The Designer's Guide Community Forum

The Designer's Guide Community Forum
https://designers-guide.org/forum/YaBB.pl
Modeling >> Semiconductor Devices >> Convert corner files to Monte-Carlo?
https://designers-guide.org/forum/YaBB.pl?num=1323697123

Message started by weber8722 on Dec 12^th, 2011, 5:38am

Title: Convert corner files to Monte-Carlo?
Post by weber8722 on Dec 12^th, 2011, 5:38am

Hi,

does anybody has a script to convert corner files defining e.g. SS, FF, FS, etc. to an equivalent MC process setup, e.g. by treating the corner values as 3-sigma values for MC? Best for Spectre! I have a pdk which has MC setup only for mismatch, but not for process variations. :-?
Would be also good if it can treat the arising correlations e.g. between VTO too.

Bye Stephan

Title: Re: Convert corner files to Monte-Carlo?
Post by Geoffrey_Coram on Dec 14^th, 2011, 8:38am

I don't think this is going to work the way you think.

Suppose your process has TOX and VFB varying for the different corners (probably, there are other parameters also, but let's keep it simple). If you assume the TOX for the SS and FF corners define the 3-sigma points of the TOX distribution, and similarly for VFB, then the MC distribution you end up with will be *very* unlikely to hit the 3-sigma points on *both* of those independent variables.

If you try to pick the SS, FF corners as 3-sigma points of the combined distribution, you'll have to make some guesses as to the relative effects of the different parameters: is the "speed" strongly dependent on TOX, but TOX is tightly controlled, or is it weakly dependent on TOX that is poorly controlled?

Let's pretend that you add the sigmas of the two parameters to get the sigma of the MC distribution (I'm sure the math is more complicated, but I think this can illustrate my point without making me spend the time to do the derivation): the 3-sigma point of the MC distribution could be:
a) the 2-sigma point of TOX plus the 1-sigma point of VFB
b) the 1-sigma point of TOX plus the 2-sigma point of VFB
c) the 0.1-sigma point of TOX plus the 2.9-sigma point of VFB

and you can see that the corners are really insufficient for making any kind of assumption about the real distributions.

Title: Re: Convert corner files to Monte-Carlo?
Post by weber8722 on Jan 2^nd, 2012, 2:07am

Hi Geoffrey, my thinking was similar and so I expect that the easy way is too easy to really allow yield simulations.

A similar problem will come up for VTO:
TOX shifts VTO but both NMOS and PMOS, leading also to extreme values for SS & FF corners. However, for SF & FS corner the VTO shifts are less extreme than in SS and FF due to correlations.

My thinking is that IF we combine all 5 cornerss, like TT,SS,FF,SF,FS than also such correlations CAN be treated good enough for most purposes.

Bye Stephan