design and develop a

low cost reliable and sophisticated zero bias 100GHz detector circuit was achieved

through exploitation of a 4times4microm2 GaAsAlAs ASPAT diode The design was performed

with the aid of Keysight ADS modelling software utilizing harmonic balance simulation

The effect of load resistance junction resistor to the detector voltage sensitivity was also

discussed in details The 90KΩ Rj value and open circuit load resistance was chosen for

high sensitivity

200

A step by step design of a W-band ASPAT detector was presented The effect of

matching circuit was discussed in detail and where an unmatched sensitivity of 843VW

is obtained which then increases to 2100VW after matching Through RF

characterization simulation a detection at 100GHz (W-band) was successfully achieved

with a relatively large device mesa area (4times4microm2) at an input power of -25dBm

(8microWatt) leading to a 2100VW voltage sensitivity a -68dBm TSS and 55dBm dynamic

range All these values are comparable to others fabricated diodes in the literature

The zero bias ASPAT detectors based on the GaAsAlAs material system in this

work are still at an early stage of development a lot of work is still required to realize

high yielding integrated millimeter and sub-millimeter wave (MMIC) detector circuits

However as this work is on-going at Manchester it is expected that fabricated ASPAT

MMICs with even higher voltage sensitivity will be fabricated in the near future through

collaborating bodies involved in this research especially the University of Cambridge

and ICS Limited

201

7 FREQUENCY MULTIPLIERS (DOUBLER) DESIGN USING

GAASALAS ASPAT DIODES

71 Introduction

Originally the key application for the ASPAT diode was for use as microwave and

millimetre wave detectors[18] This is due to the fact that such diode demonstrates

strong non-linearity low noise and high cut-off frequency features as described in the

previous chapters However these features are not only beneficial for detection purposes

but also allow them to be used and designed as microwave and millimetre wave sources

The only way to generateenhance continuous wave (CW) power using a non-linear

device is through frequency multiplication techniques It is known that the frequency

multiplier is the alternative approach (to 3 terminal transistors) using non-linear devices

that are used to generate high frequency low phase noise signals Any high quality low

frequency signal that goes through a frequency multiplier circuit can be generated to any

desired high output frequency[160] Therefore the main objective of this chapter is to

demonstrate the feasibility of the ASPAT diode as a compact source of microwave and

millimeter-wave receiver for imaging applications[161]

The study of the ASPAT diode as a power source begins with a brief explanation

of the importance of a frequency source and the lack of compact device and technologies

at high-frequency signals The state-of-the-art for frequency multiplier will also be

discussed In the next section (Section 74) the fundamentals of the frequency multiplier

architecture ie the principle of operation and appropriate devices will be presented

Since this is the first attempt at using GaAsAlAs ASPAT diodes a simple multiplier

circuit design and topology was built This will be discussed in detail in the subsequent

sections where simulation results are discussed The focus of the discussion will be to

demonstrate the possibility of a GaAsAlAs ASPAT diode functioning as a frequency

multiplier and comparison with other state-of-the-art varistor mode frequency

multipliers

202

72 Motivation and Background

Typically continuous wave (CW) sources generating below 100 GHz can be

obtained through oscillators amplifiers and pin diode comb generator Below 10THz

the sources can be made from RTD IMPATT diodes and Gunn oscillators and above 10

THz it is commonly done by photonic mixing quantum cascade laser (QCL) and gas

lasers [162] Both types of sources and their performance are plotted in Figure 71

However these conventional ways of generating millimetre and sub-millimetre waves

have their own limitations ie high cost complexity and sometimes requirement for

cryonic cooling The most effective way to tackle the limitations of conventional mm-

wave and THz sources is by implementing frequency multiplication technique using

solid state nonlinear diodes [163-165] such as SBD and ASPAT diodes

Figure 71 performance of state-of the-art millimetre wave source [166]

Twenty years ago there were only two types of diodes (SBD and P-N junction

diodes) often used for frequency multiplication To date besides the SBD there are

many types of diode that have been used as frequency multipliers These include the

high electron mobility varactor (HEMV) single barrier varactor (SBV) [167] and hetero-

structure barrier varactor (HBV) (270GHz with 90mW input power and Conversion

Efficiency of 72)[168 169] Other variants that have developed to enhance the

frequency multiplier performance of the classic SBD [170] include the Barrier-intrinsic-

203

n+ (BIN)diode and Barrier N-layer N+ (BNN) diode [171] Other diodes for use in such

applications are the planar doped barrier diode (PDB) Resonant tunnelling diode (RTD)

amp it families ie Quantum well diode (QWD) and step recovery diode which is a

modification of the P-N junction diode

Although three terminal devices ie FET GaAs MESFET and HEMT had

shown better performance and are capable of achieving greater efficiency and

bandwidth as well as having additional conversion gain features [160] two terminal

devices (ie varactor diodes) which are passive multiplier are still preferred This is due

to their simplicity and most importantly their ability to generate very little noise Among

these types of device technologies the SBD is preferable as it is mature and has been

shown to be very suitable for high-frequency applications [172 173]

The ASPAT diode is exploited to investigate the possibility and the feasibility of

generating microwave and millimetre-wave power through well-known frequency

multiplication methods The utilisation of the ASPAT in frequency multiplication will

also aid in generating local oscillator sources which are critical components in

heterodyne receivers The ASPAT diode will work in resistive I-V mode (varistor mode)

and has features to work also at zero bias condition thus offering low power handling

than traditional high-efficiency varactor diode since the varactor diode requires a large

reverse bias supply of several tens of volts

73 Frequency Multiplier Architecture the Basics

In principle a frequency multiplier is an electronic circuit that gives an output

frequency that is a multiple integer of its input frequency signal pumped from a local

oscillator as depicted in Figure 72 The ability to generate any desired multiple output

signals is realised by a nonlinear device ie diode or transistor Such devices though

also can give distortion or cause sudden change to the input frequency Additionally

these devices generate multiples of the input frequency (fout) The distortion of the

sinusoidal signal refers to an abruptsudden change versus amplitude or time which thus

generates higher frequency with lower amplitudes of the input signal Usually a

frequency multiplier circuit will include a bandpass filter to select the desired harmonic

204

frequencies and deselect undesired harmonic frequencies especially fundamental

harmonic at the output for further processing

Any non-linear device either in symmetricalantisymmetric current-voltage or

capacitance-voltage can be utilised to realise a frequency multiplier source [168 174]

Figure 73 describes the method where a nonlinear resistance is utilised to convert a

harmonic input signal into periodic output signal containing components at multiples of

the input frequency Both non-linear resistance and reactance characteristics can be

extended into power series methods

Figure 73 Principle of operation for frequency multiplier utilising a non-linear resistance [10]

The operating principle of the frequency multiplier is shown in Figure 73 where

the I-V curve converts a harmonic frequency input into a periodic frequency output

including components at multiples of the input frequency The non-linear I-V

characteristic can be explained in term of a power series at the operating fixed point of

bias voltage (VB) [174]

Frequency

Multiplier Circuit

finput

foutput

= nfinput

Figure 72 Frequency multiplier has the property that foutput is an integer multiple of fin

205

119868(119881119861 + ∆119881) = 1198860 + 1198861∆119881 + 1198862∆1198812 + 1198863∆1198813 helliphellip (71)

For a given input voltage as below

∆119881 = 119881119904 cos120596119878119905 (72)

The input signal harmonics will become

119868(119905) = 1198680 + 1198681 cos120596119904119905 + 1198682 cos 2120596119904119905 + 1198683 cos 3120596119904119905 (73)

Where t and ω are the time and angular frequency respectively based on equation (73)

the output contains both signal source and harmonics Therefore a complete frequency

multiplier circuit has to have non-linear device and filter to allow the selection of any

frequency components needed

731 Types of frequency multipliers

Frequency multipliers can be classified into passive and active multipliers This

classification is based on the ability of the frequency multiplier to yield any conversion

gainlosses The passive multiplier is the one that only produces conversion losses In

other words it can be described as a multiplier that generates an output power level

lower than the excitation input power and it is mostly dominated by passive nonlinear

devices ie Diodes On the other hand the active multipliers refer to a device that would

produce an output signal with a power level that is greater than the input signal power

This conversion of power is termed as conversion gain These types of multipliers attract

much attention as they do not only increase the frequency at the output but also the

signal power

Passive frequency multiplier can be formed by using diodes that are classified as

being of the varistor (non-linear I-V) or varactor (non-linear C-V) type [160 174] The

varistor type will influence the frequency multiplication with a non-linear resistance or

conductance (resistive diodes) and this results in a very large potential bandwidth at the

output but poor conversion efficiency The varactor diode type where the frequency

multiplications are affected by the non-linear capacitance (reactive diode) as their

reactive element typical result is high conversion efficiency A diode that is used in this

