The Designer's Guide Community Forum

The Designer's Guide Community Forum
https://designers-guide.org/forum/YaBB.pl
Modeling >> Semiconductor Devices >> CMOS models: which one ?
https://designers-guide.org/forum/YaBB.pl?num=1153214524

Message started by extractor_man on Jul 18^th, 2006, 2:22am

Title: CMOS models: which one ?
Post by extractor_man on Jul 18^th, 2006, 2:22am

Can someone help me to draw a picture of the advantages and weakness of most common MOSFET models for technologies < 0.25, 0.18 um in RF applications?
Specilal attention to: BSIM3V3, EKV2.6 and EKV3, and PSP.

In order to choose one model vs others I would like to know:

1- Maturity of the models: (still bugs to be fixed?)

2- Model implementation in EDA simulation tools like Spectre, Eldo, ADS and HSPICE (C code, VerilogA ADMS, Simkit), supported? what about the speed for each model and simulators?

3- Symetry modeling

4- parameter extraction procedure (easy or not): It seems that EKV models have less and uncorrelated parameters than any others. Is it really easier to extract parameters?

5- Accuracy, convergence issue reported?

6- Accurate modeling of IIP3 at VDS=0 and at higher voltages. Also important is to be able to simulate accurately the higher
order derivatives of the DC and RF I-V curves.

7- And ofcourse your own experiences and any comments on these models will be highly appreciated.

Many thanks for your support,

Regards,

Title: Re: CMOS models: which one ?
Post by Geoffrey_Coram on Jul 19^th, 2006, 11:24am

BSIM3v3 is certainly the most "mature" of the models. PSP and EKV3 are quite immature, in fact, I think no simulator yet has the official PSP 102 that was accepted as a standard model by the Compact Model Council.

However, BSIM3 is probably the worst of all those models in terms of symmetry, and will give you bogus IIP3 at VDS=0.

I've heard good things about EKV in terms of analog design and being able to make sense of parameters. However, I have no experience personally with using it. I think it's used more in Europe (STmicro, Infineon?).

Title: Re: CMOS models: which one ?
Post by extractor_man on Jul 26^th, 2006, 7:41am

Thanx Geoffrey.

EKV seems to be used more specificly in ATMEL.
As far as I know ST (+ freescale + Philips in Crolles II) and Infineon prepare the migration to PSP for the future generations and (RF applications).

The issue with EKV is that there are 2 versions! one (EKV2.6) with less parameters which is implemented in most simulators (but less accurate in RF applications) and EKV 3 which was one of the candidate for the 4th generation of MOS models in CMC. As PSP won , it seems that EDA companies do not anymore support and update EKV3 !!! Is it correct ?

What about PSP101 via Simkit ? Did you heard something regarding the bugs and problemes reported?

Thanks again for your support.

Title: Re: CMOS models: which one ?
Post by Geoffrey_Coram on Jul 31^st, 2006, 7:19am

I don't know off-hand whether EKV3.0 is public domain. HiSIM2 was also a candidate, and it was specifically NOT public domain (though the CMC would have required it of any candidate that was selected). EKV2.6 was not selected as a "standard model" but it is still supported; I would expect that EKV3.0 will similarly be supported, but not as quickly as PSP102 (which is the version selected as the standard). There were some changes between 101 and 102, including, I believe, a bug in how the junction diode parameters were mapped.

-Geoffrey

Title: Re: CMOS models: which one ?
Post by loose-electron on Aug 15^th, 2006, 9:26pm

A minor comment on models -

99% of the time the model is not the problem, rather it is the numbers put into the model.

Ergo, if the BSIM4.5 model gets used, and you don't set most of the parameters, then the model definition is fine, but the content of the model is not.

Am I being clear here?

An example - I recently did a foundry model audit for a client on 65nm CMOS, the models used were the BSIM4.4, after I got done digging through the nitty gritty of the models I found that none of the noise parameters were set properly, and over half of the capacitance terms were set to zero.

Total junk.

Jerry