pss: "Time steps are continuously to be extremely small."

Hi all,

I'm simulation a differential ring oscillator with a frequency of
around 1GHz.
I've successfully used pss and pnoise simulations until I changed the
transistor parameters a bit. Now pss refuses to converge. The transient
simulation shows a nice periodic waveform with a period of about 1ns.

The last messages from the simulator are:

*******************************************************************************
pss: time = 59.21 ns (97.6 %), step = 2.415 ps (234 m%)
Conv norm = 37.2e+03, max dV(I0.dummy_1.I0.load0.D1:int_a) = -63.0052
mV, took
2.61 s.


=======================================
`pss': time = (58.206 ns -> 60.2687 ns)
=======================================
Time steps are continuously to be extremely small.
Conv norm = 347e-12, max dV(I0.dummy_1.delay0.I16.M0:int_d) = -3.77476
fV, took
2.07 s.


Error found by spectre at time = 58.206 ns during periodic steady state
analysis `pss'.
After the dynamic adjustment of tstab, PSS still failed.
*******************************************************************************

Now that looks weird to me. It says "Conv norm = 347e-12, max
dV(I0.dummy_1.delay0.I16.M0:int_d) = -3.77476 fV", but fails to
converge. (It would have covered 2 periods of the signal at that time
from 58.2 to 60.2 ns).

I've tried playing with ireltol, vreltol, step, maxstep with no luck.
Anything else I might try?

Blondie

 

It sounds as if the oscillation has stopped - the conv norm could be very low
because everything ended up DC somehow. Have you looked at the saved
initial transient from the PSS (or a separate PSS run).

What version of the simulator are you using (the subversion info)? What settings
are you using? Perhaps you could post the analysis part of the netlist here?

Regards,

Andrew.