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

About this list Date view Thread view Subject view Author view

From: Scott McMorrow ([email protected])
Date: Tue Nov 16 1999 - 16:10:47 PST


Fred and all,

When an IBIS model is extracted from real silicon it can be
more accurate than Spice. It is certainly more accurate for that particular
point in process, temperature and time. Some manufacturers produce
IBIS models based on actual measurements on real silicon that has
been produced at the worst case process corners. These are some of
the most accurate and reliable models that I have ever used.

When an IBIS model is extracted from Spice it can be as accurate
as the original, under some conditions. It is never more accurate.
IBIS is often less accurate (just as Spice models are offen less accurate)
than reality. This is usually due to poor modeling methodology, invalid
assumptions, invalid theory, and difference in the full manufacturing device
process range vs. shipped device process range (which is often
overlooked).

regards,

scott

--
Scott McMorrow
Principal Engineer
SiQual, Signal Quality Engineering
18735 SW Boones Ferry Road
Tualatin, OR  97062-3090
(503) 885-1231
http://www.siqual.com

Fred Balistreri wrote:

> Muranyi, Arpad wrote: > > > > Fred, > > > > Very good comments, and I would like to put an emphasis on the > > assumption you make, IF the models are correct (big IF). It is > > also true that SPICE models can or usually have a lot more detail > > so they can be more accurate (if done correctly). But detail does > > not equal accuracy. And, it is very true that they both have their > > appropriate places for use. A circuit designer will not get anything > > done with IBIS models, but a system designer will probably never get > > his/her work done with SPICE models either. > > > > I want to point out another thing that kind of hit a nerve. You > > mentioned the number of BIRDs (64 and counting) in the IBIS case. > > To be fair, you should have also mentioned the number of SPICE > > flavors that exist. These exist mostly because each vendor has some > > (unfortunately incompatible) "value added" feature. In addition, look > > at the number of "levels" each SPICE has for the various models, some > > of which are, again, incompatible. BSIM alone has at least three levels, > > and BSIM3 of one tool does not necessarily work with another tool > > supporting BSIM3. You don't even have to use different flavors of SPICE, > > just think about one of the comments someone made about the scaling factors. > > You simply can't simulate with two SPICE models, if one of them uses > > certain scaling factors that another model doesn't need unless you rewrite > > the model (good luck). (Of course this is only a problem for those who do > > system level simulations where you must have more than just your own > > design's SPICE model). > > > > I don't claim that IBIS is perfect, it is an evolving standard, mostly > > done by volunteer efforts. I also wish that it would have a lot more > > badly needed features, that it would be more consistent and more general. > > We have serous talks now about these issues among the "IBIS officials". > > We want to fix the shortcomings. However, if we point these problems > > out for IBIS as negatives or drawbacks, we should also be honest and > > admit that SPICE DOES have similar problems, even if it may seems to be > > a more stable language... > > > > Arpad Muranyi > > Intel Corporation > > ====================================================================== > > > > I agree strongly with most of the statements but disagree with the > > accuracy issue. Assuming that the SPICE model is correct and the IBIS > > model is correct, there is no way that an IBIS model can be more > > accurate than a SPICE model. This is regardless of where the IBIS > > model came from. Keep in mind that MOST IBIS models are produced from > > the SPICE model using s2ibis. That means by default MOST IBIS models > > cannot be more accurate than the source. For one thing SPICE is a > > different animal and contains much more information regarding both the > > topology and process of the design. Secondly there have been and > > continue to be accuracy as well as other issues with IBIS. If this were > > not the case there would not be 64 birds to date and counting. > > > > Both the IBIS and SPICE model are prone to incorrect data. Being not > > correct is NOT the same as model accuracy. In my view both SPICE and > > IBIS models have their application. There is some overlap in the SI > > community. You probably should not use SPICE models to simulate a > > whole board. But accuracy is a different issue. For most SI applications > > IBIS will do. But there are cases where one may need the SPICE model. > > > > Best Regards, > > -- > > Fred Balistreri > > [email protected] > > > > http://www.apsimtech.com > > > > **** To unsubscribe from si-list: send e-mail to > > [email protected]. 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 **** > > > > **** To unsubscribe from si-list: send e-mail to [email protected]. In the BODY of message put: UNSUBSCRIBE si-list, for more help, put HELP. si-list archives are accessible at > > Your information is correct. However you are now comparing a simulator > to a standard. The nice thing about IBIS is that it is a standard. > SPICE is generic for circuit simulator. So its not a wonder that there > are many flavors. Your also getting into the issue of EDA tools not > the models. A good SPICE simulator can and does take in IBIS models and > has interfaces to CAD. So its possible for SPICE to function as a > SI simulator even at the board level, thanks to behavioral modeling and > IBIS....and in that case the SPICE simulation accuracy is limited by > the information contained in the behavioral models. I still don't see > your argument though. How can IBIS models be more accurate than SPICE > transistor level models whence they are created from? > > Your argument seems to be one of using IBIS models in an SI application > instead of transistor level SPICE. In that regard there is no argument > for the most part. But as you know there are still cases where the > SPICE model is required or does a better job. > > Best Regards, > -- > Fred Balistreri > [email protected] > > http://www.apsimtech.com > > **** To unsubscribe from si-list: send e-mail to [email protected]. 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 ****


**** To unsubscribe from si-list: send e-mail to [email protected]. 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 - 11:38:56 PST