Devices and their models
1
Course Flow!
Importance of course
Some required perquisite -device physics
Part-I Part-II
2
Course Flow!
How to extend LF Model to RF model ?
Device modeling exercises
Part-III
Seminar Talk!
Evaluated on the basis of:
– Technical contents (T) – Logical flow (L)
– Presentation skill (P) – Question/Answer (Q)
Presentation : 8-10 minutes
Questions & Answers: 2-4 minutes
25 % 25 % 25 % 25 %
Mark Distribution
Prepare talk related to the subject areas !
Continue 26/11/2016
2pm onwards in UG Class room
Part-II: MOSFET devices &
their Models
Level 1
– Known as Shichman-Hodges Model (1968).
– Accurate only for long channel devices
Level 2: MOS2
– It incorporates second order effects e.g. CLM, Body Effect etc.
computationally complex.
Level3: MOS3
– A semi-empirical model includes second order effects due to narrow width and short channel lengths
– Computationally efficient compared to Level2
BSIM Model
– BSIM stands for Berkeley Short-channel IGFET Model.
Other popular MOSFET Models:
– Philips MOS9, EKV, HiSIM etc.
BSIM Model Development
BSIM 1
– Cover sub-micron MOSFET geometries
BSIM2
– BSIM2 improves upon BSIM1 in terms of model accuracy as well as convergence behaviour in circuit simulation, however it breaks the transistor operation into several regions.
BSIM3
– Widely used model for CMOS IC Design
– Single-equation to describe device characteristics in the various regions of operation.
BSIM4
– Include parasitics at rf, which BSIM3 ignore – Used for high frequency Applications
– With improve rf Modeling capability
BSIM Model Development
BSIM 5
– Model for aggressively scaled (e.g. 45 nm ) MOSFET – Incorporate more detail physics
– Provide unified formulation the device in various region of operation
BSIM6
– Released in year 2014
– The model provides excellent accuracy in all regions of operation.
Others Models – BSIM-SOI – BSIM-CMG
HA:
Prepare a brief report of 1-2 pages on Various stages of BSIM Models development
Ref: http://www-device.eecs.berkeley.edu/bsim/?page=BSIMCMG
BSIM3 Model – A critical Look
It is a physics-based, accurate, scalable, robust and predictive MOSFET SPICE model for circuit simulation
It adopts a single-equation to describe device characteristics in the various operating regions, which eliminates the discontinuity in the I-V and C-V characteristics.
There are three versions of this model e.g. BSIM3v1, BSIM3v2 & BSIM3v3.
There are several variations within BSIM3v3 itself, which include BSIM3v3.1 BSIM3v3.2, and BSIM3v3.3 .
It is a charge-conservative model, where the sum of currents flowing into the four terminals of the MOSFET is zero.
The model is physical, with most parameters having strong correlation with the fabrication process. However, some of the parameters have weak physical significance and are introduced mainly for the purpose of fitting.
A variety of short-channel effects are included in the model like charge-sharing effects, drain-induced barrier-lowering, reverse-short-channel effects, and those effects found in short-channel and narrow-width devices
Cont..
BSIM3 is adequate for a majority of analogue and digital circuit simulation.
However, for rf applications, a sub-circuit approach to include parasitic elements is required.
BSIM3 accounts for the parasitic such as overlap and (outer) fringing capacitances as well as the bulk-source and bulk-drain diodes
Web Page: http://www-device.eecs.berkeley.edu/~bsim3.
The source and drain terminal resistances are included in BSIM3, but not the gate to substrate resistances. Likewise, the noise associated with the gate and substrate resistances are not included.
All voltages in the BSIM3 are referenced to the source terminal, not to the bulk.
It accounts for the first-order effects of temperature on the device characteristics. The degradation of the mobility and increase of threshold voltage with temperature are modelled.
CapMod=0 in the BSIM3 for the intrinsic transistor is based on long-channel approximation. Many of its device capacitances can exhibit incorrect voltage dependencies, although the circuit simulation can still be quite accurate.
What BSIM3 does not include?
It does not include self-heating effects. The device temperature, once specified, remains constant during the entire simulation. It is possible to specify the device temperature to any particular value. BSIM3 contains the parameters to calculate the proper device characteristics under various temperatures.
The model does not include the drain-induced gate noise. Only the thermal noise and the 1/f noise of the channel are considered.
BSIM3 does not include the gate-oxide tunnelling current through oxide.
In means that dc gate current is identically zero. There is no leakage path between the gate and other terminals. However, the only leakage currents included in BSIM3 are those associated with drain-bulk, the source-bulk junctions, and the substrate current brought about by impact ionisation.
