1. Introduction

Quantum key distribution (QKD) establishes secret keys from quantum no-cloning, measurement disturbance and composable-security principles, rather than from computational hardness alone. Since BB84, entanglement-based QKD, decoy-state protocols, measurement-device-independent QKD (MDI-QKD), device-independent QKD (DI-QKD), twin-field QKD (TF-QKD) and continuous-variable QKD (CV-QKD) have progressively expanded the security and implementation landscape^[1-3]. This evolution has also changed the hardware problem. A practical QKD apparatus must prepare, transmit, measure and monitor quantum states while keeping the physical device inside the assumptions of the security proof^[3-5].

QKD has moved beyond isolated laboratory links. Fiber links, free-space channels, satellite experiments and metropolitan networks, including the Tokyo QKD Network and the 4,600-km integrated space-to-ground quantum communication network, have shown that QKD can operate across multi-node and cross-domain communication infrastructures^[6-9]. These demonstrations also clarify the engineering bottleneck. Deployable systems require low optical loss, stable encoding, compact packaging, repeatable calibration, thermal robustness, manufacturable components and interfaces compatible with classical network control.

Integrated photonic chips provide a direct route to this form of hardware consolidation. On planar lightwave circuit (PLC), silicon photonic, silicon nitride, thin-film lithium niobate, indium phosphide (InP) and hybrid platforms, interferometers, couplers, beam splitters, phase and intensity modulators, attenuators, filters, monitoring paths and detector interfaces can be integrated within compact optical modules^[10-13]. The framework in Figure 1 organizes this transition from protocol demand to chip response: shortened optical paths and monolithic alignment primarily improve stability, whereas wafer-scale fabrication and on-chip monitoring turn security-relevant quantities into measurable engineering parameters. These parameters include loss, extinction ratio, phase error, polarization-dependent loss, crosstalk, timing jitter and calibration stability.

Display Full Size

Download

Figure 1. Protocol-to-hardware framework for integrated photonic QKD chips. The diagram links deployment requirements, protocol assumptions and chip-level responses, emphasizing that integration is valuable only when compact routing, stable interference, calibration access and manufacturable packaging jointly preserve the security model. QKD: quantum key distribution.

Chip-based QKD has evolved from PLC interferometric modules toward diverse photonic platforms. Early silica and silicon PLC studies integrated BB84, differential-phase-shift QKD and polarization-insensitive receiver functions, demonstrating the stability advantages of planar interferometers^[14-16]. Silicon-photonic QKD transmitters and transceivers later compressed phase, intensity and polarization encoding into compact chips, while integrated continuous-variable QKD (CV-QKD) and heterogeneous photonics extended the design space toward coherent receivers, electro-optic modulation and source-detector interfaces^[13,17,18].

Different QKD protocols impose different chip requirements. Decoy-state BB84 requires high-extinction intensity modulation, phase randomization and limited state-dependent leakage^[19-21]. MDI-QKD requires high-indistinguishability interference between independent users^[22,23], whereas TF-QKD is dominated by remote phase reference, frequency locking and long-term stability. CV-QKD relies on linear modulation, low-noise coherent detection, balanced photodetection and high-speed digital signal processing^[24,25]. A chip that performs well for one protocol may therefore be unsuitable for another if its non-idealities fall outside the corresponding security model.

Integration does not automatically improve security. Detector-blinding, time-shift and local-oscillator attacks have shown that practical QKD systems can fail when device behaviour deviates from model assumptions^[26-28]. Chip integration may suppress some alignment and drift problems, but it can also introduce new forms of reflection feedback, residual modulation, thermal crosstalk, electrical coupling and package-dependent leakage. The main challenge is therefore to align protocol models, chip metrics and deployable packaging within one security-aware evaluation framework.

Existing reviews have summarized QKD from the perspectives of practical security, realistic devices, CV-QKD systems, QKD networking and chip-based QKD^[2-4,29]. For PLC and integrated photonic QKD chips, however, the key task is to connect protocol requirements, device metrics, platform trade-offs and deployable systems. The organizing contribution of this review is a traceable connection between these levels: protocol-specific trust assumptions identify the device quantities that must be characterized; platform and package choices determine how those quantities can be measured and controlled; and system demonstrations establish the conditions under which secure keys are obtained. Table 1 and Table 2 organize protocol and platform constraints, whereas Table 3 and Table 4 distinguish system evidence from proposed reporting requirements. The framework is intended to support comparison under stated operating and security assumptions, rather than to rank platforms by integration density alone.

2. Protocol Requirements and Device-Level Constraints for Integrated Photonic QKD Chips

As QKD moves from experimental links to deployable systems, integrated photonic chips must translate protocol assumptions into device-level metrics. The relevant question is not only how many optical functions can be integrated. It is which constraints on state preparation, encoding, synchronization, interference, coherent measurement and monitoring must be preserved in hardware. Figure 2 compares the relative stringency of chip-level requirements across representative QKD protocols and encoding approaches. Higher scores indicate more stringent requirements in the corresponding dimension; for optical loss, this means a stronger requirement to minimize loss. The qualitative scores summarize the engineering requirements discussed in Section 2, while Table 1 specifies the corresponding protocol assumptions and trust boundaries. At a fundamental level, these constraints arise from quantum-state preparation, the role of loss and noise in the secret-key rate, and the mapping of receiver outcomes onto the security model^[1-3,30]. An integrated QKD chip should therefore be regarded as a protocol-constrained quantum photonic system, not as a generic optical-communication component^[3,24,29].

