Finding and transferring the value of a variable that gives the minimum value of an output in a test where more than one variable swept to another test using calcVal function in Assembler

Dear Can,

If I understand your question correctly, I think the use of the ?matchParams argument of the calcVal() function is what is needed. When ?matchParams is set to "all", calcVal() will look at all of the sweep points in  test_1. Have you tried that? In other words:

var_2 = calcVal("output_2" "test_1" ?matchParams "all")

The default value for ?matchParams is "none".

The documentation for calcVal(), including an explanation of some common errors, is at the On-line support site at URL:

https://support.cadence.com/apex/techpubDocViewerPage?xmlName=maeskillref.xml&title=Virtuoso%20ADE%20SKILL%20Reference%20--%20Output%20Functions%20-%20Troubleshooting%20calcVal&hash=pgfId-1116037&c_version=ICADVM20.1&path=maeSKILLref/maeSKILLrefICADVM20.1/measures.html#pgfId-1116037

I do not have access to the Cadence tools to verify this currently, so I cannot be sure this will accomplish your objective, but it came to mind and wanted to at least suggest it Can.

Shawn

Dear Shawn,

Many thanks for your answer. I tried to use ?matchParams "all" argument after your message but it didn't solve the issue.

Can

Dear baltaci,

baltaci said:

tried to use ?matchParams "all" argument after your message but it didn't solve the issue.

Sorry Can!! I thank you for trying, but am sorry to have wasted your time!

Well, I re-read your Forum post again and I think I may have overlooked an important fact.

baltaci said:

I can find the var_1 value that results in the minimum output_1 value by evaluating xmin(output_1) expression in the calcualtor and I get the correct var_1 value. However, when I assign the same expression to an output, I get "eval err" because this output is calculated for each swept var_1 value instead of taking the full range (0:1:7) into account.

I think the problem is that you have defined an output that is a measurement across all corners and not a value computed from one corner. There is a relatively new function measureAcross() that allows you to define an output that represents a measurement using all corners of a simulation set.

I think the syntax you want to use is something like:

For test_1; output_2 =  measureAcross( xmin(output_1) parameters list("temperature")) or

output_2 =  measureAcross( xmin(output_1 ?overall t) parameters list("temperature"))

 You can add this as an output in Assembler as detailed in the reference manual at URL:

support.cadence.com/.../techpubDocViewerPage

The function definition for measureAcross() is at URL:

https://support.cadence.com/apex/techpubDocViewerPage?path=maeSKILLref/maeSKILLrefICADVM20.1/measures.html#measureAcross

if you clear the error with your output_2 output, I think you will be able to pass it to test_2.

Shawn

Hi Shawn,

Many thanks for your answer. In fact, using measureAcross() works. I can build the test set-up correctly and transfer the output of one test to the variable of another test. However, I have a new issue now. If I run Monte Carlo mismatch analysis with the correct set-up, the second test takes very long, more than four days (still ongoing in fact)! Actually, when I look at the log files of each MC run, the simulation time is reported as 10 seconds. I have 400 MC runs in the second simulation: 400 x 10 sec = 66 minutes. If I run the second test standalone (i.e. with a pre defined var_2), it finishes in a reasonable time. However, if it is run with var_2 set by measureAcross() and calcVal(), the the time of the second test is very very long. Do you have any idea why? It seems that if a variable set using measureAcross() and the second test is run usng calcVal() in MC mismatch analysis, the algorithm spends a very large time somewhere between the MC iterations (in test_2).

To check if this issue is specific to my design or not, I created a very simple schematic and maestro view where I have one vdc and one res from analoglib. I observe a very similar behaviour.

What would you suggest regarding this issue?

Many thanks in advance.

Best regards,
Can

Dear Can,

baltaci said:

In fact, using measureAcross() works. I can build the test set-up correctly and transfer the output of one test to the variable of another test.

Excellent! Thank you for letting us know! Once again, sorry for my first response - I shold have read your Forum post more carefully before I responded.

baltaci said:

It seems that if a variable set using measureAcross() and the second test is run usng calcVal() in MC mismatch analysis, the algorithm spends a very large time somewhere between the MC iterations (in test_2).

baltaci said:

What would you suggest regarding this issue?

I do not have visibility into the Cadence CCRs and therefore do not know if this is a known issue or if there exists a workaround. Others who monitor this forum do aand may have a comment with respect this potential issue.

However, your thoughtful test case and its result would serve as an excelllent resource to provide Cadence if you were to submit an SR by contacting customer support. One item I might suggest is to try the example that Cadence provides as a tar file in the support article detailing the use of measureAcross() with a Monte Carlo run at URL:

https://support.cadence.com/apex/ArticleAttachmentPortal?id=a1O3w000009fpasEAA&pageName=ArticleContent

It would be worth checking to see if its behavior is similar.

Shawn

I wouldn't expect this behaviour - you must be using something recent if you're using measureAcross, so I would follow Shawn's suggestion and log a case via http://support.cadence.com

Probably we'd need to get some insight into your setup and see log files, and so on to understand what's going on.

Andrew

Hi Shawn, Andrew, many thanks for sharing the link. I used the recommended set-up of the link. The simulation works the same as what I had before (i.e. it is the same correct set-up). However, the simulation time is still very long. I will proceed with creating a case in Cadence Support.
Can

Dear Can,

baltaci said:

I used the recommended set-up of the link. The simulation works the same as what I had before (i.e. it is the same correct set-up). However, the simulation time is still very long. I will proceed with creating a case in Cadence Support.

Thank you for the added information! Contacting Customer Support sounds like a great idea. Good luck Can.

Shawn

Dear Can,

baltaci said:

I used the recommended set-up of the link. The simulation works the same as what I had before (i.e. it is the same correct set-up). However, the simulation time is still very long. I will proceed with creating a case in Cadence Support.

Thank you for the added information! Contacting Customer Support sounds like a great idea. Good luck Can.

Shawn