Enabling CNTFET-based analog high-frequency circuitdesign with CCAM

Martin Claus1,2, Anıbal Pacheco2, Max Haferlach2, Michael Schroter2

1 Center for Advancing Electronics Dresden2Chair for Electron Devices and Integrated Circuits

Technische Universitat Dresden, Germany

MOS-AK, Graz, Austria, 18.09.2015

CNTFET technology statusfor analog HF applications1

1M. Schroter, M. Claus, et al., ”Carbon nanotube FET technology for radio-frequency electronics:State-of-the-art overview (invited)”, IEEE Journal of the Electron Devices Society, 1(1), pp. 9–20,2013.

2 / 23

CNTFET technology overview

Multi-tube CNTFETs

high current, high power application(1000–3000 parallel tubes)

scale with tube density, finger number andwidth to desired applications

relaxed constraints for technology(800 nm channel length)

parasitic metallic tubes in the channel(20%-30%)

first prototyp technologies available(fT,peak ≈ 10 GHz, Gpower > 10 dB)

3 / 23

Status of HF CNTFET technology I

Single-tube CNTFET

HF CNTFET in GSG configuration

Multi-tube Multi-finger CNTFET

100 mm wafer

4 / 23

Status of HF CNTFET technology II

0

5

10

15

20I d(µA)

-3 -2 -1 0 1 2 3

Vgs (V)

1.8 V

1.1 V

0.75 V

0.4 V

Single tube transfer characteristic

0

5

10

15

20

25

30

I d(µA)

0.0 0.5 1.0 1.5 2.0 2.5

Vds (V)

1.0V

0.5V

0V

−0.5V

−1.0V

−1.5V

Single tube output characteristic

0

10

20

30

40

50

60

I d(mA)

-1 0 1 2 3

Vgs (V)

2V1V0.5V0.25 V

Multi tube transfer characteristic

0

10

20

30

40

50

60

70

I d(mA)

0.0 0.5 1.0 1.5 2.0 2.5

Vds (V)

3 V

2 V

1 V

0 V

Multi tube output characteristic

5 / 23

Status of HF CNTFET technology III

0

2

4

6

8

10f T,extr(GHz)

0 1 2 3 4

Vgs (V)

2V1V0.5V0.25V

Transit frequency of HF CNTFET

0

2

4

6

8

10

f max,extr(GHz)

0 1 2 3 4

Vgs (V)

2V1V0.5V0.25V

Maximum oscillation frequency

-12

-8

-4

0

4

8

12

MAG(dB)

0 1 2 3 4

Vgs (V)

2V1V0.5V0.25V

Maximum available gain

0.0

0.5

1.0

1.5

2.0

2.5

Av

0 1 2 3 4

Vgs (V)

2V1V0.5V0.25V

Intrinsic voltage gain

6 / 23

Circuit results - L-band RF amplifier

First CNT-based single-stage L-band RF amplifier2

11 dB linear gain with 10 dB input/output return loss at 1.3 GHz

-30

-15

0

15

|S|(dB)

0.5 1.0 1.5 2.0f (GHz)

meas

sim

S21

S11

S22

Good comparison between experimental results and model

2M. Eron, S. Lin, D. Wang, M. Schroter, P. Kempf, ”An L-band carbon nanotube transistor

amplifier”, Electronics Letters, vol. 47, no. 4, pp. 265-266, 2012.

7 / 23

CCAM – A compact model for HF CNTFETs3,4

3M. Claus, ..., M. Schroter, ”Critical review of cntfet compact models”, in NSTI-Nanotech (Work-shop on Compact modeling), Vol. 2, 2012.

4M. Schroter, ..., M. Claus, ”A semi-physical large-signal compact carbon nanotube fet model

for analog rf applications”, IEEE Transactions on Electron Devices, Vol. 62(1), 2015.