Display Full Size

Download

Figure 2. Protocol- and encoding-dependent requirements for integrated photonic QKD chips. The six columns compare optical loss, state indistinguishability, phase stability, synchronization, security monitoring and trust boundary. Colors and marker sizes indicate qualitative scores on a 0-5 scale, with higher scores denoting more stringent requirements. For optical loss, higher scores indicate a stronger requirement to minimize loss. The trust-boundary scores indicate the stringency of requirements for defining and maintaining the relevant trust assumptions. Scores summarize the authors’ qualitative assessment of representative implementations discussed in Section 2 and Table 1. QKD: quantum key distribution.

BB84 and decoy-state protocols provide the baseline case. Lo, Ma and Chen introduced decoy-state QKD to estimate the single-photon contribution from weak coherent pulses with different mean photon numbers, enabling practical sources to resist photon-number-splitting attacks^[19]. Wang independently formulated a decoy-state method for defeating the same attack class, reinforcing the importance of source photon-number control^[20]. At the chip level, this means that transmitters must provide intensity modulation while preserving decoy intensity, phase randomization, spectral shape and spatial mode within the security model. If modulators or attenuators introduce distinguishable temporal, spectral, polarization or spatial signatures, an adversary may obtain side information about state preparation^[3,21,29].

The representative system layouts collected in Figure 3 are drawn from three experimentally important routes: high-speed silicon-photonic discrete-variable QKD (DV-QKD), chip-based MDI-QKD and composable-security CV-QKD demonstrations^[31-33]. This comparison clarifies why phase-encoded and time-bin protocols impose direct constraints on integrated interferometers. Their key rates depend on interference visibility, phase error, path-length matching, thermal drift and polarization-dependent loss. PLC, silicon photonic and related waveguide platforms can integrate unbalanced Mach-Zehnder interferometers (MZIs), delay lines and phase modulators on the same chip, thereby improving relative path stability^[11,12,29]. Chip designs for phase-encoded QKD must therefore co-optimize low loss, low polarization-dependent loss, thermal tuning range, phase-control precision and online calibration.

Display Full Size

Download

Figure 3. Representative QKD system architectures and their chip-level implications. (a) From left to right: COW, polarization-encoded BB84 and time-bin BB84 layouts highlight source preparation and interferometric stability requirements. Reproduced from reference^[31]. CC BY 4.0; (b) MDI-QKD shifts the trusted boundary away from the measurement unit and makes independent-user indistinguishability central. Reproduced from reference^[32]. CC BY 4.0; (c) CV-QKD relies on coherent detection, shot-noise calibration and receiver linearity. Reproduced from reference^[33]. CC BY 4.0. QKD: quantum key distribution; COW: coherent-one-way; MDI-QKD: measurement-device-independent QKD; CV-QKD: continuous-variable QKD.

The three panels also identify different subsystem boundaries. In the discrete-variable (DV) layouts, the chip must prepare the intended intensity and encoded states, while any interferometric receiver must preserve the required path delay and visibility^[31]. In MDI-QKD, two user transmitters must deliver compatible states to an untrusted measurement station^[32]. In CV-QKD, the central requirements are modulation fidelity, coherent detection, and calibrated noise estimation; Figure 3c illustrates these system requirements and should not be interpreted as a fully integrated chip implementation^[33]. Coherent-one-way (COW) monitoring and phase-randomized decoy-state BB84 use different coherence assumptions, so their source specifications cannot be interchanged simply because both employ time-resolved optical circuits^[31].

MDI-QKD changes the security boundary. Lo, Curty and Qi placed the measurement unit outside the trusted boundary, removing detector side channels from the direct assumptions of the security proof^[22]. Braunstein and Pirandola similarly emphasized the importance of eliminating measurement-device trust assumptions^[23]. For integrated chips, this moves the key device metric from receiver detector characterization to indistinguishability between independent user transmitters. Wei demonstrated high-speed MDI-QKD with integrated silicon photonics, and Semenenko reported chip-based MDI-QKD, showing that multi-user interference can be implemented with compact photonic circuits^[32,34].

For chip-based DV-MDI-QKD, the design specification should therefore report pulse timing and width, spectral detuning, polarization alignment and spatial-mode overlap, together with the source-intensity settings and their uncertainties. These quantities affect two-photon interference and the detection statistics used for decoy-state estimation^[22,32,34]. Interference visibility must be reported with the input-state and background-subtraction conventions, rather than treated as a universal security threshold. Source characterization, private randomness and authenticated classical communication remain necessary even when the measurement station is untrusted^[3,22]. In particular, polarization-dependent loss can make the emitted states basis dependent; a recent two-chip experiment explicitly incorporated this imperfection into MDI-QKD key-rate estimation and demonstrated transmission over 202 km^[35].

