Characterizing Atomic Scale Strain Variations in 2D Materials with a Convolutional Neural Network

Local measurements of electronic properties, enabled by scanning tunneling microscopy (STM), have led to a better understanding of 2D materials such as graphene and MoC2. The impact of local strain on these properties has been of great theoretical and experimental interest. However, correlating these electronic properties with local strain has proven challenging, as directly measuring strain - determining picometer-scale offsets of atoms from their unstrained locations – is non-trivial. Here we present the development of a convolutional neural network to assign atom locations and local strain values to STM topographies of hexagonal lattices. Our method enables the direct visualization of atomic-scale strain variations in Moiré patterns, doping impurities, and other similar structures. We will discuss future applications as well as improvement of model accuracy through increasingly realistic simulated training sets and better post-processing techniques.