Resource-efficient fault-tolerant one-way quantum repeater with code concatenation

The study addresses the challenge of faithful long-distance quantum communication, where photon loss and operational (Pauli) errors severely limit performance over intercontinental distances. Two-way repeaters rely on probabilistic entanglement generation and long-lived memories, limiting rates. One-way repeaters use forward error correction for both loss and gate errors and can, in principle, sustain high rates set by local processing, but often require daunting resources. Prior work using only loss-tolerant photonic tree-cluster encodings relaxed resources to as few as three spin qubits per node but lacked fault tolerance to operational errors, demanding extremely low gate error rates (~10^-5). The research question is how to design a fault-tolerant, one-way quantum repeater that efficiently addresses both loss and operational errors while minimizing physical resources and complexity. The authors propose a discrete-variable concatenated-code architecture combining a tree-cluster inner code for loss and a 5‑qubit outer code with flag-based fault-tolerant syndrome extraction for Pauli errors, implemented with modest hardware using small modular processors. The goal is to achieve kHz-level communication rates over thousands of kilometers with realistic error rates (~10^-3) and reduced qubit overhead per node.

Two classes of quantum repeaters exist: two-way architectures with heralded entanglement requiring long-lived multimode memories, and one-way architectures employing error-correcting codes to combat loss and operational errors. Previous one-way schemes often faced large resource overheads. A loss-focused approach using only tree-cluster states demonstrated near-deterministic photonic interfaces and inter-node tree re-encoding with as few as three spin qubits per node but required extremely low operational errors, lacking fault tolerance. Concatenations of CV GKP with small DV codes have been proposed to improve performance and reduce resources, though high-quality GKP states remain experimentally challenging. This work instead advances a purely DV concatenation (tree inner + 5‑qubit outer) with flag-based measurements to achieve fault-tolerant operation while keeping resource demands low. The approach builds on efficient photonic cluster generation from single emitters and utilizes the smallest distance-3 code correcting arbitrary single-qubit errors (the 5‑qubit code), thus minimizing physical resources.

Architecture and protocol: The network comprises two node types. TYPE I nodes correct transmission loss at the inner (tree) code level only. TYPE II nodes correct both loss (tree decoding/re-encoding) and operational errors using the outer 5‑qubit code with flag-based fault-tolerant stabilizer measurements. A start node encodes a message qubit into the [[5,1,3]] code; each data qubit is then encoded into a photonic tree-cluster state via a Bell measurement with the tree root. Five trees are transmitted in parallel between nodes. Inner code (tree-cluster): Trees are specified by a branching vector t=[b0,b1,...,bd]. Qubits are prepared in |+⟩ and entangled via CPHASE on the graph edges. Loss-tolerant re-encoding at a repeater node uses a heralded transfer of a first-level photon to the emitter and a Bell measurement between the emitter and the root of a freshly generated tree, followed by single-qubit measurements on the incoming tree according to stabilizers. Trees are generated sequentially from a single emitter with time-bin encoding, requiring photon re-ordering via switching and delay lines so first-level photons arrive first at the next node. Outer code (5‑qubit code) and fault tolerance: TYPE II nodes decode the incoming five trees into spins, then perform fault-tolerant syndrome extraction on the 5‑qubit code using a single ancilla and one flag qubit (sequential stabilizer measurements). Flag-based circuits detect high-weight error propagation from faulty two-qubit gates; when flagged, an unflagged recovery circuit and syndrome-dependent Pauli corrections are applied at the end node via Pauli frame updating. TYPE II nodes then re-encode each data qubit into a freshly generated tree. Nonlocal gates via teleportation: TYPE II nodes require two-qubit gates between qubits residing in different emitter-cavity systems. Teleported CNOTs are implemented by heralding Bell pairs between emitters using a reflected photon and performing local gates and measurements to enact a remote CNOT up to Pauli frame corrections. Decoded data in emitter spins are first transferred to auxiliary memory spins to free emitters for Bell-pair generation. Implementation and resources: Both node types use 5 cavity-coupled emitters, each with up to 4 nearby memory spins. TYPE I nodes use 2 memory spins per emitter for depth-3 tree generation and re-encoding. TYPE II nodes similarly use 2 memory spins per emitter for tree operations; one emitter additionally has 2 extra spins serving as ancilla and flag qubits, enabling sequential syndrome extraction. Teleported gates are routed by fast optical switches. A cavity-mediated photon-emitter CPHASE realizes heralded transfer of photonic qubits. Error model: Single-qubit depolarizing channel Λ(ε) acts on tree-level qubits during transmission with εtrans derived from accumulated re-encoding errors between TYPE II nodes. Two-qubit gates are followed by a depolarizing channel Λ2(ε0) for local gates; teleported gates have effective error Λ2(3ε0). Tree re-encoding introduces an error εr at each node. Numerically, εr ≈ 3ε0 for considered parameters. Transmission error between TYPE II nodes is εtrans = 1 − (1 − εr)^n (1 − ε0)^n, where n is the number of links between TYPE II nodes. Erasure-error handling: The 5‑qubit code can also correct erasures (lost trees). TYPE II nodes support correction of up to one erasure per node; cases with two or more erasures at a single TYPE II node are treated as failures for simplicity. An explicit look-up table maps ancilla syndromes and lost-qubit position to Pauli corrections restoring the logical code state. Decoding and Pauli frame: All syndromes collected along the route are stored classically and communicated to the end node, which performs final corrections on the 5 data qubits and then decodes back to the message qubit via Pauli frame updates, avoiding mid-network active corrections and associated latency. Performance metric and optimization: Performance is assessed via the secret key rate (SKR) of the six-state QKD protocol, combining transmission probability and final qubit quality. SKR is averaged over configurations with i TYPE II nodes experiencing a single erasure each, weighted by transmission probabilities. The total processing time per TYPE II node is Ttot = Ttree + 14Tss + 26Ttele + 8Tmeas, where Ttree depends on the branching vector and emission/gate times. A cost function C = SKR^{-1}(Latt/(Tph Ltot)) − (mI + κ mII) is minimized over inter-node spacing L0, numbers of TYPE I and TYPE II nodes, and tree branching vector t, subject to constraints (L0 ≥ 1 km; mII ≤ ⌊mtot/2⌋; tree depth d=2; photon budget N ≤ 300). Hardware parameters used include Tss=100 ns, Tph=1 ns, Tmeas=1 μs, Ttele=1 μs, Latt=20 km, and ηd=0.95. Uniform interspersing of TYPE I and TYPE II nodes is assumed.

