In the 9 December 2024 issue of
Nature [
1], a team of Google engineers reported breakthrough results using “Willow,” their lat-est quantum computing chip (
Fig. 1). By meeting a milestone “below threshold” reduction in the rate of errors that plague super-conducting circuit-based quantum computing systems (
Fig. 2), the work moves the field another step towards its promised super-charged applications, albeit likely still many years away. Areas expected to benefit from quantum computing include, among others, drug discovery, materials science, finance, cybersecurity, and machine learning.
Despite steady progress in the past several years, superconduct-ing circuit-based quantum computing systems have remained unstable and highly vulnerable to errors. To reduce the errors, the Google team applied a combination of more precise hard-ware—Willow employs 105 physical qubits, more than double the number of Sycamore [
2], its previous iteration—and improved algorithms. The stability enabled by these advances allowed the chip—in another demonstration of quantum supremacy—to com-plete a highly specific task in five minutes, versus an estimated 10 septillion years—that is one followed by 25 zeros—for one of the world’s fastest supercomputers [
1,
3,
4].
“This is one of the most advanced superconducting quantum computing chips currently available,” said Daniel Lidar, professor of engineering at the University of Southern California (USC; Los Angeles, CA, USA), director of the USC Center for Quantum Informa-tion Science and Technology, and co-director of the USC-Lockheed Martin Quantum Computing Center. “It is top of the line in terms of the conventional parameters we use to characterize the quality of qubits.”
Qubits are the basic building blocks of quantum information. Unlike the bits used in classic computing, which can store informa-tion as a 1 or a 0, qubits store a 1, a 0, or a superposition of both. This versatility facilitates the design of algorithms that can quickly tackle problems that would take classical computers an inordinate amount of time to solve.
Unfortunately, qubits are quite delicate, with their superposi-tions prone to disturbance by environmental perturbations. With-out effective correction of these anticipated disturbances, or errors, qubits lose stability too quickly to perform useful computations. To counter errors in fragile qubits, researchers have developed elabo-rate schemes to create “llogical qubits’’ that spread information across multiple “physical’’ qubits in superposition, making them more resilient to noise. Though several companies and academic research groups have in recent years shown that such approaches can produce minor improvements in accuracy [
5-
7], the Google results with Willow are the first to demonstrate a “below-threshold” rate of quantum error correction. That is, Google engineers have shown that error correction improves exponen-tially as the scale—the number of qubits—increases. Typically, as the number of qubits in a system increases, so does the number of errors. Error correction can bring the system to a break-even point where when qubits are added, errors are corrected at a cor-responding pace. With Willow, however, the more qubits that are added, the more efficiently errors are corrected, bringing the system below that threshold.
“It is a feat of engineering,” said Bert de Jong, senior scientist at Lawrence Berkeley Laboratory in Berkeley, CA, USA, where he also directs the facility’s quantum systems accelerator. “However, the key advance for Willow is not just the chip itself, but being able to do the error correction smartly, quickly, and efficiently.”
In 2019, to great fanfare, Google announced the first beyond-classical computation (i.e., quantum supremacy) with its 54-qubit Sycamore quantum processor, which encodes each physical qubit in a superconducting circuit [
8,
9]. In early 2023, the company reported using a modified 49-qubit version of its Sycamore quan-tum processor to demonstrate the feasibility of quantum error cor-rection improving performance with increasing numbers of qubits [
2]. Willow, a scaled-up version of that technology, was wholly fabricated at Google’s Quantum AI campus in Santa Barbara, CA, USA [
3].
In bench tests, Google researchers demonstrated that as its chips’ qubits increased, from a 3 × 3 to a 5 × 5 to a 7 × 7 lattice of physical qubits, the error rate dropped each time by a factor of about 2.14 compared to its Sycamore predecessor [
4]. “They could not yet demonstrate an improvement going to, let us say, a 9 × 9 lattice since that would require about 150 qubits,” Lidar said. “In addition, it is possible the noise would start growing again, or their control circuitry would not keep up—this remains to be seen.”
