converge.jpg

Although SPICE is one of the most prevelant tools for analog circuit simulation, saying that it was without problems would be incorrect. The problem that I have seen is that individuals start using a cad package, such as the Cadence/Orcad suite and run into some incomprehensible problems with analog simulation with PSPICE and quickly give up on the technology deeming it unusable. What needs to be understood is that the simulation heavily relies on numerical integration and matrix inversion methods, both of which are very susceptible to numerical errors. The up-side is that there are parameters that can be tuned, and in some cases, simulation speed/performance can be sacrificed to get higher accuracy and convergence.

Charles Hymowitz, from Intusoft, has written a small article on the topic of SPICE simulation convergence. The basic idea here is to first make sure your circuit is wired as intended and that the nodes are properly labeled. He then gives an overview of some parameters and algorithms that can be tweaked to try to attain convergence. Finally, he makes a very important point: if your circuit still doesn’t converge, look back at the design, there may be something inherently wrong with it.

I agree with most of his statements and have been using SPICE for some time to get a high-level simulation of various analog circuits. Unfortunately, it is not a substitution for building a prototype, but it does offer assistance with initial design. I am thinking of writing up a few step by step guides for using the Orcad suite, including one that demonstrates PSPICE simulation with device models downloaded from vendor web sites. If anyone is interested in seeing those types of things, send an email or leave a comment and I may do it a bit quicker.

( converg.pdf )

 
del.icio.us | digg