Home > Community > Forums > Custom IC Design > monte carlo simulations on instance/output parameters

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 simulations on instance/output parameters 

Last post Thu, Nov 21 2013 12:02 PM by Marc Heise. 6 replies.
Started by reliability 19 Nov 2013 08:35 AM. Topic has 6 replies and 1085 views
Page 1 of 1 (7 items)
Sort Posts:
  • Tue, Nov 19 2013 8:35 AM

    monte carlo simulations on instance/output parameters Reply

    Hi, I am using ADEXL to conduct MC statistial simulations.I know it is easy to do MC simulation on model paramters like Vth by adding statistical distribution in model card. however, here I need to statistically simulate instance parameters like Leff, width of the nmos...  is there any way to do it? thank You in advance

    Ps. Is there anyway to do MC simulations on other variables like power supply ?

    • Post Points: 20
  • Thu, Nov 21 2013 2:52 AM

    • Marc Heise
    • Top 100 Contributor
    • Joined on Mon, May 4 2009
    • Feldkirchen, Bavaria
    • Posts 77
    • Points 1,310
    Re: monte carlo simulations on instance/output parameters Reply

     Hi,

     you can vary any parameter you want by just providing a statistic section for it inside an included model file.

    For example: Instead of giving   1k  to an analogLib resistor  you write   1k*x   and create a statistic for "x".

    The problem is, you want use several instances of that res and apply a different "x" variation. In that case you need to push
    the "x" into a subcircuit, means you need to create an additional schematic/symbol for that resistor cell. Placing the resistor and pins in the schematic and parameterizing the value with  pPar("res")*x. The you create the symbol for it.

    If you have a support account you can check:

     http://support.cadence.com/wps/mypoc/cos?uri=deeplinkmin:ViewSolution;solutionNumber=11485928;searchHash=7263fedf02d98d96b59c2a2254744246

    Kind regards,

    Marc

    • Post Points: 20
  • Thu, Nov 21 2013 10:33 AM

    Re: monte carlo simulations on instance/output parameters Reply

    Thank you so much for your help, Marc!

     

    I have one more question about providing a statistical section for one instance parameter? like the width of an NMOS fet....Should I assign it in the process section or mismatch section? 

     Is it use process statistial format to define it as follows:

     

     section stats

    simulator lang=spectre

    process {

    vary width dist=gauss std=0.1 percent=no

    }

    endsection stats

     

    • Post Points: 35
  • Thu, Nov 21 2013 10:59 AM

    • Marc Heise
    • Top 100 Contributor
    • Joined on Mon, May 4 2009
    • Feldkirchen, Bavaria
    • Posts 77
    • Points 1,310
    Re: monte carlo simulations on instance/output parameters Reply

    That depens on what you are planning to vary.

    Process:  Your device is probably produced on more than one wafer. No wafer is like the other. The production "process" will vary. This is captured by the process variation.

    Mismatch:  Now that device is existing more than once on that same wafer. But also here you have some kind of variation depending on where the device is located on the wafer. This is captured by the mismatch variation.

    Now you can apply both or just one of them, realy depends on what you are looking for.

    For a resistor the process variation can be around +/- 20% and the mismatch +/-5% ( depends on the foundry).

    Process variation is usally larger than mismatch variation.

    Kind regards,

    Marc

     

    • Post Points: 5
  • Thu, Nov 21 2013 11:17 AM

    • Marc Heise
    • Top 100 Contributor
    • Joined on Mon, May 4 2009
    • Feldkirchen, Bavaria
    • Posts 77
    • Points 1,310
    Re: monte carlo simulations on instance/output parameters Reply

     Depends what you want to vary, both or one of them.

     

    Process...variation of devices on different wafers.

    Mismatch...variation of devices on same wafer.

    If you select both, mismatch is applied ontop of process.

     

     ( hmm.. kind of double post. Got an error after the first message and expected it to be lost))

     

    • Post Points: 20
  • Thu, Nov 21 2013 11:56 AM

    Re: monte carlo simulations on instance/output parameters Reply

    thank you very much for your detailed answer, Marc!

     

    Another question is Width of the devices is a design variable is unlike threshold voltage which is defined in the model card. I tried to add the process/mismatch for width/VDD distribution in the model card, but my output does not change at all...Actually if i change the width/VDD by sweeping simulation instead of MC simulations , the output actually changes a lot... it seems to me the the statistical sentences controling the distribution of width is not affecting the MC simultion.....Do you know the answer to that?

    • Post Points: 20
  • Thu, Nov 21 2013 12:02 PM

    • Marc Heise
    • Top 100 Contributor
    • Joined on Mon, May 4 2009
    • Feldkirchen, Bavaria
    • Posts 77
    • Points 1,310
    Re: monte carlo simulations on instance/output parameters Reply

    Without some hard data (netlist, model, log) there is not much I can do. 

    • Post Points: 5
Page 1 of 1 (7 items)
Sort Posts:
Started by reliability at 19 Nov 2013 08:35 AM. Topic has 6 replies.