Google has demonstrated a technique that cuts errors in crucial quantum states by about 40 times on its Willow processor.
According to Google Quantum AI the work tackles the expensive business of preparing “magic states”, which enable operations needed for universal quantum computing.
Some estimates suggest producing these states could consume up to 95 per cent of a fault-tolerant quantum computer’s capacity. That leaves an impressive machine spending most of its time preparing to do something useful.
Google’s researchers demonstrated an alternative called magic state cultivation, developed theoretically in 2024. Their results appeared in Science on 15 October.
The experiment achieved a reported state fidelity of 99.99 per cent, a measure of how closely the prepared state matches its intended form.
A substantial qualification remains: the protocol retained only eight per cent of attempts. Faulty attempts were discarded, so the accuracy figure describes the states that survived the checks.
Cultivation begins with an imperfect state in a small region of the chip. Coordinated measurements identify flawed attempts, allowing surviving states to be developed into a form suitable for computation. The aim is to reduce the hardware and processing resources needed to supply these essential ingredients.
Willow is Google’s 105-qubit superconducting research processor, introduced in December 2024. Earlier error-correction experiments showed that grouping more physical qubits could improve the reliability of an encoded logical qubit.
Making cultivation work required the researchers to fix assumptions in their measurement equipment and better control the chip.
Google Quantum AI research scientist Emma Rosenfeld encountered results that appeared physically implausible. Electrical drift and temperature changes were interfering with the benchmark, prompting additional controls and a new measurement method.
The team demonstrated transferring prepared states between two error-correction schemes, helping connect cultivation with Google’s wider architecture. The remaining engineering includes demonstrating the technique at larger scales and supplying reliable states throughout extended computations. Google says substantial technical challenges remain before practical deployment.
