We propose and experimentally demonstrate, to the best of our knowledge, the first silicon-on-insulator (SOI) wavelength triplexer operating at 1310, 1550, and 2000 nm. The device is realized on a standard 220 nm SOI platform using a subwavelength-grating engineered 2 x 2 directional coupler with an ultra-compact footprint of only 8.374 & micro;m. Leveraging the distinct photonic band properties of the subwavelength waveguide gratings, the 1310 nm channel operates in the Bragg reflection regime, while the 1550 and 2000 nm channels are routed in the subwavelength guiding regime, enabling efficient three-channel separation within a single coupler. Experimental results show insertion loss below 1 dB and crosstalk better than-15 dB across the three channels, with bandwidths of similar to 40 nm (1310 nm), similar to 55 nm (1550 nm), and similar to 60 nm (2000 nm), respectively.

Mode-division multiplexing (MDM) is a key strategy for boosting the capacity of optical interconnects, and its use in reconfigurable networks requires multimode optical switches with low loss, low crosstalk, and wide bandwidth. We propose and experimentally demonstrate a scalable, high-performance silicon multimode optical switch based on a robust Mach-Zehnder interferometer (MZI) architecture for arbitrary mode-to-dual-mode conversion. The switch integrates an arbitrary mode-to-single-mode power splitter, a passive pi phase shifter (for odd input modes), a thermo-optic phase shifter, and a single-mode-to-dual-mode combiner, enabling any guided mode in the multimode bus to be thermally switched to either the fundamental or first-order mode. A dual-mode demultiplexer then routes these modes to the desired output port. The design requires only a minimal set of fundamental building blocks, with identical intermediate component layouts for all input modes, eliminating the need for redesign when scaling to higher-order modes. By carefully engineering these components, all intermediate devices exhibit superior performance over a bandwidth exceeding 300 nm, ensuring robustness against fabrication variations and offering bandwidth advantages over existing devices. As a proof of concept, we fabricated a 1 x 2 multimode switch capable of independently routing four TE modes (TE0-TE3). Experiments show crosstalk below -18 dB across 1500-1600 nm for all modes, with a maximum modal loss of 3 dB and a switching power consumption of only 34.5 mW. Compared with existing approaches, the proposed device achieves low loss, reduced crosstalk, broader bandwidth, and improved scalability.

We propose and experimentally demonstrate a compact, ultra-broadband and dual-polarization multimode waveguide crossing. This design employs a two-dimensional subwavelength grating array to regulate the equivalent refractive index around the crossing region, thereby forming bound states. To validate the structural model, we design and fabricate a waveguide crossing supporting 12 modes (TE0-TE5 and TM0-TM5) on the standard 220 nm silicon-on-insulator platform using a single-step etching process. The device features a compact footprint of 13 & times; 13 mu m(2) and operates within a 12-mode dual-polarization high-performance bandwidth covering 1350-1600 nm, where all modes maintain insertion losses below 0.95 dB and crosstalk levels below -20 dB. For TE polarization, the operational range further extends from 1350 to 2100 nm, while TM polarization exhibits a broad spectral response from 1200 to 1600 nm. Experimental results show that the 12-mode waveguide crossing device achieves insertion loss < 1.08 dB/<0.63 dB (TE5/TM3) and crosstalk < -20 dB in the 1500-1600 nm range. Notably, by flexibly adjusting the waveguide width and the subwavelength grating array parameter N-2, this structure can be further expanded to support higher-order modes (such as TE0-TE9, TM0-TM9), demonstrating outstanding mode-expansion capability. The proposed device combines broadband performance, dual-polarization support, and a compact footprint, setting a record among multimode waveguide crossings. It effectively addresses the key challenge of dense integration in polarization- and mode-division multiplexing systems, providing a solid foundation for high-capacity on-chip optical interconnects.

