Inspection of reticles using machine learning

Abstract

Disclosed are methods and apparatus for inspecting a photolithographic reticle. A near field reticle image is generated via a deep learning process based on a reticle database image produced from a design database, and a far field reticle image is simulated at an image plane of an inspection system via a physics-based process based on the near field reticle image. The deep learning process includes training a deep learning model based on minimizing differences between the far field reticle images and a plurality of corresponding training reticle images acquired by imaging a training reticle fabricated from the design database, and such training reticle images are selected for pattern variety and are defect-free. A test area of a test reticle, which is fabricated from the design database, is inspected for defects via a die-to-database process that includes comparing a plurality of references images from a reference far field reticle image to a plurality of test images acquired by the inspection system from the test reticle. The reference far field reticle image is simulated based on a reference near field reticle image that is generated by the trained deep learning model.