Deep Learning: A Rapid and Efficient Adaptive Design Approach for Phase-Type Optical Needle Modulators

Abstract

The radially polarized beams are modulated by phase-type optical needle modulators can be tightly focused to create needle-like focused beams, which are called optical needles. The use of optical needles with different resolutions and focal depths as direct writing heads for laser direct lithography enables periodic, cross-scale processing of high aspect ratio micro-nano structures with different line widths. The design of the phase-type optical needle modulators is the key to obtain optical needles with different resolutions and focal depths. However, the existing conventional methods for designing phase-type optical needle modulators rely on the physical model for generating optical needles and the defined fitness function, which makes their design time long and not adaptive. Based on the deep learning, a novel phase-type optical needle modulator design (PONMD) approach is proposed in this paper. The results show that the PONMD method takes 0.5526ms to design a phase-type optical needle modulator, and the similarity between the designed and target values is 96.73%. Compared with the conventional methods, the time consumption is reduced by about 8 orders of magnitude, and the similarity is improved by 11.19%. The PONMD approach has the advantages of adaptability, more efficient, less time-consuming, and less computational resource-consuming.