Monte Carlo simulations: how many?

[Humungus]

I come again with another phylosophical question;-)

Normally, Monte Carlo methods assure that 30 runs are enough to be sure at a 90 percent that 96 percent of your designs will be between the maximum and minimum values found by the Monte Carlo simulations.

I supose that this is for only one parameter varying, isn't it?

What can the industrial guys say?

For a system where more than 20 parameters can be varying at same time, how many simulations we should do?

 

[jako123]

how many MonteCarlo runs!?

Hello!

I'm doing allways 200 run for my analog designs -> at least!!!

to cover ererything

ciao

 

[bastos4321]

This is an information from Alcatel Made matching tech. note:

*******************************************
How many Monte Carlo simulation runs are needed ?
Error on the standard deviation as function of the number of Monte Carlo simulations:
35 runs for 25% accuracy with 95% confidence level
100 runs for 15% accuracy with 95% confidence level
250 runs for 10% accuracy with 95% confidence level
850 runs for 5% accuracy with 95% confidence level
e.g. The offset of an OpAmp is simulated with a 100 run Monte Carlo matching analysis.
The resulting matching sigma = 2.58 mVolt
From the above Table, the accuracy is 15%
Design for 4 sigma (matching): the OpAmp offset specification is: 2.58mV x 1.15 x 4 = 11.87 mVolt.
********************************************************

Bastos

 

[bastos4321]

Alcatel Made matching tech. note:

Alcatel Made matching tech. note: