Regarding Monte-Carlo Simulations

Trying to guess an appropriate number of Monte Carlo points is very hard. How many is enough will depend on your yield target - the higher the yield target, the more points you are likely to need. I don't know why 100 would be chosen (100 is somewhat arbitrary, and it's almost certainly not enough for any reasonable yield target, but it may give you a reasonable idea about the sensitivities of your design).

In general you want to use the yield verification choice when using ADE Explorer/Assembler. The Monte Carlo setup form for ADE Explorer/Assembler now asks you what you're trying to do - and it can do three things:

1. Run a fixed number of points. This assumes you have a fixed budget (maybe time) for simulations and so you need to get a rough idea but aren't being very precise about the yield
2. Verify the yield - you can then enter the yield target as either a percentage or in sigma, and it will run enough simulations to be confident that it either meets or fails to meet the target. By default it would use the Virtuoso Variation Option license to allow sample reordering to efficiently run the worst samples first, but you can also choose "basic auto stop" on the advanced options which can be run without an additional license (it will still be able to stop early if the confidence limit has been reached, but most likely will need more simulations than the sample reordering flow). The yield verification also supports high-sigma (e.g. up to 6 sigma) methods which use a different approach. Don't be tempted to say that your yield target is 100% because that needs an infinite number of simulations ;-). All of this requires you to have specifications on your output measurements, by the way.
3. Statistical Corner creation - this is an automated way to get (say) 3 sigma corners for each of your measurements. These can then be used in non MC sims to tune your design.

A good place to understand all this is in the Rapid Adoption Kit here: Advanced Statistical Analysis for Variation Aware Design

Regards,

Andrew.

I am thankful to you for guiding so well.

Dear Kabes,

For your information, my approach to answer your question regarding what is considered a “sufficient” number of Monte Carlo simulations is to initially perform a very large number of Monte Carlo simulations on a small circuit whose statistical behavior mimics the circuit under study. If your circuit under study is small, you could simply use it. Using the raw output data for the parameter of interest, I compute the estimate of the standard deviation as a function of the number of simulations and plot the result. I have attached an example of such a result. From this data, and knowing the statistical requirements for your estimate of the standard deviation, you may better estimate the number of Monte Carlo simulations to perform.

Does this address your question?

Shawn

Thanks for your reply.

I will work with this strategy.