Re: model accuracy

I believe that the behavior a driver exhibits in an EDA tool's transmission
line simulator is a function of both the IBIS model and the tool's IBIS
simulation algorithms. The tool must translate the simple data points (I/V and
dv/dt tables) in the IBIS model into a complicated and dynamic behavioral
interaction with the behavior of the transmission line algorithms in the tool.
This is rocket science. There are quite a few transmission line algorithms in
use, and quite a few algorithms to get useful behavior from the IBIS model data
points. So far we haven't seen much discussion of the accuracy differences
between simulators, but I expect that to come as IBIS becomes a commodity.

Manufacturers can verify that the I/V curves and dv/dt curves match hardware or
spice, manufacturers can build spice models to mimic the IBIS model in spice
but unless they check the behavior of the IBIS model on a few diverse
transmission line networks in several transmission line simulation tools -
against spice or hardware - they are essentially only placing the data sheet in
electronic form. Unfortunately the form is so useful that you can extrapolate
some very realistic behavior form it. Realistic, but not necessarily
representative of the manufacturer's hardware.

Trust the IBIS model to tell you as much usefull information as the data sheet.
Do your own qualification of the transmission line simulator and the IBIS
models you wish to use against whatever other, more reliable information you
can obtain on a few benchmark networks before you push the edge of your design
using them.

Joe
 
Received on Fri May 30 19:54:02 1997