U-Net and molecular dynamics-guided reconstruction and simulation of Ag NPs metasurfaces for high-performance UVC photodetectors

Ultraviolet C (UVC) light sources are central to micro/nano fabrication. Current research focuses on optimizing exposure by increasing irradiance, improving precision, and implementing real-time dose control, creating an urgent demand for high-performance UVC photodetectors. To enhance responsivity, many UV detectors leverage metal nanoparticles (NPs) formed by solid-state dewetting on photosensitive substrates. However, limited characterization and simulation capabilities hinder detailed analysis of how these metasurfaces enhance responsivity, constraining device design and optimization. Therefore, developing methods to simulate and analyze the optical properties of metal NPs is essential. In this study, we propose an approach: reconstructing models of metasurfaces formed by solid-state dewetting and simulating them to support UVC photodetector design. Specifically, we combine U-Net deep learning with molecular dynamics (MD) simulations to construct accurate metasurface models, and employ the finite-difference time-domain (FDTD) method to evaluate the optical properties within each functional layer. Guided by these simulations, we fabricated a p-i-n UVC photodetector based on an ITO/Si structure, achieving an ultrashort response time (7.04 <i>µ</i>s) and high sensitivity (0.84 A/W). Crucially, this detector also serves as an experimental platform to elucidate the mechanism by which localized surface plasmon resonance (LSPR) enhances responsivity. We achieved this by correlating the measured UVC electrical response with simulated metasurface absorption and LSPR effects. Our "algorithm-guided simulation" strategy provides a framework for designing high-performance UVC photodetectors. Furthermore, it strengthens the theoretical basis for LSPR effects through closed-loop "simulation-device-mechanism" verification.