Accurate and Efficient Proximity Effect Correction for Electron Beam Lithography Based on Multilayer Perceptron Neural Network

Abstract

This paper proposes a proximity effect correction (PEC) method for electron beam lithography (EBL) using multilayer perceptron (MLP) neural network (NN). By leveraging the symmetric characteristics of the point spread function (PSF), several annular regions divided around the exposure point are used as the input of NN. The exposure dose after traditional model-based PEC is used as the output when training NN. The PEC inference error of trained NN for grating with different periods can reach the same level as model-based PEC method (relative error is less than 1%). Meanwhile, the inference speed of the NN-based PEC is more than 7~10 times faster than that of the model-based PEC, which can significantly enhance the efficiency of PEC.