Hi all,
I submitted a bug in a GitHub Issue on the Xyce GitHub repository, but it seems that's not the preferred way to notify the Xyce team, so I'll link to it here:
(has evidence and test cases)
The short summary is that Xyce seems to only carry out binning using total W, but BSIM4 bins according to W/NF, which means that the bins are chosen according to a large W, but with NF>1 the actual widths used by BSIM4 are very small, often smaller than the bin minimum. This causes bizarre solver errors because of misbehaved equations out of bin.
This affects sky130 which is very binned.
thanks,
Matt