Between the mountainous and coastal vistas of Goleta, California, sits an unassuming office on the side of a building next to the freeway. It could belong to any Southern California company; workers sit in grey cubicles beneath fluorescent lights, and there’s a rack to hold employees’ bikes and surfboards. But at those desks are physicists and computer scientists developing a computer like none you’ve ever seen before. Behind a set of double doors, cylindrical machines hold computer chips at temperatures colder than the vacuum of space.
Here, Google’s scientists have been toiling to create a computer processor that can solve a problem that’s too hard for the world’s best supercomputers. Today, they announced they’d succeeded: Their Sycamore quantum computer was able to complete a problem in 200 seconds that a supercomputer would need 10,000- years to solve, according to their estimates.
It’s a single, contrived problem, and the chip would fail in a race with a supercomputer to add two and two together. But Google’s scientists think they’ve achieved a historical computing milestone.
“One criticism we’ve heard a lot is that we cooked up this contrived benchmark problem — [Sycamore] doesn’t do anything useful yet,” Hartmut Neven, a Google engineering director dressed in a puffy silver coat reminiscent of a space suit, told journalists at a press event today. “That’s why we like to compare it to a Sputnik moment. Sputnik didn’t do much either. All it did was circle Earth. Yet it was the start of the Space Age.”
Today, Google gave journalists a first look at the device and how it was able to complete the experiment.
While classical computers use transistors to represent data in zeroes and ones, quantum computers represent data using artificial atoms, called qubits. Rather than simply using the rules of logic, these qubits interact via the weird mathematics of quantum mechanics. They take on zero or one and produce long strings of binary code just like classical computers do, but during the calculation, they can take on states between zero and one, which determine how likely you are to get zero or one on the final measurement.
Each qubit is made from a tiny, plus sign-shaped loop of superconducting wire. Not only does current travel without resistance through these systems, but it’s almost as if the whole unit acts like a single electron. Each plus sign touches four other plus signs in a lattice shape.
The chip (which looks much like any common processing chip to the untrained observer) sits in a casing at the bottom of a structure shaped like an upside-down wedding cake, held in a vacuum chamber. The environment is progressively colder with each tier until it’s at the 15-milliKelvin operating temperature. A mess of wires sends tiny microwave pulses to the qubit, causing it to take on excited states that are measured by another tiny component attached to the plus sign.
Google’s scientists first designed the quantum supremacy experiment back in 2016. The premise: Set up a random circuit with these quantum gates. Re-measure the same circuit thousands to millions of times, and certain strings of zeroes and ones will become more likely than others, via an effect called quantum interference. Make a supercomputer simulate the quantum computer, and tell it to try to create a similar probability distribution of these strings.
With each additional qubit (and with each additional operation), it becomes much harder for the supercomputer to keep up. Google’s scientists felt comfortable that, running 53 of Sycamore’s 54 qubits (one wasn’t working), they’d soundly beaten the supercomputer. Confirming that the answer is correct is a matter of slightly decreasing the complexity of the circuit, running it in a way that the supercomputer can check, and then extrapolating.
With the help of University of Texas physicist Scott Aaronson, they even devised a use for this quantum supremacy experiment. It outputs random bits, and randomness is important in areas like cryptography and the lottery. But what if it’s not really random — what if someone can secretly guess the supposedly random number? With this experiment, Google can verify for you that a regular computer could not have devised these random numbers.
Despite the technological accomplishment, the computer is prone to errors. Any interaction with the outside world can cause the qubit to spit out the wrong values. But the experiment demonstrated that as they add more qubits, the number of errors will increase in a predictable way. The layout, specifically the lattice of plus signs, is meant to be compatible with a future when they can anticipate and work around these problems.
“We’ve shown that we have an understanding of these errors,” said Google scientist Marissa Giustina. “That’s a key engineering and physics piece of the breakthrough.” (FYI, Giustina was the only woman scientist in the room).
I got to program the computer, too. Similar to IBM’s Q experience, you use a regular computer interface to drag pulse-generating, qubit value-altering operations onto each qubit, like music notes on the staff. Oscilloscopes showed the shape of the pulses that I was sending to the qubits. I watched as the probability of each final qubit delivering a zero or a one changed with each additional operation.
Plenty of scientists have already levied critiques that classical computers actually can run the supremacy experiment in less time or that the right classical algorithm just hasn’t been found yet. Neven responded to IBM’s claim that it would take a classical supercomputer 2.5 days, rather than 10,000 years, to run the supremacy experiment:
“Ever since we published the suggestion of quantum supremacy, there was a steady stream of improvements on the classical side that have become a benchmark of classical supercomputers,” he said. He explained that researchers at NASA, Oak Ridge National Lab, and elsewhere are working on improving classical computing algorithms so that the Google device has state-of-the-art supercomputers to compete against.
On the scientific front, Google has demonstrated a large, convoluted quantum system, far more complex than has been shown before. And on the computing front, we’ve entered uncharted territory: Quantum computers are now devices that, maybe, can do something a classical computer can’t.
Said Giustina: “We’ve reached a space in computation that’s new, that no other tool can reach.”