Abstract
The current yield of defect-free extreme ultraviolet (EUV) masks in high-volume manufacturing (HVM) is limited by multilayer defects, which can be mitigated through defect compensation. In this process, characterizing the geometric morphology parameters, defect types (bump/pit), and the number of defect-induced deformed layers is crucial for compensation. EUV dark-field microscopy (EUV-DFM) is a widely adopted technique for mask defect inspection, offering high sensitivity and throughput. However, existing EUV-DFM studies have yet to fully exploit the image information, resulting in limited characterization capabilities. This paper proposes a novel defect characterization method that integrates EUV-DFM with neural networks. Developed, trained, and evaluated using simulated EUV-DFM data, the proposed method enables the characterization of detailed defect morphology, including both deformed multilayer surfaces and smoothed ones. Simulation results demonstrate that the proposed method achieves a mean absolute percentage error (MAPE) of 0.75% and 1.40% for geometric morphology reconstruction of bump and pit defects, respectively, and a MAPE of 0.35% for the number of defect-induced deformed layers, while also maintaining robustness. This method demonstrates the potential to enhance the defect characterization capability of EUV-DFM and provides a novel approach for multilayer defect characterization in HVM. This methodology is expected to play a significant role in realizing the ‘zero-defect’ EUV mask manufacturing process.