MSE Talks: Arun Kumar Mannodi Kanakkithodi

Speaker: Dr. Arun Kumar Mannodi Kanakkithodi

Title: Semiconductor Discovery using Multi-Fidelity DFT-ML and Crystal Graphs

Abstract: Challenges of environmental pollution, global energy shortage, and over-reliance on fossil fuels can be addressed by innovation in solar technology, such as new absorbers for increasing solar cell efficiency and improved photocatalysts for hydrogen production and CO2 reduction. Novel semiconductors that show bulk stability, promising optoelectronic properties, defect tolerance, and suitable dopability, are desired as substitutes for current candidates used in these applications, but the atom-composition-structure space of potential materials is practically infinite and not conducive to brute-force experimentation necessitates the use of data-driven strategies combining large computational datasets and state-of-the-art machine learning (ML), prior to experimental validation and discovery. Provided a sufficiently diverse dataset (typically hundreds or low thousands of data points) of accurate property estimates can be generated from density functional theory (DFT), and materials can be suitably encoded using chemical “descriptors”, DFT-ML models can be trained to ensure that potentially millions of new predictions could be made at DFT accuracy at a mere fraction of the cost of running even a single DFT computation. Such DFT-ML models can be trained synergistically within a multi-fidelity framework by combining data from different functionals (that differ in computational expense and accuracy) with collaborative experiments, ultimately leading to high-fidelity selection of the “needle in the haystack”, i.e., the best candidates out of infinitely many. In this talk, I will talk about different ongoing projects in the Mannodi research group at Purdue University that use multi-fidelity ML, combining DFT and experimental data, to accelerate the prediction of electronic, optical, and defect properties of semiconductors. I will further discuss the application of crystal graph-based neural networks to improve the screening of low energy defective structures in binary and ternary semiconductors, as well as to drive the discovery of novel halide perovskite crystal structures with desirable properties for solar absorption. Finally, I will talk about advancing open-science and education via datasets and tools created as part of nanoHUB, a nanotechnology repository housed at Purdue.

Bio: Arun Mannodi Kanakkithodi is an assistant professor in the department of Materials Engineering at Purdue University. He received his PhD in and Engineering from the University of Connecticut in 2017 and worked as a postdoctoral researcher at Argonne National Laboratory from 2017 till 2020. His research primarily involves applying first principles simulations and methods rooted in data science and machine learning for materials design. He is a contributor to and co-organizer of machine learning resources and hands-on workshops for nanoHUB, and a regular organizer of materials informatics tutorials at Materials Research Society (MRS) fall and spring meetings. Arun is a recipient of the 2020 Distinguished Young Investigator award from Argonne National Lab, the 2023 Functional Materials Division (FMD) Young Leaders Professional Development Award from The Minerals, Metals & Materials Society (TMS), and a 2023 DOE Solar Energy Technology Office (SETO) Small Innovation Projects in Solar (SIPS) awardee.