Last month, physicists at Toronto-based startup Xanadu published a curious experiment in Nature in which they generated seemingly random numbers. During the pandemic, they built a tabletop machine named Borealis, consisting of lasers, mirrors, and over a kilometer of optical fiber. Within Borealis, 216 beams of infrared light bounced around through a complicated network of prisms. Then, a series of detectors counted the number of photons in each beam after they traversed the prisms. Ultimately, the machine generated 216 numbers at a time—one number corresponding to the photon count in each respective beam.
Borealis is a quantum computer, and according to the Xanadu researchers, this laser-powered dice roll is beyond the capability of classical, or non-quantum, computing. It took Borealis 36 microseconds to generate one set of 216 numbers from a complicated statistical distribution. They estimated it would take Fugaku, the most powerful supercomputer at the time of the experiment, an average of 9,000 years to produce a set of numbers from the same distribution.
The experiment is the latest in a series of demonstrations of so-called quantum advantage, where a quantum computer defeats a state-of-the-art supercomputer at a specified task. The experiment “pushes the boundaries of machines we can build,” says physicist Nicolas Quesada, a member of the Xanadu team who now works at Polytechnique Montréal.
“This is a great technological advance,” says Laura García-Álvarez of Chalmers University of Technology in Sweden, who was not involved in the experiment. “This device has performed a computation that is believed hard for classical computers. But it does not mean useful commercial quantum computing.”
So what, exactly, does Xanadu’s claim of quantum advantage mean? Caltech physicist John Preskill coined the concept in 2011 as “quantum supremacy,” which he has described as “the point where quantum computers can do things that classical computers can’t, regardless of whether those tasks are useful.” (Since then, many researchers in the field switched to calling it “quantum advantage,” to avoid echoes of “white supremacy.” Xanadu’s paper actually calls it “quantum computational advantage” because they think “quantum advantage” implies that the computer performed a useful task—which it didn’t.)
Preskill’s words suggested that achieving quantum advantage would be a turning point, marking the beginning of a new technological era in which physicists would begin devising useful tasks for quantum computers. Indeed, people anticipated the milestone so hotly that the first claim of a quantum computer outperforming a classical computer—by Google researchers in 2019—was leaked.
But as more researchers claim quantum advantage for their machines, the meaning of the achievement has become murkier. For one thing, quantum advantage doesn’t mark the end of a race between quantum and classical computers. It’s the beginning.
Each claim of quantum advantage has set off other researchers to develop faster classical algorithms to challenge that claim. In Google’s case, its researchers performed a random-number-generating experiment similar to Xanadu’s. They wrote that it would take a state-of-the-art supercomputer 10,000 years to generate a collection of numbers, while it took their quantum computer only 200 seconds. A month later, researchers at IBM argued that Google used the wrong classical algorithm for comparison, and that a supercomputer should take just 2.5 days. In 2021, a team using the Sunway TaihuLight supercomputer in China showed they could complete the task in 304 seconds—just a hair slower than Google’s quantum computer. That same year, the developers of that algorithm presented an even faster method. A larger supercomputer would be able to execute this algorithm in dozens of seconds, says physicist Pan Zhang of the Chinese Academy of Sciences, who helped develop both algorithms. That would put the classical computer on top again.
“If you say you've gotten quantum advantage, you’re saying that no one will ever simulate your experiment as accurately as your experiment was,” says physicist Jacob Bulmer of the University of Bristol. “It's a big scientific moment when you make that claim. And big claims require strong evidence.”
A 2020 quantum advantage claim from researchers at the University of Science and Technology in China met similar criticism. The team, led by physicist Pan Jian-Wei, also used their quantum computer to generate numbers according to a set probability distribution. In their paper, they claimed that their quantum computer could generate a set of numbers in 200 seconds, while the world’s most powerful supercomputer would take 2.5 billion years. In January, Bulmer led a team to show that it would actually take a supercomputer 73 days.
