(126A) Data-driven decision making for autonomous materials synthesis

Targeted solid-state synthesis plays a major role in driving materials development. However, synthesizing novel compounds with high purity is challenging. Reaction products generally vary with changes in precursors, heating profile, atmosphere, and quality of mixing. Exploring all possible combinations of these variables in a trial-and-error fashion requires many time-consuming and costly experiments, slowing materials discovery. We have therefore developed a more systematic approach to materials synthesis whereby precursors and conditions are suggested to maximize the driving force associated with formation of the target phase. In cases where these experiments fail, our algorithm learns from the synthesis results and updates its ranking of precursors and conditions accordingly using pairwise reaction analysis.

To design and test the decision-making algorithm, we have prepared an unprecedented dataset containing results from 188 solid-state synthesis experiments targeting YBa2Cu3O7, a well-known high-temperature superconductor. With this data, we show that our physics-informed algorithm outperforms state-of-the-art black-box optimization methods. Considering its success, we plan to integrate the decision-making algorithm with an autonomous synthesis platform known as the “A-Lab,” where robots will perform experiments in high-throughput to discover new materials at accelerated rates.