Google’s Willow chip cuts quantum errors 40-fold in study

Google Quantum AI has shown that "magic state cultivation," a technique its researchers proposed in theory in 2024, works on real hardware. On the 105-qubit Willow processor it cut errors about 40-fold compared with uncultivated states, according to a paper published October 15 in Science. The team said it produced T-states with a record logical fidelity of 99.99%. In Google's announcement, Alphabet's quantum unit called the work "the first demonstration of a practical solution" to one of the field's worst resource bottlenecks.blog.google

Magic states are needed for the operations that let a quantum computer do things classical machines cannot. They are expensive to make. Some estimates suggest that producing them could take up to 95% of a quantum computer's capacity. Each one can only be used once.newscientist.com

How cultivation works

The standard method, called distillation, combines many imperfect states into fewer, more accurate ones. Cultivation instead works on a single state. The team encoded a rough magic state in a few qubits and checked whether it was the right state. States that passed were spread across more qubits for stronger protection. The experiment used 54 of Willow's 105 physical qubits. Without cultivation, about four attempts in 1,000 produced a faulty state, Scientific American reported.scientificamerican.com

The researchers also moved states from a distance-3 color code to a distance-5 surface code, which are two error-correction schemes. Google said this shows that purified states can survive the transfer on its hardware. During testing, lead researcher Emma Rosenfeld found that lab noise from equipment drift and temperature changes was skewing a standard benchmark. To get around it, the team created a method that uses cultivation to measure itself.blog.google

Praise and caveats

Isaac Kim of the University of California, Davis, told New Scientist that the approach "can make fault-tolerant quantum computing substantially cheaper." Dominik Hangleiter of ETH Zurich called it a potential "gamechanger" for machines with hundreds of logical qubits.newscientist.com

The method keeps only the states that pass its quality checks. Only 8% of runs produced usable magic states, and Imperial College London's Zhenyu Cai said Willow's error rates must fall further before it can tackle useful problems, New Scientist reported. The team is currently at about one error in 10,000 attempts. Breaking RSA-2048 encryption would require about one failure in 10 billion, and simulating molecules would require one in a million to one in 100 million, according to Scientific American.scientificamerican.com+1

Google acknowledged that cultivation still has to be shown at a larger scale. "This is the first demonstration that the concepts work," Rosenfeld said. "We're not done."blog.google+1