How to extract S-parameters without EM field solvers?

Hello Everyone,

I would like to extract S-parameters of a handful of traces routed on a Silicon Interposer (permittivity ~4). Traces length will be less than 8mm and the signal frequency is 28GHz. The goal is to simulate signal integrity among other things.

I do not own an EM Field Solver or any RC extractors (xRC, Quantus and so on) because of a very limited budget. So I was exploring some free tool options (fast henry, fast cap) and hoped to extract a cascaded RLCk network for each traces. Then my plan is to do a manual conversion (hand calculation) to get cascaded S-parameters.

This whole process seems fairly laborious and I also question the reliability of the free tools I mentioned above.

I would like to hear any recommendations/suggestions or even critics that could help me get where I want !

Thanks !

If you don't have more than 4-ports, you can download Sonnet Lite to simulate those structures.

BigBoss said:

If you don't have more than 4-ports, you can download Sonnet Lite to simulate those structures.

Click to expand...

I will look into that, however our circuit will definitively have more than 4-ports.

parps said:

I would like to hear any recommendations/suggestions or even critics that could help me get where I want !

Click to expand...

If you want reliable results, you need tools + expertise .
I am doing a lot of such EM extraction work as an external consultant (EM expert).

Your company is doing RFIC design in 28GHz frequency range and you don't have access to a decent EM simulator !?
Should know, they can design the best individual RF dies in the world, but if the dies are stacked and have poor connection between them, everything is compromised.
As was mention above, "better call an expert"

Hi vfone and Volker!

It's a fair question. My company has never done design beyond 8GHz (16Gbps), and everytime the design was confined within the IC and the distances involved were negligible compared to our signal wavelength so that we didn't need EM FS. The routing outside the IC were done by the package company who provided us the S-param data.

My company also do 2.5D/3D integration and it is the first time we have been asked to do routing at that frequency. I am well aware how this assembly work can compromise our customer's design, however because of very tight budget our options are limited.

Given the answers I read, it sounds like I can't go around getting an EM FS.. Can you provide a cost range for tools like EMX, ADS momentum or HFSS so I can get a better sense of the budget required for this project? Are there lower cost tool options that i didn't mention?
I appreciate your feedback,

parps

parps said:

Given the answers I read, it sounds like I can't go around getting an EM FS.. Can you provide a cost range for tools like EMX, ADS momentum or HFSS so I can get a better sense of the budget required for this project? Are there lower cost tool options that i didn't mention?

Click to expand...

Purchase of an EM solver for this type of work is somewhere in the range $25k to $100k for a perpetual license.
A full 3D solver that you didn't mention is Empire XPU, price is at the lower end of that range.
https://muehlhaus.com/products/empire-3d-em

If you don't need the solver on a regular basis, I can also extract your model as a service.

To my understanding you and your company design High Speed PCB Boards for your customers, right ??
If it's so, you don't need a real 3D EM simulator because your signals are always digital and you just need some timing properties of the signals such as delay,overshoot,undershoot etc.
In order to simulate such kind of properties, you need a Signal Integrity package from either Cadence ( Sigrity) or Mentor Graphics ( Hyperlinks) or any kind of similar software package.
They are relatively cheaper and some of them have "pay as you go" feature so you don't have to pay since you don't use them
Think about them..

parps said:

the distances involved were negligible compared to our signal wavelength so that we didn't need EM FS.

Click to expand...

Ohh, not again. This is like in Selena Gomez song "The Same Old Love": "I'm so sick of that same old love, my body's had enough"
For tens of years now, analog designers keep saying that short length traces don't need attention (read impedance control).
Unfortunately (for the same period of time), we, the RF guys, say the opposite.

vfone said:

Ohh, not again. This is like in Selena Gomez song "The Same Old Love": "I'm so sick of that same old love, my body's had enough"
For tens of years now, analog designers keep saying that short length traces don't need attention (read impedance control).
Unfortunately (for the same period of time), we, the RF guys, say the opposite.

Click to expand...

Interesting, I had no ideas RF guys felt so strong about this ;-). How do you draw the limits then? I often thought that distance lower than 1/20th the wavelength were safe enough :grin:

Hi Vfone,
It was not a facetious question, as an analog designer I only considered RC or RLCk models (depending upon frequency) for almost all of my designs using static extractor tools (xRC, xACT, Quantus and Quantus FS). I always neglected propagation effect for distances below 1/20th to 1/12th the wavelength, I assumed/was told it was a good design practice.

However, I was genuinely surprised by your previous statement and would be interested to hear your opinion for when you would start considering propagation effect? Any other feedback/critics are welcomed !

parps, I was also surprised to read that statement.

I agree on your 1/20 wavelength approximation. This is when transmission line effects become visible. Independent of transmission line effects, even short interconnect can cause trouble by adding series inductance (narrow strip) or shunt capacitance (wide strip, stub). It depends on circuit impedance level how much of those parasitics you can tolerate.