Overview of CMOS Device Behavior and Modeling for Mixed ... · PDF fileAdvanced Device...
Transcript of Overview of CMOS Device Behavior and Modeling for Mixed ... · PDF fileAdvanced Device...
Overview of CMOS Device Behavior and Modeling
for Mixed-Signal/RF Circuit Design
Yuhua Cheng
Siliconlinx, Inc.Bridging the gap between IC designers and foundries
2 Yuhua Cheng
________________________________Outline
• CMOS Technology Trends
• Device Behavior Overview
• Device Modeling Challenges
• Figures of Merit and Model Validation
• Summary
3 Yuhua Cheng
________________________________• CMOS has been in nanoscale era.
• Silicon CMOS is still the mainstream IC technology in the next 7-10 years before othernano devices play roles.
1990 1995 2000 2005 20100.01
0.1
1
22nm
Micro-scale Technology Regime
Nano-scale Technology Regime30nm
45nm65nm
90nm0.13µm
0.18µm0.25µm
0.35µm0.5µm
Year
Mic
rom
eter
10
100
1000
Nanometer
Source: Intel
CMOS Technology Trends -- Nano-scale
4 Yuhua Cheng
________________________________Semiconductor Industry (SI) TrendPC -> Communication (wireless)
Intel Inside Communication Everywhere
• The data of voice, video and other information will be exchanged through any media at any time and any places
by wireless devices!
5 Yuhua Cheng
________________________________SI Trend (CONT.)IC -> Application Integration
– The revenue ratio of the system chip to overall semiconductor industry will increase to 21.9% (with a total revenue of 65.6 billions) in 2008 from 16.6% in 2002
– Product integration is to integrate different products such smart phone, camera, cell phone, PDA, TV, music player, based on one platform such as cell phone.
– The integration of end market will be the merge of markets such as computer, consumer, wired and wireless communications.
Power CPU
WLAN
3D Graphics
GPS
PDA
MovieTVVideo
Integration of Chips Integration of Applications (End Market)
6 Yuhua Cheng
________________________________CMOS Technology Trends -- RF CMOS• Recent speed improvements and better noise behavior of
CMOS transistors have made it feasible to implement RF circuits for wireless products, such as cell phones, Global Positioning System, and Bluetooth.
• CMOS is one of the best suitable technologies to ingrate RF circuits with analog and base-band digital circuits.
• Tremendous research effort is dedicated to RF CMOS, driven by System-on-Chip (SoC).
CMOS:High integration densityHigh yieldLow power consumptionLow cost
7 Yuhua Cheng
________________________________An Illustration of RF SoC on CMOS
K. Muhammad et al. “Digital RF Processing: Toward Low-Cost Reconfigurable Radios,” IEEE Communication Magazine, Vol. 43, pp. 105-113, 2005.
• RF SoC - Integrate a chipset system onto a same chip:
• RF Front End (RFFE)• Analog Based-band (ABB)• Power Management and Audio Codecs (PMAC) • Digital Base-band (DBB)
• RFCMOS implementation challenges:
• low cost• high performance
8 Yuhua Cheng
________________________________
Need to Well Understand the Device Behavior in Advanced Technology
Nodes
Need to Develop Physical and Accurate Device
Models for RF/Analog Design
Design IC in Nano-scale/RF Era
9 Yuhua Cheng
________________________________MOSFET High Frequency Behavior
0.1 110
100
0.18um
0.35um
0.13um
90nm
0.25um
0.25um
0.5um
0.18um
0.12um
0.35um
90nm
Channel Length (µm)
Cut
off F
requ
ency
(GHz
)
0.1
1
NFmin (dB)
• Today’s MOSFETs show high Ft and low NF
(Yuhua Cheng, “Challenges in Bridging Manufacturing and Design for RF applications-An Overview of Advanced Device Modeling for RF Circuit Design”, RFIC/IMS workshop, June 2004)
10 Yuhua Cheng
________________________________Leakage in Different Technology Node
• Leakage increases significantly as technology advances.
