A quick Google search came up with http://www.xmodtech.com/ for Hicum extraction.

I have heard the same story you have (hicum better for scalable extraction), but we use an enhanced Gummel-Poon model and have nice scalable models.  Maybe our characterization guy is just smarter than your average Joe.

Hmm ... that's odd.  I typed P O O N, but the software translated it to "thingy."
I'll have to check with the webmaster.

We have our own simulator, so we built the model into it -- much more flexible than subckts.

Geoffrey_Coram wrote on Jun 6^th, 2005, 8:14am:



thanks Geoffrey,
well, could you tell me more qbout you on house enhanced SGP model if it is not too confidential? For example
-Is your ft well modeled and the scalability valid ? (versus lenght, width, and configurations).
-are your parameters physical meaning?
-your max Ft? technology? (SiGe 0.35, 0.18? 0.13?)

A minor comment -

Bipolar geometry scaling does not work as nicely (in theory anyhow) as does CMOS scaling. (lets ignore geometry binning for the moment shall we? )

Consequently a lot of the models get developed fixed geometry without size scaling. (ergo lots of unique, one model fits one transistor models)

When I was part of IBM we were using the VBIC95 model for the 5HP, 5HPE, and later SiGe processes. FWIW it seemed to give pretty good corellation to silicon.

Thats out of UCSD as well as out of the foundry group in Vermont. Must be one of Larry Larson's students.

The 5HP, 5AM, 5HPE, and 6HP processes all used the VBIC95 models if I recall correctly. If they went on to different models after that I am not sure.

Good news - most model development coming out of IBM is pretty good. No complaints from me on what they do there.

Bad news - There are a lot of bad models out there. I know a lot of horror stories.