Sequential Screening in Semiconductor Manufacturing, I: Exploiting Spatial Dependence

This paper addresses the same quality management problem as Ou and Wein (1992), except that here screening is performed at the chip level, rather than at the wafer level. We analyze over 300 wafers from two industrial facilities and use a Markov random field model to capture the spatial clustering of bad chips. Chip screening strategies are proposed that exploit the various types of yield nonuniformities that are detected in the data, such as radial effects, spatial clustering of bad chips, and yield variation by chip location. The numerical results suggest that screening at the chip level is significantly more profitable than screening at the wafer level.