U-Net convolutional neural network-based modification method for precise fabrication of three-dimensional microstructures using laser direct writing lithography

Abstract

In this paper, a modification method based on a U-Net convolutional neural network is proposed for the precise fabrication of three-dimensional microstructures using laser direct writing lithography (LDWL). In order to build the correspondence between the exposure intensity distribution data imported to the laser direct writing system and the surface profile data of the actual fabricated microstructure, these two kinds of data are used as training tensors of the U-Net convolutional neural network, which is proved to be capable of generating their accurate mapping relations. By employing such mapping relations to modify the initial designed exposure intensity data of the parabolic and saddle concave micro-lens with an aperture of 24µm×24µm, it is demonstrated that their fabrication precision, characterized by the mean squared error ( MSE ) and the peak signal-to-noise ratio (PSNR) between the fabricated and the designed microstructure, can be improved significantly. Specifically, the MSE of the parabolic and saddle concave micro-lens decreased from 100 to 17 and 151 to 50, respectively, and the PSNR increased from 22dB to 29dB and 20dB to 25dB, respectively. Furthermore, the effect of laser beam shaping using these two kinds of micro-lens has also been improved considerably. This study provides a new solution for the fabrication of high-precision three-dimensional microstructures by LDWL.