New IC EM simulator integrated into Cadence

[GoaGosha]

Cadence gradually introduce their integrated EM solver for RFICs.

As I understand the basic concepts of that solver can be found in this doc:
www.cadence.com/community/virtuoso/resources/LargeScaleBroadband.pdf

It would be interesting to hear all the disadvantages of that tool.
( the advantages I hear from Cadence )

Jian - can you comment this solver , it uses IE and seems to be
most closest in concepts to IE3D.

I also will be happy if Jim Rautio and other people here which are deeply in
computational EM could commnent it.

Thanks ,

Gosha

 

[jian]

Hi, GoaGosha:

I also heard about it. However, I have not tried it and I would like not to give comment on it. Thanks!

Best regards,

 

[GoaGosha]

This tool is new so I don't expect people here having expirience with it.
I was actually more interested theoretically, I'm sure that people
deeply in computational EM will see few apparent weaknesses in above
paper.

Thanks

 

[rautio]

Hi Gosha -- I have met with the primary author of that paper on several occasions. He came to my Maxwell biography presentation in Phoenix, and we also had a very nice dinner together. In our first meeting, Feng Ling actually sat down with me at the 2004 IMS and we spent an hour or so going over details. He even demonstrated the software to me. I am quite impressed with it. Feng is exceptionally intelligent and a very pleasant person to be around.

High level, they are going for reduced accuracy analysis of very very large circuits, much larger than can be done by any of the present EM tools, including Sonnet. In doing so, they do make approximations. For example, they approximate the far field coupling between subsections by assuming the current generating the field is an equivalent dipole. For many useful circuits this is OK. At some frequency, it is not OK. It can also limit the size of a subsection because a big subsection has a larger near field that can not be estimated.

Another limitation is that they are using only triangle subsections (RWG basis functions). Triangles, as discussed in other threads, do not represent linear current well, or at least I have never seen a high quality current distribution come out of pure triangle subsections. However, a low quality current distribution (i.e., ragged appearance, indistinct high edge current) can still produce S-parameters that have a "pretty close" kind of agreement. And perhaps they have now included rectangle subsections too.

Another factor is the iterative matrix solve. When I talked with him, Feng said that they had no problem with lack of convergence. I wonder how many circuits he had tried at that point, however. Another problem with iterative matrix solve is that one solve must be done for each port. If you have 100 ports, then 100 iterative solves must be done, and they must all converge. Thus, this approach is strongest on circuits with fewer ports.

Also, this tool is an unshielded analysis. As such, the highest accuracy port calibration (de-embedding) is not available to it, but it can be approximated. While they can (and perhaps actually have) included an ability to simulate a shielding box surrounding the circuit, it is likely that it will not be useable for box resonances, filter enclosures, etc., due to approximations that must be made. If they do at some point offer a shielded mode for this analysis and it is important to you, you should check it to be sure.

I am sure there are other considerations of which I am not aware. There are also considerations that I can not discuss due to confidentiality.

I view this tool as being useful, given you are aware of both its advantages and disadvantages. Anyone who believes that this is the only EM solution you will ever need for all your planar EM problems is being seriously foolish. I view this as minimally competitive with Sonnet as we are a shielded analysis, and they are unshielded. We are also now recommending design approaches that include large numbers of internal ports (e.g., 50 or 100 or more) and I think the Cadence tool is unlikely to excel in this problem space. We go for the highest possible accuracy at all frequencies, they accept moderate accuracy at lower frequencies. Contingent on positive reports from at least a few customers I would recommend this tool to our customers who do larger circuits at lower frequencies.

From a business point of view, I think it is most interesting that Cadence is taking such a strong interest in the RF/microwave CAE field. Cadence and Agilent have a close relationship, however, this tool seems to be at least moderately competitive with Agilent Momentum, both are unshielded analysis tools. Cadence is also playing around in other areas of microwave design. The future will be interesting to watch.

One legal point, the IEEE copyright allows authors to distribute copies of their own papers, but I was told that the IEEE views posting IEEE papers on the author's company's web site as exceeding that right to distribute. Reading the actual copyright statement that the author's sign seems to leave it a bit unclear.

Several hours later...I checked out https://www.ieee.org/web/publications/rights/policies.html and ...

D. Personal Servers. Authors and/or their companies shall have the right to post their IEEE-copyrighted material on their own servers without permission, provided that the server displays a prominent notice alerting readers to their obligations with respect to copyrighted material and that the posted work includes the IEEE copyright notice as shown in Section 8.1.10A above. An example of an acceptable notice is:

"This material is presented to ensure timely dissemination of scholarly and technical work. Copyright and all rights therein are retained by authors or by other copyright holders. All persons copying this information are expected to adhere to the terms and constraints invoked by each author's copyright. In most cases, these works may not be reposted without the explicit permission of the copyright holder."

 

[GoaGosha]

Huge Thanks Jim.

One point that I'm not sure I understood is:

"approximate the far field coupling between subsections by assuming the current generating the field is an equivalent dipole. For many useful circuits this is OK. At some frequency, it is not OK"

Are you talking here about causality issue - like for instance lengthy port
whose length is not much smaller that wavelenght ?

Thanks,

Gosha

 

[rautio]

Hi Gosha -- Actually, what I am talking about is that by approximating the far field of a patch of current by a dipole, the voltage induced on one patch caused by the current on another patch will be calculated slightly in error. This is the number that goes into the MoM matrix. If this number is in error, then error is introduced to the final result. In some cases this will be small. In other cases, the error will be significant. This problem gets worse at high frequencies. Also, even at low frequencies, if you have some large subsections, they have a larger "near field" that requires a full MoM calculation for all the small subsections that are inside the large near field of the large subsection. This reduces the speed of the analysis. Of course, at all times, subsection size should be kept small with respect to wavelength.

 

[GoaGosha]

So the error will be determined by the amount of far fieled differences
between 'current element field generator' and 'dipole field generator'?

 

[rautio]

As per my understanding of Feng's description in his paper and our subsequent discussions, that is correct. Of course, there will be a judgement call as to what is far-field and what is near-field. We actually looked into doing something very much like this, but we found a small but significant percentage of our test cases changed to a degree that we considered unacceptable for the type of (high accuracy) solutions we need. There are, of course, many other error sources as well, as there are in all numerical EM analyses.