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 | M A Y / J U N E 2 0 2 2 6 MACHINE LEARNING | AI MACHINE LEARNING AIDS RARE EARTHS RESEARCH Scientists at the DOE’s Ames Laboratory and Texas A&M University trained a machine learning (ML) model to determine the stability of rare earth compounds. The team used the upgraded Ames Laboratory Rare Earth database (RIC 2.0) and high-throughput density functional theory (DFT) to build the foundation for their model. High-throughput screening allows researchers to test hundreds of models quickly, while DFT is a quantum mechanical method used to investigate thermodynamic and electronic properties. Based on this collection of information, the new ML model uses regression learning to assess the phase stability of different compounds. Ames scientist Prashant Singh says the material analysis relies on a discrete feedback loop in which the AI/ ML model is updated using the new DFT database, which is based on real-time structural and phase information obtained from the experiments. The process ensures that information is carried from one step to the next and reduces the chance of making mistakes. Project supervisor Yaroslav Mudryk notes that the framework was designed to explore rare earth compounds due to their technological importance, but its application is not limited to rare earths research. The same approach could be used to train an ML model to predict magnetic properties of compounds, develop new process controls for manufacturing, and optimize mechanical behaviors. ameslab.gov. SELF-DRIVING LAB STUDIES NANOCRYSTALS A research team from North Carolina State University and the University at Buffalo developed a “self- driving lab” that uses artificial intelligence (AI) and fluidic systems to gain knowledge regarding metal halide perovskite (MHP) nanocrystals. These nanocrystals are an emerging class of semiconductor materials that have potential for use in printed photonic devices and energy technologies due to their solution processability, unique size, and composition-tunable properties. They are highly efficient, optically active materials that are under consideration for use in next-generation LEDs. Because they can be made using solution processing, they also have the potential to be made in a cost-effective way. Doping the material with varying levels of manganese can change its optical and electronic properties, such as the wavelength of light emitted, and also introduce magnetic properties. Especially noteworthy is that the new system does all of this autonomously. Specifically, its AI algorithm selects and runs its own experiments: Results from each experiment inform which experiment it will run next—and it keeps going until it understands which mechanisms control the MHP’s various properties. “In other words, we can get the information we need to engineer a material in hours instead of months,” says NC State associate professor Milad Abolhasani. While the work demonstrated in this research focuses on MHP nanocrystals, the system could also be used to characterize other nanomaterials that are made using solution processes, including a wide variety of metallic and semiconductor nanomaterials. ncsu.edu. A newmachine learning model uses regression learning to assess the phase stability of various rare earth compounds. Courtesy of Ames Laboratory. A new self-driving lab is using AI and fluidic systems to study MHP nanocrystals. Courtesy of Milad Abolhasani.
RkJQdWJsaXNoZXIy MTYyMzk3NQ==