Main

Large-scale quantum communication networks promise secure information transfer among numerous clients over long distances2,3. Over the past few decades, notable milestones have been achieved in long-distance QKD in satellite and fibre-optic systems. Progress has been made towards various network models, for example, trusted-node networks5,6,7,8,9, point-to-multipoint networks10,11 and wavelength-division-multiplexing (WDM) entanglement networks12,13. Simultaneously scaling the number of quantum clients N and the communication distance L poses substantial challenges, hence necessitating advancements in protocols, architectures and hardware. Recently, advanced protocols, such as TF-QKD14, have been proposed to extend communication distances, enhancing key rates by the square root of channel transmittance, and experimental results have surpassed the repeaterless bound by distances of up to 1,000 km (refs. 15,16,17,18,19,20,21). Furthermore, TF-QKD operates under a measurement-device-independent (MDI) framework22, enabling untrusted central nodes and sharing of expensive resources among users on the shared central node. This architecture therefore holds the potential for large-scale cost-effective networks with large N and long L. TF-QKD implementation hinges on achieving ultrastable single-photon interference in large optical interferometers, requiring low-noise lasers and high-precision optical phase tracking. Despite notable advancements in protocols, the present implementation of TF-QKD remains primarily limited to point-to-point communications15,16,17,18,19,20,21. Practical implementations of TF-QKD networks face substantial technical challenges. These include development of many ultralow-noise coherent laser sources, massive manufacture of identical, high-performance QKD chips that maintain uniformity across all client nodes and system-level integration that enables scalable quantum node operation and quantum channel transmission, while remaining compatible with existing phase tracking. They collectively represent the fundamental requirements for TF-QKD network infrastructure, demanding simultaneous advances in photonic integration, manufacturing scalability and system control.

Here we present a proof-of-principle laboratory demonstration of a user-massively scalable and long-haul TF-QKD network enabled by integrated photonics, named the ‘Weiming Quantum Chip-Network’. Our integrated-photonics QKD network comprises a central server chip that integrates an optical microresonator frequency comb (microcomb) on a silicon nitride (Si3N4) platform and 20 copies of independent client-side QKD transmitter chips that monolithically integrate all essential components for key preparation based on indium phosphide (InP) photonic circuits. These client-side QKD chips fabricated on a 3-inch InP wafer and server-side microcomb chips fabricated on a 4-inch Si3N4 wafer both demonstrated high performance, reproducibility and operational yield, thus providing scalable quantum devices for network implementations. We use integrated microcomb technology, which has revolutionized classical photonics23,24,25, to substantially enhance the scalability and reliability of the TF-QKD network by delivering a large number of ultralow-noise, phase-coherent comb lasers for WDM-based networks across distributed client chips. We implemented pairwise TF-QKD between 20 client chips through ten wavelength-multiplexed channels aligned with microcomb lines, realizing a total network spanning capability NL/2 = 3,700 km (in which N = 20 and L = 370) without any trusted relays.

Architecture of integrated-photonics TF-QKD network

The TF-QKD14 network retains MDI properties22, effectively eliminating detector vulnerabilities using untrusted resources and maintaining the star topology around a centralized server node, and greatly enhances the communication distance15,16,17,18,19,20,21. In typical TF-QKD, frequency references are distributed among user nodes to generate coherent light pulses and secure keys are then derived from single-photon interference measurements at the untrusted server node. Figure 1a shows a WDM-based TF-QKD network enabling parallel key distribution between paired clients with identical wavelengths. Clients in one metropolitan zone connect to intermediate nodes, transmitting WDM weak-coherent pulses through backbone fibres for long-haul transmission between different zones. At the server, light pulses are demultiplexed, interfering at beamsplitters and secure keys are extracted through single-photon detections.

