TIME-DOMAIN AC CHARACTERIZATION OF SILICON CARBIDE (SiC) NANOELECTROMECHANICAL SWITCHES TOWARD HIGH-SPEED OPERATIONS

Tina He1†, Vaishnavi Ranganathan1, Rui Yang1, Srihari Rajgopal1,

Swarup Bhunia1, Mehran Mehregany1, and Philip X.-L. Feng1† 1Electrical Engineering, Case Western Reserve University, Cleveland, OH 44106, USA

†Email: ting.he@case.edu; philip.feng@case.edu ABSTRACT

We report an experimental study on AC measurements of contact-mode switches based on silicon carbide (SiC) nanoelectromechanical systems (NEMS). We describe the development of circuits and measurement techniques for recording long cycles of AC switching characteristics of SiC NEMS featured by ultrasmall device movable volumes (at ~1μm3 level) and contact areas (only ~0.01–0.1μm2), and challenging contact resistances (can be from ~10kΩ to ~100MΩ). We perform time-domain AC characterization of SiC NEMS switches with operating speeds up to 1kHz and high on/off current ratios of ~106. For multiple devices, we have recorded the complete time evolution of AC switching data traces of >106 cycles at 1kHz, without failure in ambient air. Beyond these long cycles the devices are still alive, which demands even higher-speed, accelerated AC measurements for long-lifetime recording.

KEYWORDS

Nanoelectromechanical Systems (NEMS), Silicon Carbide (SiC), Switch, AC Measurement, Contact INTRODUCTION

Nanoscale contact reliability and device lifetime are among the most important open challenges in today’s active research on rapidly emerging contact-mode switches enabled by nanoelectromechanical systems (NEMS) [1-5]. Scaled NEMS offer ideally abrupt switching with minimal leakage, and ultra-small footprints. SiC has been proven to be a unique material for its exceptional electromechanical properties and promises for long lifetime in ambient air [6].

High-precision, high-speed time-domain measurement and long-cycle recording are the key to understanding time dependent details of switching dynamics, contact evolution and lifetime limitations. One major factor that limits the lifetime of mechanical switches is the failure mode related to contact degradation [6], in which the contact resistance of the NEMS switches increases with number of switching cycles and eventually appears to be an open circuit.

Nanocontacts degradation depends on the material, contact area and local temperature in the contact region. The actual contact areas are usually very small for NEMS switches; and as current passes through the contact, local temperature increases and affects the contact area and other properties. At high operating speed the time for current passing through the switch, which changes the nanocontact properties in each switching cycle, becomes shorter.

However, time-resolved high-speed measurements of NEMS switches have been particularly challenging, due to the abruptly varying nature of impedance (e.g., from open circuit to ~kΩ) in each switching event. This requires a detection scheme with a wide range of adaptive resistance

matching and very fast, low-noise response, which brings greater complications to the measurement circuit design. The tradeoffs between readout range, precision, and operation speed impose major practical challenges in high-speed AC measurement of NEMS.

SiC NEMS DESIGN AND FABRICATION

The SiC NEMS switches in the work are three- terminal, gate-controlled nanocantilevers with widths (w), thicknesses (t), and gaps (g) within ~200300nm range. Figure 1 shows the design of the switch, and the scanning electron microscope (SEM) image of a typical device. We apply a voltage VG on the gate (G), and the movable cantilever beam (source, S) is clamped to ground. When VG is greater than a threshold voltage Von, the electrostatic force actuates the cantilever beam till the tip makes contact to the drain (D) to create a current path. This threshold voltage Von is called the switch-on voltage. At contact, a bias voltage VD applied on D drives a current from D to S; and the switch is at “on” state. We have also recently explored different designs with split and localized gates for enhancing gate control over device motion [7].

L

S

G

D(a)

5µm(b)

Figure 1: Three-terminal lateral electrostatic SiC NEMS switch. (a) A 3D illustration of the SiC NEMS switch. (b) SEM image of a nanofabricated SiC NEMS switch.