TF-QKD and sending-or-not-sending protocols shift the long-distance problem from single-link attenuation to remote phase reference and interference stability. Lucamarini proposed TF-QKD to overcome the linear rate-transmittance scaling of repeaterless quantum communication through single-photon interference^[36]. Wang later reported TF-QKD over 830 km of fiber, and Liu pushed experimental TF-QKD to the 1,000 km scale^[37,38]. These results indicate that chip-based TF-QKD transmitters must prioritize frequency stability, phase tracking, reference isolation and long-term interferometric reproducibility.

CV-QKD is closely connected to coherent optical communication. The coherent-state protocol of Grosshans and Grangier, together with Gaussian-modulated coherent-state experiments, linked QKD to homodyne and heterodyne detection^[39,40]. Qi introduced a local local-oscillator (LLO) approach that reduces engineering dependence on transmitting a strong local oscillator, and Jain demonstrated practical CV-QKD with composable security^[25,33]. Integrated CV-QKD depends on linear modulators, balanced coherent receivers, shot-noise calibration, analogue bandwidth and digital signal processing, rather than on single-photon counting alone^[41,42].

For CV-QKD, a receiver specification should state whether the local oscillator is transmitted or generated locally, the shot-noise normalization, the trusted detector parameters, and the calibration procedure^[25,43,44]. Receiver bandwidth and symbol rate describe acquisition capability, not the final secret-key throughput. A reported rate must also be identified as asymptotic or finite-key and linked to the modulation and detection models used in the security analysis^[33,45-47].

The protocol comparison in Table 1 further shows that trust boundaries are not interchangeable across QKD families. DI-QKD bases security on Bell nonlocality and minimal device trust, but current experiments remain constrained by detection efficiency, loss, finite-key effects and long-term stability^[5,48]. MDI-QKD mainly removes detector-side assumptions, decoy-state BB84 constrains source photon-number statistics, and CV-QKD requires accurate modelling of excess noise and coherent detection^[3,22,24,39]. The security design of integrated QKD chips cannot rely on a single universal metric. It must define trusted boundaries separately for the source, channel, measurement and control subsystems.

These protocol requirements constrain integrated photonic QKD chips in four ways. First, optical loss, crosstalk and mode distinguishability determine whether prepared states remain compatible with the protocol model. Second, modulation accuracy, phase stability and synchronization determine encoding quality. Third, receiver bandwidth, timing jitter and detector or homodyne noise determine measurement fidelity. Fourth, monitoring and calibration determine whether device parameters can be bounded during operation. The following sections therefore evaluate chip platforms and systems by their ability to satisfy protocol-specific constraints rather than by integration density alone.

3. Photonic Platforms and Integration Architectures

A photonic integrated chip denotes an optical circuit fabricated on a substrate; a chip-based QKD system combines one or more such circuits with the off-chip optics, electronics and processing needed to execute the protocol. Heterogeneous integration combines distinct material or device technologies, whereas a packaged module includes the optical, electrical and thermal interfaces required for operation^[10,18,49].

Protocol constraints become deployable hardware only through material platforms, device blocks and transceiver architectures. The radar charts in Figure 4 compare the qualitative capabilities of representative photonic platforms and receiver integration routes in low-loss operation, modulation, integration, source/receiver integration and stability^[3,10,29,50]. The same protocol can follow different implementation routes: silica PLCs favour low-loss and stable interferometry, silicon photonics supports dense complementary metal-oxide-semiconductor (CMOS)-compatible circuits, thin-film lithium niobate enables high-speed electro-optic modulation, and InP or hybrid platforms support active sources and detector interfaces^{[11,12,18,51]}.

Display Full Size

Download

Figure 4. Qualitative capability profiles of representative photonic platforms and receiver integration routes for QKD. The radar charts compare low-loss capability, modulation capability, integration capability, source/receiver integration and stability. Larger radial values indicate stronger capability in the corresponding dimension on a five-level scale. The scores summarize the authors’ qualitative assessment of representative implementations discussed in Section 3 and Table 2. Panels show (a) silica PLC; (b) silicon photonics; (c) SiN; (d) TFLN; (e) InP/hybrid integration; and (f) detector/coherent-receiver integration. QKD: quantum key distribution; PLC: planar lightwave circuit; InP: indium phosphide; TFLN: thin-film lithium niobate.

Silica PLCs remain important because QKD interferometry benefits directly from fiber-compatible processing, low propagation loss, weak polarization sensitivity and long-term stability. Nambu, Hatanaka and Nakamura proposed a BB84 system based on silica PLCs, and Honjo used a PLC MZI for differential-phase-shift QKD^[14,15]. Kimura demonstrated single-photon interference over 150 km using silica integrated-optic interferometers, confirming the value of PLC path stability in long-distance coherent interference^[52]. Nambu, Yoshino and Tomita then moved PLCs toward practical encoder and decoder units^[53]. These results explain why PLCs remain useful for phase-encoded, time-bin and polarization-insensitive receiver modules, even when active modulation is supplied by other platforms.

