CMOS Modeling

1
Confidential!
CMC MeetingBoston, May 8th, 2006
PSP101 Model Evaluation Joachim
Assenmacher Infineon Technologies AG COM BTS TD
DIF
Infineon
Never stop thinking.
2
Outline

• Sub-threshold Current Modeling
• Extraction of Global Parameters for NEFF and
  DPHIB L-Scaling
• Depletion Capacitance Modeling
• Gate-to-Channel Capacitance in Depletion and
  Accumulation
• Conclusions

3
PSP100 Weak and Moderate Inversion Region
Modeling log Id-Vg of short channel NMOS devices
PSP100
transfer characteristics
VdsVdd saturation
Vds50mV linear mode
Sub-threshold current, incl. sub-Vt slope and Vt
for higher back-biases, of short channel devices
cant be modeled accurately with PSP100.
4
PSP101 Weak and Moderate Inversion Region
Modelinglog Id-Vg of short channel NMOS devices
PSP101
transfer characteristics
VdsVdd saturation
Vds50mV linear mode
Modeling of sub-threshold current is
significantly improved with PSP101, due to
introduction of length dependence of coulomb
scattering and DPHIB (offset of FB)
5
PSP101 L-dependence of CS Analog Gain Factor K
Modeling for Short Channel Devices
Improved K modeling for short channels in the
moderate inversion region with PSP101, due to
introduction of L-dependence of CS (and DPHIB).
CS L-scaling
Kgmsat2 /(2?Idsat)
NMOS 2µm/0.06µm
moderate inversion
moderate inversion
PSP101
PSP100
6
PSP100 NSUB vs. PSP101 NEFF L-Scaling
Extraction of length scaling global parameters
for effective doping
NSUB and F0 mainly influence Vt and NSUB partly
based on body effect
NEFF sets Vth0 and models body effect
PSP100 NSUBF0 vs. channel length
PSP101 NEFF vs. channel length
In PSP101, the local parameter values of NEFF
shows scattering and not a similar trend as
NSUBF0 in PSP100, whereas the NEFF scaling
equation is identical to NSUBF0 scaling equation
(F0 lateral gradient factor coefficient). Assumpt
ion Because of the changed FB implementation
(new parameter DPHIB), NEFF in PSP101 and F0NSUB
in PSP100 shows different behavior.
7
PSP100 AF vs. PSP101 DPHIB L-Scaling
Extraction of length scaling global parameters
for effective doping
DPHIB is for Vt adjustment
AF is for body effect adjustment
PSP100 DPHIB vs. channel length
PSP101 AF vs. channel length
The L-scaling trend of PSP100 AF versus PSP101
DPHIB is nearly the same, but the electrical
effect in the model is different. Why has DPHIB
now, almost nothing to do with body effect
anymore (same reason as before)? Note The
lateral gradient factor F (AF body effect of F)
from PSP100 has been removed for PSP101
(motivated by reciprocity requirement).
8
PSP100 vs. PSP101 Depletion Capacitance
Modeling(Wide Long Device)
Cgg
NMOS 150x 5µm x 5µm island area cap.
PSP100
PSP101
The depletion capacitance modeling with PSP101
has introduced an offset in the total intrinsic
gate capacitance. Assumption Drawback of NEFF
implementation in PSP101. PSP101 Cdep
(esi/d)A (0.5qesiNEFF)1/2 () instead of
PSP100 Cdep (esi/d)A (0.5qesiNSUB)1/2
() ? simple formulae just for illustration.
9
PSP100 vs. PSP101 Cgds-Modeling in Depletion and
Accumulation Mode of Short Channel Devices
NMOS 120x 32µm x 0.08µm
Cgds (zoom in)
PSP100
PSP101
Since the inner fringe capacitance has been
removed in PSP101, the modeling of bias-dependent
overlap capacitance is worse than in PSP100.
10
Conclusions

• Modeling of the subthreshold and moderate
  inversion region has been significantly improved
  with PSP101, due to introduction of L-dependence
  of CS and DPHIB.
• Side-effects of PSP101 w.r.t. depletion
  capacitance needs to be investigated by the PSP
  team.
• Would be great, if the PSP team find a way to
  re-introduce inner fringe capacitance (like in
  PSP100), without sacrifying reciprocity of caps
  at Vds0V. This will lead to higher accuracy in
  the modeling of the bias-dependent overlap
  capacitance (i.e. more flexibility to adjust
  slope of Cgds in depletion and accumulation).

11
Backup PSP101 Geometry Dependent Bulk
Potential and Doping

• Picks up some functionalities originally placed
  in geometry dependence of lateral gradient factor.

Note PSP101 model code has a bug in DPHIB
scaling rule (see eq. above) must be (DPHIBO
DPHIBL)(1DPHIBW) according to the manual!