The research, conducted by Ivan Cvitić and Dragan Peraković of the University of Zagreb together with Armando Nolasco Pinto of the University of Aveiro and the Instituto de Telecomunicações in Portugal, analyzes and classifies 26 distinct routing strategies proposed between 2013 and 2024 for terrestrial fiber networks, satellite links, and hybrid combinations of the two. Rather than treating routing as a single technique, the authors organize the field within a three-dimensional taxonomy spanning the type of network involved, the optimization strategy employed, and the routing objective being pursued. The result is both a survey of a decade of work and a conceptual blueprint for what the authors argue must come next: routing algorithms that weigh multiple competing criteria simultaneously rather than fixating on a single metric such as hop count or path length.

The core difficulty that distinguishes QKD routing from ordinary packet routing lies in the nature of the commodity being moved. Quantum keys cannot simply be amplified and forwarded like data packets, because the no-cloning theorem of quantum mechanics forbids copying unknown quantum states. In practical deployments, this limitation forces networks to rely on trusted relay nodes, where keys arrive, are temporarily stored in classical key pools, and are then re-transmitted over the next hop. Every additional relay increases the number of locations where security depends on physical and procedural trust rather than on physics alone. It also means each network link consumes keys from a finite pool at a rate determined by the link’s quantum bit error rate and the underlying key generation hardware. A routing algorithm blind to these constraints can direct traffic down paths whose key pools are exhausted, causing service rejections even when the network topology appears healthy.

The new study demonstrates precisely how much can be gained by making routing aware of these quantum-specific realities. In an illustrative simulation introduced by the authors, key-aware overflow routing, which shifts key delivery sessions away from links whose key pools are running low, reduced service rejection rates by 25 to 40 percent across the practical quality-of-service operating range compared with static shortest-path approaches. This finding aligns with previously reported results from software-defined networking based dynamic routing experiments, suggesting that the performance advantage of key-awareness is robust rather than an artifact of any particular simulation setup. The mechanism is conceptually simple but consequential: by incorporating real-time key-pool availability and link error rates into path selection, the network treats keys as the scarce, perishable resource they actually are, much as congestion-aware routing treats bandwidth in conventional networks.

The authors also quantify the security trade-off at the heart of multi-path routing strategies. Distributing key material across several disjoint routes mitigates the risk that the compromise of a single trusted node exposes an entire session, since an adversary would need to intercept shares of the key on multiple independent paths. The price is increased key consumption. The study’s simulation shows this overhead reaching 30 to 60 percent when roughly a quarter to half of all sessions adopt dual-path relay schemes. That overhead translates directly into reduced network capacity, because every key consumed on a redundant path is a key unavailable for other sessions. The finding underscores a recurring theme of the analysis: in QKD networks, security, efficiency, and resilience are not independently optimizable but must be balanced against one another, often under fluctuating operating conditions.

Software-defined networking emerges from the survey as the critical enabler for this kind of adaptive, multi-objective decision-making. SDN architectures separate the control plane from the data plane, allowing a central or distributed controller to maintain a global view of network state and reconfigure routing policies dynamically. The authors point to operational deployments, including a heterogeneous SDN-QKD network running in production facilities in Madrid, as evidence that this orchestration model is viable outside the laboratory. It becomes especially important for hybrid infrastructures that integrate terrestrial fiber with intermittent satellite links, where contact windows with low-Earth-orbit satellites are fleeting and require pre-computed, rapidly deployable routing plans. Recent demonstrations of integrated space-to-ground quantum communication networks spanning thousands of kilometers show the scale such hybrids can reach, but they also expose how poorly static routing schemes cope with links that exist for minutes at a time.

Building on the gaps identified across the 26 surveyed strategies, the researchers propose a conceptual multi-criteria routing optimization framework designed for SDN-orchestrated hybrid networks. The framework represents the QKD network as a weighted graph and jointly models four competing objectives: key delivery performance, trust exposure along candidate paths, resource cost in terms of key consumption, and resilience against node or link failures. To combine these objectives, it merges subjective criteria weighting, reflecting operator priorities and policy, with objective weighting derived from measured network state, and couples this to adaptive link-state estimation. The approach draws on established multi-objective optimization methodology from engineering, adapting it to the particular constraints of quantum key relay. The framework is presented as conceptual rather than fully validated, but it provides a concrete structure for algorithm developers who until now have optimized single objectives in isolation.

