Making quantum computers reliable

CALCULATING machines that run on quantum bits (known as qubits, for short) are, by some accounts, the future of computation. Quantum computers have the theoretical advantage that they can solve with ease certain mathematical problems, such as the factorisation of large numbers, which are hard or impossible for classical machines. This is possible thanks to a qubit’s ability to remain, through the peculiarities of quantum mechanics, in many quantum states simultaneously. The more qubits a computer has, the more mind-bogglingly gigantic are the calculations it can handle. Finance, medicine, chemistry and artificial intelligence are thus all expected to be transformed by quantum computing.

And where the future is, there surely will Google be also. The firm sets great store by its quantum-computing project, which it calls Project Bristlecone. This is intended to develop a “quantum-supremacy device”, ie, one that is palpably and provably faster than a traditional computer of equivalent size at solving particular mathematical problems. Google is currently preparing such a device, which will have 49 qubits. It is rumoured that this will be ready for demonstration before the year is out. Though not confirming such rumours, John Martinis, Google’s quantum-computing boss, told the annual meeting of the American Association for the Advancement of Science, in Austin, Texas, about some of the problems involved in doing so.

Leadership in the quantum-computing race, Dr Martinis said, is typically measured in terms of the number of qubits that a machine can handle. Less attention is paid to those machines’ error rates. Since the people building such experimental machines are usually physicists, rather than engineers, they typically cite their best measurements when reporting qubit error rates, in order to show the machine’s capability. That number is of little interest to Dr Martinis and his team, though. They are thinking like engineers, attempting to build a robust, working device. In this case, Dr Martinis says, it is the worst error, not the slightest, that is important.

Project Bristlecone’s main concern is the quality of the qubits. The theoretical ability to beat a classical computer is of little use if the hardware that serves as the physical representations of those calculations is misfiring or dodgy.

At the moment, that is sometimes the case. Preparing and maintaining qubits is a delicate and fiddly process, akin to classical computing in the days before silicon. The slightest puff of interference from the outside world risks disrupting a qubit and scuppering a calculation. Every one of Google’s qubits is held in a chip that has 120 wires coming out of it, each of which is capable of introducing troublesome noise. Nor can quantum computers rely on the error-correction techniques that classical computers use. Those duplicate data outputs and check them against each other. Duplicating the output of qubits would mean having to measure them prematurely. That would change the qubits’ quantum states, wrecking the calculation. Instead, everything must be and remain perfect.

Such reliability has been mostly achieved in classical computing. Hardware problems are not unheard of, but they are relatively rare. (The software is another matter.) But that dependability is the result of 60 years of continuous improvement of solid-state silicon transistors. Quantum computing is now in the equivalent of the days of vacuum tubes running calculations in room-sized computers. And that was a world in which the tubes often blew, and bugs in the system were literal ones, namely insects that caused short circuits. Such behemoths were able to turn into today’s sleek machines because, at every stage of the journey, they were useful. And that, ultimately, is the standard quantum computers will have to match.