Application of neural network to controlling three-dimensional electron-beam exposure distribution in resist

Electron-beam (e-beam) lithography is often employed for the fabrication of binary patterns with nanoscale features and grayscale structures. In such applications, exposure distribution along the depth dimension of the resist layer can significantly affect the geometry of the written patterns or structures. Most of the previous e-beam dose control schemes adopted a two-dimensional exposure model, ignoring the depth-dependent exposure variation. In this paper, a method which utilizes a neural network for explicitly controlling three-dimensional (3D) (vertical as well as lateral) distribution of exposure is described. A neural network allows one to achieve more balanced exposure distributions compared to a conventional region-by-region recursive approach. Through an extensive computer simulation, performance of the proposed approach to controlling (3D) exposure distribution has been analyzed in detail.