Exploring new useful phosphors by combining experiments with machine learning.

Abstract

New phosphors are consistently in demand for advances in solid-state lighting and displays. Conventional trial-and-error exploration experiments for new phosphors require considerable time. If a phosphor host suitable for the target luminescent property can be proposed using computational science, the speed of development of new phosphors will significantly increase, and unexpected/overlooked compositions could be proposed as candidates. As a more practical approach for developing new phosphors with target luminescent properties, we looked at combining experiments with machine learning on the topics of emission wavelength, full width at half maximum (FWHM) of the emission peak, temperature dependence of the emission spectrum (thermal quenching), new phosphors with new chemical composition or crystal structure, and high-throughput experiments.