Scientists are close to building a quantum computer that can beat a conventional one
It's a Sunday afternoon in September, and the two co-founders of ionQ, a quantum computing startup, are meeting for a strategy session with their first hire: their new CEO. Sitting in comfy leather chairs in the Physical Sciences Complex at the University of Maryland (UMD) in College Park, the two founders are experiencing a touch of culture clash. Lifelong research scientists, UMD physicist Chris Monroe and Jungsang Kim, an electrical engineer at Duke University in Durham, North Carolina, are relaxed and talkative about their company's plans, even in the presence of a reporter. They tick off reasons why trapped ions, their specialty, will make for a great quantum computer—perfect reproducibility, long lifetimes, and good controllability with lasers.
Their CEO, David Moehring, whom Monroe and Kim have just hired away from the U.S. Intelligence Advanced Research Projects Activity, is more guarded. He warns Monroe and Kim against divulging information that he thinks a startup should keep secret—including exactly how much money they received from the venture capital firm New Enterprise Associates. (He is willing to confirm that it's several million dollars.) Kim nods at Moehring and chuckles. "At some point this guy will implement a policy that we need to get his approval to talk."
These unlikely partners share a common conviction: that quantum computing—which aims to harness quantum mechanics to vastly accelerate computation—is ready for prime time. They are not alone. Tech giants Intel, Microsoft, IBM, and Google are all plowing tens of millions of dollars into quantum computing. Yet the contenders are betting on different technological horses: No one yet knows what type of quantum logic bit, or qubit, will power a practical quantum computer.
...there's a market, there's a hunger for these devices.Google, often considered the field's leader, has signaled its choice: tiny, superconducting circuits. Its group has built a nine-qubit machine and hopes to scale up to 49 within a year—an important threshold. At about 50 qubits, many say a quantum computer could achieve "quantum supremacy," a term coined by John Preskill, a physicist at the California Institute of Technology in Pasadena, to denote a quantum computer that can do something beyond the ken of a classical computer, such as simulate molecular structures in chemistry and materials science, or tackle certain problems in cryptography or machine learning.
IonQ's team isn't ruffled by Google's success. "I'm not worried that Google will declare next month that the game's over," Kim says. "Or maybe they can declare it, but it won't be over." Still, ionQ, which lacks a building or even a website, is a decided underdog. The startup is sticking with trapped ions—the technology behind the world's very first quantum logic gates, which Monroe himself helped create in 1995. With precisely tuned laser pulses, Monroe can knock ions into quantum states that last for seconds—far longer than Google's qubits. Kim has developed a modular scheme for connecting groups of ions together, which might allow ionQ to scale up faster than many expect. But so far, its leaders have joined just five qubits into a programmable machine. Trapped ions are "a bit of a black sheep right now," Monroe admits, "but I think in the coming years people will be flocking to it."
One thing is certain: Building a quantum computer has gone from a far-off dream of a few university scientists to an immediate goal for some of the world's biggest companies. And Monroe and his colleagues are among many who hope to cash in. Although superconducting qubits may have taken a momentary lead among industry players, experts agree that it's far too early to declare a winner. "It's a good thing that these different technologies are being developed in parallel," says Preskill, an unofficial dean of quantum information science. "Because there could be surprises that really change the game."
A bit of the action
In the race to build a quantum computer, companies are pursuing many types of quantum bits, or qubits, each with its own strengths and weaknesses.
(GRAPHIC) C. Bickel/Science;(DATA) Gabriel Popkin
This massive parallelism would not help with many tasks—nobody thinks quantum computers will revolutionize word processing or email. But it could dramatically speed up algorithms designed to explore vast numbers of different paths simultaneously, and solve problems that include searching through large data sets, discovering new chemical catalysts, and factoring large numbers used to encrypt data. Quantum computers may even find a role simulating black holes and other phenomena in physics.
There is a major catch, however. Quantum superpositions and entangled states are exquisitely fragile. They can be destroyed by slight perturbations from the environment—or by attempts to measure them. A quantum computer needs protection from what Robert Schoelkopf, a physicist at Yale University, calls "a sea of classical chaos."
Though theoretical ideas started appearing in the early 1980s, experimental quantum computing got going only in 1995, after Peter Shor, a mathematician at Bell Labs in Murray Hill, New Jersey, showed that a quantum computer could quickly factor large numbers—a capability that would render much of modern cryptography obsolete. Shor and others also showed that it was theoretically possible to keep fragile qubits stable indefinitely by using neighboring qubits to correct their errors.
