January_2021_AMP_Digital

A D V A N C E D M A T E R I A L S & P R O C E S S E S | J A N U A R Y 2 0 2 1 7 MACHINE LEARNING | AI CREATING HIGH-ENTROPY ALLOYS WITH AI A research team at Pohang Univer- sity of Science and Technology, South Korea, developed a technique for phase prediction of high-entropy alloys (HEAs) using artificial intelligence (AI). While traditional metal alloys are made by mixing one principal element for the desired property with two or three oth- er elements, HEAs are made with equal or similar proportions of five or more el- ements. They are theoretically infinite and have exceptional mechanical, ther- mal, physical, and chemical properties, such as corrosion resistance and high strength. Researchers focused on devel- oping models of HEAs with enhanced phase prediction and explainability by using deep learning, which was applied through three perspectives: model op- timization, data generation, and param- eter analysis. In particular, the empha- sis was on building a data-enhancing model based on the conditional gener- ative adversarial network. This allowed the AI models to reflect samples of HEAs that have not yet been discovered, thus improving phase prediction accuracy compared to conventional methods. In addition, the team developed a de- scriptive AI-based HEA phase predic- tion model to provide interpretability to deep learning models, thus acting as a black box while also providing guid- ance on key parameters for creating HEAs with certain phases. www.pos- tech.ac.kr/eng. MACHINE LEARNING FOR ELECTRONIC MATERIALS Researchers at Northwestern Uni- versity’s McCormick School of Engi- neering, Evanston, Ill., developed a new computational approach to accelerate the design of materials exhibiting met- al-insulator transitions (MITs). These rare electronic materials have poten- tial to deliver faster microelectronics and quantum information systems, key technologies behind Internet of Things (IoT) devices and large data centers. The new strategy integrates techniques from statistical inference, optimi- zation theory, and compu- tational materials physics. The design of many different materials proper- ties has been accelerated with data-driven methods aided by high-throughput data generation coupled with methods like machine learning. However, these approaches have not been available for MIT materials, which are classified by their ability to reversibly switch between electrically conducting and insulating states. Most MIT models describe a sin- gle material, making model generation challenging. At the same time, conven- tional machine learning methods have shown limited predictive capability for MIT materials due to the absence of available data. The new method, called advanced optimization engine (AOE), bypasses traditional machine learning-based dis- coverymodels by using a latent variable Gaussian process modeling approach, which only requires the chemical com- positions of materials to recognize their optimum nature. This allows the Bayes- ian AOE to efficiently search for mate- rials with optimal band gap tunability and thermal stability. To validate their approach, the team analyzed hundreds of chemical combinations using densi- ty function theory-based simulations and found 12 previously unidentified compositions of complex lacunar spi- nels that showed optimal properties and synthesizability. These MIT materi- als are known to host unique spin tex- tures, a necessary feature to power fu- ture IoT devices and other resource- intensive technologies. mccormick. northwestern.edu. AI is being used to develop high-entropy alloys. The advanced optimization engine could help accelerate design of materials with metal-insulator transitions.