Google Cuts Errors 40-Fold in a Key Quantum Computing Step on Its Willow Chip
Google Quantum AI showed on real hardware that a technique called "magic state cultivation" works, producing the special quantum states needed for full-scale quantum computing with about 40 times fewer errors; the results were published in the journal Science.
Google's quantum computing team showed that a technique called "magic state cultivation" works on its Willow quantum chip, cutting errors about 40-fold. The results were published in the journal Science.
Quantum computers store information in qubits, which can hold combinations of 0 and 1 at once, but they make errors easily. To run useful programs, a quantum computer needs a supply of special, extremely clean quantum states, which researchers call "magic states." Making them has been one of the field's biggest costs: some estimates put it at up to 95% of a machine's capacity.
Cultivation works like a filter. The chip starts with a rough version of the state in a small patch of qubits, repeatedly checks it for errors, throws out flawed attempts and keeps the clean ones. On Willow, the kept states reached at least 99.99% accuracy, which Google calls a record, with about 8% of attempts kept.
Emma Rosenfeld, a Google research scientist and the paper's first author, spent months tracking down test data that seemed to defy the laws of physics; the cause turned out to be noise in the lab from drifting electrical equipment and temperature changes. The method still has to be shown at larger scale before it can support practical machines.
Mountain View, Calif. — Google's quantum computing team showed that a technique called "magic state cultivation" works on its Willow quantum processor, producing a crucial ingredient for full-scale quantum computing with about 40 times fewer errors than without it.
The results were published in the journal Science. The paper's authors are researchers at Google Quantum AI, the company's quantum computing lab.
Ordinary computers store information in bits that are either 0 or 1. Quantum computers use qubits, which can hold a blend of both at once and can be linked together, which lets them tackle certain problems, such as simulating molecules for chemistry, that would overwhelm any ordinary machine. Qubits are also fragile, and stray heat or electrical noise knocks them into errors. Quantum error correction spreads one reliable "logical" qubit across many physical qubits so mistakes can be caught and fixed.
Error correction handles most operations well, but not all of them. To run any program a quantum computer might need, it must also perform a few operations that error correction cannot protect directly. The standard workaround is to prepare special quantum states, called "magic states," and feed them into the calculation. Each one is used up when it is consumed, and making them clean enough has been expensive. According to Google, some estimates found that up to 95% of a quantum computer's capacity could go to producing magic states, leaving little for actual computing. Researchers have been working to lower that cost since 2004.
In 2024, Google researchers Craig Gidney, Noah Shutty and Cody Jones worked out the cultivation method on paper. The chip injects a noisy magic state into a small region of qubits, then runs a series of tightly coordinated measurements that check for errors and discard any flawed attempt. What survives is an exceptionally clean state that can then be moved into a larger, more protected encoding. The Science paper reports that cultivation cut the error by a factor of 40 and produced states with a fidelity, or accuracy, of at least 99.99%, while keeping about 8% of attempts. The team also moved the cleaned states from one kind of error-correcting code, called a color code, into another, the more widely used surface code, and showed that they survived the transfer.
Getting there took a detour. Emma Rosenfeld, a Google Quantum AI research scientist and the paper's first author, was seeing test data that seemed to defy the laws of physics. After months of simulations and checks, she found that drifting electrical equipment and temperature changes in the lab were throwing off the team's standard way of measuring results. The team added extra control to reduce that noise and invented a new benchmark that uses cultivation to measure itself.
Google calls the work the first demonstration of a practical solution to one of quantum computing's most severe resource bottlenecks. The method still has to be shown at larger scale before it can support practical machines, and the company says "plenty of technical challenges remain."