Suddenly, physicists and their funders had both a concrete reason to build a quantum computer and a sign that the machine wouldn't dissolve into a pile of cascading errors. David Wineland, a Nobel Prize-winning physicist at a National Institute of Standards and Technology (NIST) laboratory in Boulder, Colorado, had already pioneered methods to use lasers to cool ions and control their internal quantum states. Within a year of Shor's discoveries, Wineland and Monroe, a NIST staff scientist at the time, built the first quantum mechanical logic gate, using lasers to manipulate electron states in a beryllium ion. Because of Wineland's experience with ions, the chance to seize the lead in early quantum computing experiments "fell in our laps," Monroe says.
We like to joke that once we have a quantum computer, we're going to use it to design the next quantum computer.As millions of government research dollars began flowing to quantum physics groups around the world, other kinds of qubits began to appear. By the early 2010s, trapped ions faced a strong challenge from a new darling: circuit loops made out of superconductors—metallic materials that can carry an oscillating electric current without resistance when chilled nearly to absolute zero. The 0 and 1 of the qubit correspond to different current strengths. Adding to their appeal, the loops can be seen with the naked eye, controlled with simple microwave electronics rather than finicky lasers, and fabricated using techniques from conventional computer chip manufacturing. They also operate very quickly.
At least at first, however, superconductors had a fatal weakness: Environmental noise, even from the electronics used to control them, can disrupt their quantum superpositions in a small fraction of a microsecond. But engineering refinements have improved the circuits' stability by more than a million times, so that they now can remain in a superposition state for tens of microseconds—though they still collapse far faster than ions.
In 2007, D-Wave Systems, a startup company in Burnaby, Canada, surprised just about everybody by announcing that it had built a quantum computer, with 16 superconducting qubits. D-Wave's machine didn't entangle all the qubits, and it couldn't be programmed qubit by qubit. Instead, it relied on a technique called quantum annealing, in which qubits are entangled only with near neighbors and interact to produce not a set of parallel computations, but a single overall quantum state. D-Wave developers hoped to map complicated mathematical problems onto such states and use quantum effects to find minimum points, a promising technique for solving optimization problems such as efficiently routing air traffic.
Almost instantly critics cried foul: D-Wave did not even attempt to do certain things that many thought essential to quantum computing, such as error correction. But several companies, including Google and Lockheed Martin, bought and tested D-Wave devices. A tentative consensus emerged: They did something quantum, and, for certain specialized tasks, they might perform faster than a conventional computer. Quantum or not, D-Wave jolted the private sector awake. "It was really eye-opening," Monroe says. "[D-Wave] showed that there's a market, there's a hunger for these devices." Within a few years, companies started lining up behind technologies that aligned with their in-house expertise.
Intel made one of the biggest bets, announcing in 2015 that it would invest $50 million into research at QuTech, an offshoot of Delft University of Technology in the Netherlands. The company is focusing on silicon quantum dots, often called "artificial atoms." A quantum dot qubit is a small chunk of material in which, as in an atom, the quantum states of an electron can represent 0 and 1. Unlike ions or atoms, however, a quantum dot doesn't need lasers to trap it.
Early quantum dots were made from nearperfect crystals of gallium arsenide, but researchers have turned to silicon, hoping to leverage the massive manufacturing infrastructure of the semiconductor industry. "I think [Intel's] heart is with silicon," says Leo Kouwenhoven, scientific director of QuTech. "That's what they're good at." But silicon-based qubits are well behind those based on ions or superconductors, with the first two-qubit logic gate reported only last year by a group at the University of New South Wales in Sydney, Australia.
Microsoft went for what many consider an even longer shot: topological qubits based on nonabelian anyons. These aren't objects at all—they're quasiparticles, traveling along the boundary between two different materials—and their quantum states are encoded in the different braiding paths they follow in time. Because the shapes of the braided paths lead to the qubit superpositions, they would be "topologically protected" from collapse, similar to how a shoelace stays tied even if nudged or bumped.
This meant that theoretically, a topological quantum computer wouldn't need to devote so many qubits to error correction. As early as 2005, a Microsoft-led team proposed a way to build a topologically protected qubit in hybrid semiconductor-superconductor structures, and Microsoft has funded several groups to try to make one. Recent papers from these groups and from a separate effort at Bell Labs have shown hints of the crucial anyon in the patterns of electrical currents that flow in their specialized circuitry, and the scientists are close to demonstrating an actual qubit, Preskill says. "I think in a year or two, we can consider it to be nailed: Topological qubits exist."
Google, for its part, recruited John Martinis, a superconducting qubit expert at the University of California, Santa Barbara (UCSB), who had studied D-Wave's operation and shortcomings. In 2014, the company swallowed his UCSB research team whole, hiring about a dozen people. Soon afterward, Martinis's team announced they had built a nine-qubit machine at UCSB, one of the largest programmable quantum computers so far, and they are now trying to scale up. To avoid creating an unwieldy jumble of wires, they are rebuilding the system into a 2D array that will sit on top of a wafer with control wires etched into it.
In July, Martinis's team—now up to about 30 scientists and engineers—used three superconducting qubits to simulate the ground state energy of a hydrogen molecule, demonstrating that quantum computers can simulate simple quantum systems as well as classical computers. The result points to the coming power of a machine with quantum supremacy, he says. Martinis calls the 1-year timetable for reaching a 49-qubit computer a "stretch goal," but he believes it may be possible.
Meanwhile, Monroe is grappling with the challenges that come with trapped ions. As qubits, they can remain stable for seconds, thanks to vacuum chambers and electrodes that stabilize them even in the presence of external noise. Yet that isolation also means it is a challenge to get the qubits to interact. Monroe recently entangled 22 ytterbium ions in a linear chain, but so far he is not able to control or query all ion pairs in the chain, as a quantum computer will require.
The complexity of controlling the ensemble rises with the number of ions squared, so adding many more is impractical. The way forward, Monroe believes, is to go modular and use fiber optics to link traps holding perhaps 20 ions each. In such a scheme, certain qubits within each module would act as hubs, reading out information from the rest of the qubits and sharing it with other modules; this way, most qubits could remain shielded from external interference.
On a recent afternoon, Monroe toured his six lab spaces at UMD. In his three older labs, electrical wires and vacuum lines descend in tangles from above. On oversize tables, a bewildering array of lenses and mirrors shape and direct laser light toward portals in small steel vacuum chambers containing the all-important ions. Overhead heating, ventilation, and air conditioning (HVAC) equipment—necessary to keep dust down and stabilize the lab temperature—gives off a steady drone. "I'm passionate about HVAC," Monroe says.
The three newer labs are, by contrast, tidy and eerily empty. Instead of Rube Goldberg optics tables, most of the lasers are integrated, plug-and-play units from companies like Honeywell—prototypes for the kinds of turnkey systems that ionQ needs to perfect if it is going to succeed. "The lasers we use now have only one knob, and it's 'on,'" Monroe says. He is antsy to get ionQ's labs up and running, so he can transition his highly paid research scientists onto ionQ's payroll and set them to perfecting technologies they've developed at UMD—which, thanks to an unusual agreement, ionQ can license exclusively and royalty-free. Next year he will take his first-ever sabbatical to focus on building ionQ. The private sector money flowing into quantum research, he says, "is the biggest deal in my career."
Even as money has poured in, quantum computing is a long way from becoming a secretive commercial field. The major research groups—even those affiliated with big companies—are still publishing results and presenting at conferences. They say they have a mutual interest in publicizing their advances, not least so that potential customers can think about how they could use a quantum computer. "We all need a market," Monroe says.
What's more, nobody knows enough about quantum computing yet to go it alone with a single qubit type. Every approach needs refining before quantum computers can be scaled up. Superconductor- and silicon-based qubits need to be manufactured with more consistency, and the refrigerators that chill them need streamlining. Trapped ions need faster logic gates and more compact lasers and optics. Topological qubits still need to be invented.
A future quantum computer could well be a hybrid, with ultrafast superconducting qubits running algorithms, then dumping output to more stable ion memory, while photons shuttle information among different parts of the machine or between nodes of a quantum internet. "One can imagine we'll have an environment in which several types of qubits exist and play different roles," says Krysta Svore, a Microsoft researcher in Redmond, Washington.
A quantum computer is so new, and so strange, that even the world's top quantum physicists and computer engineers do not know what a commercial one will ultimately look like. Physicists will need to simply build the most complex computer possible with current technology, then confront the new challenges that are sure to crop up, Svore says. Build, study, and repeat. "We like to joke that once we have a quantum computer, we're going to use it to design the next quantum computer."
No comments:
Post a Comment