Phase diagram construction supported by artificial intelligence

Many researchers may first want to understand the phase diagram to develop new materials. However, it is necessary to conduct many experiments to create a phase diagram from scratch, which is a very time-consuming process. We have developed a method for efficiently suggesting “next experimental points” using artificial intelligence technique, in order to determine a detailed phase diagram with the fewer number of experiments as possible. In this method, uncertain points in the phase diagram are found by machine learning, and these points are selected as candidates for the next experiment. Validation against well-known phase diagrams shows that the number of experiments to determine phase diagrams can be reduced, and new phases which are not found at the initial state can also be found efficiently. We have applied the method to the creation of new phase diagrams for the deposition of Zn-Sn-P films and for a crosslinked polymer.