Quantum repeaters are the relay nodes needed to break the exponential photon-loss limit of optical fibre, and while their individual components (memories, sources, swapping, purification) now work at high fidelity in the lab, no source here shows a full link delivering useful rates.
Summary
Photons sent down optical fibre are lost exponentially with distance, and because quantum states cannot be copied, classical amplification is unavailable. A quantum repeater splits a long link into shorter segments, generates entanglement across each segment, stores it in quantum memories until neighbouring segments are ready, then joins the segments by entanglement swapping (a Bell-state measurement at the intermediate node). Errors accumulated along the way are removed by entanglement purification/distillation, or, in later-generation designs, by encoding the transmitted qubits in an error-correcting code so that loss and dephasing are corrected rather than filtered.
The parameters that decide the technology are narrow and unforgiving. Photons must be indistinguishable, because two-photon interference fidelity sets the quality of every Bell-state measurement and fusion gate; recent work reports raw Hong-Ou-Mandel visibilities above 0.95 from a tin-vacancy centre in diamond with an inferred intrinsic indistinguishability up to 0.999. Photon sources must be spectrally matched to the memory, which for rare-earth and atomic ensembles means bandwidths of hundreds of MHz or less, hard to reach with integrated photonics. Memories must combine efficiency, fidelity, storage time and, critically, multimode capacity, since the rate of a repeater scales with the number of modes that can be multiplexed into one attempt entanglement distribution and quantum storage of more than 8.
There is also a hard theoretical boundary. A 2026 no-go theorem shows that a repeater chain built only from Gaussian operations, homodyne measurement and classical communication cannot exceed the quantum capacity of direct transmission over a pure-loss channel, so the experimentally demanding non-Gaussian resources (single-photon detection, matter qubits, or non-Gaussian bosonic codes such as GKP) are not optional. The engineering question is therefore not whether a cheap all-optical shortcut exists, but how fast and how cheaply the expensive route can be made to run.
Evidence base
- 4 June 2026: a no-go theorem proves that repeater chains built from Gaussian operations, homodyne measurement and classical communication cannot exceed the quantum capacity of direct transmission over a pure-loss channel, forcing non-Gaussian hardware.
- 14 August 2026: a Yb:YSO repeater node (250 MHz bandwidth, 76.6 µs lifetime) entangled a telecom photon through 25.3 km of fibre with a photon stored 125 µs across 16,340 temporal modes, plus a 5.66 km field deployment on the Geneva metropolitan network.
- 31 July 2026: entanglement swapping with integrated narrowband sources achieved net HOM visibility 0.99 ± 0.01 and net swapped-state visibility 0.88 ± 0.06, sufficient to violate a Bell inequality.
- 17 August 2026: an 11-dimensional spatial-mode memory reached >80% efficiency and >99% storage fidelity, with an estimated 3.56 bits distributed over a 1000 km repeater link in one minute.
- 27 July 2026: a tin-vacancy centre in diamond produced raw HOM visibilities above 0.95 with intrinsic indistinguishability inferred up to 0.999, with residual limits attributed to technical rather than decoherence effects.
- 17 July 2026: the NSF Engineering Research Center for Quantum Networks received $6,000,001 to build quantum-internet foundations, PI Saikat Guha.
Open questions
- Can multimode memories combine the demonstrated large mode counts (16,340 temporal modes) with storage lifetimes long enough for segments of hundreds of kilometres, or does mode capacity trade directly against lifetime and efficiency?
- Given the Gaussian no-go theorem, which non-Gaussian resource wins on cost per node: matter qubits with cryogenics, or bosonic GKP encodings claimed to be room-temperature compatible?
- When does a two-node repeater experiment demonstrate a measured end-to-end advantage over direct transmission on the same fibre, rather than component fidelities plus an extrapolated rate?
- Can source and memory be co-fabricated with matched bandwidths at yields and stability suitable for field operation, beyond the single-chip silicon-carbide demonstration at 88.1% visibility?
Assessment drafted 2026-08-31 from up to 18 KB sources using the technology-scorecard framework; scores are a draft read pending review.