Silicon photonics provides a well-developed route to dense and scalable transceiver chips, and its development shows a clear progression from device physics to link-ready subsystems. The device basis was established early, with high-speed silicon modulation through carrier-dispersion effects and the positioning of silicon quantum photonics as a platform for integrated interference and state manipulation^[54,55]. On this basis, chip-scale QKD links have been demonstrated, including high-speed silicon-photonic QKD, a composable secret-key rate of 157 kbit/s over a 43 km link at 16.4 dB loss, and a photonic integrated quantum secure communication system^[12,31,56]. More recent silicon-electronic co-integration and CMOS-compatible silicon QKD chips extend this progression toward link-ready subsystems^[57,58]. Integrated multiuser receiver architectures offer a complementary route to sharing detection resources^[59].

Silicon nitride and thin-film lithium niobate fill complementary gaps. Silicon nitride offers low-loss routing, broad optical bandwidth and microresonator-based quantum light sources; this makes it suitable for delay lines, filters, frequency-bin resources and multiwavelength quantum photonics^[60-64]. Thin-film lithium niobate adds strong Pockels-effect modulation. Wang demonstrated CMOS-compatible lithium-niobate electro-optic modulators, and He reported hybrid silicon-lithium-niobate MZI modulators for 100 Gbit/s and beyond^[51,65]. These capabilities are directly relevant to decoy modulation, CV-QKD coherent-state preparation and TF-QKD phase control, provided that RF packaging, optical coupling and process uniformity are controlled.

InP, III-V/silicon and broader heterogeneous integration provide routes to active sources and multi-material functionality. Smit described InP generic integration, where lasers, semiconductor optical amplifiers, modulators and detectors can share a common platform^[66]. Liang, Roelkens and Fathpour showed that III-V gain and silicon passive circuits can be combined to overcome single-platform limitations^[67-69]. Hybrid integrated quantum photonics further suggests that sources, routing, modulation and detection can be assigned to materials according to function^[49]. Table 2 condenses these trade-offs into device-level selection criteria for QKD chips, highlighting where each platform is naturally strong and where hybridization is likely to be necessary.

Platform comparisons must separate waveguide propagation loss, single-interface coupling loss and packaged insertion loss, because they describe different parts of the link budget. Reported values require a wavelength, waveguide geometry and measurement configuration; a best-case passive-waveguide result should not be treated as the loss of a complete QKD module^{[10,18,60,61]}. Likewise, process maturity should be described through available fabrication processes, design tools, reproducibility and packaging evidence, rather than inferred from a single working device. Heterogeneous integration adds interface alignment, back-reflection and thermal-management constraints, so the benefit of combining materials must be assessed at module level^[49,67-69].

Transmitter and receiver architectures convert those platform capabilities into protocol states and measurement outcomes. Figure 5 brings together device examples that span high-speed silicon-photonic MDI-QKD transmitters, modulator-free transmitter design, receiver integration and GHz-rate detector-integrated reception^{[34,50,70,76]}. The comparison shows that an integrated QKD architecture is a closed measurement system, not a simple aggregation of optical blocks. Source interfaces, intensity, phase, polarization or time-bin encoding, attenuation, monitoring taps, decoding interferometers, filters, detector interfaces, coherent receivers and electronic control must be specified together^[11-13,29]. Transmitter and receiver non-idealities jointly determine state-preparation error, measurement error, mode leakage, stability and the final secret-key rate^[3,29].

Display Full Size

Download

Figure 5. Representative transmitter and receiver implementations in integrated QKD. (a) Integrated silicon-photonic transmitter architecture for high-speed MDI-QKD, showing how modulation and interference preparation are brought onto chip. Reproduced from reference^[34]. CC BY 4.0; (b) Modulator-free transmitter design, illustrating an alternative route that reduces high-speed external modulation. Reproduced from reference^[70]. CC BY 4.0; (c) Transfer integration of a membrane-supported superconducting nanowire single-photon detector onto a photonic waveguide. The waveguide guides photons to the detector, gold contacts provide electrical connections for biasing and readout, and the tungsten probe positions the detector membrane during transfer. Reproduced from reference^[50]. CC BY 4.0; (d) Optical micrograph of a detector-integrated on-chip QKD receiver. Reproduced from reference^[76]. CC BY 4.0; (e) Packaged transmitter and receiver modules, emphasizing the transition from chip layout to deployable hardware. Reproduced from reference^[34,50]. CC BY 4.0. QKD: quantum key distribution; MDI-QKD: measurement-device-independent QKD.

The transmitter chip maps protocol variables onto physical quantum states. In decoy-state BB84, the analyses by Lo, Ma and Chen and by Wang require weak coherent sources to switch among different mean photon numbers so that the single-photon contribution can be estimated^[19,20]. At the chip level, this requirement couples high-speed intensity modulation, variable attenuation, phase randomization and monitoring taps. Paraïso reduced reliance on high-speed external modulators with a modulator-free transmitter chip, whereas Ma and Cai demonstrated silicon-photonic polarization-encoded transmitter and transceiver circuits^[70,71,72]. These designs show that transmitter miniaturization is useful only when drive calibration, decoy-state indistinguishability and monitoring are preserved.

