2026/02/05 by Yuki K. Wakabayashi, Takuma Otsuka, Yoshiharu Krockenberger +1
Engineering · Materials Science · #Bayesian inference #Bayesian optimization #Bayesian probability #Electronic and Structural Properties of Oxides #Machine Learning in Materials Science #Oxide #Semiconductor materials and devices #Weighting
paper · pdf · doi:10.1038/s41427-026-00668-1
openalex publication_date 2026/07/15 · openalex created_date 2026/07/16 · openalex updated_date 2026/08/05
We present a physics-informed Bayesian optimization (PIBO) with a concise modification to its acquisition function to incorporate the prior physics knowledge. Specifically, this method multiplies the expected improvement (EI) by a weight encoding prior crystal growth physics. When applied to LaAlO3 molecular-beam epitaxy, the weighting function defines a flat stoichiometric window and penalizes off-window proposals, thereby steering the optimization toward physically plausible regions while maintaining controlled exploration. In a closed-loop optimization, the weighted EI constrains the search toward stoichiometric regions while retaining sufficient flexibility to explore neighboring conditions, eventually identifying an optimum slightly beyond the stoichiometric window. Within only 15 growth runs, the lattice constant of the grown LaAlO3 film converged to the bulk value and the optimized film showed an increase in X-ray diffraction peak intensity ten times larger than that of the unoptimized film, indicating efficient and rapid identification of growth conditions that yield a bulk-like lattice constant and improved crystallinity. We demonstrate the efficacy of our weighted EI method through comparison with standard BO using bare EI where the bare EI required 43 runs to reach the best condition. Because physics knowledge is incorporated solely through the weighting function, the approach requires only minimal modification to standard BO workflows and provides a practical route to physics-guided optimization in thin-film growth. Recent advances in artificial intelligence have revolutionized materials discovery, particularly in optimizing thin-film growth processes. This study introduces a physics-informed Bayesian optimization (PIBO) framework to enhance the epitaxial growth of complex oxides, focusing on achieving precise stoichiometric control. The authors developed a method that integrates crystal growth physics into the Bayesian optimization algorithm, specifically targeting the LaAlO₃ (LAO) thin-film growth via molecular beam epitaxy (MBE). By incorporating a stoichiometry-based weighting function, the framework efficiently guides experiments toward optimal growth conditions. The results show that after only 15 growth cycles, the LAO films achieved a lattice constant matching the bulk value, with a tenfold increase in X-ray diffraction peak intensity, indicating improved crystalline quality. This approach not only accelerates the optimization process but also enhances structural quality, offering a promising direction for AI-assisted materials synthesis. This summary was initially drafted using artificial intelligence, then revised and fact-checked by the author. Physics-informed Bayesian optimization accelerates oxide thin-film growth by multiplying the expected-improvement acquisition function by a stoichiometry-based weight. Applied to LaAlO3 molecular-beam epitaxy, this soft constraint steers experiments toward physically plausible La/Al ratios while retaining controlled exploration. The proposed method identified the optimized condition within 15 runs, whereas standard Bayesian optimization using bare expected improvement required 43 runs, yielding films with a bulk-like lattice constant and more than tenfold higher X-ray diffraction intensity than unoptimized films.