Achieving High-Crystallinity YSZ Thin-Films on Si Using Explainable AI and Bayesian Optimization in Resource-Constrained Settings

Abstract

The growth of epitaxial yttria-stabilized zirconia (YSZ) thin-films on Si substrates is critical for integrating functional oxides into advanced electronic and energy devices. However, identifying the optimal growth conditions for YSZ thin-films remains challenging because of the complex and interdependent nature of deposition parameters in thin-film growth processes. Herein, we propose an integrated approach that combines Shapley additive explanations (SHAP), which is an explainable artificial intelligence (XAI) method, with Bayesian optimization (BO) to efficiently identify key growth parameters and optimize crystallinity in resource-constrained settings. By performing SHAP analysis of literature-derived data, oxygen partial pressure was identified as the most influential parameter, followed by temperature. Guided by this ranking, BO was employed to minimize the full width at half maximum (FWHM) of the YSZ 002 rocking curve. High-quality epitaxial films with an FWHM of 0.74° were achieved with only nine growth attempts. The optimized films exhibited atomically smooth surfaces (root-mean-square roughness of ∼0.48 nm) and uniform chemical compositions, as confirmed by Raman and energy-dispersive spectroscopy. This study shows the power of integrating XAI and BO for efficient and interpretable process optimization of thin-films.