V-AMS Compact Modeling Extensions subcommittee
Minutes of Oct. 21, 2003

Attendees:
Geoffrey Coram, Analog Devices
Ilya Yusim, Cadence
Jeroen Paasschens, Philips
John Moore, Agilent
Laurent Lemaitre, Motorola
Colin McAndrew, Motorola
Marek Mierzwinski, Tiburon
Boris Troyanovsky, Tiburon
Peter Liebmann, Xpedion
Jim Barby, U Waterloo
Al Davis, Kettering U

----

1) Approval of previous minutes (Sept 23, Oct 8)

2) Model Cards / Paramsets

The paramset proposal is on the designers-guide web site.
The main idea is that we need to propose some standard
syntax for Verilog netlists that will cover the
functionality of the Spice .model card.  We hope to do it
in such a way that we can suggest how a simulator vendor
might interpret a .model card for a Verilog-A compact
model in a Spice netlist.

Jim pointed out that large mos circuits with lots of devices
having the same l and w are not handled well in the current
Spice .model card.  Since paramsets can reference other
paramsets, one could make a paramset for a specific l and w
that referenced the main mos paramset and get better
efficiency.

Someone mentioned that Cadence's original Verilog specification
had something to say about model cards, but this was ultimately
not included in the IEEE standard.  Ilya promised to look into
the issue.

3) M-factor and rmin

Rmin sets a minimum resistance to avoid convergence problems
in Spice-like simulators (large conductance in the Jacobian
matrix and finite-precision arithmetic).  We want the simulator
to handle the m-factor automatically, but the model writer needs
to handle parasitic resistances that end up below the threshold
due to m.

I proposed $m_scaled_value_above(value, thresh, expon)
which returns (value*m**expon > thresh).  With this, one
could do parasitic resistors as well as capacitors, etc.
It's rather verbose; I'm open to a cleaner name.

We probably can't expect the simulator to detect when a
resistor is parasitic; in particular, it might be a nonlinear
resistor that is nevertheless negligible in some cases.
Also, in power electronics, small resistors are important,
and we want to allow maximum control over when they are
elimimated.  Rmin should be an instance parameter, not just
a global option to the simulator.

Jeroen took issue with vendors implementing rmin differently;
this conflicts with the idea of a standard model.  However,
Colin pointed out that "standard" models won't give the same
answer on HP, Sun, Linux; with different compiler optimization
levels (-O2); etc.  You have to accept the numerical differences,
including those due to rmin and gmin.