1

fT = 688 GHz and fmax = 800 GHz in

Lg = 40 nm In0.7Ga0.3As MHEMTs

with gm_max > 2.7 mS/m

D.-H. Kim, B. Brar and *J. A. del Alamo,Teledyne Scientific Company, *MIT

IEDM December-6th, 2011

III-V HEMT: record fT vs. time

2

Current record:ft=660 GHzLeuther IPRM 2011 (Fraunhofer Inst.)

(and MHEMT)

For >20 years, record fT obtained on InGaAs-channel HEMTs.

Well balanced devices:ft=644 GHz, fmax=680 GHz at same bias pointKim EDL 2010(MIT)

InGaAs-channel HEMTs offers record balanced fT and fmax.

RON ~ 420 Ohm-m

Strategy to improve fT

3

• In typical HEMTs: – RON: 350 ~ 450 Ω-m– T-Gate: Stem height = ~ 150 nm

Kim, EDL 2008

0 10 20 30 40 50 60-0.4

-0.3

-0.2

-0.1

0.0

0.1

0.2

0.3

0.4

0.5

Electron density [cm-3]

CB

Pro

file

[eV]

Vertical depth [nm]

Single delta doping

Zero bias

0

2

4

ns = 3 x 1012 /cm2

EC and n incross section

x 1018

Lg = 30 nm

Cgext

4

Contents

1. Introduction

2. Device Technology

3. DC and RF Characteristics

4. Analytical fT Model

5. Conclusions

5

- SiO2 assisted T-gate

Lg = 40 nm Gate-stem > 250 nm

- Two-step recess (InP = 6 nm)

- Pt (3 nm)/Ti/Pt/Au Schottky

- QW: 10 nm In0.7Ga0.3As n,Hall > 10,000 cm2/V-s

- *In0.52Al0.48As/In0.7Al0.3As spacer

- **Dual Si -doping

Device Technology

S

Cap

D

Etch stopper

Barrier

Channel

Buffer

tins

OxideLside

tch

Stem

Buried Pt

100 mm InP Substrate

KIM, *Electron Lett 2011 ** IEDM 2010

6

TEM Images

- Mo-based S/D with 2 m

- Gate Stem > 250 nm

GaAs Substrate

Graded Buffer

HEMT Epi

- Lg = 40 nm, Lside = 100 nm

- tins = ~ 4 nm

Buried Pt

Ti

Pt

Au

SG

D

7

DC of Lg = 40 nm InGaAs MHEMTs

- Maximum ID > 1 mA/m- RON = 280 Ω-m

- gm > 2 mS/m @ VDS = 0.3 V- gm_max = 2.75 mS/m

@ VDS = 0.8 V

0.0 0.2 0.4 0.6 0.8 1.00.0

0.4

0.8

1.2

0.5 V

I D [m

A/

m]

VDS [V]

VGS = 0.6 V

0.0 0.2 0.4 0.60

1

2

30.8 V

0.2 V

g m [m

S/m

]

VGS [V]

VDS = 0.1 V

- Output characteristics - - gm characteristics -

-0.2 0.0 0.2 0.4 0.610-10

10-9

10-8

10-7

10-6

10-5

10-4

10-3

VDS = 0.05 V

VDS = 0.8 V ID

I D, I

G [A

/m

]

VGS [V]

IG

8

Subthreshold characteristics- Lg = 40 nm - - gm_max scalability-

gm saturates.

1001000

1500

2000

2500

3000tins = ~ 4 nm

VDS = 0.5 V

In0.7Ga0.3As MHEMTs [This work] In0.7Ga0.3As PHEMTs [IEDM-10] InAs PHEMTs [IEDM-08]

g m,m

ax [

S/m

]

Lg [nm]

- VT = 0.02 V @ VDS = 0.5 V- S = 100 mV/dec., DIBL = 105 mV/V

As Lg ↓,

This work

109 1010 1011 10120

20

40

Frequency [Hz]

Gai

ns [d

B]

0

1

2

3

4

k

Ug

MSG

Stability Factor (k)

h21

fT = 602 GHz

9

fT & fmax: Lg = 40 nm, Wg = 2 x 20 mCalibration: LRRM, De-embedding: OPEN/SHORT

- fT already approaches to 600 GHz @ VDS = 0.35 V.

- VGS/VDS = 0.4/0.35 V -

109 1010 1011 10120

20

40

Frequency [Hz]

Gai

ns [d

B]

0

1

2

3

4

fT = 688 GHz

ID = 0.4 mA/mk

Ug

MSG

Stability Factor (k)

h21

- Record fT = 688 GHz @ VDS = 0.6 V.

