Title: PSP and HiSIM2 Model Evaluation Yutao Ma, MinChie Jeng
1PSP and HiSIM2 Model EvaluationYutao Ma,
Min-Chie Jeng
2OUTLINE
- Model comparison of Hisim2 and PSP
- General impression
- Gate current model
- Rout model
- RF (Rg/Rb) network
- STI effect
- NQS model
- Code quality
- Version control
- Cadence support status on Hisim2/PSP
3General impression
- Both models have good symmetric characteristics
around Vds0 - Smooth transition between different operation
regions - Speed PSP Hisim2, slower than Bsim4
- Model documentation
- PSP Consistent with code
- Hisim2 Need improvement for completeness
- Extraction documentation
- PSP More detailed extraction guideline needed
- Hisim2 Provides better document, but still
inadequate - Release stability
- PSP Consistent between releases
- Hisim2 Too frequent releases with some
inconsistency( e.g. Rs/Rd implementation)
4Gate current modeling issue Missing Igbinv in
both models
5Rout modeling Missing SCBE
- SCBE ( Substrate Current Induced Body Effect ) is
missing in both PSP(?) and Hisim2 model - People might have to rely on Isub to fit Id
increase at high Vds bias.
6RF (Rg/Rb) network
- Gate resistance and substrate resistance networks
- Important for RF modeling
- Missing in hisim2 model
- Provided in PSP but only available in NQS model,
which is not implemented in C-code yet. - Suggestion to Hisim2 adding Rg/Rb network
- Suggestion to PSP move Rg/Rb network to non-nqs
version ( Possible in Verilog-A flow ? )
7STI ( LOD) effect
Comparison STI model in PSP, Hisim2 and Bsim4
8NQS model comparison
- Single NFET with parasitic resistance is used to
investigate NQS model. - S node voltage is plotted
- Netlist
- m D G S 0 nch w10u l3u
- rd 1 D 1k
- rs S 0 1k
- vd 1 0 2
- vg G 0 pwl(0 0 0.01n 0 0.011n 2)
- .print tran v(S)
9NQS model simulation results
- Both models show convergence to None NQS output
eventually - PSP results show humps in NQS model.
PSP NQS simulation results
Hisim2 NQS simulation results
10NQS model performance
- Verilog-A code is used for PSP model ( C-code NQS
not available ) - NQS model performance degradation is significant
for both PSP and Hisim2 model - Both model have room to improve NQS model
performance
Relative simulation time for NQS model w.r.t no
NQS case ( based on limited cases, more
investigation is underway )
11Discontinuity issue in Hsim2 caused by small size
effect
- The following code is causing problem
if ( WL lt model-gtHSM2_wl0 ) Leff Leff
model-gtHSM2_wl1 / pow( WL, model-gtHSM2_wlp )
12Performance comparison
- PSP has 2X speed improvement since last CMC
meeting. - PSP and Hisim2 have similar speed now within the
testing scope - Both PSP and Hisim2 are slower than Bsim4
13Code quality
- PSP
- Full derivatives of current and charge generated
by machine - Good parameter range clamping for almost all the
model/instance parameters. - Poor code readability and maintainability
- No NQS model in C-code
- Hisim2
- Readability and maintainability of the code is
good - More protection in code is needed to prevent
numerical problems like overflow and
divided-by-zero
14Version control
- Critical for user adoption and keep backward
compatible. - Neither of the models supports version control at
this moment. - Version control is more critical for PSP
- EDA vendors can not afford to add version control
in machine generated code. - The only practical way is to add version control
in verilog-A code before generating new C-code - Suggest both models to add version control in
future releases
15PSP and HiSIM2 Support in Cadence tools
- PSP (psp1000,psp100 and juncap200) level 100.1 is
supported in Spectre/Ultrasim/BsimPro - Hisim2 version 2.20 is supported in
Spectre/Ultrasim/BsimPro - PSP will be formally released in MMSIM60 USR2 in
Dec.2005. - Formal release of Hisim2 depends on when Hisim2
goes to public - Simulation/Optimization flow for PSP and Hisim2
in BsimProP are available now. - Extraction flow for PSP and HiSIM2 in BsimPro is
planned to be available in 1H06