To support the massive bandwidth and deterministic low-latency requirements of next-generation intra-data center interconnects for distributed artificial intelligence (AI) clusters, optical transmission is evolving toward ultra-wideband (UWB) operation over hollow core fibers (HCF). However, the chromatic dispersion slope of HCF creates a wavelength-dependent equalization complexity for high-speed intensity modulation and direct detection (IM/DD) systems. To overcome the rigidity of conventional static equalizers and empower the receiver with intelligent, spectrally-aware processing capabilities, dynamic adaptation to this non-uniform complexity scenario is essential. In this paper, we propose an autonomous band-adaptive equalization framework using a mixture-of-experts neural network (MoENN) architecture. By employing an end-to-end multi-task learning strategy enabled by a soft-gating training mechanism, we integrate a gating router that autonomously identifies the operating spectral band and a parallel bank of four expert networks with hierarchical model capacities. We experimentally demonstrate 100 GBaud PAM4 transmission over a 6.1 km HCF link covering the O, S, C, and L bands. The results verify that the proposed MoENN achieves bit error rate (BER) performance superior to a global neural network across the entire spectrum. Furthermore, by transitioning to a hard-gating inference mechanism during deployment, the system realizes spectrally-aware computational scaling, reducing complexity by approximately 80.4% in the O band and 49% in the S/C bands relative to the static baseline. This establishes the MoENN as a scalable, energy-efficient enabler for wavelength-agnostic reconfigurable optical switching fabrics.

Efficient on-chip mode manipulation is essential for scalable and flexible filtering devices in mode-division and hybrid multiplexing systems. A multimode manipulation strategy based on band engineering in one-dimensional grating-based photonic crystal waveguides (PCWs) is proposed. By strategically widening the transverse dimension of a silicon PCW, both the number of supported Bloch modes and their bandgaps are expanded, enabling broadband mode control across three distinct regimes: all-pass, selective-pass, and all-block. Based on this principle, a scalable higher-order mode (HOM) pass-filtering architecture and three compact HOM-pass filters (TE1-pass, TE2-pass, TE3-pass) are demonstrated. Devices fabricated on standard silicon-on-insulator (SOI) exhibit measured bandwidths exceeding 200 nm, 123 nm, and 70 nm, respectively, with extinction ratios (ERs) greater than 20 dB and insertion losses (ILs) below 2 dB. Leveraging these filters as building blocks, we further demonstrate their capability to achieve high ER, low IL filtering devices for polarization and wavelength applications, including a polarizer, a polarization beam splitter, and a wavelength demultiplexer. This work establishes a scalable multimode bandgap-engineering framework for compact, broadband, and multifunctional silicon photonic devices for mode, polarization, and wavelength manipulation.

We experimentally demonstrate an ultra-wideband and high-speed optical transmission system based on intensity modulation and direct detection (IM/DD) with Nyquist 4-ary pulse amplitude modulation (PAM4), enabled by a 6.1 km span of hollow-core nested antiresonant nodeless fiber (NANF). This NANF exhibits significantly lower attenuation compared to previously reported ultra-wideband variants. To mitigate the impairments of high-baud rate signal induced by the accumulated chromatic distortion, a Tomlinson-Harashima precoding (THP) in conjunction of linear FFE scheme is applied. Experimental results show the successful transmission of PAM4 signal exceeding 200 Gb/s per wavelength across four major optical communication bands—the O, S, C and L bands for the first time. This achievement marks a critical advancement toward high-capacity IM/DD optical interconnects using hollow-core fiber technologies.

