RE: comments on BIRD 31.1

To: David Fogel & IBIS Forum

David raises good questions. My responses are inserted in his text below,
set off by "+++++" for easy location.

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: owner-ibis
To: ibis
Subject: comments on BIRD 31.1
Date: Thursday, April 11, 1996 4:38PM

To: ibis@vhdl.org

Subj.: Mated connector model, BIRD 31.1

Regarding the proposed BIRD31.1 specs, I have the following
questions/comments

1. Power and ground pins assignment will affect L,C values

1. Ground (and power) pins assignment will affect both the self LC values
of a pin and the LC coupling matrix, based on the position of each pin
relative to the ground pins.
Consider the following three connector pins example:

                o o o
                ^ ^ ^
                | | |
               gnd 1 2(or gnd)

The pins are 50 mils apart.

When only the center pin is a signal and the other two are ground I get:

SINGLE CONDUCTOR LC:
____________________

 i j Lij Cij Ze Zo Se So Fwdx Rvsx
from to (nh/in) (pf/in) (ohms) (ohms) (ns/ft)(ns/ft) (s/s) (v/v)
 --------------------------------------------------------------------------
  1 1 10.537 3.066 58.63 - 2.16 - - -
;

and when only one side pin is a ground I get:

DUAL CONDUCTOR LC:
_________________

 i j Lij Cij Ze Zo Se So Fwdx Rvsx
from to (nh/in) (pf/in) (ohms) (ohms) (ns/ft)(ns/ft) (s/s) (v/v)
 --------------------------------------------------------------------------
  1 1 15.880 3.066 88.36 - 2.16 - - -
  1 2 10.903 1.502 130.90 33.01 2.46 1.81 0.303 0.597
  2 2 22.247 2.188 123.78 - 2.16 - - -
;

Which clearly demonstrates the fact that you can not assign LC values to
connector pins just by their physical location in the connector. You also
need to know their position relative to the ground pins.

This can be accomplished by depicting several most likely to be used
signal-gnd
configurations and providing the LC matrices to all of them. In a dual
row connector you may end up with about 10 different combinations.
This kind of information is precisely what's needed for high speed
simulations and will make the modeling effort worthwhile from system
engineering point of view.

+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
+ From a pure academic perspective, you are correct, but there is another
way. We at AMP have devised proprietary methods that enable us to develop
SPICE models that behave correctly without having defined the ground
locations. You get the right behavior by just connecting the model to the
actual circuit wiring pattern. There are full series and mutual parameters
in the model for every line of the connector. Obviously, I can not share
those methods but I feel that they are applicable and should be permitted by
IBIS. If they are not permitted, there will have to be an overwhelming mass
of specifically modeled configurations for each connector.
+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

2. Another level of complication in a mated model is the plug-in daughter
card. The fingers themselves are part of a PC board. How long are those
fingers in the mated model? is there any definition for it?
The electrical characteristics can be affected by the presence (or absence)
of a ground plane in between the daughter cards' top and bottom layers (the
finger layers). How is this going to be reflected in the model?

+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
+ This is exactly my point for the need to define model boundaries. We have
definitions for each case. For card edge connectors, the model includes
the recommended pads (as shown on the AMP customer drawing). There are
separate models for boards with and without ground planes where those are
common practices.
+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

3. Another issue is measurements and verifications of the model itself.
Since we are talking about relatively small LC values, the test set-up
can greatly affect the measured LC values.
You may build a very elaborate test fixture at your facilities and obtain
very precise figures. But can every vendor or user do the same?
How do we decide which values are the right ones. I believe that we need a
well defined measurement methodology to be included as part of the
specs.

+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
+ This is a real challenge. EIA/JC-15 has been struggling with this for
packages for some time now. Capacitance is easy compared to inductance
measurements. You can not measure a package or connector to get the
inductances that are in the model. All measurements find the effects of all
series and mutual inductances in the entire set of current loops. The
values in the model are the partial parameters that work together to behave
correctly. I do agree that some defined validation process should be sought
but I suggest that it should measure behavior, not partial parameters.
+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

4. Another method for model development is to use the mechanical drawings
and
feed the physical dimension to a field solver. If precise mechanical
dimensions,
including the ER of the connector body, can be provided, then a field solver
can
be used to extract the LC matrices. This can be an alternative to the above
methods.
The issue here is which field solver to use? 2D? 3D? from which CAE vendor?
etc.

+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
+ I believe there are at least three major problems with this approach.
 First, I know we do not want to release manufacturing level drawings or
files to the world. I think other companies would feel the same way.
 Second, field solvers are expensive, slow, require considerable expertise
to use correctly, and generate very large model files. Also, there are many
common structures that can not be analyzed directly with 2D solvers and the
3D solvers are really difficult to use correctly. Finally, how do you know
the model is good? Validation is another expensive and time consuming
engineering specialty, as is modeling. Bottom line, the suppliers should
provide validated models.
+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

David Fogel
Received on Fri Apr 12 11:37:56 1996