May

29

Poisson’s or Murphy’s yield models?

Filed Under Semiconductor Industry, Video Gallery 

A yield prediction model is used to estimate the yield for a given die area. Traditionally, Poisson’s model is the most dominant model used by most fabs. Recently, I came across some fabs exactly used Murphy’s model instead of Poisson’s model. What is Murphy’s yield model? I have not heard of this yield model before. After googling on the internet, I found that Murphy’s yield model was actually a general yield methodology and it was studied in detail way back in the 80’s and 90’s by a group of IBM’s researchers [Ref].

By assuming a certain defect clustering distribution, it is possible to work out an analytical yield model based on Murphy’s yield methodology. If we assume a uniform defect distribution across the wafer, we get the traditional Poisson’ yield model. If we assume a triangular defect profile, then we got the Murphy’s yield model [Ref]. A comparison of the 4 major yield models is shown in the figure. The question is “What is the most accurate yield model should the fab use?” It is difficult to tell, it depends on the defect distribution. From the figure, poisson’s yield model shows the most pessimistic yield prediction for larger die area . If we want to quote the worst case, then we should use Poisson’s yield model. However, some fabs may want a better yield number for marketing purpose, and they might prefer to other models :)

It seems that IBM has done a lot of fundamental research way back in the 80’s. It is amazing. Take a look at the IBM 12″ Fab in the following video.

Related posts

Tags: , , , ,

Comments

Leave a Reply