Discrete spectral Bayesian deconvolution for thermal parameter inversion in micro-nanoscale heterostructures

Abstract

Obtaining the thermophysical properties and film thickness of micro-nanoscale heterostructures is essential for thermal management in semiconductor devices. However, non-invasive and numerically efficient extraction of multiple thermal parameters remains challenging due to the complexity of the inversion problem. In this study, we propose a parameter inversion technique based on a novel approach to spectral analysis of thermal transient responses. First, the Discrete Spectral Bayesian Deconvolution method is introduced to extract well-resolved spectral lines from junction thermal responses, forming the foundation of a parameter inversion algorithm. Second, it is applied to two-layer heterostructures with imperfect thermal contact, where an analytical solution for transient heat conduction is derived and the numerical accuracy of the inversion solver is verified. Finally, case studies are conducted on nanoscale Silicon-on-Insulator (SOI) and microscale GaN-on-SiC using simulation data, as well as microscale SOI using experimental data. The inverted thermophysical properties and film thickness exhibit relative errors below 10% when compared to the reference values, highlighting the algorithm’s robustness and generalizability. This study offers a non-destructive and adaptable framework for thermal parameter measurement across diverse material compositions and feature sizes.