In a new study, two quantum computers fashioned from dramatically different technologies have competed head-to-head in an algorithm-crunching exercise. One computer was more reliable, and the other was faster. But what’s most important, some scientists say, is that for the first time, two different quantum computers have been compared and tested on the same playing field.
“For a long time, the devices were so immature that you couldn’t really put two five-qubit gadgets next to each other and perform this kind of comparison,” says Simon Benjamin, a physicist at the University of Oxford in the United Kingdom, who was not involved in the study. “It’s a sign that this technology is maturing.”
One of the computers is built around five ytterbium ions held in an electromagnetic trap and manipulated by lasers. It belongs to a lab led by Chris Monroe, a physicist at the University of Maryland in College Park, and co-founder of the startup company ionQ. The other computer belongs to IBM. At its heart are five small loops of superconducting metal that can be manipulated by microwave signals. It is also the world’s only quantum computer that can be programmed online by users, rather than exclusively by scientists in the lab—a fact that allowed Monroe’s team to design the experiment.
Neither device has much computing power, but they demonstrate the principle that many think will eventually make quantum computers a major technology. Unlike conventional computers’ bits, which can be in states of only 0 or 1, quantum computers rely on quantum bits, or qubits, that can be teased into combinations, or “superpositions,” of both 0 and 1. In Monroe’s computer, each qubit is an ion in which an electron can be placed at one energy level to signify 0, another to signify 1, or both levels at once. In each of IBM’s superconducting circuits, electric current can circulate with one of two different strengths, or at both levels simultaneously. It’s also possible to join the superposition states of many qubits. This gives a quantum computer a potential calculating power that grows exponentially with every added bit.
But the states of qubits are also fragile: Small perturbations from the outside world can easily collapse the superpositions to just a 0 or a 1. So computers must carefully maintain superposition states as a computation proceeds. In the test, the two computers both had two-qubit “gate fidelities,” or probabilities of successfully completing a single two-qubit logical operation, of about 97%—considerably below what will ultimately be needed for any real world operation.
To test their performance, Monroe’s team ran a set of standard algorithms on each device, and compared the output. The ion computer got the right answer more often in each case. For one particular exercise, the contrast was especially dramatic: The ion computer achieved a 77.1% success rate, whereas the superconducting computer succeeded only 35.1% of the time. The scientists published their results last week on arXiv.
The performance difference arises not from the qubits themselves, but from how they are wired together, Monroe says. Each of his ions can interact with every other ion, reducing the number of operations needed for many tasks—and the chances that a superposition will collapse. In the IBM computer, by contrast, four of the superconducting loops were connected only to one central one, often necessitating additional operations to swap information among the loops. Because no operation is 100% reliable, the overall success rate declines as the number of operations grows. “Connectivity matters,” Monroe says.
Jerry Chow, a physicist who leads IBM’s quantum computing team at the company’s Yorktown Heights, New York, lab, acknowledges that connectivity is important. But ultimately, he expects that qubit superposition states will last longer and be more “coherent”—which would mean that his computer’s lower connectivity won’t necessarily drag down its overall reliability in the long run. “If you have enough coherence, it doesn’t matter how long the whole operation of your algorithm might take.” He also notes that IBM’s online computer now features more qubit connections than it did when Monroe’s team ran its test, which would likely bring it closer to equaling the ion computer’s performance. And both labs are already working on more reliable next-generation devices with more qubits.
Indeed, the study compares “the embryonic form” of the two “front-running” approaches to quantum computing, Benjamin says. A practical device made of either ions or superconducting loops will need thousands of qubits, and the web of interconnections between them will grow far more complex. He also notes that whereas the ion computer is more reliable at the moment, the superconducting computer is faster. IBM’s device completes a two-qubit operation in 250 to 450 nanoseconds, up to 1000 times faster than the ion computer.
The study also provides food for thought for quantum software designers like Krysta Svore, a Microsoft researcher in Redmond, Washington. Understanding how quantum computers’ specific architectures affect performance will be critical for optimizing future algorithms, she says. “It’s a great step to start that conversation.”