Accelerating the combinatorial optimization process for phosphor materials by Bayesian optimization

Abstract

Abstract We investigated the acceleration of the combinatorial optimization process for phosphor materials using a machine learning method based on Bayesian optimization. Combinatorial pulsed laser deposition can be used to create a library of single-crystalline films with varying chemical compositions. However, the systematic evaluation of the target functional properties requires a long measurement time, impairing rapid material screening. In this study, Bayesian optimization was applied to sequential measurements of the photoluminescence (PL) properties of EuxY2−xO3 films to accelerate the combinatorial high-throughput evaluation. Although a conventional combinatorial PL evaluation of a binary composition-gradient film is composed of a sequential measurement of 80 points, the autonomous PL mapping technique based on Bayesian optimization drastically reduced the measurement points to only six points, demonstrating that the optimum chemical composition can be identified in a shorter experimental time.