Transient results as starting point in PSS

Prakash,

Both PSS and tran have a capability to "save" and "recover" the state at the end of (or arbitrary points throughout) a transient. There are parameters "saveperiod", "savetime", "saveclock", "savefile" and "recover" for this. These save the entire state and can be restarted reliably from where it left off. Unfortunately you can't use a save file from transient in PSS or vice versa - you can only use them in the same analysis.

The alternative is a less perfect solution which you might be able to get away with but has to be used with care. On the transient use "writefinal" to save a final nodeset/ic file. Then on the PSS use "readic" to point to that filename, and set skipdc=yes - but allow enough time for any imperfections to settle out. Since this doesn't capture quite everything, sometimes it needs some time to get over initial discontinuities caused by the missing information in the ic file.

Regards,

Andrew.

Hi Andrew, 

In ADE Assembler I am using transient results from "Writefinal" as a starting point for my PSS sims "readic". I specify a file location/name for witefinal and use the same location/name in redic and that seems to be working when I run nominal corner. When I try to run multiple corners I only get one writefinal file which correspond to the last transient corner that finished. Is there a way to save "Writefinal" per corner in transiant sim and use that data as a starting point for the corresponding PSS sim?

Best,

Hamed

Dear Hmjafari,

Hmjafari said:

When I try to run multiple corners I only get one writefinal file which correspond to the last transient corner that finished. Is there a way to save "Writefinal" per corner in transiant sim and use that data as a starting point for the corresponding PSS sim?

I am not Andrew, and will never hope or pretend to have the insight and knowledge he has! Nevertheless, have you tried using the methodology provided in the Cadence Sourcelink at URL:

https://support.cadence.com/apex/ArticleAttachmentPortal?id=a1Od0000000uDrGEAU&pageName=ArticleContent

for using a custom recover and restore file for snapshots? I assume this will work in Assembler, but can't say I have verified it. I have used it with a pss file in the former ADE-XL to save a re-run a pss analysis and has saved a lot of simulation time.

Shawn

Hi Hamed,

Regarding:  Is there a way to save "Writefinal" per corner in transiant sim and use that data as a starting point for the corresponding PSS sim?

I wanted to make sure that you realized this:  You cannot use transient results as in input to pss.  See my article:   Can I use a transient output as an input to pss or hb analysis instead of using tstab? 

You may want to look at this article:   Running multiple PAC (or PXF, Pstb, etc.) analyses from a single PSS analysis. 

Also see:  Spectre Circuit Simulator and Accelerated Parallel Simulator RF Analysis in ADE Explorer User Guide -- Single Input Large and Small-signal Analyses - Reuse Tab 

Best regards,

Tawna 

Thanks a lot Tawna, 

The reason I used transient sim "Writefinal" as initial condition for PSS is that transient  sim runs much faster compared to PSS for the same amount of tstab time setting. The pss results are used for pnoise and pxf sim.

I have a regulator + some low pass filter in my test-bench that force me to run pss tstab for ~2us. This take roughly 4-5 days on 4 core APS. Same 2us runs in few hours for transient sim (both use dynamic errpreset, liberal at start and conservative at the last 10% of sim time). 

In my test case, I first ran transient sim for 2us, then use the "writefinal" as the inital condition for pss. I set pss tstab for 100ns and it seems that pss is using that initial condition from transient sim and settling within 100ns. If this is a wrong flow as you described in your article, is there a way to speed up the pss tstab sim? That way I don't need to run transient sims at all. 

Thanks a lot for your helps. Much appreciated.

Best,

Hamed

Thanks a lot Sawn. 

Hi Hamad,

FIrst thing... always start a new thread rather than append a 7+ year old thread.    

Did you read the article that I mentioned?

Can I use a transient output as an input to pss or hb analysis instead of using tstab? 

I'm surprised (actually shocked) if you were able to use "regular" transient as an input for pss, as the formats are incompatible. Also, transient doesn't contain trending and history information (whereas tstab does).  Are you positively certain it wasn't the tstab (transient assist)? 

You can use dynamic parameters in the tstab interval, which you are already doing.    Use the tstab save/recover.   

For most efficient simulation of SpectreRF, I recommend this Article below.  It has the latest "best practices" that include things that'll help you speed up your simulation.

Getting the Most Out of Spectre® APS RF 19.1: Maximizing Performance 

In Spectre 20.1, we have SpectreX RF - that is another option to speed up performance.

If you have more detailed questions, I recommend filing a Case on Cadence Online Support https://support.cadence.com .

Best regards,

Tawna

Hi Tawna,

Tawna said:

I'm surprised (actually shocked) if you were able to use "regular" transient as an input for pss, as the formats are incompatible. Also, transient doesn't contain trending and history information (whereas tstab does).  Are you positively certain it wasn't the tstab (transient assist)? 

Maybe I am misinterpreting Hamad's recent post regarding his methodology, but I believe his pss tstab simulation only utilizes the node voltages of the final state of his conventional transient analysis as the .ic values (i.e., initial conditions) for the 100 ns TSTAB portion of the pss analysis. Hence, I think this should be expected to work. In essence, he is not attempting to recover the state file from a conventional transient analysis and use it as the starting state of the pss TSTAB analysis. The former only forces the initial voltages of the nodes in the TSTAB portion of the pss simulation to some value at the start of the simulation.

Is my understanding incorrect?

Hamad,

Hmjafari said:

This take roughly 4-5 days on 4 core APS. Same 2us runs in few hours for transient sim (both use dynamic errpreset, liberal at start and conservative at the last 10% of sim time). 

One other subtle point to consider when using a dynamic analysis to vary the errpreset if I may and it is not obvious to you. There is a trade-off between varying the errpreset dynamically and simulation time. When the errpreset is varied, it will appear as a transient step and will require some settling time governed by your circuit time constants. Therefore, each time you change the errpreset in time adds a time step that must settle. If the magnitude of the step and its time constant are significant, the use of the varying the errpreset may not be justified and the use of a single errpreset for the 100 ns tstab period may allow you to further reduce the tstab period to less than 100 ns.

Shawn

Good points, Shawn.  I certainly could have been misinterpreting what Hamed said.  Yet one more good reason to file a Case with Support...the AE can look at your spectre.out logfile, your settings, etc. and give a more informed answer once they understand the problem in more detail.    :-)    Also note... those of us answering questions on the community page often do so in our "spare time".  If you file a Case, you get a dedicated AE to research and answer your question.

Be careful when modifying errpreset during dynamic tstab.  If you set it too loose, you can have convergence issues.

best regards,

Tawna

Dear Tawna,

Wow! Thank you for even taking your valuable time to read my response - not to even mention your invaluable insights! Hopefully, Hmjafari will follow through with your recommendations and, possibly in addition, pursue a support case.

Shawn

Dear Tawna,

Wow! Thank you for even taking your valuable time to read my response - not to even mention your invaluable insights! Hopefully, Hmjafari will follow through with your recommendations and, possibly in addition, pursue a support case.

Shawn

Thanks a lot Tawna and Shawn, I really appreciate your helps with this. Yes your assumption is correct Shawn. I will file a case with the support as Tawna suggested. 

Best,

Hamed