re: Quest for IBIS models

Greg,

My experience in getting Spice and/or IBIS models has varied dramatically from
vendor to vendor. As this might also be of some interest to the various tool
vendors, here's my feedback...

My starting point is usually the vendor's rep. Inevitably the rep has no clue
what we're requesting, though they do come up to speed fairly quickly. Within
a day or so I get feedback as to what's available. An increasing number of
companies have Spice information readily available, and a small number have
v1.1 IBIS stuff available. I've yet to see a v2.1 model (though Intel has
provided equivalent Quad and HSpice models for buffers that couldn't be
accurately modeled in the earlier format). Interestingly enough, many of the
companies I've encountered that have IBIS models also are willing to provide
the original Spice info.

I make a point to try to get my hands on Spice models for several reasons
(even though I won't be using Spice to run the sims). First off, we've had
some success in getting vendors to "guarantee" their Spice models. I rarely
get guaranteed IBIS models (it's happened once, and wasn't a memorable
experience). Secondly, I can gain some insight into the model accuracy by
looking through the netlist and device models. Though this can be misleading
it can also be quite beneficial. I have a big issue with non-guaranteed
models, but that's another subject. Finally, with most Spice models you're
free to set the operating conditions (process point, voltage, and temperature)
according to the each system's specifications. IBIS min/max conditions are
often predefined and accordingly are too conservative for some applications.

When Spice models aren't available or can't be obtained in an acceptable time
period (the lawyers got involved), I take one of two different approaches
depending on my needs. Sometimes I'll ask for IBIS models, while other times
I'll ask the vendor to perform a series of buffer characterization sims and
simply send me the raw results. The second approach has worked well when the
vendor doesn't have IBIS models, doesn't know what IBIS models are, has v1.1
models that aren't accurate enough, or has the wrong operating conditions in
their standard models. These two approaches have worked well at times, while
other times the whole experience is a big pain in the [insert favorite
anatomical reference here].

As to Zeelan models, they could come in handy at times. I've yet to purchase
a copy of their library for two reasons: it doesn't contain min/max models
and doesn't contain all the parts I think it should (DRAMs, cache RAMs, etc.).
 There have been times, however, when it would've been beneficial to have
their library available.

As to industry trends, I've seen a positive shift over the last 6 months or
so. In general most of the companies have a long way to go, but I get the
impression that more and more of 'em are heading the right way. I know I've
been going to a lot of trouble to educate a number of 'em...

Mark Leonard
Lead Engineer, Signal Integrity Group
Compaq Computer Corporation
Received on Thu Feb 9 15:17:59 1995