Development of a Flux-Method Process Informatics System and Its Application in Growth Control for Layered Perovskite Ba5Nb4O15 Crystals

Composition selection and crystal-growth control are the most important issues in material development. Crystal growth using a molten salt as the solvent, which is otherwise known as the flux method, is expected to be one of the key technologies for developing crystalline materials, because this method can grow single crystals of inorganic materials with control of the crystal habit and size. In the flux method, the number of combinations of experimental conditions usually exceeds 10,000 owing to the wide variety of experimental variables, such as solvents and heating conditions. Due to such large combinations, the conventional trial-and-error approach based on hypothesis testing would take several years to obtain the optimal crystal, and the crystal control region is also limited. As a result, the development of crystalline materials is risky, and their use is limited. To overcome this issue, we propose a novel crystal growth approach called flux-method process informatics (FPI). The FPI approach generates a data-driven experimental proposal based on machine learning predictions. More specifically, regression analyses are carried out using the experimental process as the explanatory variable and the crystal shape/size as the objective variable. In this study, FPI was applied to the layered perovskite-type oxide Ba5Nb4O15. It was found that the experimental efficiency of the FPI approach was five times higher than that of humans, and it was able to quantitatively control the shape of low-aspect-ratio crystals, which is usually difficult for anisotropic structures. These achievements provide new insights into the rapid development of high-performance crystalline materials.