Deep learning driven ultra-broadband solar absorber based on Ti–Si3N4 composite multilayer structure

Abstract

Efficient harvesting and utilization of solar energy is a key strategy for addressing the global energy crisis. However, current solar absorbers still require improvements in both broadband spectral response and absorption efficiency. In this work, we propose a novel composite stacked structure and optimize its structural parameters using deep learning to achieve a near-ideal absorption spectrum. The absorber unit cell utilizes titanium (Ti) substrate and integrates Ti-Si₃N₄ stacked four-lobed structure, Ti arc-faced cubic pillars structure and Ti cylindrical structure. To facilitate device design and achieve near-perfect absorption spectra, we construct a deep learning-based design methodology and establish an inverse mapping between the ideal absorption spectra and structural parameters. The mean squared error (MSE) between the designed and target spectra based on the optimized structural parameters is on the order of 10⁻⁴. Simulation results show the design achieves an average absorption of 99.06 % in the 310–3010 nm, with absorption consistently above 95 % across a 2700 nm bandwidth. The solar absorption efficiency reaches 98.75 % under the AM1.5 conditions. In terms of thermal radiation performance, it achieves a thermal emission efficiency of 99.44 % at 1300 K operating temperature, demonstrating excellent photothermal conversion capabilities. Moreover, the proposed solar absorber exhibits polarization insensitivity and wide-angle tolerance, and maintains high absorption across a wide range of structural parameter variations, indicating good fault tolerance. Thus, our design holds significant potential for applications in efficient solar energy collection, photothermal conversion, and thermal radiation.