MOSFET DC Models
HO #17: ELEN 251 - MOS DC Models Page 1S. Saha
• In this set of notes we will
– summarize MOSFET Vth model discussed earlier
– obtain BSIM MOSFET Vth model
– describe Vth model parameters used in BSIM
– develop piece-wise compact MOSFET IDS models:♦ basic equations
♦ BSIM equations
– describe IDS model parameters used in BSIM
– develop substrate current Isub models to characterize device degradation due to high field effects.
BSIM Vth Model
HO #17: ELEN 251 - MOS DC Models Page 2S. Saha
We know that for uniformly doped substrate, Nsub, the long channel threshold voltage is:
A unified expression for Vth used to model the non-uniform vertical channel doping profile is:
where K1 and K2 model the vertically non-uniform doping effect on Vth.
tcoefficieneffect body 2
0 @ where 0
==
=≡
ox
subSi
BSthTH
CNqε
VVV
γ
( )φφγ sBSsTHth VVV −−+= 0 (1)
( ) BSsBSsTHth VKVKVV 210 −−−+= φφ (2)
BSIM Vth Model
HO #17: ELEN 251 - MOS DC Models Page 3S. Saha
We derived Vth model due to non-uniform vertical and lateral doping profile as:
( ) φφφ sLX
BSsBSsTHth LNKVKVKVV ⎟⎟
⎠
⎞⎜⎜⎝
⎛−++−−−+= 111210
(3) where
– VTH0 is a model parameter extracted from the measure IDSvs. VGS data at VBS = 0
– K1 and K2 models the effect of non-uniform vertical channel doping profile on Vth and are fitting parameter extracted from the measured data
– NLX models the non-uniform lateral profile on Vth and is a fitting parameter extracted from the measured data.
Modeling SCE due to DIBL
HO #17: ELEN 251 - MOS DC Models Page 4S. Saha
By solving Poisson Eq along the channel, Vth shift due to DIBL can be shown as:
∆Vth = θth(L)[2(Vbi - φs) + VDS] (4)
Vbi = built-in voltage of the S/D junctions given by:
where NDS = source-drain doping concentration NCH = channel doping concentration
Eq (4) shows that ∆Vth depends linearly with VDS. And,Vthdecreases as VDS increases due to DIBL.
⎟⎟⎠
⎞⎜⎜⎝
⎛= 2ln
i
DSCHbi n
NNq
kTV (5)
Modeling SCE due to DIBL
HO #17: ELEN 251 - MOS DC Models Page 5S. Saha
η = fitting parameter accounts for the approximations used to obtain lt.
In order to account for non-uniform vertical channel doping concentration, (15) is modified so that from (13):
( )CH
BSsSidep
ox
depOXSit
t
lL
lL
th
qNVX
XTl
leeL tt
−==
+=−−
φεηε
ε
θ
2 with
:bygiven length sticcharacteri a is where2)( Now, 2 (15)
( )
( ) DSBSTABTAlL
Dl
LD
sbil
LD
lL
D
VTdth
VVEEee
VeeDV
t
effSUB
t
effSUB
t
effVT
t
effVT
+++
−+=∆
−−
−−
02
20
)2(
)2(
00
11
φ
(16)
Narrow Width Effect - An Empirical Model
HO #17: ELEN 251 - MOS DC Models Page 6S. Saha
Vth shift due to narrow width effect can be shown as:
To model both narrow width and reverse narrow width effects, the model is expressed in terms of fitting parameters: K3, K3B, and W0 to get:
where W'eff = effective channel width. Introducing SCE in narrow devices, we add to (18):
seff
OXthW W
TV φ∝∆ (17)
seff
OXBSBthW WW
TVKKV φ0
33 )(+
+=∆ (18)
( )sbil
LWD
lLW
D
WVTWLth VeeDV tw
effeffWVT
tw
effeffWVT
φ−+=∆′
−′
−
)2(11 2
0 (19)
Complete Vth Model
HO #17: ELEN 251 - MOS DC Models Page 7S. Saha
(30)
( )
φφ
φφ
seff
OXbsBs
eff
LX
bssbssTHth
WTVKK
LNK
VKVKVV
′++⎟
⎟⎠
⎞⎜⎜⎝
⎛−++
−−−+=
)(11 331
210
1
2 3
( ) ( ) DSBSTABTAl
LD
lL
D
sbil
LD
lL
D
VT VVEEeeVeeD t
effSUB
t
effSUB
t
effVT
t
effVT
++−−+−−−−−
022
0 )2()2( 011
φ
4
( )sbil
LWD
lLW
D
WVT VeeD tw
effeffWVT
tw
effeffWVT
φ−+−′
−′
−
)2(11 2
0
Small L, W effect
“1” ⇒ vertical non-uniform channel doping effect “2” ⇒ lateral non-uniform channel doping effect “3” ⇒ narrow width effect “4” ⇒ SCE due to DIBL
Vth Model Parameters
HO #17: ELEN 251 - MOS DC Models Page 8S. Saha
TOX gate oxide thickness TOXM nominal TOX at which parameters are extracted XJ junction depth NCH channel doping concentration NSUB substrate doping concentration VTH0 threshold voltage @ Vbs = 0 for large L VFB Flat band voltage K1 first-order body effect coefficient K2 second-order body effect coefficient K3 narrow width coefficient K3B body effect coefficient of K3
Vth Model Parameters
HO #17: ELEN 251 - MOS DC Models Page 9S. Saha
W0 narrow width parameter NLX lateral non-uniform doping coefficient DVT0W first coefficient of narrow width effect on Vth at
small L DVT1W second coefficient of narrow width effect on
Vth at small L DVT2W body-bias coefficient of narrow width effect
on Vth at small L DVT0 first coefficient of SCE on Vth
DVT1 second coefficient of SCE on Vth
DVT2 body-bias coefficient of SCE on Vth
VBM maximum applied body bias in Vth calculation
Piece-Wise IDS Model: Inversion Layer Conductance
HO #17: ELEN 251 - MOS DC Models Page 10S. Saha
n
VG
Lx
W
P
xI
Let us apply a large potential at the gate of an MOS capacitor to cause inversion, i.e. VG > Vth.
