Google Quantum AI has demonstrated magic-state cultivation on its Willow processor, reporting a reduction in errors for states needed by universal quantum computations. The Google account of the work says the research has been published in Science.
Target a costly quantum resource
Magic states support operations needed for useful, error-corrected quantum programs. Producing sufficiently reliable states consumes resources, and a state cannot be reused after it has been consumed.
Cultivation aims to reduce that cost by starting with a noisy state, checking it through coordinated measurements and discarding flawed attempts. Google's researchers developed the approach in earlier theoretical work before demonstrating it on hardware.
Separate the reported measurement from the wider goal
Google reports a 40-fold error reduction relative to uncultivated states. It also reports T-state logical fidelity of 99.99% for the demonstrated protocol.
The account describes a measurement problem encountered during development: equipment drift and temperature variation interfered with an assumption in the team's benchmark. The researchers added control and developed a benchmarking approach using cultivation to measure itself.
That detail helps explain why the result depends on measurement design as well as the quantum operations. A numerical improvement is meaningful within the protocol and comparison being reported.
Larger-scale operation remains to be shown
The team transferred states between color-code and surface-code encodings. The reported demonstration grafted a state from a distance-3 color code into a distance-5 surface code, testing a connection needed by the proposed architecture.
Google says larger-scale cultivation still needs to be demonstrated. The work addresses a resource bottleneck on real hardware, but does not establish that a generally useful error-corrected quantum computer is already available.
For readers tracking the field, the next evidence to watch is whether the protocol preserves its benefit as the implementation scales and joins the other parts of a complete computation.