Stats

  • Replies 2
  • Subscribers 131
  • Views 426
  • Members are here 0

Is random sampling necessary if LHS LDS exist

CC202602246636

 Monte Carlo Sampling Method 

 Fast Yield Analysis and Statistical Corners 

I was studying about Monte Carlo Sampling methods and luckily I came by these two posts.

It seems that LHS and LDS give similar results for a limited number of runs when it comes to estimating the dataset, due to spreading out limited samples in the population.

Mathematically, when the number of runs reach infinite, sampling methods do not matter.

However, for a limited number of runs (which is usually the case) LHS and LDS are preferred over random sampling.

So are there any cases, as an analog designer, random sampling is preferred over LHS and LDS?

If no, then why does random sampling still exist as an option?

Thanks!

  • Andrew Beckett

    Two scenarios where random sampling is still used:

    1. When using Spectre's Fast Monte Carlo, it essentially generates a very large number of samples (especially when there are high sigma requirements) to feed into the model used to help find the tail samples. Because so many samples are generated, there is no benefit of LDS or LHS, and there is some overhead in using LDS (and LHS) compared with random.
    2. Spectre has a feature called "mismatch id" which allows you to assign separate random number generators to different blocks (or the same random generator to multiple instances of the same block) - this is useful when you want to look at a test bench where you have multiple instances representing different scenarios but you want these to represent the same occurrence in the design and so should be correlated. This mismatch id mechanism only supports random.

    There's not really any benefit of using LHS over LDS. LHS has some downsides in fact because you have to know in advance the number of samples and you need to have generated all the samples to properly cover the surface (i.e. it doesn't handle auto-stopping if needed, or adding additional samples later). So I'd say that for all practical purposes (unless you are running a very large number of samples), then LDS is the choice to go for, unless you\re using mismatch id. Or if you're using the Fast Monte Carlo system.

    Andrew

  • CC202602246636
    in reply to Andrew Beckett

    Hi Andrew. Thank you so much for clearing my confusion.

Community Guidelines

The Cadence Design Communities support Cadence users and technologists interacting to exchange ideas, news, technical information, and best practices to solve problems and get the most from Cadence technology. The community is open to everyone, and to provide the most value, we require participants to follow our Community Guidelines that facilitate a quality exchange of ideas and information. By accessing, contributing, using or downloading any materials from the site, you agree to be bound by the full Community Guidelines.