The growing computational demands of Convolutional Neural Networks (CNNs) in artificial intelligence applications-particularly in hyperspectral classification and medical imaging-have surpassed the processing capabilities of von Neumann electronic devices, due to their inherent memory and power limitations. Although optical computing offers inherent parallelism advantages for convolution operations, traditional implementation schemes still exhibit limitations in spectral dimensions and adaptability. To overcome these constraints, we propose a multidimensional optical field modulation method that enables parallel multispectral image processing through a spatiotemporal transformation optical convolution system. This architecture employs a Fourier-domain axial multi-kernel design combined with joint displacement-spectral modulation technology, which not only physically models inter-kernel coupling effects across spectral bands but also ensures computational accuracy through compensation mechanisms. Simulation analysis indicates that the system's theoretical minimum wavelength capacity range typically reaches similar to 1000nm@NA=0.2 (without axial shift), achieving classification accuracies of 98.59%, 87.98% on MNIST and FashionMNIST datasets, and 92.45% (human recognition) and 93.7% (ship recognition) on maritime search and rescue image datasets. Experimental validation using liquid crystal modulators with narrowband characteristics (reflectivity/transmissivity >90%) processed spatially distributed multispectral inputs with minimal crosstalk. After calibration, the classification accuracy for maritime search and rescue tasks reached 85% (human recognition) and 85.25% (ship recognition). This work establishes a theoretical foundation for spectral-spatial-temporal co-optimization, not only expanding the dimensional limits of optical convolution but also providing an efficient hardware solution for hyperspectral medical imaging and high-dimensional AI computing in the post-Moore era.

Programmable photonic integrated circuits (PPICs) offer a versatile platform for implementing diverse optical functions on a generic hardware mesh. However, the scalability of PPICs faces critical power consumption barriers. Therefore, we propose a novel non-volatile PPIC architecture utilizing MEMS with mechanical latching, enabling stable passive operation without any power connection once configured. To ensure practical applicability, we present a system-level solution including both this hardware innovation and an accompanying automatic error-resilient configuration algorithm. The algorithm compensates for the lack of continuous tunability inherent in the non-volatile hardware design, thereby enabling such new operational paradigm without compromising performance, and also ensuring robustness against fabrication errors. Functional simulations were performed to validate the proposed scheme by configuring five distinct functionalities of varying complexity, including a Mach-Zehnder interferometer (MZI), a MZI lattice filter, a ring resonator (ORR), a double ORR ring-loaded MZI, and a triple ORR coupled resonator waveguide filter. The results demonstrate that our non-volatile scheme achieves performance equivalent to conventional PPICs. Robustness analysis was also conducted, and the results demonstrated that our scheme exhibits strong robustness against various fabrication errors. Furthermore, we explored the trade-off between the hardware design complexity of such non-volatile scheme and its performance. This study establishes a viable pathway to a new generation of power-connection-free PPICs, providing a practical and scalable solution for future photonic systems.

We propose a high-precision photonic computing-in-memory scheme enabled by 2D block-addressable VCSELs for optical neural networks, and experimentally demonstrated the multiply-accumulate operations using 3-bit input operands and 1-bit weights.

Photonic computing-in-memory (PCIM) based on non-volatile phase change materials (PCMs) offers a promising route to bypass the von Neumann bottleneck. However, its scalability and practical deployment are constrained by the integration complexity of the external optical devices required for PCM programming. Here, we propose and experimentally demonstrate an integrated PCIM module that overcomes this limitation by integrating a 2D-addressable vertical-cavity surface-emitting laser (VCSEL) array with a PCM memory bank via flip-chip technology. This approach eliminates the need for discrete fiber-optic components for PCM control. Concurrently, the 2D row-column addressing scheme reduces the required electrical interconnects for an N × N array from O(N2) to O(N), significantly alleviating wiring congestion. As a proof-of-concept, we developed a 32 × 32 VCSEL array heterogeneously integrated with a SiN-SOI crossbar memory bank comprising GST cells. The integrated cells exhibit over 70 distinct levels (>6 bits) within an 11 dB optical dynamic range, showcasing excellent multilevel capability. Leveraging this hardware, we partitioned the memory bank to simultaneously execute discrete Fourier transform (DFT), discrete cosine transform (DCT), and convolution operations via wavelength-division multiplexing (WDM), demonstrating high-fidelity parallel signal and image processing. This work establishes a scalable and high-density integration pathway for PCIM, addressing a critical barrier and paving the way for its practical deployment in advanced optical computing.

In the big-data-driven artificial intelligence era, similarity search, as a core operation in machine learning and data mining, demands high speed, energy efficiency, and scenario adaptability. Conventional electronic content-addressable memory (ECAMs) suffer from inherent RC delay bottlenecks, whereas existing optical content-addressable memory (OCAMs) are restricted by fixed bit-widths and limited distance metrics. In this work, we propose a variable bit-width all-optical CAM leveraging multi-segment modulators and phase-change material (PCM) Sb2Se3. The multi-segment memory unit (MSMU) therein compresses N-bit binary data into a single analog photonic unit, supporting direct data writing/loading without digital-to-analog converters (DACs) and flexible trade-offs between precision, storage capacity, noise immunity, and energy while enabling Hamming and nonlinear distance metrics. A six-element three-bit OCAM prototype was fabricated on a silicon nitride silicon-on-insulator (SiN-SOI) platform. Despite the absence of integrated high-speed phase shifters, the device still achieves reliable optical data storage and retrieval. K-nearest neighbor (kNN) simulations based on experimentally derived statistical data—validated on the iris, wine, and breast cancer datasets—show that the three-bit operating mode achieves classification accuracy comparable to Manhattan/Euclidean distances at high signal-to-noise ratios (SNRs), while the one-bit mode exhibits strong noise robustness. Energy consumption is 364 fJ/bit (3-bit) and 890 fJ/bit (1-bit). This work provides a high-speed, energy-efficient, and reconfigurable all-optical similarity search solution with experimentally verified device performance and dataset-validated applicability, showing great potential for widespread deployment in data-intensive machine learning and data-mining applications.

In the AI era of big data explosion, similarity search—a core task in machine learning and data mining—requires high speed, energy efficiency, and scenario adaptability. Conventional electronic CAMs face RC delay bottlenecks, while existing OCAMs are limited by fixed bit-widths and limited distance metrics. Here, we demonstrate a variable bit-width all-optical CAM architecture employing phase-change material Sb₂Se₃ integrated with Mach-Zehnder Interferometers (MZIs). The proposed multi-segment memory unit (MSMU) compresses N-bit binary data into a single analog photonic unit, supporting direct data writing/loading without DACs and flexible trade-offs between precision, storage capacity, noise immunity, and energy, while enabling Hamming and non-linear (NL) distance calculations. A 6-element 3-bit OCAM fabricated on a SiN-SOI platform realizes reliable storage and retrieval. kNN simulations on iris, wine, and breast cancer datasets show that the 3-bit mode achieves accuracy comparable to Manhattan/Euclidean distances under high SNR, while the 1-bit mode offers robust noise immunity. Energy consumption is 364 fJ/bit (3-bit) and 890 fJ/bit (1-bit). This architecture provides a high-speed, energy-efficient, and flexible all-optical similarity search solution, promising wide applications in machine learning and data mining.

The fractional Fourier transform (FrFT) plays a crucial role in multidimensional signal processing for applications ranging from synthetic aperture radar to optical encryption. However, conventional implementations-such as those based on bulky 4f lens systems or kilometer-scale fiber arrays-suffer from limitations in reconfigurability and integrability. In this work, we present a programmable discrete FrFT (DFrFT) processor based on a fixed array of basic transformation units (BTUs) interconnected via a dynamically reconfigurable architecture. Leveraging the order additivity of DFrFT, this processor synthesizes arbitrary DFrFT matrices with orders spanning the full 0 to 2π range at π/8-order resolution. Using the inverse design method, four BTUs (π/8, π/4, π/2, π) were designed with fidelity values exceeding 0.995 in simulation. Numerical analysis of assembled high-order DFrFT matrices demonstrates fidelities>0.989 across all 16 transformation orders. Experimental characterization on a silicon photonic platform verified both individual BTU performance (fidelity >0.85) and assembled DFrFT operations (fidelity>0.8), confirming scalable performance for integrated photonic signal processing. Order mapping strategy and fabrication tolerance are further evaluated to confirm the system's modularity, reconfigurability, and robustness. This processor establishes a scalable and programmable platform for integrated optical signal processing, particularly suited to time-frequency transformations and non-stationary signal analysis in future optical computing systems.

We propose in-NIC AllReduce to lower communication times, adapting to optically switched GPU network for distributed training. The experimental results show a 1.89. acceleration aligning with a theoretical analysis of 1.90. acceleration on average.

A polarization-splitting grating coupler (PSGC) is a key component for silicon photonic integrated circuits (PICs), which achieves light coupling from/to a fiber for both polarizations. We propose a high-performance PSGC designed by an inverse design method. The PSGC is designed to be able to be fabricated through a single etching step with outstanding fabrication tolerance. Experimental results indicate a peak insertion loss (IL) of-1.05 dB and a polarization-dependent loss (PDL) of 0.04 dB at 1534.7 nm, which are the lowest on current 220nm-silicon on insulator (SOI) platforms, to the best of our knowledge. Additionally, the 1-dB bandwidths of IL and PDL are 48 nm and 68 nm respectively. It is worth mentioning that the PSGC works under 0 degrees injection of fiber, which reduces packaging costs. (c) 2025 Optica Publishing Group under the terms of the Optica Open Access Publishing Agreement

We present an inverse-designed integrated optics device achieving a π/4-order discrete fractional Fourier transform (DFrFT) with fidelities of 0.995 (simulation) and 0.833 (experiment). This scalable design enables the implementation of higher-order DFrFTs through additivity.

Based on the principle of HiDM, we propose a new LUT construction method that achieves low loss and low complexity rate adaptation under 16QAM and 64QAM modulation.

Fabrication imperfections must be considered during the configuration of programmable photonic integrated circuits[PPICs].Therefore,characterization of imperfections is crucial but challenging,especially for PPICs based on recirculating waveguide meshes.In this Letter,we propose a characterization method based on an optimization method assisted by a step-by-step parameter space reduction technique,capable of greatly broadening the range of characterized parameters compared to existing methods.Our method ensures precise characterization,enabling the modeling of defective meshes with an error of 0.35 dB.Furthermore,the method was tested under various scenarios to evaluate its stability and robust-ness.Finally,we applied our method to implement six different types of finite/infinite impulse response[FIR/IIR]filters to demonstrate its effective application in off-chip configuration.

We present a virtual-physical hybrid testbed for system-level validation of joint optical transmission and switching in LEO satellite networks, confirming SDN-controlled reconfigurable paths with switching module insertion loss < 18 dB.

A polarization-splitting grating coupler (PSGC) is a key component for silicon photonic integrated circuits (PICs), which achieves light coupling from/to a fiber for both polarizations. We propose a high-performance PSGC designed by an inverse design method. The PSGC is designed to be able to be fabricated through a single etching step with outstanding fabrication tolerance. Experimental results indicate a peak insertion loss (IL) of -1.05 dB and a polarization-dependent loss (PDL) of 0.04 dB at 1534.7 nm, which are the lowest on current 220nm-silicon on insulator (SOI) platforms, to the best of our knowledge. Additionally, the 1-dB bandwidths of IL and PDL are 48 nm and 68 nm respectively. It is worth mentioning that the PSGC works under 0 degrees injection of fiber, which reduces packaging costs.

We proposed a programmable in-memory photonic computing scheme, where large PCMs array was controlled by using spatial light from the VCSELs array. Experiment of edge detection verified its feasibility.

We present a simulation-based study of ground-to-LEO uplinks using a steady optical beam (SOB) to mitigate turbulence-induced power loss. By comparing SOB and conventional Gaussian beam (GB) coupling into single-mode fibers, we characterize their received-power distributions and evaluate photon-noise-limited spectral efficiency. Across weak, moderate, and strong turbulence, the SOB yields narrower, higher-power statistics (an average power gain of 2–3dB, with up to 5.5dB guaranteed at the 90% level) and 1–2bps/Hz gains in guaranteed capacity. These results demonstrate SOB’s promise for robust, high-capacity satellite uplinks.

The flat topology of LEO satellite networks can impact communication performance. This paper demonstrates through experiments that elastic optical link switching enables path optimization. Additionally, we proposed a topology reconfiguration algorithm. Through comparison with the + Grid, our proposed method demonstrates better performance.

Atmospheric turbulence significantly hinders the reliability of satellite-to-ground laser communication, posing a particularly severe challenge for Geostationary Earth Orbit (GEO) links. Adaptive optics (AO) combined with mode diversity reception (MDR) is an effective turbulence mitigation strategy (referred to as AO+MDR). This paper leverages a GEO downlink simulation model to systematically investigate the power gain and spectral-efficiency gain achieved by AO, MDR, and their combined approach across various turbulence intensities. A key focus is a systematic examination of the interaction mechanisms within the AO+MDR method. Notably, we establish a novel finding: within the AO+MDR framework, AO and MDR exhibit a mutually weakening effect under weak turbulence, while demonstrating a mutual enhancement under strong turbulence. This significant observation has been further corroborated through desktop experiments.