Stepwise regression for identifying sources of variation in a semiconductor manufacturing process

In any semiconductor manufacturing process identifying tools that are significant sources of variation is critical to developing and maintaining a high yield process. Unfortunately, the most common tool for accomplishing this, analysis of variance (ANOVA), is not well suited for this task. We propose to apply stepwise regression (SWR), a well-known statistical modeling technique, for identifying tools with large contributions to the overall variance. In this paper we explain how SWR can be used to analyze the semiconductor manufacturing process and then discuss simulations which show that in many common situations SWR performs better than ANOVA.