Machine-Learning Analysis of the Electronic Structure of Atomically Thin MnTe Films from Scanning Tunneling Spectroscopy

Abstract

Layered two-dimensional (2D) van der Waals magnetic crystals based on 3d transition metals and chalcogenides are typically synthesized by chemical vapor transport, while establishing epitaxial, layer-by-layer growth under vacuum remains essential for device applications. Here, we study Mn–Te as a model system motivated by its intriguing topological and magnetic properties. Our findings using scanning tunneling microscopy and spectroscopy (STM/STS) combined with density functional theory calculations reveal that film growth initially results in a mixture of Te-, Mn-, and MnTe-related metastable phases. Upon post-annealing up to ～650 K, these states transform into a thermodynamically stable MnTe alloy phase, yielding an atomically flat 2D surface. This work further demonstrates that integrating machine learning with STM/STS enables reliable chemical classification and identification of local density of states in structurally complex surfaces.