TechGlobal.news

Science publishes first cross‑platform benchmark of quantum error correction performance

Priya Nair ·

The journal Science on Thursday published a blinded, cross‑platform benchmark of quantum error correction that measures how well logical qubits outperform physical qubits on current hardware. The study, coordinated by the U.S. National Institute of Standards and Technology with academic and industry partners, reports results from trapped‑ion, superconducting, and neutral‑atom processors tested in the first half of 2026.

The work matters because it replaces headline qubit counts with comparable logical error metrics that procurement teams and funding agencies can use. It defines test circuits and reporting formats that manufacturers can adopt without revealing proprietary calibration details.

The team measured logical error per circuit layer, break‑even conditions where a logical qubit fails less often than the best physical qubit, and how error scales with code distance under repeated two‑qubit operations. Platforms implemented small surface, color, and Bacon‑Shor codes with distances three to seven, using between 25 and roughly 200 physical qubits.

Across devices, the study found that trapped‑ion systems achieved consistent logical improvement over the best physical qubit for sequences up to tens of entangling layers. Neutral‑atom arrays achieved logical improvement for short circuits and shallow depths, with performance limited by correlated errors and crosstalk during parallel operations. Superconducting processors reduced error on idle logical memories but did not yet show net logical gain under the study’s multi‑layer entangling circuits.

The authors caution that the results are a snapshot of calibration and control as of mid‑2026. They report that correlated two‑qubit errors and leakage dominate logical failure across all modalities, which lowers the effective threshold compared with simple depolarizing models used in simulations.

Participating groups included NIST Boulder, the University of Maryland Joint Quantum Institute, Harvard University, and industrial labs from IBM, Quantinuum, QuEra, and others. Named contributors include Scott Glancy and Emanuel Knill at NIST, Sarah Sheldon at IBM, Brian Neyenhuis at Quantinuum, and Mikhail Lukin at Harvard.

The benchmark specifies open test circuits, error reporting templates, and a dataset that laboratories can reuse. It recommends targets for two‑qubit gate error below one part in a thousand with bounded correlations and verified leakage detection to achieve logical gains at modest code distances.

For the broader technology landscape, the paper gives buyers and regulators a consistent way to evaluate quantum hardware. It shifts attention from raw qubit counts to logical fidelity under workload, which is the metric that will determine which platforms secure early commercial and public sector contracts.