3D Lithography Using Talbot Effect

- Controlling the Periodicity of Structures by Deep Learning -

抄録

In recent years, 3D nano-periodic structures have attracted much attention because the structures have a function; for example, entropic cages near a nanopore can trap DNA. To fabricate periodic structures, this paper focused on Talbot lithography which has efficiency for processing. Previously, convolutional neural network (CNN) is utilized to estimate phase and amplitude of incident light for exposurement in Talbot lithography. However, light intensity distribution estimated by the incident light is not in good agreement with design distribution because of too much information of datasets. Moreover, the estimated distribution does not have periodicity. To overcome the problems, this paper focused on CNN with binarized datasets to improve flexibility of Talbot lithography. An estimation result trained by CNN with a binarized dataset is in good agreement with design distribution. Additionally, it is possible to control local intensity distribution by controlling threshold for processing.