A Conditional Latent Diffusion Approach for Solving Phase Retrieval in Ptychography
OHSA/E13
Abstract:
Extreme Ultraviolet (EUV) lithography is a critical technology for semiconductor device manufacturing that promises to uphold Moore’s law. EUV mask metrology, which comprises inspection and review of the EUV mask defective sites, is one of the major challenges that need to be addressed for a successful deployment of EUV lithography. At Paul Scherrer Institute (PSI), the RESCAN microscope is used for EUV mask inspection. RESCAN employs coherent diffraction imaging (CDI) and ptychography to detect and characterize defects, with a demonstrated resolution down to 50x50 nm² size. Ptychography solves the phase retrieval problem using an iterative algorithm that requires access to a vast amount of data (diffraction patterns measured at overlapping probe positions), and it is a time-consuming process. In this talk, I will present our work based on deep generative models, which is efficient both in terms of compute time and storage space, requiring a comparatively small number of diffraction patterns and iterations to achieve a satisfactory reconstruction. More specifically, I will delve into our proposed solution which is based on conditional latent diffusion models (LDMs). By implementing these advanced models, we aim to develop a cost-effective, high-speed data processing framework that will significantly impact future semiconductor manufacturing technologies.
ZOOM ID: https://psich.zoom.us/j/68363377588
Laboratory for Simulation and Modelling
SDSC Hub at PSI