Multilayer optimization and scheduling using model predictive control: application to reentrant semiconductor manufacturing lines

Abstract A two-layer production control method applied to discrete event reentrant semiconductor manufacturing lines is investigated. A modified l 1 -norm predictive controller/optimizer is proposed as a coordinator in the highest layer and a distributed control policy is used as a follow-up controller in the lowest layer. The use of a model predictive control (MPC) formulation allows the scheduling algorithm to simultaneously solve the production optimization and in-process inventory control problems at each sampling time. As an optimizer, the scheduler maximizes production rate and, as a controller, it addresses variability. Using this control-oriented framework, an optimal trade-off between production rate and cycle time is obtained. An l 1 -norm cost function allows the implementation of the optimization layer as a mixed integer linear program (MILP) which is solved at each time shift. The approach is applied to a one-product six-step, five-machine reentrant discrete event semiconductor manufacturing line whose specifications were provided by Intel Corporation.