Multiple convergence tests possible?

[Mehmet Cirit (Ceridli)]

I have some cases where transient simulation is very slow, several hours where it should be several seconds.  Logically the circuit should have stabilized. I am guessing that

there is some noise which can not be reduced by time step reduction. Most likely you have voltage and current convergence tests. Is it possible to assume convergence is achieved if one of them indicates convergence, and others do not?  Sometimes satisfying all of them

may not even be possible.

[xyce-users]

Hi Mehmet,

when this sort of things happens, there are a number of things that could be going wrong.  For example, in transient, if you are getting really slow behavior you might check if there are any sawtooth oscillations in the signals.  If there are, then this is a numerical artifact of the trapezoid method, which is the default time integration method.    This can be fixed by changing the method.   You can do that with the netlist command:

.options timeint method=gear

Another thing to check; if this is a netlist originally intended for Hspice.  We have an hspice compatibility option. It is limited, but one thing that is very important (covered by this option) is the "atto" unit suffix, which is the letter a.  In Hspice this means "atto" or 1.0e-18, but in Xyce (by default) it is a unit, so it just means amps with no multiplier.   As you can imagine this can make a big difference.   You can invoke the hspice toggle using the command line option:

Xyce -hspice-ext all  netlist <return>

thanks,

Eric