Method of regenerating diffraction signals for optical metrology systems

A method of enhancing accuracy of an optical metrology system including an optical metrology tool and an optical metrology model, the optical metrology model including a model of the optical metrology tool and a profile model of a sample structure;

• the sample structure formed on a substrate, the profile model having profile parameters, the optical metrology tool having an illumination beam, the illumination beam having an angle of incidence and an azimuth angle, the method comprising:
• (a) using a processor, generating a first library comprising Jones and/or Mueller matrices or components (JMMOC) and corresponding profile parameters, the library generated using ray tracing and a first set of beam propagation parameters;
• (b) using the processor, training a machine learning system (MLS) using the first library, with one or more profile parameters and beam propagation parameters as input and corresponding JMMOC as output;
• (c) loading and initiating the MLS;
• (d) providing one or more profile parameters and a range of beam propagation parameters to the MLS as input;
• (e) regenerating the corresponding JMMOC as output of the trained MLS;
• (f) generating a simulated diffraction signal using the regenerated JMMOC, a ray tracing algorithm, and the optical metrology model;
• (g) if matching one or more criteria of the simulated diffraction signal compared to a measured diffraction signal are met, accessing the one or more profile parameter associated with the matching simulated diffraction signal;
• else, modifying the one or more profile parameters and/or the range of beam propagation parameters and iterating operations (d), (e), (f), and (g) until the matching one or more criteria are met.