Applying Deep Reinforcement Learning to Semiconductor Manufacturing Scheduling Problems

In this paper, we proposed a scheduling method for semiconductor manufacturing systems by extending that for a job shop scheduling problem (JSSP) using deep reinforcement learning approach. The proposed method can handle the unique characteristics of the semiconductor manufacturing process, i.e., the reentrancy and the flexibility. The advantage of the proposed method is scalable and can effectively handle increasing problem sizes. From the results of experiments, we confirmed that even when small-sized tasks were used in the training, the scheduling performance for large-sized tasks was equal to or better than the case using large-sized tasks in the training.