I forgot to mention :- I built a fully featured version of ngspice from this git hash ```tom@tom:~/r...
t
I forgot to mention :- I built a fully featured version of ngspice from this git hash
Copy code
tom@tom:~/repos/edatools/ngspice$ git log -1
commit 2af390f0b12ec460f29464d7325cf3ab5b02d98b (HEAD -> master, tag: ngspice-43, sf/master, origin/master, origin/HEAD)
Author: Holger Vogt <holger.vogt@uni-due.de>
Date:   Sat Jul 13 10:14:02 2024 +0200

    Update copyright info
tom@tom:~/repos/edatools/ngspice$
h
Using SKY130A I did the following changes to tb_cp_tran.spice: Remove .options reltol=1n Use .tran 10p 100n uic Remove .options savecurrents Remove .save all Use .save v1 v2 v3 v4 net1 net2 clk_ref The first two have been inevitable to get both variants running (with or without xpfd1). The others save time and memory and may be changed back later when things are running.
t
Thank you very much for the fix @Holger Vogt! What is the reason for having to remove the reltol option and specify uic in order to get it to work? It's not clear to me how these are related to the issue I'm seeing. Is this a temporary workaround for some underlying issue or is there something else I'm not understanding?
h
Standard (default) reltol is 0.001 (aka 1m). reltol=1n is simply wrong.
Specifying uic skips the operating point calculation (which sometimes creates convergence problems) and starts transient immediately with voltages starting at 0V (if not specified by .ic).