| Aug 21, 2025 |
AI speeds discovery of atomic defects in 2D materialsA deep learning model identifies atomic-scale defects in MoS2 with 95% accuracy, offering a faster route to quality control and quantum material research.(Nanowerk News) Researchers from the Changchun Institute of Optics, Fine Mechanics and Physics (CIOMP) at the Chinese Academy of Sciences have developed an artificial intelligence system that automatically identifies atomic-scale defects in molybdenum disulfide (MoS₂), a promising two-dimensional semiconductor for advanced electronics. The study, published in Molecules ("Point Defect Detection and Classification in MoS2 Scanning Tunneling Microscopy Images: A Deep Learning Approach"), combines machine learning with quantum simulations to streamline defect analysis—a process that has traditionally been slow and error-prone. |
| Defects in 2D materials play a decisive role in their electronic and optical behavior, influencing everything from device efficiency to quantum properties. Until now, pinpointing and classifying these irregularities in scanning tunneling microscopy (STM) images required painstaking manual inspection. |
| The research team addressed this challenge with a hybrid pipeline that pairs the Segment Anything Model (SAM) for automatic defect localization with a convolutional neural network (CNN) for classification. By applying noise-reduction and contrast-enhancement techniques to STM images before analysis, the model was able to distinguish between sulfur vacancies, interstitial atoms, and Moiré patterns with 95.06% accuracy—despite being trained on a relatively small dataset of 198 augmented images. |
| To validate the results, the team performed density functional theory (DFT) calculations, confirming that sulfur vacancies create localized electronic states consistent with the features identified by the AI. Compared to conventional methods such as OpenCV, the new approach proved more reliable in handling limited data and subtle structural variations. |
| Beyond improving efficiency, this framework provides a scalable pathway for defect analysis across a wide range of two-dimensional materials. The authors suggest that the method could be applied to semiconductor quality control, defect engineering in quantum devices, and surface science studies where data scarcity is a persistent obstacle. |
| Future work will focus on expanding the dataset to include additional defect types and extending the approach to other layered materials. By integrating automated imaging with physics-based modeling, this research marks a step toward intelligent, high-throughput characterization at the atomic scale. |
| Source: Changchun Institute of Optics, Fine Mechanics And Physics (Note: Content may be edited for style and length) |
