The Designer's Guide Community Forum

The Designer's Guide Community Forum
https://designers-guide.org/forum/YaBB.pl
Simulators >> Circuit Simulators >> Sequencial MonteCarlo Analyses using previous analysis results
https://designers-guide.org/forum/YaBB.pl?num=1276348343

Message started by pancho_hideboo on Jun 12^th, 2010, 6:12am

Title: Sequencial MonteCarlo Analyses using previous analysis results
Post by pancho_hideboo on Jun 12^th, 2010, 6:12am

For improving yields againt process variations, I use laser fuse trimmings for some register's settings.
I would like to emulate this operation and confirm yields after trimmings using Monte Carlo Analysis.

Assume trimmings are done according to Vth value of MOS FET.
This is actually done by DC analysis under Monte Carlo Trials.

After trimmings, I would like to evaluate dynamic characteristics such as Gain or Output Power Level by using AC analysis or HB analysis under Monte Carlo Trials.

How can I do these analysis using Monte Carlo trials ?

For Corner Analysis, I can easily do AC analysis or HB analysis with trimming register's values as design variables.
But I would like to confirm yield's improvemnet by trimmings under more random process variations.
Here trimmings are also random.

I can accept any simulator engine and any post processing environment.

Title: Re: Sequencial MonteCarlo Analyses using previous analysis results
Post by sheldon on Jun 13^th, 2010, 5:14am

Pancho,

This capability is supported in IC614, contact your local
support for details if you are using Cadence ADE.

Best Regards,

Sheldon

Title: Re: Sequencial MonteCarlo Analyses using previous analysis results
Post by pancho_hideboo on Jun 13^th, 2010, 5:34am

sheldon wrote on Jun 13^th, 2010, 5:14am:

This capability is supported in IC614

Do you understand my need correctly ?
My need is like following:

for(MC_Trial=1; MC_Trial<=1000; ++MC_Trial) {
// MC_Trial=n
DC Analysis in random variables set for "MC_Trial=n"
Tran Analysis in same random variables set as DC Analysis
and design prameter value's set determined by DC Analysis.
}

I'm now trying my need by using Golden Standard HSPICE.
We can not use full ability of HSPICE if we use it in Cadence ADE.

Cadence Spectre's netlist for MonteCarlo Analysis is like following.

Quote:

mc2 montecarlo donominal=no variations=mismatch seed=1234 numruns=200 {
dcop2 dc
tran2 tran start=0 stop=1u
export slewrate=oceanEval("slewRate(v(\"vout\"),10n,t,30n,t,10,90 )")
}

I don't think there is a special ability in Cadence Spectre for my need.

If task of my need is possible to resolve without adding special ability to simulator engine,
it can be possible by any simulator with Ocean Scripts or Unix Shell Scripts.

I will survey Cadence IC6.1.4.

Title: Re: Sequencial MonteCarlo Analyses using previous analysis results
Post by pancho_hideboo on Jun 13^th, 2010, 7:37am

sheldon wrote on Jun 13^th, 2010, 5:14am:

This capability is supported in IC614

I can find out nothing for resolving my needs in any of followings.
- Virtuoso Analog Design Environment GXL: New Features in IC 6.1.4
- Virtuoso Analog Design Environment L : New Features in IC 6.1.4
- Virtuoso Analog Design Environment XL : New Features in IC 6.1.4

If you mean following, it is useless for my need.

Quote:

Virtuoso ADE GXL: What's New
Product Version 6.1.4
November 2009
New Features in IC 6.1.4
Following new analyses were introduced in IC 6.1.4:
Following enhancements were made in Sensitivity analysis:
Support for sensitivity analysis with statistical variables and DC operating point parameters as input parameters.
Improved sensitivity analysis results display with support for filtering results based on search criteria,
separate tabs for visualizing data based on correlation of specifications vs. specifications,
specifications vs. statistical parameters, and so on.

Title: Re: Sequencial MonteCarlo Analyses using previous analysis results
Post by pancho_hideboo on Jun 13^th, 2010, 7:45am

I can make simulation methodology for resolving my need now.

I do mutiple monte carlo trials without using mutiple trials run provided for Monte Carlo ability.
Instead, I do multiple monte carlo trials by Ocean Scripts or Unix Shell Scripts using single monte carlo trial with fixed seed and incrementation of "Starting Run #" for random variables.

Here I use some Verilog-A modules for dumping design parameter's values as results of process variation judgement in DC analysis.
This task of setting design variables according to DC Analysis results is also possible without using Verilog-A module.

I have to merge many single trial monte carlo files to one file.
If monte carlo trial number is 1000, 1000 files are created.
But I don't do that, instead I add result to one file after each trial, so it does not create many files.
It is easy to do this if we use ascii text data.

I can apply this methodology for any simulator such as Synopsys HSPICE, Agilent RFDE(ADSsim), Agilent GoldenGate and Cadence Spectre, etc.