0 20 40 60 80 100 120 140 16010-1
100
101
102
103
104
0.25µm
0.18µm
0.13µm
90nm
J off (n
A/µm
2 )
Temperature (oC)(Yuhua Cheng, “Challenges in Bridging Manufacturing and Design for RF applications-An Overview of
Advanced Device Modeling for RF Circuit Design”, RFIC/IMS workshop, June 2004)
11 Yuhua Cheng
________________________________MOSFET Gate Leakage
0.0 0.5 1.0 1.5 2.0 2.5 3.010-910-810-710-610-510-410-310-210-1100101102103104
Tox=2.5nm
Tox=2.0nm
Tox=1.4nm
VDS=0V
J gat
e (A
/cm
2 )
Vgate (V)
• In today’s MOSFETs, gate leakage increases by orders for the decrease of Tox
(Yuhua Cheng, “Challenges in Bridging Manufacturing and Design for RF applications-An Overview of Advanced Device Modeling for RF Circuit Design”, RFIC/IMS workshop, June 2004)
12 Yuhua Cheng
________________________________
100 200 300 400 500 600 700 800 90010-12
10-11
10-10
10-9
10-8
10-7
10-6
10-5
NFETPFETIo
ff (A
)
IDsat (mA)
IDsat/Ioff Universal Curves
• A trade-off between IDsat and Ioff.• Better device design and material selection to reduce the slope
of the universal curves.
Better Device Design
13 Yuhua Cheng
________________________________Layout Dependent Device Performance
• Effects different to nfet and pfet.• Impact both digital (changing drive current) and
analog (changing both Gm and matching).
(C. H. Diaz et al., “Process and circuit design interlock for application-dependent scaling tradeoffs and optimization in the SoC era,”, IEEE J. of Solid-State Circuits, Vol. 38 , Pages:444–449, 2003)
14 Yuhua Cheng
________________________________Layout Dependent Device Performance (CONT.)
• Irregular device layouts are used in digital (even analog) cell libraries.
• The layouts of the devices in the cell libraries may be different from those for device modeling
(C. H. Diaz et al., “Process and circuit design interlock for application-dependent scaling tradeoffs and optimization in the SoC era,”, IEEE J. of Solid-State Circuits, Vol. 38 , Pages:444–449, 2003)
15 Yuhua Cheng
________________________________HF Behavior – Parasitic Effects
trench trench trenchtrench
SB DB
RSB RDB
RDSB
RDS
RS RD
G
RG
DSB DDB
CGSO CGDO
N+ N+ P+P+
P-SUB
N- N-
• Parasitic components need to be well understood at HF.
16 Yuhua Cheng
________________________________HF Behavior: Effective Rg• Effective RG extracted by:
• Effective RG does not follow equation below when Lf is long:
• Significant increase in effective RG due to non-quasi-static (NQS) effect.
)(,α
+=f
extff
GshpolyG
WWLN
RR
}Im{}Im{}Re{
1211
12
YYYRG =
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.60
10
20
30
40
50
~LF/W
F~W
F/L
F
VGS
=0.8V
1.0V
1.8V
VDS
=1.0V
VBS
=0V
WF=6µm
NF=10
Frequency=2GHz
Gat
e R
esis
tanc
e (O
hm)
Ldrawn
(µm)
(Yuhua Cheng et al., “High frequency characterization of gate resistance in RF MOSFETs”, IEEE Electron Device Letters , Vol. 22, pp. 98-100, 2001.)
17 Yuhua Cheng
________________________________HF Behavior: NQS Effects (1)• Effective Gm, normalized:
• Short L, Gm is “constant” over f.
• Longer L, Gm becomes frequency dependent;
• The longer the L, the stronger the frequency dependence.
