DG review slides - epfl.ch
Transcript of DG review slides - epfl.ch
Principle Investigators: Prof. Chenming Hu and Prof. Ali Niknejad
Current Team: Sriramkumar V., Yogesh Chauhan, Muhammed Karim
Previous Graduates: Darsen Lu, Chung-Hsun Lin, Mohan Dunga
UC Berkeley
BSIM-CMG A Turnkey Compact Model for Common Multi Gate Devices
Dec. 16, 2011
Nano-Tera Workshop, EPFL, Lausanne
Why next generation transistors?
Gate cannot control the leakage paths far from the gate
Thinning the oxide down was not enough
Dopant fluctuations => Increased variability
Leakage!
Scaling
UC Berkeley- 2
New MOSFET Structures
Leakage!
UC Berkeley- 3
Bulk Planar MOSFET
FinFET UTBSOI
Substrate/ Back Gate
New MOSFET Structures - Demonstration
K. Cheng et al. IEDM 2009 - (IBM / ST)
UC Berkeley - 4
Y. Choi et al. IEEE EDL 2000- (UC Berkeley) X. Huang et al. IEDM 1999 (UC Berkeley)
C.C. Wu et al. IEDM 2010 (TSMC)
FinFET UTBSOI
Advantages for SoC Design
No doping required in the channel
Superior short channel control
Lower DIBL
Steeper on-off
Improved mobility
Lower vertical field needed
Less scattering in the channel
Lower gate-induced drain leakage
Next Gen.: Scale by thinning body
UC Berkeley- 5
FinFET
UTBSOI
Improves lower-Vdd
logic operations
Stand-by power
savings in HVT
devices
Substrate / Back Gate
Another MOSFET architecture
UC Berkeley- 6
DRAM Cell Transistor – 4F2 layout
Versatile Multi-Gate Compact Model: BSIM-MG
Fin
BOX
Gate
1
Gate
2
BOX
p-sub
P+ back-gate
BSIM-IMG
BSIM-CMG
UTBSOI
BG-ETSOI
UC Berkeley - 7
FinFETs on Bulk and
SOI Substrates
Vertical Fin
IMG
Challenges in developing a new model
New Physics
Fully depleted channel
Quantum confinement etc.
When to include them?
Support Multiple Device architectures
Inertia with BSIM4
Large user base
Familiarity with the parameters -> a language in itself
Speed
Convergence – Model behavior in extreme cases
Balance Physics and Flexibility
Balance Speed and Accuracy
UC Berkeley - 8
BSIM-CMG
UC Berkeley - 9
Introduction
Core Model
Surface Potential Equation
Drain Current
Capacitance Model
Real Device Effects
Model Validation & QA
Conclusion
BSIM-CMG Core Models
Four device architectures
Three core models
Intrinsic Double Gate Core (Y. Taur et al, IEEE EDL, 2004)
Perturbation based DG Core for high-doping
Cylindrical Gate Core
Bulk and SOI Substrate
Double Gate Double Gate /
Trigate / FinFET
Quadruple Gate Cylindrical Gate /
Nanowire FET
UC Berkeley - 10
NANA
Surface Potential Core – Double Gate
Surface potential is obtained from the solution of Poisson’s equation and Gauss’s Law.
Poisson’s equation inside the body can be written as (Vch is channel potential)
Body doping complicates the solution of the Poisson’s equation.
Perturbation approach is used to solve this problem.
