Machine learning-driven metamaterial absorbers for solar energy harvesting and UV to infrared radiation shielding

This paper introduces the design and electromagnetic analysis of a compact metamaterial absorber (MMA) that operates across the ultraviolet to near-infrared spectral range for solar energy harvesting and electromagnetic interference (EMI) shielding. The proposed metal-dielectric-metal structure with an ultra-compact unit cell dimension of 20 × 20 × 5 nm 3 uses an interlocked rotated square ring resonator made of nickel on an aluminum nitride (AlN) substrate. Numerical results show that there is a broad absorption with an average value of 97.45% in 200-900 nm and a peak absorption of 99.71% at 400 nm. Surface current and electric-magnetic fields analyses elucidate the mechanism of resonance and the energy dissipation behavior. The absorber has polarization-insensitive performance in both the TE and TM modes and has stable absorption up to an incidence angle of 60°. The high solar energy harvesting efficiency of 96.33% is obtained. The polarization conversion ratio remains negligible (PCR ≈ 0.00043), ensuring minimal re-radiation. Additionally, notable shielding effectiveness is obtained, which amounts to 50.49 dB at 400 nm and to an average value of 42.67 dB in the range 300-900 nm. A physics-guided machine learning framework, termed the Physics-Constrained Residual Network (PCRN), is employed to approximate the absorption response over the simulated wavelength and substrate-thickness range. The learning component is intended as a surrogate prediction tool within the defined parameter domain and is used to complement, rather than replace, the full-wave electromagnetic analysis.