Detector and coherent-receiver interfaces define the measurement side of the architecture, and the relevant figures of merit are detection efficiency, dark-count rate, timing jitter and, for coherent reception, bandwidth and linearity^[73-75]. On the single-photon side, travelling-wave superconducting-nanowire single-photon detectors (SNSPDs) have been embedded in nanophotonic circuits, a detector-integrated on-chip QKD receiver has been operated at GHz clock rates, and a heterogeneously integrated superconducting silicon-photonic platform has been proposed for MDI-QKD^[76,77,78]. On the coherent side, integrated photonic-electronic receivers now reach beyond 20 GHz for balanced detection and 9 GHz for squeezed-light measurement, and a 10 GBd (gigabaud) CV-QKD experiment with an integrated receiver estimated key rates of 0.737 Gbit/s over 5 km and 0.315 Gbit/s over 10 km, including finite-size parameter-estimation effects in its collective-attack analysis^[41]. The broadband detector demonstrations establish receiver capability separately^[79,80]. Chip-scale and time-domain homodyne detectors further show that bandwidth, linearity and shot-noise clearance are receiver-level rather than isolated photodiode metrics^[81-83].

Chip architectures are also expanding beyond a single degree of freedom. High-dimensional QKD using multicore fiber and silicon photonic integrated circuits brought spatial encoding into an on-chip framework^[84]. Orbital-angular-momentum multiplexing extends the spatial encoding options^[85], while entanglement-based integrated multiuser networks demonstrate a distinct networking application^[86]. These developments broaden the role of integrated QKD chips from compact BB84 transmitters to configurable quantum-communication nodes. The security challenge is that added degrees of freedom can also create added leakage channels, so monitoring and calibration must scale with architectural complexity.

4. Module Engineering, Calibration and Security-Parameter Mapping

A QKD chip becomes a deployable module only when its optical, electronic and thermal responses remain measurable and stable after packaging. Phase drift, thermal crosstalk, polarization drift, coupling loss, drive crosstalk and calibration bias enter state preparation, interference measurement, excess-noise estimation and the boundary of the security proof^[3,4,29]. Bare-chip insertion loss, modulation bandwidth or detector efficiency therefore indicate device potential, but deployment value depends on whether the packaged module remains calibratable under temperature variation, vibration, ageing, fiber coexistence and unattended operation^[10,17,18].

Phase stability is a common constraint for phase-encoded, time-bin, MDI-QKD, TF-QKD and CV-QKD chips. Sibson and Bunandar showed that shortened on-chip paths can support stable high-speed and metropolitan QKD, but these advantages depend on trackable phase response under modulation, tuning and package stress^[11,12]. The silicon-photonic CV-QKD platform of Zhang and the high-speed integrated QKD system of Sax further show that phase compensation, drive synchronization and stable calibration must operate as part of the link, not as occasional offline adjustments^[13,87]. Phase-remapping studies by Fung and Xu, together with calibration-security analysis by Jain, show that imperfect phase modulation can become an exploitable side channel if it is not bounded in the security model^[88-90].

Thermal control and electro-optic modulation play different roles. Silicon thermo-optic phase shifters provide compact tuning for drift compensation, as demonstrated by Harris, but they are better suited to slow control than to carrying high-speed encoding^[91]. Lithium-niobate and hybrid silicon-lithium-niobate devices use electro-optic modulation for high-speed phase and intensity control, with 100 Gbit/s-class demonstrations providing a device basis for high-rate QKD encoding and CV-QKD modulation^[51,65].

Packaging must preserve polarization, coupling and monitoring fidelity. QKD is more sensitive than classical communication to polarization-dependent loss, polarization-mode dispersion and basis mismatch because these effects modify state distinguishability and measurement statistics^[3,4]. Wang demonstrated an integrated dynamic polarization controller, Zhang reported polarization-insensitive QKD using PLC chips, and Choe implemented a compact 1 × 4 polarization beam-splitter module for free-space BB84 reception^[16,92,93]. These studies indicate that package design should be treated as a security-relevant part of the optical system, not as a final mechanical step.

These effects enter security analysis through different observables. Phase drift changes interference statistics and, in CV-QKD, contributes to the inferred phase-noise budget; polarization-dependent loss changes transmission in a state-dependent manner^[35,43,94]. Thermal or electrical crosstalk should be characterized under simultaneous channel operation, since a calibration of isolated components need not bound correlated preparation errors. We recommend reporting the measurement interval, environmental conditions and uncertainty alongside each security-relevant parameter, and verifying that the resulting bounds remain valid during key acquisition.

