Hybrid-Channel Quantum Repeaters

• Hybrid-channel quantum repeaters are architectures that integrate distinct physical systems—such as long-lived matter memories, photonic carriers, and wavelength converters—to generate, store, and extend entanglement.
• They employ various elementary-link mechanisms like single-photon heralding and coherent-state probes, combined with multiplexing and error correction, to optimize entanglement fidelity and distribution rates.
• By partitioning tasks across different hardware components, these repeaters balance protocol efficiency with hardware constraints, paving the way for scalable long-distance quantum communication.

Hybrid-channel quantum repeaters are quantum-repeater architectures in which entanglement distribution is not assigned to a single homogeneous physical layer. Instead, long-lived matter systems, photonic carriers, wavelength converters, multiplexed memories, or heterogeneous transmission media are combined so that elementary entanglement can be generated, stored, synchronized, and extended over long distances. In the recent literature, this includes trapped-ion memories interfaced to telecom fibre through quantum frequency conversion, visible-memory/telecom-transmission schemes, discrete-variable memories coupled to continuous-variable coherent-state channels, and network stacks that join metropolitan fibre access with free-space or ultra-low-loss backbones (Liu et al., 9 Feb 2026, Gu et al., 2024, Bernardes et al., 2010, Liu et al., 21 Jul 2025). The common repeater requirement is that one entangled link must remain usable while adjacent links are created and then connected by entanglement swapping and, in some architectures, entanglement purification or encoded processing (Liu et al., 9 Feb 2026).

1. Conceptual scope and historical lineages

The term hybrid entered repeater theory through protocols that explicitly combined discrete-variable matter memories with continuous-variable optical carriers. In the coherent-state or “qubus” line, entanglement generation is mediated by conditional phase rotations of traveling coherent pulses, followed by optical measurements that herald memory–memory entanglement. “Optimal entanglement generation for efficient hybrid quantum repeaters” established a two-probe linear-optics protocol whose output contains only one type of error and whose success–fidelity tradeoff saturates the theoretical boundary, with

$$P_s(\alpha)=1-e^{-2T\alpha^2\sin^2(\theta/2)}, \qquad F(\alpha)=\frac{1+e^{-2(1-T)\alpha^2\sin^2(\theta/2)}}{2}$$

in the ideal-detector case (0811.3100). “Rate analysis for a hybrid quantum repeater” then treated the same general architecture at the network level, assuming perfect memories, optimal probabilistic entanglement generation, and deterministic swapping, and concluded that one round of purification at the first nesting level can outperform multiplexing in relevant high-fidelity regimes (Bernardes et al., 2010).

A second early line combined discrete-variable heralding with continuous-variable cat-state processing. “A Hybrid Long-Distance Entanglement Distribution Protocol” generated elementary Bell-like states by single-photon detection, converted them into entangled coherent-state superpositions by homodyne-conditioned growth, and used auxiliary cat states to make swapping arbitrarily close to deterministic (Brask et al., 2010). “Hybrid Quantum Repeater Protocol With Fast Local Processing” reordered the protocol so that cat-state growth became local rather than nonlocal, thereby eliminating classical communication during growth and improving the 1000 km rate from about $0.004$ pairs/min to about $0.08$ pairs/min at $r_{\rm rep}=1$ MHz under the paper’s assumptions (Borregaard et al., 2012).

Later work generalized the hybrid idea rather than replacing it. “Hybrid quantum repeater with encoding” introduced repetition and CSS-code variants and identified a triple trade-off between code efficiency, memory decoherence time, and local gate errors (Bernardes et al., 2011). “Hybrid quantum repeater based on resonant qubit-field interactions” replaced dispersive gates by resonant cavity-QED interactions and field postselection, aiming at $F>0.999$ over distances up to $900$ km with rates between $5\times 10^{-4}$ and $23$ pairs per second (Bernád, 2017). “A hybrid quantum repeater for qudits” extended the framework beyond qubits; for qutrits, it reported that for repeater spacing up to $10$ km the realistic lossy qutrit entanglement is even larger than any ideal loss-free qubit entanglement in negativity, and that after purification and swapping near-unit-fidelity pairs can be distributed over $1280$ km at rates on the order of $0.004$0 Hz (Bergmann et al., 2017).

