Enhancing defect detection in semiconductor manufacturing through physics-based synthetic imaging for micro- and nano-scale structures

This study investigates the application of physics-based synthetic imaging to enhance defect detection in semiconductor manufacturing. Leveraging optical principles, this approach simulates realistic interactions such as diffraction and interference, addressing limitations of generative neural networks like overfitting and lack of diversity. Virtual environments modeled micro- and nano-scale trench isolation structural variations, enabling data generation with minimal reliance on real-world defect samples. Validation experiments using pinhole artifacts demonstrated the method's effectiveness in enhancing neural network training under constrained data conditions.