)Re()Re(
)0(21
21,
fnormalizedm
YYG =
0 2 4 6 8 10 12 140.0
0.2
0.4
0.6
0.8
1.0
1.2
L=1.35 µm
L=0.85µm
L=0.35 µm
WF=15 µm
NF =10
VGS
=1.8V
VDS
=1.5V
VBS
=0V
Rea
l(Y21
)/Rea
l(Y21
)(f0)
Frequency (GHz)
(Yuhua Cheng et al., “Frequency-dependent resistive and capacitive components in RF MOSFETs ,” IEEE Electron Device Letters , Vol. 22, pp. 333 –335, July 2001.)
18 Yuhua Cheng
________________________________HF Behavior: NQS Effects (2)• Effective Gate Capacitance
Cgg,eff:
• Short L, Cgg,eff is “constant”over f.
• Longer L, Cgg,eff becomes frequency dependent.
• The longer the L, the stronger the frequency dependence.
ω=
}Im{ 11,
YC effGG
• (Yuhua Cheng et al., “Frequency-dependent resistive and capacitive components in RF MOSFETs ,” IEEE Electron Device Letters , Vol. 22, pp. 333 –335, July 2001.)
0 2 4 6 8 10 120.002
0.003
0.004
0.005
L=1.35 µm
L=0.85 µm
L=0.35µm
WF=15µmNF=10
VGS=1.8VVDS=1.5VVBS=0V
CG
G (F
/m2 )
Frequency (GHz)
19 Yuhua Cheng
________________________________HF Behavior: Substrate Coupling
trench trench trenchtrench
SB DB
Rsb Rdb
Rdsb
G
Dsb Ddb
P-sub
Signal at the drain coupling to the nearby source diffusions and to the substrate terminal through the junction capacitance and substrate resistance
• At HF, signal is coupled to substrate through junction capacitance.
• This effect impact mainly the output impedance.
20 Yuhua Cheng
________________________________HF Behavior: Bias and f Dependence of Rsub
0 2 4 6 8 10 120
50
100
150
VGS
=0.0V, 0.2V, 0.4V, 0.6V
Wdrawn
=15µm
Ldrawn
=0.35µm
VDS
=0V
VBS
=0V
Rsu
b (Ohm
)
Frequency (GHz)0 2 4 6 8 10 12
0
50
100
150
VDS
=0.0V, 0.5V, 1.0V, 1.5V, 2.0V
Wdrawn
=15µm
Ldrawn
=0.35µm
VGS
=0V
VBS
=0V
Rsu
b (O
hm)
Frequency (GHz)
• Substrate resistance is a very weak function of biases at the gate and drain.
• Up to 10GHz, substrate resistance is not sensitive to frequency.
(Yuhua Cheng et al., “On the high-frequency characteristics of substrate resistance in RF MOSFETs ” IEEE Electron Device Letters , Vol. 21, Page(s): 604 –606, 2000 )
21 Yuhua Cheng
________________________________HF Behavior: Thermal Noise
• HF noise reduces as the technology advances (and shorter L).
• Higher Gm, lower HF noise.• Low Nf (<1dB) can be obtained at normal gate biases.
(Jamal Deen, Chih-Hung Chen; Yuhua Cheng; “MOSFET modeling for low noise, RF circuit design” Proceedings of the IEEE 2002 Custom Integrated Circuits Conference, pp. 201 – 208, May 2002. )
22 Yuhua Cheng
________________________________HF Behavior: Induced Gate Noise
• Channel noise is frequency independent and induced gate noise is frequency dependent.
• Induced gate noise is not negligible in devices with long L or devices with short L but at very high frequency.
(Jamal Deen, Chih-Hung Chen; Yuhua Cheng; “MOSFET modeling for low noise, RF circuit design” Proceedings of the IEEE 2002 Custom Integrated Circuits Conference, pp. 201 – 208, May 2002. )
23 Yuhua Cheng
________________________________HF Behavior: High “Low Frequency Limit”
(Tzung-Yin Lee, and Yuhua Cheng, “High-Frequency Characterization and Modeling of Distortion Behavior of MOSFETs for RF IC Design”, IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 39, NO. 9, pp. 1407 – 1414, 2004.)