M. Dunga et al., IEEE TED, Vol. 53, No. 9, 2006
M. Dunga et al., VLSI 2007
Mohan Dunga, PhD Dissertation, UC Berkeley UC Berkeley - 11
Start with a long channel transistor
Assumptions Gradual Channel Approximation
No influence of holes (SOI like structure)
Fully Depleted Body
Gauss Law + Correction for Polysilicon Gate-Depletion
Implicit equation in Qi (channel charge)
Surface Potential Equation
1-D Poisson’s Equation
Surface Potential Core – Cylindrical Gate
S.Venugopalan et. al., SSE, Vol 67, Issue 1, Jan 2012
UC Berkeley - 12
Core Drain Current Model
Drain Current (Pao-Sah, No charge sheet approximation)
0.0 0.2 0.4 0.6 0.8 1.0
-2
-1
0
1
2
Vds=1.0V
Ids E
rro
r R
ela
tive
to
TC
AD
(%
)
Gate Voltage (V)
v104.0 and prior
v105.0
0.0 0.2 0.4 0.6 0.8 1.0
-2
-1
0
1
2
Vds=1.0V
Gm
Err
or
Re
lative
to
TC
AD
(%
)
Gate Voltage (V)
v104.0 and prior
v105.0
0.0 0.5 1.0 1.50.0
0.4
0.8
1.2
1.6
2D TCAD
I-V Model
Vg = 0.9V
Vg = 1.2V
Vg = 1.5V
Tsi = 20nm
Tox = 2nm
Na = 3e18cm-3
Dra
in C
urr
ent (m
A)
Drain Voltage (V)
UC Berkeley - 13
Core Capacitance Model
Model inherently exhibits symmetry Cij = Cji @ Vds= 0 V
Model overlies TCAD results
No tuning parameters used
Accurate short channel behavior – RF Design
Needs additional tuning parameter flexibility
Symbols: TCAD Results; Lines: Model UC Berkeley - 14
BSIM-MG
UC Berkeley - 15
Introduction
Core Model
Surface Potential Equation
Drain Current
Capacitance Model
Real Device Effects
Model Validation & QA
Conclusion
Short Channel (2D) Effects
Along the channel – 2D
Quasi-2D analysis
Similar expression for Double Gate and FinFET/Trigate
Analytical expressions model
Threshold Voltage roll off
Drain induced barrier lowering (DIBL)
Sub-threshold swing degradation
Auth and Plummer, IEEE EDL, 2007
Symbols: TCAD Results; Lines: Model
si
acfbgsc qNVV
dy
d
22
2
Characteristic Length
UC Berkeley - 16
Quantum Mechanical Effects
Predictive model for confinement induced Vth shift due to band splitting present in the model
Can choose to use an effective tox that accounts for charge centroid behavior with bias
Effective Width model that accounts for reduction in width for a triple / quadruple / surround gate structure
Width reduction due to structural confinement of inversion charge. (Dotted lines represent the effective width perimeter)
BOX
UC Berkeley - 17
Asymmetric Device Model
S/D under-lapped FinFETs used as PU and AX transistors in SRAMS show
Improved RSNM at same WSNM
Decreased Time-to-Write (but increased cell read access time)
Asymmetric Halo could be used to create I/O FinFETs for special applications
V.P.Hu et al VLSID 2006
UC Berkeley - 18
Asymmetry Model : Parameters Identified
Identified five existing parameters and created their reverse mode equivalents
Doping Gradient in Channel
On-current degradation parameter
Source-side barrier different – DIBL parameter
Output conductance parameter
Top/Bottom Electrode Asymmetry
Quasi-saturation amplification / suppression – Rs/d Parameters
UC Berkeley - 19
S. Venugopalan et al., NVMTS 2011
n1
nwell
ni
STI STI
p+
n1
nwell
ni
STI STI
FinFETs on Bulk require an implant positioned just below the fin to prevent punch-through (and a retrograde profile).
Creates S/D junctions to have two metallurgical junctions of the form p+-n1-nwell or n+-p1-psub.
Simulations indicate that the depletion region beneath the junction could traverse the second junction too.
Source/Drain Junctions in FinFETs
Punch-through
implant
Cross-section at
middle of channel
Cross-section
at S/D junction
UC Berkeley - 20
Enhanced BSIM4 style junction charge model with just 2 additional parameters – Cj02 and m2
S/D Junction Cap in FinFETs - Model
Cross-section at S/D junction
1
1
1
/
101,1
11
1
m
V
CQ
m
b
dbs
bjrevjn
2
1
2
/
202
1
1
1
1011
11
1
11
21
m
V
Cm
V
C
m
b
dbs
bj
m
b
bc
bj
1
1
1
1
101
2102
2
1
02
01
1
1
1
m
b
bcj
bj
b
m
j
j
bbc
VmC
mC
C
CV
where,
Vbc<=Vbs/d<=0
Vbs/d<= Vbc
n1
nwell
ni
STI STI
p+
Vbc: the voltage at when the depletion
edge reaches the n1-nwell boundary
UC Berkeley - 21
0.0E+00
5.0E+28
1.0E+29
1.5E+29
2.0E+29
2.5E+29
3.0E+29
0 0.5 1 1.5 2
Retrograde Profile
|Vbs/d|
1/C
jn2
Unified Framework for FinFET Parasitic Resistances and Capacitances
Source Drain
Source Drain
Extension (under
spacer)
Gate
Gate
UC Berkeley - 22
In sub-45nm tech.
