Controlling the morphologies of femtosecond laser-induced periodic surface structure on silicon by combining deep learning with energy deposition model

As an important patterning method, femtosecond laser-induced periodic surface structure (LIPSS) has attracted widespread attention in recent years. Due to the complex physical processes involved in the femtosecond laser scanning process, it is difficult to predict the required LIPSS morphologies, which hinders rapid customization of the large-area LIPSS patterns. In this study, a structural optimization method combining data-driven deep learning with energy deposition model was proposed to control LIPSS morphologies of silicon. After femtosecond laser processing under dynamic irradiation condition, four LIPSS morphological types were well defined based on experimental results. Deep learning model was constructed to extract data features by labeling and encoding sample data. The accuracy of self-correlation validation reached 98.0% and the accuracy of cross-correlation validation reached 91.9%. Our results show that the distribution of LIPSS morphological types exhibits the dependence of energy deposition, which is the joint effect of the effective pulse number and accumulative fluence. Moreover, the period of groove (super-wavelength LIPSS) increases with the increase of energy deposition, while the period of LSFL (low spatial frequency LIPSS) slightly decreases with the increase of energy deposition. Through this structural optimization method, customization and online monitoring of large-area LIPSS patterns in the future industrial applications.