A deep learning-based predictive simulator for the optimization of ultrashort pulse laser drilling

Abstract Ultrashort pulse laser drilling is a promising method for the fabrication of microchannels in dielectric materials. Due to the complexity of the process, there is a strong demand for numerical models (simulators) that can predict structures produced under specific processing conditions in order to rapidly find optimal processing parameters. However, the validity of conventional laser drilling simulators for dielectrics has been confined to a range of strict interpolations of the data used during the construction of the model, and thus, their usefulness is limited. Here, we demonstrate simulator-based optimization for ultrashort pulse laser drilling in dielectrics based on an iterative deep neural network which is trained to predict microchannel structure after a small number of irradiated pulses. Our approach predicts the development of hole shapes over a wide variety of conditions and allowed the discovery of 20% more energy efficient processing strategies than in the initial experimental data. More broadly, our approach can address realistic problems in laser processing considering a variety of processing parameters, and thus enabling improved performance of next-generation smart laser processing systems.