TechGlobal.news

Google team reports lower-than-physical error rates for two logical qubits in Science

Priya Nair ·

Google Quantum AI and academic collaborators report in Science today that they executed a sequence of two-qubit logical gates whose error rate fell below that of the same operations on constituent physical qubits. The work used a superconducting chip at Google’s Santa Barbara lab, and the team says it is the first time their platform has shown entangling logical operations that outperform the underlying hardware.

The result addresses a long standing gap between demonstrations of a single logical qubit with extended lifetime and the ability to build low error logical circuits. Engineers require both to plan architectures, control electronics, and facility footprints for fault-tolerant systems.

What the experiment measured

The group implemented a pair of distance-5 surface-code logical qubits with repeated syndrome extraction, then performed a calibrated entangling gate through lattice surgery. They measured logical error per entangling operation across gate sequences of increasing depth, and compared it with the aggregated physical error of the constituent two-qubit gates on the same device.

Julian Kelly, hardware co-lead at Google Quantum AI, and Hartmut Neven, who oversees the program, are senior authors on the paper. The team includes collaborators from the University of California, Santa Barbara, and EPFL. The paper builds on Google’s 2023 Nature report that showed exponential suppression of error for a single logical qubit as code distance increased, and shifts the focus to logical two-qubit performance that gates practical algorithms.

Why it matters for roadmaps

IBM, IonQ, Quantinuum, and neutral-atom groups have reported progress on error detection and small codes in Nature and Science over the past three years. The new Science paper provides a cross check on whether surface-code stacks can maintain an advantage once entangling gates and circuit depth enter the picture, which is the condition engineers care about for real workloads.

The authors do not claim useful algorithms. They chart the overhead needed to push logical error rates down further, including tighter control of crosstalk, improved calibration automation, and faster cryogenic control. For policymakers, the result narrows uncertainty around when fault-tolerant prototypes might run small chemistry or optimization kernels, and it reinforces the case for continued investment in cryo electronics and fabrication yield.

For the broader technology landscape, this moves quantum computing conversations from headline qubit counts to circuit-level error budgets, which affects procurement language, cloud service SLAs, and which toolchains developers start testing against in the second half of the decade.