anyone familiar with hspice? Help on Monte carlo sweep

[sunjiao3]

Hi, dear friends, I am doing a Monte Carlo simulation for my hysteresis comparator to analyse the mismatch effect on the width of the hysteresis window. There are 2 questions.
First, I set the width of one of my differential pair transistors as something like Gauss(,,,),and I sweep it for 60 times. To my surprise, the samples not ranger equally beside the original value I choose. I mean, the histogram of this parameter is far from symmetry. So, I add the number of my sweep to 100, the problem seem not alleviated. Why? Should I go on adding the sweep number to get more "realistic" Gauss specs?

Second, I do 2 dc sweeps to get both the possitve and negtive trig point of this hysteresis comparator. And I use very small steps, like 0.01mv to get detailed out. Due to this small step and the large sweep number, like 60, the simulation take a very long time__it is a DC sweep!! So, is there any method to speed up it? I mean, only use small steps near the trig points to get detailed information. I know Hspice has some funciton like it in Tran sweep, but here, it is DC sweep.

Could any one of you give me some suqqestions on the questions? Thank you in advance.

Added after 30 minutes:

Here are the histogram graph of the parameter I set as Gauss distribution like Gauss(8u,0.1,3). The first graph is the histogram of 60 samples and the last is the histogram with 100 samples. The unsymmetry problem seems not solved.

 

[hunk]

Actually, A reasonable number is 30. 60 is very large. then the time consuming doesnt surprise. The statistical significance of 30 iterations is quite high. If the circuit operates correctly for all 30 iterations, there is a 99% probability that over 80% of all possible component values operate correctly.

 

[sunjiao3]

Thank hunk for your clear answer. I read in some paper that the more times Monte Carlo is did, the more accurate anticipation we can get from the result of span. And, it is recommended that at least 1000 times sweep in that paper. And I did use 1200 times sweep. Is there some standard on this topic? Or is there some rule of thumb?

 

[beargebinjie]

coulde tell me how to extract the paramter viariations(Vth and w etl.) form the process lib file? thanks!

 

[heverlee]

I would say the important thing is whether or not you want to catch those extreme values or "outliers". If they are not important, 30 is enough to have an idea

Actually, A reasonable number is 30. 60 is very large. then the time consuming doesnt surprise. The statistical significance of 30 iterations is quite high. If the circuit operates correctly for all 30 iterations, there is a 99% probability that over 80% of all possible component values operate correctly.