Abstract
One novel methodology of reinforcement learning run-to-run control used for reducing the Chemical-mechanical polishing (CMP) rework rate was proposed in this study. The rework rate of polished wafers is the key issue of advanced semiconductor CMP manufacturing. Thus, we proposed one artificial intelligence (AI) embedded manufacturing approach to maximize the efficiency of CMP manufacturing. The proposed hybrid approach was composed of a reinforcement learning mechanism and a CMP run-to-run controller. The operation know-how of the CMP process was applied as the novel reinforcement learning run-to-run controller. In this study, we manipulated the wafer quality and quantity during key periods of the CMP manufacturing cycle with deep reinforcement learning. The proposed novel run-to-run CMP controller was superior to the existing model with the stochastic filter controller. Simulation results confirmed the advantage of advanced semi-conductor FAB throughput enhancement by double digits.