Quantum supremacy: what Google's 2019 milestone claimed and why IBM pushed back

Google's claim rests on one number: 200 seconds against an estimated 10,000 years. The paper was posted to NASA.gov and reported by Fortune on 20 September 2019. Google's researchers wrote that their processor takes about 200 seconds to sample one instance of the quantum circuit a million times. A state-of-the-art supercomputer, they estimated, would need approximately 10,000 years for the equivalent task. That comparison is the whole basis of the term quantum supremacy: a quantum machine solving a problem a classical computer cannot solve in any practical amount of time.

The machine was called Sycamore and it carried 53 qubits. Google had previously designed a 72-qubit device named Bristlecone and scaled back from it. The researchers also estimated that running the same experiment on a Google Cloud server would take 50 trillion hours. On the quantum processor it took 30 seconds.

At first, Google confirmed nothing. A spokesperson declined to confirm the authenticity of the paper and its results, and NASA did not respond immediately to a request for comment. A source at Google familiar with the situation suggested NASA had accidentally published the paper early, before the team's claims could be thoroughly vetted through scientific peer review, a process that could take anywhere from weeks to months. The Financial Times first reported the news.

Two years of build-up

Google had been signalling this target since at least 2017. In April that year, MIT Technology Review reported that John Martinis, who led the Google research group, had given himself just a few months to reach a milestone in the history of computing. Martinis joined Google in late 2014 from the University of California, Santa Barbara, where he remains a professor. His roughly 25-strong group had built a six-qubit chip arranged in a two-by-three configuration. That chip tested a manufacturing method in which qubits and the conventional wiring controlling them are made on separate chips and later bump bonded together. The approach is meant to eliminate extra control lines that can interfere with qubit function.

"That process is all working," Martinis told MIT Technology Review at the time. "Now we're ready to kind of move fast." He said designs for devices with 30 to 50 qubits were already in progress, and that he would need a grid of 49 qubits for his quantum supremacy experiment. By December 2017, speaking at the Q2B conference at NASA Ames, he was warning against reading progress as a qubit count. "Press releases always talk about quantum space race in number of qubits," Gizmodo quoted him saying. "It's more than just quantity, it's qubit quality." Gizmodo reported that Google was then fabricating its 49 or 50 qubit device and would begin testing within two weeks.

Other companies were moving on the same ground. IBM announced in November 2017 that its scientists had built and measured a processor prototype with 50 qubits. Dario Gil, who led IBM's quantum computing and artificial intelligence research division, said it was the first time any company had built a quantum computer at this scale. Seth Lloyd, an MIT mechanical engineering professor not involved in IBM's research, said IBM likely still had glitches to work out but that the announcement was a sign of significant progress.

When Google's paper surfaced, IBM's Dario Gil, by then head of IBM Research, argued against using quantum supremacy as a metric at all. "The experiment and the 'supremacy' term will be misunderstood by nearly all," he told Fortune. He described the work as a highly special case laboratory experiment with no practical applications. "Quantum computers will never reign 'supreme' over classical computers, but will rather work in concert with them, since each have their unique strengths."

Confirmed in Nature, and what it did not mean

On 23 October 2019, Google's blog confirmed that Nature had published the result in its 150th anniversary issue. The blog described the test computation as taking 200 seconds and repeated that the best known algorithms on the most powerful supercomputers would need thousands of years. It framed the result as the "hello world" moment the field had been waiting for. It also stated plainly that there is a long way to go between lab experiments and practical applications, and that it will be many years before a broader set of real-world applications can be implemented.

Google's own researchers had made a similar point in the paper. "We are only one creative algorithm away from valuable near-term applications," they wrote. The blog noted a 13-year effort: Hartmut Neven started exploring quantum computing for machine learning in 2006, Martinis and his Santa Barbara team joined in 2014, and Sergio Boixo published a paper two years later that focused the effort on the well-defined computational task of quantum supremacy.

Outside researchers were already splitting the milestone from the marketing. Chris Monroe, a University of Maryland professor and cofounder of the quantum computing startup IonQ, told MIT Technology Review in 2017 what achieving the experiment would amount to: "It'll be an academic milestone. Afterward you still have to figure out how to make it more scalable and programmable." Jim Clarke, Intel Labs' director of quantum hardware, called Google's 2019 update "a notable mile marker" and said a commercially viable quantum computer would require many R&D advancements first. "While development is still at mile one of this marathon, we strongly believe in the potential of this technology," he said.

The qualifier that tends to get lost is scale. Quantum processors would need to be much larger than 50 qubits to be capable of useful work, MIT Technology Review noted. The supremacy experiment was designed to be hard for a classical machine precisely because it is not a useful industrial problem. It sampled randomly generated numbers through a specialised scenario involving quantum phenomena, calculating the output of certain specialised circuits, rather than solving anything a business needs solved.

That is why IBM kept pushing a different yardstick. By March 2019, ZDNET reported, IBM was outlining its highest Quantum Volume to date, a performance metric combining qubit count, connectivity, coherence time, gate and measurement errors, device cross talk and circuit software compiler efficiency. IBM said its Q System One, with a 20-qubit processor, produced a Quantum Volume of 16, double the current IBM Q at 8. The company said Quantum Volume would need to double every year to reach Quantum Advantage, the point where quantum applications deliver significant advantages over classical computers, within the next decade. IBM had doubled the power of its quantum computers annually since 2017.

Google's framing of the moment was grander. The blog compared it to building the first rocket that left Earth's gravity to touch the edge of space, and to the questions that followed about why go there at all. "It showed us what was possible and nudged the seemingly impossible into frame," it said. The paper's authors went further, predicting quantum computing power would grow at a double exponential rate, beating even the exponential rate that defined Moore's Law.

The peer review caveat never fully disappeared from the story, but Nature's publication settled the question of whether the result would be formally vetted. What it did not settle is the argument over the word. Google's own blog called quantum supremacy a term of art with a narrow meaning. IBM's objection was that the narrow meaning would be read as a broad one, and that reading has followed the field ever since.