BSIM level 69 XDM problem

[Isbarak]

I am trying to use a MOSFET BSIM 69 ver 103.3 model in xyce. This model has the non-quasi-static parameter SWNQS, which  is not converted using XDM 2.7.0. 

For a workaround, Is there an equivalent parameter name in XYCE?

log file output:

...Param removed. No param defined internally in XML: SWNQS...aram removed. No param defined internally in XML: MUNQSO...Param removed. No param defined internally in XML: FNTEXCL

thanks

ilan

[xyce-users]

Hello, thanks for your interest in Xyce.

FYI, the model you are attempting to use is not a BSIM model.  It is the PSP.   Not a big deal, but it took me an extra step to figure out what model you were talking about.

Regarding the NQS part of the PSP; I have checked and can verify that this part of the PSP isn't in the Xyce implementation.    My guess (as I didn't do the implementation) as to why this is;  probably that part of the model used verilog-ams features that were not supported by ADMS.    ADMS has been the model compiler we've used to import verilog-A models.

Unfortunately, I'm not aware of a workaround for you.

We are working on a new model compiler, which will be able to support more of the standard.    That model compiler isn't ready for prime-time yet, so it isn't a short term solution. 

thanks,

Eric

[Tom Russo]

The PSP 103 NQS model makes use of the "idt" function (time integration), which Xyce/ADMS does not support.

The primary reason for this is that Xyce itself doesn't have a one-size-fits-all time integration feature that could be leveraged easily to work with Verilog-A.

There is no reason that Xyce *couldn't* have such a feature, but it didn't exist at the time that Xyce/ADMS was under development and as far as I know doesn't exist yet.  Every hand-coded model in Xyce that does some sort of time integration did so via some one-off code that doesn't map easily onto a generic feature.

This is the main reason that none of the Verilog-A derived models in Xyce have NQS enabled.

There is no quick and dirty workaround.  If, however, Xyce were to have an easily accessible means of coding time integration so that Verilog-A's "idt" function could readily be mapped onto it, it wouldn't be hard to add to Xyce/ADMS or whatever replacement gets implemented.

--
You received this message because you are subscribed to the Google Groups "xyce-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to xyce-users+...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/xyce-users/6d8496da-be61-4611-9173-10062bbb570en%40googlegroups.com.

--

Tom Russo    KM5VY 
Tijeras, NM 

 echo "prpv_a'rfg_cnf_har_cvcr" | sed -e 's/_/ /g' | tr [a-m][n-z] [n-z][a-m]

[Isbarak]

Thanks Tom and Eric

I will atempt to understand the diffeence in device performance with and without NQS and if I have success will report here

ilan

[Isbarak]

I came across this: XYCE release notes

it states that Xyce™ 6.11 include the following.
...
New Devices and Device Model Improvements
....
PSP version 102 and 103 (MOSFET levels 102 and 103)

....

should I deduce that  NQS is supported?

ilan 

[xyce-users]

No, the presence of updated variants of the PSP 102 and 103 in Xyce 6.11 does not mean that NQS is supported.  That version of Xyce is pretty old at this point, and was released in 2019.  So, if you aren't finding NQS parameters in a recent version of Xyce (like version 7.9) it would not have been present in older versions like 6.11.

NQS-style approximations have been appearing in MOSFET models since (at least) the late 90s.  For example, the BSIM3 (a late-90s model) has its own version of NQS.   What they do (and possibly this is obvious from the name) is make it so that the model's response is less instantaneous.    There are various ways of approximating this effect on the compact model level.  The PSP style of doing it is different than the BSIM3 for example.    

If you are not running rapid transients and/or high frequency circuits, you might not need NQS.  For slower transients my guess is that results with NQS turned on vs. turned off will be the same.

thanks,

Eric

[xyce-users]

By the way, if you are curious about the details of the PSP NQS approximation, that has been documented in papers like these:

https://www.researchgate.net/publication/3075071_A_Unified_Nonquasi-Static_MOSFET_Model_for_Large-Signal_and_Small-Signal_Simulations

https://www.researchgate.net/publication/224625316_Unified_non-quasi-static_MOSFET_model_for_large-signal_and_small-signal_simulations

thanks,

Eric

[Isbarak]

Thanks Eric

I am running a passive FET switch mixer at 10 GHz. I understand from the litrature that especially IM3 is incorrect in BSIM models. However, I am not sure I need the NQS partof the PSP to properly model this mixer. Can you comment?

thanks

ilan

[Thomas Russo]

I will point out, however, that the NQS model in the BSIM3 (or the BSIM4, for that matter) remains unimplemented in Xyce.  If you set NQSMOD=1 in that device, the code will warn that it's unsupported and reset it to zero.

As far as I know, none of the MOSFETs in Xyce actually have NQS implemented.  It is certainly the case that no models derived from Verilog-A (PSP, BSIM-SOI 4.X, and so forth) do, because they all seem to want to use that idt() verilog function to do time integration.

On Tuesday, April 22, 2025 at 6:29:38 AM UTC-6 xyce-users wrote:

[xyce-users]

Yes, that is true, none of the MOS models in Xyce have NQS implemented.  We seldom have people request it, so I've inferred that they aren't used much.

Whether or not you need the NQS part of the PSP model depends on the size of the transistor and how accurate you need the simulation to be.  It is hard to give precise guidance on that.    The upper frequency limit for going without NQS will be proportional to something like \omega = \mu * (Vgs-V_T)/ (\alpha * L^2), but this is just a rough guideline.  The book "Operation and Modeling of the MOS Transistor" by Tsividis and McAndrew has some discussion of this.   (that is where this equation is from).

Eric