The value of σ in cadence monte carlo simulation

[rosaeidi]

Hi

I use two simple voltage sources for input offset in my design.
How can I set their σ value in cadence monte carlo simulation? I want to use gaussian function.

Rosa

 

[erikl]

Depends on how much of the Gauss distribution you want to cover: with ±1σ you include ≈68% of all possible cases under it (you could also say: 68% confidence level, or trustability), with ±3σ about 99.7% of all cases are covered, see e.g. this Wiki explanation.

 

[rosaeidi]

I don't know how I can apply it on monte carlo simulation of cadence. I use "tools/monte carlo" icon.

Rosa

 

[erikl]

See the C@dence Monte Carlo tutorial below, which shows the setup for several examples. In the statistics { process, mismatch } section, the parameter distribution type (gauss, lnorm, ...) and the standard deviation (std) are stated. std is identical to σ, see the a.m. Wiki explanation.

View attachment Cadence_Monte_Carlo_simulation_tutorial.pdf

Don't be bothered by the low std values shown there e.g. on p. 13; these are example values only; the correct σ values should be given by the foundry. For the variation of your own parameters - like the variation value of a voltage source - you can state an arbitrary σ=std value, which should comply to the expected variation.

The mean value of the parameters is to be given in the parameters section.

 

[timof]

Also, don't forget that if one voltage source has standard deviation sigma1, and another voltage source - sigma2, then the difference or the sum of these voltages (assuming they are statistically independent) is sigma=sqrt(sigma1^2+sigma2^2). (it's not clear form the description of the initial post how do these two voltages define the input offset).

 

[erikl]

Actually - for input offset effect analysis - it would be sufficient to put one single voltage source in series with one of the two inputs of a differential input amplifier. Mean value = 0 , and a (process-dependent) ±1σ value of -say- 10mV .

If you then run an MC analysis with -e.g.- 400 runs over a ±2σ distribution, you cover about 95% of all possible cases - with a max. error of 5% - normal (=Gauss) distribution assumed.

 

[Alchemist_]

erikl, nice tutorial