Fig. 1: An integrated-photonics TF-QKD network system.
Fig. 1: An integrated-photonics TF-QKD network system.
Full size image

a, Schematic diagram for a many-user long-haul TF-QKD WDM-based star network system. This network could support a large number of QKD clients (N) and long client-to-client communication distances (L). It can mitigate detector vulnerabilities at the central (untrusted) server node, while enabling cost-efficient resource sharing among many legitimate QKD users. b, Conceptual diagram of a large-scale integrated-photonics TF-QKD network, ‘Weiming Quantum Chip-Network’, with both large N and long L. QKD client chips integrate local lasers, PMs, IMs and VOAs, delivering high-speed operation, full integration, cost-effectiveness, miniaturization and mass-manufacturability for the TF-QKD network. The server chips encompass an integrated microcomb, as well as linear-optic circuit (LOC; including optical switches, wavelength multiplexers and beamsplitters) modules, single-photon electro-optic frequency shifters (EOFSs), SNSPD modules and electronic logics. The integrated microcomb provides a broad range of ultralow-noise coherent Hz-level comb lines for TF-QKD, enabling WDM-based networks with inherent phase coherence between QKD client chips. Microcomb and client chips are clock synchronized within the network, with the microcomb driving many chip-based transmitters to reliably share quantum-secure keys across long-haul fibre channels. c, Photograph of integrated-photonics quantum chips used in our network, including an integrated Si3N4 microcomb chip (unpackaged) at the centre and 20 copies of identical InP QKD client chips (two InP chips are packaged on one printed circuit board). Both server-side microcomb chips and client-side QKD transmitter chips can be massively manufactured with wafer-scale reproducibility. These 20 InP QKD chips used in this experiment were randomly selected from a set of 24 copies fabricated on a 3-inch wafer, without any preselection.

We introduce an integrated-photonics TF-QKD network, as illustrated in Fig. 1b. At the central server node, an integrated optical microcomb serves as the heart, generating a broad range of ultralow-noise coherent dark-pulse frequency combs with linewidths at the Hz level in the telecommunication band. The microcomb can be generated through the self-injection locking of a semiconductor laser to an ultrahigh-quality (Q) Si3N4 microresonator26,27, eliminating the necessity for complex electronic controls and bulky table-top lasers and hence leading to cost-effectiveness, compact footprint and high scalability. Microcomb lines are dispersed across the network by means of the backbone fibres and wavelength demultiplexed at intermediate nodes before reaching the client-side chips. Our InP-based TF-QKD client transmitter chips monolithically integrate all essential components, encompassing lasers, electro-optic (EO) phase modulators (PMs), intensity modulators (IMs) and variable optical attenuators (VOAs). Using the injection-locking technique21,28,29, the original lasers with MHz-level linewidths on QKD chips can be locked to the Hz-linewidth comb lines to align their frequencies and phases, locally regenerating low-noise light fields. Following this, QKD chips implement the encoding and transmit wavelength-multiplexed pulses back to the server node. As shown in Fig. 1c, 20 copies of InP QKD chips are synergistically used in this network. At the server’s detection module, a full integration of quantum devices is possible, encompassing linear-optic devices (for example, switches, wavelength multiplexers and beamsplitters), quantum frequency shifters (see Supplementary Information Section 5.2 for details), superconducting nanowire single-photon detectors (SNSPDs) and fast electronic logics, further bolstering scalability and cost efficiency. In our experiment, the integration of devices other than microcombs on the server is still underway. A Supplementary Video is provided to illustrate the entire procedure of the integrated-photonics TF-QKD network.

We emphasize several unique capabilities of integrated-photonics QKD networks. (1) Ultralow-noise coherent dark-pulse microcombs provide ultralow-noise lasers required for TF-QKD and establish a globally stable phase reference for the entire network. (2) Broadband and equally spaced comb lines (with intervals adjustable to International Telecommunication Union standards) can facilitate massive wavelength-multiplexed QKD between a large number of client chips (at present, we use a 30-GHz spacing for the proof-of-concept demonstration). (3) High-yield fabrication and operation of fully functional QKD transmitter chips ensuring high performance, along with the wafer-scale reproducibility of microcomb chips, critically advance the scalability and reliability of quantum networks.

