Recent changes to patches

#121 Continue Newton iteration until currents settle

Keith OHara — Wed, 22 Apr 2026 12:46:42 -0000

It is, as are others of the BSIM* devices. It seems it was too confusing for me the first time I generated a list of non-NULL DEVconvTest pointers, but it is straightforward now, so this evening I can finish completing the patch, recheck, and repost.

#121 Continue Newton iteration until currents settle

Holger Vogt — Wed, 22 Apr 2026 12:27:09 -0000

If we look at b4cvtest.c, lines 56 ff. , is this the same issue?

#121 Continue Newton iteration until currents settle

Keith OHara — Wed, 22 Apr 2026 10:22:01 -0000

Attaching a completion of the repairs started in the patch already applied to pre-master-47,
making the same fix in devices bjt, dio, hicum2, vbic, vdmos. I see no similar errors on other devices ( mos6, etc.). Clean make check and in my sampling through the examples, everything seems to work to me.

The regression test in this patch now uses a 'dio' device instead of a user-written 'asrc' function representing a diode. Improving performance on the simulation using this more rigorously physically-based model. is a hint that this repair might provide real benefit, as well as doing the test described in documentation §1.4.2.

I do think it wise to apply this patch, to let people using pre-master find problems or benefits

Not all device-types apply the current-convergence criterion in §1.4.2. Switches for example, in swinit.c set their DEVconvTest = NULL, as do many other devices (including the linear devices that don't need the §1.4.2 test). This makes sense to me for the switches and similar with conceptually discontinuous I-V curves where the solution being sought can be at the discontinuity.

Bug #824 showed a case where the initially desired behaviour was to ignore §1.4.2 convergence of a current representing an internal state. I'm looking at that example, and how users could meet its goal with or without §1.4.2 being applied to 'asrc' B-devices. Maybe the asrc devices belong with the switches, and should not have had their current-convergence test written (but not applies) in the first place.

#121 Continue Newton iteration until currents settle

Giles Atkinson — Mon, 20 Apr 2026 19:24:56 -0000

Thanks! G.

#121 Continue Newton iteration until currents settle

Keith OHara — Mon, 20 Apr 2026 18:14:41 -0000

2 files, attached, go in tests/regression/misc
They are in the patch up-thread, so it might be easier reapply and 'git add' the two new files.

I did follow up on my comment "there is a similar error in the tests for other nonlinear devices" from the top post, and in the end found only one or two copies of this error. I'll try within one day to find and recheck that completion of this patch, and then post it here.

#121 Continue Newton iteration until currents settle

Giles Atkinson — Mon, 20 Apr 2026 17:47:26 -0000

The patch was missing a file, convergence.cir. Keith OHara, please post the file.

#126 OSDI array parameters lead to segmentation fault

Holger Vogt — Tue, 14 Apr 2026 16:34:13 -0000

status: open --> closed-fixed

#121 Continue Newton iteration until currents settle

Robert Turnbull — Thu, 09 Apr 2026 20:27:44 -0000

This patch has eliminated the false convergence Giles noted on bug #824. My test circuit now fails to converge with or without the added 0V source.

#121 Continue Newton iteration until currents settle

Francesco Lannutti — Mon, 06 Apr 2026 15:36:11 -0000

Hi Holger,
I think it's that one, but it's a 10years old work, so I don't remember exactly! :)
It's part of my PhD, attached, along with LnL Separation (I think there is a branch for this too).

Thank you,
Fra

#121 Continue Newton iteration until currents settle

Francesco Lannutti — Mon, 06 Apr 2026 15:23:57 -0000

Hi Holger,
no, no, because it was experimental for some models. It needs code model modification, like KLU...
But we could brainstorm offline among all of us.

Thank you,
Fra