0.0001 0.001 0.01 0.1-100
-80
-60
-40
-20
0
W=12µmL=0.36µmFinger=10
P3
P2
P1V
DS=0.5V
Pin=-10dBm
Solid lines: fin=50MHz
Dashed lines: fin=100MHz
Square : fin=900MHz
+: fin=1.8GHz
Pou
t (dB
)
IDS
(A)
• MOSFET has much higher “low frequency limit” (LFL)• HF distortion characteristics (<fLFL) can be described
by its low-frequency behavior
24 Yuhua Cheng
________________________________Device Modeling for RF Applications
• Device Modeling at HF should at least include the modeling for:– MOSFET– Passive devices (R, C, inductor,varactors)– Special devices (LDMOS and PNP BJTs)– Interconnect– Substrate– Circuit blocks (behavior modeling)
• Today, we will mainly talk about MOSFET modeling.
25 Yuhua Cheng
________________________________MOSFET Modeling Challenges• Core model with good continuity, accuracy and scalability over
wide biases, temperatures and geometries.
• Modeling of classic well-known short channel and narrow width effects and recent advanced physical effects such as velocity overshoot, self-heating, channel charge quantization, tunneling, layout dependent behavior, …
• Scalable parasitic capacitance and resistance models.
• Predicts accurately the small signal AC and noise and the distortion behavior.
• Non-quasi static (NQS) effects.
• Statistical modeling
26 Yuhua Cheng
________________________________Modeling Challenges – Gm & Gm/ID
• Gm, the most important parameters in analog design.• Gm/ID, a universal characteristic of MOSFETs to evaluate the
transconductance efficiency.• Inversion Coefficient (IC), ratio of ID/IS, is proposed as a
measure of MOS inversion level.
10-2 10-1 100 101 102 103-5
0
5
10
15
20
25Weak Inversion
Moderate Inversion
Strong Inversion
W=10µm
GM
(mS)
L=0.35µm
L=0.13µm
IC
-5
0
5
10
15
20
25
GM
/ID (1
/V)
(Yuhua Cheng, "A study of figures of merit for the high frequency behavior of MOSFETs in RF IC applications", Eighth International Conference on modeling and Simulation of Microsystems, Anaheim, May 8-12, 2005)
27 Yuhua Cheng
________________________________Modeling Challenges - Capacitance
• Gate Capacitance is not constant in strong inversion. • Bias dependence is caused by Poly-depletion effect.• Both poly-depletion (PD) and channel quantization
(CQ) effects will impact Cgg.
-3 -2 -1 0 1 2 30.0
0.1
0.2
0.3
0.4
0.5
0.6
Lines: Simulated resultsSymbols: Measured data
With PD and CQ EffectsWith PD Effect No PD and CQ Effects
Gat
e C
apac
itanc
e (p
F)
Gate Voltage (V)(Yuhua Cheng et al.,” ICM-an analytical inversion charge model for accurate modeling of thin gate oxide MOSFETs,”
Simulation of semiconductor Processes and Devices,Page(s) 109 –112, 1997.)
28 Yuhua Cheng
________________________________Modeling Challenges – Scalable Rsub
Signal at the drain coupling to the nearby source diffusions and to the substrate terminal through the junction capacitance and substrate resistance
B B
Rsb Rdb
Csb Cdb
Si Di
Rdsb
(Intrinsic source) (Intrinsic source)
• Rsub should be scalable in terms of channel width, length and fingers.
• All parameters should be extracted easily.
(Yuhua Cheng, “An Overview of Device Modeling in Bridging Manufacturing and Design for RF applications”, International Conference on Solid-state Integrated Circuits and Technologies, D4.3-1 – D4.3-6, Beijing, China, Oct 2004 )
29 Yuhua Cheng
________________________________Modeling Challenges - NQS Effects• NQS will significantly
impact Y11 and Y21behavior.