The conductance, 1/RI of the inversion layer is:
dxxnx
qL
WR
g InI
I
I
)(1
0∫== µ (1)
Where, nI(x) = e- density in the inversion layer xI = depth of the inversion layer µn = e- mobility in the inversion layer
Inversion Layer Conductance
HO #17: ELEN 251 - MOS DC Models Page 11S. Saha
The inversion layer mobility, referred to as the surface mobility, µs ≈ 1/2 of bulk mobility. Now, the inversion layer charge per unit area is given by:
Assuming µs = constant, we get from (1) and (2):
The inversion layer resistance of an elemental length dyis:
Eq. (4) can be used to derive drain current for MOSFETs under appropriate biasing conditions.
dxxnx
qQ II
I
)(0∫−= (2)
QLWg IsI µ= (3)(− ⇒ e−)
QWdy
dRIs
I µ= (4)
Strong Inversion Region
HO #17: ELEN 251 - MOS DC Models Page 12S. Saha
n+ n+S D
G
P
Let us consider the following MOSFET structure. The gate bias VG provides the control of surface carrier densities.
If VG > Vth, an inversion layer exists. ∴ a conducting channel exists from D → S and current ID will flow. Vth is determined by the properties of the structure and is given by:
For VG < Vth, the structure consists of two diodes back to back and only leakage current flows (≈ Io of PN junctions); i.e., ID ~ 0.
CNq
C
QV
ox
FSiSUBF
ox
fMSth
)2(22
φεφφ ++−= (5)
Strong Inversion Region
HO #17: ELEN 251 - MOS DC Models Page 13S. Saha
Note:• The depletion region is wider around the drain because of
the applied drain voltage VD.• The potential along the channel varies from VD at y = L to 0
at y = 0 between the drain and source.• The channel charge QI and the bulk charge Qb will in
general be f(y) because of the influence of VD, i.e. potential varies along the L only ⇒ Gradual channel approximation.
+VD+VG
P
ID
QI(y) Qb(y)n+ n+
Inversionlayer
Depletionregion
y
x
0=y Ly =
Drain Current Model
HO #17: ELEN 251 - MOS DC Models Page 14S. Saha
From (4), the voltage drop across an elemental length dyin the MOSFET channel is:
Now, at any point in silicon the induced charge due to VGis: Qs(y) = QI(y) + QB(y) (7)
Again, VG = VFB − Qs/COX + φs (8)
Here VFB includes φMS and Qf. Combining (7) and (8): QI(y) = −{VG − VFB − φs(y)}COX − QB(y) (9)
Since the surface is inverted, φs ≅ 2φB (≡ 2φF) plus any reverse bias between the channel and substrate (due to VD or VBS). ∴φs(y) = V(y) + 2φB (10)
)(yQWdyIdRIdV
In
DID µ
== (6)
Drain Current Model
HO #17: ELEN 251 - MOS DC Models Page 15S. Saha
Also, we know from MOS capacitor analysis that the bulk or depletion charge at any point y in the channel is:
Note that as we move from S → D, V(y)↑ due to IR drop in the channel.∴ Xdep↑ as we move toward the drain and
Qb↑ as we move toward the drain.
Substituting (10) and (11) into (9), we get the expression for inversion charge:
]2)([2)()( BSUBSidepSUBb yVNqyXqNyQ φε +−=−= (11)
]2)([2
}2)({)(
BSUBSi
OXBFBGI
yVNq
CyVVVyQ
φε
φ
++
−−−−=
(12)
Basic Drain Current Model
HO #17: ELEN 251 - MOS DC Models Page 16S. Saha
Substituting (12) in (6), we get:
If we assume Qb(y) = constant, i.e. neglect the influence of channel voltage on Qb, then:
In (13) Vth includes the effects of φB, φMS, and Qb(y = 0).