Parasitic Resistance – significant portion of RON
Total Parasitic Caps ~ Intrinsic Gate Capacitance
Geometrically scalable parasitic resistance and capacitance model present in BSIM-CMG
FinFET Rds Modeling
UC Berkeley - 23
Darsen Lu, PhD Dissertation, UC Berkeley
Contact Resistance
Spreading Resistance
Extension Resistance
Silicide contact
Heavily-doped S/D
regions (HDD)
Rcon
Gate
Gate
Oxide
S/D Extension
Region Channel Region
Gate
Rsphdd Rext
-60 -30 0 30 60 900
250
500
750
1000
1250
Rch (
-m
)
Lg (nm)
Vgs - Vth = 0.1V . . . 0.45V
in steps of 0.05V
Decreasing
Vgs - Vth
Cfg: fin gate
Ccg=Ccg1+Ccg2+Ccg3: contact gate
Both Cfg and Ccg agree well with 2D numerical simulations
0 10 20 30 40 50 60 70 800.16
0.18
0.20
0.22
0.24
0.26
Lext
=20nm Lext
=25nm
Lext
=30nm Lext
=35nm
Lext
=40nm
Cfg
(fF
)
Wg (nm)
20 25 30 35 40 45 500.06
0.09
0.12
0.15
Ccg (
fF)
Wg (nm)
Tc = 20nm --> 40nm
in steps of 2nm
Increasing Tc
Fin
Epi
-Si
Gate
Cfg
Ccg2
Ccg1
Wg
Lext
Tc
Ccg3
0 25 50 75 1000.0
0.1
0.2
0.3
0.4
0.5
Tc (nm)
Ccg
1 (
fF)
Increasing Lext
Lext = 15nm --> 40nm
in steps of 5nm
Tc = Wg + Tox
Tox = 1nm
Wfin=1um
UC Berkeley - 24
FinFET Cfr Modeling – TCAD Verification
Real Device Effects
Core
SPE
I-V C-V
Short
Channel
Effects
GIDL Current
Channel Length
Modulation and
DIBL
Mobility
Degradation
Direct tunneling
gate current Overlap
capacitances
Temperature
Effects
Velocity
Saturation
S/D Resistance/
Parasitic
Resistance
Fringe
Capacitances
Noise models
Impact Ionization
current
Quantum
Effects
UC Berkeley - 25
BSIM-CMG
UC Berkeley - 26
Introduction
Core Model
Surface Potential Equation
Drain Current
Capacitance Model
Real Device Effects
Model Validation & QA
Conclusion
SOI FinFET Fitting
-0.5 0.0 0.5 1.00.0
300.0µ
600.0µ
900.0µ
1.2m
GIDLVds=50mV
Vds=1V
Vds=50mV
Vds=1VLg = 90nm
Dra
in C
urr
ent (A
)
Gate Voltage (V)
1n
100n
10µ
1m
0.0 0.2 0.4 0.6 0.8 1.00.0
0.4
0.8
1.2
1.6
Lg = 90nm
Vgs = 0.4V
Vgs = 0.6V
Vgs = 0.8V
Vgs = 1.0V
Dra
in C
urr
ent (m
A)
Drain Voltage (V)
0.0 0.2 0.4 0.6 0.8 1.00.0
0.2
0.4
0.6
Vds = 50mV
Vds = 1V
Lg = 90nm
Tra
nsco
nd
ucta
nce
(m
S)
Gate Voltage (V)
0.0
0.6
1.2
1.8
2.4
SOI FinFETs are fabricated at TI
