Data-Driven Lot Sizing with Random Yield

Abstract

Motivated by yield uncertainty in semiconductor manufacturing, we explore a data-driven lot sizing problem under random yield. Our focus is on the case where the random yield rate of the process depends on a number of features that can be observed before the lot sizing decision is made. We develop and compare the performance of different estimation and optimization approaches for this problem.  We calibrate and test the methods on a publicly available data set for feature-dependent semiconductor yield data which presents challenges that are typical in prescriptive analytics in operations: a large number of features and relatively few observations. 

Authors: Fikri Karaesmen, Koç University (joint work with Bijan Bibak).