Home > Community > Forums > Custom IC Design > monte-carlo simulation running one seed

Email

* Required Fields

Recipients email * (separate multiple addresses with commas)

Your name *

Your email *

Message *

Contact Us

* Required Fields
First Name *

Last Name *

Email *

Company / Institution *

Comments: *

 monte-carlo simulation running one seed 

Last post Tue, Jun 18 2013 12:44 AM by Frank Wiedmann. 11 replies.
Started by frogconsultant 09 May 2012 01:35 AM. Topic has 11 replies and 3310 views
Page 1 of 1 (12 items)
Sort Posts:
  • Wed, May 9 2012 1:35 AM

    • frogconsultant
    • Top 200 Contributor
    • Joined on Thu, Feb 18 2010
    • Bourron Marlotte, Seine-et-Marne
    • Posts 40
    • Points 605
    monte-carlo simulation running one seed Reply

    Hi All,

    I am using 6.1.5 environment and I have ran a monte-carlo simulation over 100 runs yesterday and I load back my results. Some runs have failed and I wish to see more data on these particulars seeds. I went to the monte-carlo options and choose the specific seed number and re-run only this one with monte-carlo.

    As a result, I do not have the correct seed environment and so the same results.

    How could I re-run a specific seed number from a monte-carlo run with all data such dc voltage, dc operating point, waveforms specified, ...?

    Thanks in advance for your help,

    Fabian

    • Post Points: 20
  • Wed, May 9 2012 1:45 AM

    Re: monte-carlo simulation running one seed Reply

    Hi Fabian,

    The best way is not to mess around with the "seed". If the seed was not set, it defaults to 12345. It's the start run number that you'd want to use, rather than the seed. The seed alters the sequence of random numbers that are generated, and the start run number tells you how far that sequence you are.

    The easiest way is take your full 100 run monte carlo results, switch to Detailed view (rather than Yield), and then over the particular run numbers that failed, do a right-mouse-button->Create Statistical Corner. This will then add into the corners the specific runs which you're interested in.

    Then if you run in Single Run, Sweeps and Corners mode, and have the corners enabled, it will re-run the specific monte carlo points that you're interested in - making it easier to debug them.

    Regards,

    Andrew.

    • Post Points: 20
  • Wed, May 9 2012 2:07 AM

    • frogconsultant
    • Top 200 Contributor
    • Joined on Thu, Feb 18 2010
    • Bourron Marlotte, Seine-et-Marne
    • Posts 40
    • Points 605
    Re: monte-carlo simulation running one seed Reply

     Hi Andrew,

    Thanks for your quick reply.

    I select the run number of interest and create a statistical run after reloaded the monte-carlo run from yesterday (reload results). Then run only this new run as a Single Run, ...

    The outcome results are not the same as the specific one selected. How could I make sure this new run created is the one which failed in the monte-carlo run?

    Regards,

    Fabian

    • Post Points: 20
  • Wed, May 9 2012 3:14 AM

    Re: monte-carlo simulation running one seed Reply

     Fabian,

    They should be the same. If they're not, you should log a service request with us to take a look. Potentially there are some situations where the circuit has multiple operating points and it might converge due to subtle differences in starting conditions to one or the other - but it's hard to tell without seeing the data.

    Regards,

    Andrew.

    • Post Points: 20
  • Thu, Feb 7 2013 7:03 AM

    • pirateKing08
    • Not Ranked
    • Joined on Fri, Sep 16 2011
    • Cirencester, Gloucestershire
    • Posts 6
    • Points 105
    Re: monte-carlo simulation running one seed Reply

    Hi Andrew

    I use statistical corners and I can re-produce the same montecarlo run number result when re-run as single statistical corner run on the same cct. What concerns me however is har far can you change your cct before that worst case statistical corner is no longer valid? I understand your flow suggests using statitstical corners to parametric optimize a cct  but is that as long as the letlist doesn't change? Is the seed also related to the netlist? I end up with loads of worst case statistical corners when developing a circuit as the previous worst case statistical corners no longer produce the worst result in a later MC run if the cct has changed. Also grabbing the worst case statistical conrers for a schematic run then extracting the layout and running just the statistical corner extremes gives better results which I am aprehensive about. I still feel I need to run a full MC again as I don't believe the worst case statistical corners are still the worst case.