Control electronics and calibration strategies must be designed together with the protocol model. Yuan's directly phase-modulated light source reduces the conventional modulation chain, but it requires accurate modelling of source dynamics^[95]. For MDI-QKD, polarization-multiplexing and phase-reference-free schemes reduce calibration burden, while CV-QKD LLO and self-coherent reference-sharing schemes move phase handling into receiver estimation and compensation^[96-98]. Local-oscillator phase attacks and phase-noise models identify requirements for the phase-reference treatment^[43,99]. Receiver imbalance, shot-noise calibration and trusted detector-noise studies address distinct receiver assumptions^[44,100,101]. We recommend recording the corresponding calibration bounds and using them consistently in key-rate estimation.

Reliability evaluation should expand from device metrics to module-level and security-level metrics. Photonic packaging usually emphasizes coupling loss, bandwidth, thermal resistance and mechanical robustness. QKD modules also require records of phase error, extinction-ratio drift, decoy-intensity drift, polarization-dependent loss, monitoring-tap stability, detector-gating drift, excess-noise bias and confidence intervals for calibration parameters^{[3,29,88,102]}. Practical security analyses of chip-based CV-QKD and CV-QKD with imperfect phase-reference calibration show that device imbalance, noise modelling and calibration cannot be treated as ordinary engineering imperfections^[94,100,102].

Acceptance limits must be distinguished from representative device specifications. A quoted extinction ratio, visibility or noise value is not a protocol-independent guarantee of security. The admissible range depends on the source and detector models, channel loss, observed statistics, block size and security parameters^{[3,33,46,103]}. For decoy-state operation, photon-number statistics and finite-key estimation must be consistent with the characterized source^[19,20,104]. For CV protocols, the shot-noise calibration and the treatment of trusted detector noise must match the security model^[44,47,101].

Representative numbers illustrate why these categories must remain separate. A 20-mm thin-film lithium-niobate modulator reported a half-wave voltage of 1.4 V, an electro-optic 3-dB bandwidth above 45 GHz and on-chip optical loss below 0.5 dB^[51]; these are device specifications, not a demonstrated QKD security limit. A silicon-photonic MDI-QKD experiment measured Hong-Ou-Mandel visibility up to 48.4% and obtained a finite-key rate of 31 bit/s at 36 dB channel loss^[34]. That visibility was an operating result for the stated sources and measurement configuration, not a universal acceptance threshold. For a particular module, an acceptance test should instead propagate the measured uncertainty intervals into the chosen key-length calculation and require a positive lower bound at the declared security level and operating loss.

Network coexistence adds another reliability boundary. Karinou discussed routes toward integrating CV-QKD into deployed optical networks, and Kumar studied coexistence between CV-QKD and intense dense wavelength-division multiplexed (DWDM) classical channels^[105,106]. DWDM crosstalk studies examine coexistence-induced noise^[107], while upstream CV-QKD access experiments address a distinct network configuration^[108]. Reconciliation methods contribute another processing constraint on the achieved throughput^[109]. These topics should be evaluated separately when specifying synchronization, noise control and classical-channel isolation. Packaged QKD chips should therefore report packaged insertion loss, environmental drift, calibration interval, coexistence conditions and security-parameter traceability.

5. Chip-Based QKD System Demonstrations and Network Deployment

System experiments are the point where protocol assumptions, chip metrics, package stability, synchronization, post-processing throughput and network conditions become inseparable. Table 3 summarizes representative link results and identifies their integration scope or comparison boundary. Metropolitan, satellite and space-to-ground demonstrations show that QKD can operate in multi-node and cross-domain infrastructures, but they also expose the size, interconnection, maintenance and calibration burden of discrete optical systems^[6-9]. Integrated chips address this burden only if packaging, control and security monitoring remain stable during operation.

Discrete-variable chip-based systems first showed that integrated transmitters and receivers can carry real protocol functions. Integrated transmitter encoding has been realized on a photonic chip; a metropolitan link delivered 157 kbit/s over 43 km at 16.4 dB loss; and a 1,550 nm silicon-photonic transmitter supported daylight free-space QKD over 145 m at a quantum bit error rate (QBER) of ~0.5% and a ~30 kbit/s average secret-key rate^[11,12,110]. Chip-based QKD has since been pushed toward high-speed operation, including 2.5-GHz silicon-photonic system demonstrations^[87,111]. The deployment examples in Figure 6 matter because they test chip functions under field conditions, including fiber loss, daylight background, calibration stability and link-level key extraction.

Display Full Size

Download

Figure 6. Field-oriented demonstrations of silicon-photonic QKD. (a) Daylight free-space implementation using a 1,550 nm silicon-photonic transmitter, illustrating compact terminal operation under background-light noise. Reproduced from reference^[110]. CC BY 4.0; (b) Metropolitan field test over a 43 km dark-fiber link, showing how chip-based encoding can be evaluated in a realistic loss and calibration environment. Reproduced from reference^[12]. CC BY 4.0. QKD: quantum key distribution.

Differences in Table 3 cannot be attributed to chip material alone. Detection technology, clock or symbol rate, coupling and channel attenuation, modulation format, data-block size, reconciliation assumptions and offline versus real-time processing all affect the reported throughput. Integrated CV receiver estimates should therefore be compared with other estimates using compatible noise and security models, whereas composable DV results require their stated finite-key conditions.

