I'm Ali Elshorbagy and I'm working at Mabrains, I ...
# xycea
Ali Elshorbagy
10/17/2021, 12:27 PMI'm Ali Elshorbagy and I'm working at Mabrains, I made a small testcase to document a weird behaviour that i found when using .data inside xyce.
I'm simply doing dc sweep on sky130 nfet transistor on different widths and lengths. I have 2 parameters for width and length to sweep on them.
I found an unexpected behaviour. Initial value for width and length parameters affect solution values.
Normal behaviour during simulation is initial values will be replaced by values given in .data line but surprisingly i found that initial values for parameters affect the results.
In this git repo i have 2 netlists (netlist1.net and netlist2.net) exactly the same but initial value for width and length parameters are different. When i invoke Xyce to run those two netlists i have a completely different results
@User
https://github.com/mabrains/xyce_sky130/tree/main/testcases/Xyce_data_behaviour
e
Eric Keiter
10/18/2021, 10:44 PM@Ali Elshorbagy thanks for this report. I have verified that this incorrect behavior is happening in Xyce. I’ll open an internal bug on the issue.
Eric Keiter
10/18/2021, 10:45 PMYou probably figured this out; but the second netlist fails many of the DC sweep points.
Eric Keiter
10/18/2021, 11:21 PMI have entered an issue in our internal bug tracker now. My guess is that this will be (yet again) a problem with “scale”. That is a relatively new feature that seems pretty easy to break.
Eric Keiter
10/18/2021, 11:22 PMAnd in fact, I’ve fixed at least two other “scale” related bugs since I started participating on the Sky130 slack channel.
Eric Keiter
10/18/2021, 11:33 PMOK. Here is the issue, or at least part of it. I think that this is a binning problem. Xyce isn’t adjusting the bins when the L,W parameters are updated.
a
Amro Tork
10/19/2021, 7:52 AM@Eric Keiter appreciate all the help.
e
Eric Keiter
10/19/2021, 7:35 PM@Amro Tork no problem. The binning issue that I mentioned may be a headache to fix, but we’ll do it.
Eric Keiter
10/19/2021, 7:35 PMThe problem is that Xyce currently does all the binning during parsing and throws away models it doesn’t need. So, later, once the simulation is up and running, if it needs to update L,W, it no longer has enough information to re-bin the models.
Eric Keiter
10/19/2021, 7:37 PMSo, the fix will have to involve refactoring that a bit.
a
Amro Tork
10/20/2021, 6:24 AM@Eric Keiter appreciate all the help