Designer’s Guide Community

mgehring

New Member

Offline

Posts: 6
Portland, OR

Convergence of PSS with "writefinal", &q
Nov 18^th, 2004, 1:50pm

I am using version 5.0.33.144

I have noticed often that there are convergence difficulties ( especially with large circuits ) when using writepss/readpss or writefinal/readic. I cannot concieve how there can be difficulty when using these with a circuit that is unchanged. I have recently seen, on a small circuit, PSS not converge when loading a "final" state with readic and skipdc=yes ( identical circuit and options ). It had to "use a few more tstab steps" to converge.

Also, I have see with a large circuit that loading a writepss with readpss requires MIC refinement. Again, this makes no sense - why does it not save the refined result ? These refinement steps take as long as the original PSS iterations.

If anyone has a solution to this, please let me know.

Thanks,

Mark

Back to top

IP Logged

Paul Geraedts Community Member Offline GIGO Posts: 60 Enschede, Netherlands	Re: Convergence of PSS with "writefinal" Reply #1 - Dec 28^th, 2004, 3:01am Similar experiences here. In general, writing and reading of the circuit state seems problematic within Cadence (DC, PSS and QPSS). From theory, it seems a nice way to speed up similar simulations (like in Envelope Analysis). Maybe somebody from Cadence could shed light on this issue? Thanks in advance! Paul
Back to top	IP Logged

Andrew Beckett

Senior Fellow

Offline

Life, don't talk to
me about Life...

Posts: 1742
Bracknell, UK

Re: Convergence of PSS with "writefinal"
Reply #2 - Jan 3^rd, 2005, 8:36pm

The issue with readpss doing a MIC always again was resolved in IC5141 from PCR 604941. A new option "checkpss" was also added, which you can set to no to stop it trying to check if the readpss data is consistent or not.

As for writefinal/readic type problems, this unfortunately does happen sometimes. A true state saving mechanism for transient has been added in the MMSIM60 version of spectre (released a few weeks ago). From spectre -h tran:

Quote:

State-file parameters
12 write File to which initial transient solution is to be
written.
13 writefinal File to which final transient solution is to be
written.
14 ckptperiod Checkpoint the analysis periodically using the
specified period.
15 saveperiod save the tran analysis periodically on the smulation
time.
16 saveclock=1800 s save the tran analysis periodically on the wall clock
time.
17 savetime=[...] save the analysis states into files on the specified
time points.
18 savefile save the analysis states into the specified file.
19 recover specify the file to be restored.

(note, it's the options with "save" or "recover" in the name which are new).

I expect it's in the release notes too. However, I just noticed that pss doesn't have similar options, so I don't think it will work (yet) with pss (i.e. if you want to recover a saved state (saved at the end of a transient) as the starting point for a pss); I'll file a PCR for that.

Andrew.

Back to top

IP Logged

mgehring

New Member

Offline

Posts: 6
Portland, OR

Re: Convergence of PSS with "writefinal"
Reply #3 - Jan 5^th, 2005, 9:32am

I guess I will have to try this new release. In any case, the bottom line for me is that I should achieve the original final convergence norm(s) when I load a final state ( readic ) or readpss with an identical circuit, without requiring any additional computations. If that is the case, then I aqree this issue has been resolved. Otherwise, one is not really loading the "true" final state.

Thanks, Mark

Back to top

IP Logged