A review of the application of machine learning in short-pulse and ultrashort-pulse laser etching

Short Pulse Laser (SPL) and Ultrashort Pulse Laser (USPL) etching are pivotal for fabricating high-precision components in extreme manufacturing. However, their industrial deployment faces significant challenges. The highly non-linear nature of laser-material interactions necessitates expensive trial-and-error, while pure data-driven models suffer from “black-box” interpretability issues and data scarcity, and traditional static optimization fails to adapt to dynamic disturbances during real-time processing. To address these issues, machine learning (ML) and metaheuristic algorithms (MA) have been widely used in SPL and USPL processing. The primary objective of this review is to systematically synthesize and critically evaluate the applications of Machine Learning (ML) and MA in SPL/USPL etching, specifically focusing on drilling, microchannel fabrication, and Laser-Induced Periodic Surface Structures (LIPSS) . We categorize core strategies into “Forward Modeling” (quality prediction) and “Reverse Design” (parameter optimization) to elucidate how these algorithms mitigate the aforementioned challenges, and providing a forward-looking perspective, highlighting Physics-Informed Machine Learning (PINNs) for enhancing interpretability with sparse data. This review expands on the little-covered part of the existing review literature on the application of ML and MA in the field of laser processing and summarises the effectiveness of different ML models and MA for SPL and USPL in terms of etching. It further explores the potential of emerging, multiple ML converged processing and provides an outlook on novel trends and challenges at the intersection of laser etching and ML.