High-precision synchronization, empowering sub-centimeter positioning and simplified digital signal processing (DSP), is essential for next-generation multifunctional, high-capacity and energy-efficient fronthaul networks. However, simultaneously achieving sub-picosecond time synchronization and high-throughput fronthaul transmission within a unified network remains a technical challenge. Here, we propose an electro-optic comb-based bidirectional fronthaul architecture that seamlessly integrates advanced telecommunication and time-frequency synchronization. Pilot comb lines, acting as high-fidelity carriers for clock delivery, are bidirectionally transmitted, enabling real-time phase drift compensation and achieving 0.25 ps root-mean-square time synchronization stability. Meanwhile, the forward pilots also facilitate remote-site comb regeneration, providing local oscillators for self-homodyne detection. Moreover, by harnessing this dual capability, we achieve network-synchronous one-sample-per-symbol coherent detection at the Nyquist limit across all parallel channels, eliminating carrier-phase and sampling-clock-related DSP while substantially reducing the complexity of analog-to-digital conversion and the remaining DSP. Our proposed architecture offers a scalable prototype for 6G multifunctional coherent-lite fronthaul networks. The work demonstrates an electro-optic comb-based fronthaul architecture that unifies subpicosecond time-frequency synchronization with high-capacity transmission, enabling energy-efficient one-sample-per-symbol coherent detection for 6G networks.

Beyond supporting ultra-high-capacity data transmission, metropolitan and access networks are expected to enable real-time infrastructure monitoring, driving the emergence of integrated sensing and communication (ISAC). Distributed acoustic sensing (DAS) has proven to be well-suited to urban sensing application requirements, yet its seamless integration into ISAC remains challenging—conventional high-peak-power sensing pulses in DAS induce nonlinear crosstalk in communication channels. DAS inherently suffers from interference fading due to single-frequency laser sources, which limits sensitivity. Here, we propose an ISAC architecture based on an electro-optic (EO) comb and a 7-core fiber, achieving nonlinearity-suppressed self-homodyne transmission and fading-suppressed DAS. Unmodulated comb lines and sensing pulses are polarization-multiplexed into orthogonal polarization states within the central core to minimize nonlinear crosstalk while delivering local oscillators (LOs) for wavelength division multiplexing (WDM) coherent transmission within six outer cores—achieving 10.56 Tbit/s capacity. In addition to supporting WDM transmission, the EO comb’s wavelength diversity is also exploited to enhance DAS performance. Specifically, a dual-pulse probe loaded onto four comb lines yields a 6 dB signal-to-noise ratio gain and a 64% reduction in fading occurrences, achieving a sensitivity of 1.72 pε/Hz with 8 m spatial resolution. Moreover, our system supports simultaneous multi-wavelength backscatter detection in sensing and simplified digital signal processing in self-homodyne communication, reducing receiver complexity and cost. Our work presents a scalable, energy-efficient ISAC framework that unifies high-capacity communication with high-sensitivity sensing, providing a blueprint for future intelligent optical networks.

We demonstrate win-win integrated-sensing-and-communication architecture enabled by a shared linear-frequency-modulated probe, simultaneously achieving unprecedented DAS sensitivity of 0.6$p\varepsilon /\sqrt {{\text{Hz}}} $ @10m spatial resolution, reconfigurable resolution down to 0.52m, and comb-based phase-related-DSP-free coherent-lite transmission of 62.4-Tb/s CPRI-equivalent-rate signals. © 2025 The Author(s)

We demonstrate electro-optic comb-based 18λ×384Gb/s superchannel for AIDC distributed training. It achieves clock synchronization and <0.9-ps sampling instant jitter, enabling low-power baud-rate-sampling coherent-lite reception with negligible 0.12dB long-term SNR penalty over 1 hour.

We propose an integrated fronthaul architecture that simultaneously realizes 1-ps timing jitter clock distribution and DSP-simplified 2.88-Tb/s self-homodyne transmission by comb cloning, yielding simple, low-cost solution for multi-functional 6G-network, empowering both large-capacity communication and high-precision positioning.