The study does not shy away from the field’s most fundamental vulnerability: trusted nodes themselves. The authors identify reducing dependence on trusted relays as a critical research direction, highlighting two complementary paths. Quantum repeaters, which would use entanglement swapping and quantum memories to extend secure links without any trusted intermediate, remain largely experimental, with satellite-based entanglement distribution over 1,200 kilometers representing the current state of the art for long-distance quantum links. In the nearer term, hybrid architectures that combine trusted relays with post-quantum cryptographic wrapping offer a pragmatic mitigation, layering mathematically hard problems on top of quantum-secured links so that a compromised relay cannot expose key material in usable form. Field demonstrations of post-quantum and QKD hybridization in commercial fiber networks indicate that this layered defense is already moving toward deployment.

Beyond trusted nodes, the authors flag two further priorities for the coming years. The first is standardization of key management interfaces, which remains fragmented despite early work by international telecommunication standards bodies; without common interfaces, multi-vendor QKD networks risk becoming islands of incompatible hardware. The second is the adoption of artificial intelligence driven predictive routing, in which machine learning models anticipate fluctuations in key generation rates and demand, allowing the network to reposition key resources before bottlenecks form. Reinforcement learning approaches to QKD routing have already shown promise, and the survey suggests they represent a natural evolution from today’s reactive algorithms toward genuinely predictive network orchestration.

The significance of this work lies less in any single algorithm than in its synthesis of a field that has grown faster than its organizing principles. Quantum networks in Vienna, Tokyo, and across a 46-node metropolitan deployment in China have proven that QKD can operate at meaningful scale, and satellite links have stretched its reach across continents. What those deployments have lacked is a shared vocabulary for comparing routing approaches and a principled way to balance the competing demands of throughput, trust, cost, and resilience. By providing the taxonomy and the framework in a single treatment, the Zagreb-Aveiro team has offered network engineers a practical map of the terrain. As nations invest in quantum-protected communication backbones, the humble routing algorithm, long an afterthought behind headline-grabbing quantum hardware, is taking its place as a decisive factor in whether the promise of unconditionally secure communication survives contact with real-world networks.

The challenge arises from the unusual nature of quantum information. A classical bit can be copied, measured and retransmitted relatively directly, but an unknown quantum state cannot be cloned without disturbing it. Quantum networks therefore rely on methods such as entanglement distribution, teleportation and quantum error correction rather than simply amplifying a weakening signal. Optical fibre is particularly hostile to single photons, with transmission losses accumulating exponentially as distance increases. Conventional repeaters can divide a long channel into shorter segments, but many proposed architectures require classical signals to travel backward through the network before a repeater knows whether an operation succeeded. Over thousands of kilometres, those round-trip communications introduce latency and can reduce the rate at which useful quantum states are delivered. One-way repeaters seek to avoid that bottleneck by processing incoming quantum information continuously, without waiting for two-way confirmation. That makes them faster in principle, but it also places a heavy burden on the encoding scheme: the system must tolerate loss and operational errors as the quantum state moves forward from node to node.

The architecture examined in the study combines photonic tree structures with a small stabilizer code. In broad terms, a photonic tree spreads information across multiple photons arranged in a branching pattern. If some photons disappear in transit, measurements on surviving branches can provide enough information to reconstruct the logical state or determine which parts of the encoded state have been erased. The outer stabilizer code then adds another layer of protection against errors introduced during processing and re-encoding. Stabilizer codes work by measuring carefully chosen parity-like properties of a group of physical qubits. These measurements, called a syndrome, reveal information about the error without directly revealing the logical quantum state. A decoder uses the syndrome to infer a correction operation. The design is therefore not simply sending one photon through a fibre; it is distributing a logical qubit across multiple physical carriers and repeatedly using structured measurements to keep that logical information intact.