Integrated microcombs and QKD client chips

The advancement of integrated quantum photonic technologies has sparked notable interest in high-speed, cost-effective, miniature and readily manufacturable devices for quantum communications, with the specific emphasis on practical QKD applications30. Previously, protocols of prepare-and-measure QKD31,32,33,34,35,36,37, MDI-QKD38,39,40 and entanglement-based QKD13,41,42,43 have been implemented with integrated-photonics devices but primarily in the point-to-point scenario. Notably, chip-based point-to-point TF-QKD has been demonstrated using partially integrated InP photonic chips44 and recent work shows the potential of microcombs in MDI-QKD networks45.

Integrated microcombs at the server node

Our integrated microcomb that emits tens of ultralow-noise comb lines is realized through the hybrid integration of an InP distributed feedback (DFB) laser with an ultrahigh-Q Si3N4 microresonator (see Fig. 2a,b). This is enabled by the self-injection-locking process, which allows the DFB laser to lock onto the resonant mode and narrows laser linewidths through direct backscattering of the intracavity field26,27. The microresonator was fabricated on 100-nm-thick Si3N4 waveguides (see the scanning electron microscope image in Fig. 2b) having an ultralow loss of 1.7 dB m−1. Figure 2c reports the measured intrinsic Q factor for the modes around λ0 at 1,549.78 nm, with a mean Q of 20 million. Its free spectral range was measured to be 30.03 GHz. Fabrication and characterization details for the wafer-scale microresonators are provided in Supplementary Information Section 1, exhibiting consistently high Q factors and well-controlled dispersion. By adjusting the air gap between the DFB laser and the microresonator, the dynamics of nonlinear phase could be tailored, which then resulted in linewidth narrowing and dark-pulse comb generation. Figure 2d reports the measured spectrum for the generated dark-pulse microcombs (see simulation data in Supplementary Fig. 4). It exhibits a flat-top multichannel spectrum with a total output power of around 10 dBm. Coloured comb lines (partly from λ−5 to λ+8) were selected for TF-QKD implementations. To assess the spectral stability, which is crucial for practical implementations, we conducted continuous system operation over 12 h. The measured flat temporal output intensity profile of the microcomb in Fig. 2e demonstrated high system-level stability, which can be further improved through co-packaging hybrid integration of the DFB laser and microresonator26. Figure 2f presents measured frequency noises for the comb lines46. In comparison with the free-running DFB laser, the comb lines exhibit more than 25 dB noise suppression and yield a white-noise floor around 13 Hz2 Hz−1, indicating short-term linewidths around 40 Hz, which is essential for TF-QKD.

Fig. 2: Integrated-photonics microcomb and QKD transmitter chips.
Fig. 2: Integrated-photonics microcomb and QKD transmitter chips.
Full size image

a, Schematic diagram of chips and networks. Server-chip microcomb with Hz-level linewidths is used to drive and phase-lock slave lasers on QKD client chips separated at metropolitan areas A and B corresponding to group A and group B chips. Client chips locally regenerate low-noise light field for phase reference (at λ0, yellow pulses) and weak coherent states for quantum key coding (at λi, coloured pulses) using phase-locked slave lasers. Paired client chips of {Ai, Bi} send quantum states through the λi channel in long-haul upstream fibres (in which all wavelength channels are multiplexed), with single-photon interference measurements at the server node returning secure keys. A complete set-up diagram is provided in Supplementary Fig. 5. b, Device and experimental set-up of integrated microcombs by self-injection-locking a DFB laser to a Si3N4 ultrahigh-Q microresonator. Scale bar, 100 nm. c, Measured intrinsic Q values for the resonate modes of the Si3N4 microresonator with a mean value of 20.0 ± 2.8 million. Mode number specifies the ordinal position relative to the central mode. d, Measured spectrum of dark-pulse microcombs. Coloured channels indicate those used in the present network. e, Normalized intensity fluctuation of the microcomb output under the spectral stability characterization. The spectrum of the microcomb was collected every minute with an optical spectrum analyser over consecutive 12 h. The normalized intensity fluctuation shows excellent system stability, maintaining a mean relative intensity of 0.994 ± 0.006. f, Frequency noise characterization of microcomb lasers at the server. Power spectral densities were measured using the correlated self-heterodyne method46. g, Summary of intrinsic linewidths for the microcomb and 20 phase-locked slave lasers on the client chips denoted as {Ai, Bi}. Linewidths are derived from the white-noise floor of measured power spectral densities in f and i. Points are experimental data and dashed lines are their mean values. h, Device and experimental set-up of InP-based QKD client chips, featuring monolithic integration of all components as shown in a. i, Frequency noise characterization of phase-locked slave lasers on the QKD client chips, showing comparable performance with the microcomb (see summary in g). In f and i, free-running lasers (black) are plotted for comparison. j, Spectra of locked DBR lasers on QKD client chips. The top plot shows the continuous tuning spectrum of a single client laser, aligning and phase-locking to specific comb lines. The middle and bottom plots show spectra of injection-locked slave lasers on different client chips. k, Characterization of 120 EOPMs on 20 independent InP client chips (each with six EOPMs, forming IMs). Points are experimental data and lines are their mean values for the measured Vπ and interference visibility of IMs. Data points of zero Vπ or visibility indicate three hardware failures of modulators. Error bars denote ±1 standard deviation of measured data.

