Google shows it can grow quantum computing's rarest ingredient on its Willow chip

Quantum computers promise huge computing power by working through many possibilities at once, Scientific American explains. The machines we have today are still far from that ideal.

The main reason is errors. Qubits, the quantum version of bits, make mistakes often. That keeps today's quantum computers limited, and it's why so much research goes into error correction: ways to spot and fix mistakes while a program runs.

Google says that to reach "universal" quantum computing, where a machine can run any useful program, you need to run and error-correct one specific set of operations. Magic states are the hard part of that set.

Regular computers build every calculation from a few simple logic gates, like AND, OR and NOT. Quantum computers have their own basic set, called Clifford gates.

Here's the twist. A normal computer can imitate Clifford gates. So on their own, they don't give a quantum computer any special power.

"But there is one particular gate that unlocks the quantum computer's power," Emma Rosenfeld, the study's lead author at Google Quantum AI, told Scientific American. Physicists call it a non-Clifford gate, most often a T gate.

To run a T gate, you need a magic state. That is a qubit set up in a very specific blend of 0 and 1. It gets fed into a calculation and used up.

• Single use: once a magic state is used, it's gone.
• No room for flaws: any error in the state becomes an error in the gate.
• Precision: "It has to be made very, very precisely," Rosenfeld said.

Google's blog puts it simply: you need magic states to go beyond what classical computers can do.

Making clean magic states is expensive. Researchers saw this coming back in 2004, long before quantum computers were a real engineering project.

By the numbers: some estimates say up to 95% of a quantum computer's capacity could go into making magic states, Google says. That would leave very little room for actual computing.

The standard recipe is called distillation. It takes many imperfect states and boils them down into fewer, more accurate ones. It works, but it's hungry.

Think of it like making maple syrup. You boil a huge pot of sap to end up with a small jar. Distillation is the sap-boiling approach: most of the effort goes into the pot.

For two decades, researchers have been chipping away at that cost.

In 2024, Google researchers Craig Gidney, Noah Shutty and Cody Jones refined an idea from earlier research into what they named magic state cultivation. It works on a single state instead of many.

Google describes it as a filter:

• Plant: inject a noisy magic state into a small region of the chip.
• Check: run a series of tightly coordinated measurements to look for errors.
• Weed: throw away any attempt that fails a check.
• Grow: spread the states that pass across more qubits for stronger protection.

There's a clever trick in the checking step. Measuring a quantum state usually collapses it, forcing it to pick 0 or 1. So instead of asking "Are you 0 or 1?", Scientific American explains, the team asked: "Are you the right magic state: yes or no?"

That's where the gardening name comes from. The state starts small, gets checked, then grows.

In real life Picture a bakery that only sells perfect croissants. Every batch gets inspected, and anything slightly off goes in the bin. You waste dough, but every croissant on the shelf is flawless. Cultivation makes the same trade.

On paper, cultivation looked great. Making it work on real hardware took a lot more.

The catch: when Rosenfeld first tested the protocol, her data seemed to defy the laws of physics. Eventually she traced it to a wrong assumption in the team's go-to measurement method. Noise from drifting electrical equipment and temperature changes in the lab was skewing the benchmark.

It took months of simulations, discussions with colleagues and consistency checks to find it. Google says the fix required a pivot in both its theory and its engineering.

The fix had two parts. The team added extra control to reduce that noise. Then it invented a new benchmark that uses cultivation to measure itself.

The results, according to Google and Scientific American:

• Errors: about 40 times lower than magic states made without cultivation.
• Fidelity: T-states with a record logical fidelity of 99.99%.
• Before and after: about four faulty states in 1,000 without cultivation, roughly one in 10,000 with it.
• Hardware: 54 of Willow's 105 physical qubits.

Google calls it the first demonstration of a practical solution to one of quantum computing's most severe resource bottlenecks.

The team also showed it could move a purified state between two kinds of error correction. It grafted a state from a distance-3 "color code" (a newer method that suits cultivation) onto a distance-5 "surface code" (the more established one). The state survived the transfer, which suggests the theory fits Google's hardware design.

"The Google experiment shows that cultivation works in practice," Kenneth Brown, director of the Duke Quantum Center at Duke University, told Scientific American. He was not involved in the study.

Brown also pointed out the cost. The quality check rejects low-quality states, and it discards 92% of attempts. That's a lot of croissants in the bin.

Still a long way. Scientific American lays out the targets:

RSA-2048 is the encryption used in website security, banking and electronic ID cards. So no, your bank account is not in danger this week.

Rosenfeld says molecule simulation is the more reachable goal. Google's blog adds that it still needs to show cultivation at a larger scale, and plenty of technical challenges remain.

What's next: the team wants to push toward "near-term scientific applications." In Rosenfeld's words: "We're more galvanized than ever."

For Google, the stated goal is a useful, error-corrected quantum computer that solves problems ordinary machines can't. This result is one step on that road.

• Nothing changes today: this is lab research, not a product. No quantum computer is running your apps.
• Smaller machines, sooner: cheaper magic states could shrink the quantum computer needed for practical tasks, Google says.
• Encryption is safe for now: today's error rates are about a million times too high to break RSA-2048.
• Watch for chemistry: molecule simulation is the first goal researchers call realistic.