Machine learning-assisted high-throughput prediction and experimental validation of high-responsivity extreme ultraviolet detectors.

Abstract

Identifying materials with optimal optoelectronic properties for targeted applications represents both a critical need and a persistent challenge in optoelectronic device engineering. Machine learning models often depend on extensive datasets, which are typically lacking in specialized research domains such as extreme ultraviolet (EUV) radiation detection. Here, we demonstrate a Cross-Spectral Response Prediction framework that leverages existing visible and ultraviolet (UV) photoresponse data to predict more efficient material's performance under EUV radiation. Our predictive model, based on Extremely Randomized Trees, correlates physical descriptors with performance across different spectral regions using a comprehensive dataset of 1927 samples. Through this approach, we identified promising materials such as α-MoO₃, MoS₂, ReS₂, PbI₂, and SnO₂, achieving responsivities varying from 20 to 60 A/W, exceeding conventional silicon photodiodes by ~225 times in EUV sensing applications. Monte Carlo simulations revealed double electron generation rates (~2×10⁶ electrons per million EUV photons) compared to silicon, with experimental validation confirming the effectiveness of our prediction framework for accelerating the discovery of other high performing materials for diverse spectral applications.