One error check, measured across two separate quantum computers

Oxford's two-module ion-trap machine used entanglement made over a photonic link to measure a stabiliser on data qubits sitting in different processors. It detected phase flips on a logical qubit spread over both modules, in real time, and corrected any single-qubit Pauli error on a Bell state shared between them.

Research desk University of Oxford, arXiv preprint Read on 2 November 2026

Schematic of two trapped-ion modules about two metres apart, each holding a strontium network ion and a calcium data ion joined by a local gate, with a dotted photonic link between the modules used to measure one check across both data ions. Below, two panels: detect, a phase flip on a two-qubit code spread over both modules; correct, any single-qubit Pauli error on a shared Bell state. — Our schematic. The 96.4% and 82.5% figures are the paper's own; module layout and spacing are drawn, not measured.

One box will not hold a useful machine

A quantum computer that can run useful algorithms will need many thousands of physical qubits, and an error-corrected one needs them to check each other constantly. No single ion trap or chip is expected to hold all of them. One answer is to build smaller modules and join them with light: each module sends photons down a fibre, and the photons leave qubits in different modules entangled. The open question has been whether error correction, the checks that keep a logical qubit alive, can actually be measured across that gap.

What Oxford reports

A University of Oxford team led by Ellis Ainley, Gabriel Araneda and David Lucas used two trapped-ion modules about 2 m apart. Each holds a strontium ion that talks to the photonic link and a calcium ion that holds the data. Entanglement made over the link let them measure a stabiliser, one of the parity checks an error-correcting code is built from, on data qubits sitting in two different processors. The paper describes this as the first experimental demonstration of distributed quantum error detection and correction.

Two experiments. First, a [[2,1,1]] repetition code, one logical qubit spread over both modules, with phase flips detected in real time: when the check flags an error, the run is aborted and the logical qubit is prepared again. Second, the same remote checks combined with real-time feedforward to correct any single-qubit Pauli error on a Bell state shared between the modules. With no error added, one check gave the right answer 96.4(6)% of the time; the shared Bell state was prepared at 82.5(15)% fidelity. Each data point is 1,000 runs.

What this does not show

The codes are as small as codes get. A two-qubit repetition code detects one kind of error and cannot fix it; the correction experiment protects a Bell state, not an arbitrary logical qubit. The authors note that protecting an arbitrary logical superposition this way needs at least five physical qubits.
Not good enough yet, in the authors' words. They write that the fidelities “are not sufficient for typical fault-tolerant implementations”, limited mainly by the local gates between the two ion species and by the quality of the network link.
Speed. Making remote entanglement is slow next to the gates inside a module, and the authors flag network latency as a cost that can dominate an error-correction cycle.
A preprint. This is on arXiv, revised on 15 October 2026, and has not yet been through peer review.

Why it matters

Most of the field's roadmaps now assume modules joined by light, and the codes that save the most qubits need checks between qubits that are far apart. Until a check had been measured across two machines, that plan rested on theory. This is the proof of principle: small and noisy, and the first of its kind on our desk. The ion-trap side of the race is on our race page, and remote entanglement in ion traps also appears in our note on 10 km of fibre.

Sources

We read the sources above ourselves. Where a figure or number is ours, the article says so. If you find a mistake, tell us: a wrong sentence here gets logged like any other.