Deep reinforcement learning based scheduling within production plan in semiconductor fabrication

Abstract

In the semiconductor industry, efficient production planning and scheduling decisions are required to enhance the manufacturing productivity of a company as the system is complicated due to a re-entry characteristic and requires a long production lead time. Production planning is implemented before scheduling and is important for successful manufacturing operations. However, if scheduling at the operation level cannot execute the production plan, failures occur because of inconsistent decisions. Therefore, scheduling needs to fulfill the production plan to ensure realistic decision-making processes for the companies aiming for economic growth and global competitiveness. In this study, deep reinforcement learning (RL) is employed to deal with a scheduling process operating within the production plan. As the algorithm of the deep RL, Deep Q-network is conjugated, and a novel state, action, and reward are suggested to optimize the scheduling policy. As a result, the performance of the proposed deep RL method is in comparison with other dispatching rules, and the proposed method outperforms the other scheduling methods in diverse cases.