Integrated transmitters at the user node

Under the MDI framework, TF clients rely on the precise preparation and encoding of quantum states using trusted devices. These trusted devices locally generate weak coherent states, implement phase randomization for the decoy-state method and then apply phase and intensity modulation for encoding. As a result, a fully integrated photonic TF-QKD transmitter typically includes local lasers, PMs, IMs and VOAs. The InP client chips monolithically integrate all of these devices for the preparation of signal or decoy states and the execution of dual-λ global phase tracking17,18,19,20,47,48. As illustrated in Fig. 2a, client chips consist of two distributed Bragg reflector (DBR) lasers I and II, three IMs I, II and III and one PM. Twenty InP chips (footprint 4.6 × 2 mm2 each; Fig. 2h) are divided to group A and group B, to implement pairwise TF-QKD between the two chips {Ai, Bi}, in which i = (1, 2,…, 10) represents paired users sharing identical wavelengths \({\lambda }_{i}^{{\prime} }\). \({\lambda }_{i}^{{\prime} }\) are aligned with microcomb lines, remapping from λ−5 to λ−3 and λ+2 to λ+8. We first characterize all lasers and modulators on these 20 QKD chips (randomly selected from a 3-inch wafer) to demonstrate their reproducibility, performance and uniformity. It shows that 117 of 120 modulators function as designed across all 20 chips, with only three hardware failures. These modulators exhibited consistent performance, as reported in Fig. 2k, with an average Vπ of 5.8 ± 1.4 V and interference visibility of 33.8 ± 5.6 dB (details can be found in Supplementary Information Section 1). An eye diagram to characterize the PM bandwidth is provided in Supplementary Fig. 7. Further, the original DBR InP lasers exhibit MHz-level linewidths introducing high-frequency noise incompatible with TF-QKD, which requires low-noise lasers with typically Hz-level linewidths. Scalable TF-QKD networks require numerous such low-noise lasers with identical spectra. To address this, we use microcomb lines as seed light to locally drive and lock the slave DBR lasers using optical-injection-locking techniques21,28,29. The server-chip microcombs are first distributed to different zones by means of downstream backbone fibres, with each client chip receiving two microcomb lines as seeds to implement the dual-λ phase tracking17,18,19,20,47,48 (Supplementary Information Section 4.3). Specifically, laser II is locked to the reference comb line λ0, whereas laser I is locked to another line as the quantum light for pulsation and key encoding, as illustrated in Fig. 2a. We characterize the regeneration of ultralow-noise light in those wavelength-tunable slave lasers across 20 InP chips, after injection-locking to microcomb lines (see Fig. 2i for frequency noise tests). Figure 2g summarizes the measured linewidths for microcomb lines and locked local lasers, demonstrating substantial frequency noise suppression that achieves intrinsic linewidths of around 60 Hz for TF-QKD. As shown in Fig. 2j, one client-chip locked DBR laser reaches 3.2 nm continuous wavelength tuning, spanning λ−5 to λ+8. All DBR lasers across the 20 QKD chips demonstrate independent wavelength tunability and injection locking to the microcombs, with each achieving >30 dB side-mode suppression in Fig. 2j. The lasers all operated properly and the EOPMs exhibited a high yield of 97.5%, with the adaptable circuit design ensuring that the TF-QKD protocol remains fully executable (Supplementary Information Section 4), enabling all 20 randomly selected client chips to function properly in the network. These results also show the capability of massive integration and identical operation of low-noise lasers for scalable WDM-based TF-QKD networks.

