Google and IBM's quantum milestones, and the long road to useful machines

Google's claim rests on an experiment run on a 53-qubit processor called Sycamore. The paper, obtained by Fortune and later published by Nature, says the chip sampled a randomly generated circuit one million times in about 200 seconds. Google estimates a state-of-the-art supercomputer would need roughly 10,000 years for the same task.

That gap is the whole point. Fortune obtained the paper before NASA briefly posted it. In it, Google researchers wrote that "quantum speedup is achievable in a real-world system and is not precluded by any hidden physical laws."

The result did not arrive out of nowhere. In April 2017, MIT Technology Review reported that John Martinis, the physicist leading Google's quantum group, had given himself until the end of that year to reach the same milestone. His team then had a six-qubit chip, arranged two by three. It tested a manufacturing method in which qubits and their control wiring are built on separate chips and then bump bonded together. Martinis said the approach was meant to eliminate extra control lines that can interfere with how qubits function. He also said he would need a grid of 49 qubits for the supremacy experiment, and that designs for 30 to 50 qubits were already in progress.

IBM gets to 50 qubits first, then questions the metric

Google was not alone. The Associated Press reported in November 2017 that IBM had built and measured a processor prototype with 50 qubits. Dario Gil, who led IBM's quantum computing and AI research division, called it the first time any company had built a quantum computer at that scale. Seth Lloyd, an MIT mechanical engineering professor not involved in the work, told the AP that IBM likely still had glitches to work out, but that the announcement was a sign of significant progress.

By March 2019, IBM was pushing a different yardstick. ZDNET reported that IBM's Q System One, a 20-qubit processor, had produced a Quantum Volume of 16. That was double the Quantum Volume of 8 recorded by its earlier IBM Q system. Quantum Volume accounts for qubit count, connectivity and coherence time, plus gate and measurement errors, device cross talk and compiler efficiency. IBM said the metric would need to double every year to reach Quantum Advantage within a decade, and that it had doubled the power of its quantum computers annually since 2017.

When Google's paper surfaced in September 2019, IBM's Gil was blunt. He told Fortune the experiment was a highly special case "laboratory experiment" with "no practical applications," and advised against using quantum supremacy as a progress metric. "Quantum computers will never reign 'supreme' over classical computers, but will rather work in concert with them, since each have their unique strengths," he said.

Intel Labs' director of quantum hardware, Jim Clarke, called Google's update "a notable mile marker." He said a commercially viable quantum computer would require many more research and development advances. "While development is still at mile one of this marathon, we strongly believe in the potential of this technology," he said.

Google's own account, published on its blog in October 2019, was careful about the limits. The company described the result as a "hello world" moment and said it would be many years before a broader set of real-world applications could be implemented. It framed the work as a 13-year effort. It began in 2006, when Google scientist Hartmut Neven started exploring quantum computing for machine learning, and continued in 2014, when Martinis and his University of California, Santa Barbara team joined.

The gap between a milestone and a product is not a technicality. Quantum processors would need to be far larger than 50 qubits to do useful work, and error correction remains unsolved at scale. Chris Monroe, a University of Maryland professor and cofounder of the startup IonQ, told MIT Technology Review in 2017 that the supremacy experiment would be "an academic milestone." His assessment of what came next: "Afterward you still have to figure out how to make it more scalable and programmable."