Significant discrepancies between the s-parameters obtained from Momentum and Sonnet for a spiral inductor

[Hilbertstrauss]

So I basically designed this inductor ( 2 turns, inner radius=60u, spacing=4u, width=10u) in both sonnet and momentum. The metal layer M9 has a thickness of 3.4u. the under bridge M8 has a thickness of 0.9u. I also manually designed the stack up in both softwares.

I set the maximum cell size in momentum ( RF mode) to 5u (metal and via both are set to 3d model). The cell size in Sonnet is 2.5u (thick conductor model with 2 sheets is used for M8 and M9).

For momentum, I used 2 TML ports with calibration lengths of 85.96u. In sonnet, I used 2 box wall ports also with a calibration length of 85.96u. I used 4 vias for each of the two connections of the underbridge.

**broken link removed**

I run a single frequency simulation at 2.4G in both sonnet and momentum.

The results obtained from momentum differ significantly from those obtained with sonnet. Both are in dB.

For sonnet:

For momentum:

It's been many days now and I still can't find the cause of this inconsistency. At first I thought it could be related to the ground plane used in momentum, so I designed a custom ground plane under the inductor and assigned the ports as it should, this didn't change the results by much (no more than about ~10%).

Any help would be more than appreciated. (The sonnet project file and the momentum workspace project are both attached under).

 

[volker@muehlhaus]

Port concepts are different in these tools. From many years of RFIC inductor simulation experience with both tools: In Momentum, do NOT use port calibration and reference shift. From my experience, Momentum port cal does not work well over silicon, and is not needed. Instead, just place the pins at the inductor and switch OFF port calibration (Feed Type: Direct). Leave the port ground reference as global ground at the bottom of the silicon - that sounds strange for Sonnet users, but works fine in Momentum.

In your Momentum stackup, via8 looks wrong because it does not connect to metal8. It should go down one more layer. At least in previous releases, Momentum thick metal needed to be thinner than the dielectric above them. Not sure if that was changed, and M8 can now go up all the full 0.91µm to the layer above?

Also check your stackup by a few simple testcases:
- resistance of a line on each metal layer
- shunt capacitance of a metal plate M8 to M9 and M8 to substrate

Evaluate your results for series L and series R, so that you can plot data over frequency and see if DC resistance is off or it is an RF effect that causes trouble. Single frequency S-Params are not useful for troubleshooting.

Adding a ground plane under the inductor will change fields a lot, so that is not a useful.

Good luck!
Volker