NIST and University of Maryland use AI to accelerate materials discovery

NIST researcher Gilad Kusne and colleagues at the University of Maryland used machine learning to investigate materials based on their composition and temperature. They applied an AI algorithm called the Autonomous Materials Search Engine (AMASE) to autonomously direct X-ray analysis across a combinatorial wafer containing hundreds of thousands of different ratios of tin and bismuth.

Each point on the wafer represents a distinct material composition, and AMASE analyzes each point to infer properties under varying conditions. The team described this combinatorial library as analogous to selecting from a large collection of books, where each point corresponds to a different 'book' to read.

According to the announcement, AMASE can extrapolate beyond the available dataset to predict material behaviors, a capability the researchers note is not inherent to many AI systems. This predictive capacity is intended to accelerate discovery by reducing the need for exhaustive experimental measurements.

The method is positioned to support applications in automated manufacturing and quantum materials, where understanding material behavior under diverse conditions is critical.