Agilent GoldenGate provide an utility command, "mc_merge" for Running Monte Carlo Trials in Parallel.
If I use this, I can merge results as binary data very easily.

Title: Re: Sequencial MonteCarlo Analyses using previous analysis results
Post by sheldon on Jun 14^th, 2010, 6:14am

Pancho,

Been doing this type of simulation for a while. In the past, cdsSpice
had some nice features to support this type of simulation. Now, if I wanted to use scripting, I would use SpectreMDL. There was a SpectreMDL presentation at CDNLive! San Jose a couple of years ago
provided an example. If you want to do it in the environment ,
contact your local support for help with tools.

Best Regards,

Sheldon

Title: Re: Sequencial MonteCarlo Analyses using previous analysis results
Post by pancho_hideboo on Jun 14^th, 2010, 6:19am

sheldon wrote on Jun 14^th, 2010, 6:14am:

In the past, cdsSpice had some nice features to support this type of simulation.
Now, if I wanted to use scripting, I would use SpectreMDL.
There was a SpectreMDL presentation at CDNLive! San Jose a couple of years ago provided an example.
If you want to do it in the environment, contact your local support for help with tools.

No, I don't need it at all.
If I use Synopsys HSPICE, I can do more superior tasks than "cdsSpice" or "Specte + SpectreMDL" can do.

I can create more versatile scripts which are not restricted to specific vendor's simulator by myself.

Title: Re: Sequencial MonteCarlo Analyses using previous analysis results
Post by sheldon on Jun 15^th, 2010, 2:15am

Pancho,

You missed the point, but that can't be helped.

Best Regards,

Sheldon

Title: Re: Sequencial MonteCarlo Analyses using previous analysis results
Post by pancho_hideboo on Jun 15^th, 2010, 9:19am

sheldon wrote on Jun 15^th, 2010, 2:15am:

You missed the point

What did I miss ?
I can't miss any point since there can not be any interest or novel point.

sheldon wrote on Jun 14^th, 2010, 6:14am:

contact your local support for help with tools.

I don't need any support which is no more than tool usage and I can easily understand if I read documents.
I don't think our local Cadence support can provide something more than easy tool usage which I can easily find out only if I read documents.

Again If I use Synopsys HSPICE, I can do more superior tasks than "cdsSpice" or "Spectre + SpectreMDL" can do.
List up tasks which can not be possible by HSPICE but possible by "Spectre + SpectreMDL", if any.

Again I can create more versatile scripts which are not restricted to specific vendor's simulator by myself without such valueless support.

Previously you sent me a mail that "VPCD was remarkably improved, so contact local support",
after http://www.designers-guide.org/Forum/YaBB.pl?num=1169632704/8#8.

But I could not find out any remarkable improvement in VPCD(Virtuoso Passive Component Designer) at all.
http://www.edaboard.com/viewtopic.php?t=335639

At last, VPCD has been EOL.
http://www.designers-guide.org/Forum/YaBB.pl?num=1244552575

Title: Re: Sequencial MonteCarlo Analyses using previous analysis results
Post by August West on Jun 15^th, 2010, 6:45pm

My experience is that the time-step selection in HSpice is weak, and you have to make up for it by setting delmax small, and as a result the simulations are considerably slower for the same accuracy. This is particularly true if you are interested in currents.

For me the poor accuracy is a big deal, and as such I would never consider HSpice my 'golden' simulator.

Also, SpectreMDL is more sophisticated in its scripting, and gets better results much faster than HSpice. At least, that has been my experience.

Spectre and SpectreMDL just seem to work better.

Title: Re: Sequencial MonteCarlo Analyses using previous analysis results
Post by Andrew Beckett on Jan 2^nd, 2011, 10:14am

pancho_hideboo wrote on Jun 13^th, 2010, 7:37am:

sheldon wrote on Jun 13^th, 2010, 5:14am:

This capability is supported in IC614

The reason why it is not in the documentation is because this is an engineering release in IC614, and we wanted to work with customers to get their valuable feedback before releasing it to everyone. That's why Sheldon recommended contacting your local Cadence office.

It's controlled via a cdsenv variable, and the capability is fully released in the upcoming IC615 release (and is then documented).

It allows you to do exactly this:

Quote:

for(MC_Trial=1; MC_Trial<=1000; ++MC_Trial) {
// MC_Trial=n
DC Analysis in random variables set for "MC_Trial=n"
Tran Analysis in same random variables set as DC Analysis
and design prameter value's set determined by DC Analysis.
}

In fact you can have complex calibration steps at each point in the monte-carlo, for example doing a binary chop calibration run involving multiple simulations. I've presented this at a Technology on Tour, and it was also presented at a recent CDN Live! event in 2010, I believe.

Regards,

Andrew.