When gate-to-source voltage VGS is zero or negative, the sub-threshold current continue to decrease exponentially towards zero as VGS decreases, even when VGS become negative. i.e. BSIM3 can be used to predict the OFF current (the drain current at zero VGS)
There is no parameter which specifies the maximum operating voltages for VGS, VDS and source-to-bulk VBS.
Care must be exercised to ensure that the voltages lie within the breakdown values.
In developing a model the mathematical formulation is used to describe the transistor behaviour. A circuit designer, in contrast, prefers to visualize a transistor with its equivalent circuit. In the following sections, equivalent circuits based on BSIM3’s mathematical formulation will be described. It includes equivalent circuits for dc, large-signal, small-signal ac, and noise analysis.
What BSIM3 does not include?
Types
I-V
(Static behavior)Dynamic
(Include energy storage
elements) Noise
Model Formulation
Modelling is defined as the process by which the electrical properties of a semiconductor device are represented by means of mathematical equations, circuit representations
How equation for these models are formulated ?
Ref: Y. Cheng et. al. , MOSFET Modeling & BSIM3 User Guide
dc I-V
The dc model used in BSIM3 is shown here.
IDS
Isub
RD
GMIN RS
GMIN
Source Drain
Gate
Bulk
SB
Ij,
Ij,DB
dc model does not account for any leakage current between gate and channel.
Ref: Y. Cheng et. al. , pages: 106-141
dc I-V
• i.e. in BSIM3, there is a complete dc open circuit between the gate and the other side of the oxide.
• The dc gate current is always zero.
• It add G
MIN(default value 10
-12
-1) across drain- bulk and the source-bulk just to aid convergence
• These conductances do not bear physical significance but they added purely to aid convergence of numerical solution during computer iteration.
• I
JDB& I
JDSare modelled
by a normal diode
equation.
Cont…
• The source resistance R
Sand drain resistance R
Dare included and modelled based on sheet resistance of the source and drain contacts.
• The drain-to-source current I
DSincorporates short-channel effects and is generally a complex function of voltages.
• The I
DScan be formulated based on channel charge
density and the velocity-field relationship.
Channel Charge
gsteff OX
chso C V
Q
When V
DS=0, the channel charge Q
chs0can be written as:
V
vgseffbecome V
GS-V
thin the strong inversion, and follows an
exponential relationship in the sub-thresold region.
Cont..
When V
DSis applied unified charge, Q
chs(y)can be expressed as:
) 1
( ( )
) (
b y F chso
y
chs V
Q V
Q
bulk t gsteff
b A
nv V V
Where
Abulk is a parameter accounting for the bulk charge effect due to V
DS V
F(y)stands for the quasi-Fermi potential at any given
point, y along the channel with respect to the source.
Velocity-Field
The velocity and electric field is related through the equation:
sat y
y
y E E E
v( ) v(y) vsat Ey Esat
Where is the mobility including the influence of the lateral electric field Eyand is given by
) 1
/(
sat y
eff E
E
Where eff is the effective mobility, which is complex function gate bias and body bias
Now, combining mobility and charge expressions, Id valid in different regions can be expressed as:
dy Q dV
W
Id(y) eff chs(y) F(y)
Assuming Rds = 0 and integrating the above equation from source (i.e. y=0 ) to drain (i.e. y=Leff ), the drain current, Ids0 expression valid from sub-threshold to strong inversion region can be written as:
Cont..
) 1
(
2 ) 1
0 (
0
eff sat
DS eff
b DS DS
chs eff eff ds
L E L V
V V V
Q W
I
When Rds>0 , IDS can be modified as :
DS ds ds ds DS
V I R I I
0 0
1
Now when Rds>0 and SCEs are carefully accounted, then IDS current valid from sub-threshold to the strong inversion region can be derived as:
) 1
)(
1 (
1 0( )
) ( 0
ASCBE dsat DS
A dsat DS
dsat Vdsat ds ds
Vdsat ds
DS V
V V
V V V
V I R
I I
Where Vdsat is the saturation voltage. VA and VASCBE are the Early voltages accounting for various short-channel effects.
Dynamic Formulation
• Before proceeding to the discussion of BSIM3 dynamic formulation used for transient and ac analysis, it is instructive to examine the various device capacitances, which play an important role in MOSFET dynamic performance.
• Generally, MOSFET capacitance can be divided into two groups, intrinsic and the extrinsic capacitances.
• The intrinsic capacitance
is related to the region
between the metallurgic
source and drain
junctions.