Researchers challenge quantum advantage claims with two main strategies. In one technique, they use a supercomputer to simulate the quantum computer itself in order to compare how quickly each can complete the desired task. In Xanadu’s case, the supercomputer simulates the light beams, the network of prisms, and the photon counting detectors to generate numbers. The faster computer wins. In the other technique, known as “spoofing,” researchers generate numbers by any means possible without simulating the quantum computer. The classical computer wins when its generated numbers follow the desired probability distribution more closely than its competitor’s numbers.
Every time a quantum computing team lays their hands on the trophy, their rivals try to yank it back. Because of this dynamic, announcements of quantum advantage have become less like triumphant declarations than invitations for public critique. In fact, Xanadu’s team tried to anticipate the critiques by having their own researchers challenge their claim before they published their paper. The claim stood up to their internal spoofing, yet in their paper they acknowledged that the quantum computer’s lead may not last. “We leave as an open question to the community whether better … algorithms for spoofing can be developed,” the Xanadu researchers wrote.
The back-and-forth pushes researchers to make better quantum computers, says physicist Jonathan Lavoie of Xanadu: “I think this kind of competition is very healthy.” But the experiments misrepresent the anticipated purpose of quantum computers. “People emphasize too much the competition between classical and quantum,” he continues.
Quantum computers are not meant to replace supercomputers; instead, experts want them to tackle specific tasks inaccessible to classical computers. For example, one near-term goal is to have quantum computers simulate complex molecules for drug discovery or battery design, which are resource-intensive tasks for supercomputers to perform accurately. Researchers might perform these simulations using a future supercomputer that would contain a quantum computing chip. The quantum chip would handle a specific part of the simulation, while the supercomputer does the rest.
A single quantum advantage claim demonstrates an incremental advance in the field. In particular, each claim indicates that “people are making progress in terms of scaling up the hardware,” says Alicia Welden, a researcher who develops quantum computing algorithms for startup QC Ware. Even if Xanadu’s claim doesn’t hold up, they’ve demonstrated the potential of designing quantum machines that encode information in photons, rather than superconductors, as Google’s quantum computer does. The experiment is a small step on the way to building a so-called “fault-tolerant” quantum computer, meaning one that is robust to errors and can run arbitrarily long algorithms. Existing machines, by contrast, cannot hold on to information for very long and have no way of correcting errors.
So if claims of quantum advantage can be rapidly leapfrogged, and the tasks themselves have no practical application, perhaps it’s time for more informative ways to evaluate progress. Physicists have already begun judging quantum computers based on their environmental footprint. In 2020, one team showed that a supercomputer used 50,000 times more energy than a quantum computer to perform a specific task. Another metric might be how well these tasks edge toward practical utility. Last month, a collaboration led by researchers at Caltech and Google claimed quantum advantage in performing a machine-learning task, where they studied a simplified model of a material.
These complicated discussions highlight the long road ahead to making a useful quantum computer. Governments and private investors have already promised billions of dollars to the field in anticipation of its challenges, chief of which is simply making the hardware work. Unlike classical computers, which store information as 1s and 0s, quantum computers store information in superpositions of 1s and 0s. This “quantum” information is extremely fragile. Reading the information alters it, so the quantum computer must be extremely precise and intentional to avoid accidentally destroying it. “It’s so hard, but that’s what is so beautiful about it,” says Quesada of the Xanadu team.
In fact, some researchers aren’t convinced that a fault-tolerant quantum computer is the ultimate goal. García-Alvarez, for example, is motivated to do quantum computing research because she believes the work could spawn or bolster other new technologies, such as improved measuring tools and sensors. “The development of the technology can give rise to other applications that maybe we don't foresee right now,” she says. It’s tough to devise a good metric to judge quantum computing when the future is so far away.
Update 6-12-2022 10:29 am ET: This article was updated to correct information about the algorithms developed by Pan Zhang's group.
