Centralized model predictive control strategies for inventory management in semiconductor manufacturing supply chains

Centralized strategies based on model predictive control (MPC) are applied to inventory management problems associated with semiconductor supply chains. Specifically, two benchmark problems of relevance to semiconductor manufacturing are examined. The first one is a single product, two node problem consisting of a fab/sort and an assembly/test facility controlled with a predictive controller using anticipation. The performance of the control scheme under conditions of plant-model mismatch, and unforeseen demand are evaluated. The insights gained from this problem are used in the design of a centralized MPC controller for a four node problem involving two interconnected fab/sort and assembly/test facilities. In this latter problem, inventory management of wafer, die, and package inventories are considered.