Generation and verification of optimal dispatching policies for multi-product multi-tool semiconductor manufacturing processes

Semiconductor manufacturing is one of the fastest-growing industries today. As the recent requirements for feature sizes and wafer sizes change rapidly, it becomes imperative to configure increasingly intricate control schemes to maintain product quality and tool utilization rate. For this purpose, it is assumed in this study that a semiconductor production environment can be viewed as multiple queues operated in parallel and, also, the EWMA controllers can be implemented independently to adjust the process recipes of different products in each queuing system. Based on these assumptions, a MINLP model is formulated to determine the optimal dispatching policies. Systematic numerical simulation procedure is also devised to confirm the validity of the dispatching model. Since accurate estimates of the model parameters may not always be available, the effects of model mismatch have been analyzed and the proper range of controller tuning parameter is recommended to achieve an acceptable level of process capability.

► A MINLP model has been constructed to determine the optimal dispatching policy.
► The model-predicted policies have been verified in numerical simulation studies.
► Results show product quality can be improved without sacrificing utilization rate.
► Extension to more complex systems is straightforward.