Google slashes errors by 40 times in a critical quantum computing operation

Google demonstrates on its Willow chip a method reducing quantum state preparation errors by 40 times, with a logical fidelity of 99.99%.

Producing a quantum state reliable enough to perform a computation can consume a large portion of a quantum computer's resources. With its Willow chip, Google Quantum AI experimentally demonstrates a method designed to reduce this constraint. Dubbed magic state cultivation, it allows the preparation of specific quantum states with 40 times fewer errors than the baseline method used in the experiment, along with a reported logical fidelity of 99.99%. The results, published in the journal Science, address one of the technical hurdles to building quantum computers capable of executing complex calculations with error correction. States essential to certain quantum computations

Quantum computers use qubits to manipulate information. Unlike classical bits, they can exploit the phenomena of superposition and entanglement.

These properties offer unique computational capabilities, but they also make systems sensitive to perturbations.

Interactions with the environment, hardware imperfections, and measurement operations can introduce errors into the results.

To build a quantum computer capable of reliably running complex programs, researchers must combine multiple operations and protect them against these errors.

Certain operations can be performed using quantum gates that are relatively simple to correct. Others require additional resources, specifically special states known as magic states.

These states notably enable the execution of so-called non-Clifford operations, which are necessary to achieve a universal set of quantum operations.

However, their preparation is costly, as they must be produced with very high precision before being used. Up to 95% of resources dedicated to their preparation

The magic state problem has been recognized for over twenty years.

Traditional methods rely in particular on distillation, which involves using multiple imperfect states to obtain more reliable ones.

This preparation can mobilize a significant number of physical qubits and correction operations.

According to theoretical estimates cited by Google, up to 95% of a quantum computer's capacity could be devoted solely to producing these states.

Such a proportion would leave relatively few resources to run the computations for which the machine is intended.

Another difficulty stems from their consumption: a magic state used to perform one operation cannot be reused for another.

Programs requiring numerous non-Clifford operations must therefore have a continuous supply of sufficiently reliable states.

Reducing the number of qubits, operations, and time required for their preparation is thus a major goal for error-corrected quantum architectures. Cultivating a quantum state rather than distilling it

The method tested by Google builds on theoretical work by Craig Gidney, Noah Shutty, and Cody Jones, developed in 2024.

Their approach, called magic state cultivation, consists of gradually preparing a high-quality quantum state within a small region of a processor.

The protocol begins with the injection of an imperfect state, which is susceptible to containing errors.

A series of coordinated measurements then allows controlling its evolution and detecting certain anomalies.

When an attempt shows signs of error, it can be rejected. States that pass the checks are retained and progressively integrated into a larger quantum encoding.

This method differs from distillation techniques that rely on combining multiple states to improve their quality.

Magic state cultivation combines error detection with the preparation and extension of the logical state.

The process nevertheless requires repeating attempts when checks detect an unsatisfactory result.

In the published experiment, approximately 8% of the attempts are retained after this selection. This proportion is important for evaluating the actual resources required for steady production. Forty times fewer errors on Willow

Google Quantum AI researchers implemented the protocol on Willow, the superconducting quantum processor developed by the group.

The demonstration establishes an approximate 40-fold reduction in the error rate compared to states prepared without the cultivation process.

The team also achieved T states with a logical fidelity of at least 99.99%.

Fidelity here measures the closeness between the prepared quantum state and the expected theoretical state.

This result pertains to a specific resource required for certain quantum operations, and not to the overall reliability of all computations executed on Willow.

Similarly, the factor of 40 describes an error reduction. It does not mean that the processor's operating cost, power consumption, or hardware requirements have been divided by forty.

The scientific paper, titled Magic state cultivation on a superconducting quantum processor, describes the experimental methods and measurements performed on Google's hardware. A measurement issue that required a new method

Implementing the protocol also revealed difficulties in how to evaluate its precision.

Emma Rosenfeld, a researcher at Google Quantum AI, observed results during initial experiments that seemed inconsistent with expected physical behavior.

After several months of simulations, verifications, and exchanges with other researchers, the team identified an incorrect assumption in the measurement method being used.

Certain environmental perturbations in the laboratory, related in particular to electrical drifts and temperature fluctuations, were affecting the validity of the measurements.

These phenomena could give a misleading representation of the quality of the prepared states.

Researchers therefore tightened the control of their system to limit the influence of these perturbations.

They also developed a new evaluation method using the cultivation process itself to measure the properties of the resulting states.

This adaptation made it possible to better characterize the results and proceed with the experimental demonstration.

Google presents this work as an advancement in quantum