Error correction crossed a processor boundary
University of Oxford researchers have demonstrated error detection and active correction across two separate trapped-ion quantum processors. In a version 1 preprint submitted September 11, the team reports using entanglement between the modules to measure whether remote data qubits had suffered particular errors. Classical communication then carried the measurement result needed to reject or correct the affected state. The experiment connected several operations that must eventually work together in a modular fault-tolerant computer: remote entanglement, local gates, mid-circuit measurement, syndrome extraction and real-time feedforward.
This is a meaningful integration result, but it is deliberately small. The processors were separated by approximately two meters in one laboratory. The first test used a two-qubit repetition code that could detect phase flips but could not correct a general error on an unknown logical qubit. The second actively corrected injected errors on one specified Bell state. The authors describe the work as the first experimental distributed quantum error detection and correction. That priority claim comes from the research team and has not been independently established.
Different ions handled networking and memory
Each processor trapped one strontium-88 ion and one calcium-43 ion. The strontium ion acted as the network qubit because its internal state could be entangled with the polarization of an emitted photon. The calcium ion supplied a long-lived data qubit and an auxiliary encoding used during local operations. This division let the apparatus use one species as an optical interface while reserving another for storing the quantum information being protected.
To connect the modules, the researchers excited both strontium ions and interfered their emitted photons in a Bell-state analyzer. A successful detection event heralded entanglement between the distant network ions. That shared Bell pair then served as a distributed ancilla, which is a temporary quantum resource used to learn an error syndrome. Local controlled gates mapped the parity of the calcium data qubits onto the strontium pair. Measuring the network ions revealed the joint parity without requiring the data qubits themselves to be measured and destroyed.
A syndrome reports an error, not the encoded value
Quantum error correction cannot simply inspect a qubit to see whether it is right. Directly reading an unknown quantum state would disturb it and reveal the information the code is meant to preserve. Stabilizer measurements avoid that problem by asking a narrower question about a group of qubits. The answer indicates whether the state remains inside the code space or has moved into a pattern associated with an error, without disclosing the logical value stored inside it.
The Oxford experiment used this principle across the network. Its first protocol encoded one logical state in two calcium data qubits, one per module, and applied controlled phase-flip errors. A remote stabilizer measurement distinguished even from odd parity. When the syndrome indicated an error, the control system abandoned that attempt, reinitialized the logical state and repeated the procedure. When the syndrome indicated no detected error, the experiment retained the state and could perform another round. This was error detection with real-time selection, rather than correction of every failed attempt.
Repeated measurements suppressed injected phase errors
Conditioning on the remote syndrome increased the measured occupation of the intended code space and reduced the logical error across the tested range of injected phase-flip probabilities. With no intentional error applied, the paper reports a 96.4(6) percent probability of obtaining the target two-qubit outcome after one stabilizer measurement. The corresponding measured error was 3.6(6) percent. The researchers' independently characterized error model predicted 3.5(2) percent, with the largest listed contributions coming from local mixed-species entangling gates.
The team also repeated the remote stabilizer measurement after applying a phase-flip channel with probability 0.3 to each data qubit. Successive accepted rounds increased the measured stabilizer value while the logical error rate remained approximately constant within the experiment's resolution. This matters because a useful syndrome operation must be repeatable. A measurement that removes one error while steadily damaging the stored information would not support continuous protection. The demonstration shows repeated filtering in this minimal code, not an indefinitely operating quantum memory.
Active correction required two remote checks
A second experiment tested the complete correction loop on a Bell state shared by the calcium data qubits. The prepared state had an average measured fidelity of 82.5(15) percent before the team added controlled depolarizing noise. The researchers then measured two remote stabilizers in sequence. One exposed bit-flip behavior and the other exposed phase-flip behavior. The classical outcomes selected an X or Z recovery operation on one module, with the phase correction implemented virtually by updating the qubit's control reference.
With those conditional operations, the estimated fidelity to the target Bell state remained approximately constant across the applied range of depolarizing-noise strength. Without correction, it decreased as the injected noise increased. The correction circuit imposed its own fixed error, so the corrected result was worse at zero added noise than a direct measurement of the uncorrected state. Each shot also needed three rounds of remote entanglement generation: one to prepare the distributed Bell state and one for each stabilizer. Error correction helped only after the injected noise became large enough to outweigh that overhead.
The code protected a known state, not arbitrary information
The active result should not be generalized beyond its encoding. A single error on either data qubit transformed the chosen Bell state into the same orthogonal Bell state, so the syndrome uniquely identified the recovery needed to restore the target. The authors note that the known Bell state could instead have been prepared again. Preserving it through measurement and feedforward was useful because it exercised the complete distributed correction chain, but it did not protect an arbitrary logical superposition against every single-qubit error.
The two-qubit repetition code has similarly narrow power. It can reveal the dominant phase-flip channel chosen for the experiment, yet its [[2,1,1]] parameters do not provide the distance needed for general correction. Protecting unknown logical information requires more physical qubits and additional independent stabilizers. Large quantum low-density parity-check codes motivate the modular approach because they can encode information efficiently while demanding connections that are difficult to arrange locally. This experiment demonstrated one remote measurement primitive, not one of those full codes.
Network speed and local gates remain bottlenecks
The reported fidelities are not sufficient for typical fault-tolerant implementations. Coherent calibration errors in local controlled gates shifted measured parity fringes, and the experiment did not compensate those offsets. Probabilistic entanglement generation also left the ions waiting, during which their motion could heat and reduce later gate quality. The team interleaved entanglement attempts with cooling. That preserved operation but reduced the reported remote-pair production rate from roughly 100 per second to roughly 10 per second in the relevant sequence.
Those delays become more serious as a code requests many remote checks. Data qubits continue accumulating errors while the network produces entanglement, and an active Bell-state correction in this experiment consumed three remote pairs per shot. The underlying datasets and analysis code are available only from the authors on reasonable request, limiting immediate external reproduction. The paper is also a preprint without peer review or independent replication. Its strongest conclusion is therefore architectural: networked modules can exchange and act on quantum error syndromes in one functioning workflow.
A practical bridge between networking and correction
Modularity offers a credible response to the physical limits of one processor. Instead of forcing every qubit, optical channel, control line and cooling mechanism into a single device, engineers could build smaller modules and connect them. Remote stabilizers could preserve the connectivity of efficient error-correcting codes even when the participating data qubits live on different chips or traps. The Oxford experiment gives that proposal a laboratory implementation using real-time decisions rather than a circuit simulated after data collection.
The next evidence must show that the primitive improves a genuinely encoded quantum memory after all network and gate costs are counted. That will require more data qubits, correction of unknown logical states, lower-error local gates, faster remote-pair generation and comparisons against equivalent uncorrected operation. Tests over larger separations and multiple modules will also reveal whether latency and synchronization can remain manageable. The present work does not solve distributed fault tolerance, but it closes a concrete gap between sending entanglement through a link and using that entanglement to protect computation.