Leander shalom,
My name is Itai and I am work for Ansoft's rep in Israel, and fammiliar with Gosha's work on this subject.
Instead of a "one-size-fits-all" answer, please consider the following 6.5 points:
1. What percent of your custom inductor designs will use symmetric/differential spirals (that do not require a ring retrun path at all) ?
2. Assuming you do not have the layout yet and you're using a non-symmetric spiral, you'll need the ring back.
Start the smallest possible ring (like in the UMC example), and take a typical non-symmetric inductor that you would be using (the same number of turns/typical size). Next, replace your inductor with just one straight line. Compare the results with the inductor and without the inductor.
How much L (in percent) has the ring added to the original model ?
3. Look at the layout of the specific design that you intend to use the non-symmetric spiral. Can you identify the "real" retrun path ?
For example, let's assume that you have GND on one side of the inductor, VCC on the other side, and a bypass capacitor between them. Import the inductor layout with the GND,VCC nets and capacitor geometry. If you do not want to simulate the capacitor, just fill its airgap with metal. First, you'll be using the "real" return path in the simulation. Second, the non-ideal EM field will take the capacitor plate into account. The inductor results will include this return path effect.
In a benchmark, here in Israel, a customer has achieved a 2% accuracy of the VCO frequency shift, after extracting the VCO layout in HFSS. The circuit simulation and lab measurements were done by the customer ! With HFSS what you put in is what you get out, no need to "tweak" it.
4. Had you chosen another simulator over HFSS,
Would it be able to correctly simulate the return path through the silicon. In high frequencies some of the return current flows in the silicon, since it's the shortest way to go.
Would it be able to correctly calculate the losses inside the metal?
Unless it is a quasi-static extractor it has a return path (there has to be one, even if it's implicit). Are the simulator assumptions correct (or does it assume that the return path is 300um under the silicon were no current has ever gone before).
Does it simulate correctly above the inducor resonance frequency (3rd harmonic) ?
5. Have you consulted your local Ansoft AE? I can tell you that Ansoft is very proud of is their worldwide technical support engineers. When you buy Ansoft, you also buy their technical support. "Benchmark" them too. If your local AE does not have enough expirience with exporting layout from your favorite layout tool into HFSS, ask him to drop me an email to
itaifrenkel@yahoo.com
6.5. Have a great day.
Itai