Google demonstrates magic state cultivation on Willow: 40 times fewer errors

Researchers at Google Quantum AI have shown for the first time on a real quantum chip that magic state cultivation works. This technique makes special states that a quantum computer needs to be able to carry out all kinds of calculations. On Google's Willow chip, the errors in these states became about 40 times smaller than without the technique. The team published the result in mid-October in the journal Science. An earlier version of the study had been online since December 2025.

Why magic states are so difficult

A quantum computer calculates with qubits. These are very sensitive to disturbances from outside. To calculate reliably anyway, the computer spreads each piece of information over a small group of qubits that together can detect and repair errors. With this protection, many operations work well, but not all of them. A few important operations need an extra aid: a magic state, a state that is prepared beforehand. It is used up when it is used, so every operation needs a new one.

Making such states pure enough takes many qubits. According to estimates that Google cites, up to 95 per cent of a quantum computer could be busy making them. That leaves little for the actual calculations. According to Google, researchers have seen this bottleneck coming since 2004.

Make, check and start again

The usual method is called distillation: you take many imperfect magic states and make a few better ones from them. This takes many qubits. Cultivation, devised in 2024 by the Google researchers Craig Gidney, Noah Shutty and Cody Jones, works differently. The chip makes one state and checks it again and again, with measurements that detect errors without destroying the state itself. If a check fails, the chip throws the attempt away and starts again. With the strictest checks, about 8 per cent of the attempts were left.

The states that were left were of exceptionally high quality. The fidelity, a measure of how close a state is to the intended state, came out at about 99.99 per cent. Google calls that a record.

A puzzle in the measurements

Measuring such small errors is difficult. Researcher Emma Rosenfeld, the first author of the article in Science, got results in the first experiments that, according to Google, seemed to defy the laws of nature. After months of simulations and checks, it turned out that an assumption in the usual measuring method was wrong: small fluctuations in the electronic equipment and in the temperature of the lab disturbed the measurements. The team brought the equipment under better control and devised a new measuring method, kickback tomography. It uses the technique itself to measure the remaining errors and needs far fewer measurements than the usual method.

The team also showed that a cultivated state can be transferred to a larger protected group of qubits, of the kind Google also uses in its other error-correction experiments on Willow. According to the researchers, this is a first demonstration that this works in principle.

The next step

This is an experiment on a small part of one chip. It is not yet a complete quantum computer that fully controls its own errors. The next step is to make the technique work on a larger scale, in a machine that makes many of these states one after another and uses them straight away.

What does this mean for the road to the singularity?

Quantum computers will one day have to solve problems that ordinary computers cannot. This needs many reliable operations, and so a great many magic states. If these can be made with far fewer qubits, a useful quantum computer can become a good deal smaller and simpler. This research shows on real hardware for the first time that such an economical approach works. Now the task is to scale it up, and that brings the first truly useful quantum computer one step closer.

What is the technological singularity? Read the explanation →