This suggests that hybrid-channel quantum repeater now denotes a family of architectures rather than a single protocol class. In current usage it may refer to DV–CV interaction structure, to heterogeneous hardware roles within a node, or to the use of distinct physical channels optimized for storage, transmission, and long-haul networking (Gu et al., 2024, Liu et al., 9 Feb 2026, Liu et al., 21 Jul 2025).

2. Physical architectures and elementary building blocks

Recent hardware proposals distribute repeater functions across distinct physical subsystems. The architectural pattern is to separate long-lived storage, fast photonic interfacing, and low-loss transmission, rather than demanding that one platform perform all three optimally.

Representative architecture	Hybridization principle	Repeater role
Trapped-ion elementary link (Liu et al., 9 Feb 2026)	trapped-ion memory + telecom QFC + single-photon heralding	elementary repeater link
Rb/Tm repeater (Gu et al., 2024)	visible-memory + telecom-transmission channels	multiplexed long-distance chain
Hot alkali–noble-gas repeater (Ji et al., 2022)	hot-vapor interface + noble-gas long memory	non-cryogenic repeater
Dual-species trapped-ion repeater (Dhara et al., 2021)	communication ion + memory ion	multiplexed fibre chain
QD/G4V repeater chain (Strocka et al., 4 Nov 2025)	deterministic QD source + diamond memory	distilled solid-state chain
Atom-QPU + AFCQM chain (Sun et al., 25 Dec 2025)	SPDC/AFC multiplexing + deterministic matter processing	high-rate elementary-link generation

The trapped-ion demonstration of “A building block of quantum repeaters for scalable quantum networks” is explicit about its status as an elementary repeater link. Each node hosts a single $0.004$1 ion, uses a single-photon heralded protocol, converts the emitted 393 nm photons to 1550 nm in periodically poled lithium niobate waveguides, and then transfers the heralded entanglement into a more robust auxiliary state protected by KDD dynamical decoupling (Liu et al., 9 Feb 2026). Its significance is not merely remote Bell-pair creation, but the fact that the pair is kept alive beyond the average time required to create the next pair.

The Rb/Tm design adopts a different division of labor. A single $0.004$2 atom coupled to two nanophotonic cavities emits visible–telecom time-bin-entangled photon pairs; the visible photon is stored locally in a Tm-doped crystal memory, while the telecom photon propagates through standard fibre and participates in heralding at the elementary-link midpoint (Gu et al., 2024). Because the visible photon is directly resonant with the Tm memory around 795 nm, the architecture avoids frequency conversion entirely. The same paper assigns entanglement swapping to spin-based local logic after the stored photonic excitation is transferred into single-spin systems (Gu et al., 2024).

A third example is non-cryogenic. The hot hybrid alkali–noble-gas repeater uses hot alkali atoms for optical storage and retrieval, transfers the excitation into noble-gas nuclear spins for long waiting times, and places the cell in a ring cavity chosen to be resonant for the signal and anti-resonant for the anti-Stokes field, $0.004$3, $0.004$4, in order to suppress four-wave-mixing noise (Ji et al., 2022). In dual-species trapped-ion repeaters, the same separation appears at the ion-species level: $0.004$5 communication ions generate ion–photon entanglement and are frequency-converted to telecom wavelengths, whereas $0.004$6 memory ions store the entanglement and participate in local swapping (Dhara et al., 2021). Solid-state and atom–photon chain proposals preserve the same architectural principle: bright or multiplexed photonic generation is front-ended by memories or processors that handle synchronization, swapping, and distillation (Strocka et al., 4 Nov 2025, Sun et al., 25 Dec 2025).

