Rapid morphology characterization of two-dimensional TMDs and lateral heterostructures based on deep learning

TL;DR

The paper addresses the need for fast, accurate characterization of MoS2-MoSe2 lateral heterostructures and MoS2 morphology. It presents a YOLO-based deep learning pipeline (notably YOLOv11m-seg) for detection and instance segmentation, augmented by transfer learning to adapt across materials. The approach yields high accuracies (heterostructure >95%, shape 94.67%, thickness 97.3%) and supports real-time, local analysis integrated with optical microscopy, including robust generalization under varied imaging conditions. This work offers a practical, scalable tool for rapid 2D material characterization with potential to accelerate synthesis optimization and device development.

Abstract

Two-dimensional (2D) materials and heterostructures exhibit unique physical properties, necessitating efficient and accurate characterization methods. Leveraging advancements in artificial intelligence, we introduce a deep learning-based method for efficiently characterizing heterostructures and 2D materials, specifically MoS2-MoSe2 lateral heterostructures and MoS2 flakes with varying shapes and thicknesses. By utilizing YOLO models, we achieve an accuracy rate of over 94.67% in identifying these materials. Additionally, we explore the application of transfer learning across different materials, which further enhances model performance. This model exhibits robust generalization and anti-interference ability, ensuring reliable results in diverse scenarios. To facilitate practical use, we have developed an application that enables real-time analysis directly from optical microscope images, making the process significantly faster and more cost-effective than traditional methods. This deep learning-driven approach represents a promising tool for the rapid and accurate characterization of 2D materials, opening new avenues for research and development in material science.

Figures (8)

• Figure 1: Experimental workflow and deep learning architecture.(a) Key steps: Material preparation $\rightarrow$ structural characterization $\rightarrow$ dataset annotation $\rightarrow$ YOLOv11 training $\rightarrow$ real-time deployment on optical microscopy systems. (b) YOLOv11 architecture schematic showing data input (left), model architecture (center), and inference results (right).
• Figure 2: Performance comparison of heterostructure characterization models. Among them, box refers to boundary box detection, and mark refers to mask segmentation. The left image shows mAP50, and the right image shows mAP50-90.
• Figure 3: Characterization of lateral heterostructures of MoS$_2$-MoSe$_2$ and YOLO model inference results. (a) Optical images of the heterostructures. (b) Raman mapping of the heterostructures. (c) Heterostructures predicted based on the YOLO model. (d) Raman spectra, where the measurement areas are marked in red, blue, and black in (a).
• Figure 4: Identification results of MoS$_2$ with different shapes: (a) triangle, (b) hexagon, (c) dendritic. (d) Optical image inference results of MoS$_2$.
• Figure 5: Training results of the MoS$_2$ thicknesses dataset. (a) Loss curves, where the solid line indicates the training set loss and the dashed line represents the validation set loss. (b) mAP50-90 curves, with the solid line showing the mAP50-90 for the boundary box detection and the dashed line for mask segmentation. In (a) and (b), the red line represents the original model, and blue line denotes the transfer learning model. (c) YOLOv11m-seg model inference results. (d) Inference results of YOLOv11m-seg model after transfer learning. Differences between (c) and (d) are highlighted with red boxes.