Re: package and connector model questions

From: Kellee Crisafulli <kellee@nwlink.com>
Date: Thu Sep 23 1999 - 20:45:00 PDT

Hi Aubrey all,

At 03:59 PM 9/23/99 -0500, Aubrey_Sparkman@Dell.com wrote:
>The first question has to do with Kellee's outstanding item below. I.E.,
>why not use the RLC matrix description in section 7 on package modeling? I
>am not sure why Connector modeling isn't (or couldn't be) a subset of the
>package modeling spec. After all, a package is a connector that connects
>the silicon to a PCB....or socket, which is again another connector. And
>before someone comments "un-mated connector" I also wonder why I need the
>model for an unmated connector or a package out of the socket, etc.

Wow good comments... here is some of our thinking but many more
heads and ideas made these decisions than just mine.

The connector group made several decisions up front:
1) The specification should be stand-alone to remove the overhead
     associated with the IBIS IC specification which is getting
     very complex. Several of the connector companies that were interested
     in using the connector specification expressed concern over the
     need to understand the full IBIS specification in order to make a
connector
     model.
2) The matrix descriptions must be compatible with the IBIS .pkg
    specification.
3) The general syntax must be the same as the IBIS specification.
4) Some of the concepts needed for connectors were missing from
    the .pkg specification and were much easier to accomplish as a stand
    alone specification. You mentioned a few. The biggest concern I
personally had
    is that since the specification was constrained to just connectors
there was less
    disagreement over what and how to do it. We attempted this once before
    integrated in the main specification and the project failed. One of
the reasons
    was difficulty getting agreement on how to proceed with various aspects.
    Another was different groups had different needs i.e. cables v.s.
connectors.
    v.s. IC packages and that added another set of problems that caused
disagreements.

I do believe the connector specification concepts and keywords could
be used to augment the .pkg model since the connector specification
is mostly a super set and contains many improvements.
The people in the connector group fully expected many of the features to
migrate directly to the .pkg specification over time.
We also have many other new features like support for .jpg images of the
connector
and web site and email paths that can be added directly into the file so a
user
can easily get the latest version model or contact the department that created
the model. These features should probably migrated back into the IBIS IC
specification. Can you image being able to access the correct web site by
just clicking on your connector model picture. Or being able to email the
correct
department of the company that created the model by clicking on an email icon.
way... cool. if the simulator folks pick up on it.

I found it is often a mistake to try and make one size shoe fit everyone.

If we can use the same shoe leather, laces and polish and manufacturing
equipment we are doing very good. (i.e. the syntax is compatible and we
could re-use the concepts and keywords directly). If the syntax is left
compatible the simulators should have an easy time supporting either
with the same code and parsers.

>But there are some excellent modeling enhancements in the connector spec
>such as Begin/End_Cn_Swath. Great way to simplify what could be an
>unnecessarily large matrix. Can we put this in the package model section?
>At least where possible, let's keep (or copy) everything that is or could be
>the same. And I also like the ability to name and therefore have more than
>one RLC matrix.
Speaking for the connector sub group we appreciate the kind words .

>Which brings me to my second (and more important to me) question. As I read
>section 7 of Ver 3.2, I can have only one RLC matrix set. I can't put two
>RLC sets in series. Or, if I want to do distributed effects, I can do
>sections (a wire bond section, a trace section, a via section, another trace
>section, etc) but I can NOT have "trace to trace" coupling effects. And I
>can not connect several uncoupled sections to the single RLC matrix which
>would at least give me some coupling. Is this a correct statement of my
>options?
>I was hoping for the ability to connect more than one matrix in series.
>And since the point of having a matrix is the coupling, I was also looking
>for the typ, min, and max coupled matrix for L and C.

The connector specification allows for as many series matrices as
you desire and the matrices may all be of different types. So the
first matrix might be simple inductors to represent bond wires and
then a series of diagonal matrices to represent transmission lines
and then a series of full matrices to represent a coupled region.
This method could be added to the .pkg file very easily if the IBIS
group desired.

I should be able to complete the final integration of feedback and release
the connector specification to the full IBIS group very soon.
  We have received comments from the full group already at
DesignCon 99 and we are hoping for rapid acceptance.
I am only working 3 days a week so my schedule is limited and it
is taking longer than I would like to complete. Hopefully within the next
one to two weeks.

best wishes...
Kellee
Received on Fri Sep 24 00:15:07 1999

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