The researchers compared two quantum codes that encode one logical qubit while correcting a single physical-qubit error. The five-qubit code, written as [[5,1,3]], is the smallest quantum error-correcting code capable of correcting arbitrary single-qubit errors. The notation indicates a block of five physical qubits encoding one logical qubit, with a distance of three, meaning that the code can detect errors affecting up to two qubits and correct any single-qubit error. The Steane code, written as [[7,1,3]], also has distance three but uses seven physical qubits. At first glance, that larger block appears disadvantageous. More qubits mean more photons or hardware operations, a larger amount of information to manage, and more opportunities for faults. Yet the study focuses on a subtle structural difference between the codes: the Steane code has an underpopulated syndrome space, whereas the five-qubit code is described as having a fully populated syndrome space. That unused capacity in the Steane code can be exploited by its decoder to identify and correct all single-qubit erasures, along with a subset of two-qubit errors.

An erasure is different from an ordinary unknown error. In an erasure event, the system knows that a particular qubit has been lost, even though it does not know the state that qubit carried. Photon loss and failed detection often produce this kind of information: a detector registers no photon, or the architecture identifies a missing branch in the photonic tree. Because the location of the missing qubit is known, an erasure can be easier to correct than an arbitrary error, whose location and type must both be inferred. The Steane code’s syndrome structure gives the decoder additional room to exploit that knowledge. According to the study, this lets the code correct every single-qubit erasure and some cases involving two-qubit errors. The distinction matters in a repeater because loss is not a rare edge case but a central feature of long-distance optical transmission. A code that uses information about where the loss occurred can therefore deliver a higher logical transmission success rate, even if it requires more physical qubits per encoded message.

The study’s central comparison involved the re-encoding error rate, represented by εr. This parameter describes the probability that an error is introduced when a quantum state is re-encoded at a repeater node. Re-encoding is essential in a one-way architecture: each node must transform the incoming information into a form that can be forwarded, and imperfect gates, measurements, photon sources and detectors can all corrupt the process. The simulations found that the Steane-based repeater became particularly advantageous at realistic re-encoding error rates of at least 0.05 per cent. At εr = 0.2 per cent, the performance gap was striking in the researchers’ stated cost comparison. The Steane design maintained a competitive threshold out to 5,000 kilometres, while the five-qubit baseline reached only about 800 kilometres. “Cost” here refers to the resource burden required to achieve useful transmission performance, rather than a direct financial price. That burden can include the number of physical qubits, photons, operations and repeater resources needed to preserve a logical message.

The result illustrates why quantum-network design cannot be judged by qubit count alone. The five-qubit code has an unbeatable minimality advantage, but a code that is smaller on paper may become less efficient when its limited syndrome structure leaves it less able to handle the dominant failure modes of the network. The Steane code pays an overhead by encoding the logical qubit into seven rather than five physical qubits, but its stronger erasure-handling capability can compensate for that overhead as distances and operational noise increase. The finding is especially relevant to hybrid systems that combine photonic loss tolerance with discrete-variable quantum error correction. Such systems are designed around the reality that no single layer can solve every problem: photonic trees address transmission loss and known missing components, while stabilizer codes address residual errors in the surviving quantum information. The study’s algorithms include procedures for constructing logical states, generating error-correction operators, producing flag-based correction tables and building erasure-correction tables, providing a computational framework for comparing these layers.

Still, the findings should be read as a modelling result rather than a demonstration that quantum messages can now be sent 5,000 kilometres. The article reports no experimental data, and its data-availability statement says that no datasets were generated or analysed during the study. A practical repeater would need reliable single-photon sources, high-efficiency detectors, low-loss optical interfaces, accurate synchronisation and quantum operations with error rates low enough for the assumed model. The authors also acknowledge that the Steane code’s larger block size creates additional overhead, even as its erasure-correction capacity improves robustness. Real devices may experience correlated errors, imperfect photon distinguishability, memory decay, detector dark counts and hardware-specific noise patterns that are not captured by a single re-encoding parameter. Future experiments will need to test whether the predicted advantage survives those complications. Even so, the work points to a provocative route for quantum networking: rather than always chasing the smallest possible code, engineers may gain more by matching a code’s syndrome structure to the actual pattern of photon loss and repeater faults. For one-way architectures, that could turn a modest increase in hardware into a major extension of communication distance.