Data-driven prediction model for the femtosecond laser ablation threshold of metal film-metallic substrate structures

Aiming at the critical challenge of predicting the ablation threshold in femtosecond laser processing of thin film–substrate structures, a data-driven deep neural network prediction model is proposed. By leveraging the COMSOL two-temperature model (TTM), we systematically simulate the influence of substrate thermophysical parameters (electron thermal conductivity, lattice heat capacity, and electron–lattice coupling coefficient) on the thin film ablation threshold, constructing a comprehensive dataset integrating thermophysical characteristics and threshold values. The trained neural network model achieves an outstanding prediction performance on the test set, with a mean absolute error (MAE) of 0.0221 J/cm2 and a coefficient of determination (R2) of 0.966. Parametric sensitivity analysis reveals that the substrate’s electron heat capacity exerts the dominant influence on the ablation threshold, surpassing other thermophysical properties. After verifying the reliability of the two-temperature model through experiments, a neural network model is trained based on simulation data, achieving high-precision prediction on the test set. This provides a tool for the rapid acquisition of the ablation threshold of thin films on metallic substrates. We focus on the typical bilayer structure where a metal film is deposited on a metallic substrate. The core conclusions are applicable to similar layered metallic material systems and do not cover nonmetallic substrates or multicomponent composite film scenarios for the time being. They actually pave a way to the development of reliable predictive neural models with potential use in industry.