Beyond providing user access to the core network,the radio access network(RAN)is expected to support precise positioning and sensing for emerging applications such as virtual reality(VR)and drone fleets.To achieve this,fronthaul-the link connecting the central units/distributed units(CUs/DUs)to wireless remote units(RUs)in centralized RAN—must realize both high-capacity transmission and low-timing-jitter clock synchronization be-tween RUs.However,existing solutions fall short of supporting these functions within one simple,cost-effective network.In this work,we propose a solution that simultaneously achieves picosecond-level timing jitter clock distribution and Tb/s data transmission with simplified DSP,using an electro-optic(EO)comb cloning technique to enable multifunctionality in fronthaul systems.Through the delivery of pilot comb lines,a 1 ps(integrated from 1 Hz to 40 MHz)low-timing-jitter 100 MHz clock is distributed by the beating of adjacent pilot comb lines and subsequent frequency dividing,realizing frequency synchronization between the CUs/DUs and RUs.Moreover,the delivery of pilot comb lines also facilitates self-homodyne structures through EO comb cloning,and supports wavelength division multiplexing(WDM)transmission with a line capacity of 2.88 Tb/s and a net capacity of 2.5 Tb/s.Thanks to the clock-synchronized and self-homodyne structure,DSP is streamlined,with digital timing recovery,carrier phase estimation,and frequency offset estimation all omitted.This work lays the technical foundation for implementing a 6G WDM fronthaul architecture that integrates ultra-wide wireless bandwidth with precise positioning and sensing.

We propose and experimentally demonstrate an integrated sensing and communication (ISAC) system based on an electro-optic (EO) comb and 7-core fiber, simultaneously achieving 10.56 Tbit/s self-homodyne transmission and high-fidelity ϕ-OTDR sensing. By co-propagating communication local oscillators (LOs) and high-peak-power dual-pulse probes in orthogonal polarization states within the central core, we suppress pulse-induced nonlinear effects while preserving spatial channels for both dense Wavelength division multiplexing (DWDM) and space division multiplexing (SDM) transmission. Frequency-diverse sensing using four EO comb lines yields a 6 dB SNR improvement and a 64.4% reduction in fading. This architecture demonstrates a high-capacity, high-sensitivity, and low-complexity ISAC solution for metro-scale networks.

Optoelectronics could be used to develop fast and wideband information systems. However, the large frequency mismatch between optically synthesized signals and electronic clocks makes it difficult to synchronize optoelectronic systems. We describe an on-chip microcomb that can synthesize single-frequency and wideband signals covering a broad frequency band (from megahertz to hundreds of gigahertz) and that can provide reference clocks for the electronics in the system. Our synchronization strategy, which aligns optically synthesized signals and electronics, can provide signal manipulation precision and data transmission without coherent digital signal processing. To illustrate the capabilities of this approach, we create a wireless joint sensing and communication system based on a shared microcomb-based transmitter. An on-chip microcomb that provides reference clocks for all synthesized signals and electronics within the system can be used to unite time–frequency references in optoelectronics.

We demonstrate unprecedented 0.25-ps RMS long-term time-frequency synchronization for multifunctional WDM fronthaul using an EO comb-based bidirectional feedback architecture, simultaneously achieving 16.9-Tb/s CPRI-equivalent rate, the first power-efficient one-sample-per-symbol self-homodyne coherent-lite detection, and enabling sub-cm positioning. (c) 2024 The Author(s)

We propose and experimentally demonstrate a microwave sensing synchronized by a fully on-chip microcomb. This synchronization strategy shows enhanced precision in signal manipulation and enables a series detection and imaging experiments.

We demonstrate a wireless communication system and unite the time-frequency reference by using a microcomb. This synchronization strategy enhances precision in signal manipulation and reduces coherent digital signal processing. © 2024 The Author(s)

Based on an AI-accelerated silicon slow-light modulator chip, we realize 400 Gbps PAM-4 optical transmission per wavelength in a standard silicon photonic platform for the first time, leading to a total data capacity of 3.2 Tbps with an on-chip data-rate density of 1.6 Tb/s/mm(2). (c) 2025 The Author(s)

The next-generation radio access network is envisioned to support ultra-wide bandwidth, be highly reliable, and with low-latency wireless signal delivery in mobile fronthaul. In this work, we introduce a high-capacity self-homodyne digital-analog radio-over-fiber fronthaul architecture employing multi-core fiber. By delivering the unmodulated local oscillator (LO) through one of the cores, the influence of laser phase noise can be eliminated after coherent beating provided optical path matching. We experimentally demonstrate a single-wavelength 23.6-Tb/s common public radio interface (CPRI)-equivalent rate supporting the 1024-QAM modulation format. The remote LO can also serve as the optical source for up-link transmission with an aligned central wavelength. Moreover, we characterize and discuss the impact of the optical path mismatch-induced phase noise and random polarization rotation of remote LO, respectively. Further directions and practical implementations of DA-RoF are discussed. The results reveal the potential to upgrade the future fronthaul in terms of capacity, fidelity, and simplicity.

Silicon photonics is a promising platform for the extensive deployment of optical interconnections, with the feasibility of low-cost and large-scale production at the wafer level. However, the intrinsic efficiency-bandwidth trade-off and nonlinear distortions of pure silicon modulators result in the transmission limits, which raises concerns about the prospects of silicon photonics for ultrahigh-speed scenarios. Here, we propose an artificial intelligence (AI)-accelerated silicon photonic slow-light technology to explore 400 Gbps/λ and beyond transmission. By utilizing the artificial neural network, we achieve a data capacity of 3.2 Tbps based on an 8-channel wavelength-division-multiplexed silicon slow-light modulator chip with a thermal-insensitive structure, leading to an on-chip data-rate density of 1.6 Tb/s/mm2. The demonstration of single-lane 400 Gbps PAM-4 transmission reveals the great potential of standard silicon photonic platforms for next-generation optical interfaces. Our approach increases the transmission rate of silicon photonics significantly and is expected to construct a self-optimizing positive feedback loop with computing centers through AI technology.

