RE: Re[2]: Connector spec swathing

I agree with Mikhail. I actually push the connector vendors for smaller
matrices, for 30X30 connector matrices cause very long run times in HSPICE,
even using Ultra 60 Sparcs with 500MB of RAM. It's very unreasonable to
expect larger than 30X30 connector LRC matrices.
Regards,
Fabrizio Zanella
Hardware Engineering
EMC Corporation
508-435-2075, x14645
FAX: 508-497-8027
fzanella@emc.com

                -----Original Message-----
                From: Mikhail Khusid
[mailto:Mikhail_Khusid@notes.teradyne.com]
                Sent: Thursday, June 15, 2000 2:29 PM
                To: Chris Rokusek
                Cc: IBIS Mailing list
                Subject: RE: Re[2]: Connector spec swathing

                Chris,

                Whether a connector company can make a full matrix model of
a connector
                depends more on the connector than the tools. In fact, if a
connector is
                complicated enough to warrant several L and C sections, and
if a connector
                has many pins, providing a full matrix model can be very
unpractical.

                As a representative of the connector company where a 570 pin
connector
                is considered a small one, I can assure you that it's
impossible for me
                to generate dozens of 570x570 matrices. Furthermore, I
doubt that any
                simulator will be able to solve a problem with full matrices
of that size.
                Doing a swath method allows me to generate reasonable size
matrices,
                up to 30x30, which should respresent sufficient couplings
inside
                the connector, and thus is an effective way to simulate the
connector
                behavior in reasonable time.

                Lastly, I agree with Kellee that a standard should probably
include
                a description of a "golden" way to generate a full matrix
model out of a swath,
                however, I would welcome simulator companies attempts to
avoid using
                this method for every connector.

                Michael Khusid
                Teradyne Connection Systems
                http://www.teradyne.com

                "Chris Rokusek" <crokusek@innoveda.com> on 06/15/2000
12:27:55 PM

                To: "IBIS Mailing list" <ibis@eda.org>
                cc: (bcc: Mikhail Khusid/NNH/Teradyne)
                Subject: RE: Re[2]: Connector spec swathing

                Gus,

                Given that some companies actually do possess the tools to
create Full
                Matrix connectors (& packages) accurately and efficiently
and given that
                simulators MUST come up with algorithms to reduce this full
matrix (perhaps
                efficiently with an accuracy tradeoff), simulator companies
are just trying
                to avoid extra work in having to write two algorithms that
do very similar
                things--one to expand the swath to arbitrary pins and one to
reduce the full
                matrix to arbitrary pins. Many of us prefer to solve the
most
                general/accurate problem first (Full Matrix).

                I don't think we're saying "don't use the swath", I think
all we're saying
                is provide us with a recommended algorithm (and perhaps
build it into the
                parser) for expanding the swath into a full matrix so that
each simulator
                can potentially get the same answer for the most general
case. It seems
                like that would keep everybody happy.

                Chris Rokusek
                Innoveda

> -----Original Message-----
> From: apanella@molex.com [mailto:apanella@molex.com]
> Sent: Thursday, June 15, 2000 4:37 AM
> To: fred
> Cc: Kellee Crisafulli; IBIS Mailing list
> Subject: Re[2]: Connector spec swathing
>
>
> Fred,
>
> I am sure that if you have this question, so do others...
as such I have
> prepared a fairly lengthy reply.
>
> The "basic" answer to the question is...
>
> Find a FIELD simulator that can generate a 100 pin
connector model from a
> connector with a general current path length of around
2.5cm.
>
> Make sure it is a true 3D simulator.
>
> OK... now find one that can do such using less than 8GB of
RAM
> and 4 xxxx GHz.
> processors
> - This is pretty much what one might call a high end
> PC/Workstation. Maybe
> even not "typical" hardware.
>
> OK... Using what has been found above... have the
problem solve
> in the FIELD
> simulator in less than a week. (BTW, a week is solve
time, it
> does not include
> reports, empirical confirmation, or support documentation)
>
> OK... now put that full matrix model that is generated
into a circuit
> simulator.... Setup the rest of the problem.... go away
for the
> weekend...
> comeback Monday.... still not done... Comeback
Wednesday...
> Ooopps found out
> that a termination resistor was misplaced... restart the
simulation.
>
>
> The point is... connector companies would be perfectly
happy with smaller
> models...
>
> But from what I have been told my customers which are the
same as many SI
> simulators NEED to be able to model 50, 100, 200, 500,
1000, 5000
> pins... (Yes,
> these are all real connector sizes today).
>
>
> As a work around for this problem.... we suggest the
preverbal
> "critical net
> analysis". Which will still take 8 hours to solve using
something
> in the order
> of a 40 pin connector model.
>
> Also, I as a connector manufacture make parts with
different
> circuit sizes...
> sometime from 10 to 1000 in 10 pin increments... OK...
there we have 100
> models... Lets say we just go up to 100 pins... that is
10
> models... at 3 days
> AVERAGE per model (smaller models take less time)... that
means
> one month of
> FIELD SIMULATION time... and we STILL can not support
any thing
> over 100 pins.
>
>
> Not to mention that connector companies have about 40,000
> different product
> lines.. lets say conservatively that only 1% of the 40,000
> require models...
> That's 400.... a conservative estimate would be that there
are 10
> circuit sizes
> for each of those 400 connectors... as such 4,000
different
> models. OF EACH
> Type. There are three basic types Single Line models,
MultiLine
> Models, and
> Cascaded Models... then there are TWO VERSIONS of each
type...
> distributed and
> lumped... grand total 24,000 SPECIFIC models.
>
> But wait, Now model makers and simulators also have to
database
> and revision
> control the models.
>
> Point being... an auto swath will give end uses access to
more
> models with more
> pin options than they ever had before (to answer the
customers
> requests). And
> GREATLY reduce the redundancies such that the 24,000
models above
> can be done in
> 1 model per connector family.. or 400 files.
>
>
> _gus: 630-527-4617
>
>
> ____________________Reply Separator____________________
> Subject: Re: Connector spec swathing
> Author: fred <fred@apsimtech.com>
> Date: 6/15/00 1:33 AM
>
> I know the connector specification committee has spent
much time
> and effort in
> comming up with the specification. However the current
swat
> matrix approach
> seems overly complicated and technically less than
desirable.
> Why not give the
>
> full matrix and let the simulation SI tool decide which
part of
> the matrix to
> choose
> for simulation based on what pin and coupling is desired.
We (simulation
> vendors)
> only need the data. We can decide how and when to use
what. What
> we need is
> the committee to do is identify the connector pins to the
matrix diagonal
> entrys. If
> the connector is very large in terms of number of pins
then
> whether one gets a
> full
> or sparse matrix will depend on the field solver
capabilities. This is not
> intended to
> be critical of the committee which has worked long and
hard to
> come up with a
> spec
> in the first place while hence keeping everybody happy.
>
> Best Regards,
>
> Kellee Crisafulli wrote:
>
> > Hi Chris,<SNIP>
>

                
Received on Fri Jun 16 05:24:48 2000