TFIN=22nm, HFIN=60nm, TOX=2nm
Drain Current v.s Vgs
(Lg = 90nm) Drain Current v.s. Vds
(Lg = 90 nm)
Transconductance
(Lg = 90nm)
Symbols: Data
Lines: Model
0.2 0.4 0.6 0.8 1.00
20
40
60
Vds
= 50 mV
Vds
= 1 V
L = 1m
gm E
ffic
ien
cy, g
m/I
d (
V-1)
Gate Voltage (V)0.0 0.2 0.4 0.6 0.8 1.0
1n
1
1m
Lg = 90nm
Vgs = 1.0V --> 0.0V
(in steps of 0.2V)
Ou
tpu
t C
on
du
cta
nce
(S
)
Drain Voltage (V)0.0 0.2 0.4 0.6 0.8 1.0
1p
1n
1
1m
Lg = 1m
Vgs = 1.0V --> 0.2V
(in steps of 0.2V)
Ou
tpu
t C
on
du
cta
nce
(S
)
Drain Voltage (V)
Gm / Ids (Lg = 1 μm) Output Conductance (Lg = 90 nm) Output Conductance (Lg = 1 μm)
µ
UC Berkeley - 27
Bulk FinFET Fitting
Bulk FinFETs are fabricated by TSMC
TFIN=25nm, HFIN=27.5nm, EOT=2.42nm
Drain Current vs. Vgs (Lg = 50nm) Drain Current vs. Vds (Lg = 50 nm) Output Conductance (Lg = 50nm)
0.3 0.6 0.9 1.20
10µ
20µ
30µ
40µ
50µ
Vds = 1.2V
Vds = 50mV
Vds = 1.2V
Vds = 50mV
Lg = 50nm
Dra
in C
urr
ent (A
)
Gate Voltage (V)
1p
1n
1µ
1m
0.0 0.3 0.6 0.9 1.20
10
20
30
40
50
Lg = 50nm
Vgs
= 0.4V
Vgs
= 0.8V
Vgs
= 0.6V
Vgs
= 1.0V
Vgs
= 1.2V
Dra
in C
urr
ent (
A)
Drain Voltage (V)0.0 0.3 0.6 0.9 1.2
10n
1
100 Vgs = 1.2V --> 0.2V
(in steps of 0.2V)
Lg = 50nm
Outp
ut C
on
du
cta
nce (
S)
Drain Voltage (V)
Symbols: Data
Lines: Model
0.0 0.3 0.6 0.9 1.21p
1n
1
1m
Vgs = 1.2V --> 0.4V
(in steps of 0.2V)
Lg = 0.97m
Ou
tpu
t C
on
du
cta
nce
(S
)
Drain Voltage (V)
0.0 0.3 0.6 0.9 1.20
200n
400n
600n
800n
L = 0.97m
Vds = 1.2V
Vds = 50mV
Tra
nsco
nd
ucta
nce
, g
m (
S)
Gate Voltage (V)
0
2µ
4µ
6µ
8µ
10µ
0.0 0.3 0.6 0.9 1.21p
10p
100p
Vds = 1.2V
Lg = 50nm
Su
bstr
ate
Curr
ent (A
)
Gate Voltage (V)
Transconductance (Lg = 0.97μm) Output Conductance (Lg = 0.97 μm) Substrate Current (Lg = 50nm)
µ
µ
UC Berkeley - 28
Vertical GAAFET Overlay
Transconductance
Output Conductance Temperature Dependence
Drain Current vs. Vds
Drain Current vs. Vgs
Lg=120nm, D=80nm, Tox=3nm Symbols: Data
Lines: Model
UC Berkeley - 29
Global Extraction Procedure
θSAT[L] = θSAT0 × (1 + CθSAT х exp ( - Leff / DθSAT ) )
mobility & series
resistance
short channel
effect
velocity
saturation DIBL output conductance
...
e.g.
...