We have developed a surface nanomachining process

for fabrication of these SiC NEMS, from a 500nm-thick polycrystalline SiC (poly-SiC) film grown on 500nm SiO2 on 4-inch Si wafers, enabled by low-pressure chemical vapor deposition (LPCVD). We define and transfer the designed patterns of devices by wafer-scale electron-beam lithography (EBL) and reactive ion etch (RIE), and finally we release the NEMS in a high-yield vapor hydrofluoric (HF) acid etching of SiO2.

MEASUREMENTS AND RESULTS

I-V Characterization in Quasi-DC Measurement To evaluate the basic performance of a SiC NEMS

switch, we first examine the individual switching events by linearly sweeping up and down the gate voltage VG. Figure 2 demonstrates the I-V characteristics of a typical SiC

M4C.004

978-1-4673-5983-2/13/$31.00 ©2013 IEEE 669 Transducers 2013, Barcelona, SPAIN, 16-20 June 2013

NEMS switch measured in such quasi-DC operations. We connect a source measurement unit (SMU) to G to provide the sweeping VG, and another SMU to D for the biasing VD, and S is grounded as shown in Fig. 1a. We sweep VG from 0V up to a voltage level Vact>Von and then back to 0V; and we record the gate current IG and drain current ID simultaneously throughout the sweeping of VG.

Figure 2b & c show plots of two cycles of hysteretic characteristics. We observe clear on and off states in the switch operation, with switch-on voltage Von=23.1V and 23.9V, respectively. The on-state current Ion is larger than 10-6A, the current limit we set in the measurement system. The observed on/off current ratio is Ion/Ioff >106.

To explore the time evolution of nanocontacts and switch lifetime, we first examine multiple switching cycles in real time with high-precision SMUs by applying a continuous square-wave gate voltage with peak amplitude VG,pk>Von, to switch the device on and off for multiple cycles. Figure 2d displays the results of this high-precision measurement that demonstrates the device switching on and off for seven cycles, with a dwelling time of several seconds. The drain biasing voltage is VD=100mV and the on-state current Ion of these cycles is greater than 10-6A. Hence the on-state resistance Ron is less than 100kΩ.

0

10

20

30

Ga

te V

olta

ge,

VG (

V)

0 20 40 60 8010-14

10-12

10-10

10-8

10-6

10-4

Time (s)

VG

ID

Dra

in C

urre

nt,

ID (

A)

0 5 10 15 20 25

0.0

3.0x10-7

6.0x10-7

9.0x10-7

1.2x10-6

ID,1st

IG,1st

ID,2nd

IG,2nd

Cu

rren

t, I D

, I G

(A

)

Gate Voltage, VG (V)

10-16

10-14

10-12

10-10

10-8

10-6

Cu

rre

nts,

ID, I

G

(A)

Von

(a)

(c)

(b)

(d)

Figure 2: Quasi-DC characterization of a typical SiC NEMS switch. (a) Measurement scheme. (b) & (c) Hysteresis I-V curves of two switching cycles; currents are plotted in logarithmic and linear scale, respectively. (d) Quasi-DC measurement of drain current following the square-wave gate voltage, where VG,pk>Von. Magenta trace is gate voltage and blue is drain current.

The first two cycles in Fig. 2d have no switching. Two possible explanations are: (i) the cantilever does not bend to the shape that allows the tip to make good contact with the drain in the first two cycles; or (ii) the tip of the cantilever beam made contact with the drain but the contact resistance was too high for any observable current to flow in the first two cycles, and after two cycles of “breaking in”, the contact resistance becomes lower so the on-state resistance Ron (mainly contact resistance) decreases and on current starts to emerge in the measurement.

