A report released Dec. 4 by the National Academies of Sciences, Engineering, and Medicine throws some cold water on the hype smoldering around quantum computing. The report finds no commercially viable applications for near-term quantum computers that cannot already be tackled with conventional computers. But those developing algorithms for these computers are more optimistic about the usefulness of the devices, particularly for chemical simulation.
Unlike conventional computers, which represent information digitally as 1s and 0s in electronic bits, quantum computers use qubits. During quantum calculations, qubits can hold more information than a simple 1 or 0, existing in a combination of states called a superposition. Because of these superpositions, quantum computers should be able to solve certain problems exponentially faster than classical supercomputers and perform calculations that cannot be performed by conventional devices. With an ideal quantum computer, chemists could model systems currently unmodellable on regular supercomputers, accurately simulating complex chemical bonds and predicting the structures of new drugs, semiconductors, and efficient catalysts.
Physicists and computer scientists are sure that, once high performance quantum computers can be built, the devices will be able to tackle these complex simulations. What’s less certain is what intermediate quantum computers currently in development will be capable of in the next few years. Start-ups as well as large companies like IBM, Intel, and Google are building quantum computers with tens of qubits. Researchers who are developing these noisy intermediate-scale quantum (NISQ) systems are optimistic that they will be able to do something useful. Google has been particularly bullish, says Peter Love, a quantum information scientist at Tufts University. The company pledged to demonstrate that their systems can do something classical computers cannot in 2017, and then in 2018, but has yet to prove it, Love says.
The report committee, made up of academic and corporate researchers working in the field, studied the current state of quantum computing hardware and software. The group concluded that we’re at least a decade away from powerful quantum computers that will definitively run laps around conventional supercomputers. The report highlights several key challenges. One is error correction.
Qubits are sensitive devices, and prone to error. To get the right answer from a quantum computer, researchers either have to repeat a calculation an unreasonable number of times, or build a quantum computer with millions of qubits. And the challenge in scaling up is not in the fabrication—after all, the semiconductor industry can pack billions of devices in small areas—but in the operation. Even tens of qubits nestled together in one array start interfering with one another, causing errors. This error correction problem, along with other issues, led the report’s committee to conclude that “there is no publicly known application of commercial interest based upon quantum algorithms that could be run on a near-term analog or digital NISQ computer that would provide an advantage over classical approaches.”
Alán Aspuru-Guzik, a quantum chemist at the University of Toronto who served on the committee, says he is personally more optimistic. Indeed, he’s the founder of Zapata Computing, a startup company with a commercial interest in developing algorithms that help classical and NISQ computers work together on chemical simulations.
This approach to simulation uses a NISQ’s qubits to simulate the wave functions of a molecule, and a classical computer to tune them again and again until the simulation is accurate. But today it takes an impractical number of such iterations to get the quantum computer spitting out accurate calculations—and that’s why classical supercomputers still beat the NISQ systems. Error correction is too onerous. “If each step takes one second and you have 1019 of them, there’s not enough time,” in existence, Love says.
By chopping up the algorithms in different ways, quantum programmers hope to reduce the number of steps. Meanwhile, NISQ computer hardware is getting better. With more and better qubits, fewer such repetitions will be necessary. Zapata Computing and others working on these blended algorithms have only recently had access to the NISQ computers needed to test them experimentally. That’s why it’s so exciting to work in quantum computing right now, Love says. Even five years ago, without experimental systems to test their algorithms, this work was mostly theoretical, and it was hard to get motivated, he says. Over the next five years, Love plans to test his algorithms on a 32-qubit NISQ computer, then a 64-qubit one.
But on top of the technical challenges, a possible economic problem could undercut the development of quantum computing in the US, according to the National Academies report. Companies building NISQ systems are hoping to make money by letting academic and industry users access them over the cloud. But if these early, error-prone quantum computers do not gain footing in the market over the next few years, private companies will not be able to sustain R&D in quantum computing, and “government investment will be essential,” said Mark Horowitz, an electrical engineer at Stanford University and chair of the committee that authored the report. Horowitz spoke at a webinar hosted by the National Academies on Dec. 4.
Aspuru-Guzik says this economic message is the most important one in the report. If commercial funding lags, and the government doesn’t keep pace with efforts in China, Europe, and elsewhere, the US will miss out on future technological and economic benefits of quantum computing.
Love says he’s happy the National Academies are “throwing some cold water into the discussion.” Excited as he is about the possibilities, Love says it’s possible that NISQ systems will indeed never do anything useful in themselves, and will merely serve as a stepping stone along the way to future universal quantum computers. Right now, researchers are exploring what’s possible with NISQ computers, and he expects they’ll know in five years what works and what doesn’t.