Metasurfaces, owing to their capacity to significantly enhance localized electric fields, have emerged as crucial tools for improving sensor sensitivity and enabling efficient biomolecular detection in the terahertz range. However, most terahertz metasensors in current research focus only on a single molecular absorption peak of the analyte, limiting their ability to distinguish substances with similar absorption peaks. Achieving metasensors that precisely match molecular fingerprint spectra requires simultaneously optimizing multiple resonance modes, which remains a considerable challenge. Here, we introduced a hybrid algorithm-driven metasensor design approach (HAMD) that combines deep learning diffusion model with a genetic algorithm to efficiently design highly integrated metasensors capable of matching multiple molecular absorption peaks. The deep learning model incorporates the attention mechanism, enabling not only accurate prediction of sharp spectra but also high-fidelity inverse design of multiresonant spectra. The genetic algorithm further refines the local characteristics, enhancing the detection accuracy of the metasensor. We also experimentally demonstrated the sensing performance of the designed metasensors in detecting isomers, where Rabi splitting is observed and fingerprints of isomers are clearly recognized. Our approach contributes to advancing the design of molecule-specific metasensors and promotes the application of machine learning in the design of other photonic devices.
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号