Re: [SI-LIST] : receiver jitter

About this list Date view Thread view Subject view Author view

From: D. C. Sessions (dc.sessions@vlsi.com)
Date: Wed Jan 19 2000 - 11:44:11 PST


Scott McMorrow wrote:
> Chris cheng wrote:

> > believe it or not if there is a will, there is a way. most of the issue
> > of
> > not supplying spice model is related to the ignorance of the tech
> > support
> > people (they simply don't know how to run spice) and not because ip or
> > accuracy as they told u.

Chris, I hate to break it to you but I'm on the semiconductor side, not
the PWB side. I can stat equite confidently that there IS a problem with
providing *my* customers with SPICE models because I've been through the
mill with *my* management trying to do just that and failed. Semi companies
are (perhaps justifiably) paranoid about letting process data out of the
fab; even I have to sign away my grandchildren to get each update of the
process characterization tables.

Keep in mind that not everyone doing high-speed design is Dell. They're
big enough that having Legal spend a few days working on the NDA and getting
a VP to bless it can be justified. Multiply that by a dozen or so major
component suppliers and it adds up fast. If you're a small shop you're just
plain SOL.

Oh, and OF COURSE Intel uses SPICE, as do I. We're the ones doing the
transistor-level work, after all --- those behavioral models have to come
from somewhere.

> > if there is technically capable people on the
> > other side of the table, you can get spice model most of the time. i am
> > always under the impression spice is a standard circuit and transistor
> > language and the only thing that's lagging is an industrial standard of
> > encrypting it. why can't we have spice encryption committee instead of
> > ibis committee ?
> > i have to say for the last time i am not against ibis. it's a perfect
> > tool
> > for designing mickey mouse buses like pci or jtags etc. its just not
> > good
> > enough for highspeed design.
>
> Unfortunately, when SPICE models are encrypted it it next to impossible
> to resolve fundamental simulator issues, such as convergence, without
> support from the vendor. If a vendor's models were to exist in a vacuum
> then this might not be a bad thing. However, many of my SPICE issues
> stem from interoperability of multiple vendor supplied models within
> the same simulation.
>
> Perhaps SPICE would be more useful if there were some standards
> to provide better encapsulation of necessary information about the model
> such as the extraction assumptions, and subcircuit interface documentation.

One of my favorite problems is namespace collision. Every semi vendor in
the world calls their (encrypted) transistor model "N" or "NMOS" or some
such and names their (encrypted) I/O cell "IO" or something equally original.
Then you try to do a system-level simulation with four different processes
each using the same name for different transistors and six ICs each calling
its transceiver "IO" and watch the fun. IBIS eliminates the process model
problem and at least lets you rename the flipping [Model] sections.

-- 
D. C. Sessions
dc.sessions@vlsi.com

**** 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 ****


About this list Date view Thread view Subject view Author view

This archive was generated by hypermail 2b29 : Thu Apr 20 2000 - 11:34:44 PDT