Smart MonteCarlo

Discussion in 'Cadence' started by JD, Nov 1, 2007.

  1. JD

    JD Guest

    Hi,

    Just had a look at the new virtuoso simulator spec sheet. It says with
    the Advanced Monte Carlo algorithms, the Smart MonteCarlo analysis can
    speedup the MC analysis by 10x.

    Is this new feature available in Cadence 6? What excatly is the
    advanced monte carlo algorithms?

    Thanks,

    JD
     
    JD, Nov 1, 2007
    #1
  2. This is (almost certainly) referring to the "Latin Hypercube Sampling" method
    that is available (look on wikipedia for a short overview on what latin
    hypercube is all about).

    You can do this in both IC5141 and IC61X - provided you're using an MMSIM611 (I
    think) or later version of spectre with it. For IC5141 you'd need to be using a
    version from 2007 (support for this was added early this year). There's a choice
    on the monte-carlo form to use "lhs".

    Regards,

    Andrew.
     
    Andrew Beckett, Nov 2, 2007
    #2
  3. JD

    danmc Guest

    On this topic, when I try to use LHS with spectre-6.2.0.420, I get an
    error that says numBins is not a valid parameter for an instance of
    montecarlo. numBins gets set if I click the LHS button in the
    montecarlo setup form in analog artist.

    Also, I couldn't find any documentation at all in the spectre
    reference or user manual for 6.2.0.420.

    Do I have the wrong mmsim release?

    Something I also don't understand is the form lets you specify both
    the number of bins as well as the number of runs. I thought with LHS,
    you get 1 run per bin?

    Thanks
    -Dan
     
    danmc, Nov 2, 2007
    #3
  4. JD

    danmc Guest

    On this topic, when I try to use LHS with spectre-6.2.0.420, I get an
    sorry, it is a warning about numBins, not an error.
    I should have been more specific. I can't find documentation for
    numBins or a real description other than the spectre -h output for
    sampling=.
     
    danmc, Nov 2, 2007
    #4
  5. JD

    JD Guest

    Thanks very much for the reply. I really don't know Spectre montecarlo
    can do LHC. It is not on the spectreref and spectreuser manual.

    JD
     
    JD, Nov 2, 2007
    #5
  6. In solution 11365046 Cadence say that numBins should have been spelled
    numbins in the input.scs and that this is fixed in versions after
    5.10.41.500.5.94. Interesting to see that you have this in 6.2xxx

    You find one line of doc in spectre -h montecarlo if you run it on a
    new enough version
    You find the same line of doc in spectrePN.pdf (Whats new doc) in a
    new enough version.
    And you find a few more lines in the aatoolsuser.pdf in a new enough
    version.

    Remember that spectre is a part of mmsim now, and that the doc in the
    ic* doc dir not nescessarily is reflecting this. I think IC* has its
    own spectre, but that is not the newest one, but Andrew will probably
    correct me if I am wrong.
     
    Svenn Are Bjerkem, Nov 3, 2007
    #6
  7. It's pretty new - that's why...
    Documentation for features added during a release sometimes takes a while to
    catch up.

    Andrew.
     
    Andrew Beckett, Nov 15, 2007
    #7
Ask a Question

Want to reply to this thread or ask your own question?

You'll need to choose a username for the site, which only take a couple of moments (here). After that, you can post your question and our members will help you out.