Re: convergence problem


Subject: Re: convergence problem
From: Ingraham, Andrew (Andrew.Ingraham@compaq.com)
Date: Fri Jan 25 2002 - 04:48:52 PST


SPICE convergence problems (and HSPICE is no exception here) do happen,
and can be difficult to get around.

There are some suggestions in the HSPICE manuals which may help, but
there is no single solution that always works. If there was, they would
have hard-coded it into SPICE and nobody would have convergence
problems. HSPICE is said to have several algorithmic enhancements that
were added to help. Alas, we still have non-convergence problems, some
small, some major.

Sometimes it is due to mathematical difficulties, but often it is
because the models you are using, have something bad like
discontinuities in them, which can be next to impossible to avoid
(without fixing the models). Occasionally, you get devices which work
just fine by themselves, but which fail to converge when combined into
the same simulation.

SPICE has a number of tolerances (options with *TOL) that you can try to
make looser, but the risk then is that the results become less accurate.

Regards,
Andy



This archive was generated by hypermail 2b28 : Fri Jan 25 2002 - 05:09:25 PST