Rapid morphology characterization of two-dimensional TMDs and lateral heterostructures based on deep learning
Junqi He, Yujie Zhang, Jialu Wang, Tao Wang, Pan Zhang, Chengjie Cai, Jinxing Yang, Xiao Lin, Xiaohui Yang · 2025
Read Rapid morphology characterization of two-dimensional TMDs and lateral heterostructures based on deep learning by Junqi He, Yujie Zhang, Jialu Wang, Tao Wang, Pan Zhang, Chengjie Cai, Jinxing Yang, Xiao Lin, Xiaohui Yang (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-00470.txt.
Introduction
Junqi He and co-authors teach a vision model to inspect two-dimensional materials under the microscope in a 2025 paper. Atom-thin crystals and their stitched heterostructures have prized properties, but characterizing flakes by hand is slow. Their deep-learning system targets molybdenum sulfide flakes and its lateral junctions with molybdenum selenide, judging shape and thickness from optical images.
Detection accuracy passed ninety-four percent. Transfer learning from related materials improved performance further, and the model resisted interference and generalized across scenarios. The team wrapped it in an application that analyzes microscope images in real time. They pitch the tool as a faster, cheaper stand-in for traditional characterization, ready to speed routine materials screening in the lab.
Catalog record
| CREATOR | Junqi He, Yujie Zhang, Jialu Wang, Tao Wang, Pan Zhang, Chengjie Cai, Jinxing Yang, Xiao Lin, Xiaohui Yang |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 25 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2503-00470
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Rapid morphology characterization of two-dimensional TMDs and lateral heterostructures based on deep learning Junqi He, Yujie Zhang, Jialu Wang, Tao Wang, Pan Zhang, Chengjie Cai, Jinxing Yang, Xiao Lin, Xiaohui Yang 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 MoS 2 -MoSe 2 lateral heterostructures and MoS 2 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. keywords 2D material, TMDs, lateral heterostructure, deep learning, instance segmentation, morphology characterization † † affiliation: These authors contributed equally to this work. † † affiliation: These authors contributed equally to this work. † † affiliation: Department of Physics, China Jiliang University, Hangzhou 310018, P. R. China † † affiliation: Hangzhou Key Laboratory of Quantum Matter, School of Physics, Hangzhou Normal University, Hangzhou 311121, China † † affiliation: ZJU-Hangzhou Global Scientific and Technological Innovation Center, College of Integrated Circuits, Zhejiang University, Hangzhou, 311215, China. † † affiliation: Department of Physics and Texas Center for Superconductivity, University of Houston, Houston, TX 77204, USA † † affiliation: Department of Physics, China Jiliang University, Hangzhou 310018, P. R. China † † affiliation: Key Laboratory for Quantum Materials of Zhejiang Province, Department of Physics, School of Science, Westlake University, Hangzhou 310030, P. R. China † † affiliation: Institute of Natural Sciences, Westlake Institute for Advanced Study, Hangzhou 310024, P. R. China † † email: [email protected] † † affiliation: Department of Physics, China Jiliang University, Hangzhou 310018, P. R. China † † affiliation: Zhejiang Province Key Laboratory of Quantum Technology and Device, Zhejiang University, Hangzhou 310027, P. R. China † † affiliation: Key Laboratory for Quantum Materials of Zhejiang Province, Department of Physics, School of Science, Westlake University, Hangzhou 310030, P. R. China † † suppinfo: 1 Introduction Two-dimensional (2D) materials have attracted significant attention due to their excellent mechanical, electrical, thermal, and optical properties, making them ideal candidates for next-generation technologies Novoselov et al. 2004 ; Novoselov et al. 2005 ; Chhowalla et al. 2013 ; Zeng et al. 2018 ; Li et al. 2020 ; Rickhaus et al. 2020 . Among these, transition metal dichalcogenides (TMDs) such as MoS 2 and MoSe 2 stand out for their tunable bandgaps and strong light-matter interactions Zhang et al. 2014 ; Chhowalla et al. 2013 ; Ganatra and Zhang 2014 ; Li and Zhu 2015 ; Eftekhari 2017 ; Wu et al. 2020 . The properties of 2D materials are highly sensitive to thickness and morphology