Integrating within the limits:
( )dVyVNqCyVVVWdyI BSUBSioxBFBGsD ]2)([2}2)({ φεφµ +−−−−−=
(13)[ ]dVyVVVCW
dVC
yVNqyVVVCWdyI
thGoxs
OX
BSUBSiBFBGoxsD
)(
]2)([22)(
−−−=
⎟⎟⎠
⎞⎜⎜⎝
⎛ +−−−−−=
µ
φεφµ
⇒⎭⎬⎫
⎩⎨⎧
==
⎭⎬⎫
==
DVVLy
toVy
00
DD
thGoxsD VVVVCL
WI ]2
[ −−= µ (14)
Basic Drain Current Model: Linear Region
HO #17: ELEN 251 - MOS DC Models Page 17S. Saha
Eq. (14) is the Level 1 MOSFET IDS model. If we define, κ ≡ µsCox = process transconductance β ≡ µsCox(W/L) = κ(W/L) = gain factor of the device
If VD < 0.1 V, we can simplify (15) to:
(16) shows that current varies linearly with VD. This is defined as the linear region of operation. From (16), the effective resistance between the source and drain is:
(15)DD
thGD VVVVI ]2
[ −−= β
(16)( ) DthGD VVVI −≅ β
(17)( )thGD
Dch VVI
VR−
≅=β
1
HO #17: ELEN 251 - MOS DC Models Page 18S. Saha
• ID Vs. VD from (15) with different VG show ID↓ for higher VD. While the measured ID saturates at higher VD. This discrepancy is due to the breakdown of gradual channel approximation near the drain–end of the channel at high VD.
• Eq. (15) is valid only as long as an inversion layer exists all the way from S → D.
Basic Drain Current Model: Linear Region
• The maximum value of ID vs. VD plots occurs at:VD = VG − Vth ≡ VDSAT = drain saturation voltage (18)
• For VD > VDSAT, a channel will not exist all the way to the drain.
HO #17: ELEN 251 - MOS DC Models Page 19S. Saha
Basic Drain Current Model: Saturation
HO #17: ELEN 251 - MOS DC Models Page 20S. Saha
Basic Drain Current Model: Saturation• When VD > VDSAT,
– e- travelling in the inversion layer are injected into the pinch-off region near the drain-end.
+VD > VDSAT+VG
P
n+
Inversion layerends (pinch off)
Depletionregion
n+
length ofpinch =Dl
VD
ID 6
5
4321
VG - Vth
LinearRegion Saturation
Region
– The high ε-field in the pinch-off region pulls e- into the drain.
– further increase of VD do not change ID (to a first order).
∴ ID ≅ constant for VD > VDSAT
– The boundary between the linear and the saturation regions is described by: VG − Vth = VDSAT.
Basic Drain Current Model: Saturation
HO #17: ELEN 251 - MOS DC Models Page 21S. Saha
Substituting (18) in (16), we can show that the saturation drain current is given by:
⇒ square law theory of MOSFET devices.
At VD > VDSAT, channel is pinched-off and the effective channel length is given by: Leff = L − ld = L(1 - ld/L)
Typically, ld << L, then from (20):
( )2
2 thGOXnDSAT VVCL
WI −≅ µ (19)
( ) ( )⎟⎠⎞
⎜⎝⎛ −
=−−
≅∴
Ll
IVVClL
WId
DSATthGOXn
dD
122µ (20)
⎟⎠⎞
⎜⎝⎛ +≅
LlII d
DSATD 1 (21)
Basic Drain Current Model: Saturation
HO #17: ELEN 251 - MOS DC Models Page 22S. Saha
Since ld depends on (VD - VDSAT), we can write:
Here, λ = channel length modulation (CLM) parameter.
)(11 DSATDd VVLl
−+≡+ λ (22)
( ))(1 DSATDDSATD VVII −+≅∴ λ (23)
From (23), ID = 0 at (VD−VDSAT) = −1/λ.
∴ λ can be extracted from the VD-intercept of ID vs. VD plots at ID = 0 as shown in Fig.
VG5
VG4
VG3
VG2
VG1
1/λ VD
ID VG5 > VG4 > VG3 > VG2 > VG1
0CLMtodue ageEarly volt1
=≡ CLMAVλ
Basic Drain Current Model: Summary
HO #17: ELEN 251 - MOS DC Models Page 23S. Saha
• Current Eq:
• Model Parameters:– VTO = threshold voltage at VB = 0– KP = process transconductance– GAMMA = body factor– LAMBDA = channel length modulation factor– PHI = 2|φB| = bulk Fermi-potential.
(24)
( ) ( ).12
]2
[
0
2DthG
DD
thG
VVV
VVVV
λβ
β
+−
−−
VG ≤ Vth (cut-off region)
VG ≥ Vth, VD ≤ VDSAT (linear region)
VG ≥ Vth, VD ≥ VDSAT (saturation region)
=DI
Basic Drain Current Model: Summary
HO #17: ELEN 251 - MOS DC Models Page 24S. Saha
• Assumptions: – gradual channel approximation (GCA) is valid– majority carrier current is negligibly (such as hole current for
nMOSFETs is neglected)– recombination and generation are negligible– current flows in the y-direction (along the length of the
channel) only– carrier mobility µs in the inversion layer is constant in the y-
direction– current flow is due to drift only (no diffusion current)– bulk charge Qb is constant at any point in the y-direction.
The accuracy of The basic model is poor even for long channel devices.
