Accelerated Process Development of YBa2Cu3O7 film with Bayesian Optimization Augmented by Transferred Data

Efficient process optimization is a key to the development of YBa₂Cu₃O₇ films, whose properties are influenced by many factors such as substrate, nanostructure, and composition. The simplest optimization is exhaustive search, which requires many experiments. Bayesian optimization is more efficient but requires initial data. Here we propose the Bayesian optimization augmented by domain-aligned sampling, data transfer, and peak screening (transfer-based Bayesian optimization), which efficiently acquires the initial data. In this method, the preliminary data in a system with a consistent process nature is used as the initial data. The preliminary data of the critical current density (J_c) were obtained in the YBa₂Cu₃O₇/SrTiO₃ by exhaustive search, and the J_c of YBa₂Cu₃O₇/LaAlO₃ was optimized. A Gaussian process regression model was constructed from the preliminary data of YBa₂Cu₃O₇/SrTiO₃ and the domain-aligned sampling data of the YBa₂Cu₃O₇/LaAlO₃. The initial local maxima were included as candidates for the global maximum in the peak screening to avoid overlooking potentially high J_c. The transfer-based Bayesian optimization found an optimal J_c as high as that by the exhaustive search with a smaller number of iterations. Thus, the optimal property was obtained efficiently by transfer-based Bayesian optimization, making the optimization of coated conductors and advanced materials more efficient.