ADVANCED MATERIALS & PROCESSES | JANUARY/FEBRUARY 2024 10 MACHINE LEARNING | AI GOOGLE’S AI LAB AMPS UP MATERIALS DATABASE The Materials Project, an open- access database founded at the DOE’s Lawrence Berkeley National Laboratory in 2011, computes the properties of both known and predicted materials. Google DeepMind, Google’s artificial intelligence (AI) lab, is now contributing nearly 400,000 new compounds to the project. The dataset includes each compound’s crystal structure and formation energy. To generate the new data, Google DeepMind developed a deep learning tool called Graph Networks for Materials Exploration (GNoME). Researchers trained GNoME using workflows and data that were developed over a decade by the Materials Project and improved the algorithm through active learning. Researchers ultimately produced 2.2 million crystal structures, including 380,000 stable ones they are adding to the Materials Project, making them potentially useful in future technologies. Some of the computations from GNoME were used alongside data from the Materials Project to test A-Lab, a facility at Berkeley Lab where AI guides robots in making new materials. Over 17 days of independent operation, A-Lab successfully produced 41 new compounds out of an attempted 58. To make the compounds, A-Lab’s AI created new recipes by combing through scientific papers and using active learning to make adjustments. Data from the Materials Project and GNoME were then used to evaluate predicted stability. The Materials Project is the most widely used open-access repository of information on inorganic materials in the world. The database holds millions of properties on hundreds of thousands of structures and molecules, and more than 400,000 people are registered as site users. The contribution from Google DeepMind is the biggest addition of structure-stability data since the Materials Project began. lbl.gov. MACHINE LEARNING FOR ENERGY STORAGE Researchers at the DOE’s Oak Ridge National Laboratory (ORNL) designed a powerful supercapacitor material that stores four times more AI-guided robots created more than 40 new materials predicted by the Materials Project. Courtesy of Berkeley Lab. energy than today’s best commercial material. Commercial supercapacitors have an anode and cathode that are separated and immersed in an electro- lyte. Double electrical layers reversibly separate charges at the interface between the electrolyte and the carbon. The materials of choice for making electrodes for supercapacitors are porous carbons, as the pores provide a large surface area for storing electrostatic charge. The ORNL study used machine learning to guide discovery of the promising material. Colleagues from the University of California, Riverside built an artificial neural network model and trained it to set a clear goal: Develop a “dream material” for energy delivery. The team designed an extremely porous doped carbon that would provide huge surface areas for interfacial electrochemical reactions. Then they synthesized the new material, an oxygen-rich carbon framework for storing and transporting charge. The synthesized material had an impressive capacitance of 611 farads per gram. The data-driven approach allowed the scientists to achieve in three months what would typically take a year or more. ornl.gov. Graphic depicts machine learning finding an energy storage material. Carbon framework shown in black, functional groups with oxygen in pink, and nitrogen in turquoise. Courtesy of ORNL.
RkJQdWJsaXNoZXIy MTYyMzk3NQ==