Bulk Charge Effect
HO #17: ELEN 251 - MOS DC Models Page 25S. Saha
In reality, Qb varies along the channel form the source at y = 0 to drain at y = L because of the applied bias VD. Then from (11) with back bias VBS we have:
For the simplicity of computation, we simplify (25) by Taylor series expansion and by neglecting higher order terms, we get:
)(2
]2)([2)()(
yVVC
VyVNqyXqNyQ
BSBox
BSBSUBSidepSUBb
++−=
++−=−=
φγ
φε
(25)
( ))(2)(2)( yVVyVVVCyQ BSBBFBGoxI ++−−−−−=∴ φγφ (26)
( )( )
( )( )
[ ])(.2)(2
5.02
...2
)(212)(
yVVCyVV
VC
VyVVCyQ
BSBoxBSB
BSBox
BSBBSBoxb
δφγφ
φγ
φφγ
++=⎥⎥⎦
⎤
⎢⎢⎣
⎡
+++≅
⎥⎥⎦
⎤
⎢⎢⎣
⎡−
+++=
(27)
Bulk Charge Effect
HO #17: ELEN 251 - MOS DC Models Page 26S. Saha
BF V+≅
φδ
25.0 where
,(27) and (26) From( )( )
{ }( )( )( )
( ))()(122
)(.22)(
)(22)(
)(2)()(
yVVVCyVVVVC
yVVyVVVC
yVVyVVVC
yQyVVVCyQ
thGox
BSBFFBGox
BSFBFBGox
BSFBFBGox
bBFBGoxI
αδγφγφ
δφγφ
φγφ
φ
−−−=
+−+−−−−=
++−−−−−=
++−−−−−=
+−−−−=
(28)
BSB V++≅≡
φγα
25.01factor chargebulk where (29)
models. advanced develop to(28) use willWe DSI
MOSFET IDS Model with Bulk Charge Effect
HO #17: ELEN 251 - MOS DC Models Page 27S. Saha
• Qb variation along the channel offers more accurate IDmodeling in the linear and saturation region.
• The simplified MOSFET drain current model using the bulk-charge factor α as a fitting parameter is given by:
(30)( ) ( ).1
2
]2
[
0
2DthG
DD
thG
VVV
VVVV
λαβ
αβ
+−
−−
VG ≤ Vth (cut-off region)
VG ≥ Vth, VD ≤ VDSAT (linear region)
VG ≥ Vth, VD ≥ VDSAT (saturation region)=DI
(31)( ) α/Where thGDSAT VVV −=
BSB V++=
φγα
25.01 (29)
High-Field Effects in IDS
HO #17: ELEN 251 - MOS DC Models Page 28S. Saha
• Surface mobility µs = constant assumed in MOSFET current expressions is not true under high VG and VD.
• As the vertical field Ex and lateral field Ey increase with increasing VG and VD, respectively, carriers suffer increased scattering. ∴ µs = f(Ex, Ey)
• For the simplicity of ID calculation an effective mobilitydefined as the average mobility of carriers is used:
(32)∫∫=invinv XX
seff dxyxndxyxnyx00
),(),(),(µµ
dVQL
WIDSV
IeffD ∫=∴0
µ (33)
Effect of High Vertical Electric Fields
HO #17: ELEN 251 - MOS DC Models Page 29S. Saha
In reality, µs is highly reduced by large vertical e-fields. The vertical e-field pulls the inversion layer e- towards the surface causing: VD
+VG
P
n+n+S
yx
Ex1 more surface scattering2 coulomb scattering due to
interaction of e- with oxide charges (Qf, Nit).
Since e-field varies vertically through the inversion layer, the average field in the inversion layer is: Eeff = (Ex1 + Ex2)/2 (34)
where Ex1 = vertical e-field at the Si-SiO2 interface Ex2 = vertical e-field at the channel-depletion layer interface.
From Gauss’ Law: Ex1 − Ex2 = Qinv/Ksεo and Ex2 = QB /Ksεo (35)
∴ From (34) & (35) we can show: Eeff = [(0.5Qinv + QB)]/Ksεo (36)
Universal Mobility Behavior
HO #17: ELEN 251 - MOS DC Models Page 30S. Saha
• In general, Eeff = [(QB + ηQinv)]/Ksεo
where for <100> Si η = 1/2 for electrons η = 1/3 for holes.
• Measured µeff vs. Eeffplots at low VD show:– a universal behavior
independent of doping concentration at high vertical fields.
– dependence on: 1) doping concentration and 2) interface charge at low vertical fields.
universal behavior
Universal Mobility Behavior
HO #17: ELEN 251 - MOS DC Models Page 31S. Saha
The experimentally observed mobility behavior is due to the relative contributions of different scattering mechanisms set by the strength of vertical e-fields:1 Coulomb
scattering by ionized impurities and oxide charges.
2 Phonon scattering
3 Surface roughness scattering at the Si-SiO2interface.
Universal Mobility Behavior
HO #17: ELEN 251 - MOS DC Models Page 32S. Saha
• Surface roughness scattering↑ as carrier confinement close to the interface↑ at high vertical e-fields. ∴ µeff↓ as Eeff↑.
• The experimentally observed universal behavior occurs because phonon scattering is weakly dependent on vertical e-fields.
