[SI-LIST] Re: HSPICE S-paramter questions
- From: "Dmitriev-Zdorov, Vladimir" <vladimir_dmitriev-zdorov@xxxxxxxxxxx>
- To: "Scott McMorrow" <scott@xxxxxxxxxxxxx>
- Date: Thu, 17 Mar 2005 09:24:11 -0800
Hi Scott,
Thanks for your reply. I would not discuss the silicon model related
issues, this is not my area.
Just a short note about passivity:
* the user is simulating pure passive interconnect. (a good
application of your software for passive characterizations.
There is no requirement that the interconnect model is passive. It can
be active and non-reciprocal. The only requirement is that the model is
linear and stationary, i.e. everything described as a touchstone.
BTW, ELDO supports slightly extended touchstone, where you can define
normalizing impedance individually for each port. Useful feature when we
need to put interconnect and power distribution models together.
Just a short question.
1.) At least one 3rd party solution using Hspice recursive convolution
is extremely fast.
Do you mean that it is HSPICE that performs recursive convolution?
Regards,
Vladimir
-----Original Message-----
From: Scott McMorrow [mailto:scott@xxxxxxxxxxxxx]
Sent: Thursday, March 17, 2005 9:18 AM
To: Dmitriev-Zdorov, Vladimir
Cc: ray.anderson@xxxxxxxxxx; silist
Subject: Re: HSPICE S-paramter questions
Vladmir,
Your points are well taken. However, there are several issues.
1.) At least one 3rd party solution using Hspice recursive convolution
is extremely fast. Extraction times are equivalent to yours, as are the
extraction times of passivity corrected models.
2) ELDO does not work with encrypted HSPICE models, which makes it an
unacceptable solution for most simulations with complex Silicon.
I have followed the progress of your s-parameter modeling approach and
am impressed. But, it still does not solve the basic problem of
simulation of real circuits in real systems with real silicon models.
Ray does have the advantage of being at a silicon company where he can
use ELDO models for the silicon. But, even he will have issues if he
wants to simulate the interoperability of their silicon with that of
other vendors, when the other vendors' models are in encrypted HSPICE.
Your software is generally good if:
* the user is simulating with IBIS models. (in which case the
accuracy is extremely limited)
* the user is simulating pure passive interconnect. (a good
application of your software for passive characterizations.
*
* the user has ELDO models of the silicon available.
Whenever accurate models of the silicon are required, HSPICE must be
used. It may not be the technically superior product, but it is the VHS
to your Betamax.
best regards,
scott
Scott McMorrow
Teraspeed Consulting Group LLC
121 North River Drive
Narragansett, RI 02882
(401) 284-1827 Business
(401) 284-1840 Fax
http://www.teraspeed.com
Teraspeed(r) is the registered service mark of
Teraspeed Consulting Group LLC
Dmitriev-Zdorov, Vladimir wrote:
Hi Ray,
So far you consider only two alternatives: using S-model based on
touchstone file or the n-port "circuit" model whose frequency response
approximates the touchstone data. Both have their own limitations. E.g.
direct use the tabulated data (touchstone) for transient simulation
leads to convolution-based approach that is slow (super-linear by
complexity) and often inaccurate. Possible trade-off between accuracy
and performance normally is not successful.
The equivalent n-port models (sub-circuits) produced by the third-party
model tools are good from those points but they add a large amount of
the circuit components (and sometimes internal nodes) that slow down
simulation from the other end. Such models are also not free from
accuracy issues. Although they do not suffer from inaccuracy inherent
for convolution, they develop LTE (local truncation error), as any other
circuits with LC elements or LAPLACE-type controlled circuits.
When solving transient for a long enough with models that have sharp
resonances, such errors become considerable and/or force using smaller
step. Even worse, the step selection mechanism cannot correctly predict
how much this error will accumulate over time.
There is a simulator (ELDO) that utilizes a third solution, free from
above limitations. In addition, it does not need third party conversion
tools and makes everything by itself, including passivity enforcement.
For re-use purpose, it generates the compact model in the intermediate
format that is neither touchstone nor circuit. Such S-model does not
generate convolution-specific error or LTE. On average, it is 5-7 times
faster (per step) than the corresponding equivalent circuit.
Example. Given the 158-port touchstone file of 214MB size (fully
populated S-matrix) it took 5 minutes to convert it into such
intermediate format (however 30 minutes if with passivity enforcement).
Then, simulation with 10,000 output points takes 3 minutes. The user
should not care about conversion: if needed it is performed
automatically on the initialization stage. The converted model is stored
and made re-usable. During re-simulation, the converted model is loaded
in seconds.
With the equivalent circuit, built from the same converted model, it
takes about 25 min for the simulator only to parse and check errors in
the subcircuit, before any simulation starts. This time cannot be
avoided if simulated repeatedly. Solution itself takes about 15 minutes.
Convolution-based approach (also available in ELDO) fails on that model.
Not because of poor implementation but due to inherent problems.
Vladimir
-----Original Message-----
From: si-list-bounce@xxxxxxxxxxxxx [mailto:si-list-bounce@xxxxxxxxxxxxx]
On Behalf Of Ray Anderson
Sent: Wednesday, March 16, 2005 11:57 AM
To: si-list@xxxxxxxxxxxxx <mailto:si-list@xxxxxxxxxxxxx>
Cc: scott@xxxxxxxxxxxxx
Subject: [SI-LIST] Re: HSPICE S-paramter questions
Scott-
I agree with you completely regarding the speed issue. Another factor to
consider is stability/passivity. S-parameters originating from
measurements (and quite often from simulation/extraction) can exhibit
stability/passivity issues. I do not believe the S element checks and/or
corrects these issues. Most of the available s-parameter to n-port model
tools available have passivity enforcement capability such that the
synthesized n-port model is guaranteed passive. Some of the tools also
have the capability of regenerating a passivity enforced set of
s-parameters from the original input data set. This is accomplished by
"nudging" the problematic parameters until passivity is achieved. In
most cases the response of the resulting data set is acceptably close to
the original (but not passive) data set.
-Ray
Scott McMorrow wrote:
>Ray,
>The other reason for using these other tools is speed. For example,
>after conversion of a Xilinx differential s-parameter model to a
>Laplace pole-zero model, I typically see a 20X speed up in performance
>when compared to the Spice s-element simluations. If you are doing
>only one simulation, then the translation time dominates. But if you
>are doing multiple trace length, connector type, backplane sweeps in
>simulation, then the performance advantages of the Laplace pole-zero
>model is significant, with no decrease in accuracy.
>
>scott
>
>
>
>
--
Raymond Anderson
Senior Signal Integrity Staff Engineer
Product Technology Dept.
Package Engineering Group
Xilinx Inc.
------------------------------------------------------------------
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 FAQ wiki page is located at:
http://si-list.org/wiki/wiki.pl?Si-List_FAQ
List technical documents are available at:
http://www.si-list.org <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
------------------------------------------------------------------
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 FAQ wiki page is located at:
http://si-list.org/wiki/wiki.pl?Si-List_FAQ
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:
