Google Cuts Willow Quantum Processor Errors 40-Fold in Magic-State Test

Google Quantum AI says its latest Willow experiment is a step toward making quantum computations more reliable, after researchers showed they could prepare and filter "magic states" with much lower error rates than before. The work, published in Science, used a staged cultivation process on the superconducting processor rather than producing a large batch all at once.

The accepted magic states had a reported fidelity of 99.99%, but only about 8% of attempts passed the researchers' checks. Attempts with detected errors were discarded, and the team started again.

Researchers transferred cultivated states from a distance-3 color code into a distance-5 surface code, testing how they moved between two error-correction schemes. The experiment used no more than 54 of Willow's 105 qubits.

The work also faced measurement challenges: temperature changes in the lab and drift in electrical equipment interfered with the results. The team refined its system monitoring and developed a new method to assess the cultivation process.

The experiment does not yet establish how many resources the approach could save in a larger system. Google plans to test magic state cultivation at a greater scale.

The reported error rate also remains far from the level cited as necessary for factoring RSA-2048: that estimate requires magic-state errors about one million times lower than those achieved in the experiment.

What would make this result convincing as a practical advance: a higher pass rate, a larger-scale test, or a clearer estimate of resource savings?