Jaramillo et al. 2007 ; Mas-Balleste et al. 2011 ; Gupta et al. 2015 ; Tang et al. 2021 ; Chhowalla et al. 2013 . For example, single-layer MoS 2 has a direct bandgap, making it suitable for optoelectronics, while multilayer MoS 2 switches to an indirect bandgap, altering its electronic behavior Chhowalla et al. 2013 ; Ganatra and Zhang 2014 ; Li and Zhu 2015 . The performance of 2D materials also depends on shape and size, with large-area flakes showing superior charge transport properties, while nanostructures with edge defects or vacancies exhibit enhanced catalytic activity Lin et al. 2018 ; Gao et al. 2018 ; Jaramillo et al. 2007 ; Zhang et al. 2014 ; Ye et al. 2016 ; Tang et al. 2021 . The inherent limitations of individual 2D materials have prompted extensive research into heterostructures, which combine distinct materials to enhance or tailor their properties while preserving the beneficial characteristics of the constituent components Liu et al. 2019 ; Zhang et al. 2021 ; Wang et al. 2019 . These heterostructures are typically classified as lateral or vertical, with lateral heterostructures offering advantages such as reconfigured band structures, reduced interlayer scattering, and superior optoelectronic performance Wang et al. 2019 ; Liu and Hersam 2018 ; Schneider et al. 2024 ; Behranginia et al. 2017 . Such attributes make lateral heterostructures particularly promising for next-generation technologies in electronics, optoelectronics, and flexible devices Liu et al. 2023 ; Hoang et al. 2022 ; Liu and Hersam 2019 . TMD heterostructures, like MoS 2 -MoSe 2 , have been particularly attractive for semiconductor applications due to their remarkable optoelectronic and thermoelectric properties Duan et al. 2014 ; Chen et al. 2017 ; Barik and Pal 2022 ; Darboe et al. 2022 ; Bellus et al. 2018 ; Jia et al. 2019 ; Xu et al. 2024 . Due to the diverse physical properties, 2D materials and their heterostructures require highly accurate and efficient characterization techniques. Traditional techniques such as atomic force microscopy (AFM) and Raman spectroscopy are effective but time-consuming Liu and Hersam 2018 ; Liang and Meunier 2014 . In contrast, artificial intelligence, particularly machine learning and deep learning, offers data-driven solutions for efficient material characterization Yang and Yao 2020 ; Li et al. 2019 ; Masubuchi and Machida 2019 ; Ryu et al. 2022 ; Lu et al. 2024 ; Si et al. 2023 ; Choudhary et al. 2022 ; Dong et al. 2022 . Models like CNNs, including Mask R-CNN, U-Net, and YOLO, have been used for analyzing optical images of 2D materials, enabling accurate identification of layer thickness, defects, and heterostructures He et al. 2024 ; Masubuchi et al. 2020 ; Dong et al. 2021 ; Shi et al. 2022 ; Yang et al. 2024 ; Zhu et al. 2022 . These methods reduce time and cost compared to traditional measurements and also offer the potential for characterizing physical properties. In the identification of heterostructures, support vector machine (SVM) have been used to distinguish graphene-MoS 2 heterostructures, while artificial neural network (ANN) and conditional generative adversarial network (cGAN) have been applied to identify WS 2 -MoS 2 heterostructures Lin et al. 2018 ; Zhu et al. 2022 ; Bhawsar et al. 2024 . Despite their success, current approaches face challenges in portability and require extensive dataset preprocessing, with inference results often needing manual validation. In this study, we propose a deep learning-based approach for the rapid and accurate characterization of MoS 2 flakes and MoS 2 -MoSe 2 lateral heterostructures. Using YOLO models (YOLOv8 and YOLOv11) for object detection and instance segmentation, our method surpasses traditional techniques with faster inference, smaller model size, higher accuracy, and reduced data requirements. We synthesized the materials via chemical vapor deposition (CVD) and characterized them using Raman spectroscopy. An annotated optical image dataset was created, achieving over 96% recognition accuracy for heterostructures. We also observed that synthesis conditions significantly affect MoS 2 morphology, which in turn influences its properties. We developed a shape recognition method using deep learning, achieving over 94% accuracy in detecting and quantifying