Image to Properties: Extracting Atomic Structure Information from Band Dispersion Images of Semiconductor Heterostructures Using Machine Learning

Abstract: The atomic environments of semiconductor heterostructures can be highly varied as various structural imperfections, lattice mismatch and non-uniform strain environments are generally present. The computational costs of first-principles modeling techniques make it challenging to fully explore how atomic environments tune the electronic bands of heterostructures. We present a ML-assisted first-principles modeling framework that establishes a direct relationship between the atomic environments and the electronic bands of semiconductor heterostructures. The framework combines a forward and a reverse model: The forward model predicts how the atomic environments tune electronic bands; The reverse learning model extracts information about the atomic environments that is associated with an input band structure image, such as the ones obtained with angle-resolved photoemission spectroscopy. We demonstrate the framework using silicon/germanium-based superlattices and heterostructures. Our framework offers a physics-informed approach to designing heterostructures for new phenomena and device possibilities for diverse technologies, going beyond trial-and-error approaches.