3. Entanglement generation and link-connection primitives

Hybrid-channel repeaters employ several distinct elementary-link mechanisms. One major family uses single-photon heralding. In the trapped-ion telecom link, the relevant Bell component is

$0.004$7

with $0.004$8 fixed by the optical path difference between Alice and Bob. The experiment stabilizes this phase with both wavelength-division multiplexing and time-division multiplexing: a continuous-wave 1548 nm reference co-propagates with the 1550 nm signal to suppress fibre phase noise, and a weak 393 nm reference pulse tracks slower drifts. With both in place, the interference contrast reaches $0.004$9 after transmission through a 10 km fibre link (Liu et al., 9 Feb 2026).

Another family uses coherent-state probes and unambiguous state discrimination. In the optimal two-probe protocol, Bob’s measurement task is an unambiguous discrimination problem, and the protocol is proven optimal within a broad class in which the final state lies in the Bell subspace spanned by $0.08$0 after local unitaries (0811.3100). The later rate-analysis paper adopts optimal USD as the elementary-link primitive and expresses its failure probability as

$0.08$1

so that the elementary-link success probability is $0.08$2 (Bernardes et al., 2010). In encoded versions of the same family, the coherent pulse is retained, but the distributed object becomes a logical Bell state over repetition or CSS code blocks (Bernardes et al., 2011).

A third family converts heralded single-photon entanglement into cat-state resources. In the Brask-type protocol and its modification, weakly squeezed sources and single-photon detection first create $0.08$3, then homodyne-conditioned growth generates approximate cat states, and finally homodyne-based swapping exploits the continuous-variable part of the architecture (Brask et al., 2010, Borregaard et al., 2012). The principal advantage claimed there is not only high-efficiency measurement, but also the possibility of near-deterministic swapping when auxiliary cats are inserted between beam splitters and homodyne detectors (Brask et al., 2010).

Not all current repeater logic is swapping-based. “Merging-Based Quantum Repeater” proposes a resource-state repeater in which a 1D cluster state is grown recursively by fusion type-I operations,

$0.08$4

rather than by recursively collapsing Bell pairs into longer Bell pairs (Mor-Ruiz et al., 6 Feb 2025). If the merge succeeds, two $0.08$5-qubit cluster states become a cluster of size $0.08$6. If it fails, the existing entanglement is not globally destroyed; instead, a local gap appears and is repaired by a fixed patching block. This alters the dominant failure mode from global restart to local repair (Mor-Ruiz et al., 6 Feb 2025).

4. Memories, multiplexing, and error management

The central repeater bottleneck is memory survival during stochastic link generation. The trapped-ion telecom experiment addresses this directly: over 10 km the average entanglement generation time is $0.08$7 ms, the entanglement coherence time is $0.08$8 ms, and the entanglement remains above fidelity $0.08$9 for up to $r_{\rm rep}=1$0 ms. The work defines a quantum link efficiency as entanglement generation rate divided by decoherence rate and reports a value of $r_{\rm rep}=1$1, above the critical threshold of $r_{\rm rep}=1$2 required for deterministic delivery of remote entanglement (Liu et al., 9 Feb 2026). This is a strict repeater criterion rather than a mere entanglement-generation benchmark.

A complementary response to probabilistic waiting is massive multiplexing. In the Rb/Tm proposal, the required number of memory modes is

$r_{\rm rep}=1$3

with $r_{\rm rep}=1$4 MHz and $r_{\rm rep}=1$5. Under optimized spacing, the required multimode capacity ranges from $r_{\rm rep}=1$6 to $r_{\rm rep}=1$7 modes, and the headline design uses up to $r_{\rm rep}=1$8 repeater stations, each equipped with two Tm memories capable of holding up to $r_{\rm rep}=1$9 storage modes and four single Rb atoms (Gu et al., 2024). The same paper chooses AFC memories precisely because their multimode capacity does not depend strongly on optical depth in the way EIT-based storage does (Gu et al., 2024).

