[SI-LIST] Re: How accurate is HSPICE's field solver?
- From: Matthew Herndon <mherndon@xxxxxxxxx>
- To: si-list@xxxxxxxxxxxxx
- Date: Thu, 6 Feb 2003 10:44:43 -0800
I did some simple comparisions among 2D field solvers as part of a
paper at DesignCon in January. If anyone wants details, email me
offline.
Briefly, I looked at 7 field solvers: Ansoft, HSPICE (2001.4), IFS,
Quad, Greenfield, SigXp, and ApsimRLGC. Presence on the list is not an
endorsement, nor is absence from the list a criticism; these were
simply the 7 I had easy access to. I looked at two SIMPLE
configurations: a 1-conductor microstrip case and a 2-conductor
stripline case (nothing as complicated as what Scott describes below).
The results were that all the solvers were consistent within 2.5%, and
most results were within 1%. My conclusion was that accuracy is not a
concern with these simple structures; rather, when choosing a 2-D field
solver, one should look at ease of use and fit into existing design
flows, and price.
One member of the audience at my presentation commented that he saw
good correlation between 2-D solvers and measurements of boards, but
that the measured characteristic impedance within a package was often
off by 15% from the 2-D solver (I forget if he said higher or lower).
Does anyone have related experience?
One question for you, Scott: I work mostly at the board level, and
usually don't see much crosstalk except from the traces nearest the
victim (and so wouldn't need structures with more than 3 coupled
elements). Do you see significant crosstalk from farther away? Am I
missing something important?
-matt
Matt Herndon
ECAD Group
Apple Computer
On Friday, January 31, 2003, at 10:27 AM, Scott McMorrow wrote:
>
> Hassan,
>
> We've done some characterization work with it and found the solver to
> be
> quite sensitive to the mesh size that is used. This is controlled by
> the Accuracy and Gridfactor statement in the field solver call. (Yes,
> and do not believe the comment in the pdf documentation of the
> Gridfactor, it does in fact work.) The field solver uses a filament
> method, similar to the old Pacific Numerics field solver and equivalent
> to the FastHenry solver for resistance and inductance. This type of
> solver is sensitive to the number of filaments used in the extraction,
> which can also greatly increase the execution time. Just like these
> solvers, there is no adaptation of the mesh for the problem that you
> are
> solving, so you are left to your own devices to determine if the result
> is accurate enough, unlike a finite element field solver that is slow,
> but uses a minimum energy approach to determine the optimal mesh.
> Unfortunately, the Hspice solver does not give you any visability into
> the way in which filaments are divided, how many their are ... etc, so
> all you can do is to "play" with the settings and compare the results
> to
> another "golden" field solver. We've done this and compared it to
> Ansoft Maxwell 2D.
>
> We found that for simple models of single or differential transmission
> lines, the solver gives reasonable accuracy with 2% using the standard
> settings. (I am talking about versions 2001.4 or later. Earlier
> versions had some significant flaws.) But for large coupled models
> with
> 6 or more coupled elements, the accuracy falls off with the standard
> settings. Increasing the gridfactor will increase the resolution and
> give convergence to Ansoft Maxwell 2D results, but at the cost of
> increased solve time.
>
> Ultimately, for frequency dependence the solver only outputs an
> equivalent Rs or Gd, which is perfectly acceptable for lower frequency
> work (below 1 GHz), but is quite lacking at higher frequencies where
> losses are dominant. At that point we switch to w-element table
> models,
> with a table of frequency dependent RLGC tables. THe Hspice field
> solver will not produce this sort of model. What we do is to use
> Ansoft
> Maxwell 2D, create a frequency dependent parameter table to sweep
> against the trace geometry, and then extract the matrices from the
> Ansoft Admittance and Impedance results for each frequency, from 1 Hz
> to
> 100 GHz. These are then placed into a w-element table model format and
> simulated to a high degree of resolution.
>
>
> regards,
>
> scott
>
>
> Hassan O. Ali wrote:
>
>> Hi All,
>>
>> Has anybody checked the accuracy of differential RLGC parameters
>> computed by HSPICE's
>> internal field solver against those computed by other full-wave field
>> solvers such as
>> Ansoft Maxwell 2D Extractor, HFSS, etc, over a broad frequency band?
>>
>> Thanks.
>>
>> Hassan.
>> ------------------------------------------------------------------
>> 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 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
>>
>>
>>
>>
>
> --
> 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 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 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:
