Google AI researchers reduce error rates in key step toward future of quantum computing

Open this photo in gallery:

Quantum computers promise to one day change the world by performing types of calculations that are mathematically out of reach of conventional digital systems. One of the challenges to making that future a reality is the fact that quantum systems can easily be disrupted by the slightest nudge from the outside world, making them highly error-prone.

Now researchers at Google AI, a division of the internet search giant, have demonstrated that they can correct errors in a quantum computer at a rate that improves as the power of the computer increases. The result could offer a path to the industry’s ultimate goal: a quantum computer that can be used for a wide range of commercially valuable applications.

“The only way to achieve this is by introducing quantum error correction,” said Hartmut Neven, who leads Google’s Quantum Artificial Intelligence Lab in Santa Barbara, Calif., at an online press briefing on Tuesday. “I would call this is a necessary rite of passage that any maturing quantum computing technology has to go through.”

Quantum computing star D-Wave faces cash crunch, investors say

The achievement comes as Google GOOGL-Q and other companies are working to develop quantum computers based on a range of underlying technologies. At this point it is not yet clear which approach may lead to the most commercially competitive platform.

In 2019, Google became the first to claim a quantum advantage by performing a specific numerical calculation faster than a digital system, using a chip based on superconducting circuits that are cooled to liquid helium temperature. Last year, Toronto-based Xanadu Quantum Technologies Inc. reached a comparable milestone with a room-temperature machine that is based on interacting waves of light.

But however a quantum system is designed, the importance of suppressing errors is a recurring theme in the field.

Open this photo in gallery:

Even ordinary computer systems can develop errors as they run through their operations. One basic way to combat this is to protect bits of information – the ones and zeros that are at the heart of any digital calculation – with three layers of redundancy. For example if a single bit is encoded three times instead of once then any error in the way the bit is recorded and stored by the computer’s hardware is outweighed by the other two versions if they were encoded correctly. In such a set-up, a computer looks at a trio of bits such as 011 and interprets it as a 1.

In a quantum computer, system bits are replaced with qubits whose quantum properties allow them to hold ambiguous values, each a mixture of one and zero, while a calculation is under way. Maintaining such a state is a fine balance that is constantly at risk from physical vibrations, radio waves and other forms of system noise, right down to the microscopic level. Quantum systems can be made more robust if the information contained in one qubit is spread across a group of qubits that are linked together.

In this scenario, the qubits that are linked together are referred to as physical qubits. The group they create acts as a single entity, called a logical qubit, in the operations of the computer.

The challenge for those seeking to build a fault-tolerant quantum computer – one that can keep errors in check – is finding a way to add qubits without adding even more instability to an already fragile system.

In their latest effort, the Google AI team developed a strategy for combining the qubits on a version of its superconducting chip. In the experiment, when each logical qubit consisted of 17 physical qubits, the error rate was measured at just over 3 per cent per computation cycle. When the number of physical qubits per logical qubot was increased to 49, the error rate decreased to about 2.9 per cent.

“The improvement that we see is quite small,” said Julian Kelly, a team member. “What we want to do is make that improvement, as we add more error correction, to be quite significant.”

In a research paper describing the work, published Wednesday in the journal Nature, the Google team suggested that as quantum computers scale up in size, the method they employed to control errors will allow those computers to perform reliably.

“It’s certainly an advance,” said Daniel Gottesman, a professor of theoretical computer science at the University of Maryland who was not involved in the work. “They managed to put everything together with good enough accuracy that they can actually see the hope for improvement.”

It remains to be seen if the hope is borne out as quantum machines get progressively larger. While the kinds of systems that researchers are working with today are at the level of dozens to hundreds of qubits, it is estimated that it will require machines with one million or more qubits performing billions of operations to realize the real potential of quantum computing.

Dr. Gottesman said that one milestone he is still waiting for is the development of an error-corrected qubit that is unequivocally better than a qubit that uses no error correction at all.

Another open question is whether a method of error correction that improves one type of quantum system will be just as good for another. For example, the approach employed by the Google team is known as “surface code” because it is well-suited to qubits that are connected on a two-dimensional plane, like superconducting circuits on a chip.

Surface code can also be used in a way that is applicable to the kind of light-based quantum computer developed by Xanadu among others. However, such systems can exploit other ways that its qubits interact in three-dimensional space to find approaches to error correction that lead to better performance than surface code.

“We’re very much focused on building an error-corrected quantum computer,” said Zachary Vernon, Xanadu’s chief technology officer for hardware, in an interview. “A major part of that focus for us is developing codes that leverage three-dimensional connectivity.”