Design of efficient high-order immersed metagratings using an evolutionary algorithm

Immersed reflection gratings improve spectral resolving power by enabling diffraction within a high refractive index medium. This principle has been widely adopted to make grating spectrometers more compact. Conventional immersed gratings have blazed profiles which typically show the highest efficiency for one main design wavelength. In addition, the blazed profiles tend to cause significant polarization sensitivity. In this work, we propose an alternative approach for designing an immersed grating composed of sub-wavelength structures, designed to increase diffraction efficiency and reduce polarization dependence. For a theoretical demonstration, a reflective metagrating immersed in silicon is optimized over the short-wave infrared band-3 (SWIR-3, here $2.304~\mu$m-$2.405~\mu$m), targeting the same diffraction angles as the immersion grating used in the Sentinel-5 Earth observation mission. The structure is optimized using a modified Covariance Matrix Adaptation Evolution Strategy (CMA-ES). The optimized immersed metagrating achieves an average efficiency of (over the SWIR-3 band) $\sim 78\%$, compared to $\sim 62\%$ for the conventional immersed blazed grating, and reduces polarization sensitivity from roughly $\sim 15\%$ to $\sim 5\%$. A manufacturing tolerance analysis is also conducted to evaluate the design's performance under systematic manufacturing errors, which revealed a degradation of $\sim 10\%$ efficiency at feature size errors of $\pm 25{nm}$ and almost negligible effect on the efficiency at $-10{nm}$ and of $\sim 5\%$ at $+10{nm}$.