In our recent work [6] we have carefully studied the I-V characteristics and lifetime of these switches in such quasi-DC modes, with recorded cycle numbers exceeding ~105106 for typical devices. The switches therein [6] are still alive after these long-cycles tests, though with clearly

noticeable contact degradations (e.g., decreased Ion).

AC Measurement Techniques One of the challenges in high-speed AC testing of the

NEMS switches is that the on-state resistance Ron can be as high as ~100kΩ-100MΩ and it tends to increase with the number of switching cycles. In some experiments, the recorded value of on-state resistance Ron goes up to ~109Ω (Ion~sub-nA) after long switching cycles. In quasi-DC measurements (Fig. 2), we evaluate switch performance by reading out the currents directly using SMUs, in slow measurement, with very high precision. However, SMUs cannot directly measure high-speed switching events. To explore high-speed operations and ultimate lifetime of the switches, we design and fabricate a measurement circuit, as illustrated in Fig. 3, on printed circuit boards (PCBs).

SR560L=8µmw=200nmgGS=300nm

PCB

AC VD

1kΩRb

0.5V DC

DAQ

Ch1 Ch2

Vin0 ~ 1V1kHz

-

+

OpAmp

Buffer

Coax

Gate

Source

Drain

Coax

Coax

0 ~ VG

1kHz Power Amplifier

gDS=300nm

VOut

Figure 3: Measurement scheme for AC characterization of SiC NEMS switches for high-speed operations. The switch connects to a PCB through coaxial probes, and the output is amplified by preamplifier (SR560) with a gain of 10.

The circuit features a high-resistance bleeding resistor Rb as a current-to-voltage converter to read out the on and off state by forming a voltage divider with Ron. We read out this voltage by employing a buffer circuit using a commercial operational amplifier (OpAmp, LF356). The OpAmp has input capacitance of ~3pF and thus reduces the parasitic capacitance effect from 12pF generally limited by measurement instruments such as oscilloscopes. We then amplify the voltage signal Vdiv with a low-noise, band-pass preamplifier (SR560), with the signal band selected at a proper filter range (300Hz to 3kHz band for 1kHz signal). Finally we record the voltage output in real time with a two-channel data acquisition (DAQ) device. The DAQ records the gate control voltage waveform in one channel, and the output voltage Vout in another channel, using the same time reference and sample rate to prevent any phase delay from DAQ (20kHz sample rate for 1kHz signal). The output voltage Vout is given by

Gdivout AVV , (1)

where AG is the gain of the preamplifier and Vdiv is the output voltage of the voltage divider (at Rb),

bonbDdiv RRRVV , (2)

where VD is the drain bias voltage. The background signal (baseline level due to feedthrough) of the AC measurement circuit before the preamplifier stage is ~100mV so the output voltage of the voltage divider Vdiv should be as large as possible for a better signal-to-background ratio. Eq. (2) indicates that Vdiv increases when VD or Rb increase. However, we also like to set the current passing through the NEMS switch to be smaller than 10-6A (a conservative

670

value) to make sure we protect the cantilever beam from any possible failure due to joule heating. This leads to

6bonDD 10 RRVI A. (3)

Therefore, in order to increase the AC detection efficiency (signal-to-background ratio) while limiting the biasing current through the ‘hot’ switch, we choose value of Rb as large as possible. The time constant for the RC circuit is

pbon CRRRC , (4)

where Cp is the parasitic capacitance, which is fixed with the chosen test circuit. The cutoff frequency fC is

pbonC 2121 CRRf . (5)

So the measurement speed of the circuit is mainly limited by the resistance Ron, since Ron dominates (Ron||Rb).

The SiC NEMS switches in this study typically have on-state resistance Ron values in ~kΩ to ~MΩ ranges at the initial state (before any testing-related contact degradation) and often increase with the number of switching cycles in the measurement [6]. Given the above discussions we choose the bleeding resistance to be ~5–10 times of Ron. For example, with a NEMS switch with Ron=100MΩ, we choose Rb=1GΩ, and VD=2V. This will give us an on-state current Ion~2nA, and Vdiv~2V. With Cp=3pF, the cutoff frequency is fC=530Hz. When Ron increases during the experiment, both the amplitude of the output voltage and the cutoff frequency of the RC circuit will decrease.

