Super-resolution reconstruction of thermal property distributions in transient thermoreflectance

Pump–probe thermoreflectance technique encounters challenges in balancing high spatial resolution and high-throughput testing during wafer-level scanning. To address this, we introduce super-resolution reconstruction (SR), an emerging computer vision technique, into semiconductor thermophysical field characterization. By analyzing the spatial isomorphism between thermal field data and visual images, we developed a “ sparse sampling-high resolution reconstruction” framework based on a learnable SRResNet architecture, achieving 4× super-resolution reconstruction of low-resolution thermal property maps obtained through automated scanning. This method overcomes the inherent limitation of traditional point-by-point scanning measurements, which requires a time that increases with the square relationship of spatial resolution (t ∝ N2), enhancing wafer-level thermal property characterization efficiency by over 16-fold at equivalent resolution requirements. Furthermore, to resolve edge smoothing artifacts caused by SR, we developed a correction strategy incorporating transient thermoreflectance edge measurement data, significantly improving the authenticity and accuracy of thermal property distributions. This work systematically explores the application potential of SR in thermal–physical characterization, providing an efficient and rapid approach for semiconductor material analysis that effectively resolves the trade-off between testing time and spatial resolution.