Mobile Navigation

Chemical Engineering

View Comments

A breakthrough in artificial intelligence-enabled materials discovery

| By Mary Page Bailey

A groundbreaking artificial intelligence (AI) algorithm, dubbed CAMEO (Closed-Loop Autonomous System for Materials Exploration), rapidly identified a potentially useful new material — a germanium-antimony-tellurium alloy (Ge4Sb6Te7) that is optimized for phase-change applications in data storage and photonic-switching devices. Tasked with evaluating 177 different materials, CAMEO performed 19 experimental cycles over ten hours, representing a nearly tenfold reduction in the time required compared to a scientist running the experiments in a laboratory. CAMEO is a self-learning AI, accessing and processing data from a combinatorial library of material compositions, using prediction and uncertainty to determine which experiments to run next. It then facilitates the experimentation procedures, such as x-ray diffraction (XRD), and collects the data. At this point, CAMEO can request additional information, such as data on a material’s crystal structure, before running the next experiment. CAMEO contains knowledge related to previous simulations and laboratory experiments, equipment operation and physical concepts, such as phase mapping, or the behavior of atomic arrangement with changing chemical composition. Since the AI runs unsupervised, it enables scientists to more easily work remotely, an ability that is especially valuable for experiments involving potentially toxic chemicals or contagious viral media.

For CAMEO’s recent breakthrough, researchers set out to determine the Ge-Sb-Te alloy that exhibited the largest difference in optimal contrast between the crystalline and amorphous states. In terms of optical contrast, the new alloy is twice as effective as a commonly used phase-change material, Ge2Sb2Te5. To achieve this discovery, CAMEO focused on the material-structure-properties relationship of various crystalline materials, effectively tracking the structural origins of a material’s function. Moving forward, researchers plan to make the algorithm capable of solving more complex problems. Several institutions contributed to the development and demonstration of CAMEO’s recent achievement, including: the National Institute of Standards and Technology (NIST; Gaithersburg, Md.; www.nist.gov); Stanford University (www.stanford.edu); the University of Maryland (www.umd.edu); the University of Washington (www.washington.edu); and the U.S. Department of Energy (DOE; Washington, D.C.; www.energy.gov).