PSpice Simulations

Figure 4 shows the PSpice simulation with the gate voltage operating at 1kHz, with Ron=1MΩ, 10 MΩ and 100 MΩ, respectively. Figure 4a illustrates the equivalent circuit model used in the PSpice simulation, where the NEMS switch is simply modeled as a resistor Ron at its ‘on’ state (ignoring the intrinsic transient that is much faster than the operating cycle here) with a square-wave voltage source, with peak amplitude same as the drain voltage VD

applied to the NEMS switch. Rb is the bleeding resistor and Cp is the equivalent parasitic capacitor. We choose Cp to be 3pF, which is the input capacitance of the OpAmp (LF356) in the circuit, because this is the main contributor to the parasitic capacitance.

Simulation results in Fig. 4bd demonstrate the effect of increasing Ron and Rb on the operation frequency. When Ron increases, Rb has to increase accordingly to have a readable voltage output, but this causes accuracy decrease at the same frequency. The output voltage follows the gate voltage at 1kHz with Ron=1MΩ and shows a phase delay and some distortion at Ron=10MΩ, and finally severe

distortion and delay due to the charging/discharging effects for the case of Ron=100MΩ.

0.0

0.4

0.8

1.2

1.6

Gat

e V

olta

ge, V

G (

V)

0 2 4 6 8 10

0.0

0.4

0.8

1.2 VOut V

G

Time (ms)

Ou

tpu

t V

olta

ge

, V

Out (

V)

0.0

0.2

0.4

0.6

Gat

e V

olta

ge, V

G (

V)

0 2 4 6 8 10

0.0

0.2

0.4

0.6

0.8 V

Out V

G

Ou

tpu

t V

olta

ge, V

Out (

V)

0.0

0.2

0.4

0.6

Gat

e V

olta

ge,

VG (

V)

0 2 4 6 8 10

0.0

0.2

0.4

0.6

0.8 V

Out V

G

Time (ms)

Out

put

Vol

tag

e, V

Out (

V)

(a)

(d)

(c)

(b) Figure 4: PSpice simulations of the AC characterization system. (a) Schematic of the equivalent circuit. Simulation results with Ron, Rb & Cp of (b) 1MΩ, 1MΩ, 3pF; (c) 10MΩ, 1GΩ, 3pF; and (d) 100MΩ, 1GΩ, 3pF; respectively. AC Measurement Results & Discussions

With measurement scheme and circuits proposed in Fig. 3, we measure NEMS switches at different frequencies. The PSpice simulation results indicate that with Ron=100kΩ and Rb = 1MΩ, it should have a safe margin for a measurable output voltage signal at 1kHz. For device shown in Fig. 2a, we previously demonstrated Ron <100kΩ. Figure 5 shows over a million (>106) switching cycles at 1kHz recorded in real time, in ambient air. The change in amplitude of output voltage over these long cycles indicates a varying on-state resistance Ron. We have found the lowest Ron after several tens of thousands switching events, suggesting a “breaking in” process of nanocontacts.

Figure 6 illustrates the details from Fig. 5. The large on-state resistance in Fig. 6a results in a distorted output voltage Vout waveform with a phase delay with respect to the VG waveform. Since Ron is large compared to 1/10 of Rb, the voltage divider formed by these two resisters does not output the maximum voltage. Note that the change in voltage amplitude also indicates that Ron is different for every cycle. Comparing the output voltage Vout waveforms with control experiment results in Fig. 6c, we confirm the electrostatically actuated nanomechanical switching.

