Google Cuts Quantum Magic-State Errors 40-Fold on Willow

Google Quantum AI reported a roughly 40-fold reduction in errors when preparing “magic states” on its Willow quantum processor, demonstrating a technique that could reduce the hardware needed for useful quantum computing.

The result, announced Oct. 15 and published in Science, achieved a state fidelity of “at least 0.9999(1)”—about 99.99%—while retaining 8% of attempts. Fidelity measures how closely the prepared state matches the target quantum state, not the accuracy of an entire computation.

For Bitcoin and Ethereum, the advance strengthens the experimental basis for resource-saving quantum computing techniques. It does not demonstrate a cryptographic attack. Project Eleven CEO Alex Pruden said cultivation was already assumed in earlier quantum resource estimates, so the result does not necessarily bring forward the date when quantum machines could break today's cryptography.

Magic states are consumed to perform operations needed for universal, error-corrected quantum computing. Producing them can absorb much of a machine's resources. Google's research post describes cultivation as a filter: researchers inject a noisy state, run coordinated measurements and discard attempts where they detect errors.

The team's earlier preprint reports an injection-only logical error probability of about 0.44%, compared with a cultivated-state error upper bound of about 0.01%. The high-fidelity benchmark included two additional cultivation rounds and three quantum error-correction cycles to check the state. Cultivation prepared a single logical qubit; Willow contains 105 physical qubits.

The Gap to Breaking Crypto Keys

The relevant threat to Bitcoin and Ethereum transaction signatures is Shor's algorithm solving the elliptic-curve discrete logarithm problem, allowing an attacker to derive a private key from a public key.

Google's March resource study modeled that problem on secp256k1, the curve used for those signatures. Its two circuit designs use up to 1,200 logical qubits and 90 million Toffoli gates, or up to 1,450 logical qubits and 70 million such gates. Logical qubits are error-corrected units built from many physical qubits; Toffoli gates are computational operations that help determine runtime.

Google's faster modeled attack would take roughly nine minutes after public-key exposure on a machine using fewer than 500,000 physical qubits, assuming superconducting hardware and advance computation of the algorithm's key-independent first half. Those are modeled resources, not demonstrated capabilities. The Defiant covered the study in March.

That study already used the cultivation proposal to estimate magic-state production costs. The 40-fold improvement reported in Science measures state error, not a 40-fold reduction in the qubit count or runtime of a key-breaking attack.

“This doesn't necessarily move Q-Day timelines forward,” Pruden wrote on X. “But it does experimentally validate one of the key assumption underlying those estimates, making them more credible.” Project Eleven works on post-quantum migration.

Google's announcement focuses on producing usable quantum states rather than claiming a new encryption-breaking capability. Researchers also demonstrated transferring states between two error-correction encodings, from a distance-3 color code to a distance-5 surface code. But scaling remains unfinished: “we still need to demonstrate magic state cultivation at a larger scale to make it realistic for practical quantum computing,” Google researchers Emma Rosenfeld and Craig Gidney wrote.