206

application must have strong nonlinearity stable electrical characteristic repeatable and

has fast enough response to an applied frequency Therefore multipliers are classified

into Doubler Tripler quintuple and so forth depending on the highest power of output

harmonic signal

In general all varactor type diodes with such characteristics will produce high

power at odd-order harmonic oscillation if any microwave signal is pumped into them

The benefit of having odd-order in multiplier design is that it reduces the complexity of

the overall circuit ie it eliminates even-order idler frequencies [175] The varactor type

diode had been shown by Manley-Rowe to a get maximum 100 conversion efficiency

for generating an ideal harmonic [176] compared to the varistor type where the

maximum efficiency achievable ideally is 1 1198992frasl where n is the multiplication factor

(output harmonics number) [174] In the case of power handling (input excitation power)

for multipliers varactor mode diode required greater power (several milli Watt) than the

varistor mode due to the fact that reverse applied voltages are very large (many tens of

volts) Therefore these types of frequency multipliers may not be suitable for the case of

high input power excitation There no report in the literature of varistor based

multipliers working with high power excitations

74 Parameters of interest for Frequency Multipliers

The simplest way to describe an equivalent circuit for a complete frequency

multiplier is by setting a Source impedance (ZS) at the input side and load impedance

(ZL) at the output side as depicted in Figure 74 below This circuit usually has the same

properties as described in the previous chapter and most of the others two ports

networks However in this case the purpose is different and is the conversion of a sine

wave signal source (Vs) with angular frequency ωs to an output signal with frequency

nωs where n is the multiplication integer or the order of multiplication

207

Figure 74 A standard system for two port frequency multiplier circuit

Referring to Figure 74 above there are few sets of parameters for the frequency

multiplier to be taken out and compared Examples are the conversion loss maximum

input signal power Impedance at source and load Bandwidth multiplication factor or

harmonic amp subharmonic content and noise conversion properties The conversion loss

(CL) is described by the ratio of available power at source (Ps) to the output harmonic

power delivered to load resistance (PL) and is normally expressed in dB It occurs due to

the nature of passive semiconductor diodes and the electronic circuit itself that are lossy

and dissipate energy On the other hand the conversion efficiency (ηn) is a ratio of the

output power at load (PL) to the available power at the input (Ps) This is often expressed

in percentage () The conversion efficiency can be determined as

120578 =

119875119878

119875119871

(74)

while the conversion loss is expressed as[174]

119871119899[119889119861] = 10 log

119875119904

119875119871= 10 log (

|1198811199042|

4 119877119890 |119885119904||1198681198712|119877119890|119885119871|

) (75)

Where Vs is the input voltage and IL is the output current amplitude Besides this the

conversion efficiency is often referred to as the inverted value of the Ln In designing a

frequency multiplier it is crucial to minimise the conversion loss and maximise the

conversion efficiency value

To achieve a perfect multiplier with minimum conversion loss the impedance of

source and load must be at an optimum level This implies that the source impedance

Frequency

Multiplier Z

L

Zout

Z

in

V

Z

SWR Г

208

(Zs) must be very close to the complex conjugate of the multiplier input impedance

(Zin) hence minimum reflection loss will occur at the input side This can be realised by

introducing an impedance matching circuit between the diode and source The power

transfer between the source and the multiplier is quantitatively described by the value of

the multiplier input reflection coefficient (Ѓ) with source Zs assumed to represent a

reference impedance This specification also can be explained in the standing wave ratio

(SWR) Both relationships are described below respectively [174]

Γ =

119885119894119899 minus 119885119904lowast

119885119894119899 + 119885119904

(76)

119878119882119877 =

1 + |Γ|

1 minus |Γ|

(77)

Where the asterisk () represents the complex conjugate of the Zs impedance

On the other hand the situation of the load impedance is different when a standard

or an optimum value is provided by the designer This will either increase the conversion

loss or decrease the output power Thus one has to keep in mind that frequency

multipliers are non-linear devices and power transfer condition both at the input and the

output depend on each other and the input signal level[174]

75 20GHz40GHz Varistor Mode ASPAT Diode Frequency Doubler

In this work a similar ASPAT diode (4times4microm2) to that in designing millimetre

wave detector in the previous chapter is used The main objective of designing the

frequency multiplier circuit was first to investigate the performance of the ASPAT

diode as a microwave or millimetre wave signal source A design is deemed successful

when the diode physical parameters are optimised and the suitable impedance matching

network is produced for each desired harmonic as well as maximising the output power

These goals however are hard to achieve when a higher frequency operation is targeted

for use

There are many types of multiplier circuit topologies that can be implemented

using GaAsAlAs ASPAT diode in varistor mode to achieve high order of multiplication

209

Examples are single diode multiplier series or parallel connected diode multiplier anti-

parallel amp anti-series connection diode pair multiplier anti-parallel-series connected

diode multiplier and bridge frequency multiplier as well as nonlinear transmission line

frequency multiplier [174] Before designing a circuit there is one most important

consideration to make Prior to choosing any mentioned circuits to be used for frequency

doubler the design considerations are made based on the capability of ASPAT diode to

receive an optimum amount of input excitation RF power From the discussions in

Chapter 6 the ASPAT diode will reach saturation level (linear regime) at power ~

-10dBm for a device size of 4x4microm2 Once the optimum input power was confirmed the

circuit topology was carefully chosen to balance between the requirements of the

ASPAT to work at high frequencies ie low Rs and Cj amp high diode cut-off frequency

as well as the desired output signal frequency that needs to be produced

To realise the first attempt of an ASPAT diode as a signal source a simple circuit

topology of a frequency doubler was deployed as depicted in Figure 75 below The

frequency doubler circuit consists of a voltage source (can be power source) input

filtering with matching network ASPAT diode output filtering with matching network

and load impedance (ZL)

Figure 75 Block diagram of frequency doubler with series ASPAT mounted diode

In order to investigate the doubler performance the Keysight ADS simulation

tool and similar procedure to obtain accurate ASPAT model using a 10th

order

polynomial equation as in Section 65 was used The circuit in Figure 75 is translated

into ADS format as illustrated in Figure 76 Once the circuit was constructed the

analysis was performed using the Harmonics Balance (HB) simulator The circuit

requirements are matched terminations at the input and output frequencies open

Input Filtering

and matching network

Output Filtering

and matching network

Zs

Vs

ZL

Pin

fin

Pout

nfout

210

circuited terminations at the higher harmonics and optimum reactive terminations (an

inductance which resonates with the junction capacitance) at the output frequencies

Figure 76 Empirical modelling of ASPAT doubler utilising Keysight ADS tool

The circuit in Figure 76 is the simplest way of constructing a frequency doubler

circuit which consists of the signal source (P_1Tone) input matching circuit (Stubs)

filter (Short stub) ASPAT diode low pass filter output matching circuit(Stubs) and ZL

(load resistance) The utilisation of the stubs is an ideal case of simulation since in the

real fabrication stubs are normally formed in large sizes Therefore a proper design such

as using CPW instead of stubs is essential in real fabrication

Again this simulation works for this particular circuit in this chapter as all the diode

parameters were extracted from on-wafer one port S-Parameter measurement described

in chapter 5 To apply them in such two ports applications may not very accurate

however it still provides adequate parameters to build and design particular frequency

multipliers but not in general applications These simulations and results are adequate for

high frequency applications due to the fact that the actual RF measurements on the

diodes were carried out up to 40GHz and the target operating frequency in this multiplier

design does not exceed 40GHz

Since the on-wafer measurement that were carried out were limited to one port

characterization applying them to two port network applications may have extra

consequences which are unknown Therefore actual MMIC frequency multiplier is

needed to be built and test to validate this work

Input Matching

Output Matching

211

To find the optimum output power initial simulation without matching circuit

was performed This simulation was run by varying the input power from -35dBm to

20dBm but fixing the input frequency at 20GHz As can be seen in Figure 77 the lowest

point in the conversion loss (CL) and the highest point of the conversion efficiency (CE)

are obtained from an input power of -1dBm However this amount of input power is too

high for the ASPAT diode The lowest CL at -1dBm may not be accurate since it was

applied without matching Note that it is difficult to achieve a matching between source

impedance and load impedance when varying the input power Therefore a lower input

power of -10dBm is chosen for this frequency doubler operation Figure 77 shows the

