Femtosecond laser processing with adaptive optics based on convolutional neural network

A wavefront aberration causes focal spot distortion leading to loss of resolution and efficiency in laser processing. Therefore, aberration compensation is important for ensuring sub-micron resolution in laser processing. In this paper, femtosecond laser processing with adaptive optics based on a convolutional neural network was demonstrated. The aberrations existing in the laser processing system were continuously predicted by the trained network with an update period of 36 ms and was compensated by a liquid crystal spatial light modulator. In the experiment, the neural network-based adaptive optics reduced the wavefront error in the laser processing system to most one-ninth. Furthermore, parallel laser processing by a computer-generated hologram displayed on the spatial light modulator was also demonstrated while dynamically compensating the aberrations in the system.

Automated summary

This paper demonstrates an adaptive optics technique based on convolutional neural network for femtosecond laser processing, continuously predicting and compensating aberrations to reduce wavefront error in the system.