Integrated-photonics TF-QKD network demonstration

We then implemented a multi-user sending-or-not-sending TF-QKD (SNS-TF-QKD) protocol49 in this integrated-photonics network. Figure 3a shows the overall experimental process, including system initialization, quantum key encoding and detection. Note that the reference light signal performs continuous phase estimation, whereas the quantum signal is temporally divided into a phase estimation period—for calibrating the phase difference ϕd between the reference ϕr and quantum ϕq signals arising from slight discrepancies in their optical paths—and a quantum communication period for TF-QKD. Details for the SNS-TF-QKD protocol and experimental implementation are provided in Supplementary Information Section 4.1.

Fig. 3: Phase tracking in the integrated-photonics TF-QKD network using a dual-λ method.
Fig. 3: Phase tracking in the integrated-photonics TF-QKD network using a dual-λ method.
Full size image

a, Experimental flow chart detailing the initialization, encoding and detection of quantum states. The server-side microcomb and client-side DBR lasers are locked by adjusting injection currents in the gain media. Sequence controls implemented on the QKD client chips are rescaled in the plots with: ➀ IM III, ➁ IM I/II, ➂ PM I, ➃ PM for reference pulse and ➄ output quantum pulse chain. For detection, the server node identifies effective events for quantum signals and performs phase estimation. Dq(r)L(R) denotes the left (L) or right (R) detection outcomes for quantum (q) or reference (r) signals after interference at the server node. b–d, Experimental characterization of phases across the fibre links between those ten pairwise client chips {Ai, Bi}: phases for the reference signals (ϕA(B),r) at the wavelength of λ0 (b); phases for the quantum signals (ϕA(B),q) at the wavelength of \({\lambda }_{i}^{{\prime} }\) (c); and their relative phase differences ϕd = ((ϕB,r − ϕA,r) − (ϕB,q − ϕA,q)) (d). The fibre network used 370-km (185 + 185 km) upstream and predominantly short downstream links, except for the {A5, B5} paired QKD, which used 120-km (60 + 60 km) downstream fibres. Phase information was extracted by means of SNSPD-based single-photon interference measurements (10 kHz sampling rate). Data were collected sequentially for each pairwise-linked QKD chips at their wavelength channels, while maintaining continuous operation of all other wavelengths multiplexed in the same fibres (filtered out before injection-locking those microcomb lines of interest). e, An example of phase distributions for the {A1, B1} paired QKD chips. Inset, zoom-in view of the initial 100-ms window. f,g, Measured phase drift rate: 2.871 rad ms−1 for ϕr and 0.033 rad ms−1 for ϕd, through short downstream links {A1, B1} (f) and 3.908 rad ms−1 for ϕr and 0.069 rad ms−1 for ϕd, through long downstream links {A5, B5} (g).

We implemented a lab-scale proof-of-principle quantum network of 20 QKD client chips linked through ten WDM channels. Phase tracking and key distribution were sequentially executed between paired chips {Ai, Bi} owing to resource constraints in electronic controls and synchronizations. Note that all ten wavelength channels maintained parallel fibre transmission connecting the 20 client chips during sequential QKD operations, with non-target channels at high intensity to estimate an upper bound for crosstalk and noise (emulating realistic scenarios). Our implementation combines established dual-λ phase tracking and postselection19,47,48, maintaining system simplicity while ensuring sufficient stability, with microcomb-enabled coherent WDM, which together provide an efficient solution for a QKD network. Details are provided in Supplementary Information Section 4.