Conversion Loss and Conversion Efficiency as a function of the available power of the

given input source

Figure 77 Conversion loss and conversion efficiency as a function of input power

The circuit in Figure 76 works with a -10dBm input power and 20GHz centre

frequency input signal is pumped from the power source (P1_tone) to the ASPAT diode

and distortionabrupt change of input waveform occurs at the fundamental frequency (f0

in this case 20GHz) Such abrupt change produces harmonics and these harmonics can

be classified into desired frequency component by placing two-quarter wavelength (λ4)

stubs (90˚) at both sides of the ASPAT diode At the input side of the diode short circuit

stubs are utilised to permit the f0 tone to reach the ASPAT diode and block the second

harmonics (2f0 in this case 40GHz) back to the input side and pushes it towards the

load resistance On the contrary at the output side of the diode the open circuit stubs are

used to ldquoopen circuitrdquo the 2f0 signals while ldquoshort circuitrdquo the f0 component Thus 2f0

212

signal will not be affected due to the open circuit stubs being half wave (λ2) long The

function of both stubs is basically to isolate the input and output signal from mixing each

other Therefore the design of input and output matching circuit can be achieved easily

The input matching circuit was designed based on the mentioned input frequency

(f0=20GHz) for an available input power (Ps=-10dBm) which is set up at the power

source by using two stubs with the same configuration as used in Chapter 6 Such

configuration is purposely deployed to increase the 50Ω coming from the P1_tone

source impedance to the conjugate thus reducing the reflection coefficient to the

ASPAT diode From the simulation without matching the input impedance to the diode

is 551Ω in magnitude for an available input power of -10dBm

On the output side of the diode output matching circuit is available to transform the 50Ω

port impedance in the optimum load impedance which provides minimum conversion

loss for the ASPAT diode The output matching circuit is designed based on expected

output frequency which is in this case 2f0 =40GHz Other than this optimum

impedance between load impedance and ASPAT will not be achieved thus resulting in

higher conversion loss

To ensure the proposed circuit is valid and suitable for the specific ASPAT diode

mesa size the response of conversion loss and efficiency are plotted as a function of

output frequency from 20GHz to 100GHz The results of both conversions are illustrated

in Figure 78

Figure 78 Conversion loss and Conversion efficiency as a function of output frequency

213

As can be seen in Figure 78 the conversion loss is obtained at the lowest point where

the output frequency is needed Meanwhile the conversion efficiency is maximum at the

same output frequency Therefore this indicates that the first attempt of an ASPAT

Doubler frequency source works well However the values obtained for Ln from this

simulation is 28dB which is rather high On the contrary the η achieved in this study is

very low with a value less than 1

Since the ASPAT is in varistor mode with no bias applied the conversion

efficiency is expected to be low due to resistive losses Another factor that may

contribute to lower η is the diode model itself as it is taken from a 10th

order polynomial

equation not from a diode model provided in the ADS software tools Thus some

properties of such tunnelling diode may not be included Hence it is necessary in due

course to fabricate and build such a compact frequency doubler in the future to verify

the simulation results

From the simulation point of view the less than 1 Conversion Efficiency

obtained is still good enough for a first attempt at a frequency doubler which utilises the

new ASPAT tunnelling diode The frequency doubler obtained from this work is suitable

for use in zero bias varistor modes for low power application The varistor mode doubler

performances from this simulation work are gathered and compared to other in the

literature as summarized in Table 71 below where fout is 40GHz η is 015 Ln is 28dB

and Pi is -10dBm

Table 71 Comparison parameter obtained from this simulation to the state-of-the-art multiplier

diode

Device fout (GHz) η() Ln(dB) Pi (dBm)

ASPAT 40 015 28 -10

SBD (Si)[177] 104 2 134 -10

SBD (GaAs)[178] 13 2 137 -16

214

The performance of the 2040GHz is compared to the literature based on their input

power below -8dBm since the ASPAT is only capable of working at low power To the

best of the author knowledge very few diodes operating in varistor mode at low power

excitation can be found in the field of research and industry Therefore this 2040GHz

ASPAT Doubler might a first for tunnel diodes if it can be fabricated and test at Ka band

and above

76 Conclusions

In this chapter another alternative application based on non-linear features of

GaAsAlAs ASPAT has been presented The simulation of a frequency multiplier

(doubler) was carried out utilizing the 4times4microm2 size ASPAT diode The theory and the

ability of the ASPAT diode to operate as a frequency source were explained in detail

A unique varistor mode frequency multiplier circuit topology for the 2040GHz

ASPAT doubler has been demonstrated and briefly discussed The details and step by

step simulation technique utilizing harmonic balance from Keysight ADS has been

presented Even though the conversion efficiency is very small at 015 and large

conversion loss of 28dB there is still space for improvement in term of design ie

different circuit topology optimized input and output matching circuit etc This design

can be a good reference for a doubler operating at very low power but produce high

frequencies in Ka band

215

8 CONCLUSION AND FUTURE WORK

81 Conclusion

The main focus of this research was the development of a new tunneling diode

namely the asymmetrical spacer layer tunnel (ASPAT) diode for process repeatability

manufacturability and reproducibility The broad study undertaken was to improve the

microwave performance technology by introducing a new type of tunneling diode

For years the asymmetrical spacer layer tunnel diode was unable to be manufactured

due to the high sensitivity of the tunneling current to the barrier thickness This changed

dramatically when the MBE method was carefully optimized to precisely control the

growth to sub-monolayer precisions When stability repeatability and reproducibility in

the epitaxial growth was achieved the next step was to qualify the fabrication process of

the diodes themselves thus ensuring high performance device can be delivered to the

market

For this purpose GaAsAlAs ASPAT diodes made of two different types of

substrates were grown The first batch was grown in the Riber V100HU SSMBE and

used doped substrates Samples XMBE307 and XMBE368 were successfully grown

and fabricated from that batch The DC characterization obtained from measurement

proved that this first batch had fully functional reproducible and manufacturable

devices Later a second batch using semi insulating substrates improvement in spacer

layer and doping concentration were grown This set of samples (9 x 2rdquo wafers grown

simultaneously) and denoted as XMBE304 also showed fully functional DC

characteristic and was used for RF characterization and detector integrated circuits

The conventional GaAsAlAs ASPAT diode structures grown on doped

substrates and developed previously in our lab were not suitable for high frequency RF

characterizations Therefore a major contribution of this work was to develop a new

fabrication technique for a new GaAsAlAs ASPAT structure using semi insulating

substrates to achieve repeatability manufacturability and reproducibility in term of

process flow DC characteristics and ultimately RF characteristics Apart from the

enhancement of the epitaxial layer the other important contribution of this research was

216

the optimization of the small 4times4microm2 emitter size by incorporating both vertical and

lateral structure based purely on low cost I-line optical lithography

To obtain a repeatable and manufacturable fabrication process of lateral

GaAsAlAs ASPAT structures the key issue was to solve the over etching of the

effective mesa area when qualifying the Air Bridge technique This issue caused all

semiconductors under the metal contact to be completely lost thus leaving the metal

contact hanging without connection to any bond pad area However the developments of

the Dielectric Bridge technique realized the true performance of the GaAsAlAs ASPAT

diode structures The samplersquos surface cleanliness as well as DC and RF performance

showed significant improvement when using Si3N4 as a dielectric layer Due to the

highly isotropic etching profile and thicker GaAs layer in XMBE304 samples and

although the smallest emitter size designed on the mask was 2times2microm2 only 4times4microm

2 were

reproducible and showed good uniformity in I-V characteristics

Upon successful optimization in the fabrication process flow of the small emitter

size diodes (4times4microm2 6times6microm

2 and 10times10microm

2) a good uniformity of better than 91

was obtained for DC measurement results within a tile containing over 1000 devices

This confirmed that the lateral GaAsAlAs ASPAT diode structure can only be realized

through the Dielectric Bridge technique These devices were further characterized with

S-parameter measurements and their intrinsic and extrinsic parameter values and

junction capacitances series resistances and junction resistances were extracted leading

to intrinsic cut-off frequencies of 600GHz 429GHz and 100GHz for the three device

sizes respectively

Temperature dependence measurements and simulations were also carried out in

order to confirm that the ASPAT diodersquos characteristics were temperature insensitive

showing less than 5 change in current at both extremes of temperatures 77K to 400K

By comparison the SBD I-V characteristics variations with temperature span orders of

magnitude Physical modelling agreed very well with measured data confirming good

and validated models that can also describe temperature effects

For the realization of the integrated ASPAT millimeter wave detector empirical

modelling using ADS simulation tools was carried out This was performed to predict

the detector performance at 100GHz to comply with the initial objective to develop a

217

millimeter wave detector circuit The simulations using the 4times4microm2 diode models led to

a successful 100GHz circuit design able to detect 100GHz incoming frequency with

2100VW voltage sensitivity

The first ever GaAsAlAs ASPAT diode frequency multiplier design was also

attempted A reasonably good result was obtained for a 20 to 40GHz frequency doubler

operating in varistor mode However the conversion efficiency obtained was less than