Trapped-ion line architectures also exploit multiplexing, but in a different resource language. The dual-species trapped-ion protocol combines spatial multiplexing depth $F>0.999$0 and temporal multiplexing depth $F>0.999$1, giving

$F>0.999$2

The point is explicit: increasing $F>0.999$3 consumes more physical channels and communication ions, whereas increasing $F>0.999$4 is governed by memory lifetime and control timing (Dhara et al., 2021). In another hybrid chain, the total multiplexing budget is fixed at $F>0.999$5 and numerically partitioned as $F>0.999$6, $F>0.999$7, balancing repeated loading attempts against frequency-parallel remote heralding (Sun et al., 25 Dec 2025).

Error management is equally architecture-specific. Encoded hybrid repeaters use repetition or CSS codes to trade larger local resources against memory protection and faster parallelized swapping (Bernardes et al., 2011). The atom-QPU/AFCQM architecture instead studies detector- and loss-driven false-heralding errors and compares photon-number-resolving detectors, extreme photon loss distillation, and a modified re-emission trick. It reports $F>0.999$8 for a representative raw elementary link, $F>0.999$9 after PNR filtering, and $900$0 after EPL distillation, with PNR+EPL selected as the preferred combined strategy across local-loss regimes (Sun et al., 25 Dec 2025). The underlying design lesson is that hybridization does not remove stochasticity; it redistributes it into domains where multiplexing, coding, or local deterministic processing can suppress it.

5. Channel engineering and network topology

In contemporary work, hybrid-channel often refers literally to distinct optical channels assigned to different functions. The trapped-ion repeater link emits at 393 nm but converts to telecom 1550 nm using PPLN waveguides, reporting a combined transmission efficiency of $900$1, noise after filtering of $900$2 cps at $900$3 mW pump power, and raw broadband noise before filtering of $900$4 photons s$900$5 nm$900$6 (Liu et al., 9 Feb 2026). This is a memory–photon–telecom hybridization in which the matter qubit remains local and only the converted photonic carrier traverses the low-loss fibre.

The Rb/Tm architecture realizes the same division of labor without conversion. The Rb source uses two cavities, one resonant at $900$7 nm and one at $900$8 nm, so that the telecom photon is already fibre-compatible and the visible photon is already memory-compatible (Gu et al., 2024). The paper explicitly characterizes this as a hybrid wavelength strategy: the visible photon interfaces with the Tm memory and the telecom photon propagates in standard fibre. A plausible implication is that wavelength-native matching can replace conversion hardware when source and memory spectroscopy happen to align.

Channel engineering can also change the optimal repeater generation. The vacuum-beam-guide study compares fibre attenuation length $900$9 km with VBG attenuation length $5\times 10^{-4}$0 km and concludes that first-generation repeaters benefit dramatically because the optimal VBG configuration uses $5\times 10^{-4}$1 and $5\times 10^{-4}$2, eliminating entanglement purification, whereas second-generation repeaters remain primarily limited by logical gate errors rather than channel loss and third-generation repeaters benefit most because improved link transmission directly aids quantum error correction (Gan et al., 18 Apr 2025). The same paper therefore treats transmission medium and repeater generation as a coupled design problem rather than separable choices (Gan et al., 18 Apr 2025).

At the largest scales, channel hybridity becomes topological. The Hybrid Quadruple-Link Quantum Repeater connects local clients to nearby ground servers through metropolitan fibre and connects distant servers by a chain of balloons at about 24 km altitude, while keeping all quantum memories, entangled-photon sources, and detectors on the ground (Liu et al., 21 Jul 2025). The optimized balloon backbone reaches about $5\times 10^{-4}$3 dB over $5\times 10^{-4}$4 km, outperforming satellite-based relays by $5\times 10^{-4}$5 dB with the same device parameters, and the paper attributes this to beam-waist optimization together with adaptive optics (Liu et al., 21 Jul 2025). Here, hybrid-channel means simultaneous use of fibre access and free-space backbone links within one repeater-enabled network stack.