Significance The rapid advancement of artificial intelligence, particularly deep learning, has created increasingly demanding requirements for hardware performance. Traditional electronic computing architectures encounter substantial limitations-including the deceleration of Moore's Law and persistent challenges from the "memory wall" and "power wall"-restricting their capacity to maintain performance improvements for large-scale, highly concurrent AI tasks. This widening gap between computational requirements and hardware capabilities necessitates the exploration of alternative computing paradigms to overcome these fundamental constraints. Optical computing, utilizing the inherent properties of photons, emerges as a highly promising solution. Among various optical computing approaches, photonic neural networks (PNNs) have attracted considerable attention. PNNs employ photons directly to perform essential mathematical operations fundamental to neural networks, such as vector-matrix multiplication, convolution, and nonlinear activation functions. This natural capability to execute computation in the optical domain provides significant advantages over conventional electronic methods, including ultra-high processing speed, extensive bandwidth for data throughput, inherent parallelism, and substantially reduced energy consumption through minimized data transfer latency. Consequently, PNNs have emerged as a critical research frontier bridging photonics, information science, and artificial intelligence, offering an innovative solution for next-generation high-performance AI hardware. This review thoroughly examines PNNs' core concepts, technological developments, and future directions. Progress This review systematically summarizes recent key technologies and progress in PNN physical implementations, organized by primary architectural types that have driven significant advancements in the field. PNNs based on diffractive optical elements, often referred to as diffractive optical neural networks (DONNs), harness the wave propagation of light through structured diffractive layers to perform all-optical deep learning inference. This architecture has demonstrated remarkable performance in tasks like complex image classification and reconstruction. Recent breakthroughs include the development of reconfigurable and programmable DONNs for multi-task learning, the integration of multi-dimensional multiplexing to significantly boost computational throughput, enhanced robustness against fabrication errors and environmental noise, and successful on-chip integration, paving the way for compact and efficient devices. PNNs based on Mach-Zehnder interferometer (MZI) arrays utilize reconfigurable MZI units to implement arbitrary linear optical transformations, establishing highly adaptable computational layers. Early theoretical designs have evolved into large-scale integrated MZI meshes that achieve high-accuracy classification and regression tasks, including complex-valued computations. Key advances include innovative architectural designs for enhanced scalability and energy efficiency, robust configurations addressing hardware imperfections and crosstalk, and sophisticated on-chip training methods for precise weight loading and adaptive operation in real-time. PNNs leveraging microring resonator (MRR) arrays utilize the distinctive wavelength-selective properties of microring resonators, particularly in wavelength division multiplexing (WDM) systems, to enable high-throughput parallel processing. The "broadcast-and-weight" architecture establishes a fundamental paradigm for MRR-based PNNs, enabling dynamic weight modulation and optical summation. Notable advances include sophisticated weight bank control for high-precision tuning, innovative architectural designs for integrated tensor computations and optical convolutions at impressive computation densities, and the integration of diverse functionalities for specialized applications, demonstrating their potential for ultra-compact and high-performance computing. PNNs based on cascaded modulator architectures achieve complex optical transformations through the sequential modulation of optical signals, offering structural simplicity and high integration potential. These architectures have demonstrated ultra-low energy consumption per operation and high accuracy in classification tasks like MNIST digit recognition. Recent advancements focus on direct cascaded modulator systems, robust hybrid optoelectronic integration for versatile control and non-linearity, coherent processing architectures for high-precision complex-valued computations, and programmable signal processors for reconfigurable and high-speed inference, pushing the boundaries of compact integrated photonic circuits. Finally, the implementation of optical nonlinear activation functions is crucial for enabling deep learning capabilities in PNNs, allowing networks to learn and process complex, non-linear relationships. Two primary categories are distinguished: optoelectronic hybrid methods, which convert optical signals to electrical for nonlinear processing before re-converting, and all-optical methods, which directly exploit intrinsic material nonlinearities or specific device effects (Figs. 20-22). Progress in this area is vital for constructing truly multi-layered PNNs that can break linearity and achieve high accuracy across diverse and challenging AI tasks. Conclusions and Prospects While PNN research has achieved significant progress, substantial challenges remain. These include achieving high level integration and scalability for complex tasks, improving power efficiency of active photonic components, enhancing robustness against manufacturing errors and environmental noise, realizing efficient all-optical nonlinear activation for deep networks, and developing practical on-chip optical memory. Future development requires multidisciplinary innovation, emphasizing novel materials and computing elements, co-design of hardware and algorithms, advanced photonic integration platforms, and expanding PNN applications into scientific computing, optimization, simulation, and advanced sensing. Addressing these challenges will enable PNNs to evolve from prototypes to practical solutions, establishing their position in post-Moore computing.

We realize a silicon photonic transmission link with an aggregate data rate exceeding 60 Tb/s and reduce the phase-related DSP consumption by 99.99999% using a self-injection locked microcomb as the light source.

A novel dual-BiGRU equalization model is proposed to enhance the performance of the implemented photonic-electrical integrated silicon transceiver, enabling 170 Gbps PAM4 signal generation per channel within a ~33 GHz electro-optical bandwidth.

We explore implementing a multilevel deep neural network to enhance the performance of a 4-channel photonic-electrical hybrid-packaged silicon transceiver. Stable transmission and reception of 150 Gbps/λ PAM4 signals are achieved, which shows the potential for beyond-400G optical interconnects.

We demonstrate unprecedented 2nm broadband ASE source-enabled digital-analog radio-over-fiber mobile fronthaul system with joint force of SOAs for intensity noise suppression and multicore fiber for self-homodyne detection. We achieve 35GHz(=7core×5GHz) aggregated bandwidth with 2Tb/s CPRI-equivalent data rate sup-porting 1024-QAM signal.