1 Further research on this is required to improve the efficiency by using other circuit

topology ie using a balun or other Co-planar waveguides Ultimately fabricating and

testing the actual multiplier circuits are essential to validate the simulation data

82 Future Work

This work has provided a foundation for a reproducible and repeatable GaAsAlAs

ASPAT (SI substrate) wafer fabrication process and recommendations for design and

simulation of ASPAT diode MMIC detectors and frequency source has also been

provided However the GaAsAlAs ASPAT diodes still remain immature and there are

many ways to improve its DC and RF performances both experimentally and in

simulations which will directly affect the detection performances

In term of fabrication process smaller size diodes ie submicron level can be

achieved using dry etching technique with proper calibration For wet etch technique the

etched profile still can be improved by thinning the doped layers so that etching time

will be reduced and hence dimensions down to 2times2microm2 or even 15times15microm

2 can be

reproducibly made

For simulations advanced AC and RF modelling utilizing physical device

simulation available software (SILVACO) must be include in future research hence

holistic study can be conducted to improve the understanding of the ASPAT

For MMIC detector and multiplier design it is vital to produce actual MMIC

devices so that the simulation results can be validated Ultimately tested devices with

good performances can be realized and manufactured

218

REFERENCES

1 Laeri F U Simon and M Wark Host-Guest-Systems Based on Nanoporous

Crystals 2006 John Wiley amp Sons

2 Łukasiak L and A Jakubowski History of semiconductors Journal of

Telecommunications and information technology 2010 p 3-9

3 Song H-J and T Nagatsuma Present and future of terahertz communications

IEEE Transactions on Terahertz Science and Technology 2011 1(1) p 256-

263

4 Hu B and M Nuss Imaging with terahertz waves Optics letters 1995 20(16)

p 1716-1718

5 Smith PR DH Auston and MC Nuss Subpicosecond photoconducting

dipole antennas IEEE Journal of Quantum Electronics 1988 24(2) p 255-260

6 Nagatsuma T Terahertz technologies present and future IEICE Electronics

Express 2011 8(14) p 1127-1142

7 Kumar S et al A 18-THz quantum cascade laser operating significantly above

the temperature of [planck][omega]kB Nature Physics 2011 7(2) p 166-171

8 Phillips T and D Woody Millimeter-and submillimeter-wave receivers Annual

Review of Astronomy and Astrophysics 1982 20(1) p 285-321

9 Whitmer HCTaCA Crystal Rectifiers McGraw-Hill book Company

London 1948

10 Young D and J Irvin Millimeter frequency conversion using Au-n-type GaAs

Schottky barrier epitaxial diodes with a novel contacting technique Proceedings

of the Ieee 1965 53(12) p 2130-2131

11 Liu L et al A broadband quasi-optical terahertz detector utilizing a zero bias

Schottky diode IEEE microwave and wireless components letters 2010 20(9) p

504-506

12 Sankaran S Schottky barrier diodes for millimeter wave detection in a foundry

CMOS process IEEE Electron Device Letters 2005 26(7) p 492-494

13 Chattopadhyay G Submillimeter-wave coherent and incoherent sensors for

space applications in Sensors 2008 Springer p 387-414

14 Anand Y and WJ Moroney Microwave mixer and detector diodes

Proceedings of the Ieee 1971 59(8) p 1182-1190

15 Syme RT Microwave Detection Using GaasAlas Tunnel Structures Gec

Journal of Research 1993 11(1) p 12-23

16 Syme RT et al Asymmetric superlattices for microwave detection in Physical

Concepts of Materials for Novel Optoelectronic Device Applications 1991

International Society for Optics and Photonics

17 Missous M MJ Kelly and J Sexton Extremely uniform tunnel barriers for

low-cost device manufacture IEEE Electron Device Letters 2015 36(6) p 543-

545

18 Syme RT et al Novel GaAsAlAs tunnel structures as microwave detectors in

Semiconductors 92 1992 International Society for Optics and Photonics

19 Schwierz F and JJ Liou Semiconductor devices for RF applications evolution

and current status Microelectronics Reliability 2001 41(2) p 145-168

219

20 HayasHi H Development of Compound Semiconductor DevicesmdashIn Search of

Immense Possibilitiesmdash SEI TECHNICAL REVIEW 2011(72) p 5

21 Mead C Schottky barrier gate field effect transistor Proceedings of the Ieee

1966 54(2) p 307-308

22 Hooper W and W Lehrer An epitaxial GaAs field-effect transistor Proceedings

of the Ieee 1967 55(7) p 1237-1238

23 Drangeid K R Sommerhalder and W Walter High-speed gallium-arsenide

Schottky-barrier field-effect transistors Electronics Letters 1970 6(8) p 228-

229

24 Pillarisetty R Academic and industry research progress in germanium

nanodevices Nature 2011 479(7373) p 324-328

25 Oxley TH 50 years development of the microwave mixer for heterodyne

reception IEEE transactions on microwave theory and techniques 2002 50(3)

p 867-876

26 Baca AG and CI Ashby Fabrication of GaAs devices 2005 IET

27 Cho AY and J Arthur Molecular beam epitaxy Progress in solid state

chemistry 1975 10 p 157-191

28 Cho A Growth of IIIndashV semiconductors by molecular beam epitaxy and their

properties Thin Solid Films 1983 100(4) p 291-317

29 Kiehl RA and TG Sollner High speed heterostructure devices 1994

Academic Press

30 Feiginov M et al Resonant-tunnelling-diode oscillators operating at

frequencies above 11 THz Applied Physics Letters 2011 99(23) p 233506

31 Chang LL L Esaki and R Tsu Resonant tunneling in semiconductor double

barriers Applied Physics Letters 1974 24(12) p 593-595

32 Kasjoo SR Novel Electronic Nanodevices Operating in the Terahertz Region

2012

33 Kanaya H et al Structure dependence of oscillation characteristics of

resonant-tunneling-diode terahertz oscillators associated with intrinsic and

extrinsic delay times Japanese Journal of Applied Physics 2015 54(9) p

094103

34 Chattopadhyay G Technology capabilities and performance of low power

terahertz sources IEEE Transactions on Terahertz Science and Technology

2011 1(1) p 33-53

35 Betz A and R Boreiko A practical Schottky mixer for 5 THz in Proceedings of

the 7th International Symposium on Space Terahertz Technology 1996

36 Yu D et al Ultra high-speed 025-spl mum emitter InP-InGaAs SHBTs with

fsub maxof 687 GHz in Electron Devices Meeting 2004 IEDM Technical

Digest IEEE International 2004 IEEE

37 Das MB Optoelectronic detectors and receivers speed and sensitivity limits in

Optoelectronic and Microelectronic Materials Devices 1998 Proceedings 1998

Conference on 1999 IEEE

38 Rodwell MJ et al Submicron scaling of HBTs IEEE Transactions on Electron

Devices 2001 48(11) p 2606-2624

220

39 Bouloukou A and S Missous Novel High-breakdown Low-noise InGaAs-

InA1As Transistors for Radio Astronomy Applications 2006 University of

Manchester

40 Bean J Materials and technologies 1990 John Wiley amp Sons New York p

13

41 Swaminathan V and A Macrander Materials aspects of GaAs and InP based

structures 1991 Prentice-Hall Inc

42 Vurgaftman I J Meyer and L Ram-Mohan Band parameters for IIIndashV

compound semiconductors and their alloys Journal of applied physics 2001

89(11) p 5815-5875

43 Dingle R W Wiegmann and CH Henry Quantum states of confined carriers

in very thin Al x Ga 1minus x As-GaAs-Al x Ga 1minus x As heterostructures Physical

