There are 2 tools that come to mind. One is ContecSPAR from
CONTEC (Now Applied Simulation Technology) and the other is part of
the picosecond modeling option for HP MDS.
I've only used the Contec program. It does a Pade' approximation to
create a 'black box' model subcircuit that is the ratio of two polynomial
controlled sources. The user has the option of specifying how complex (how
many terms) to use. The complexity is dictated by how 'well behaved' the
input s-parameters are and by how wide a bandwidth you want the model to be
useful over. You need to watch out for certain cases where you may end up
with an unstable model due to poles in the RHP, but that is true of the
Pade'technique in general.
The resulting model always works in Contec Spice but if you use
the resulting model in Hspice you can run into problems where Hspice's
ability to handle very small and very large exponents is exceeded. This
problem is seen when numbers are < 1e-38 and >1e+38 . This problem was
evident in Hspice 95.2 and as far as I know they haven't fixed it in the
current release.
HP's version ends up creating a lumped element model with user
selectable topology. I don't remeber the technique that HP is using. I did
run some sample evaluation problems on both Contec and HP a while back and
as I recall the resulting spice simulation results produced by the Contec
Pade' model and the HP lumped model were very nearly indistinguishable from
each other.
Ray Anderson
Sun Microsystems
Advanced Component Engineering