Mask 3D Parameter Prediction in EUV Lithography by Convolutional Neural Network

In extreme ultraviolet (EUV) lithography, a rigorous but time-consuming electromagnetic simulation is conventionally used to obtain the diffraction amplitude from an EUV mask.This is because the absorber thickness of the mask is comparable to the pattern size and the mask 3D effect is non-negligible.In our research, the diffraction amplitude is divided into the thin mask amplitudes and the residual mask 3D amplitude.The mask 3D amplitude is a function of the incident angle and the diffraction order, and the incident angle dependence of the mask 3D amplitude is approximated by a linear function using mask 3D parameters.Mask 3D parameters represent the on-axis and the off-axis mask 3D effects.The number of parameters that need to be predicted is more than ten thousand, and these parameters are predicted by convolutional neural networks from an input mask pattern.The image intensity on the wafer estimated using the mask 3D parameters predicted by the convolutional neural network is obtained 3,000 times faster than the electromagnetic simulation.