Prediction of surface morphology and reflection spectrum of laser-induced periodic surface structures using deep learning

Abstract

Laser-induced periodic surface structures have been extensively explored as an enabling tool for fabricating various optical surfaces because the resulting surface ripples can effectively modify surface reflectance. Here, we propose a deep-learning-based model for predicting high-quality surface scanning electron microscopy (SEM) images with detailed surface morphology corresponding to unexplored process conditions. In addition, the reflectance value at 550 nm and reflection spectra from the generated (or original) SEM images were predicted. To obtain training data, stainless steel 304 specimens were processed with femtosecond laser pulses on a large process window consisting of 32 process conditions to obtain SEM images with various surface morphologies and corresponding reflection spectra. The image prediction model is based on a conditional generative adversarial network, which generates a surface SEM image from the laser fluence and scanning speed values. The reflectance prediction model was developed based on ResNet152 and Long-Short Term Memory network. The average accuracies for the pattern period and ripple width were 98.2 and 94.6 %, respectively, and the L2 norm error for the reflection spectra was less than 4 %. It was demonstrated that the reflection spectra can be accurately predicted using only surface images, which can also be accurately generated from process parameters.