MOSFET convergence

After getting spammed, it is time for a real on-topic posting.

I have been looking at MOSFET convergence, and have some clues to 
improving it.  There is definitely a bug that sometimes produces a 
single nonsense iteration when polarity is reversed.

This problem seems to also impact the transient response if MOSFET 
circuits somewhat.  I am not sure why.  In my benchmarks, AC analysis 
matches Spice as close to exactly as can be determined from the 
printout.

In the 0.31 release notes, I mentioned a voltage sync bug.  I have a 
fix for it, but it makes little difference in the non-convergence 
cases I have.

Development has been very slow for the past few months due to a heavy 
teaching load.  I expect to look at it seriously in a few weeks.

Another problem is that it is very difficult to see the convergence 
trajectory of internal values in Spice, for comparison.  In Gnucap, 
you can probe most of these, and request transient printout at every 
iteration, making the trace a lot easier.

In some cases, convergence in Spice but not Gnucap is due to 
differences in convergence checking and step control.  Gnucap has 
about 40 checks per device.  Spice has about 2.  This means that 
sometimes incorrect results will slip by in Spice, but be caught by 
Gnucap.  (one plus for gnucap)  It also means that sometimes an 
almost insignificant parameter can slow things down a lot, because it 
too must be within tolerance.  (perhaps a minus for gnucap??)

That is not the problem in this case.  It is a real bug.