Working around lack of sweeps in MC runs

Hi Michele,

The main reason why you can't do variable/parameter sweeps at the same time as Monte Carlo is because Monte Carlo sweeps are in the same direction as the variable sweeps - you have corners in one direction, and variables/parameters/monte in the other - and this also affects how the results are displayed. There are a number of requests to support this - and I'm sure it will happen one day.

Anyway, part of the reason why it's not too bad is that there already is a way of doing this. You can sweep a variable as part of a corner. Corners do not have to be single values of variables (or model file sections). Within a corner definition, you can say you want DesVar2=1,2,3,4,5 or DesVar=1:1:5 (1 to 5 in steps of 1), or use the GUI to define more complex sweeps, within log steps, or with inclusion lists, center span, etc. The corner will then contain a set of "sub corners" - indicated by a number at the bottom of the corner definition showing the total number of sub-corners. 

Monte Carlo will then be run for each of these sub-corners, hence incorporating your variable sweeps.

This is much simpler than the ideas you were proposing.

Regards,

Andrew. 

Hi Andrew,

this is like a million times easier and more elegant! Many thanks for the tip, I did not know of this feature of the corners.

Now, I just have to understand why the MC run did not preserve my modulated freq out of the ENV simulation :(

I had saved data for family plots and all the rest.

I have a feeling that this is related to APS, which I used to speed up the runs. In a previous adventure, I struggled quite a bit with saving transient OPs from a PSS. I needed the gm Vs. time for a mixer, and the steady state (time) result came quite handy. I remembered being able to save and display it, by using the save commands in an attached file...Then, all of a sudden, nothing helped and the gm data disappeared from the saved results. As it turned out, APS was not saving those curves no matter what. If I used the normal Spectre, everything worked fine.

Let's see....And thanks!!

Michele 

Hi Michele,

Which version of spectre are you using? I'll do some checks - if this is not being saved, then it's probably a bug that should be reported via customer support.

Cheers,

Andrew. 

Hi Andrew,

just switched to MMSIM13.10.127

And I checked, APS or Spectre makes no difference. Modulated Frequency is just not there.

I look forward to hearing about your checks.

Another thing: I ran a MC for a single point overnight, saving data to allow family plots. 200 points of an ENV simulation. The whole simulation folder weights some 500MB. I asked to plot a current Vs. time this morning around 8h10...I am still waiting...

What can cause this? And more importantly, how can I exit this state without killing/restarting? A simple top command shows that the virtuoso process isn't doing anything...

Cheers,

Michele 

You might also want to take a quick look at Cadence Solution 11850044 (see also http://www.designers-guide.org/Forum/YaBB.pl?num=1391751554).

The problem that I see is that I don't get the "time" output from envelope (envlp_td) if I do monte carlo - I only get harmonic time and spectrum (both of which are from the envlp_fd results). This is a bug - I've not seen it reported before though.

I'm not sure what you mean by "Modulated Frequency" as an output of envlp though.

Frank - I'm not convinced that you post or the solution that you've mentioned is relevant here. That's more to do with .measure or MDL outputs - which isn't the case here...

As for the performance, I'm not sure why that would be. Unfortunately the envlp_fd results are still saved as PSF, and if you can cut down the number of signals saved, that would help. The "Time" output can be saved as either PSF or SST2 (not PSF XL yet, unfortunately) - do you know which it is? If you use the results browser, plot a signal from the envlp_td in ViVA, and then use the Trace Info assistant whilst selecting the trace you just plotted, it should tell you the output format.

Regards,

Andrew.

Thanks Frank! Interesting since I do have some HDL in my test bench.

Anyhow, it looks like the issue is in the retrieval of waveforms from ADEXL GUI. A simple right-click and plot yields the following:

WARNING (OCN-6038): Results envlp are not available for /myPath/simulation/myCell/adexl_MC/results/data/MonteCarlo.0/psf/myCell. Use results() for a list of available results.

I overlooked this all important message in the beginning and thought the system was stuck, but actually it was still working (although not updating the progress of another sim I was running, so that's why I thought system was stuck).

However, by using the Results Browser I can access all results and also the much needed demodulated frequency. Only thing, it is contained in a directory called:  

Andrew,

there is an output called like this among the ENV offer. I am hopefully attaching a snapshot of what I mean.

BTW, is there a way to have images displayed inline without resorting to some img hosting? I mean, upload them and reference them from within the post editor?

Cheers,

Michele

Andrew Beckett said:

Frank - I'm not convinced that you post or the solution that you've mentioned is relevant here. That's more to do with .measure or MDL outputs - which isn't the case here...

The problem described in Cadence Solution 11850044 will always occur when something is written to a file (not only measurements, but also nodesets etc.). This is something I do very often in my simulations and I thought that Michele might run into similar problems sooner or later.

MicheleA said:

BTW, is there a way to have images displayed inline without resorting to some img hosting? I mean, upload them and reference them from within the post editor?

You can attach a single image to each post via the Options tab, but apart from that, you can't reference pictures from within the post unless they are hosted elsewhere, unfortunately. The "Content Selector" button in theory should do this, but I don't think you can upload your own pictures that way.

Andrew.