High-precision and rapid detection of complex defects in transferred CVD graphene enabled by machine learning algorithms

Two-dimensional (2D) materials, such as graphene, have garnered significant attention due to their wide potential applications in optical and next-generation semiconductor devices. However, before integrated into devices, these materials must be transferred from the growth substrate to a target substrate, often introducing complex defects such as wrinkles, cracks, and residues. Ensuring high-quality graphene production is essential for these applications. Despite advances in computer vision technologies, efficiently and accurately detecting transfer-induced defects in 2D materials remains challenging due to their complex morphologies. Previous studies applied machine learning (ML) to identify layer numbers of 2D materials from optical microscopy (OM) images, but few have addressed defect detection during or after transfer processes. In this study, we focus on detecting transfer-induced defects in chemical vapor deposition (CVD) grown graphene using a ML-based automated detection system. Two distinct CVD growth methods with three different transfer techniques, we were able to construct a diverse and representative dataset capturing various levels and types of transfer-induced defects in graphene. An innovative automated segmentation tool was developed to generate defect annotations and convert them into JSON format, enhancing consistency and efficiency in data preparation. OM images were used together with YOLOv7 deep learning model, combined with customized loss function optimization, to identify and quantify irregularly shaped defects. The proposed approach achieved approximately 10 % improvement in detection accuracy through improved dataset splitting and loss function tuning. Automated segmentation reduced manual annotation time by 75 %, while maintaining high consistency in defect labeling. These results demonstrate that the system significantly enhances graphene defect detection, streamlines quality control, and improves production efficiency, supporting the development of high-performance graphene-based technologies and addressing limitations of traditional inspection methods.