Infrared spectroscopy is a technique that utilizes the interaction between matter and infrared light for material analysis, which plays a crucial role in fields such as material analysis, mass spectrometry identification, biomedicine, and military security. As information technology and related disciplines advance, the miniaturization of infrared spectrometers has emerged as a prominent research trend for both current and future applications. However, existing portable infrared spectrometers generally have narrow spectral ranges and low resolution due to the need to balance device miniaturization. Metasurfaces, as artificial two-dimensional materials, possess advantages such as small size and ease of integration. By designing the shape, size, and arrangement of metasurface atoms, researchers can create structures with special modulation effects on the phase, amplitude, and polarization of light, achieving remarkable results in fields such as achromaticity, quantum light sources, color routing, and holographic imaging. In spectral imaging, the use of metasurfaces is expected to overcome the limitations of traditional spectrometers, enabling the realization of high-precision, wide-spectrum, miniature portable spectrometers for infrared applications.

We present a miniature infrared wide-spectrum spectrometer based on the principle of spectral encoding utilizing a metasurface. The morphology of the metasurface consists of a periodic arrangement of 2×2 structural units, each of which provides a distinct modulation effect on the wide-spectrum light. When a beam of wide-spectrum infrared light carrying spectral information is incident on the metasurface, the metasurface encodes the spectral information of the incident light, ultimately yielding a light intensity value. This intensity value can be decoded at the backend using deep learning algorithms to retrieve the spectral information of the incident light. We employ a C₄-symmetric metal-dielectric-metal structure, which is insensitive to polarization. When infrared light is incident on the metasurface composed of this structure, a unique electromagnetic field is generated at the interfaces between the upper and lower metal layers and the dielectric. This interaction excites surface plasmons, which represent collective oscillation modes of electromagnetic waves and free electrons at the metal surface. The coupling of the two electromagnetic fields results in band hybridization, leading to resonance at specific frequencies. By varying the materials, heights, and geometries of the metasurface atoms, additional resonance modes can be introduced, allowing for diverse modulation effects on the incident light. To enhance the degrees of freedom of the metasurface atoms and achieve infrared wide-spectrum imaging, we pixelate the unit structure to improve encoding performance. Considering the requirements for high transmittance and ease of fabrication, ZnSe is selected as the substrate material. Simulations are conducted using the finite-difference time-domain (FDTD) method, with periodic boundary conditions applied along the x- and y-axes, and perfectly matched layer (PML) boundary conditions applied along the z-axis.

The unit structure consists of four equally spaced cylinders. By fixing one parameter, either the radius of the cylinders or the distance between them, and varying the other, we obtain different transmission spectra (Fig. 2). This indicates that both the size and arrangement of the metasurface atoms influence transmission rates. To enhance the degrees of freedom of the metasurface atoms, we pixelate the unit structure and randomly place cylinders within the defined regions, resulting in a variety of spectral transmission rates. According to the principles of spectral encoding and compressed sensing, a sparser spectral transmission matrix for each structure of the metasurface is advantageous for computational reconstruction. Therefore, we adopt the correlation coefficient as a criterion for assessing the sparsity of the spectral transmission rates, selecting structures with an average transmission rate greater than 50% as our filtering condition. This process yields four distinct structures (Fig. 3), with the correlation coefficients between the corresponding spectral transmission curves ranging from a maximum of 0.19 to a minimum of -0.16. Utilizing deep learning methods, we simulate the spectral detection process to further investigate the feasibility of our design scheme. The results indicate that for broadband spectra, the accuracy of spectral reconstruction exceeds 90% (Fig. 4). Additionally, to explore the diversity of spectral detection, we also reconstruct spectral curves with narrower unimodal characteristics (Fig. 5). The reconstruction results demonstrate that even at a signal-to-noise ratio of 27 dB, it is possible to achieve a reconstructed spectrum that closely resembles the original spectrum.

Based on the principle of spectral encoding in metasurfaces, a micro-spectral imaging system for 1?10 μm spectral band is designed in this paper. The metasurface is composed of polarization-insensitive, pixelated metal?insulator?metal (MIM) structures, which not only maintain an average transmission rate exceeding 50% across three spectral bands but also exhibit sufficiently low correlation in spectral transmission rates. This design facilitates convenient integration with commercial cameras via a patching approach. Additionally, we utilize deep learning techniques to investigate the spectral reconstruction capabilities of the system. The results indicate that for relatively smooth spectral curves, the reconstruction performance is satisfactory. Moreover, our design method demonstrates a certain degree of reconstruction ability for spectral curves characterized by narrow unimodal features. In summary, we have developed an infrared wide-spectrum miniaturized spectrometer based on metasurfaces, which possesses rapid and non-destructive analytical capabilities. This technology holds significant implications in various fields, including environmental monitoring, medical diagnostics, and national security. Although the current number of spectral reconstruction channels is somewhat limited, advancements and optimizations in reconstruction algorithms are anticipated to enable the realization of a miniaturized spectrometer that maintains high spatial resolution while achieving high spectral resolution in the future.