Intelligent ALD—Real-Time Prediction, Anomaly Detection, and Recipe Optimization

Atomic Layer Deposition (ALD) requires atomic-level precision, where slight process variations directly impact uniformity, properties, and yield. This presentation introduces an Intelligent ALD framework that integrates equipment, sensor, and metrology data into a closed-loop of learning, inference, and feedback. By combining high-frequency signals (≤100 ms) with metrology results, the system enables real-time property estimation with 30–60 second lead prediction, on-tool fault detection, and multi-objective recipe optimization. A hybrid data strategy and Bayesian Optimization with Reinforcement Learning handle rare anomaly data and continuous/discrete parameters. Validations show early anomaly detection reduces defective runs, while uniformity prediction accelerates fine recipe tuning. Currently under validation with KETI and partners, the framework demonstrates yield and productivity gains and is being extended to Dry Etch, with future scalability to CVD and CMP tools.