Integrated Hierarchical Reinforcement Learning for Enhanced CMP Manufacturing Efficiency

The application of artificial intelligence (AI) to the chemical-mechanical planarization (CMP) process control was proven to reduce rework rates and significantly enhance manufacturing efficiency in this study. A hierarchical AI CMP controller with an auto-tuning feature is proposed. This integrated hierarchical reinforcement learning (iHRL) control is designed to eliminate the CMP variations caused by both the removal rate decay and pattern density changes. The iHRL is implemented with a well-designed hierarchical wafer-to-wafer and lot-to-lot control architecture. It is demonstrated that the self-learning auto-tuning control effectively reduces the CMP rework rate significantly. In this study, comprehensive simulation studies show the average manufacturing efficiency compared with the original run-to-run controller is enhanced by more than 20%.