We now characterize phase tracking using the dual-λ method in each wavelength channel of the TF network. The QKD client chips were symmetrically connected to the server using long upstream fibres spanning up to 185 km, resulting in a maximum distance of L = 370 km between paired chips. That is, this TF network contains N = 20 QKD client chips (ten paired chips connected through ten WDM channels), achieving a total quantum networking capability NL/2 = 3,700 km. In Fig. 3b,c, the measurement results showed rapid phase fluctuations for the reference (ϕA(B),r) and quantum signals (ϕA(B),q) across ten channels. Strong fluctuations exceeding 2π were observed within a 2-s time frame, mainly caused by the channel noises, whereas laser noises have been effectively mitigated with the use of Hz-level lasers. This was confirmed by the phase difference ϕd = ((ϕB,r − ϕA,r) − (ϕB,q − ϕA,q)) in Fig. 3d, revealing slower phase variations. As an example, Fig. 3e shows the detailed results for the {A1, B1} paired chips, indicating minimal changes in ϕd over 100 ms. For experimental simplicity, short downstream fibres were used in these experiments, except for the {A5, B5} chips, for which 120-km (60 + 60 km) fibres were used to characterize the network with realistic long downstream fibres. We characterized injection locking and frequency noises, when microcombs were transmitted through long downstream fibres and consecutively reamplified (see Supplementary Fig. 9), achieving comparable performance to short-haul configurations. In Fig. 3f,g, comparative results reveal accelerated phase drift in long downstream fibres (see ϕd and ϕr for {A1, B1} and {A5, B5}). This behaviour emerges from the formation of a giant interferometer, in which injection-locked lasers replicate the phase information of the seed light. Despite increased drift, phase tracking can be achieved by simple adjustment of phase-correction cycles as the follows: ϕd was recalibrated every 2 ms within each 5-ms cycle for {A5, B5} versus every 5 ms within each 20-ms cycle for other pairs.

Secure quantum keys were then distributed across the network between 20 QKD client chips, with fibre channel lengths of 204 km (102 + 102 km) and 370 km (185 + 185 km) respectively implemented in the experiments. The Z-basis measurement was used for key generation, whereas parameter estimation was performed with the results obtained in the X-basis. Our proof-of-principle implementation used sequential transmission of encoded bins in repeated cycles to characterize both network gain and quantum bit error rate (QBER). Figure 4a–d reports the measured QBERX in the phase-sensitive X-basis for the SNS-TF-QKD over distances of 204 km and 370 km, with data collected across all 20 client chips. Experimental results show QBERX values of 2.87–3.63% for 204 km and 3.50–4.17% for 370 km. Full experimental data across all 20 client chips, including gain and Z-basis/X-basis QBER, are summarized in Supplementary Table 5. The low QBER in the Z-basis/X basis results from a combination of high TF coherence, precise quantum state encoding enabled by the integrated photonics and robust phase tracking across the network. The finite-size effect has been taken into account by sending a total of 1012 bins in each case. Furthermore, we obtained the secure key rates at the two different communication distances, as shown in Fig. 4e,f. A notable feature of TF-QKD is the ability to surpass the Pirandola–Laurenza–Ottaviani–Banchi (PLOB) repeaterless bound50, which represents a fundamental limit on the capacity of quantum communication with no quantum relays. As anticipated, in Fig. 4e, the key rates do not exceed the PLOB-repeaterless bound at a distance of 204 km but aligned well with the expected theoretical rate–distance relationship in Extended Data Fig. 1. In Fig. 4f, at a communication distance of 370 km, all TF-QKD across 20 client chips successfully surpassed the PLOB-repeaterless bound. In comparison with the traditional upper bound, the key rates of SNS-TF-QKD showed improvement from at least 51.5% to up to 251.4%, as illustrated in Fig. 4f, inset. This highlights the substantial advantage of the SNS-TF-QKD protocol over long distances, with our system adeptly supporting the advanced protocol49. In the future, the achievable communication distance and key rate can further be improved by using more protocol optimization (for example, advanced data postprocessing method51), high-performance SNSPDs and ultralow-loss fibres19 (see Supplementary Information Section 4.6 for detailed discussion).