The output voltage Vout waveform in Fig. 6b follows that of the VG and reaches the maximum range of the voltage divider output, exhibiting a uniform performance from cycle to cycle. This is because in those cycles Ron is very small so the charging/discharging effects from the RC

(A) (B)

Figure 5: Real-time recorded time-domain evolution of switching events of over 1 million (>106) cycles from one SiC NEMS. The output voltage Vout follows the gate voltage VG; drain bias voltage is VD=0.5V and amplifier gain is 10.

671

circuit is minimal (among these cases) and the circuit is able to read out the full range of the voltage.

0

5

10

15

20

25

30

VOut

Gat

e V

olta

ge,

VG (

V)

2.001 2.004 2.007 2.010-4

-2

0

2

4 VG

Time (s)

Ou

tpu

t V

olta

ge, V

Out (

V)

0

5

10

15

20

25

30 V

G V

Out

Gat

e V

olta

ge, V

G (

V)

820.000 820.004 820.008

-2

0

2

4

Time (s)

Out

put

Vo

ltag

e, V

Out (

V)

0 5 10 15 20

10-14

10-12

10-10

10-8

10-6

Gate Voltage, VG (V)

Cu

rre

nt, I

D, I

G (

A)

ID

IG

0

5

10

15

20

25

30 V

Out

Ga

te V

olta

ge,

VG (

V)

40.002 40.005 40.008-4

-2

0

2

4 VG

Time (s)

Ou

tpu

t V

olta

ge,

VO

ut (

V) (a)

(d)

(b)

(c)

Figure 6: Representative time-domain data from AC measurements. (a) & (b) Data from box (A) & (B) in Fig. 5, respectively. (c) Control experiment showing no NEMS switching when VG,pk<Von (‘sub-threshold’). (d) Hysteresis I-V curves taken after 106

switching cycles confirms that the device is still alive with on/off ratio of ~106 or larger.

The nanoscale contact may start to degrade after over

106 cycles but the device is still alive. High-precision quasi- DC measurements in Fig. 6d confirm that the device is still switching, without obvious decrease in the on-state current level after all these cycles. We then continue actuating this device at 1kHz for another 4 million cycles, and the Vout exhibits a degraded waveform; but the I-V curves measured in quasi-DC tests after the 4 million cycles once again confirm that the device is still alive, albeit with degraded contact and lower on-state current Ion.

0

5

10

15

20

25

Out

put V

olta

ge,

VO

ut (

V)

Gat

e V

olta

ge

, VG (

V)

VG

3.26 3.28 3.30 3.32 3.34 3.36-0.2

-0.1

0.0

0.1

0.2

VOut

Time (s)

-5

0

5

10

15

20

25

30

VOut

Out

put V

olta

ge,

VO

ut (

V)

Gat

e V

olta

ge,

VG (

V)

VG

1.08 1.09 1.10 1.11 1.12

-1

0

1

2

3

Time (s)

-5

0

5

10

15

20

25

30

VOut

Time (s)

Ou

tput

Vol

tage

, VO

ut (

V)

Ga

te V

olta

ge, V

G (

V)

VG

0.900 0.902 0.904 0.906 0.908 0.910-1.0

-0.5

0.0

0.5

1.0

1.5

1.08 1.09 1.10 1.11 1.12

-1

0

1

2

3

Time (s)

Ou

tput

Vol

tage

, VO

ut (

V)

-5

0

5

10

15

20

25

30

Gat

e V

olta

ge,

VG (

V)

VG

VOut

(a)

(d)

(b)

(c) Figure 7: Time-domain voltage curves from AC measurement of another SiC NEMS switch. Data taken at (a) 100Hz, and (b) 300Hz and (d) 1kHz, respectively, with VG,pk>Von. (c) Control experiments at 300Hz showing no NEMS switching when VG,pk<Von (‘sub-threshold’).