Cont..
Intrinsic Capacitance
Extrinsic Capacitance Extrinsic
Capacitance
Cont…
• Extrinsic capacitance, or the parasitic capacitance, which include bulk-drain and bulk-source diodes can be broadly grouped in three categories as follows:
– The fringing capacitance between the poly-silicon gate and source (taken identical to gate-drain fringing capacitance), C
f; – The overlap capacitances between the gate and source
( taken identical to gate-drain overlap capacitance), C
ov;
– Drain-bulk junction capacitance, C
JDBand source-bulk junction
capacitance, C
JSB.
Cont…
• These parasitic capacitances are reciprocal. If C
GS,Pdenote the parasitic gate-source capacitance, define as Q
G,P/
V
Swhere Q
G,Pis the portion of the gate charge associated with the parasitic capacitance.
• The reciprocal nature of the parasitic capacitance means that C
GS,Pis the same as C
SG,Pwhich is Q
S,P/ V
G.
• Similarly, other inter electrode parasitic capacitances can be defined as:
DB j p
db p
bd
f ov
p dg p
gd
C C
C
C C
C C
, ,
,
, ,
DB j p
sb p
bs
f ov
p sg p
gs
C C
C
C C
C C
, ,
,
,
,
0 ,
,p bg p gb
gb C C
C
• The remaining capacitances
C
gsi, C
gdiand C
gbiare the
device intrinsic capacitances
as shown in the figure in bold
face. The total gate-to-source
C
gs,t, gate-to-drain C
gd,tand
gate-to-body C
gb,tcan be
expressed as:
Cont…
f ov
i gd t
gd C C C
C , , Cgb,t Cgb,i Cgb0
T
he current-voltage relation can now be expressed as:dt dv v
Q dt
dv V
Q dt
dv v
Q dt
t dQ
i BS
BS G DS
DS G GS
GS G G
G
) (
dt dv v
Q dt
dv v
Q dt
dv v
I Q I
I t
i BS
BS DS Deff
DS GS Deff
GS Deff DB
j sub DS
D
,
) (
dt dv v
Q dt
dv v
Q dt
dv v
I Q I
I t
i BS
BS DS Beff
DS GS Beff
GS Beff DB
j SB j sub
B
, ,
) (
Where QG is the total gate charge, QDeff = QD - QJ,DB is the total effective drain charge and QBeff = QB + QJ,DB +QJ,DS is total effective bulk charge.
These equations can be solved and branch currents and voltages can be determined.
Ref: Y. Cheng et. al. , MOSFET Modeling & BSIM3 User Guide
Noise Formulation
• A useful measure of the noise performance of a system is noise factor, usually denoted by F.
The noise factor F and noise figure NF are defined as:
• Presented below the most significant noise sources associated with the MOSFET, which have been suitably modelled to accommodate the sub- micron characteristics.
source the
to due power
noise output
Total
power noise
output Total
F
(F) log
10 (NF)
Figure
Noise
10Cont…
• Channel thermal noise is the most important noise source. In order to calculate its value the channel is separated into two regions:
– One where the gradual channel approximation holds
– Another where carrier velocity saturation is present.
• The fundamental theory to calculate the channel noise
• Gamma factor () equals 2/3 for the classical theory but it deviates toward larger values for short channel devices.
0 2
4 B d
d k T g
f
i
Where kB is the Boltzmans constant, is a factor which strongly dependent on bias, and gd0 is the zero-bias drain conductance.
Cont…
• Induced gate noise, which is insignificant for large device. However, it dramatically increase as
the frequency of
operation is increased in GHz range.
• Gate resistance R
gnoise is contributed due to gate material used and modelled in BSIM3 as:
• Substrate resistance and series drain/source resistances also produce noise in a device, which need to be carefully modelled with.
• While series drain/source resistances can be kept low by increasing doping, however the contribution of the back-gate effect e.g. substrate resistance is very critical.
4 3
2 R f
T k
vg B g
• As regards noise model in BSIM3, the thermal noise is definitely accounted for.
• The induced gate noise and gate noise are not available in BSIM3 and should be accounted at microwave frequency of operation.
• For the substrate noise contribution is absent in BSIM3 but thermal noise due to source resistance is incorporated
• The proper way to include the terminal resistances, as far as noise analysis is concerned, is to declare the source and drain resistances through the sheet resistance parameters.
• The noise equivalent circuit adapted in BSIM3 is shown in the Figure.
• BSIM3 considers the parasitic junction diodes to be noise free.
BSIM3--Noise Modeling
Cont..
