RE: [SI-LIST] : IBIS Modeling

Kellee Crisafulli ([email protected])
Wed, 25 Mar 1998 12:46:12 -0800

Hi Andrew,

>In an ideal world, IBIS models *ought* to be portable such that they can
>be used on different platforms and different simulators and yield
>basically the same result, within the accuracy limits of those
>simulators.
I agree completely.

>If there are big differences such as this (where large RLC values are
>treated in a fundamentally different way on some simulators), that seems
>like a risky thing to take advantage of. It ties your IBIS model to
>that simulator. How do you prevent the model from being used on another
>simulator?
In this instance we made the models for our own customers only.
However the real issue is how good or bad it works using either
method. Please try it, demonstrate the magnitude of whatever problem
you are concerned. Based on comparative
simulations we ran we felt the result were acceptable and useful until the
version 3.x models are available and simulators are all using them. I was
one of the most vocal advocates of getting modeling into IBIS for
DIMM and SIMM modules and made several proposals to the IBIS committee to get
this done.

>In essence, you have created a proprietary *.ibs model, similar to an
>Hspice model that won't work correctly on someone else's simulator.
You have a valid point. However unlike Hspice
it will run on all the simulators that use IBIS with slight differences in
the package effects but at least they will have a model v.s.
no model. Do you think it is better not to do any modeling and have
users guess what the signals look like. And yes I think this sort of
model is an order of magnitude better than guessing.

>As a side note, is it even the right thing to treat large RLC package
>values as a transmission line? What if they represent something like a
>long bondwire and a big bonding pad? How do you make that choice
>between which way to model it in the simulator?
IBIS has never specified that the LC pin values be used as lumped values.
As you may know some packages are modeled better as a lump and some
model better as a transmission line. Neither is perfect for all packages.
however it generally isn't possible to get devices to work if the pin is
really a large lumped L so I don't see much danger here.

>Should the IBIS spec make such loopholes possible?
I don't see this as a loop hole. Each vendor needs to use the IBIS
data to create the best results they feel possible.

I don't believe we need to spend a great deal of time on this thread Andrew.
I only wish to point out that it is possible to create a useable model for
a DIMM or SIMM in V2.1 IBIS.
I lobbied to get better SIMM and DIMM modeling into the spec. for over a year
so I agree it is a much better method. I was the one that insisted that the
changes proposed by Intel for their processor models also work for DIMM and
SIMM
modules.

I understand your concerns and they are reasonable.
Best wishes..
Kellee