All-photonic quantum repeaters

All-photonic quantum repeaters

Memory-based quantum repeaters require matter qubits whose coherence time exceeds the classical-message round-trip across the link they serve — a demanding constraint that limits memory-based architectures to platforms with long-lived storage. All-photonic quantum repeaters remove the memory entirely: each repeater node generates a multi-photon entangled graph state on the fly, processes it with linear optics and heralded measurements, and forwards the resulting entanglement downstream Azuma et al. 2015 .

The protocol's appeal is that there is nothing to wait on: all operations are local at each node and the classical messages only confirm which heralded attempts succeeded. Its cost is a much higher per-node photonic-resource budget — many single photons of high indistinguishability, many high-efficiency detectors, and the photonic gates to entangle them into the required tree-state structure.

The Azuma–Tamaki–Lo protocol

At each repeater node, a tree state is produced — a many-photon graph state with a tree topology. The leaves are dual-rail photonic qubits; interior vertices encode the redundancy that lets the protocol tolerate photon loss. Photons at the appropriate leaves are sent forward toward the next repeater node, where they meet photons arriving from the upstream node at a Bell-state analyser. Successful BSM outcomes herald entanglement across the link; failed outcomes are absorbed by the tree's redundancy and don't kill the run Azuma et al. 2015 .

The end-to-end Bell pair is established without any node holding a qubit while waiting on a classical message. The protocol is structurally one-way: each node generates its tree state, performs its local operations, and forwards the relevant photons before any classical signal from downstream returns.

Resource demands

The memory-coherence budget is replaced by a photon-source-and-detector budget. Per repeater node, the protocol requires:

  • Many high-purity single photons per attempt — tree states of useful depth need tens to hundreds of indistinguishable photons.
  • Photonic entangling gates to assemble the tree state. In a linear-optics implementation, these are probabilistic and consume ancilla photons.
  • High-efficiency, low-dark-count detectors at every BSA. Detector inefficiency compounds across the network.

The 50 % linear-optics BSM ceiling applies: a Bell-state measurement built from passive linear optics and photon counting succeeds at most 50 % of the time in the ideal case. Ancilla-assisted schemes can beat the ceiling — Bayerbach et al. recently demonstrated a BSM at ~57.9 % success probability using two ancilla photons Bayerbach et al. 2023 . Either way, photon-source brightness and detector efficiency are the dominant engineering numbers for all-photonic.

vs memory-based repeaters

Memory-based (1G/2G) repeaters store qubits in matter while round-trip classical signalling coordinates swaps and purification. Their performance is set by memory T₂ and the link length. All-photonic repeaters eliminate the memory dwell-time entirely but replace it with a heavier per-attempt resource cost. Azuma's 2023 RMP review covers the comparison at length, mapping the trade-off surface across loss regimes and per-node hardware assumptions Azuma et al. 2023 .

Practical status: memory-based platforms (NV/SiV centres, ions, atomic ensembles) have multi-node experimental demonstrations today. All-photonic remains predominantly theoretical with table-top building-block demos rather than full-chain repeater experiments. The gap reflects the hardware demand asymmetry — single-photon sources at the indistinguishability and brightness all-photonic needs are still maturing.

Network role

All-photonic occupies an alternative-architecture slot to the 1G/2G memory-based repeaters and the 3G QEC-encoded transport. Where memory T₂ caps the link length of memory-based repeaters, all-photonic's reach is limited by the per-node photonic resource overhead. If high-rate deterministic photon sources become widely available, all-photonic gets practical; until then memory-based dominates the experimental frontier Kumar et al. 2025 .

Forward references:

  • Repeaters — the 1G/2G/3G memory-based and QEC-based families this subject sits beside.
  • Distribution — single-link architectures (MM/SR/MS) that any repeater family chains.
  • Metrics — photon-loss and indistinguishability numbers that set the per-node budget here.