Quantum networks distribute entanglement rather than copies of bits between distant nodes, enabling entanglement-based key distribution, distributed quantum computing and distributed sensing, and the 2026 literature shows working component demonstrations and a maturing protocol stack but no service-carrying network beyond metropolitan-scale testbeds.
Summary
A quantum network moves quantum states, and more usefully shared entanglement, between separated nodes. The defining constraint is that entanglement cannot be copied or amplified, so every classical networking assumption about buffering, retransmission and broadcast breaks. End-to-end links are built by generating entanglement over short segments and joining them with entanglement swapping, which means the network is really a scheduling and resource-allocation problem over a stochastic, decaying resource. Satellite architectures inherit the same constraint: because long-lived buffering is technologically limited in near-term devices, useful service requires fixed optical ground infrastructure and simultaneous availability of every hop on a path at once.
The stack has three layers under active work. At the physical layer sit spin-photon interfaces and quantum memories. Diamond colour defects are the most developed node candidate, combining good optical properties, fast spin control and long spin coherence, and have already supported metropolitan-scale network demonstrations. Erbium is the competing route because it emits natively in the telecom C-band and so is compatible with installed fibre; a 2026 demonstration put single implanted Er3+ ions in a silicon-carbide-on-insulator microring with roughly 70x Purcell enhancement operating at room temperature. Memories are the bottleneck: a cavity-enhanced erbium-doped thin-film lithium niobate memory reached 23.3 ± 0.5% on-chip storage efficiency for 100 ns storage, with a 277.6 ± 52.6 s atomic frequency comb lifetime, frequency-selective routing at up to 20 MHz and inter-channel crosstalk below 1e-4, and stored and retrieved time-energy-entangled telecom photons violating an entanglement witness by more than 11 standard deviations.
Above that sits link-level entanglement generation. Photonic approaches depend on fusion gates, which without ancillas succeed only half the time; a 2026 proposal reaches 3/4 total success using four ancillary single photons and passive linear optics, succeeding directly with probability 5/8 and recovering the rest by distillation. Waveguide-QED routes aim at deterministic transfer instead: giant atoms coupled to 1D waveguides give 87% transfer fidelity with two coupling points and above 99% with ten or more.
The top layer, network control, has moved fastest. There is now a full protocol stack with a Global Entanglement Module maintaining a network-wide view of entanglement resources, improving entanglement generation rates by about 20% over a globally optimal but non-adaptive fixed-tree baseline and more than two-fold over connectionless approaches, and online schedulers that defer, retry or drop entanglement reservations and beat static TDMA baselines on completion time, completion ratio and throughput while degrading gracefully under overload. The parameters that decide the field are memory efficiency and lifetime, spin-photon interface efficiency at telecom wavelengths, and how fast entanglement fidelity decays along swapped paths.
Evidence base
Open questions
- Can on-chip telecom memory efficiency rise from 23.3% towards the levels needed for multi-hop links without sacrificing the 277.6 s comb lifetime or the 20 MHz programmable routing?
- Do the room-temperature Purcell-enhanced single erbium ions in silicon carbide retain usable spin coherence and support spin-photon entanglement, or is room-temperature operation limited to bare single-photon emission?
- How short is the operational correlation length in practice, and does it cap useful entanglement-swapping paths at a handful of hops in deployed metropolitan fibre?
- Does any demand exist beyond entanglement-based key distribution, given that the sources contain no customer, funding or procurement evidence of any kind?
Assessment drafted 2026-08-31 from up to 18 KB sources using the technology-scorecard framework; scores are a draft read pending review.