• Deviations from the universal behavior occur in heavily doped substrates at low fields due to:– dominant ionized impurity scattering at low inversion charge
densities. ∴ µeff = f(Nch).– coulomb scattering by
♦ ionized impurities in the depletion region♦ oxide charges.
• Phonon scattering has the strongest temperature dependence on µeff .
Effective Mobility due to High VG
HO #17: ELEN 251 - MOS DC Models Page 33S. Saha
Substituting for Qb from Vth Eqn and QI in (36), we get:
The dependence of µeff on VG is described by an empirical relation:
Where µo = the maximum extracted value of low-field mobility at a
given doping concentration ≡ low-field surface mobility. ν ≅ 0.25 for electrons and ν ≅ 0.15 for holes. Ec ≅ 2.7x104 V/cm = critical e-field above which µo↓.
OX
thGSeff T
VVE6
+= (37)
(38)νµµ
⎟⎟⎠
⎞⎜⎜⎝
⎛+
=
0
0
1EEeff
eff
Mobility Degradation due to VG
HO #17: ELEN 251 - MOS DC Models Page 34S. Saha
By Taylor’s series expansion of (38) and introducing VBSdependence, the vertical field mobility degradation model can be shown as:
where Ua and Ub are the model parameters extracted from ID- VG characteristics.
The VBS dependence is included by model parameters Uc:
20
1 ⎟⎟⎠
⎞⎜⎜⎝
⎛ ++⎟⎟
⎠
⎞⎜⎜⎝
⎛ ++
=
OX
thGSb
OX
thGSa
eff
TVVU
TVVU
µµ (39)
( )2
0
1 ⎟⎟⎠
⎞⎜⎜⎝
⎛ ++⎟⎟
⎠
⎞⎜⎜⎝
⎛ +++
=
OX
thGSb
OX
thGSBSca
eff
TVVU
TVVVUU
µµ (40)
Effect of High Lateral Electric Fields
HO #17: ELEN 251 - MOS DC Models Page 35S. Saha
Additional complication arises because of the high lateral e-fields.
We see that for e- in silicon vd saturates near E ~ 104
V/cm and vd = µE does not hold.
Since average e-field for short channel devices > 104
V/cm.
∴small geometry MOSFET devices will operate at vd = vsat ≅ 107 cm/sec.
Since µ ≠ constant, we must account for high lateral e-field effects in the expression for ID derived from simple theory.
Mobility Degradation due to VD
HO #17: ELEN 251 - MOS DC Models Page 36S. Saha
• The velocity saturation of inversion carriers due to increased lateral field Ey causes:– current saturation sooner than predicted by VDSAT = VG – Vth
– lower IDSAT than predicted by simple theory.
• The drift velocity due to high field effect is given by:
Where EC ≡ vsat/µeff = critical lateral field for velocity saturation β = 2 for electrons; β = 1 for holes.
• For simplicity of numerical solution, β = 1 is used.
( )[ ] c
Cy
yeffd EE
EE
Ev >
+= for,
11 ββ
µ
satyeff
yeffd vE
Ev
µµ
+=∴
1 (41)
Mobility Degradation due to VG and VD
HO #17: ELEN 251 - MOS DC Models Page 37S. Saha
• The effective mobility due to the combine effect of VG and VD is given by:
Where − θC ≡ 1/LEC
– L = channel length of MOSFETs.
• Thus, µeff is modeled by the parameter set:– {(µ0, θ, θb, vsat (EC)}.
– the model parameters are obtained by curve fitting the experimental data to the model equation.
( ) DcBbthGeff VVVV θθθ
µµ++−+
=1
0 (42)
IDS Model in Strong Inversion
HO #17: ELEN 251 - MOS DC Models Page 38S. Saha
vsat
v
EcE
µo
Assume a piecewise linear model, i.e., vd saturates abruptly at E = Ec:
( )
( )csat
c
c
effd
EEv
EE
EEE
v
≥=
≤+
=
,1
,
µ
(43)
Where E = lateral e-field and Ec = critical e-field at which carriers are velocity saturated, i.e. v = vsat.
Now, the current at any point y in the channel is given by: ID = I(y) = WCox[VG − Vth − αV(y)]v(y) (44)
Where V(y) = potential difference between the drain and channel at y. v(y) = carrier velocity at any point y in the channel.
IDS Model in Strong Inversion
HO #17: ELEN 251 - MOS DC Models Page 39S. Saha
Substituting (43) in (44) we get:
Integrating (45) from y = 0 to y = L with corresponding V(y) = 0 to V(y) = VD, we get in the linear region (VDS ≤VDSAT) current:
[ ] dyydV
EIyVVVCW
IyE
c
DthGoxeff
D )(
)()( −=
−−−=
αµ (45)
⎟⎟⎠
⎞⎜⎜⎝
⎛+
⎟⎠⎞
⎜⎝⎛ −−
=
LEVL
VVVVCWI
c
DS
DSDSthGoeff
D
1
21 αµ
(46)
IDS Model in Strong Inversion
HO #17: ELEN 251 - MOS DC Models Page 40S. Saha
Let us define, VDSAT ≡ the drain saturation voltage due to vsat, i.e. at E = Ec.