8 / 23

Compact models for HF CNTFETs I

State-of-the-art of CNTFET compact models

main focus on digital applications (“beyond CMOS”)→ nanoscale channel lengths

models mostly restricted to single-tube CNTFETs and low voltages

formulations focus mostly on describing DC behavior

almost no experimental verification of model formulations

→ little emphasis on multi-tube high-frequency (HF) analog applications

9 / 23

Compact models for HF CNTFETs II

CM for MT CNTFETs includes: equivalent circuit for semiconductingtubes + metallic tubes + parasitic elements

Multi-tube CNTFET

RDcs

QtD

QtS

Isem

RScs

RDcm

CDmt

CSmtImet

RScm

RDf

CGDp2

CGSp2

RSf

CDSp

CGDp1

RG

CGSp1

D

S

G

Equivalent circuit

10 / 23

Compact modeling issues I

Trap modeling

In wafer-scale processes it is still challenging to get devices free oftraps.

For early applications: compact models for circuit design needed withwhich the trap-affected circuit behavior can be predicted

Trap model can help to define measurement conditions tocharacterize trap-free device behavior which is needed for technologyevaluation and modeling purposes

Model helps to understand experimental observation such as theapparent linearity of CNTFETs

11 / 23

Compact modeling issues II

All fabricated transistors haveSchottky-like barriers (SB) betweenmetal contacts and CNT

→ compact modeling very difficult

→ no feasible physics-based approach(for current and charge) is known

→ almost all existing compact modelsdo not consider SB properly(compared to experiments)

Two parallel approaches in our group:semi-physics based (CCAM) and

physics-based (TCAM) compact model

Ec,s

Ef,s

Ec,d

Ef,d

source channel drain

12 / 23

Compact model: CCAM

13 / 23

Compact model: CCAM

RDcs

QtD

QtS

Isem

RScs

RDcm

CDmt

CSmtImet

RScm

RDf

CGDp2

CGSp2

RSf

CDSp

CGDp1

RG

CGSp1

D

S

G

Equivalent circuit

CCAM Features

bias-dependent formulation for internalelements (i. e. large signal model)

temperature and geometry dependencefor all equivalent circuit elements

access to technology parameterse. g. fraction of metallic tubes

noise and trap model

CCAM has been implemented in Matlab and Verilog-A, making itwidely available across circuit simulators5

5M. Schroter et al., CCAM Compact Carbon Nanotube Field-Effect Transistor Model,nanoHUB, doi:10.4231 / D34F1MK28, 2015.

14 / 23

CCAM equations (not showing all)

Drain current:Isem = IDS0fGSfDS

GS dependence:

fGS =

uGS +√

u2gs + athg

1 +√

1 + athg

21 + 21 + uGS√u2GS + athg

with uGS = 1− Vthg0/vgt, vgt = VGS − Vfb

DS dependece (simple form for scattering):

fDS = uDS

(1 + |uDS|β

)−1/β

Similar smoothing functions for the charge

15 / 23

Experimental verification I

0

5

10

15

20I d(µA)

-3 -2 -1 0 1 2 3

Vgs (V)

1.8 V

1.1 V

0.75 V

0.4 V

Single tube transfer characteristic

0

5

10

15

20

25

30

I d(µA)

0.0 0.5 1.0 1.5 2.0 2.5

Vds (V)

1.0V

0.5V

0V

−0.5V

−1.0V

−1.5V

Single tube output characteristic

0

10

20

30

40

50

60

I d(mA)

-1 0 1 2 3

Vgs (V)

2V1V0.5V0.25 V

modelexp.

Multi tube transfer characteristic

0

10

20

30

40

50

60

70

I d(mA)

0.0 0.5 1.0 1.5 2.0 2.5

Vds (V)

3 V2 V1 V0 V

modelexp.

Multi tube output characteristic

16 / 23

Experimental verification II

0

2

4

6

8

10f T,extr(GHz)

0 1 2 3 4

Vgs (V)

2V1V0.5V0.25V

modelexp.

Transit frequency of HF CNTFET

0

2

4

6

8

10

f max,extr(GHz)

0 1 2 3 4

Vgs (V)

2V1V0.5V0.25V

modelexp.

