New ADE feature for statistical design variable definition: Global vs local? Correlation?

Stephan,

It's not new. It's been there for probably 5 years (see How to vary design variables with statistical distribution to be used with Monte Carlo analysis). It only creates process variations - this is because for mismatch variations, you need to have a component or model within a subckt that reference the parameter with mismatch variation so that each instance has a local variation. See A simple method of applying mismatch to ideal components and Recommended Spectre Monte Carlo modeling methodology. Because of the need to define a subckt model (done in the article above with a sub-schematic), we've not added a UI to set up the mismatch variation because you can guarantee that otherwise people would try it, not see the variation, and then there would be a flood of requests...

There is a way of creating correlations using constraints, but to be quite honest the best thing is to define your models that need variation so that they follow the methodology in the Don's document above (Recommended Spectre...) and then you can also define the statistics block - which can also have correlation coefficients - see "spectre -h montecarlo" - you can then define the process and mismatch variation (global and local, in other words) and any correlations too. Then include this in your simulation as a model file.

Andrew