Using this condition in (45), we get in the saturation region (VDS ≥ VDSAT):
Including channel length modulation (CLM):
where VACLM = channel length modulation parameter
( )2
cDSATthGoxeffD
EVVVCWI
αµ −−= (47)
(48)( ) ⎟⎟⎠
⎞⎜⎜⎝
⎛ −+−−=
ACLM
DSATDSDSATthG
coxeffD V
VVVVVECW
I 12
αµ
IDS Model in Strong Inversion
HO #17: ELEN 251 - MOS DC Models Page 41S. Saha
Since ID given by (46) and (47) must be continuous @ VDS= VDSAT, therefore, equating (46) = (47):
Note that from (43), vsat = µeffEc/2. Thus, the drain current models in strong inversion:
(49)( )
( )thGC
thGCDSAT VVLE
VVLEV−+
−=
α
DSATDSthGSDSDSthG
c
DS
oeffD VVVVVVVV
LEVL
CWI <>⎟
⎠⎞
⎜⎝⎛ −−
⎟⎟⎠
⎞⎜⎜⎝
⎛+
= ,for ,21
1α
µ
( ) DSATDSthGSA
DSATDSDSATthGsatoxD VVVV
VVVVVVvWCI >>⎟⎟
⎠
⎞⎜⎜⎝
⎛ −+−−= ,for ,1α
(50)
3. Sub-threshold Region Model
HO #17: ELEN 251 - MOS DC Models Page 42S. Saha
Simple long channel device equations assume: IDS = 0 for VGS< Vth. In reality, IDS ≠ 0 and varies exponentially with VGS in a manner similar to a bipolar transistor.
IDS
VGSVth
In order to develop a theory of sub-threshold conduction, let us consider MOS band diagram with applied source (S) and gate (G) bias measured with respect to the substrate (Sub), that is:– VSSub
– VGSub.
Ev
EF
Gate SiO2 p-Si
Eiφs
VGS
Ec
φB
VSSubVGSub
φs−VSSub− φF
Sub-threshold Region Model
HO #17: ELEN 251 - MOS DC Models Page 43S. Saha
In the sub-threshold (weak inversion) region, we know:
1 as EF is pulled above Ei, the number of minority carrier e−
at the surface increases exponentially with EF − Ei
2 when φs = 2φB, strong inversion (nsurf = pbulk) is achieved and we obtain Vth.
Ec
EiEF
EvφB
φs
3 For φs < 2φF, the dominant charges present near the surface are ionized acceptor atoms, i.e. nsurf << NA
−.♦ Thus, there is no ε-field laterally along the surface since
Poisson’s equation is the same everywhere.♦ Thus, any current flow must be due to diffusion only.
Sub-threshold Region Model
HO #17: ELEN 251 - MOS DC Models Page 44S. Saha
n+
P
n+- - - - - - -- - - - - - - - - - - - -
- - - - - --- - -- -
y →
NA−- - - - -
So few e- that εx = 0
dydnAqDID −=∴ (51)
4 The e- gradient along the channel (dn/dy) must be constant in order to maintain constant current.
∴ ID = −AqD[{n(0) − n(L)}/L] . . . (52)
Now, we can now use carrier statistics to calculate n(0) and n(L). Referring band diagram on page 41:
Where φs is the surface potential with respect to the substrate. Also, note that φs = constant along the channel [ε(y) = 0].
( )
( )kT
Vq
i
kTVq
iBDSubs
BSSubs
enLn
ennφφ
φφ
−−
−−
=
=
)(
)0( (53)
(54)
Sub-threshold Region Model
HO #17: ELEN 251 - MOS DC Models Page 45S. Saha
Since the charge in the substrate is assumed uniform (Nsub), then from Poisson’s equation:
∴ φ varies parabolically and E varies linearly with distance. Since the e- concentration falls off as e−qφ/kT away from the
surface, essentially, all of the minority carrier e- are contained in a region in which the potential drops by kT/q.∴ The depth of the inversion layer, Xinv = ∆φ/Es.
where ∆φ = kT/q, and Es is the ε-field at the surface. Again, from Gauss’ Law:
dxdEqN
dxd
s
A −==ε
φ2
2
(55)
sSUBsbss qNQE φεε 2=−= (56)
sSUB
s
sSUBs
s
inv qNqkT
qNq
kTX
φε
φε
ε
22==∴ (57)
Sub-threshold Region Model
HO #17: ELEN 251 - MOS DC Models Page 46S. Saha
Using (53), (54) and (57) in (52), we get:
Here A = WXinv. Using VDSub = VDS + VSSub,
To make use of this equation, we need to know how φsvaries with the externally applied potential VG.