Fig. 4: Experimental demonstrations of SNS-TF-QKD networks with integrated photonics.
Fig. 4: Experimental demonstrations of SNS-TF-QKD networks with integrated photonics.
Full size image

Error rates and secure key rates are measured for 20 QKD client chips {Ai, Bi} through ten quantum channels at different wavelengths \({\lambda }_{i}^{{\prime} }\). a,c, Segmented QBER in the X-basis and their statistical distributions in the 204-km upstream fibre channels. b,d, Segmented QBER in the X-basis and their statistical distributions in the 370-km upstream fibre channels. Each data point represents the error rate measured by sending 150 × 106 bins. The network used short downstream fibres, except for the {A5, B5} chips with 120-km downstream fibres. In c and d, the box plots show the mean, quartiles and extreme values. e,f, Measured secure key rates in the 204-km and 370-km fibre channels, respectively. The dotted boxes represent the theoretical value at that distance and the dashed red lines represent the PLOB-repeaterless bound at that distance. In f, it shows an enhancement of secure key rate in comparison with the PLOB-repeaterless bound. In all subplots, coloured points and bars represent different paired QKD client chips.

Discussion and conclusion

In summary, we have implemented a proof-of-principle integrated-photonics QKD network, demonstrating high scalability and reliability. Through the validation of bit transmission, the architecture and hardware have proved to be capable of supporting many-user networks accommodating 20 client chips, spanning a total networking capability exceeding 3,700 km, with each channel enabling the distribution of secure keys that surpass the repeaterless bound. Integrated-photonics quantum devices and systems offer several key features that can bolster the construction of practical QKD networks. First, the system supports a large number of QKD users, facilitated by massively manufacturable and scalable transmitter chips that combine full integration, high performance and cost-effectiveness to enable service accessibility, and microcomb-driven WDM protocols that ensure massive parallelism across the network. Second, the communication distances are extended by making use of ultralow-noise microcombs and lasers for advanced QKD protocol execution, with prospects for further improvements19. Third, network security is enhanced through the implementation of SNS-TF-QKD, maintaining MDI properties for heightened security compared with trusted-node networks. Although certain aspects under the present experimental conditions, such as imperfect phase randomization and potential vulnerabilities in the injection-locking technique, require further improvement to address possible loopholes (see Supplementary Information Section 6), these limitations do not affect the core demonstration of the scalability and reliability of the quantum networking chips. Finally, in the future, our integrated-photonics quantum devices and technologies could be extended to field-deployable QKD networks with different practical configurations. These include implementing asymmetric networks by further reducing low-frequency noise in the microcomb lasers and establishing fully connected or reconfigurable networks by combining server-side quantum frequency conversion (see Supplementary Information Section 5) with the demonstrated client-side wavelength tunability (see results in Supplementary Fig. 11 and Supplementary Table 6). A summary of existing QKD networks without trusted nodes is shown in Extended Data Fig. 2.

The range of integrated optical microcombs can be substantially broadened through more dedicated engineering52,53 and, in conjunction with broadband tunable integrated lasers54, it is promising for the network to accommodate hundreds of manufacturable QKD client chips. Through incorporating lower dark-count SNSPDs and optimized protocols51, it becomes feasible to extend individual quantum communication links to a 1,000-km distance19. An integrated-photonics quantum networking system is anticipated to span a vast quantum network system with large N and long L. Combining the demonstrated wafer-scale reproducibility of client-side and server-side chips with advanced heterogeneous integration, this platform promises further enhanced capabilities55. For example, at the server side, integrating devices such as SNSPDs56, frequency shifters57 and linear-optic circuits13,58 alongside microcombs could enhance the scalability, connectivity and cost-effectiveness of entire networks. The fusion of integrated-photonics time-frequency synchronization59,60 holds promise for advanced synchronization in quantum networking systems. Moreover, integrated-photonics QKD networking systems exhibit natural compatibility with telecommunications infrastructure through WDM technologies, paving the way for co-integrated classical–quantum communication networks that combine quantum security with classical data capacity. It is anticipated that quantum communication networks, fortified by miniaturized, scalable, high-performance and mass-manufacturable devices, are poised to increase the practicality of quantum technology.