Application Note: 60 GHz on-chip balun transformer: interleaved or overlay?

We have had quite a few topics here on RFIC EM modelling in the past, so I wanted to share an application note on 60 GHz RFIC balun design that I just finished.

In this document, two on-chip transformer layout styles with 2:1 turn ratio are compared. Target frequency is 60 GHz, nominal impedances are 200 Ohm primary and 50 Ohm secondary. Technology used for this study is IHP SG13S.

Part 1: Design and choice between interleaved and overlay configuration
http://muehlhaus.com/support/rfic-em...leaved-overlay

Part 2: Investigating the effect of dummy metal fill on balun performance
http://muehlhaus.com/support/rfic-em...-filler-effect

Please use this thread for comments or questions.

Mr. Muehlhaus
In order to catch an optimized value, How do you change the mechanical dimensions of a EM structure while it's been simulating ?
I'm really wondering this since few years.Is there any special script or something ?

Yes, this is based on scalable layout cells. I created a library of scalable inductor and transformer layout coded in ADS AEL scripting language. For such transformers with many layout options (variable number of primary and secondary), the code is a rather complex.

The basic method of creating scalable layout for ADS EM is described my appnote here:
http://muehlhaus.com/support/ads-app...es/ael-artwork

and here is an example with code:
http://muehlhaus.com/support/ads-app...urized-bpf-ael

Independent of my work, Keysight have also introduced scalable inductor & transformer layout feature in ADS 2017 (Coilsys) that can be enabled from Tools > App Manager. Coilsys layouts are different from my layouts, we have a bit different routing strategies for the transformers.

All I would like to know how the dimensions of a EM structure are changed during simulation like in Sonnet ?If the cell is parameterized, how a parameter (or parameters) is changed during simulation in order to find best approximated/exact value.Let say we have an inductor with a parameter N ( winding number) and we have to optimize this inductor.
How we arrange this parameter N that varies between let say 1 to 10 to observe the variation.

I have been using Sonnet for many years, but would not know how to build this scalable artwork workflow efficiently with an "external" solver like Sonnet.

My automated workflow is based on my scalable ADS layout cells with ADS Momentum, where the emModel block can be inserted into schematic parameter sweeps and the layout is scaled accordingly. So this is a "seemless" workflow where the EM solver can work directly on the scalable layout (no file based data exchange required) and all that is controlled from schematic level.

Initial layout parameters for a given target inductance are calculated by closed form equations (not using the optimizer), with different layout candidates that all have the required inductance at DC. That is a key component in this workflow: before going to EM, the my code creates a list of possible implementations (geometry parameters) for this inductance value. Only those inductors are EM simulated that have the required indutance (according to DC calculation).

A batch simulation of EM models with these parameters is done and evaluated. For inductors, the batch simulation is then evaluated for best Q factor, and inductor geometry parameters are fine tuned for best Q based on interpolation/extrapolation from EM batch simulation results. That fine tune also corrects for frequency dependence of L from parasitics, again with closed form math instead of using the ADS optimizer.

If you are interested in the workflow for inductors, here is a 12 minute video demo:
https://www.youtube.com/watch?time_c...&v=m1ALwB42nEU

I have done a parametric simulation in Sonnet and it's very straightforward by adding a variable.

Yes, I know, I have been selling and supporting Sonnet for many years. It gets complicated if you try to make a parametric octagon spiral, for fixed number of turns. Doing it with variable number of turns seems impossible to me. That's why I switched to ADS for this task: much easier to do this scalable artwork (custom coded artwork) stuff, and results are very similar, sometimes even more accurate because Momentum models the sidewall currents more accurately.