( ) ( )
sA
skTVq
kTVq
iD qNqkTeeDn
LWqI
FDSubsFSSubs
φεφφφφ
2⎭⎬⎫
⎩⎨⎧
−=−−−−
(58)
( )
⎭⎬⎫
⎩⎨⎧
−⎭⎬⎫
⎩⎨⎧
=−
−−kT
qVkT
Vq
sSUB
siD
DSFSSubs
eeqN
kTDnL
WI 12
φφ
φε
(59)
ox
sSUBssFBGSub C
NqVV
φεφ
2++= (60)
Sub-threshold Region Model
HO #17: ELEN 251 - MOS DC Models Page 47S. Saha
Generally, it is more common to use the source potential as the reference so that: VGSub = VGS + VSSub
φs = ψs + VSSub
The depletion layer capacitance is: Therefore, from (62):
( )
⎭⎬⎫
⎩⎨⎧
−⎭⎬⎫
⎩⎨⎧
=∴−
−kT
qVkT
q
sSUB
siD
DSFs
eeqN
kTDnL
WI 12
φψ
φε
(61)
( )ox
SSubsSUBssFBGS C
VNqVV
+++=
ψεψ
2 (62)
( )SSubs
SUBsD V
NqC+
=ψ
ε2
(63)
( ) nCC
VNq
CddV
ox
D
SSubs
As
oxs
GS ≡+=+
+= 12
11ψ
εψ
(64)
Sub-threshold Region Model
HO #17: ELEN 251 - MOS DC Models Page 48S. Saha
In order to eliminate ψs from (61) we expand VGS in a series about the point ψs = 1.5φF (weak inversion corresponds to φF≤ ψs ≤ 2φF).
Where and is obtained from (62).
Combining (64) with (61) to eliminate ψs in the exponential and using (63) to eliminate the square root of ψs in (61), we obtain:
FsGSGS VV
φψ 5.1*
=≡
( )FsGSGS nVVFs
φψφψ
5.15.1
−+≅∴= (64)
( )
⎭⎬⎫
⎩⎨⎧
−⎪⎭
⎪⎬⎫
⎪⎩
⎪⎨⎧
⎟⎟⎠
⎞⎜⎜⎝
⎛=
−⎥⎥⎦
⎤
⎢⎢⎣
⎡−
−
kTqV
kTq
nkTVVq
iDD
DSFGSGS
eenCq
kTL
WI 122 * φ
µ (65)
Sub-threshold Region Model
HO #17: ELEN 251 - MOS DC Models Page 49S. Saha
Thus, the sub-threshold current is given by:
From (47) we note that:• ID depends on VDS only for small VDS, i.e. VDS ≤ 3kT/q, since
exp[−qVDS/kT) → 0 for larger VDS.
• ID depends exponentially on VGS but with an “ideality factor” n > 1. Thus, the slope is poorer than a BJT but approaches to that of a BJT in the limit n → 1.
• NSUB and VSSub enter through CD.
( )
⎭⎬⎫
⎩⎨⎧
−⎪⎭
⎪⎬⎫
⎪⎩
⎪⎨⎧
⎟⎟⎠
⎞⎜⎜⎝
⎛=
−⎥⎥⎦
⎤
⎢⎢⎣
⎡−
−
kTqV
kTq
nkTVVq
iDD
DSFGSGS
eenCq
kTL
WI 122 * φ
µ (66)
Sub-threshold Slope (S-factor)
HO #17: ELEN 251 - MOS DC Models Page 50S. Saha
In order to change ID by one decade, we get from (47):
1.E-12
1.E-111.E-10
1.E-091.E-08
1.E-07
1.E-061.E-05
1.E-04
0.0 0.2 0.4 0.6 0.8 1.0
VGS (V)
IDS
(A)
( )ndecade
mVCC
qkTSlope
o
D
60
110ln
≅
⎟⎟⎠
⎞⎜⎜⎝
⎛+=
(@ room T)
Vth
⎟⎟⎠
⎞⎜⎜⎝
⎛+≅∴
⎟⎟⎠
⎞⎜⎜⎝
⎛+==⇒=
∂
CC
decadeImVS
CC
qkTn
qkTS
nkTVq
o
D
o
DGS
160
110ln)(10ln10ln
(@ room T) . . . (67)
Sub-threshold Model - Final Note
HO #17: ELEN 251 - MOS DC Models Page 51S. Saha
• In weak inversion or subthreshold region, MOS devices have exponential characteristics but are less “efficient” than BJTs because n > 1.
• Subthreshold slope S does not scale and is ≈ constant. Therefore, Vth can not be scaled as required by the ideal scaling laws.
• VDS affects Vth as well as subthreshold currents.
• In order to optimize S, the desirable parameters are:
– thin oxide
– low NA
– high VSub.
Sub-threshold Region Model
HO #17: ELEN 251 - MOS DC Models Page 52S. Saha
The sub-threshold current Eq. used in BSIM model is:
where Voff = offset voltage is a model parameter.
Thus, the piece-wise drain current models for different regions of MOSFET operations:
(68)( )
thGSoffthGDS
onDS VVnkT
VVVqnkTqVII <⎟⎟
⎠
⎞⎜⎜⎝
⎛ −−⎟⎠⎞
⎜⎝⎛ −−= for ,exp)exp(1
DSATDSthGSDSDSthG
c
DS
oxeff VVVVVVVV
LEVL
CW<>⎟
⎠⎞
⎜⎝⎛ −−
⎟⎟⎠
⎞⎜⎜⎝
⎛+
,for ,21
1α
µ
( ) DSATDSthGSA
DSATDSDSATthGsatox VVVV
VVVVVVvWC >>⎟⎟
⎠
⎞⎜⎜⎝
⎛ −+−− ,for ,1α
( )thGS
offthGDSon VV
nkTVVVq
kTqVI <⎟⎟
⎠
⎞⎜⎜⎝
⎛ −−⎟⎠⎞
⎜⎝⎛ −− for,exp)exp(1
=DI
MOS Threshold Voltage, Vth Extraction
HO #17: ELEN 251 - MOS DC Models Page 53S. Saha
Vth is obtained by linear extrapolation from the maximum slope to Ids = 0 of Ids - Vgs plot.
