Abstract
The reconstruction of subwavelength defects from measured images of high-NA-projection systems is demonstrated. A structure consisting of a few small unknown defects in an otherwise known mask layout is studied. The footprint of the defect, which is the measured or simulated difference between images of masks with and without defects, is used to reconstruct the position, shape, and transmission of defects. The requirement is that the few unknown defects are sparsely located in the known mask layout. The technique relies on the cost function and an appropriate optimizer. The dependency of the reconstruction results on defect sizes and types of defects is presented. Moreover, the sensitivity of the technique to noise is investigated.