As mentioned above, Willow’s robust error correction capabili-ties allowed it to perform a standard benchmark computation in less than five minutes that would take Frontier, one of the world’s fastest supercomputers [
10], an estimated 1025 years. The task— random circuit sampling—was created by Google engineers specif-ically to showcase the potential power of quantum computers. The company used the same task for its initial demonstration of quan-tum supremacy in 2019 [
9]. “It is a problem that is cherry-picked for a quantum computer to shine, a way to quantify performance without offering a practical solution to anything anyone really cares about, except random number generation, which has certain niche applications,” Lidar said. Chris Monroe, professor of electrical and computer engineering and physics at Dule University (Durham, NC, USA) and director of the Duke University Quantum Center, agreed. “It is totally useless,” Monroe said. “The only point is to prove, very indirectly, that your quantum computer is plugged in.”
Nevertheless, Willow’s stability results also suggest that scaling up the number of physical qubits in future chips will provide fur-ther improvements in the error rate. But for practically useful quantum computing, that stability needs to ramp up to one error per ten million steps [
11]. The Google Willow team estimates that achieving an error rate that low will require each logical qubit to be comprised of around 1000 physical qubits, although further improvements in error-correction techniques could perhaps bring that down to as low as 200 physical qubits [
1,
11].
Lidar said that Google has been able to make better qubits using improved engineering techniques and removing a variety of impu-rities in its chips, in part by using better materials. “Your physical qubits have to already be really, really good in order to see a logical qubit improvement like that,” Lidar said. “Certain key details about the hardware are sparse because Google, like its competitors, does not really release them.” Similarly, de Jong said he believes that Google will keep improving fidelity through better manufacturing. “Striving for perfection in materials is important because every imperfection leads to errors and noise,” he said.
On the software side, Google uses techniques like dynamic decoupling, where electromagnetic pulses are applied to the qubits to suppress environmental noise [
12,
13]. The pulses essentially freeze a quantum system in its initial state, effectively halting decoherence, thereby preserving entanglement and extending the operational lifetime of qubits.
Besides ramping up the number of physical qubits, other chal-lenges remain in the production of a practically useful quantum com-puting chip, including networking lots of logical qubits together so that they can exchange quantum states [
11]. Willow’s connectivity is relatively sparse, meaning every qubit is connected to a relatively small number of other qubits. “You have to start swapping informa-tion around, but that can lead to loss of coherence along the way,” Lidar said. “But if you can do quantum error correction successfully, then it matters a lot less that you need to do all these swaps because now you are protected against decoherence.”
Other quantum computing architectures that are based on photons, trapped ions, or neutral atoms [
14], where, in principle, every qubit is connected to every other qubit, are not as subject to errors caused by environmental noise. While an all-photon quantum computer has achieved quantum supremacy [
14], none of these alternative quantum architectures have yielded a sys-tem that experts agree can solve a real-world computing prob-lem, let alone solve one more efficiently than a classical computer.
Regardless of what architectures ultimately prove fruitful, Monroe sees a bright future for companies pursuing quantum computers. “If you fast forward 30 years from now, I would be surprised if we do not have quantum computers that work, and nobody cares what is inside,” he said. De Jong also sees long timelines regarding quantum computers that can address a wide set of industrial applications. “It is a long shot that we will see broad commercial value in the next five years, maybe even the next 10 years,” he said. “But once value is cre-ated, whatever the first application ends up being, more investment will flow into these machines to make use of them in many commer-cial applications.”
Lidar said he believes Willow’s ultimate legacy will be as a step-pingstone to something bigger. “It is a step along the path of mak-ing better and better logical qubits,” he said. “There is no question where things are headed: bigger chips, more qubits, larger logical qubits, and so on.”
PDF
(689KB)