    Regards

    Lance

    • Post Points: 20
  • Sun, Mar 24 2013 5:08 AM

    Re: monte-carlo simulation running one seed Reply

    Lance,

    Statistical corners are fine if you're using process variation. If using mismatch, then they're also fine if you only change the parameters (or global variables) in the design - if you modify the circuit itself, then the mismatch won't be identical any more - because the sequence of random numbers will be affected by the differences in the circuit.

    Regards,

    Andrew.

    • Post Points: 20
  • Mon, Mar 25 2013 2:36 AM

    • Frank Wiedmann
    • Top 100 Contributor
    • Joined on Wed, Jul 16 2008
    • Muenchen, Bavaria
    • Posts 75
    • Points 1,160
    Re: monte-carlo simulation running one seed Reply

    In my opinion, this behavior severely limits the usefulness of statistical corners in ADE XL. Solido told me that they don't have this problem because they assign mismatch parameters to components by their name and not by netlist order. I think that it would be a really good idea for Cadence to do the same.

    • Post Points: 20
  • Mon, Mar 25 2013 2:53 AM

    Re: monte-carlo simulation running one seed Reply

    I'm not sure I'd agree that it severely limits the usefulness, since it's still fine provided that you are not making topological changes to the design, but your point is understood. It's actually spectre that is doing this because a statistical corner just runs a single point monte carlo, but that's an implementation detail - ideally we can come up with something that is a bit more general whilst still retaining the performance benefit of doing the random number generation in the simulator rather than having to impose it from outside.

    Regards,

    Andrew.

    • Post Points: 20
  • Mon, Mar 25 2013 3:07 AM

    • Frank Wiedmann
    • Top 100 Contributor
    • Joined on Wed, Jul 16 2008
    • Muenchen, Bavaria
    • Posts 75
    • Points 1,160
    Re: monte-carlo simulation running one seed Reply

    Well, it seems to be a major problem for Lance and I'd guess that he's not alone. When you fix this issue, please don't forget to also deal with the correspondence between schematic and extracted views that he mentioned.

    • Post Points: 20
  • Mon, Mar 25 2013 3:54 AM

    Re: monte-carlo simulation running one seed Reply

    Frank,

    I'm not saying it's not a problem in some caes - just saying that in many cases it is still useful.

    Getting the monte carlo to correlate between the schematic and extracted views is probably quite a tall order, since there is often not a one to one correspondence between the devices anyway (for example, an m-factored transistor in the schematic will end up as separate devices in the layout).

    Regards,

    Andrew.

    • Post Points: 20
  • Tue, Mar 26 2013 1:22 AM

    • Frank Wiedmann
    • Top 100 Contributor
    • Joined on Wed, Jul 16 2008
    • Muenchen, Bavaria
    • Posts 75
    • Points 1,160
    Re: monte-carlo simulation running one seed Reply

    I agree that this is a little more difficult, but it should still be doable, for example by using information from an LVS run. Of course, a first step should be to give consistent results for a modified version of the same view (including an extracted view).

    • Post Points: 5
  • Tue, Jun 18 2013 12:44 AM

    • Frank Wiedmann
    • Top 100 Contributor
    • Joined on Wed, Jul 16 2008
    • Muenchen, Bavaria
    • Posts 75
    • Points 1,160
    Re: monte-carlo simulation running one seed Reply

    In the meantime, this problem has been submitted as CCR 1136784 (enhancement request). Anyone who also has a problem with the behavior described in this thread should consider asking Cadence support to file a duplicate of this CCR on his behalf. If more people are asking for this enhancement, it is more likely to be implemented soon.

    • Post Points: 5
Page 1 of 1 (12 items)
Sort Posts:
Started by frogconsultant at 09 May 2012 01:35 AM. Topic has 11 replies.