Flexible mode manipulation is essential for the large-scale construction of mode-division multiplexing (MDM) systems, enabling arbitrary signal exchanging, routing, and multiplexing to significantly enhance optical interconnection capacity. A compact and flexible mode manipulation scheme is demonstrated, centered on a mode exchanger (ME) optimized by intelligent algorithms. Taking four-mode exchangers based on a Bézier curve profile waveguide and corresponding mode add-drop multiplexers (MADMs) as examples, theoretical analysis and experimental verification of their flexible mode manipulation capabilities are presented. Experimental measurements on the fabricated ME01/23, ME02/13, and ME03/12 mode exchangers reveal insertion loss (IL) below 1.8 dB and crosstalk (CT) better than -10 dB, spanning the wavelength ranges of 1515-1570 nm, 1508-1580 nm, and 1519-1580 nm, respectively. All three devices feature a compact footprint of less than 6 × 3 μm2. Meanwhile, characterization of the three-mode add-drop multiplexers (MADMs) operating over the C-band (1530-1565 nm) verifies IL of less than 2.4 dB and CT exceeding -10.0 dB for all transmission modes. Notably, the proposed mode manipulation strategy offers theoretical scalability to higher-order modes, overcoming a key limitation of conventional MDM systems: the inability to directly access specific low-order modes in multimode bus waveguides. These findings thus provide critical technical foundations for integration into next-generation high-density optical communication systems.

We propose a versatile on-chip mode manipulation architecture based on arbitrary-to-single-mode power dividers (ASPDs), enabling flexible mode conversion and routing in multimode photonic systems. The architecture comprises two oppositely cascaded ASPD units arranged in a Mach-Zehnder-interferometer-like configuration, leveraging standard single-mode components such as phase shifters and mode routers to achieve highly flexible multimode manipulation. This approach offers two key advantages: a compact, scalable design based solely on ASPD units and standard single-mode components, and direct decomposition of higher-order modes into the fundamental mode, simplifying routing and removing the need for complex multimode elements. As a proof of concept, several ultra-compact ASPDs were designed using subwavelength grating structures optimized via intelligent algorithms. For commonly used low-order modes (TE0-TE3), the devices exhibit footprints below 9.5 µm and maintain excess losses (ELs) below 0.35 dB across a 300 nm bandwidth (1400-1700nm). Notably, TE0/TE1 ASPDs achieve ELs < 0.15 dB over a record-wide 500 nm range (1300-1800nm). Based on these ASPDs, we experimentally demonstrate flexible mode conversion and switching, confirming their effectiveness for multimode signal control. Furthermore, numerical results show that the scheme can be extended to higher-order modes (TE4-TE7) while maintaining excellent performance and compact footprints (8 µm), and can be combined with standard single-mode crossings for flexible multimode interconnections. These results validate the feasibility and scalability of the proposed ASPD-based platform, offering a promising route toward reconfigurable multimode photonic integrated circuits.

Quantum key distribution (QKD) makes use of the principles of quantum mechanics to enable provably secure communication1,2. One substantial challenge persists in building large-scale QKD networks with many clients over long communication distances3. Although quantum relays continue to pose practical difficulties4, existing trusted-node networks5-9, point-to-multipoint networks10,11 and wavelength-multiplexed entanglement networks12,13 encounter issues such as reliance on trusted intermediaries or limited distances. Twin-field quantum key distribution (TF-QKD) provides a compelling architecture that can overcome those issues while enhancing communication distance14. Although long-distance point-to-point TF-QKD has been achieved15-21, realizing large-scale networks requires scalable quantum devices. Here we report a proof-of-principle demonstration of an integrated-photonics TF-QKD network with exceptional scalability and reliability. This network includes 20 independent client-side QKD transmitter chips with one server-side optical microcomb chip. The microcomb generates a broad range of ultralow-noise coherent frequency combs with Hz-level linewidths, which serve as seeds and references for all client chips. Each client chip regenerates ultralow-noise light phase-locked to microcombs and prepares quantum keys. We sequentially implement pairwise QKD across 20 client chips through ten wavelength-multiplexed channels, with each surpassing the repeaterless bound at 370 km in spooled fibre, achieving a networking capability (client pairs × communication distance) of 3,700 km. We further demonstrate the wafer-scale reproducibility of both server-side microcomb chips and client-side QKD transmitter chips, together establishing system-level scalability. Combining mass-manufacturability, cost-effectiveness and high scalability of integrated photonics with long-distance quantum communication represents a viable path to large-scale quantum networks.

