Prior-primed deep neural network based EUV mask inspection.

Area of Science:

• Semiconductor Manufacturing
• Metrology
• Computational Imaging

Background:

• Actinic patterned mask inspection (APMI) is crucial for EUV lithography photomask quality control.
• Conventional APMI methods face scalability and cost challenges, hindering Moore's Law progression.
• Ptychography offers a lensless alternative but suffers from low throughput due to slow phase retrieval and data acquisition.

Purpose of the Study:

• To develop a rapid APMI method using deep neural networks (DNNs) for faster defect identification.
• To leverage prior information of photomask samples for high-fidelity image reconstruction with limited data.
• To demonstrate the feasibility of DNN-based APMI for EUV mask inspection in semiconductor fabrication.

Main Methods:

• Exploited a deep neural network (DNN) architecture trained on synthetic and experimental data.
• Utilized a small subset (<5%) of measured diffraction patterns for image reconstruction.
• Tested the DNN with a completely synthetic dataset to validate real-world applicability.
• Integrated DNN predictions as an initial guess for conventional ptychography.

Main Results:

• Achieved significant improvement in reconstruction speed compared to standard ptychography, even with prior knowledge.
• Demonstrated high-fidelity image reconstruction and defect identification using limited diffraction data.
• Successfully performed die-to-database inspection on a logic-like EUV mask pattern.

Conclusions:

• Deep neural networks offer a promising solution to accelerate ptychographic EUV mask inspection.
• The proposed rapid APMI method enhances throughput and scalability for semiconductor manufacturing.
• This approach can be integrated into existing workflows for efficient photomask quality verification.