RE: AGP model

Andy,
        It never ceases to amaze me how many IBIS models pulled from the web
need to be
fixed before they are usable. My recommendation would be (In parallel to
fixing the model
to be usable) feedback those problems and issues to the model creator !
Eventually they may not only
get fix the model, but also the model creation process at the source.

regards,
bob

p.s.
Syed Huq did an excellent paper ("Effective Signal Integrity Analysis using
IBIS models")
at designcon last year which documents common problems and solutions. It won
Best paper and I highly recommend it.

Cereva Networks
3 Network drive
Marlboro MA. 01752
Phone: 508-486-9660 X 3365
FAX: 508-486-9661
Email: rhaller@cereva.com

-----Original Message-----
From: Ingraham, Andrew [mailto:Andrew.Ingraham@compaq.com]
Sent: Thursday, August 24, 2000 4:06 PM
To: 'Johnny Pietraroia'
Cc: ibis-users@eda.org; 'si-list'
Subject: RE: AGP model

> i was just wondering, if anyone knows where i can get an IBIS model for
> the AGP port.
> i found a model at Intel's web site, but it seems to be incomplete.
> if anyone knows where i can get this model, i'd really apreciate it.
 
There are a few things you should do to the IBIS model on the Intel/AGP web
site (http://www.agpforum.org/downloads/agp4xr2.txt) before using it.
Evidently, the person who created this model doesn't know much about IBIS,
because the model, as posted, is totally useless!

Add some missing keyword lines to the file:

[IBIS Ver] 2.1
[File Name] agp4xr2.ibs
[File Rev] 0.2
[Component] AGP4X
[Manufacturer] Generic
|
[Package]
| Possible values are below; experiment!
| typ min max
R_pkg 0.1 NA NA
L_pkg 5.0nH 2.0nH 15.0nH
C_pkg 2.0pF 1.0pF 4.0pF | Max (C_pkg + C_comp) is 8pF.
|
[Pin] signal_name model_name R_pin L_pin C_pin
dummy IO_pin BAGP4X
|

With the above values, I scaled C_comp back so that its maximum was 4.0pF
(C_comp + C_pkg = 8.0pF). It's somewhat arbitrary, so experiment with
different combinations of C_comp, C_pkg, and L_pkg.

ALL of the I-V table values in the model must be scaled by 1e-03 (divided by
1000). The posted model has all the currents in milliamps, whereas they are
supposed to be in amps ... if only they had bothered to read the IBIS spec.

The [Ramp] values are not quite consistent with the I-V curves, because the
drive curves for AGP4X changed somewhat while the preliminary spec was
firming up, whereas the [Ramp] values in the model are left over from an
earlier model rev. So some simulators may give you a warning about that.
As I recall, the dV/dt values are suitable, but the dV values by themselves
don't quite agree with the I-V curves.

Play around with the component values (R_pkg, L_pkg, C_pkg, C_comp) and/or
substitute another package model to get different transient results. This
is only a crude generic model and some freedom is possible in one's actual
implementation, which may cause some subtle or not-so-subtle differences.

Of course the model is only good for the buffers that use Vddq=1.5V, so it
is no good for AGP2X chips, nor for the AGP4X pins that remain 3.3V.

What amazes me is that this totally useless model has been on the AGP site
for some 25 months now and nobody there bothered to do anything about it.

Perhaps that shouldn't surprise me; after all, for several months now the
AGP web site also has pointers to an ECR #50, but the referenced pages go no
higher than ECR #48. So where are #49 and #50? As far as I can tell, this
ECR #50 does not exist anywhere on the Web.

Regards,
Andy
 
Received on Fri Aug 25 05:32:47 2000