Smooth, continuous-profile diffractive optical elements and other regular microstructures demand advanced characterization methods. Here the problem of reconstructing the profiles with optical scat-terometry is addressed for the first time to our knowledge. We represent the profile of a blazed grating with an analytic Chebyshev series. The coefficients of the series are inferred from the far-field diffraction pattern, when the grating profile is illuminated with coherent light. An advanced neural-network model is used as a nonlinear statistical estimator. The reconstruction error is found to be less than 4% as normalized to the depth of the profile, even with realistic measurement errors included in the simulations.
展开▼