Re: BIRD 31.1

Hello John,

Thank you for your review of my comments to BIRD 31.1 and your question
(Included below). 'Minimum edge rate' is an often ignored and misunderstood
point. That is why I am also sending this answer to the IBIS Reflector.

There really seems to be two parts to your question. One is what 'Minimum
edge rate' means to a connector, and the other is how does it relate to the
model? I will answer the second question first.

Connector models are built up from sets of R, L & C matrixes. The models
currently available, represent those matrixes in SPICE Format. Once we
define a means of modeling connectors in IBIS format, they will be
represented by numeric matrixes. These matrixes actually represent pieces
of transmission lines. To understand why there is a minimum edge rate
associated with these RLC structures, we need to look at some Transmission
Line (TL) theory.

A lossless TL can be represented by a characteristic impedance and a time
delay, or by an LC pair. The first is called the distributed model and the
second, the lumped model. Distributed models are like real TL's. Lumped
models are an approximation that is useful in circuit modeling. These two
types of representations are equivalent only if the time delay through the
lumped model section is less than a small part of the rise length of the
signal going through it. What's that mean?

For digital signals, there is a distance that the signal travels during the
time it changes. That distance depends on the medium through which the
signal travels. It is called the rise length. It is significant because
there is a transition taking place all along that path. The signal, on a
circuit that is one rise length long, can not be considered a constant.
value. It you take a small enough part of the rise length, you can make
that simplifying assumption.

Now for an example. Lets look at a lossless, matched TL in a digital
system. Values of 10nH and 4pF represent a 50 ohm TL that is 200ps long.
[Z=sqrt(L/C) & T=sqrt(LC)] If a 200ps edge rate signal is placed on a real
TL like that, the signal will be fully switched at the beginning before any
of it gets to the end. In the real TL, it will come out looking the same as
it did when it started (in the matched condition: ie, 50 ohm source & load).
 If the TL is represented with one 10nH inductor and one 4pF capacitor, it
will look very different. There will be a large amount of overshoot and
undershoot. If those values are divided into small equal sections, the
simulated signal will be much more realistic. Depending on the specific
situation, values from one fifth (1/5) to one twentieth (1/20) of the rise
length are used for the subdivision. This example could be modeled with
reasonable accuracy using 10 LC sections in series with each L & C being 1nH
and 0.4pF respectively.

That is a very brief explanation of the first factor that is significant in
determining a model's Minimum Edge Rate. Simplistically, if you put too
fast an edge rate signal through a model, it will ring like a church bell
and give inaccurate answers.

There is another factor that comes into play in connector models. It has to
do with the 'resolution' to which the connector structure is represented in
the model. More of the structure becomes visible to the signal at faster
edge rates. It is therefor necessary to include an increased level of
detail in 'faster' models. Why not just provide all the detail? Computer
resources and execution time go up drastically as detail is increased.
 Also, it is not at all obvious (it is a connector modelers engineering
task) how to reduce the detailed model to the slower version. That is why
we provide a family of models for each connector.

You see now, that there is a danger of obtaining unrealistic simulation data
if adequate models are not used. (This also applies to package models.)
 That is why I feel it is important to provide a way that the simulator can
determine the validity of the model for the simulation. The 'Minimum Edge
Rate' keyword would define the upper limits for applying the model. It
could also allow the simulator to pick the simplest model for the job.
 Including it would be the fist step toward those capabilities. After that,
the simulator writers have to figure out how to use it.

The other part of your question relates to what distinguishes high speed
connectors from ordinary ones. It does not relate directly to IBIS, so I
will be brief on this part. Again it is a transmission line problem. A
good high speed connector causes the least amount of distortion to the
signals. This is accomplished through management of the EM fields,
minimizing line to line couplings, and minimizing common mode impedance, to
name just some major factors. There is no one perfect design and it costs
more to make connectors that do those things well. That is why we have so
many different types and why they cost more.

I hope this answers your questions adequately.

Hank Herrmann
Technical Staff Member
AMP Incorporated, Electromagnetic Technology
M.S. 106-14
P.O. Box 3608
Harrisburg, PA 17105-3608

            Phone: 717-986-5534
                 FAX: 717-986-5643
   INTERNET: hank.herrmann@amp.com

 ----------
From: John.Fitzpatrick
To: Herrmann Hank
Subject: Re: BIRD 31.1
Date: Thursday, April 04, 1996 9:50AM

> 1. The Minimum Edge Rate for a package, connector or cable model should
be
> specified as a required parameter. A required Keyword, such as
> [Minimum Edge Rate], should be added. Eventually, simulators should
> generate a warning message if faster signals are impinging on the
model.

Hank,

I have often wondered how a "minimum edge rate" for a connector is defined,
and how it should be used. Often it seems to be just a sales argument to
encourage
us to buy more expensive connectors :-)

Can you explain what a [Minimum Edge Rate] keyword will mean exactly? Why
should
faster signals not be allowed to impinge on a model? A simulator should
base
its
warnings upon the result of simulations.

I agree that this keyword might be useful in helping a designer choose a
connector.

Regards,
John

 --
John Fitzpatrick <John.Fitzpatrick@ln.cit.alcatel.fr>
Alcatel CIT, 4 rue de Broglie, 22304 Lannion, France
Tel: (+33)96.04.79.33 Fax: (+33)96.04.85.09
Received on Thu Apr 4 07:15:09 1996