• Many approaches are proposed to model this effect:– Multi-segment
approach– Relaxation time – Rg/Ri equivalent
circuit approach• Efficient built-in
NQS effect in intrinsic core model is preferred.
0.0 4.0x109 8.0x109 1.2x1010-0.02
-0.01
0.00
0.01
WF=10x15µm
LF=1.35µm
VGS
=VDS
=1V
VBS
=0VIm{(Y
21)}
Re{Y21
}
Symbols: DataSolid lines: Model W/ NQSDotted lines: Model W/O NQSR
e{Y 21
} and
Im{Y
21} (
A/V
))
Frequency (Hz)(Yuhua Cheng et al. , “Frequency-dependent resistive and capacitive components in RF MOSFETs ,” IEEE Electron Device Letters , Vol. 22, pp. 333 –335, July 2001. )
30 Yuhua Cheng
________________________________Modeling of Flicker noise
• Downscaling may degrade flicker noise behavior.• Modeling of flicker noise in nano-scale devices becomes
more challenging.• Accurate prediction of corner frequency, Fcorner, is
critical for circuit design.
100 101 102 103 104 10510-19
10-18
10-17
10-16
10-15
10-14
Lines: ModelSymbols: Data
L=0.13µmVGS
=0.75V
VGS
=0.45V
VDS
=0.8V
Sid
(A2 /H
z)
Frequency (Hz)(Yuhua Cheng, "A study of figures of merit for the high frequency behavior of MOSFETs in RF IC applications", Eighth International Conference on modeling and Simulation of Microsystems, Anaheim, May 8-12, 2005)
31 Yuhua Cheng
________________________________Modeling Challenges – HF Noise
• Parasitic resistance at gate, source, drain and substrate generate thermal noise
• Channel thermal noise is the dominant noise source at HF.• Understanding of noise sources is important.• Modeling of channel thermal noise and induced gate noise is the most
challenging job.
2Gi
GR
2gi
GBC
GSC
GDC
SBC
DBC
GSImVG BSImbVG
SRSUBR
DR
DSR
2Si
2di
2Di
2SUBi
D
B
G
S
iG iD
iSiB
32 Yuhua Cheng
________________________________Statistical Modeling• Process variation (even within a wafer) in today’s
advanced technologies becomes more significant.
• Need physical statistical models to predict process variation and local mismatch to optimize analog/RF circuits with high yield.
• Correlations between statistical modeling with considerations of both frontend and backend process variations and yield modeling/prediction should be developed.
• In nano-scale technologies, statistical model is more meaningful than traditional corner models.
33 Yuhua Cheng
________________________________RF Modeling in Nano-scale Era
Modeling of Small Signal, Noises, Large Signal Distortion
For RF Applications
Continuity Scalability Accuracy
ComputationEfficiency
New Physical Effects
34 Yuhua Cheng
________________________________Gm/ID: Measured vs. Fitted
• Gm/ID has been proposed a FoM for model validation for analog applications.
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.50
10
20
30
VDS
=50mV
Symbols: Measured dataLines: BSIM3v3 Model
VBS
=-3.3V
VBS
=-0.7V
VBS
=0VGm
/I D (1/V
)
VGS
(V)(Yuhua Cheng, "A study of figures of merit for the high frequency behavior of MOSFETs in RF IC applications", Eighth International Conference on modeling and Simulation of Microsystems, Anaheim, May 8-12, 2005)
35 Yuhua Cheng
________________________________fT: Measured vs. Fitted
• A standard device parameter for model validation.• However, only fT is not enough to describe HF behavior of
MOSFETs, especially at technology nodes such as 0.13um and below.
10-7 10-6 10-5 10-4 10-3 10-20
20
40
60
80
100
120Lines: BSIM3v3 ModelSymbols: Data
L=0.13µm
L=0.18µm
L=0.35µm
f T (G
Hz)
ID/width (A/µm)(Yuhua Cheng, C. H. Chen, M. Matloubian, and M. J. Deen, “High frequency small signal AC and noise modeling ofMOSFETs for RF IC design,” IEEE Transactions on Electron Devices, Vol. 49, pp. 400 – 408, March 2002.)