- VGS/VDS = 0.4/0.6 V -

Gummel technique for fT extraction

10101010

In one-pole system:

Then:

Slope gives fT

Gummel, Proc IEEE 1969

109 1010 1011 10120

20

40

Frequency [Hz]

Gai

ns [d

B]

VDS = 0.6 VID = 0.4 mA/m

k

Ug

MSG

h21

0.0 20.0G 40.0G 60.0G0.00

0.04

0.08

0.12

Imag

(h21

-1)

Frequency [Hz]

1/slope = fT= 690 GHz

11111111

109 1010 1011 10120

20

40

Frequency [Hz]

Gai

ns [d

B]

VDS = 0.6 VID = 0.4 mA/m

k

Ug

MSG

h21

Different measurement systemfor fT extraction

0.0 20.0G 40.0G 60.0G0.00

0.04

0.08

0.12

Imag

(h21

-1)

Frequency [Hz]

- 8510C @TSC: 1 ~ 50 GHz

109 1010 1011 10120

20

40

Frequency [Hz]

Gai

ns [d

B]

- PNA @UCSB: 1 ~ 67 GHz

691 GHz0.0 20.0G 40.0G 60.0G

0.00

0.04

0.08

0.12

Imag

(h21

-1)

Frequency [Hz]

1/slope = fT= 690 GHz

12

Small-signal model for fT extraction

109 1010 1011 10120

20

40

Frequency [Hz]

Gai

ns [d

B]

0

1

2

3

4

VDS = 0.6 VID = 0.4 mA/m

k

Ug

MSG

Stability Factor (k)h21

109 1010 1011 10120

20

40

Frequency [Hz]

Gai

ns [d

B]

0

1

2

3

4Stability Factor (k)

freq (1.000GHz to 67.00GHz)

S11

S21/5

S22

5xS12

- Excellent agreement, modeled fT = 680 GHzfmax = 800 GHz

Summary on fT measurements

13131313

Measurements in two different test benches:

All measurements at same bias point: VGS=0.4 V, VDS=0.6 V

Kim, EDL 2010

8510C@TSC

PNA@UCSB

fT

[GHz]

From H21 688 688

From Gummel’s approach 690 691

From Small-signal model 680

fmax [GHz] 800

14

Balance in fT and fmax

0 200 400 600 800 10000

500

1000

1500

MIT/TSC HEMT Fujitsu HEMT NGAS HEMT SNU HEMT UCSB HBT UIUC HBT TSC HBT HRL HBT ETH HBT

max ff

f max

[GH

z]

fT [GHz]

300 600 700 = favg =

TSC/MIT(This work)

→ Best-balanced fT and fmax transistor→ Record fT FET

Lai, IEDM08Kim, IEDM10

Urteaga, DRC11

15

Contents

1. Introduction

2. Device Technology

3. DC and RF Characteristics

4. Analytical fT Model

5. Conclusions

16

Analytical fT Model

• First-order fT expression for HEMT:

gmivgs

goi

RDRS

Cgs Cgd

17

Break out ‘extrinsic’ capacitances

• Capacitance components [fF/mm]:

Cgs = Cgsi + Cgsext

= Cgsi_areal × Lg + Cgsext

Cgd = Cgdi + Cgdext

= Cgdi_areal × Lg + Cgdext

[fF/m2] [fF/m2]

18

Delay time analysis

• Delay time:

• Components of delay time:

18

Intrinsic delay (transit time)

Extrinsic delay Parasitic

delay

(Cgsi_areal + Cgdi_areal) Lggmi

19

Lg-dependent model parameters

0 100 2000

1000

2000

Cgs_ext

Cgs

, Cgd

[fF/

mm

]

Lg [nm]

VDS = 0.6 VVGS - VT = 0.3V

Cgs

Cgd

Cgd_ext

1000.1

1

10

VDS = 0.6 VVGS - VT = 0.3V

goi

Lg [nm]g m

i, goi [m

S/m

]

gmi

- Linearly proportional to Lg

- Cgs_ext > Cgd_ext

← |Vgs| < |Vgd|

- gmi saturates at Lg = ~ 60 nm- goi continues to increase

As Lg↓,

→ gmi/goi ↓

1

Cgsi_areal

20

Delay components of Lg=40 nm InGaAs MHEMT

20

Delay time from ft: ~231 fs• Intrinsic delay: ~81 fs• Extrinsic delay: ~ 99 fs• Parasitic delay: ~50 fs• Unaccounted: ~9 fs

least significant, yields ve=5 x107 cm/s

most significant

42.08%2.52%

33.84%

21.56%

accounted

Transit

ext

par

34 %

42 %

22 %

2 %

21

Scaling of delay components

0 100 2000

200

400

600VDS = 0.6 VVGS - VT = 0.3V

Del

ays

[fs]

Lg [nm]

Transit

ext

par

ext and par do not scale, become dominant for Lg < ~ 60 nm.

100

200

400

600

800

1000

VDS = 0.6 V VDS = 0.5 V VDS = 0.4 V

f T [GH

z]

Lg [nm]

• Intrinsic delay:

Lg ↓ (without degrading gmi), ve ↑ channel engineering

• Extrinsic delay:

Cgsext, Cgdext ↓, or alternatively gmi ↑ (harmonious scaling)

22

Options to improve fT

22

• Parasitic delay:

RS+RD ↓, increase electrostatic integrity: goi/gmi↓

23

100200

400

600

800

1000

1200

VDS = 0.6 V

30% reductionin all the parasitics

Measured fT

Modeled fT

Model Projection

f T [GH

z]

Lg [nm]

1 THz

Model Projection

fT = 1 THz is feasible at Lg = ~ 25 nm.

24

Summary

40-nm In0.7Ga0.3As MHEMTs on GaAs substrate

- RON = 280 Ω-m, gm_max > 2.7 mS/m @ VDS = 0.8 V

- S = 100 mV/dec., DIBL = 105 mV/V

- Measured fT = 688 GHz (Record in any FET)

- fT/fmax = 688/800 GHz (Best-balanced transistor)

Analytical fT Model- Excellent description of fT behavior in III-V HEMTs

- Guidance to improve fT beyond 1 THz