Planning wafer allocation for CMOS process development. A nonparametric approach

In this paper, we present techniques that can be used to answer the following two questions: (1) how many wafers need to be allocated per treatment to detect a given difference in a device performance metric and (2) how can one determine if a given treatment significantly improved a performance metric? The approach presented here does not make any assumptions regarding the shape of the distribution or the spatial dependency structure for the within-wafer performance measurements and remains applicable for a variety of performance metrics, such as mean, variance, and median. The analysis method can be used in decisions regarding the appropriateness of allocating half or quarter wafer splits to a treatment. Furthermore, the approach allows us to evaluate and compare within-wafer sampling strategies for comparing performance metrics from competing flows.