36 Yuhua Cheng
________________________________Gmax and fmax: Measured vs. Fitted
• fmax contains the impacts from parasitics such as gate and substrate resistance and is a better FoMthan fT
10-7 10-6 10-5 10-4 10-30
5
10
15
20
25
VD=0.5V
VD=2.5VVD=2VVD=1.5VVD=1V
L=0.35µm
Lines: BSIM3v3 ModelSymbols: Data
Gm
ax (d
B)
ID/Width (A/µm)10-8 10-7 10-6 10-5 10-4 10-3 10-20
10
20
30
VD=1.5V
VD=2.5VVD=2V
VD=0.5V
VD=1VL=0.35µm
Lines: BSIM3v3 ModelSymbols: Data
fmax
(GHz
)
ID/Width (A/µm)
(Yuhua Cheng, "A study of figures of merit for the high frequency behavior of MOSFETs in RF IC applications", Eighth International Conference on modeling and Simulation of Microsystems, Anaheim, May 8-12, 2005)
37 Yuhua Cheng
________________________________C-parameters: Modeled vs. Fitted
• C-parameters are more sensitive to the bias dependence of gate resistance and capacitance.
• Useful FoMs for model validation.
10-3 10-2 10-1100
101
102
103
104
L=0.35µm
Lines: ModelSymbols: Data
Rin
Rfb
Rout
R (O
hm)
ID (A)10-5 10-4 10-3 10-2 10-1
50
100
150
200L=0.35µm
Lines: ModelSymbols: Data
Cfb
Cout
Cin
Capa
cita
nce
(fF)
ID (A)(Yuhua Cheng, "A study of figures of merit for the high frequency behavior of MOSFETs in RF IC applications", Eighth International Conference on modeling and Simulation of Microsystems, Anaheim, May 8-12, 2005)
38 Yuhua Cheng
________________________________Large Signal Behavior: Measured vs. Fitted
• Below certain (the “LFL”) frequency, the distortion behavior of MOSFETs is primarily determined by transconductance and capacitances.
• With careful parameter extraction at DC and HF small signal, a model can well predict the large signal distortion behavior.
-50 -40 -30 -20 -10 0 10-100
-80
-60
-40
-20
0
20Lines: ModelSymbols: Data
P3
P1
P2
VDS=1VVGS=0.83V
Freq=1.8GHz
W=10x12µmL=0.36µm
Pou
t (dB
m)
Pin (dBm)10-3 10-2 10-1
-60
-50
-40
-30
-20
-10
0
10
Lines: ModelSymbols: Data
VDS=1V
Pin=0dBmFreq=1.8GHz
Wf=12µmLf=0.36µmFinger=10
Pou
t (dB
m)
IDS (A)(Tzung-Yin Lee, and Yuhua Cheng, “High-Frequency Characterization and Modeling of Distortion Behavior of MOSFETs for RF IC Design,” IEEE Journal Of Solid-State Circuits, VOL. 39, NO. 9, pp. 1407 – 1414, September 2004)
39 Yuhua Cheng
________________________________
Validate Models Based On Meaningful
Figures-of-Merit
Something worth mentioning in RF Modeling
40 Yuhua Cheng
________________________________Summary
• While all the requirements for continuity, scalability, accuracy and computation efficiency need to be met for device models for circuit simulation, the new physical effects in nano-scale devices make compact model development very challenging.
• A lot of additional modeling efforts to predict HF noise and large signal distortion behavior are needed for RF circuit design.
• It would be desirable in the future the modelers use certain FoMs at HF to qualify the device models targeted for analog/RF applications.
• Device modeling has become very critical component in the technology platform for RF SOC design in nano-scale technologies.
Thank you for your attention