Physics-informed deep learning approach for nanoindentation-based thin film analysis

Abstract

This study presents an application of a physics-informed deep learning framework to improve and accelerate the yield stress characterization of thin films used in microelectronics to ensure long-term mechanical reliability via nanoindentation measurements. By combining finite element modeling (FEM) with neural networks, an accurate model for thin film yield stress has been demonstrated. This model offers comprehensive insights into the mechanical properties and plasticity of thin films under various loading conditions. The decision-making process of the model is investigated using explainable AI visualization techniques, enhancing the model's transparency and interpretability. Nanoindentation experiments on metal and dielectric thin films validate the high accuracy of the proposed deep learning models. This approach allows for the rapid analysis of load-displacement curves in milliseconds while providing high accuracy in yield stress estimations. Consequently, the proposed methodology significantly accelerates the characterization process and provides accurate yield stress estimations for thin film nanoindentation measurements, which is crucial for applications in microelectronics and the reliability of semiconductor devices.