sonnet em vs hfss

Hi everyone! I am a newbie in both RF & HFSS. Therefore, before I start performing simulation for my new design using HFSS, I tried to verify my setting (in HFSS) by comparing the result obtained with those of SONNET and Momentum (published in a journal paper). It's a simple problem concerning a square spiral inductor.

I found out that SONNET gave a great comparison with the Momentum result. However, HFSS result seemed to be very different, particularly for the inductance value. The self-resonant frequency of the spiral inductor seemed to fall at a very high frequency compare with those in SONNET & Momentum.

I then noticed that in most journal papers the inductance peak is not as sharp as the one which appeared in HFSS result. I wonder whether I did something wrong in HFSS modeling. I actually more or less followed the tutorial in HFSS. Attached please find the results and HFSS file.

Any advice or comment is truly welcomed and appreciated. Thanks.

Without going through your model in each of the programs, my first reaction is "of course there will be differences".

Each of the software suites uses a different mathematical method to analyze the model. HFSS uses 3D FEM, Sonnet uses FFT-Based analysis, and Agilent's Momentum uses Method of Moments (MoM).

You're comparing different mathematical assumptions, and calculation methods. The only way to verify which one is the closest model for your needs is to actually build a prototype, and compare it with your calculation results.

Use Google and enter - "compare results HFSS momentum sonnet" (without the quotes). You'll get a long list of papers that have been written over the years doing exactly what you are doing - comparing and questioning the results from the various software packages available.

Based on the uploaded results there is a fundamental (possibly wrong) assumption in the ground conditions in the HFSS setup.

L increases significantly and with considerably high slope as it approaches resonance. Qualitatively, those abrupt changes on the way to resonance are usually due to the ground assumptions you have made (or boundary conditions). Theoretically, total absence of a ground reference (this is impossible!) would result in extremely high resonance frequency.

I would suggest re-checking your simulation setup for HFSS.

Oh, thanks for showing me the direction! I have been puzzled by this substantial increase of inductance for days.

I actually drew a 2-D rectangle at the bottom of the silicon substrate, and set it as ground plane. So, you mean that's inappropriate? May I know what's the typical way to set the ground condition? (I am actually from aerospace engineering background...)

Added after 8 minutes:

I appreciate your advice. I am actually aware of the fundamental difference between those softwares. Little difference in result is unavoidable. However, the difference between my HFSS result and those of other softwares were just too large! And the strange thing was that, the Q-factor predicted by my HFSS result matched rather closely with other softwares' results. It's just that the inductance value increases too much near to the resonant frequency region, resulting in an extremely sharp peak. I therefore suspect I made some mistakes in the bc setting. I hope you can enlighten me on that. Thanks in advance.

Actually, both Sonnet and Momentum use the Method of Moments. Momentum uses formulation of MoM that assumes that substrates go to infinity in all directions, and uses an Integral Technique to solve the problem. Sonnet uses a formulation of MoM that assumes that the problem space is bounded by PEC walls, and because of this can use FFTs to compute the Green's function integral. But both are variations on the MoM technique.

--Max

Hi all,

Actually I met the same problem in HFSS when I simulate a stacked metal on-chip spiral inductor. The bottom of the silicon is set to PEC ground. Lump ports are used from ground ring (like the one in HFSS inductor tutorial) to inductor. But compare to measurement results, the inductance and Q factor are much larger in HFSS (20% at least). I tried to increase mesh density (by using solve inside, skin depth and increase pass numbers), however the results have really small changes.

All parameters are followed with the "reality" (silicon resistivity, different conductivity for different metal layers and so on). This mismatch problem has bothered me for days. And no good solutions yet...

my question: is the ground ring (PEC in simulation) can affect the inductance and Q of the inductor or not? if it does affect the results, how can I de-embedded it?

Thanks in advance!

Best Regards,
Kai

Check the definitions of the "inductance" you ploted. I think you are comparing apple to orange (the HFSS "inductance" is not the Sonnet or Momentum inductance).

Inductance and Q (especially Q) are very sensitive to small errors in S-parameters. So, try directly comparing the S-parameters from HFSS, Sonnet, and Momentum. All are basically correct programs and should give about the same answer. If there is a large difference in S-parameters, now you know you have set the problem up wrong in some way in one tool or the other. My impression is that setting ports is really complicated (there are so many different kinds, and many wrong ways to do them) in most non-Sonnet tools, so that is the first place I would look for problems. Be prepared to do some reading, and call support if you have any uncertainty.

Sonnet's niche (I work for Sonnet) is the very highest possible accuracy. To quantify Sonnet's error, just cut the cell size in half. The EM analysis error cuts about in half in nearly all cases. This allows you to quantify the EM analysis error in Sonnet. This technique allow us, to eliminate EM analysis error from the equation. We have, for example, then been able to zero in on, say, a measurement calibration problem for spiral inductors on silicon. See:

J. C. Rautio, and R. Groves, "A potentially significant on-wafer high-frequency measurement calibration error," IEEE Microwave Magazine, December 2005, pp. 94 - 100.

I will email a copy of the paper to anyone who can not download it from IEEE Xplore. I did the above work with IBM and we successfully eliminated a non-obvious, but major calibration error in their Si RFIC on-wafer measurement process. BTW, two other EM tools gave about the same result as the bad measurement, suggesting that those EM tool calibrations have a similar error.

I do know that all unshielded EM tools (i.e., analysis is in open space) must necessarily use approximate port calibration. They can actually do pretty good, but not perfect. Perfect is useful if you want to look at things like Q and inductance. Sonnet port calibration is perfect (to within numerical precision) as long as no port connecting lines are overmoded. In fact, we can even do tightly spaced groups of internal ports with perfect calibration. This lets you put tuning ports inside your inductor/filter/amplifier and tune it up without having to repeat lots of EM analysis. Watch for the December IEEE Microwave Magazine for a major paper on this topic.

So, you can get an answer with Sonnet (or free SonnetLite) and put error bars on the result. If the HFSS or Momentum results fall outside those error bars, then you know where to look for the problem.

Kai, have you found any good solution?

By the way, thank you for your comment, Rautio. I appreciate that.

Here is a link to an article comparing simulated and measured inductance and quality factor of a spiral inductor. They used IE3D.
http://www.gaasmantech.org/Digests/2...9.2abs2005.htm

The example of HFSS is wrong!

in fact HFSS also added the ground ring to in calculation.
then it added a pec line which will increase the inductance and increase the Q

solution:
deleting the ground ring and connect the indcutor port to outer, and using wave port(one side connect to buttom Pec ground)
and comparing the result again!

Wave port? How will you draw the integration line? I still doubt where the ground/PerfE sheet should be placed? In real case such like flip-chip applications, the ground sheet is not at the silicon substrate's side. The PCB ground is at the other side of the component.