Deep Learning-Based Data Processing Method for Transient Thermoreflectance Measurements

Pump-probe thermoreflectance has been commonly applied for characterizing the thermal properties of materials. Generally, a reliable and efficient non-linear fitting process is often implemented to extract unknown thermal parameters during the pump-probe thermoreflectance characterizations. However, when it comes to processing large amounts of data acquired from similar structural samples, non-linear fitting process appears to be very time-consuming and labor-intensive to search for the best fitting for every testing curve. Herein, we propose to apply deep learning (DL) approach to nanosecond transient thermoreflectance technique for high-throughput experimental data processing. We first investigated the effect of training set parameters (density and bounds) on the predictive performance of the DL model, providing a guidance to optimize the DL model. Then, the DL model is further verified in the measurement of the bulk sapphire, SiC, diamond samples, and GaN-based multilayer structures, demonstrating its capability of analyzing the results with high accuracy. Compared to the conventional non-linear fitting method (such as Global Optimization), the computation time of the new model is 1000 times lower. Such a data-driven DL model enables the faster inference and stronger fitting capabilities and is particularly efficient and effective in processing data acquired from wafer-level measurements with similar material structures. (c) 2024 Author(s). All article content, except where otherwise noted, is licensed under a Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).