Review Letters 1974 33(14) p 827

44 Sze SM and KK Ng Physics of semiconductor devices 2006 John wiley amp

sons

45 Tyagi MS Introduction to semiconductor materials and devices 2008 John

Wiley amp Sons

46 Schubert E Delta doping of IIIndashV compound semiconductors Fundamentals

and device applications Journal of Vacuum Science amp Technology A Vacuum

Surfaces and Films 1990 8(3) p 2980-2996

47 Rhoderick EH Metal-semiconductor contacts IEE Proceedings I-Solid-State

and Electron Devices 1982 129(1) p 1

48 Piotrowska A A Guivarch and G Pelous Ohmic contacts to IIIndashV compound

semiconductors A review of fabrication techniques Solid-State Electronics

1983 26(3) p 179-197

49 Rideout V A review of the theory and technology for ohmic contacts to group

IIIndashV compound semiconductors Solid-State Electronics 1975 18(6) p 541-

550

50 Baca A et al A survey of ohmic contacts to III-V compound semiconductors

Thin Solid Films 1997 308 p 599-606

51 Higman T et al Structural analysis of AundashNindashGe and AundashAgndashGe alloyed

ohmic contacts on modulation‐doped AlGaAsndashGaAs heterostructures Journal of

applied physics 1986 60(2) p 677-680

52 Chen KJ et al High-performance enhancement-mode InAlAsInGaAs HEMTs

using non-alloyed ohmic contact and Pt-based buried-gate in Indium Phosphide

and Related Materials 1995 Conference Proceedings Seventh International

Conference on 1995 IEEE

53 Berlin L The man behind the microchip Robert Noyce and the invention of

Silicon Valley 2005 Oxford University Press

54 Goodhue W et al Large room‐temperature effects from resonant tunneling

through AlAs barriers Applied Physics Letters 1986 49(17) p 1086-1088

55 Kerr A and Y Anand Schottky diode MM detectors with improved sensitivity

and dynamic range Microwave Journal 1981 24 p 67-71

56 Davies R Simulations of the current-voltage characteristics of semiconductor

tunnel structures Gec Journal of Research 1987 5(2) p 65-75

221

57 Kelly M Tunnel structures and devices over the coming decade Philosophical

Transactions of the Royal Society of London A Mathematical Physical and

Engineering Sciences 1996 354(1717) p 2291-2293

58 Wilkinson V and M Kelly Manufacturability of quantum semiconductor

devices in High Performance Electron Devices for Microwave and

Optoelectronic Applications 1995 EDMO IEEE 1995 Workshop on 1995

IEEE

59 Wilkinson V M Kelly and M Carr Tunnel devices are not yet

manufacturable Semiconductor Science and Technology 1997 12(1) p 91

60 Eaves L and MJ Kelly The current status of semiconductor tunnelling devices

Philos trans of the Roy soc of London Ser A Math phys and eng sciences

1996 354(1717)

61 Billen K V Wilkinson and M Kelly Manufacturability of heterojunction

tunnel devices further progress Semiconductor Science and Technology 1997

12(7) p 894

62 Kelly M The engineering of quantumndashdot devices Philosophical Transactions

of the Royal Society of London A Mathematical Physical and Engineering

Sciences 2003 361(1803) p 393-401

63 Kelly M New statistical analysis of tunnel diode barriers Semiconductor

Science and Technology 2000 15(1) p 79

64 Hayden R et al Ex situ re-calibration method for low-cost precision epitaxial

growth of heterostructure devices Semiconductor Science and Technology

2002 17(2) p 135

65 Dasmahapatra P et al Thickness control of molecular beam epitaxy grown

layers at the 001ndash01 monolayer level Semiconductor Science and Technology

2012 27(8) p 085007

66 Hayden R M Missous and M Kelly Precision growth for the manufacture of

semiconductor heterostructure devices Semiconductor Science and Technology

2001 16(8) p 676

67 Shao C et al Highly reproducible tunnel currents in MBE-grown

semiconductor multilayers Electronics Letters 2012 48(13) p 792-794

68 Abdullah MR et al GaAsAlAs tunnelling structure Temperature dependence

of ASPAT detectors in Millimeter Waves and THz Technology Workshop

(UCMMT) 2015 8th UK Europe China 2015 IEEE

69 Ariffin KZ et al Asymmetric Spacer Layer Tunnel In0 18Ga0 82AsAlAs

(ASPAT) Diode using double quantum wells for dual functions Detection and

oscillation in Millimeter Waves and THz Technology Workshop (UCMMT)

