Fast lithography aerial image calculation method based on machine learning

Aerial image calculation for thick masks is an indispensable but time-consuming step in most lithography simulations. This paper develops a fast thick-mask aerial image calculation method based on machine learning for partially coherent lithography systems. First, some sparse sampling points are chosen from the source plane to represent the partially coherent illumination. Then, the training libraries of thick-mask diffraction near-fields are built up for all sampling points based on a set of representative mask features. For an arbitrary thick mask, we calculate its aerial image using the nonparametric kernel regression technique and the pre-calculated training libraries. Subsequently, a post-processing method is applied to compensate for the estimation error and improve the computational accuracy. In addition, this paper also studies the impacts of several key factors on the accuracy and efficiency of the proposed method. Finally, the proposed method is verified by the simulations at 45 nm and 14 nm technology nodes.