RE: [SI-LIST] : IBIS Subcommittee Formed to Review Spice to IBIS

From: tomda <tom_dagostino@mentorg.com>
Date: Wed Jun 09 1999 - 11:43:47 PDT

There are several questions that beg to be asked to fuel this discussion.
 How accurate does a model (simulation) need to be? If the accuracy
requirement can be nailed down then a meaningful discussion can be had on
the number of points required. So far we have not addressed this part of
the problem. Once we start down this path we will have to start
understanding the algorithms used, different mathematics will generate
different results given input data.

I've seen too many models that have way too much precision build into the
model. E.g., 43.0156898745 mA When people insist on generating lots of
decimal points there is a tendency to generate lots of data points to
justify the precision in the data. In many cases of models I have seen
with this there are datapoints that show the difference in the 6th decimal
point.

I agree that disk space is free and MIPS are getting there too. Artificial
limits on the number of points are not good for models. But maybe there
are other ways of solving this problem. Equations for VI curves may be the
way to go and let the simulator decide the voltage step is an alternative.
 Like Jon, I use algorithms to reduce the number of points in our VI tables
with very good results. I've not seen VI curves that needed over 100
points after processing. Jon, what kinds of parts require more than 100
points?

Tom Dagostino
Mentor Graphics

-----Original Message-----
From: jonp@pacbell.net [SMTP:jonp@pacbell.net]
Sent: Wednesday, June 09, 1999 8:59 AM
To: Kellee Crisafulli
Cc: scott@vasthorizons.com; ibis@eda.org
Subject: Re: [SI-LIST] : IBIS Subcommittee Formed to Review Spice to IBIS

Here is a comment from a model creator:

I have written extensive software to deal with the ideal selection of
non-redundant data. I have found that the 100 point limit is over
restrictive for
both IV and VT curves. (that is, I have example models that cannot be
constructed
inside the 100 point limit and still be as accurate as I want (ie. as
accurate as
the XTK model I made at the same time)). I don't think that model size or
transmission time should really be a major consideration. Disk space if
free.
Bandwidth is soaring. The IBIS format should be relieved of artificial
barriers.
Our goal should be to make it easy for people to make good IBIS data sets,
not a
big pain.

jon

Kellee Crisafulli wrote:

> Hi all,
>
> At the risk of repeating myself there are several very good
> reasons to limit the number of points in the IV and VT tables.
> e.g. disk size, transfer time, overall accuracy improvement.
>
> The last one may surprise you but I do feel that including for
> example a 3rd VT table buys more than increasing the number of points
> for most models.
>
> I do agree the number of points could be increased to say 256 to
> pick yet another arbitrary spot on the wall. I would prefer to
> see a model creator like Arpad propose a new limit based on
> actual demonstrated requirements where some form of data reduction is
> used.
>
> I feel we should keep this number as small as we can and still create
> accurate models to insure the files sizes and download times are
reasonable.
>
> I think the biggest problem here is spice to ibis. People are always
> generating models with the maximum number of points. I see many many
models
> that will work EXACTLY the same with 1/3 this many points if they are
placed
> sensibly instead of using a linear spread.
>
> I am worried that if we set the limit at 1000 we will have all the new
> models with 1000 points......
> Just doesn't make sense to me sorry... especially with more and more
> VT tables being used.
> best wishes...
> Kellee
>
> At 05:00 PM 6/8/99 -0700, Scott McMorrow wrote:
> >I second Jon's opinon.
> >
> >There is no real reason to limit the number of points in
> >IV or VT tables. I have often used a simulator which allows
> >an arbitrary number of points over 100 to create high
> >resolution simulations. 256 to 512 is often useful to
> >capture rising edge timing information accurate to 1ps, along
> >with driver overshoot behavior to within 1 mv of HSpice
> >simulations.
> >
> >Model size makes no difference to me. It is the result that
>
> >matters.
> >
> >Scott McMorrow
> >SiQual
> >
> >jonp@pacbell.net wrote:
> >
> >> I agree with Weston.
> >>
> >> Either that or remove the quite artificial limit of number of points
in
> an IBIS
> >> file.
> >> I have actually been faced with the problem that even with software to
> remove
> >> extra points I cannot
> >> create IBIS models to the accuracy that I desire with the 100 point
> limitation.
> >>
> >> jon powell
> >> Viewlogic Consulting Services
> >>
> >> Beal, Weston wrote:
> >>
> >> > Kellee,
> >> >
> >> > You're disagreeing a mute point. Your point was the same point I
was
> trying
> >> > to state by saying, "the IBIS creator tool should have an algorithm
to
> >> > remove points in linear regions." We should delete all these extra
data
> >> > points. The problem that s2ibis has now is that all points are
equal
> >> > difference. They should be closer in knee points and very sparse is
> linear
> >> > regions. The IBIS creator tool should have an algorithm to clean
out
> extra
> >> > data.
> >> >
> >> > So, I agree with you and now you can agree with me.
> >> >
> >> > Later,
> >> > Weston Beal
> >> > Signal Integrity Engineer
> >> > Compaq Computer Corp.
> >> >
> >> > -----Original Message-----
> >> > From: Kellee Crisafulli
[mailto:kellee@hyperlynx.com]
> >> > Sent: Monday, 07 June, 1999 2:19 PM
> >> > To: Beal, Weston; ibis@eda.org
> >> > Subject: RE: [SI-LIST] : IBIS Subcommittee
> Formed to
> >> > Review Spice to IBIS
> >> >
> >> > Hi
> >> >
> >> > I have a comment on Weston's comment on the number
of
> data
> >> > points.
> >> > I disagree on one point. Instead of increasing the
> number
> >> > of points generated
> >> > in the IBIS file the spice to IBIS tool should
improve
> the
> >> > quality of the
> >> > points
> >> > choosen.
> >> >
> >> > I can hand build files with 50 points that out
perform
> files
> >> > with 400 points.
> >> > The big gain is that the FILE SIZE is reduced with
higher
> >> > accuracy.
> >> > Large IBIS files will take up many megabytes on
1000's of
> >> > hard drives.
> >> > The V/I and V/T tables are the largest part of IBIS
> files.
> >> > Increasing them 4 x will generally increase the file
> size 3
> >> > x or more.
> >> >
> >> > >The third thing that I changed was the number of
> points in
> >> > the time
> >> > >waveforms from 50 to 101. I have to remove one
point to
> >> > comply with the
> >> > >golden parser, but I get better detail. On the
issue of
> >> > number of points in
>
> >> > >a curve or waveform, I think the user should be
able to
> >> > define any degree of
> >> > >granularity and the IBIS creator tool should have
an
> >> > algorithm to remove
> >> > >points in linear regions. If the resulting curve
> still has
> >> > too many points,
> >> > >then increase the linearity tolerance and try it
again.
> >> >
> >> > Best wishes...
> >> > Kellee
> >
> ---------------------------------------------------------
> Have a great day....
> Kellee Crisafulli
> HyperLynx, a division of Pads Software Inc.
> SI,EMC,X-talk and IBIS tools for the Windows platform
> E-mail: <mailto:kellee@hyperlynx.com>
> web: <http://www.hyperlynx.com>
> ---------------------------------------------------------
Received on Wed Jun 9 11:49:46 1999

This archive was generated by hypermail 2.1.8 : Fri Jun 03 2011 - 09:52:30 PDT