Polarimetry, the measurement of the state of polarization (SoP) of light, is crucial in various fields. Traditional polarimetry systems, however, often involve bulky components like polarizers and waveplates, hindering miniaturization. Metasurfaces, two-dimensional arrays of subwavelength meta-atoms, offer a promising solution for compact polarimetry. Existing metasurface-based polarimetry methods often rely on spatial multiplexing, which limits spatial resolution or complex fabrication processes. This paper introduces a novel approach that uses a single chiral metasurface to independently modulate three polarization channels (RCP to LCP, LCP to RCP, and co-polarized) without spatial multiplexing. By simultaneously measuring amplitude contrast and phase difference between right-hand circular polarization (RCP) and left-hand circular polarization (LCP) components, the SoP can be determined. The incorporation of a deep convolutional neural network enhances the accuracy and robustness of the polarimetry, particularly for spatially non-uniform polarizations.

The paper reviews existing polarimetry techniques, categorizing them into division-of-time and division-of-space methods. It highlights the limitations of traditional methods, such as their bulky size and limited spatial resolution. It then discusses various metasurface-based polarimetry approaches, pointing out their drawbacks, including waveguide restrictions, complex fabrication (dual-layer configurations), reflection/diffraction modes unsuitable for direct sensor integration, and the trade-off between pixel size and crosstalk in metalens array designs. The authors note that most existing methods focus on intensity measurements of multiple polarization bases, whereas their proposed method directly measures amplitude contrast and phase shift, simplifying the process and improving efficiency.

The proposed method employs a chiral metasurface designed to generate three focal lines corresponding to RCP to LCP, LCP to RCP, and co-polarized components. The spatial arrangement of these lines allows for simultaneous extraction of amplitude contrast and phase difference between RCP and LCP components via interferometry. The intensity distributions at specific intersection points (A, B, and C) of these focal lines are used to calculate the amplitude contrast and phase difference. The authors employ a single planar anisotropic meta-atom, analyzing the Jones matrix in both Cartesian and circular polarization bases to demonstrate independent phase modulation of the three channels. The design utilizes planar chiral meta-atoms to break mirror and rotational symmetry, enabling independent phase manipulation. FDTD simulations (Lumerical) were used to generate a meta-atom library, selecting meta-atoms with optimal phase delays and efficiency (approximately 71%). A 25x25 array of these meta-atoms was fabricated, and its optical and SEM images are presented. Experiments used a white-light laser with a 470 nm filter, a polarizer, and a quarter-wave plate to generate various polarization states. The focal plane was captured by an image sensor using an objective lens. To improve accuracy, circular polarization analysis was performed using CP filters. A neural network is used to refine the analysis, particularly for spatially inhomogeneous polarization profiles. This analysis included uniform and inhomogeneous polarization.

Simulations and experiments were conducted for various uniform polarization states (linear, circular, and elliptical). The experimental results showed good agreement with simulations, demonstrating the ability of the system to accurately determine the SoP. The average transmission of the metasurface was about 80%, and the average diffraction efficiency was approximately 70%, resulting in a total efficiency of about 56%. The high spatial resolution of the system was demonstrated through measurements of spatially non-uniform polarization states generated by a specially designed metasurface. The application of the proposed polarimetry in distinguishing between two similar glasses was successfully demonstrated. The neural network was crucial in improving the accuracy and robustness of the polarimetry, particularly in the presence of noise and non-uniformity.

The proposed method provides a significant advancement in polarimetry by achieving high spatial resolution in a compact and efficient manner. The non-interleaved design and the use of a single chiral metasurface simplify the system and improve its robustness. The incorporation of a neural network further enhances accuracy and enables the measurement of spatially inhomogeneous polarizations. The relatively high efficiency of the system makes it suitable for real-world applications. The demonstrated ability to distinguish between similar glasses showcases its potential in various sensing and detection tasks.

This work presents a novel high-spatial-resolution polarimetry method based on a non-interleaved chiral metasurface and neural network assistance. The method is compact, efficient, and robust, offering significant improvements over existing techniques. Future research may focus on further enhancing the efficiency, miniaturizing the system for integration onto chips, and expanding applications to other wavelengths and polarization states.

The current system's efficiency is limited by the fabrication imperfections of the metasurface and the inherent losses in the optical components. The accuracy of the polarimetry could be further improved by developing more sophisticated neural network architectures and training datasets. While the neural network helps mitigate noise, the system's performance may be affected by extreme noise levels or highly complex polarization patterns. The current design operates at 470nm; exploring broader spectral ranges is a potential area for future development.