As a key routing component in optical interconnection systems, waveguide crossings directly affect the transmission efficiency and reliability of multimode optical signals. Existing multimode waveguide crossings face difficulty in simultaneously balancing integration density, transmission loss, and bandwidth compatibility. To address this challenge, this study has, for the first time, introduced the smooth geometric properties of Bezier curves into the optimization of multimode cross structures. By integrating with an optimization algorithm, coordinated regulation of four transmission modes and dual polarizations was achieved, thus enabling the successful development of 2×2 and 3×3 multimode waveguide crossing devices. Experimental results demonstrate that at 1550 nm, the insertion losses (ILs) of the 2×2 and 3×3 devices are as low as 1.23 dB and 1.08 dB, respectively, while the crosstalk (CT) values are as low as −29.67 dB and −23.99 dB. Stable transmission performance is maintained across the wide wavelength range of 1500–1600 nm. The device footprints are only 7.2 × 7.2 μm2 (for the 2×2 device) and 11.3 × 10.6 μm2 (for the 3×3 device), which exhibit significantly higher integration density compared to existing designs. This design overcomes the inherent performance trade-offs of traditional structures. The integration of Bezier curves with optimization algorithms provides a novel design approach for multimode photonic devices. Its excellent process compatibility and superior performance advantages are expected to accelerate the deployment of integrated photonic systems in data centers, artificial intelligence (AI) chips, and other related fields.

Mode exchangers serve as critical functional units in on-chip mode-division multiplexing systems, where their compactness and performance metrics directly influence the integration density and transmission capacity of communication systems. To address the technical bottlenecks of existing devices in mode scalability and conversion efficiency, this study proposes a novel design of ultra-compact multi-mode exchangers based on Bezier curve waveguides. Through parametric optimization of the functional region using a direct binary search algorithm, three novel mode exchangers (ME00/12, ME01/22, and ME02/11) are successfully designed, with all physical dimensions confined within 4.8 x 1.8 mu m(2). Theoretical simulations reveal that these devices achieve insertion losses < 1.0 dB and crosstalk < - 10.0 dB across wavelength ranges of 196 nm (1400-1596 nm), 187 nm (1513-1700 nm), and 155 nm (1515-1670 nm), respectively. These operational bandwidths cover the C + L primary communication bands and parts of the S/U extended bands. Experimental validation within the 1500-1580 nm range confirms the theoretical predictions. Compared to previously reported designs, the proposed devices exhibit improved mode conversion performance. Furthermore, a mode add-drop multiplexing system incorporating ME00/12 and ME02/11 was constructed. Experimental results demonstrate insertion losses < 2.4 dB and crosstalk < - 10.8 dB across all mode channels within the 1530-1570 nm operational window. The presented mode exchangers and mode add-drop multiplexing systems provide engineering-viable solutions for constructing high-performance mode-division multiplexing systems.

We present the first single-channel 1.001-Tb/s DP-36QAM-PCS recirculating transmission over 73 loops of 146.77-km ultra-low-loss low-IMI DNANF-5 fiber, achieving a record transmission distance of 10,714.28 km.

