Re: [SI-LIST] : IBIS datasheets for PCI and DDR

About this list Date view Thread view Subject view Author view

From: Stephen Peters (sjpeters@ichips.intel.com)
Date: Tue Nov 16 1999 - 17:59:25 PST


Fred Ballisari wrote:

<much snippage>

> Although there are IBIS models made from measurment, your claims seem
> far fetched. If there are Silicon vendors out there making IBIS models
> from measurement and making process corners from real silicon lets
> here from them. Some of the best IBIS models I've encountered come from
> Intel. Maybe they can comment.

  Producing model from straight measurement is close to the process that was
used to create the final Pentium(r) Pro FSB models. Starting with data
extracted from the xsistor level model, the data (V/T, I/V) was compared to
real silicon and adjusted as required. Remember, a simulation at worst case
process corners may produce absurdly pessimistic or optimistic models. In
real life, the I/V and V/T data has to be compared with and adjusted to fit
the actual buffer performance once the process folks figure out the volume
manufacturing recipe.. not to mention how tester fallout effects the
process skews.

 The real key, however, is that the model parameters were *also* adjusted to
account for such effects as SSO, power distribution, etc. Furthermore, the
model underwent *extensive* validation (and further minor tweaks) that
verified that the models' predicted performance in a system environment
(such things as Vol and edge rate) bounded the measurements we were getting
off of our test platforms.

  Easy and quick process? No. But in this case the IBIS model predicted
actual system performance much better than straight, unadjusted transistor
level models.

  I don't disagree that a transistor level model has the *potential* to
be more accurate than a behavioral model -- by it very nature it incorporates
more detail than a behavioral model and there are effects (gate source
modulation for example) that IBIS does not yet address. But potential does
not mean that it is... and no matter the form, the accuracy of a model depends
a great deal on how much care went into the modeling process.

    Regards,
    Stephen Peters
    Intel Corp.

**** To unsubscribe from si-list: send e-mail to majordomo@silab.eng.sun.com. In the BODY of message put: UNSUBSCRIBE si-list, for more help, put HELP. si-list archives are accessible at http://www.qsl.net/wb6tpu/si-list ****


About this list Date view Thread view Subject view Author view

This archive was generated by hypermail 2b29 : Tue Feb 29 2000 - 11:38:57 PST