ABSTRACT:

Shrinking process windows in very large scale integration semiconductor manufacturing have already necessitated the development of control systems capable of addressing sub-lot-level variation. Within-wafer control is the next milestone in the evolution of advanced process control from lot-based and wafer-based control. In order to adequately comprehend and control within-wafer spatial variation, inline measurements must be performed at multiple locations across the wafer. At the same time, economic pressures prompt a reduction in metrology, for both capital and cycle-time reasons. This paper explores the use of modeling and minimum-variance prediction as a method to select the sites for measurement on each wafer. The models are developed using the standard statistical tools of principal component analysis and canonical correlation analysis. The proposed selection method is validated using real manufacturing data, and results indicate that it is possible to significantly reduce the number of measurements with little loss in the information obtained for the process control systems.