High-performance polarization beam splitters (PBSs) are essential for maintaining signal integrity and minimizing interference in advanced photonic applications. This paper experimentally demonstrates a high-performance, ultra-broadband on-chip silicon PBS that integrates a mode conversion Bragg grating (MC-BG) with an adiabatic asymmetric directional coupler (ADC). The MC-BG is implemented by etching a hole-structured grating array into a multimode waveguide, serving as a polarization-dependent broadband mode router. It directs the input TM 0 and TE 0 modes in opposite directions while simultaneously converting the TE 0 mode into a higher-order TE 2 mode. The transformed TE 2 mode is then extracted from the multimode waveguide via a specially designed broadband adiabatic ADC, exiting through the cross port. This design achieves theoretically high performance with an extinction ratio (ER) of >30 dB/15 dB and an insertion loss (IL) of <0.9 dB/2 dB over bandwidths of 265 nm (1410-1675 nm) and 375 nm (1300-1675 nm), respectively. Fabricated on a standard 220 nm silicon-on-insulator (SOI) wafer using a single-step etching process, experimental results reveal that, for a device length of 91.7 μm, the bandwidth with ERs exceeding 30 dB spans more than 220 nm. To our knowledge, this is the first demonstration of a PBS on a 220 nm SOI platform achieving an experimentally verified ER exceeding 30 dB over a bandwidth greater than 200 nm. Furthermore, our analysis indicates that performance can be further improved by incorporating a TM 2 mode stripper, extending the 30 dB bandwidth to nearly 295 nm.

Multifunctional integrated photonic devices are essential for advancing high-density photonic circuits. In this work, we propose and experimentally demonstrate an ultra-broadband device that simultaneously enables polarization separation and 3 dB power splitting for the TM mode. The design exploits the anisotropic properties of subwavelength gratings (SWGs), incorporating transverse and longitudinal SWGs in both transition and coupling regions. TE-polarized light is guided through the central path, while TM-polarized light is symmetrically split into lateral SWGs. The device functions as a broadband polarizer for TE input (1260–1675 nm), with excess loss (EL) below 1 dB and crosstalk (CT) below −20 dB. For TM input, 3 dB splitting is achieved over 1400–1660 nm, with EL < 0.5 dB and CT < −22 dB at 1550 nm. The compact footprint (21 × 6 µm 2 ) and CMOS-compatible fabrication further enhance its practical relevance. Experimental results confirm efficient dual-function performance, with low loss and high polarization extinction across a wide bandwidth.

We propose and experimentally demonstrate, to the best of our knowledge, the first silicon-on-insulator (SOI) wavelength triplexer operating at 1310, 1550, and 2000 nm. The device is realized on a standard 220 nm SOI platform using a subwavelength-grating engineered 2×2 directional coupler with an ultra-compact footprint of only 7.99 μm. Leveraging the distinct photonic band properties of the SWG, the 1310 nm channel operates in the Bragg reflection regime, while the 1550 and 2000 nm channels are routed in the subwavelength guiding regime, enabling efficient three-channel separation within a single coupler. Experimental results show insertion loss below 1 dB and crosstalk better than –15 dB across the three channels, with bandwidths of ~40 nm (1310 nm), ~55 nm (1550 nm), and ~60 nm (2000 nm), respectively.

Spectrum manipulation is central to photonic systems, where advanced computing and sensing applications often demand highly complex spectral responses to achieve high throughput. Conventional methods for enhancing spectral complexity typically rely on cascading discrete photonic components, resulting in a complexity that scales only linearly with the number of components. Here, we introduce hyper-spectral photonic integrated circuits (HS-PICs), in which spectral complexity scales exponentially with the number of components. This is achieved through recursive inverse design - a system-level inverse design strategy that exploits intricate inter-component interactions as design freedoms, thereby substantially expanding the design space for spectral engineering. Using this approach, we demonstrate that even a single waveguide structure can resolve spectra with sub-picometer resolution, surpassing the performance of current state-of-the-art spectrometers. This performance bridges optical and microwave frequencies in spectral analysis, enabling simultaneous monitoring of optical and radio signals within a single device. Our work establishes a transformative framework for next-generation computing and sensing technologies.

