Is Google’s new Willow quantum computer really such a big deal?

Google has announced a new quantum computer, reasserting its lead in the race to prove that these unusual machines can beat even the world’s best conventional supercomputers. So does this mean we’ve finally arrived at a useful quantum computer?

Researchers at the tech giant unveiled their quantum computing chip Sycamore in 2019, becoming the first in the world to demonstrate this feat known as quantum supremacy. But since then, supercomputers have caught up and left Sycamore behind. Google is currently building a new quantum chip called Willow, which Google Quantum AI’s Julian Kelly says is the company’s best yet.

“You could think of this as having all the benefits of Sycamore, but when you look under the hood, the geometry has changed…We’ve rethought the processor,” he says.

The latest version of Sycamore boasted 67; The quantum bits, or qubits, that process information have been upgraded to Willow’s 105 qubits. Ideally, larger quantum computers should be more powerful, but researchers have found that qubits in larger devices struggle to remain coherent and lose their quantum nature. I discovered it. This is also the case with competitors IBM and California-based startup Atom Computing, both of which recently debuted quantum computers with more than 1,000 qubits.

For this reason, the quality of the qubits is a big focus for the team, and Willow’s qubits can store complex quantum states, reliably encoding information more than five times longer than Sycamore’s qubits, Kelly said. says.

Google uses a specific benchmark task called RCS to evaluate the performance of its quantum computers, and Willow excelled in this respect, said Hartmut Neven, also of Google’s Quantum AI. This task involves verifying that the distribution of numerical samples output by programs running on the chip is as random as possible. For several years, Sycamore was able to do this faster than the world’s best supercomputers, but in 2022 and again in 2024 a new record was set by a conventional computer.

Google says Willow’s task took five minutes on a chip, once again widening the gap between quantum and conventional machines, but the company says its prior technology would take 10 septillion years, or the age of the universe. We estimate that it will take much longer than the square of the supercomputer.

For this comparison, the researchers modeled a Frontier supercomputer (recently downgraded to only the second most powerful supercomputer in the world) with more memory than is currently available. This only emphasizes Willow’s computational abilities. says Naven. Although Sycamore’s record has been broken, he is confident Willow will remain champion for much longer as traditional computing methods reach their limits.

What remains to be seen is whether Willow can actually do anything useful, given the lack of practical use for RCS benchmark tests. Kelly said that success in benchmarks is a “necessary but not sufficient” condition for a quantum computer’s usefulness, but that chips that fail to perform well in RCS are unlikely to be used in the future.

But the Google team has another reason to believe in Willow’s bright future. That said, Willow is very good at correcting her own mistakes. Quantum computers’ propensity for error is one of the biggest current problems preventing them from fulfilling their promise of being more powerful than other types of computers. To improve this, researchers, including teams at Google, are grouping physical qubits together to form “logical qubits,” which are much more resilient to errors.

Using Willow, the team showed that as logical qubits get larger, they become more error-proof, with about half as many errors as the physical qubits that make up logical qubits. Moreover, when the size of the logical qubit was roughly doubled, the error rate was further halved. In this way, Google researchers believe they can continue to increase the number of qubits, making quantum computers larger and larger and capable of performing increasingly greater calculations than previously trending. Threshold reached.

“In my opinion, this is a signature achievement, and although we are still far from demonstrating a practical quantum computer, it is an important and necessary step towards that goal,” said Andrew Cleland of the University of Chicago. states.

Martin Wides, from the University of Glasgow in the UK, said the research points the way towards building “fault-tolerant” quantum computers, meaning they can catch and correct all errors. Although challenges remain, he says these advances pave the way for innovative applications in quantum chemistry, such as cryptography and machine learning, as well as drug discovery and materials design.

The increased focus on error correction in academic labs and across the burgeoning quantum computing industry has made advances in logical qubits a key point of comparison for today’s best quantum computers. In 2023, a team of researchers from Harvard University and the startup QuEra set a record for the most logical qubit ever created using a qubit made from cryogenic rubidium atoms. did. Earlier this year, researchers at Microsoft and Atom Computing linked a record number of logical qubits through quantum entanglement.

Google’s approach is different. Because instead of maximizing the number of single logical qubits, the focus is on making single logical qubits bigger and better. “We could have split the chip into smaller logical qubits and run the algorithm, but we really wanted to reach this threshold. There are all the fundamental challenges in science and engineering,” Kelly says.

But ultimately, the biggest test of Willow’s impact will be the goal that all other quantum computers also pursue: reliably computing things that are useful but impossible for classical computers. The question will be whether it can be achieved. Neven said Sycamore was already used for scientific discoveries such as quantum physics, but the team is setting its sights on more real-world applications with Willow. “We are moving toward new calculations and simulations that could not be performed on classical computers.”