6. Performance regimes, applications, and persistent constraints

A direct application already realized on a repeater building block is device-independent quantum key distribution. On the trapped-ion telecom link, each heralded entanglement event defines one DI-QKD round; over 10 km the experiment collects $5\times 10^{-4}$6 rounds, obtains $5\times 10^{-4}$7 and $5\times 10^{-4}$8, and after reconciliation and privacy amplification distills $5\times 10^{-4}$9 secret bits. The corresponding key rate is approximately $23$0 per round, with asymptotic key rate tending to $23$1 per round. At 101 km, the same architecture reports $23$2, $23$3, $23$4, and an asymptotic key rate of $23$5 per round (Liu et al., 9 Feb 2026). These are unusually stringent benchmarks because finite-key DI-QKD requires both Bell violation and memory-supported repetition.

Other proposals occupy very different rate–distance regimes. The Rb/Tm repeater estimates about $23$6 secret bits per second across distances of up to $23$7 km with up to $23$8 repeater stations (Gu et al., 2024). The QD/G4V chain finds that a network with thousands of memories across several repeater nodes could achieve a secret-key rate of $23$9 bit/s over $10$0 km (Strocka et al., 4 Nov 2025). The balloon-backed H4QR reports a sub-Hz entanglement distribution rate between clients separated by $10$1 km (Liu et al., 21 Jul 2025). Earlier hybrid protocols occupy still other operating points: the analytic qubus repeater targets near-maximally entangled pairs with $10$2 over $10$3 km at rates of the order of $10$4 Hz under ideal-memory assumptions (Bernardes et al., 2010), encoded HQR reports roughly $10$5 Hz per memory for 20 km repeater spacing, 1280 km final distance, and final fidelity about $10$6 (Bernardes et al., 2011), and the qutrit HQR reports rates on the order of $10$7 Hz over 1280 km after three rounds of purification (Bergmann et al., 2017).

No consensus architecture dominates across all hardware and channel regimes. The literature instead emphasizes regime dependence. Merging-based repeaters reduce waiting times by preserving most of the entangled structure after failed operations, but they also usually require longer memory storage times, which can reduce the final fidelity and secret key fraction (Mor-Ruiz et al., 6 Feb 2025). VBG-based repeaters suppress channel loss, yet second-generation architectures remain primarily limited by logical gate errors rather than channel loss (Gan et al., 18 Apr 2025). Encoded HQRs expose a triple trade-off between code efficiency, memory decoherence time, and local gate errors (Bernardes et al., 2011).

Hardware incompatibilities remain equally central. In the QD/G4V chain, the dominant challenge is the bandwidth mismatch between GHz-scale QD photons and $10$8 MHz memory transitions; filtering and magnetic-field optimization improve fidelity but lower efficiency, and the SnV optimum requires cooperativity $10$9 (Strocka et al., 4 Nov 2025). The hot alkali–noble-gas repeater offers hours-long noble-gas storage at room temperature and comparatively easy multiplexing, but the spin-exchange interface is inherently slow because $1280$0 is based on weak collisions (Ji et al., 2022). These constraints do not negate the hybrid-channel approach; they define the concrete engineering space in which it operates.

Hybrid-channel quantum repeaters are therefore best understood as a design doctrine rather than a single protocol family: assign transmission, storage, synchronization, and local processing to different physical resources, and then optimize the interfaces between them. The field has moved from asking whether a Bell pair can be made over a lossy channel to asking which combination of memories, photonic carriers, multiplexing layers, and transmission media permits the Bell pair to survive long enough, and at sufficient quality, to support higher-level repeater operations and applications (Liu et al., 9 Feb 2026, Mor-Ruiz et al., 6 Feb 2025).