We present and experimentally demonstrate a compact, ultra-broadband, and scalable multi-port waveguide crossing architecture that supports dual-polarization operation across both the O–L communication bands and the emerging 2 μm mid-infrared (MIR) window. The design is based on a composite subwavelength grating structure and implemented on a standard 220 nm silicon-on-insulator (SOI) platform using a single-step etching process. A unified MMI-based building block enables highly compact 2×2, 3×3, and 4×4 configurations with symmetric layouts, achieving footprints as small as 14×14 μm² for the international telecommunication union (ITU) bands and 16×16 μm² at 2 μm. The 2×2 crossing demonstrates record-wide bandwidths of ~380 nm (1300–1680 nm) and ~400 nm (1760–2150 nm), with low insertion losses of <0.3 dB / <0.2 dB (TE/TM) and crosstalk below −20 dB / −30 dB across the ITU bands. Even better performance is achieved for the TE mode in the 2 μm MIR range. To the best of our knowledge, this is the first demonstration of a dual-polarization waveguide crossing with such extensive bandwidth coverage in the ITU bands and the widest reported in the 2 μm MIR regime. The proposed architecture offers a scalable, fabrication-friendly solution for next-generation high-density photonic integrated circuits with polarization- and wavelength-division multiplexing capabilities across multiple spectral domains.

To overcome the polarization sensitivity bottleneck of mode splitters in traditional mode-division multiplexing (MDM) systems, this study proposed and experimentally demonstrated a silicon-based dual-polarization mode splitter. Fabricated on a silicon-on-insulator (SOI) platform, the device adopts a subwavelength circular hole array as its functional core. Optimized via the direct binary search (DBS) algorithm, which converged after six iterations, the device achieved ultra-compact integration with a footprint of 5.1 × 2.4 µm2. Without polarization preprocessing, it enables simultaneous separation of TE0/TE1 and TM0/TM1 mode pairs. Experimental results show that at 1550 nm, the insertion losses (IL) of TE0 (TE1) and TM0 (TM1) were 1.25 dB (0.7 dB) and 1.29 dB (1.09 dB), respectively, with corresponding crosstalk (CT) values of -21.54 dB (-19.3 dB) and -12.0 dB (-13.34 dB). Across the 1500-1580 nm wavelength range, the CT of all operating modes remains below -10.9 dB. To the best of our knowledge, this is the first reported dual-polarization mode splitter with orderless conversion, multi-polarization compatibility, compact integration, and broadband operation, which lays the foundation for the development of the next generation compact, high-performance, high-capacity optical interconnection systems.

Anti-resonant hollow-core fiber (AR-HCF) has exhibited the revolutionary potential in wide-spectrum optical transmission. We propose and demonstrate the co-transmission of O, S, C, L-band signals in a single AR-HCF. Enabled by 80Gbaud PAM8 modulated four independent tunable lasers in O, S, C, L-bands and linear equalizers, we achieve the 960Gb/s (240Gb/s/λ × 4 λ) optical interconnect over 6.2km AR-HCF at 20% FEC limit of 2.4 × 10-2. By adjusting the laser wavelengths, the measured useful bandwidths within the 20% FEC limit are 35nm, 35nm, 30nm and 20nm in O, S, C, L-bands, respectively, which provides an upgrade solution to achieve high-capacity optical interconnects with 200Gb/s/λ and beyond.

Recently, the NGPON2 standards organization has decided to deploy LDPC code in upstream and downstream channels in next-generation Ethernet passive optical network (NG-EPON) systems. However, PON upstream channels operate in burst-mode, hence the turn-on effects in optical network units (ONUs) may introduce burst errors, which are difficult for LDPC to deal with. One of the solutions is introducing an interleaver to convert consecutive burst errors into separate random errors. The traditional interleavers such as the block interleaver and the random interleaver, only consider dispersing the consecutive errors. In this paper, we present a design methodology for channel-adaptive interleaving pattern by joint optimizing of interleaving and LDPC decoding in PON systems deploying LDPC codes. After determining the interleaving pattern where the bits can be recovered with more reliable messages during LDPC iterative decoding, we map the bits from the potential locations of burst errors to the interleaving positions. By doing so, we can not only disperse the burst errors but also make full use of the LDPC code to improve its decoding performance. The numerical results show that the proposed interleaving scheme has a better performance under any burst-error length.