2015 8th UK Europe China 2015 IEEE

70 Liboff RL Introductory quantum mechanics 2003 Addison-Wesley

71 Esaki L Discovery of the tunnel diode IEEE Transactions on Electron Devices

1976 23(7) p 644-647

72 Landau LD LEM Quantum Mechanics Non-relativistic Theory Pergamon 3

73 Landau LD et al Quantum Mechanics Non‐Relativistic Theory Vol 3 of

Course of Theoretical Physics 1958 AIP

222

74 Syme R Tunnelling devices as microwave mixers and detectors Philosophical

Transactions of the Royal Society of London A Mathematical Physical and

Engineering Sciences 1996 354(1717) p 2351-2364

75 Syme R et al Tunnel diode with asymmetric spacer layers for use as

microwave detector Electronics Letters 1991 27(23) p 2192-2194

76 Brown E W Goodhue and T Sollner Fundamental oscillations up to 200 GHz

in resonant tunneling diodes and new estimates of their maximum oscillation

frequency from stationary‐state tunneling theory Journal of applied physics

1988 64(3) p 1519-1529

77 Reddy M Schottky-collector resonant tunnel diodes for sub-millimeter-wave

applications 1997 University of California Santa Barbara

78 Cox R and H Strack Ohmic contacts for GaAs devices Solid-State Electronics

1967 10(12) p 1213IN71215-1214IN81218

79 Valdes LB Resistivity measurements on germanium for transistors

Proceedings of the IRE 1954 42(2) p 420-427

80 Schroder DK Semiconductor material and device characterization 2006 John

Wiley amp Sons

81 Klootwijk J and C Timmering Merits and Limitations of Circular TLM

structures for contact resistance determination for novel 111-V HBTs Proc

fEEE 2004

82 Marlow GS and MB Das The effects of contact size and non-zero metal

resistance on the determination of specific contact resistance Solid-State

Electronics 1982 25(2) p 91-94

83 Murrmann H and D Widmann Current crowding on metal contacts to planar

devices IEEE Transactions on Electron Devices 1969 16(12) p 1022-1024

84 Berger H Models for contacts to planar devices Solid-State Electronics 1972

15(2) p 145-158

85 Reeves G and H Harrison Obtaining the specific contact resistance from

transmission line model measurements IEEE Electron Device Letters 1982

3(5) p 111-113

86 Shur MS GaAs devices and circuits 2013 Springer Science amp Business

Media

87 Popescu D and B Odbert The Advantages Of Remote Labs In Engineering

Education Educatorrsquos Corner-Agilent Technologies-application note 2011 p

11

88 DataSheet Karl Suss- PM5 Probe System Datasheet 2013

89 Keysight IC-CAP Device Modeling Software 2016 Available from

httpwwwkeysightcomenpc-1297149ic-cap-device-modeling-software-

measurement-control-and-parameter-extractioncc=USamplc=eng

90 DataSheet Anritsu 37369A Vector Network Analyzer Datasheet 2016 Available

from httpwwwtestequipmenthqcomdatasheetsANRITSU-37397D-

Datasheetpdf

91 Packard H HP 4142B Modular DC SourceManual Operation Manual 1992

Available from httpcpliteratureagilentcomlitwebpdf04142-90010pdf

223

92 Microtech C Cascade Microtech- Wincal High Performence RF calaibration

Software (Official Website) 2016 Available from

httpswwwcascademicrotechcom

93 Singh J Electronic and optoelectronic properties of semiconductor structures

2007 Cambridge University Press

94 Whyte W Cleanroom design 1999 Wiley Online Library

95 Vieu C et al Electron beam lithography resolution limits and applications

Applied Surface Science 2000 164(1) p 111-117

96 La Fontaine B Lasers and Moorersquos law SPIE Professional October 2010 p

20

97 Madou MJ Fundamentals of microfabrication the science of miniaturization

2002 CRC press

98 Serway R Physics for Scientists and Engineers 1996 Saunders Publ

Philadelphia

99 Jalali B and S Pearton InP HBTs growth processing and applications 1995

Artech House Publishers

100 Shih YC et al Effects of interfacial microstructure on uniformity and thermal

stability of AuNiGe ohmic contact to n‐type GaAs Journal of applied physics

1987 62(2) p 582-590

101 Zawawi M Advanced In0 8Ga0 2AsAlAs Resonant Tunneling Diodes

forApplications in Integrated mm-waves MMIC Oscillators 2015

102 Zawawi MAM et al Fabrication of Submicrometer InGaAsAlAs Resonant

Tunneling Diode Using a Trilayer Soft Reflow Technique With Excellent

Scalability IEEE Transactions on Electron Devices 2014 61(7) p 2338-2342

103 Silvaco I ATLAS Users Manual Device Simulation Software 2010 Santa

Clara CA

104 Kyono C et al Dependence of apparent barrier height on barrier thickness for

perpendicular transport in AlAsGaAs single‐barrier structures grown by

molecular beam epitaxy Applied Physics Letters 1989 54(6) p 549-551

105 Yang K JR East and GI Haddad Numerical modeling of abrupt

heterojunctions using a thermionic-field emission boundary condition Solid-

State Electronics 1993 36(3) p 321-330

106 Varshni YP Temperature dependence of the energy gap in semiconductors

Physica 1967 34(1) p 149-154

107 Handbook LLM Precision DC Current Voltage and Resistance

Measurements Keithley Instruments Inc[online] 6th revision Ohio 2004

108 Lipsky SE Microwave passive direction finding 2004 SciTech Publishing

109 Howell CM and SJ Parisi Principles Applications and Selection of Receiving

Diodes MACOM Semiconductor Products Division Application note AG314

110 Schulman J D Chow and D Jang InGaAs zero bias backward diodes for

millimeter wave direct detection IEEE Electron Device Letters 2001 22(5) p

200-202

111 Zhang Z et al Sub-Micron Area Heterojunction Backward Diode Millimeter-

Wave Detectors With 018$ rm pWHz^12 $ Noise Equivalent Power IEEE

microwave and wireless components letters 2011 21(5) p 267-269

224

112 Jin N et al High sensitivity Si-based backward diodes for zero-biased square-

law detection and the effect of post-growth annealing on performance IEEE

Electron Device Letters 2005 26(8) p 575-578

113 Shashkin VI et al Millimeter-wave detectors based on antenna-coupled low-

barrier Schottky diodes International journal of infrared and millimeter waves

2007 28(11) p 945-952

114 Zhao P et al GaN Heterostructure Barrier Diodes Exploiting Polarization-

Induced $delta $-Doping IEEE Electron Device Letters 2014 35(6) p 615-

617

115 Pozar DM Microwave engineering 2009 John Wiley amp Sons

116 Koolen M J Geelen and M Versleijen An improved de-embedding technique

for on-wafer high-frequency characterization in Bipolar Circuits and

Technology Meeting 1991 Proceedings of the 1991 1991 IEEE

117 Cao M et al RF characteristics uniformity of GaAsAlAs tunnel diodes in

Infrared Millimeter and Terahertz waves (IRMMW-THz) 2016 41st

International Conference on 2016 IEEE

118 Gao J RF and microwave modeling and measurement techniques for field effect

transistors 2010 SciTec

119 Ren T et al A 340-400 GHz Zero-Biased Waveguide Detector Using an Self-

Consistent Method to Extract the Parameters of Schottky Barrier Diode Applied

Computational Electromagnetics Society Journal 2015 30(12)

120 Fobelets K et al High‐frequency capacitances in resonant interband tunneling

diodes Applied Physics Letters 1994 64(19) p 2523-2525

121 Diebold S et al Modeling and Simulation of Terahertz Resonant Tunneling

Diode-Based Circuits IEEE Transactions on Terahertz Science and Technology

2016 6(5) p 716-723

122 Yong Z et al Design of a 220 GHz frequency tripler based on EM model of

Schottky diodes JOURNAL OF INFRARED AND MILLIMETER WAVES

2014 33(4) p 405-411

123 Louhi JT and AV Raisanen On the modeling and optimization of Schottky

varactor frequency multipliers at submillimeter wavelengths IEEE transactions

on microwave theory and techniques 1995 43(4) p 922-926

124 Guo J Z Zhang and C Qian Modeling of commercial millimeter wave

Schottky diodes in Microwave and Millimeter Wave Technology (ICMMT) 2016

IEEE International Conference on 2016 IEEE

125 Schneider M Metal-semiconductor junctions as frequency converters Infrared

and Millimeter Waves 1982 6 p 209

126 Muth C et al Advanced technology microwave sounder on NPOESS and NPP

in Geoscience and Remote Sensing Symposium 2004 IGARSS04 Proceedings

2004 IEEE International 2004 IEEE

127 Putley E Thermal detectors in Optical and Infrared Detectors 1977 Springer

p 71-100

128 Martin DH Spectroscopic techniques for far infra-red submillimetre and

millimetre waves in Spectroscopic Techniques for Far Infra-red Submillimetre

and Millimetre Waves 1967

225

129 Lucas W Tangential sensitivity of a detector video system with RF

preamplification in Proceedings of the Institution of Electrical Engineers 1966

IET

130 Balocco C et al Low-frequency noise of unipolar nanorectifiers Applied

Physics Letters 2011 99(11) p 113511

131 Benford D T Hunter and TG Phillips Noise equivalent power of background

limited thermal detectors at submillimeter wavelengths International journal of

infrared and millimeter waves 1998 19(7) p 931-938

132 Papoušek D Vibration-rotational Spectroscopy and Molecular Dynamics

Advances in Quantum Chemical and Spectroscopical Studies of Molecular

Structures and Dynamics Vol 9 1997 World Scientific

133 Nyquist H Thermal agitation of electric charge in conductors Physical review

1928 32(1) p 110

134 Turner CS Johnson-Nyquist Noise url httpwww claysturner

comdspJohnson-NyquistNoise pdf(Letzter Abruf Juli 2012)

135 Voss RF 1f (flicker) noise A brief review in 33rd Annual Symposium on

Frequency Control 1979 1979 IEEE

136 McWhorter AL 1f noise and related surface effects in germanium 1955

137 Hooge FN 1ƒ noise is no surface effect Physics letters A 1969 29(3) p 139-

140

138 Van der Ziel A Noise Sources characterization measurement Prentice-Hall

Information and System Sciences Series Englewood Cliffs Prentice-Hall 1970

1970

139 Hooge F 1f noise sources IEEE Transactions on Electron Devices 1994

41(11) p 1926-1935

140 Der Ziel A Theory of shot noise in junction diodes and junction transistors

Proceedings of the IRE 1955 43(11) p 1639-1646

141 Schottky W Small-shot effect and flicker effect Physical review 1926 28(1)

p 74

142 Cowley A and H Sorensen Quantitative comparison of solid-state microwave

detectors IEEE transactions on microwave theory and techniques 1966 14(12)

p 588-602

143 Schulman J et al 1$f $ Noise of Sb-Heterostructure Diodes for Pre-Amplified

Detection IEEE microwave and wireless components letters 2007 17(5) p

355-357

144 Lynch JJ et al Passive millimeter-wave imaging module with preamplified

zero-bias detection IEEE transactions on microwave theory and techniques

2008 56(7) p 1592-1600

145 Su N et al Sb-heterostructure millimeter-wave detectors with reduced

capacitance and noise equivalent power IEEE Electron Device Letters 2008

29(6) p 536-539

146 Westlund A Self-Switching Diodes for Zero-Bias Terahertz Detection 2015

Chalmers University of Technology

147 Yajima T and L Esaki Excess noise in narrow germanium pn junctions

Journal of the physical society of Japan 1958 13(11) p 1281-1287

226

148 Sommers H Tunnel diodes as high-frequency devices Proceedings of the IRE

1959 47(7) p 1201-1206

149 Miraftab V and A Abdipour Harmonic balance analysis of a microwave

balanced power amplifier in Electrical and Computer Engineering 2001

Canadian Conference on 2001 IEEE

150 Patrashin M et al GaAsSbInAlAsInGaAs Tunnel Diodes for Millimeter Wave

Detection in 220ndash330-GHz Band IEEE Transactions on Electron Devices 2015

62(3) p 1068-1071

151 Hesler JL and TW Crowe Responsivity and noise measurements of zero-bias

Schottky diode detectors Proc ISSTT 2007 p 89-92

152 Su N et al Temperature dependence of high frequency and noise performance

of Sb-heterostructure millimeter-wave detectors IEEE Electron Device Letters

2007 28(5) p 336-339

153 Lynch J et al Unamplified direct detection sensor for passive millimeter wave

imaging in Proc of SPIE Vol 2006

154 ZHANG Z et al A physics-based tunneling model for Sb-heterostructure

backward tunnel diode millimeter-wave detectors International Journal of High

Speed Electronics and Systems 2011 20(03) p 589-596

155 Bahl IJ and P Bhartia Microwave solid state circuit design 2003 John Wiley

amp Sons

156 Yeom K-W Microwave Circuit Design A Practical Approach Using ADS

2015 Prentice Hall Press

157 Xie L et al A W-band detector with high tangential signal sensitivity and

voltage sensitivity in Microwave and Millimeter Wave Technology (ICMMT)