• The concatenated DV repeater (tree inner + 5‑qubit outer with flag-based syndrome extraction) achieves fault-tolerant one-way quantum communication over several thousand kilometers, with potential up to 10,000 km by interspersing TYPE I (loss-only) and TYPE II (loss + Pauli) nodes.
• With realistic operational errors around 10^-3 and re-encoding error εr ≈ 0.1%, secret key rates in the kHz range are achievable. Example: at Ltot ≈ 10^3 km and εr = 0.1%, SKR ≈ 5.5 kHz, vastly outperforming a homogeneous tree-only scheme (<1 Hz at the same distance, even after compensating for parallelization).
• Resource efficiency: Each node requires only 5 cavity-emitter systems; each processor has 1 emitter and up to 4 memory spins. TYPE I nodes use 2 memory spins per emitter; TYPE II nodes require one additional pair of ancilla and flag spins on a single emitter (sequential stabilizer extraction), limiting qubit overhead.
• Error relation and placement: Numerical modeling indicates 3ε0 ≈ εr; increasing εr both raises transmission error and degrades teleported gates (effective 3ε0), favoring sparser TYPE II placement and shorter inter-node distances to mitigate loss. Optimal designs place fewer TYPE II nodes as εr increases, balancing introduced gate noise against error suppression benefit.
• Cost-performance optimization: The concatenated architecture yields significantly lower cost at comparable or superior SKR versus homogeneous tree-only repeaters. Even when TYPE II nodes are weighted more expensive (κ up to 10), SKR degradation is minimal while the optimal design shifts toward fewer TYPE II and more TYPE I nodes, maintaining overall performance with reduced cost.
• Erasure tolerance: Correcting single erasures at TYPE II nodes significantly boosts transmission probability and SKR. The 5‑qubit code’s ability to correct up to two erasures is noted, but only single-erasure corrections are included due to rarity of double erasures under optimized configurations.
• Bottleneck processing time is set by TYPE II nodes; TYPE I nodes can be implemented with similar hardware without needing faster operation, simplifying large-scale deployment.

The proposed concatenated-code architecture directly targets the asymmetric error landscape of quantum communication: high photon loss in fibers and moderate operational errors from local gates. Tree-cluster encoding provides strong loss tolerance with efficient, conceptually simple re-encoding, while the 5‑qubit code with flag-based syndrome extraction delivers fault-tolerant suppression of accumulated Pauli errors due to non-fault-tolerant re-encoding. By interspersing simple TYPE I nodes with more capable TYPE II nodes, the network minimizes resource overhead while maintaining high SKR over long distances. The optimization framework shows that even when TYPE II nodes are costlier, the architecture’s performance remains robust by adapting the TYPE I:TYPE II ratio and inter-node spacing. The approach’s DV nature circumvents the experimental challenges of high-quality GKP states while retaining fault tolerance, making it applicable to current spin-photon interface platforms. Overall, the findings demonstrate that tailored code concatenation and judicious node specialization can substantially relax experimental demands for intercontinental quantum networks.

This work introduces a resource-efficient, fault-tolerant one-way quantum repeater using DV code concatenation: a loss-tolerant tree-cluster inner code and a 5‑qubit outer code with flag-based error correction. The architecture achieves kHz-level secret key rates over thousands of kilometers with realistic gate error rates (~10^-3) and modest qubit counts per node (five emitter modules with a few memory spins each). Optimization over node spacing, node type ratio, and tree structure shows that interspersing simple TYPE I with fewer TYPE II nodes maximizes performance at reduced cost, even when TYPE II nodes are more expensive. The DV-only design avoids the need for high-quality GKP states, enhancing feasibility across many platforms. Future research directions include exploring alternative outer codes ([7,1,3], [9,1,3], [4,2,2]) for improved trade-offs, analyzing and mitigating correlated errors in tree generation, extending to deeper trees or dynamic node placement strategies, and experimental validation on solid-state spin-photon hardware.

• Correlated errors during sequential photonic tree generation (due to repeated use of a single emitter) are neglected; the model assumes effective suppression, treating each photon as subject to independent depolarizing noise.
• Single-qubit gate and readout errors are assumed negligible relative to two-qubit gate errors; readout errors are effectively incorporated via associated two-qubit stabilizer operations.
• Only single-erasure corrections at TYPE II nodes are included in performance optimization; multiple erasures per TYPE II node are treated as failures despite the code’s theoretical capability to correct more.
• Uniform interspersing of TYPE I and TYPE II nodes is assumed; more complex placement strategies might yield further improvements.
• The performance and optimization rely on specific parameter choices (e.g., Latt=20 km, ηd=0.95, timing assumptions) and a particular error model (depolarizing channels). Thresholds and optima may shift under different hardware characteristics or biased noise.
• Maximum tree depth and photon budget constraints (depth d=2, N≤300) limit the explored design space; deeper trees or different branching could alter optimal configurations.