Learning Forum

[wen bo]

How can extracting S-parameters for a large number of ports in SIWAVE improve its causality?

In addition to selecting the enforce causality option, are there any other points to note and methods available?

[Mrinmoy Bharadwaj]

Besides the "Enforce Causality" option in SYZ set-up, there is another "Enforce Causality" option while exporting a macromodel from the SYZ results. To know more about that, kindly refer to the help document of the tool under section heading "Exporting Macro Model".

[Dan Dvorscak]

Its not directly related to causality, but numerics can play a part in it, but you've got way more points in your decade sweep than I would normally use. Also, I would stop the decade sweep at 10MHz.

My usuall set up is two sweep. 

• A 10MHz linear step sweep starting at 0 Hz (so you don't need a separate line just for the DC point) going up to the max frequency.
• Then a 10 points per decade sweep from 100 Hz (Though that depends on the application. If it is a PDN, then I might start at 1 Hz to be safe.) to 10MHz. 

One more thing to double check, if you are using external s-parameter models for components, you should also run a causality test on those.

[aalmeida]

Hi Dan, 

I'm also running into a similar issue where I cannot extract causal PDN Z-parameter models from my package or board design.

If possibe, could you please post a screenshot of your recommended "Compute SYZ-parameters" setup window for the following senario?

PDN Z-parameter extraction up to 10GHz

I reverted all of my capacitors models to C / ESR / ESL approximations to keep things simple. 

Besides the frequency setup range, I am also curious what you recommend for the other setup options in this window.  For reference, I have tried about a dozen different combinations of settings and none of my results can pass 2.5% causality tolerance.

[Dan Dvorscak]

Hello Alexaner,

Ansys employees are not allowd to attach images on this forum, sorry. Similar to the above, this is what I would recommend as the starting point for a PDN:

• A 10MHz linear step sweep starting at 0 Hz (so you don't need a separate line just for the DC point) going up to the max frequency.
• Then a 10 points per decade sweep from 100 Hz (Though that depends on the application. If it is a PDN, then I might start at 1 Hz to be safe.) to 10MHz. 

Additionaly, do you have the compute exact DC point button selected? That can sometimes incur causality issues. More so in older versions. On that note, we do recommend using the latest release available. Are you on 2024R1, or something older? 

Best regards,

-Dan

[aalmeida]

Hi Dan,

Thanks for these pointers!  I was able to extract causal Z-parameters from both package and board designs using your suggestions.  For referenece, I am running Ansys 2023R2, and unselected compute exact DC point.

For my own educational benefit, what is your reasoning behind these sweep recommendations?  You double sweep select frequency regions, is that what helps to generate causal results?

Best,

Alex

[Dan Dvorscak]

Hi Alex,

The short version is that shap discontinuities in the s-parameters appear as being numerically non-causal. The point of the two tiered sweep, log at low frequency and linear at high frequency, is the best way we've found to capture the resonances withouth grossly oversampling. PDN effects are best captured by the log sweep and signaling (and other strutures interesting at higher freqeuncies) by the linear sweep.

If we try to do just one sweep, say log, you would have to add 100s of points per decade to not undersample the higher frequencies, and that would add way, way more points you need a low frequencies. 

This sampling is about more than just causality but a general best practice. So it does more than just help preserve the causaliyt in the model. 

[aalmeida]

I can't thank you enough, this is extreamly valuable information to help inform my future simulations.  Much appreciated!

Best,

Alex