2010 International Conference on 2010 IEEE

158 Fay P et al High-performance antimonide-based heterostructure backward

diodes for millimeter-wave detection IEEE Electron Device Letters 2002

23(10) p 585-587

159 Hrobak M et al Planar zero bias Schottky diode detector operating in the E-

and W-band in Microwave Conference (EuMC) 2013 European 2013 IEEE

160 Maas SA Nonlinear microwave and RF circuits 2003 Artech House

161 Appleby R and RN Anderton Millimeter-wave and submillimeter-wave

imaging for security and surveillance Proceedings of the Ieee 2007 95(8) p

1683-1690

162 Crowe TW et al Opening the terahertz window with integrated diode circuits

IEEE Journal of Solid-State Circuits 2005 40(10) p 2104-2110

163 Raisanen AV Frequency multipliers for millimeter and submillimeter

wavelengths Proceedings of the Ieee 1992 80(11) p 1842-1852

164 Erickson NR Diode frequency multipliers for terahertz local-oscillator

applications in Astronomical Telescopes amp Instrumentation 1998 International

Society for Optics and Photonics

165 Mehdi I et al Terahertz local oscillator sources performance and capabilities

in Astronomical Telescopes and Instrumentation 2003 International Society for

Optics and Photonics

166 Tonouchi M Cutting-edge terahertz technology Nature photonics 2007 1(2)

p 97-105

227

167 Nilsen SM et al Single barrier varactors for submillimeter wave power

generation IEEE transactions on microwave theory and techniques 1993 41(4)

p 572-580

168 Xiao Q et al A 270-GHz tuner-less heterostructure barrier varactor frequency

tripler IEEE microwave and wireless components letters 2007 17(4) p 241-

243

169 David T et al Monolithic integrated circuits incorporating InP-based

heterostructure barrier varactors IEEE microwave and wireless components

letters 2002 12(8) p 281-283

170 Lieneweg U B Hancock and J Maserjian Barrier-intrinsic-N+(BIN) diodes

for near-millimeter wave generation in Conference Digest for the Twelft

International Conference on Infrared and Millimeter Waves 1987

171 Lieneweg U et al Modeling of planar varactor frequency multiplier devices

with blocking barriers IEEE transactions on microwave theory and techniques

1992 40(5) p 839-845

172 Chattopadhyay G et al An all-solid-state broad-band frequency multiplier

chain at 1500 GHz IEEE transactions on microwave theory and techniques

2004 52(5) p 1538-1547

173 Maestrini A et al A 17-19 THz local oscillator source IEEE microwave and

wireless components letters 2004 14(6) p 253-255

174 Faber MT J Chramiec and ME Adamski Microwave and millimeter-wave

diode frequency multipliers 1995 Artech House Publishers

175 Frerking MA and JR East Novel heterojunction varactors Proceedings of the

Ieee 1992 80(11) p 1853-1860

176 Penfield P and RP Rafuse Varactor applications 1962

177 Palazzi V et al Low-power frequency doubler in cellulose-based materials for

harmonic RFID applications IEEE microwave and wireless components letters

2014 24(12) p 896-898

178 Presas SM Microwave frequency doubler integrated with miniaturized planar

antennas 2008

228

APPENDICES

Appendix I Doped substrate process details

Mask Stage Process Stage Process step Process detail Equipment

Mask 1 (Mesa

Etch)

Sample clean NMP 10 min 80˚C Beaker

Acetone 5 min Beaker

Isopropanol (IPA) 5 min Beaker

Apply Resist Prebake Bake for 5mins 150C Hotplate

1st Resist S1805 - Program 4 Laurell Spinner

Post bake 110C for 1 Min Hotplate

Photolithography Mask Align mask to wafer

MA4 Mask

aligner

Expose

1 min 09mW iline

(Compensation error set to 1)

MA4 Mask

aligner

Develop MIF 319 for 1 min Beaker

Post Exposure

Bake Oven or Bake 120C for 15

minutes Hotplate

Etch Etch Cal Cal Sample - etch for 2 minutes Beaker

Etch

H3PO4H2O2H2O 3150

time is determined by the etch cal Beaker

Resist Strip Acetone - 5 min and IPA - 5 min Beaker

Mask2 (Top

Contact)

Sample clean NMP ultrasonic for 15 Min Power 3 Beaker

Acetone ultrasonic for 5 Min Power 3 Beaker

Isopropanol (IPA) ultrasonic for 5 Min Power 3 Beaker

Apply Resist Prebake Bake for 5mins 150C Hotplate

1st Resist AznLoF - 2um grade - Program 6 Laurell Spinner

Post bake 110C for 1 Min Hotplate

Photolithography Mask Align mask to wafer

MA4 Mask

aligner

Expose

55 seconds 09mW i-line

(Compensation error set to 1)

MA4 Mask

aligner

Post Exposure bake 110C for 1 Min Hotplate

Develop MIF 326 for 1 mins Beaker

Clean O2 Plasma Etch

20sec 100W forward (340)

60mTorr Plasma Tech

Surface De-oxide HCLH2O 11 40 sec Beaker

Metallisation Evaporation Clean trays + metals Beaker

Load sample and metals

Edward

Evaporator

AuGe (100mg) deposits 55nm

Edward

Evaporator

Ni (1cm) deposits 15nm

Edward

Evaporator

Au (10cm) deposits 200nm

Edward

Evaporator

Lift-off

NMP1165 for 24 hours (25C) or

20 min 80C Beaker

Anneal 420˚C 2min Furnace

229

TLM

Measurement ICCAP

Measurement

Bench

Mask 3

(BottomBacks

ide Contact)

Sample clean Acetone Optional Beaker

Isopropanol (IPA) Optional Beaker

Apply Resist Prebake

Bake for 5mins 150C to dry

the sample Hotplate

Resist (top side) S1813 - Program 6 Laurell Spinner

Soft bake 110C for 1 Min Hotplate

Photolithography Mask Align mask to wafer

MA4 Mask

aligner

Exposure

10 seconds 09mW I-line

(Compensation error set to 1)

MA4 Mask

aligner

Post Exposure bake 110C for 1 Min Hotplate

Develop MIF 319 for 2 mins Beaker

De-scum O2 Plasma Etch

20sec 100W forward (340)

60mTorr Plasma Tech

De-oxidise

Surface De-oxide

HCLH2O 11 40

sec Beaker

metallisation Evaporation Clean trays + metals Beaker

Load sample and metals

Edward

Evaporator

AuGe (100mg) deposits 55nm

Edward

Evaporator

Ni (1cm) deposits 15nm

Edward

Evaporator

Au (10cm) deposits 200nm

Edward

Evaporator

Lift-off

NMP1165 for 24 hours (25C) or

20 min 80C Beaker

Clean Water 3min Beaker

TLM

Measurement ICCAP

230

Appendix II Four Mask step Process Flow

Mask Stage Process Stage Process step Process detail

Mask1 Top

Contact)

Sample clean NMP ultrasonic for 15 Min Power 3

Acetone ultrasonic for 5 Min Power 3

Isopropanol (IPA) ultrasonic for 5 Min Power 3

Apply Resist Prebake Bake for 5mins 150C

1st Resist AznLoF - 2um grade - Program 6

Hot Plate 110C for 1 Min

Photolithography Mask Align mask to wafer

Expose 55 seconds 09mW iline (Compensation

error set to 1)

Post Exposure bake 110C for 1 Min

Develop MIF 326 for 1 mins

Clean O2 Plasma Etch 20sec 100W forward (340) 60mTorr

Surface De-oxide HCLH2O 11 40 sec

Metalisation Evaporation Clean trays + metals

Load sample and metals

AuGe (100mg) deposits 55nm

Ni (1cm) deposits 15nm

Au (10cm) deposits 200nm

Lift-off NMP1165 for 24 hours (25C) or 20 mnt 80C

Mask 2 (Mesa

Etch)

Sample clean NMP Optional

Acetone Optional

Isopropanol (IPA) Optional

Apply Resist Prebake Bake for 5mins 150C

1st Resist S1805 - Program 4

HotPlate 110C for 1 Min

Photolithography Mask Align mask to wafer

Expose 1 min 09mW i-line (Compensation error set

to 1)

Develop MIF 319 for 1 min

Post Exposure Bake Oven or Bake 120C for 15 minutes

Etch Etch Cal Cal Sample - etch for 2 minutes

Etch H3PO4H2O2H2O 3150 time is

determined by the etch cal

Measure TLM

Resist Strip Acetone - 5 min and IPA - 5 min

Mask 3(Isolation) Sample clean Acetone Optional

Isopropanol (IPA) Optional

231

Apply Resist Prebake Bake for 5mins 150C

Resist S1828 - Program 4

Hot Plate 115C for 1 mins

Photolithography Mask Align mask to wafer

Expose 9 mins 09mW iline (Compensation error set

to 1)

Develop MF 319 3 mins

Post Bake Oven Bake 120C for 15mins

Etch Etch Cal Refer to Etch Cal instr tab

Sub-collector Etch H3PO4H2O2H2O 3150 time is

determined by the etch cal

Resist Strip Acetone 5mins + IPA 5 mins in ultrasonic bath

power 1

inspection Microscope

Sample clean Acetone Optional

Mask 4 (Bottom

Contact)

