arXiv subjects

Chenxuan Xiang

Publications and source records attributed to Chenxuan Xiang.

3 recordsLinked to original sources

arXiv2025

Metasurface-based all-optical diffractive convolutional neural networks

The escalating energy demands and parallel-processing bottlenecks of electronic neural networks underscore the need for alternative computing paradigms. Optical neural networks, capitalizing on the inherent parallelism and speed of light propagation, present a compelling solution. Nevertheless, physically realizing convolutional neural network (CNN) components all-optically remains a significant challenge. To this end, we propose a metasurface-based all-optical diffractive convolutional neural network (MAODCNN) for computer vision tasks. This architecture synergistically integrates metasurface-based optical convolutional layers, which perform parallel convolution on the optical field, with cascaded diffractive neural networks acting as all-optical decoders. This co-design facilitates layer-wise feature extraction and optimization directly within the optical domain. Numerical simulations confirm that the fusion of convolutional and diffractive layers markedly enhances classification accuracy, a performance that scales with the number of diffractive layers. The MAODCNN framework establishes a viable foundation for practical all-optical CNNs, paving the way for high-efficiency, low-power optical computing in advanced pattern recognition.

physics.optics↗

arXiv2025

Integrated Spectropolarimeter by Metasurface-Based Diffractive Optical Networks

Conventional spectrometer and polarimeter systems rely on bulky optics, fundamentally limiting compact integration and hindering multi-dimensional optical sensing capabilities. Here, we propose a spectropolarimeter enabled by metasurface-based diffractive optical networks that simultaneously performs spectrometric and polarimetric measurements in a compact device. By leveraging the wavelength- and polarization-dependent phase modulation of metasurfaces, our system encodes the spectral and polarization information of incident light into spatially resolved intensity distributions, which are subsequently decoded by a trained deep neural network, enabling simultaneous high-accuracy reconstruction of both spectral compositions and Stokes parameters through a single-shot measurement. Experiments validate the proposed network's accurate reconstruction of the spectral and polarization information across a broad wavelength range, and further confirm its imaging capability. Notably, we demonstrate a chip-integrated sensor prototype combing both measurement functionalities into a commercial CMOS image sensor. This integrated platform provides a compact solution for on-chip multi-dimensional optical sensing, holding significant potential for versatile sensing, biomedical diagnosis, and industrial metrology.

physics.optics↗

arXiv2024

Multiplexed Metasurfaces for Diffractive Optics via Phase Correlation Method

The multiplexing capability of metasurfaces has been successfully demonstrated in applications such as holography and diffractive neural networks. However, identifying a suitable structure that simultaneously satisfies the phase requirements across multiple channels remains a significant challenge in many multiplexing design scenarios. In this study, we propose an innovative phase correlation method for metasurface multiplexing design that utilizes a multi-layer perceptron to establish phase correlations across multiple channels. This approach reduces the difficulty of multi-channel phase training by converting it into a simpler single-channel optimization task, thereby reducing design complexity and computational cost. Using the proposed method, we design a dual-wavelength multiplexed diffractive neural network and a multi-wavelength metasurface color holography under a linear polarization. The designed multiplexed metasurface achieves up to 90% classification accuracy in image recognition and exhibits good performance in color holography.

physics.optics↗