Re: IBIS suitability for over GHz waveform simulation

Yamagishi-san,

I was a part of a committee that investigated IBIS accuracy for about one
year. Although I have not compared behavioral and structural
(transistor-level) simulations at 1.2 gb/s, I do have some opinions that I
think still apply at high data rates. The question you're asking could be
framed as follows: "Is a behavioral model accurate at data rates in excess
of 1 gb/s?" IBIS is really just a format for storing behavioral model
data, and it's up to the EDA vendor to determine how to implement this data
using a behavioral model (which the user usually never sees).

1 gb/s implies that you need timing accuracy on the order of 10-50 ps. I
think that you can achieve this kind of accuracy using a behavioral model
if your driver and receiver circuits are well-behaved! What I mean by
well-behaved is that the driver uses a traditional push-pull output stage
without any fancy feedback or edge-rate control circuitry and the receiver
can be modeled as a capacitance. Now, IBIS has provisions for model data
that describe more complex circuitry such as multi-staged drivers and
receivers that use dynamic clamping. However, these circuits require more
modeling effort and implementation support in the simulator. I personally
don't have much experience with these features, so I can't comment in
detail. I'm not saying that 10-50 ps accuracy is impossible with these
more complex circuits; if it is, it will certainly mean more work. My main
point is this: a behavioral simulation can be accurate enough IF the
behavioral model is well-matched to the physical circuit and IF the
modeling engineer spent enough effort developing the model data.

You may or may not be aware of an effort under way in the IBIS committee to
expand the flexibility of IBIS, allowing coverage of even more complex
kinds of I/O circuits. This effort is known as IBIS-X, and you can learn
about it by attending the IBIS Open Forum conference call.

I think you're correct in your assessment of the importance of packaging
and loss effects at these frequencies. Given well-behaved drivers and
receivers, package and loss effects may be the more challenging problems
from an accuracy perspective. I wish you well in your efforts.

p.s. If you're interested in reviewing the work of our committee, you can
find it on the IBIS web site: http://www.vhdl.org/pub/ibis/accuracy. Look
at handbook.pdf.

Sincerely,

Greg Edlund
Advisory Engineer
Electrical Packaging
IBM Server Technology Development
3605 Hwy. 52 N, Dept. HDC
Rochester, MN 55901
gedlund@us.ibm.com
---------------------- Forwarded by Gregory R Edlund/Rochester/IBM on
09/08/2000 07:42 AM ---------------------------

Keitarou Yamagishi <yamagi@isl.melco.co.jp> on 09/07/2000 09:51:41 PM

To: ibis@eda.org, ibis-users@eda.org
cc: yamagi@isl.melco.co.jp

Subject: IBIS suitability for over GHz waveform simulation

Hi,

I've gotten an IBIS model for 1.2Gbps serial data tranceiver IC, which
means that the 1st order frequency element is GHz class. But is IBIS
itself suitable for over GHz transmission waveform simulation? I think
that, for example, package loss element or frequency responsibility of
transister is not enough reflected.

How dou you think?

Thanks.

_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/
_/ 　　 _/
_/ /\ Keitarou Yamagishi (Senior Engineer) _/
_/ / \ Mitsubishi Electric Corporation _/
_/ \ / Information Technology R&D Center _/
_/ \/ 5-1-1 Ofuna, Kamakura, _/
_/ /~~~/\~~~\ Kanagawa 247-8501 Japan _/
_/ / / \ \ Email：yamagi@isl.melco.co.jp _/
_/ ~~~~ ~~~~ _/
_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/

 
Received on Fri Sep 8 06:21:25 2000