Re: BIRD29 Banded_matrix Correction

Eric:

Thank you very much for your comments that are attached below. Since you
were the originator of the coupling matrix extension to IBIS, your assessment
of any change is critical.

I interpret your response that you would favor such an "expansion" to IBIS.
I will provide a revision to BIRD29 per your comments. It would be helpful
to me if you could review some wording changes (and/or provide alternative
text) to correctly deal with the Banded_matrix designation in IBIS.

PROPOSED ADDITIONAL TEXTS DESIGNATED BY "|*" :

| Banded_matrix
|
| A Banded_matrix is one whose entries are guaranteed to be zero if they
| are farther away from the main diagonal than a certain distance, known
| as the "bandwidth." Let the matrix size be N, and let the bandwidth
| be B. An entry [I,J] of the matrix is zero if:
|
| | I - J | > B
|
| where |.| denotes the absolute value.
|*
|* The Banded_matrix format conforms to this definition, except that it also
|* allows for the off-diagonal entries to describe coupling between the first
|* and last pins. The Banded_matrix format supports a constant-sized band
|* of entries for each row and is not limited to be in strict compliance with
|* the traditional mathematical definition.
|
| The bandwidth for a Banded_matrix must be specified using the [Bandwidth]
| keyword:
|
|==============================================================================

The BIRD29 issue is not one of mathematical purity, but of practical reality.
One of the main applications is for a simple user interface format for
flat pack package configurations, especially those with high pin counts and
small spacing. Often pin 1 is centered along one side and next to the
highest numbered pin. Even flat pack configurations for the standard
logic families have pin 1 centered along one edge. I think that the
"Banded_matrix" format is the natural and appropriate format, and it should
be defined to support the actual coupling between adjacent pins. There
is no value and there is actually harm to wait for Version 3.0 to make
the technical inclusion proposed in BIRD29 because the IBISCHK2 parser would
test and enforce a non-practical restriction while IBISCHK3 would operate
in a different manner.

Bob Ross,
Interconnectix, Inc.

***************************************************************************

> Date: Sun, 28 May 1995 09:20:58 -0400
> From: J Eric Bracken <bracken@kevily.ece.cmu.edu>
> Status: RO

> I have two comments on the BIRD:

> Comment 1:
> ---------

> This BIRD is more of an expansion in functionality than a "correction."
> I think it's highly useful--especially for PGA's, where the pins are laid
> out in a "circle" that closes back on itself. Please note, however, that
> the resulting matrix is no longer "banded" in the traditional sense of
> the word, because the pattern of entries will be something like this:

> XX X <-- this is not present in a "banded" matrix
> XXX
> XXX
> XXX
> XXX
> XXX
> XXX
> XXX
> X XX

> ^
> |
> nor is this

> That's why the original spec is the way it is--I thought people would
> use a Sparse Matrix for this type of structure. Incidentally, this is
> a simple form of a "bordered block diagonal" matrix.

> Could we add a phrase or two to the spec which clarifies that the New
> and Improved "banded matrix" is a generalization of the concept that's
> used in most of the world's literature? This might help to avoid some
> confusion with terminology.

> Comment 2:
> ---------

> I don't like the rewording of the comment on the last paragraph of
> the Sparse Matrix description. Particularly, the "for example" seems
> clumsy. How about:

> ** With this convention, please note that the N'th row of an
> ** N x N matrix has just a single entry (the diagonal entry.)

> __________________________________________________________________
> ________ ________ __ ________
> /\_______\ /\_______\ /\_\ /\_______\ J. Eric Bracken, Ph.D.
> / / ______// / ____ // / // / ______/ Visiting Lecturer, Dept. of ECE
> / / /__\ / / /__\/ // / // / / Carnegie Mellon University
> / / ____/ / / ___ _// / // / /___ Pittsburgh, PA 15213
> / / /____\ / / / \/ / / / // / /____\ (412) 268-8944 FAX (412) 268-3204
> \/_______/ \/_/ \/_/ \/_/ \/_______/ eric.bracken@ece.cmu.edu

> http://www.ece.cmu.edu/afs/ece/usr/bracken/.home-page.html
> __________________________________________________________________
Received on Mon May 29 11:11:01 1995