Chip-based MDI-QKD experiments emphasize reproducibility of multi-user interference. Wei demonstrated high-speed MDI-QKD with integrated silicon photonics, and Semenenko demonstrated chip-based MDI-QKD, moving transmitter and Bell-state-measurement functions toward compact implementations^[32,34]. Cao's integrated silicon-photonic MDI-QKD system further supports the view that stable two-photon interference, rather than device count alone, is the central metric for multi-user QKD nodes^[112]. For deployed MDI-QKD, the relevant operating quantity is long-term indistinguishability of independent user pulses in timing, frequency, polarization and optical mode, together with stable protocol-dependent intensity and encoding settings.

CV-QKD system experiments shift the integration problem toward coherent modulation, balanced detection and photonic-electronic reception. Coherent modulation and receiver structures have been integrated on silicon photonics, a 10 GBd CV-QKD experiment has used an integrated receiver with offline digital processing, CV-QKD has been realized with on-chip light sources, and a silicon-photonic CV-QKD receiver platform has been reported^{[13,41,113,114]}. In parallel, long-distance CV-QKD over 202.81 km highlights the importance of excess-noise control and phase compensation for distance extension^[42]. These results show that CV-QKD chips must be evaluated by receiver bandwidth, shot-noise calibration, electronic noise, DSP complexity and secure throughput together.

Recent demonstrations further separate integration scope from the reported security regime. A PLC-module experiment combined a one-decoy method with advantage distillation at 625 MHz, reporting composable secret-key rates of 3.32 Mbit/s at 2 dB loss and 1.85 kbit/s at 18 dB loss^[115]. A silicon-photonic transmitter–receiver implementation using 8-PSK modulation operated at 1.5625 GBd and reported asymptotic secret-key rates of 31.05 Mbit/s over 25.8 km and 5.05 Mbit/s over 50.4 km^[116]. A further integrated CV-QKD demonstration reported an average 0.737 Mbit/s over 50 km during 12 h of operation, with the laser remaining external^[117]. These examples illustrate complementary progress in post-processing, integrated coherent operation and stabilization.

Resource-efficient silicon-photonic BB84 also demonstrates how synchronization and polarization compensation can share the encoding and measurement resources, reducing auxiliary hardware requirements^[118]. Chip integration extends to related cryptographic applications: a three-node quantum-digital-signature network used integrated encoding and decoding over distances up to 200 km^[119].

Networked and high-rate QKD benchmarks prevent overinterpretation of chip integration. Discrete systems have reported high-rate QKD exceeding 110 Mbit/s, high secret-key-rate operation enabled by fast single-photon detectors and real-time key distillation, and fiber QKD over 307 km and 421 km^[120-123]. Chip-based QKD may not immediately set every record for rate or distance, but it offers opportunities to improve module size, stability, manufacturing consistency and maintainability.

Network experiments now provide more direct evidence of scalability. An optical-frequency-comb MDI network reported an average 267 bit/s per user pair at approximately 30 dB link attenuation and demonstrated simultaneous communication between two pairs^[124]. An integrated-photonic TF-QKD network used 20 client transmitter chips and a server microcomb chip, with sequential pairwise demonstrations through ten wavelength channels over 370 km of spooled fiber^[125]. Recent long-distance photonic-integrated and microcomb-driven fully connected network studies extend this direction^[126,127].

6. Practical Security and Shared Research Challenges

6.1 Practical Security and Evaluation Criteria

Practical security remains a design principle for future chip systems. Detector-blinding, time-shift and local-oscillator studies identify exploitable differences between physical devices and assumed detector or calibration models^[26-28,128]. Trojan-horse bounds and loss-tolerant source analyses address complementary source-side risks^[103,129]. The systems highlighted in Figure 7 build on two complementary lines of work: protocol-compatible chip-based QKD integration and high-dimensional silicon-photonic QKD using multicore fiber^[11,84]. Together they show why future chips should expose measurable quantities for on-chip reflection, residual modulation, thermal crosstalk, polarization-dependent loss, monitoring-tap stability and package feedback, so that security proofs can use measured module parameters rather than idealized component assumptions.

Display Full Size

Download

Figure 7. Integrated QKD architectures viewed through a practical-security lens. (a) Multi-protocol-compatible system using an integrated InP encoder and a SiₓNᵧ decoder, illustrating how heterogeneous integration can support protocol flexibility. Reproduced from reference^[11]. CC BY 4.0; (b) Silicon photonic integrated circuit for noise-resistant high-dimensional QKD, showing how spatial encoding can extend the chip design space while introducing new calibration requirements. Reproduced from reference^[84]. CC BY 4.0. QKD: quantum key distribution; InP: indium phosphide.

Chip-specific modulation physics introduces additional tests. Kang et al. demonstrated that externally injected femtosecond pulses can exploit two-photon absorption in silicon photonic modulators to change emitted decoy intensities, causing legitimate users to overestimate the key rate under the uncorrected model^[130]. This is a concrete example of a material-level mechanism affecting a security-relevant source parameter. Isolation and monitoring must therefore be evaluated over the relevant injection conditions, and their residual bounds must be included in the source model; integration alone does not establish resistance to such perturbations.

