Re: Modeling - Fast Tr

Wai Yeung Yip ([email protected])
Fri, 6 Sep 96 13:36:05 PDT

I disagree with Ed that one rarely knows what bandwidth to consider the S
parameters over for package interconnect modeling. If that was the case, we
would not know when to switch from a single lump model to a more elaborate one.
We certainly can measure or
model to determine the output waveform of the driver. We can then compute the
frequency spectrum of the signal. One can then define the required bandwidth
for the model. Of course, it does not have to be a 3dB bandwidth, it can be
anything with which one feels comfortable. Granted the drivers and receivers
are nonlinear. But that does not come into the picture in package interconnect
modeling. The package itself is a linear structure.

Linear system analysis would not yield the correct frequency spectrum of the
driver output when the loading conditions change. But under no load condition,
the driver output should be the fastest and thus with the highest bandwidth.
This can thus be used to establish an upper bound on the bandwidth of the driver
output under all loading conditions.

By the way, I think I speak for all to suggest subscribers to this list to
update the administrator of any address change ASAP. Otherwise, the amount of
undeliverable, returned messages to sender would eventually discourage people
from posting to the list.

Wai-Yeung Yip

> From [email protected] Thu Sep 5 15:06:23 1996
> Date: Thu, 5 Sep 1996 15:58:21 -0400
> X-Sender: [email protected]
> X-Mailer: Windows Eudora Light Version 1.5.2
> To: [email protected] (Wai Yeung Yip), [email protected]
> From: "Dr. Edward P. Sayre" <[email protected]>
> Subject: Re: Modeling - Fast Tr
> Content-Length: 1952
> I do not agree that s-parameter is the best approach since you rarely know
> what bandwidth to consider the s-parameters over. The microwave idea of a 3
> dB bandwidth has little or no meaning when one is considering the non-linear
> nature and threshold of digital drivers and receivers. Based on many years
> of high performance interconnect design for digital systems, I agree with
> Fabrizio Zanella's approach. It is accurate, and if properly done, does not
> ignore coupling or any other crosstalk effects and is suitable for use in
> either IBIS or SPICE simulations.
> The choice of how big to make the subsections is entirely dependent on the
> "electrical size" of the interconnect and can be directly related to the
> risetime of the signals. The rule of thumb of 10 time samples per risetime
> translates into something like a small fraction of a wavelength of a fifth
> harmonic of the fundamental. Those of us with a background in fields know
> this corresponds to a "electrically small" object in the frequency domain.
> Measurements directly compared to theory shows this to be an excellent
> approach. On the cautionary side, one does not want to over sample the
> connector. More is not always better since the Green's functions describing
> the connector may not converge well of the subsection length is too short.
> The rule of thumb stated above works for most field simulators and has the
> theoretical basis for being correct.
> ed sayre
> +~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~+
> | ------------------------------------- |
> | "High Performance Engineering & Design" |
> +~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~+
> | Dr. Ed Sayre e-mail: [email protected] |
> | NESA, Inc. |
> | 636 Great Road Tel +1.508.897-8787 |
> | Stow, MA 01775 USA Fax +1.508.897-5359 |
> +~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~+