[SI-LIST] Re: Convergence problem after inserting detailed connectormodel
- From: "Scott McMorrow" <scott@xxxxxxxxxxxxx>
- To: a.ingraham@xxxxxxxx
- Date: Fri, 20 Feb 2004 09:01:38 -0800
All,
I'd like to re-quote Dr Liu of Apache. For those of you who do not know
him, he is the designer and developer of NSPICE, so he does know what he
is talking about. Before I do, I will say a couple of things about
connector models and modeling in general. First, lumped element
connector models can often have dynamic range issues, because the
developer failed to remove small off diagonal coupling elements which
tend to cause ill-conditioning of the matrix in simulation. This problem
can be fixed by removing some of the extremely small valued couping
elements from the model.
Second, and foremost, occasionally you will find a connector or package
model where no matter what you do under some circumstances the model
oscillates. In this case, you are running into a passivity problem.
Developers will sometimes create these models with huge numbers of
coupled elements but do not take into consideration the errors involved
in the creation process with a field solver. Now, as long as there are
uncoupled R, L and C, errors are not a problem. But, as soon as
coupling is involved with mutual C's and K's it is now possible for the
errors to accumulate and cause non-passivity within certain frequency
bands.
One way to check this is to run the model in the frequency domain and
compute the eigenvalues of the S-parameters at each frequency point
with I - S*S'. If there are any negative eigenvalues then the matrix
is not passive. Dr Liu has had some experience with this, since his
software will simulate with standard SPICE models and with
s-parameters. He has found that not only can s-parameters be
non-passive, but also that lumped element models with coupling can be
non-passive. The culprit is usually mutual L. A simple test for this
in a multi-section connector model is to comment out all K elements,
re-run the simulation and see if it is stable. If the simulation is
stable then it is most likely non-passive. More formally it should be
run through an eigenvalue check, which Dr Liu reports to me that he has
done on a number of occasions. In that case, the model is bad and needs
to go back to the developer.
Third, there are problems with the HSPICE w-element time step and
bandwidth control algorithm. For long transmission lines the algoriithm
trys to rachet back both the internal timestep and bandwidth. This also
can cause interoperabilty problems with other large lumped element and
laplacian models. To force the algorithm to do the right thing you can
use the following dummy code which places a fast edge rate pulse source
in the simulation and forces the w-element to use a small internal
stepsize and high bandwidth for the inverse transformation algorithm.
This will also have the added benefit of forcing the algorithm to
produce the correct time domain wave shapes, which when left to it's own
devices it does not do. Here is the code:
vfrog frog 0 pulse (1 0 0 25p 25p 75p 200p)
rfrog frog 0 50
I talk about this in the Teraspeed/Samtec joint paper at DesignCon:
"Advances in Design, Modeling, Simulation,
and Measurement Validation of
High-Performance Board-to-Board 5-to-10 Gbps Interconnects"
So here is the promised Dr Liu quote. Notice the overlooked discussion
of non-passivity:
There is a myth about SPICE convergence problem like this one. Actually,
for a strictly passive network consists linear RLC elements, the
convergence problem should never occur - where is the nonlinearity in
the network? The "time step too small" is likely caused by large local
truncation error (LTE) rather than non-convergence. On the other hand,
if the network is not strictly passive (even a linear RLCK network can
exhibit non-passivity), and unfortunately we have seen this happens
quite common in real cases by poor modeling, you can also run into this
"time step too small" problem.
best regards,
scott
--
Scott McMorrow
Teraspeed Consulting Group LLC
2926 SE Yamhill St.
Portland, OR 97214
(503) 239-5536
http://www.teraspeed.com
------------------------------------------------------------------
To unsubscribe from si-list:
si-list-request@xxxxxxxxxxxxx with 'unsubscribe' in the Subject field
or to administer your membership from a web page, go to:
//www.freelists.org/webpage/si-list
For help:
si-list-request@xxxxxxxxxxxxx with 'help' in the Subject field
List technical documents are available at:
http://www.si-list.org
List archives are viewable at:
//www.freelists.org/archives/si-list
or at our remote archives:
http://groups.yahoo.com/group/si-list/messages
Old (prior to June 6, 2001) list archives are viewable at:
http://www.qsl.net/wb6tpu
Other related posts:
