Machine learning-based laser heterodyne photothermal displacement method: simultaneous estimation of silicon thermal diffusivity and carrier lifetime

Abstract The laser heterodyne photothermal displacement (LH-PD) method, a recently developed photothermal technique, enables measurement of absolute surface displacements, which are otherwise challenging to measure with other photothermal methods. This method offers significant potential for quantifying physical properties that are difficult to achieve with traditional photothermal methods. In this study, we aimed to estimate the thermal diffusivity and carrier lifetime of Si using a machine learning model based on the time variation of the displacement obtained using the LH-PD method. By leveraging the machine learning model, we generated predictive mappings of thermal diffusivity and carrier lifetime of a pattern-etched Si wafer from the displacement mappings. Furthermore, our findings demonstrated that fine-tuning the model enabled accurate predictions of the carrier lifetime. While traditional simulations require tens of hours to estimate the material parameters, machine learning reduced this process to only a few seconds.