Machine learning-guided optimization of atomic layer deposition process

Abstract Optimizing atomic layer deposition (ALD) process requires precise control of multiple parameters, whose nonlinear interactions often demand extensive experimental iterations. Here, we present a machine learning (ML) framework for developing ALD model that accelerates process optimization. The framework is applied to predict hafnium oxide (HfO x ) film properties—thickness (T ox ), refractive index (RI), and wet etch rate (WER)—from ALD process parameters. Trained on 215 experimental data points using a deep-neural network (DNN), the model achieves prediction accuracy above 90% within ranges of ±1.0 nm for T ox , ±0.04 for RI, and ±0.9 Å/min for WER. Beyond numerical prediction, the framework generates film-property maps that enable process visualization and support autonomous ALD optimization. Comparable performance is retained with 100 training data points, effectively capturing nonlinear trends and extending predictability beyond experimentally measured ranges. This approach on ML-driven ALD model significantly reduces experimental time and resource consumption while effectively addressing complex multi-parameter challenges in ALD. Importantly, these results lay the groundwork for integrating ML into ALD digital twin platforms, establishing a foundation for next-generation autonomous semiconductor process development.