Two-dimensional (2D) materials and their in-plane and out-of-plane (i.e., van der Waals, vdW) heterostructures are promising building blocks for next-generation electronic and optoelectronic devices. Since the performance of the devices is strongly dependent on the crystalline quality of the materials and the interface characteristics of the heterostructures, a fast and nondestructive method for distinguishing and characterizing various 2D building blocks is desirable to promote the device integrations. In this work, based on the color space information on 2D materials' optical microscopy images, an artificial neural network-based deep learning algorithm is developed and applied to identify eight kinds of 2D materials with accuracy well above 90% and a mean value of 96%. More importantly, this data-driven method enables two interesting functionalities: (1) resolving the interface distribution of chemical vapor deposition (CVD) grown in-plane and vdW heterostructures and (2) identifying defect concentrations of CVD-grown 2D semiconductors. The two functionalities can be utilized to quickly identify sample quality and optimize synthesis parameters in the future. Our work improves the characterization efficiency of atomically thin materials and is therefore valuable for their research and applications.
Keywords: artificial neuron networks; defect concentration; fast characterization; heterostructures; two-dimensional materials.