UC Berkeley - 30
Shijing Yao et al., ICMTS 2010
Dra
in C
urr
ent
(mA)
Gate Voltage(V)
0.05
0.10
0.15
0.20
0.25
Dra
in C
urr
ent
(A)
0.1
0.2
0.3
0.4
0.5
0.6
Tra
nsc
onduct
ance
(mA/V
)
Gate Voltage(V)
0.2
0.6
1.0
1.4
0.0
0.4
0.8
1.2
Dra
in C
urr
ent
(mA)
0.5
1.0
1.5
2.0
2.5
0 0
0.0
Tra
nsc
onduct
ance
(mA/V
)
Gate Voltage(V) Gate Voltage(V)
Vds=0.05V Vds=0.05V
Vds=1.0V Vds=1.0V
Dra
in C
urr
ent
(A)
NMOS Leff from 30nm to 10um
Leff Leff
Leff Leff
Symbols: Data
Lines: Model
UC Berkeley - 31
0.1
0.2
0.3
0 Dra
in C
urr
ent
(mA)
Gate Voltage(V)
0.1
0.2
0.3
0.4
0.5
0.6
1E-13
1E-8
1E-3
1E-18
Tra
nsc
onduct
ance
(mA/V
)
Dra
in C
urr
ent
(A)
Gate Voltage(V)
0.0
0.5
1.0
1.5
2.0
5E-16
1E-11
1E-2
1E-20
5E-7
Gate Voltage(V)
Dra
in C
urr
ent
(A)
Dra
in C
urr
ent
(mA)
1.0
2.0
3.0
Tra
nsc
onduct
ance
(mA/V
)
Gate Voltage(V)
0
0.0
PMOS Leff from 35nm to 10um
Leff
Leff
Leff Leff
Symbols: Data
Lines: Model
UC Berkeley - 32
Drain Voltage(V)
Drain Voltage(V)
Dra
in c
urr
en
t(m
A)
Ou
tpu
t R
esis
tan
ce
(Ω)
NMOS Leff = 30nm PMOS Leff = 35nm
Short Channel Output Characteristics
Symbols: Data, Lines: Model
Drain Voltage(V)
NMOS Leff = 250nm PMOS Leff = 250nm
Drain Voltage(V) Drain Voltage(V)
Drain Voltage(V)
Dra
in c
urr
en
t(m
A)
Drain Voltage(V) Drain Voltage(V)
NMOS Leff = 10um
PMOS Leff = 10um
UC Berkeley - 33
Symmetry / Continuity Tests
Model passes both DC and AC Symmetry Tests
Drain Current Capacitances (Cgg and Csd)
C.C.McAndrew, IEEE TED, Vol. 53, No. 9, 2006
UC Berkeley - 34
Speed Tests
Circuit # MOSFETs Model Runtime per iteration per transistor (normalized)
1-Transistor Id-Vds 1 BSIM4 BSIM-CMG
1.00 1.22
17-Stage Ring 34 BSIM4 BSIM-CMG
1.00 1.31
Coupled Rings 2020 BSIM4 BSIM-CMG
1.00 1.24
Averaged over multiple runs on the same machine
Speed of BSIM-CMG v105 and BSIM v4.5 compared
Both model compiled with the in-built Verilog-A compiler of HSpice
Note: Each model uses its own default parameter. Parameters are not extracted for a real technology.
UC Berkeley - 35
BSIM-CMG
UC Berkeley - 36
Introduction
Core Model
Surface Potential Equation
Drain Current
Capacitance Model
Real Device Effects
Model Validation & QA
Conclusion
Roadmap of BSIM-CMG
BSIM-CMG100.0 was released in October 2006
BSIM-CMG105.03 is under balloting stage at the Compact Model Council for standardization
Verilog-A and a well documented manual
Available in major commercial simulators
BSIM-CMG105.04 was released Dec 2011
UC Berkeley - 37
Summary
BSIM-CMG is a Turnkey , Production Ready model
Final Phase of Standardization at the CMC
Physical, Scalable Core Models for multiple device architectures
Supports both SOI and Bulk Substrates
Many Real Device Effects captured
Validated on Hardware Data from different technologies
Available in major EDA tools
UC Berkeley - 38
Acknowledgement
SRC, Industry Supporters, EDA Vendors