6.2 Shared challenges and future directions

Future QKD networks will require chip modules that can coexist with classical infrastructure and network management. QKD network-security studies discuss operational and trust challenges^[131,132], while provider-network testbeds investigate integration into existing infrastructure^[133]. Untrusted-relay MDI and TF networks address specific measurement-station trust assumptions^{[124,125,134]}; they should not be conflated with arbitrary trusted-node-free routing. Satellite-to-ground and integrated space-to-ground demonstrations provide a separate class of deployment evidence^[7,8]. Integrated chips should therefore be assessed not only by peak key rate, but also by pluggable interfaces, remote calibration, power consumption, environmental retention, compatibility with network control and traceability of security parameters.

Artificial intelligence (AI) may provide an additional control and monitoring layer for future chip-based QKD networks. By learning from optical, electrical, thermal and protocol-level telemetry, AI-assisted systems could automatically identify abnormal patterns associated with hacking attempts, side-channel exploitation and device imperfections, including unexpected phase drift, modulation errors, polarization-dependent loss, thermal crosstalk and detector anomalies. In parallel, real-time AI-based resource management could optimize channel allocation, wavelength assignment, routing, key-pool utilization, power consumption and calibration schedules, while dynamically configuring quantum communication networks in response to changing traffic loads, link conditions and security risks. Such functions should be implemented as security-aware decision-support mechanisms rather than as substitutes for composable security proofs. Their training data, false-positive and false-negative rates, response latency, robustness against adversarial manipulation and interaction with the trusted boundary must be explicitly characterized. Any AI-driven reconfiguration should also be auditable and constrained by predefined security policies, so that network optimization does not introduce unmodelled side channels or invalidate the assumptions used in key-rate estimation^[135,136].

At the platform level, one promising route is heterogeneous integration combined with security-oriented co-design. No single material platform easily provides low loss, high-speed modulation, on-chip sources, detector interfaces and low-cost packaging at the same time^[17,49]. Integrated QKD and quantum random number generation (QRNG) technologies, hybrid silica-lithium-niobate transmitter chips, silica-silicon PLC decoding chips and polarization-based silicon encoder-decoder chips indicate that practical systems will allocate functions across materials and packages^[137-140]. A QKD chip specification should thus include low-loss passive routing, low-voltage modulation, low-noise reception, on-chip monitoring, thermal management, calibration logging and an explicit mapping from measured parameters to security assumptions.

Beyond the platform and network directions above, a further group of recent security, protocol and integrated-photonics studies is best read as a set of evaluation dimensions that a chip specification should report explicitly. Table 4 consolidates these dimensions, the evidence each line of work establishes, and the corresponding requirement on the chip specification, while distinguishing published evidence from proposed reporting practices. These recommendations do not constitute a validated certification standard.

The remaining challenges are shared across material and protocol choices. First, the useful loss budget includes propagation, coupling, filtering and monitoring losses after packaging, not only the best passive-waveguide value^[10,18]. Second, a reproducible module requires stable calibration across devices and operating conditions; wafer-scale and network demonstrations provide evidence beyond a single optimized chip^[125]. Third, integration must be evaluated together with the remaining laser, detector, control-electronic and digital-processing resources^[41,49]. We recommend reporting these resources, operating duration, calibration intervals and parameter uncertainties alongside the secret-key rate. DI-QKD security studies and multiuser TF-QKD architectures motivate longer-term research, but their assumptions and level of experimental implementation must be distinguished^[48,134].

7. Conclusion

This review has summarized the protocol requirements, material platforms, integrated devices and system demonstrations of chip-based QKD. Complementary advances in silicon, silicon nitride, silica, thin-film lithium niobate and InP have enabled compact transmitters and receivers, with progress extending to metropolitan links and multi-user networks. Secure-key performance depends on accurate state preparation, low optical loss, stable interference and well-characterized detection, together with reliable synchronization and post-processing. Further development of heterogeneous integration, packaging, automated calibration and photonic–electronic co-design will help translate these advances into scalable and maintainable QKD modules for quantum-secure communication networks.

Acknowledgements

GPT-5.6 was used during preparation of the original manuscript for language polishing and for generating icon elements in conceptual illustrations (Figure 1).

Authors contribution

You J: Conceptualization, investigation, writing-original draft.

Yin X: Visualization, writing-review & editing.

Yang S: Formal analysis, methodology.

Hu C: Investigation.

Cheng P: Project administration, funding acquisition.

An J: Supervision, project administration, funding acquisition.

Conflicts of interest

The authors declare no conflicts of interest.

Ethical approval

Not applicable.

Consent to participate

Not applicable.

Consent for publication

Not applicable.

Availability of data and materials

Not applicable.

Funding

This work has been supported by Quantum Science and Technology-National Science and Technology Major Project (Grant No. 2021ZD0300701).

Copyright

© The Author(s) 2026.