Therefore, there are no noise sources associated with g
j,DBand g
j,SB,
which are the conductances of the junction diodes. There is no noise
source associated with G
MIN, either. The conductance G
MINis added to
aid numerical convergence.
29
Optimization of F
MOSFET
YS F
~ Source
Source admittance
OPT S
OPT S
MIN i.e.B B &G G F
F
MOSFET
YOPT
F
MIN~ Source
Source admittance
With YOPT F=FMIN
Y
OPTfor MOSFET can be expressed as:
S 2
2 S
s n
MIN G B
G F R
F GOPT BOPT
See TH Lee for detail
T 2
T
MIN
ω
2.3 ω 1
c - 1 ω γδ
ω 5 1 2
F ~
2
gs
OPT
1 - c
5γ αωC δ
G B
opt 0
Cont..
Minimizing the noise is qualitatively similar to maximizing power transfer, however the source admittances leading to these two conditions are almost never the same.
Therefore, one generally can not enjoy maximum power
gain and minimum noise figure simultaneously.
Concluding Remarks
• Introduced the industry standard model, BSIM3
• Mathematical formulation for dc, dynamic and noise have been covered
• BSIM3 as such is not suitable for rf
applications without incorporating
parastics especially around gate and
substrate.
32
Course Flow!
How to
extend BSIM3 Model for RF ?
Part-III
Ref:
M.S. Alam, and G.A. Armstrong, Accurate Substrate Modelling of RF CMOS, International Journal of Numerical Modelling, Volume 19, Issue 3 May/June 2006, pp 257-269.
M. S. Alam, and G.A. Armstrong, Extrinsic Resistance Parameter Extraction and RF Modelling of CMOS, Solid State Electronics, Elsevier Publication, Volume 48, Issue 5, 2004, pp. 669-674.
33
RF Model
A MOSFET is a good candidate for RF IC application because of its low cost and high integration potential.
As the gate-length of a MOSFET reduces, its high frequency characteristics improve.
The BSIM3 model has been used successfully as an accurate and scalable silicon MOSFET model in the low frequency range (<1GHz).
One of the powerful strategy is to extend
BSIM3 for rf is to add relevant parasitics
especially around gate and substrate .
34
RF Model
Substrate Network Rg
Intrinsic Model
B G
D
Gate and substrate parasitics are accounted
But for simplicity they have been modeled as a lumped elements
R
GR
POLY+ R
CHGate Resistance
Distributive
>1GHz
Signal of frequency >1GHz at the gate not ONLY see the R
POLYbut R
CHalso influence and they are distributive in NATURE
R
Gis a complex combination of R
POLYand R
CHSignificant bearing on input impedance and noise.
See page 87 of Ytterdal Book!
Substrate parasitics are not important <1GHz.
Distributed in nature but can be approximated using lumped components.
Contribute significantly to output admittance (20%) and affect noise at rf.
Substrate Parasitics
Among various models , 3-elements model is more popular
3-elements Model
R
SB= Model source-bulk signal activity
R
DB= Model drain-bulk signal activity
R
DSB= Signal activity along the channel
db
dsb
R
R
sb
dsb
R
R
db sb
SUB
R R
R
RSUB
Cont..
At GHz, normally source is connected to the body and grounded
If RS is of low value
sb SUBsb
SUB dbdb SUBSUB
j C R j C R
C j R
C y j
1
1
Csb
Cdb
RSUB RS
……
B
RD
S D
……
Csb
Cdb
RSUB RD
D
Common S/B
YSUB
Cont..
sb
SUB
sC
R 1
sub db
db
SUB
j C R
C y j
1
~
Csb
Cdb
RSUB
Common S/B
YSUB
Cdb
RSUB
YSUB
i.e. most of the signal activity is
confined to R
SUBgmbV2 V1
Core
RF Model
Parasitics
Parasitics have NOW been
identified !
They have been modeled through
resistive elements and they
strongly alter GAIN and NOISEgmbV2
V
1R
SUBC
DBR
SR
DR
Grf Model (>1GHz)
Core Model (<1GHz)
BSIM3 + Parasitics rf
Question?
The model is not know as its components are NOT known!
These components are determined by a technique called EXTRACTION
Step-I
Measured S-
Parameters of the DUT
Step-II
Convert S to Y or Z or H
Step-III
Use appropriate EC
Carry out small-signal analysis
Express model elements in terms of known Ys or Zs
Following steps are involved:
gmbV2 RG
V1 Cdb
CSUB
V
2G
S
S/D heavily doped!
Simplified Model !