Isopropanol (IPA) Optional

Apply Resist Prebake Bake for 5mins 150C 1st

Resist AznLoF - 2um grade - Program 6

HotPlate 110C for 1 Min

Photolithography Mask Align mask to wafer

Expose 10 seconds 09mW i-line (Compensation

error set to 1)

Post Exposure bake 110C for 1 Min

Develop MIF 326 for 2 mins

Clean O2 Plasma Etch 20sec 100W forward (340) 60mTorr

Surface De-oxide

HCLH2O 11 40 sec

Metalisation Evaporation Clean trays + metals

Load sample and metals

AuGe (100mg) deposits 55nm

Ni (1cm) deposits 15nm

Au (10cm) deposits 200nm

Lift-off NMP1165 for 24 hours (25C) or 20 mnt 80C

232

Appendix III Epitaxial Layer XMBE277

TABLE I The epitaxial structure for sample XMBE277

Layer Thickness (nm) Doping Concentration (cm-3)

n+- In053Ga047As 45 200 x 1019

n- In053Ga047As 25 300 x 1018

In053Ga047As 20 undoped

AlAs 13 undoped

In08Ga02As 45 undoped

AlAs 13 undoped

In053Ga047As 20 undoped

n- In053Ga047As 25 300 x 1018

n+- In053Ga047As 400 100 x 1019

Semi-insulating InP

233

Appendix IV SilVaco (Atlas) Simulation Code

go atlas

---------------------------------------------------------

Structure parameter definition (Constants) values in um

---------------------------------------------------------

Thicknesses

set t_contact1=0

set t_ohmic1=03

set t_emitter=004

set t_spacer1=0005

set t_barrier=000283

set t_spacer2=02

set t_collector=004

set t_ohmic2=045

Doping concentrations

set d_ohmic1=4e18

set d_emitter=2e17

set d_collector=2e17

set d_ohmic2=4e18

set d_gap=2

set d_mesa=4

set d_device=10

set d_etch=008

Layers

set I=$t_contact1

set A=$I+$t_ohmic1

set B=$A+$t_emitter

set C=$B+$t_spacer1

set D=$C+$t_barrier

set E=$D+$t_spacer2

set F=$E+$t_collector

set G=$F+$t_ohmic2

-------------------------------------

Mesh generator

-------------------------------------

mesh diagflip width=45

xmesh location=0 s=1

xmesh location=1 s=1

xmesh location=2 s=1

xmesh location=4 s=1

xmesh location=5 s=1

xmesh location=6 s=1

xmesh location=7 s=1

xmesh location=8 s=1

xmesh location=$d_mesa s=1

xmesh location=$d_mesa+$d_gap s=1

xmesh location=$d_device s=1

Ohmic1

ymesh l=0000 s=005

ymesh l=$I s=005

234

ymesh l=$A s=0005

ymesh l=$B s=0005

ymesh l=$C s=00005

ymesh l=$D s=00005

ymesh l=$E s=0009

ymesh l=$F s=0005

ymesh l=$G s=0005

-----------------------------------

SECTION 2 Regions Structure definition

-----------------------------------

region num=1 name=contact1 material=Gold ymin=0 ymax=$I

region num=2 name=ohmic1 material=GaAs ymin=$I ymax=$A

region num=3 name=emitter material=GaAs ymin=$A ymax=$B

region num=4 name=spacer1 material=GaAs ymin=$B ymax=$C

region num=5 name=barrier material=AlAs ymin=$C ymax=$D xmin=0 xmax=$d_mesa

calcstrain qtregion=1

region num=6 name=spacer2 material=GaAs ymin=$D ymax=$E

region num=7 name=collector material=GaAs ymin=$E ymax=$F

region num=8 name=ohmic2 material=GaAs ymin=$F ymax=$G

region num=9

name=etch material=Air ymin=0 ymax=$F+$d_etch xmin=$d_mesa xmax=$d

_device

---------------------------------

Electrodes

---------------------------------

electrode num=1 name=anode xmin=0 xmax=$d_mesa ymin=0 ymax

=$I material=Gold

electrode num=2 name=cathode xmin=$d_mesa+$d_gap xmax=$d_device ymin=$F+

$d_etch ymax=$F+$d_etch material=Gold

--------------------------------

Doping

--------------------------------

doping uniform ntype conc=$d_ohmic1 Region=2 ymin=$I ymax=$A

doping uniform ntype conc=$d_emitter Region=3 ymin=$A ymax=$B

doping uniform ntype conc=$d_collector Region=7 ymin=$E ymax=$F

doping uniform ntype conc=$d_ohmic2 Region=8 ymin=$F ymax=$G

--------------------------

Contacts

--------------------------

interface sc region=1

interface ss region=2

interface ss region=3

interface si region=4

interface si region=5

interface ss region=6

interface ss region=7

interface sc ymin=$F ymax=$F xmin=$d_mesa+$d_gap xmax=$d_device

interface tunnel region=5 dytunnel=0001

contact name=cathode

contact name=anode

235

------------------------------------------

SECTION 3 Material amp Models Definitions

------------------------------------------

material material=AlAs

permittivity=10 eg300=28 mc=004 affinity=305 nc300=4e19 nimin=1e1

material material=GaAs permittivity=139 eg300=14 mc=0067

affinity=407 nc300=09e17 nimin=1e6

BAND DIAGRAM

output tquantum bandparam qfn qfp valband conband charge polarcharge flowlines

STRUCTURE GRAPHIC

solve init

save outf=XMBE304+real2str

tonyplot XMBE304+real2str

------------------------------------------

SECTION 4 ANALYSIS

------------------------------------------

trap acceptor structure=top elevel=03 density=48e17 degenfac=10 sign=284e-17 sigp=1e-17

ymin=$A ymax=$C xmin=0 xmax=$d_mesa

trap acceptor structure=BOTTOM elevel=035 density=47e17 degenfac=10 sign=284e-17 sigp=1e-17

ymin=$E ymax=$F xmin=0 xmax=$d_mesa

models sisel sisnlderivs qtregion=1 print

method climit=1e-4 itlimit=50 maxtraps=20

DC ANALYSIS

log outf=XMBE304log

solve init

solve vanode=0 name=anode vstep=001 vfinal=15

save outf=XMBE304str

log off

tonyplot XMBE304str

tonyplot XMBE304log

Page 6: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 7: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 8: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 9: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 10: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 11: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 12: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 13: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 14: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 15: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 16: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 17: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 18: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 19: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 20: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 21: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 22: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 23: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 24: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 25: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 26: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 27: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 28: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 29: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 30: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 31: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 32: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 33: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 34: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 35: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 36: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 37: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 38: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 39: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 40: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 41: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 42: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 43: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 44: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 45: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 46: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 47: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 48: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 49: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 50: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 51: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 52: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 53: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 54: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 55: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 56: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 57: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 58: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 59: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 60: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 61: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 62: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 63: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 64: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 65: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 66: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 67: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 68: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 69: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 70: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 71: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 72: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 73: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 74: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 75: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 76: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 77: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 78: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 79: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 80: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 81: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 82: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 83: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 84: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 85: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 86: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 87: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 88: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 89: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 90: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 91: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 92: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 93: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 94: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 95: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 96: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 97: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 98: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 99: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 100: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 101: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 102: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 103: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 104: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 105: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 106: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 107: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 108: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 109: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 110: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 111: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 112: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 113: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 114: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 115: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 116: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 117: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 118: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 119: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 120: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 121: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 122: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 123: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 124: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 125: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 126: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 127: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 128: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 129: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 130: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 131: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 132: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 133: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 134: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 135: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 136: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 137: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 138: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 139: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 140: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 141: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 142: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 143: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 144: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 145: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 146: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 147: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 148: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 149: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 150: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 151: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 152: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 153: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 154: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 155: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 156: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 157: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 158: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 159: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 160: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 161: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 162: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 163: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 164: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 165: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 166: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 167: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 168: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 169: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 170: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 171: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 172: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 173: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 174: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 175: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 176: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 177: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 178: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 179: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 180: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 181: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 182: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 183: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 184: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 185: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 186: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 187: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 188: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 189: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 190: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 191: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 192: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 193: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 194: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 195: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 196: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 197: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 198: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 199: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 200: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 201: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 202: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 203: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 204: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 205: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 206: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 207: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 208: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 209: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 210: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 211: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 212: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 213: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 214: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 215: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 216: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 217: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 218: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 219: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 220: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 221: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 222: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 223: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 224: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 225: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 226: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 227: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 228: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 229: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 230: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 231: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 232: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 233: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 234: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre
Page 235: GaAs/AlAs ASPAT Diodes for Millimetre and Sub Millimetre