We have also carefully measured a number of devices,

and compared switching data from typical SiC NEMS switches, at 10Hz, 100Hz, 300Hz, 1kHz and higher, by engineering the circuit in Fig. 3. Figure 7 shows the AC characteristics measured from one other NEMS switch. Hysteresis I-V measurement shows that this device switches on at Von=21.9V and the initial on-state resistance is Ron=10MΩ. We have performed AC measurements at

100Hz, 300Hz and 1kHz, respectively, with the same bleeding resistance Rb=1GΩ. Measured data demonstrate that the output voltage distorts in both amplitude and phase as operation speed increases. Control experiment at 300Hz in Fig. 7c confirms that no mechanical switching happens with VG,pk=10V<Von. Further AC measurement of this device shows that the device turns on and off in ambient air for >105 cycles at 100Hz, and DC measurements later have confirmed that the device is still alive and functional, with an on-state resistance increased to Ron=2GΩ.

CONCLUSIONS

We have investigated AC electronic measurement techniques for operating and accelerating the tests of long-lifetime NEMS switches at high speeds up to 1kHz. Combining experiments and circuit simulations, we have examined the effects of circuit parameters and parasitic effects, while exploring the technical limits toward approaching both high-speed and high-precision tests that can efficiently approach the intrinsic lifetime of the robust SiC NEMS switches. The available operating speeds and switching characteristics (e.g., ‘on’ & ‘off’ states, transient) of these SiC NEMS switches are limited by extrinsic, practical issues in the measurement circuits. To approach even higher-speed operations, the readout circuits should be highly efficient, fast, sensitive, and adaptive to capture the device intrinsic speed and fast-varying characteristics.

ACKNOWLEDGEMENTS

We thank DARPA (NEMS Program, D11AP00292), NSF (CCF-1116102), & Keithley Fellowship for support. REFERENCES [1] T. Rueckes, K. Kim, E. Joselevich, G. Y. Tseng, C.-L.

Cheung, C. M. Lieber, “Carbon nanotube-based nonvolatile random access memory for molecular computing,” Science, vol. 289, pp. 94-97, Jul. 2000.

[2] M. L. Roukes, "Mechanical computation, redux?” in Tech. Digest, IEEE Int. Electron Devices Meeting (IEDM 2004), pp. 539-542, San Francisco, CA, Dec. 13-15, 2004.

[3] X. L. Feng, M. H. Matheny, C. A. Zorman, M. Mehregany, M. L. Roukes, “Low voltage nanoelectromechanical switches based on silicon carbide nanowires,” Nano Letters, vol. 10, pp. 2891-2896, Aug. 2010.

[4] T.-H. Lee, S. Bhunia, M. Mehregany, "Electromechanical computing at 500°C with SiC," Science, vol. 329, pp. 1316-1318, Sep. 2010.

[5] R. Yang, T. He, C. Marcoux, P. Andreucci, L. Duraffourg, P. X.-L. Feng, “Silicon nanowire and cantilever electromechanical switches with integrated piezoresistive transducers,” in Proc. 26th IEEE Int. Conf. on Micro Electro Mechanical Systems (MEMS 2013), pp. 229-232, Taipei, Taiwan, Jan. 20-24, 2013.

[6] T. He, R. Yang, S. Rajgopal, M. A. Tupta, S. Bhunia, M. Mehregany, P. X.-L. Feng, “Robust silicon carbide (SiC) nanoelectromechanical switches with long cycles in ambient and high temperature conditions,” in Proc. 26th IEEE Int. Conf. on Micro Electro Mechanical Systems (MEMS 2013), pp. 516-519, Taipei, Taiwan, Jan. 20-24, 2013.

[7] T. He, R. Yang, S. Rajgopal, S. Bhunia, M. Mehregany, P. X.-L. Feng, “Dual-gate silicon carbide (SiC) lateral nanoelectromechanical switches”, in Proc. 8th IEEE Int. Conf. on Nano/Micro Engineered & Molecular Systems (NEMS 2013), pp. xxx-xxx, Suzhou, China, Apr. 7-10, 2013.

672