R
S=R
D0
1 m
V g C
gdC
gsV
1G
S
D
g
ds 2 mbV g
V
2R
gC
dbR
subVS
I
sI
2SCAnalysis
V
2SC=0
2 0
11
SC
S S V
y I
V
2 211 g
(
gs gd) (
gs gd)
y R C C j C C
Under the assumption 2 (Cgd+ Cgs)2 Rg2 1
1
2 0
2 21
SC
SC S V
y I
V
21 m
-
gdy g j C
If <<1/ Rg(Cgd + Cgs)
2 Carefully look at these
equations and express model elements in terms of Ys
1 m
V g C
gdC
gsV
1G
S
D
g
ds 2 mbV g
V
2R
gC
dbR
subVS
Is I1SC
Analysis
V1SC=0
12
-
gdy j C
If <<1/ Rg(Cgd + Cgs)
1 0
1 12
SC
SC S V
y I
V
3
1 0
22
SC
S S V
y I
V
Under the assumption that 2 Rsub2Cdb21
2 2
Re( y
2222) g
ds C R
db sub(1 g R
mb sub) Im( y ) C
gd C
db(1 g R
mb sub)
4Carefully look at these
Equation and express model components in terms of Ys
22 0
ds
Re( y )
2g
Using (4)
Using (1)
2 11
11
g
(Im( y ))
) y
R Re(
Using (2)g
m Re( y
21)
Im(
12)
gd
C y
Using (3)
0.1 0.2
mb m
g g
Im(
11)
gs gd
C y C
Using (1)
Extraction Result
gmbV2 RG
V1 Cdb
RSUB V2 G
S Known
Unknown
Going through various steps involved for EXTRACTION
Elements shown in green enclosure have been determined but BLUE enclosure elements are not known yet!
2 2
Re( y
22) g
ds C R
db sub(1 g R
mb sub) Im( y
22) C
gd C
db(1 g R
mb sub)
Defining Equation!
)) y
Im(
) y (Im(
R
g )
y C Re(
12 22
sub
ds 22
db
-
) g
) y
(Re(
g ))
y Im(
) y (Im(
g )
y R Re(
ds 22
mb 2
12 22
ds 22
sub
-
-
Re-Cap!
Taking you through various steps involved in EXTRACTION
What if source & drain are not heavily doped?
There effect need to be carefully included !
gmbV2
RS
RG RD
V1
Improved Model
Cdb
CSUB V2
Known
Unknown
Unknown
Components in GREEN enclosure are known Components in BLACK enclosures are NOT known
How to determine components in BLACK enclosure?
gmbV2
RS
RG RD
V1
Analysis
Cdb
CSUB V2
2
VS
Is
1
V1OC
12 2 2 2
Re( ) s BCgd
Z R
A B
1 0
1 12
OC S I
Z V
I
22 2 2 2
( )
Re( ) s d Cgs Cgd B
Z R R
A B
1 0
22
s s I
Z V
I
Where A and B are constant
Source Resistance R S
As
0
B A
R BC )
Z
Re(
12 s 2 2 gd 2
Re (Z
12)
RS
R
S Re( Z
12)
Asymptotic plot of Re(Z12) gives the value of Rs
Re (Z
22)
RD+RS
2 2
2
gd gs
d s
22
A B
B ) C
C R (
R )
Z
Re(
As
0
Re(
22)
S D
R R Z
Drain Resistance R D
Asymptotic plot of Re(Z22) gives the value of Rs+Rd With Rs already known; Rd can be determined
gmbV2
RS
RG RD
V1
RF Model
Cdb
CSUB V2
Model is known
RG=5 Cgs= 100fF Cgd = 30fF gds= 0.01mS gm =0.1mS gmb=0.01mS RS=RD=4 Typical value for CMOS process
gmbV2
RS
RG RD
V1 Cdb
CSUB V2
Question?
Which of the model elements contribute to the noise?
Various parasitics PLUS e-activities in the channel
gmbV2 RG
V
1 CdbRSUB
V
2Problem!
Cds
For the given model, show that
2 SUB 2
db 2
db gd
22
ds
1 C R
C C )
y C Im(
VS
I
sI
1SCV
1SC=0
2 2
2 2
22 2 2 2 2 2 2
1 1
db sub db
ds gd g gd ds
db sub db sub
C R j C
y g C R j C j C
C R C R
Under the assumptions
2(C
gd+ C
gs)
2R
g2 1 and
2R
sub2C
db2 1, y
22can be simplified to:
1 0
22
SC
S S V
y I
V
Hint: Use imaginary part