We know,
Define, Vth ≡ Vgs @ Ids = 0 and Vds = 50 mV
dsds
thgsds VVVVI ⎟⎠⎞
⎜⎝⎛ −−=
2β
2
02
dsgsth
dsds
thgsds
VVV
VVVVI
−=⇒
=⎟⎠⎞
⎜⎝⎛ −−=∴ β ID
VG
VD = 50 mV
2ds
gsthVVV −=
Substrate Bias Dependence of Vth
HO #17: ELEN 251 - MOS DC Models Page 54S. Saha
Vds = 0.05VTox = 1.5nm
Substrate Bias Dependence of Vth
HO #17: ELEN 251 - MOS DC Models Page 55S. Saha
Substrate bias dependence of Vth for uniformly doped substrate is given by:
γ is obtained from:
( )φφγ FSubFthth VVV 220 −±±=
factorbody2
where
≡=C
KNq
ox
ossub εγ
γ
Fφγ 2BSF V+φ2
0thth VV −
FBSFthth VVsVV φφ 22.)( 0 −+−
φγγ
2interceptslope
where
=−
=
y
Substrate Bias Dependence of Vth
HO #17: ELEN 251 - MOS DC Models Page 56S. Saha
Substrate bias dependence of Vth for non-uniformly doped substrate is given by the plots:
where γ1(k1) = body effect due to channel concentration
γ2(k2) = body effect due to substrate concentration.
Vth is given by:
FBSFthth VVsVV φφ 22.)( 0 −+−
)( 11 kγ
Fφγ 2BSF V+φ2
0thth VV −)( 22 kγ
( ) ( )φφγφφγ FSubFFSubFthth VVVV 2222 210 −−+−−+=
Drain Induced Barrier Lowering (DIBL)
HO #17: ELEN 251 - MOS DC Models Page 57S. Saha
• DIBL is defined as the shift in Vth due to Vds, especially, in short channel devices. DIBL is defined as:1) ∆Vth ≡ Vth(Vds-low) − Vth(Vds = Vdd)
2)
• DIBL is calculated from:– log(Ids) Vs. Vgs plots at
Vds-low = 50 mV Vds = Vdd.
Example: From Figure we get, ∆Vth ≅ (0.32 - 0.24) V
= 80 mV
)()()(
lowdsdd
dddsthlowdsth
ds
th
VVVVVVV
VV
−
−
−=−
≡∂∂
Sub-threshold Slope (S)
HO #17: ELEN 251 - MOS DC Models Page 58S. Saha
• S is the inverse of log(Ids) Vs. Vgs plot at Vds-low and is given by:
• To extract S:– extract:
Ids1 = Ids(Vth)Ids2 = 2 dec. below
Ids1
– Calculate slopeS = 1/slope.
( )⎥⎦⎤
⎢⎣
⎡⋅≡
)log(3.2
ds
gs
IddV
S
Ion and Ioff
HO #17: ELEN 251 - MOS DC Models Page 59S. Saha
• Ion and Ioff can be extracted from Ids − Vgs plot at Vds = Vdd.
• Ion ≡ Idsat atVds = Vgs = Vdd
• Ioff ≡ Ids atVds = Vdd
Vgs = 0
• Example: From Fig.,– Ion ≅ 440 µA/µm– Ioff ≅ 3 nA/µm
Home Work 7: Due June 2, 2005
HO #17: ELEN 251 - MOS DC Models Page 60S. Saha
VGS (V) VDS (V) VBS (V) ID (mA)2 5 0 405 5 0 5365 5 -5 3605 8 0 6445 5 -3 420
1) For a silicon MOSFET, considering bulk charge effect in current ID calculate (a) VTO, (b) LAMBDA, (c) GAMMA, and (d) BETAfrom the measured data shown in table.
2) If Eeff = [0.5Qinv + QB]/εsi, where Qinv and QB are the inversion charge and bulk/depletion charge under the gate, respectively and εsi is the dielectric constant of silicon. The dependence of surface mobility, µs on process parameters such as TOX , Nsub etc. and terminal voltages is lumped in Eeff. Assume VGS > Vth and small VDS:
(a) Show that Eeff ≅ (VGS + Vth)/6TOX.(b) If the effective mobility is modeled by: µeff = µ0/[1 + Eeff/E0)]η, where µs = µ0 @ VGS = 0
and E0 and η are parameters determined from the measured data. Use the expression for Eeff in part (a) to show that:
Where Ua and Ub are the model parameters that are determined experimentally.
20
1 ⎟⎟⎠
⎞⎜⎜⎝
⎛ ++⎟⎟
⎠
⎞⎜⎜⎝
⎛ ++
=
OX
thGSb
OX
thGSa
eff
TVVU
TVVU
µµ