Haystack syndrome avoidance on massive correlation for probe vs. E-test data through the concurrent use of tree base models and trellis graphics. Application on sub-micron mix-signal product for the determination of the best process conditions for yield maximisation

Abstract

In an environment which requires an intensive capitalisation like the semiconductor industry, time-detection/time-reaction to any kind of yield degradation is a key issue. Abundant literature covers the existent methodologies and strategies to prevent, detect and react to the cosmetic defects. Even hardware solutions are available in the market which offer the possibility through product or control inspections to monitor them for the analysis of electrical measurements, performed at the end of the process sequence, and their relationships with yield predictors. The analysis of hundred of variables associated with a yield descriptor presents an important challenge from the statistical standpoint. We have developed a solution that allows an easier differentiation of the main contributor variables to the yield descriptor explanation as well as a graphical output of the analysis. The graphical output also includes the concept of multivariate analysis. Multivariate analysis accounts for the relationships of several different variables against each other. Trellis library provides excellent graphical solutions to this type of analysis.