Rigorous electromagnetic simulator for extreme ultraviolet lithography and convolutional neural network reproducing electromagnetic simulations

BackgroundMask 3D (M3D) effects distort diffraction amplitudes from extreme ultraviolet (EUV) masks. Rigorous electromagnetic simulation is required to incorporate the M3D effects. However, electromagnetic simulation is too slow for optical proximity correction applications.AimWe explain the details of our electromagnetic simulator and convolutional neural network (CNN) model which reproduces the results of the electromagnetic simulations. CNN prediction is much faster than the simulation. We provide the codes for the simulator and CNN model.ApproachOur electromagnetic simulator uses the 3D waveguide model. The model solves Helmholtz-type wave equations. Weakly guiding approximation is applied to accelerate the simulation. M3D parameters are extracted from the result of the simulation, which represents the M3D effects. CNN model is constructed to predict M3D parameters from the input mask pattern.ResultsThe computation time of the electromagnetic simulation is ∼5 min for a 512×512 nm area on a wafer. The computation time using CNN is ∼0.1 s.ConclusionsEUVlitho is the open source of our EUV lithography simulator and CNN model. It is available on GitHub.