When
Tuesday, April 21, 2026, at 4:00 p.m.
Ishraque Zaman Borshon
PhD Candidate
Department of Aerospace and Mechanical Engineering
University of Arizona
"Physics-Informed Deep Learning for Atomic-Scale Electron Microscopy of Battery and High-Entropy Materials"
AME Lecture Hall, Room S202 | Zoom link
Abstract: Quantitative interpretation of atomic-resolution electron microscopy remains a major bottleneck in materials research, especially for chemically complex systems where manual analysis is slow and subjective. This seminar presents physics-informed computational approaches that combine density functional theory (DFT), atomistic structure generation, multislice image simulation and data-driven analysis to extract physically meaningful information from TEM and STEM images. For the learning-based studies, these methods generate physically realistic synthetic training data without manually labeled experimental images. Several examples of applying various machine learning models to TEM images processing will be presented. The first example uses a fully convolutional network trained on DFT- and evolutionary algorithm-generated STEM images to predict atomic column heights and elemental distributions in (Mn, Fe, Ni, Cu, Zn)3O4 high-entropy oxide nanoparticles. The second uses a hierarchical transformer trained on simulated cryo-TEM images to segment grains and grain boundaries in the solid-electrolyte interphase of lithium-metal batteries. The third integrates in situ liquid-cell TEM with DFT and nudged elastic band calculations to probe elements diffusion in the grain boundary of Pt–Cu–Ir–Ni multi-principal element alloys. Together, these studies show how physics-informed computation can both generate training data for learning-based microscopy and provide a mechanistic context for experiment, enabling more quantitative and physically grounded characterization of complex materials.
Bio: Borshon is a PhD candidate and graduate researcher at the University of Arizona, where he develops simulation-informed machine learning and computational frameworks for atomic-scale materials characterization in the Energy Storage and Conversion (ESC) Laboratory under the supervision of Vitaliy Yurkiv. His research focuses on simulation-informed machine learning for electron microscopy, spanning energy storage materials, high-entropy oxides and multi-principal element alloys. He holds an MTech in energy systems engineering from the Indian Institute of Technology Bombay (India) and a BTech in mechanical engineering from National Institute of Technology Kurukshetra (India). His work has been published in npj Computational Materials, Advanced Materials Interfaces and Computational Materials Science.