Physics Colloquium: Combinatorial Experimentation and Machine Learning for Materials Discovery

Prof. Ichiro Takeuchi, Department of Materials Science & Engineering, Quantum Materials Center, University of Maryland

Abstract: Throughout the history of mankind, scientists and engineers have relied on the slow and serendipitous trial-and-error approach for materials discovery. In 1990s, the combinatorial approach was pioneered in the pharmaceutical industry in order to dramatically increase the rate at which new medical compounds are identified. The high-throughput concept is now widely implemented in a variety of fields in materials science. We have developed combinatorial thin film synthesis and characterization techniques in order to perform rapid survey of previously unexplored materials phase space in search of new inorganic functional materials with enhanced physical properties. Over the years, the challenges in the high-throughput approach has evolved from synthesis of large number of disparate compounds to developing quantitatively accurate rapid characterization tools to analysis and digestion of large amount of data churned out by the methodology. To address the last challenge, we are increasingly relying on machine learning techniques including pattern recognition within diffraction data to construct phase  diagrams and mining experimental databases to look for trends in materials properties for future predictions. Topics we are now pursuing include superconductors, spintronic materials, and phase change memory materials. I will also discuss our latest effort where active learning is used to design and steer the sequence of experiments in order to maximize attainable knowledge, minimize experimental resources, and as a result further speed up the materials discovery process. In this manner, we have recently demonstrated a true closed-loop autonomous materials discovery. This work is performed in collaboration with A. Gilad Kusne, V. Stanev, and A. Mehta, and is funded by NIST, DOD, and SRC.