A superconducting quantum processor can now hold its coherence long enough to run roughly 5,000 consecutive two-qubit operations plus 5,000 measurement-and-reset cycles without collapsing. That is the headline finding from IBM and five US academic institutions in Nature Physics — and it points to an error-correction mechanism that seems to grow stronger, not weaker, as quantum systems scale up.
- The hardware: The experiments ran on IBM’s Quantum Heron processor, a 156-qubit device, with the toughest trials using chains of up to 100 qubits.
- The mechanism: Alternating mixing and measurement+reset produced an abrupt phase transition at a roughly 50-50 balance between the two, with measurement driving the system toward order.
- The implication: The transition sharpened as the system grew, suggesting error correction becomes more effective — not noisier — with more qubits.
Mixing, measurement and a sharp phase transition
The work involved teams from Rutgers University, IBM, the City College of New York, Iowa State University and Penn State. They configured the processor to run two processes in continuous alternation: mixing, which exchanges information between neighbouring qubits, and measurement-plus-reset, which reads out each qubit and returns it to a known state.
The headline result is a sharp phase transition around an even split between the two. Lean toward mixing and the system drifts into chaos, qubits accumulating errors as information degrades. Lean toward measurement and order wins: errors are detected and corrected. The boundary is crisp rather than gradual, closer to the freezing point of water than to a slow drift.
For practical quantum computing, the direction of travel matters most. The transition became cleaner as the system expanded, and larger experiments showed more clearly defined frontiers between the two regimes. The correction machinery appears to scale in the right direction — a direct challenge to the intuition that more qubits necessarily mean more noise.
When measuring becomes part of the computation
Error correction remains the central bottleneck. Qubits are fragile, they couple to their surroundings, accumulate faults and shed the coherence that makes quantum computation possible. Current schemes require tens or hundreds of physical qubits to encode a single reliable logical qubit.
The experiment identifies an operating regime in which periodic measurement and reset keep the system ordered while it continues to execute complex operations. That inverts the usual mental model in which observation interrupts a calculation: here, observation is part of the calculation.
Justin Wilson of Louisiana State University, a co-author, notes that the experiment operates at a scale beyond what classical computers can simulate directly. The processor is already testing hypotheses that cannot be verified by other means — an early form of genuine quantum advantage, not in the sense of solving a problem faster than a classical machine, but of tackling one a classical machine cannot address at all.
The long road to a practical machine
“An important step toward practical error correction,” says IBM researcher Maika Takita — a careful framing that stops well short of declaring victory.
Heron is an experimental processor, not a fully fault-tolerant one, and 5,000 operation pairs, impressive by historical standards, remain short of the scale required by algorithms with known practical applications: factoring RSA keys, simulating molecular dynamics or quantum optimisation.
Other architectures are advancing in parallel. A peer-reviewed photonic design called QGATE, published on 8 October 2026 in Quantum Science and Technology by UK startup Aegiq Ltd and the University of Sheffield, reports photon-loss error thresholds of 10.36 ± 0.02% and 25.98 ± 0.28%, and reduces algorithm compile time from exponential to linear scaling with circuit complexity. Photons operate near room temperature, unlike superconducting qubits that need dilution refrigerators chilled to fractions of a degree above absolute zero.
The path from scientific demonstration to usable tool is still long. But the IBM result shortens it, because it validates a mechanism — dynamic correction through measurement and reset cycles — that scales in the right direction. Major advances in quantum computing rarely arrive as a single moment of alignment; they arrive as proof that the underlying machinery works. Conventional silicon continues to improve on its own curve, and quantum hardware is now producing evidence that its own curve bends in the right way.
Sources & References