Maximum oscillation frequency

0

5

10

15

20

25

30

gm,peak(mS)

0 200 400 600 800 1000

wgf (µm)

exp.

model

Scaling of peak gm with gate width

0

2

4

6

8

10

f T,peak(GHz)

0 200 400 600 800 1000

wgf (µm)

exp.

model

Scaling of peak fT with gate width

17 / 23

Modeling of trap effects

Empirical trap model included in CCAM6

Electron capturing in traps and the resulted tube shielding is modeledas a threshold voltage shift Id = f (VGS − Vtr )

Dynamics of capture and emission modeled with RC network

ItrC0

R0

C1

R1

Cn

Rn

Vtr

Empirical model for trap current Itr = αVGS + βVds + γ fitted to stepresponse measurements

Model parameters of intrinsic part adjusted to pulsed measurements

6M. Haferlach M. Claus, A.Pacheco, et al., Nanotech, Workshop on Compact Modeling (WCM),

2014.

18 / 23

Comparison with experimental data

Non-pulsed mode: chargesare trapped and shield tubepotential from the externalvoltages

for high VGS and VDS

tube potential and currentstay almost constant

Pulsed mode: measurementcycles too fast for trappingprocesses

tube potential directlyfollows external voltages

0

20

40

60

80

I d(mA)

-1 0 1 2Vgs (V)

non-pulsed

0.25V

1V

3V

0 1 2 3Vgs (V)

pulsed

Transfer characteristicssymbols – exp. results, lines – model

→ CM predicts non-pulsed and pulsed behavior(with one single parameter set for non-pulsed and pulsed mode)

19 / 23

Benchmark circuit design studies

20 / 23

Circuit results - Power amplifier7

Class-A power amplifier designed at Vgs = 0.5 V (low saturationvoltage) and Vds = 2 V for 2 GHz applications

150 similar devices are connected in parallel to have an output powerof 16 dBm

vin

C1 R1

VGG

L1

T1

Vgs

C2

R2C3

Vdsvout

L2

VDD

PA circuit with matching andstabilization subcircuits

0

2

4

6

8

I d(A)

0 1 2 3 4

Vds (V)

20%

10%

0%

Output characteristicfor various mfrac

-80

-60

-40

-20

0

20

Pout(dBm)

-40 -30 -20 -10 0 10 20

Pin (dBm)

0%

10%

20%

Output power vs inputpower for various mfrac

Power gain only for less than 10 % metallic tube fraction

7M. Claus, et al., ”High-frequency benchmark circuit design for a sub 50 nm cntfet technology”,

IMOC 201321 / 23

Circuit results - L-band RF amplifier2

First CNT-based single-stage L-band RF amplifier

11 dB linear gain with 10 dB input/output return loss at 1.3 GHz

-30

-15

0

15

|S|(dB)

0.5 1.0 1.5 2.0f (GHz)

meas

sim

S21

S11

S22

Good comparison between experimental results and model4

2M. Eron, ..., M. Schroter, ”An L-band carbon nanotube transistor amplifier”, ElectronicsLetters, Vol. 47(4), 2012.

4M. Schroter, ..., M. Claus, ”A semi-physical large-signal compact carbon nanotube fet model

for analog rf applications”, IEEE Transactions on Electron Devices, Vol. 62(1), 2015.

22 / 23

Conclusions

CNTFET technology is suitable for HF applications.

CCAM shows an excellent agreement with DC as well as with biasand frequency dependent AC data of fabricated SB CNTFETs

Trap model included in CCAM to predict the impact of traps oncircuit behavior

CCAM predicts non-pulsed and pulsed behaviorTemperature dependence to be published soon

CNTFET circuit design is ongoing

CCAM is used to optimization and projectionDiscrete circuit design by means of the CCAM model

CCAM available at nanoHUB (doi:10.4231 / D34F1MK28, 2015)

23 / 23