Hi!
I am trying to run some simulations with 7 nm transistors in NGSPICE v41, and would like to know where does the parameters used in "examples\osdi\bsimcmg\Modelcards\modelcard.pmos" come from. I am asking this because there are some large differences to the values used in ASAP7, and also some parameters are defined which give "unrecognized parameter" warnings, yet used in "modelcard.pmos" (nseg/etaqm).
The details are here:
https://electronics.stackexchange.com/questions/678378/is-this-ngspice-simulation-correct-with-asap7-transistor-models
In the meantime I have found the documentation of the parameters in "BSIM-CMG_111.2.1_06062022.tar.gz/BSIM-CMG_111.2.1_Technical_Manual.pdf", and for example now I know that geomod=0/1 means 2 or 3 fins. But some of the other differences still make no sense (deltavsat).
I would also like to determine somehow the version of the BSIM-CMG OSDI model compiled into the NGSPICE v41 distribution, as I could not find any ".va" files in the source. Where is the source?
Any help would be appreciated. Thanks!
the files in case I delete them from GDrive
Example modelcards are never updated from the Berkeley model group since their version 106 in 2016. Also the predictive modelcard ASAP7 you mentioned seems for the older version 107.
Latest public released va models, ready to use osdi binaries and ngspice examples are available under VA-models.
I traced back the change when it was added, and it has not much sense to me:
So in essence the above commit added the file on 12/17/2022, and used the latest version of the modelcard.pmos from the BSIMCMG releases (from 2016 or 2019), but removed the 3 parameters to replace them with a slightly different default values. Not that it matters, as BSIMCMG explicitly states that the file does not represent any technology node from any manufacturer anyways.
I could not figure out from the source what puts the .osdi files into the release 7z file, maybe it is done manually? So I have no ide what the version of the VA model...
Anyways, I see no use case for including this model in the NGSPICE distribution, as for running simulations it is worthless, and nothing of the above is documented. I guess either it should be documented or removed, as currently it just confuses people like me.
The model available in the ngspice distribution is BSIM-CMG 111.1.0 released by Avirup Dasgupta on 01/14/2021.
The model parameters stem from BSIMCMG110.0.0_20160101.tar.gz. Changes to parameters probably have been made while trying to generate a model via ADMS. I will upload the original ones.
Last edit: Holger Vogt 2023-08-28
If anybody wants to know some more info:
In this distribution https://github.com/dwarning/VA-Models the modelcard.pmos is the same as the one in NGSPICE, but 2 out of the 3 parameters added back (FPITCH, TFIN but not L).
It contains two versions of the VA model:
1. VA-Models\code\bsimcmg\vacode110: this is equivalent to the one in BSIMCMG110.0.0_20160101.tar.gz, but some parameters have limits defined, and there are some #ifdef-ed changes for ADMS and such.
2. VA-Models\code\bsimcmg\vacode111: this is equivalent to the one in BSIM-CMG_111.0.0_09122019.0.0_20160101.tar.gz, with some parameters have limits defined but fewer changes relative to the original.
3. There is only a single VA-Models\code\osdilibs\bsimcmg.osdi file, it is undocumented? which one was used to generate it.
FPITCH and TFIN are both model and instance parameter wrt. documentation. So the example parameter set should contain the parameters. L is still only instance parameter.
Changes in parameter limits are made to prevent numerical problem with dividing by zero or negative argument for sqrt.
The provided osdi file is derived from latest public available model, in this case version 111.0.0.
I think about a way this to indicate by name or similar. Work is in progress.
Naming it bsimcmg1110.osdi would be enough (and changing the name in the spinint as well).
In the meantime I wrote an email to the professor of the ASAP7 paper wether the transistor parameters are expected to work with later models (like 112.1).
I think it would make the most sense to include the ASAP7 models in the default NGSPICE distribution because those parameters are similar to the values derived from real fabs and are intended to be used in simulations. (Unlike the parameters in the bsimcmg model distributions.)
Dietmar, see http://bsim.berkeley.edu/models/bsimcmg/ where BSIM-CMG 111.2.1 is available.
Thanks Holger,
Have to correct: 111.0.0 is not latest public available version. This will be 111.2.1 and soon available.
Version 11.2.1 is now under https://github.com/dwarning/VA-Models available. Works with ngspice test cases. Ancient model card is updated by removing parameter ETAQM, which is not actual anymore.