A team from the University of Innsbruck and RWTH Aachen University has demonstrated the first complete fault-tolerant quantum algorithm that never stops to measure itself. The result, published in Nature Communications, replaces the interrupt-driven architecture that has defined fault-tolerant quantum computing with something closer to continuous operation: error information processed in real time using standard quantum gates, no pause required.
The work matters because mid-circuit measurement is one of the most expensive operations in encoded quantum systems. When a quantum computer encodes logical qubits across many physical ones to detect and correct errors, the standard approach requires periodically stopping the computation, measuring some qubits to extract error information, classically processing that data, and then deciding what corrective gates to apply. Each pause introduces latency and new opportunities for noise. The Innsbruck-Aachen team proposed and built a version of this cycle that runs entirely within the quantum processor.
"Rather than stopping the computation to read out error information and classically deciding on a correction, the new approach processes error information coherently," said Friederike Butt, one of the paper's lead authors, in a university statement. "That happens entirely within the quantum computation itself, using only standard quantum gate operations."
The researchers ran Grover's quantum search algorithm on three logical qubits encoded across eight physical qubits of a 16-qubit trapped-ion processor. The experiment identified the correct solution states, according to the arXiv preprint, providing a proof-of-concept for measurement-free fault-tolerant operation. Thomas Monz, who leads the Innsbruck group, called it "a new paradigm for quantum error correction."
The headline result comes with the usual fine print that anyone following fault-tolerant quantum computing needs to hold carefully. Physical two-qubit gate fidelity on the trapped-ion hardware sits around 99.6 percent, which is excellent. The logical gate fidelities achieved in the demonstration are lower: up to 93 percent for state teleportation and 95 percent for a logical Hadamard gate. That gap between physical and logical performance is the cost of encoding and it is not trivial. Anyone telling you that fault-tolerant quantum computing is just a matter of running the same gates on more qubits should be asked about this number.
The team also demonstrated a practical advantage: resetting an auxiliary qubit used for error detection took 1.7 milliseconds, compared to approximately 30 milliseconds for a mid-circuit measurement on the same hardware. The Nature Communications paper notes that this 18-fold speedup matters because error detection in the [[8,3,2]] code used here requires repeated qubit resets throughout the algorithm. Faster resets reduce the total overhead from error management.
The theoretical framework came from Butt and Markus Müller at RWTH Aachen University and Forschungszentrum Jülich. The experimental implementation ran at Innsbruck on hardware from Alpine Quantum Technologies, a spin-off commercializing the university's trapped-ion technology. The architecture is not a replacement for measurement-based error correction. It is a complement that works well where measurement is slow relative to gate times, which describes trapped-ion hardware well. Whether it transfers to superconducting qubit systems, where measurement is faster but gate fidelity has historically been lower, remains an open question the paper acknowledges directly.
The result is a real one: three logical qubits running Grover fault-tolerantly without a single interruption. It is also a small algorithm on a small machine. The gap between 99.6 percent physical fidelity and 93-95 percent logical fidelity is the cost of encoding, and it is not small. Replication and extension by other groups will determine whether the approach scales.
The paper is "Demonstration of measurement-free universal logical quantum computation," Butt et al., Nature Communications (2026) 17:995, DOI: 10.1038/s41467-026-68533-x.