Google cuts quantum error rates 40-fold with new chip technique
Macro & Markets ·
A Science paper published Thursday describes how Google's Willow processor slashed errors in the scarce computational resources quantum machines depend on, narrowing a gap that has long stood between today's hardware and practical, large-scale quantum computing.
Google's Quantum AI team reported that a method called magic state cultivation lowered error rates by roughly 40 times on the Willow chip, reaching a logical fidelity of 99.99% for the resulting states, according to Decrypt. Magic states are single-use building blocks required for the harder operations that standard error correction cannot protect on its own; without them, a quantum computer cannot run many of the programs considered genuinely useful. Google has said some estimates place the cost of generating these states at up to 95% of a quantum computer's total capacity, meaning most of a machine's effort can go toward producing fuel rather than doing calculations.
The conventional fix, known as magic state distillation and first outlined by physicists Sergei Bravyi and Alexei Kitaev in 2004, combines many flawed magic states into fewer clean ones but requires large banks of qubits dedicated solely to that purpose. Cultivation, a method Google researchers proposed in 2024, instead works more like a selective filter: a rough magic state is created in a small section of the chip, checked repeatedly for errors, discarded if any flaw turns up, and allowed to improve only when it survives the checks. In the new experiment, only about 8% of attempts passed every check, meaning more than nine in ten tries were thrown out to yield one usable, clean state.
The significance extends beyond laboratory efficiency. Cheaper magic-state production shortens the path toward quantum computers capable of tasks such as drug and materials design, but it also moves closer the point at which a sufficiently powerful machine could undermine encryption protecting bank accounts and Bitcoin wallets. Cultivation is one of the techniques behind a 2025 estimate, noted in the same report, that RSA-2048 encryption could theoretically be broken using fewer than a million noisy qubits, down sharply from an earlier 2019 estimate of 20 million. Actually breaking that encryption, however, would require magic states with error rates near one in 10 billion, far beyond what has been demonstrated so far.
Until this experiment, cultivation had only been validated through theoretical modeling and simulation rather than on physical hardware. The cluster around this report notes four distinct sources covering the development, though it remains unresolved how quickly cultivation can scale to the error rates needed for real cryptographic risk, or how soon such capability might move from research chips to operational systems.