What Microsoft’s error-correction milestone means for usable quantum computing

Microsoft, in collaboration with quantum-computing company Quantinuum, announced that they have reached a new high in quantum error correction, which finally moves the industry into a new phase of quantum computing.

Existing quantum computers all fall into the first stage of quantum computing: the foundational level, also known as “noisy intermediate scale quantum,” or NISQ, says Krysta Svore, vice president of advanced quantum development at Redmond, Wash.-based Microsoft Corp. “The significance of this announcement is that we’re achieving level two: resilient quantum computing,” she says.

That means that commercially viable quantum computing, which requires at least 1,000 reliable, logical qubits, is now years away, instead of decades, Svore says. And scientifically useful quantum computing, which only requires 100 qubits, is even closer.

Hardware vs. software quantum solutions

There are two main challenges standing in the way of usable quantum computing: the number of quantum bits – or qubits – that a single computer has, and how reliable those qubits are. This is very roughly similar to how traditional computer manufacturers are always trying to put more gates on a single chip and having them work reliably.

The twin problems of needing more qubits, and needing more reliable qubits, are actually related. Most companies working on quantum computers address the reliability problem by using multiple physical qubits to do the same calculation. In effect, they can turn two physical qubits – or a thousand physical qubits – into one logical qubit. That’s not a typo. Qubits are so error-prone that it could take a thousand, or even more, to have one reliable logical qubit. Since each additional qubit is so hard to build, this high error rate dramatically impacts progress.

Quantum computing companies are addressing the problem on both the physical level and in software. On the hardware side, they try to keep errors from occurring the first place by reducing temperature fluctuations and vibrations, or by designing qubits that are more stable from the start.

Another hardware-based approach is to build in physical error correction. Earlier this year, three vendors announced breakthroughs on that front. For example, instead of having backup qubits, they would use backup photons – specifically, microwave-range photos bouncing around inside hollow mirrored spheres, or bouncing around inside circuits, all linked on a quantum level to the qubit they’re providing backup for.

Microsoft partner Quantinuum says it has lower physical error rates than other computers, as well as the smallest-known fault-tolerant circuit. Last September, the company announced a breakthrough in performing math on a fault-tolerant system with three logical qubits.

In a separate announcement on March 5, the company announced that it has also solved the wiring problem. In quantum computers, each qubit requires a large number of control signals, making it progressively more difficult to add additional qubits to a system. Quantinuum said it is able to reduce the wiring to one digital input per qubit, plus a fixed number of analog signals, thus finally putting scalability into reach.

Microsoft’s breakthrough relied on this hardware advantage, says Svore. In addition to the new error-correction algorithms, the new high in reliability was achieved thanks to “collaboration between the hardware and software implementations,” she says, “and by using physical qubits that have high fidelities and connectivity.”

As a result of the collaboration, the total number of physical qubits required to create one logical qubit was reduced by a factor of up to 800.

They call their software error-correction approach the “Carbon code.”

They’re not the first companies to do software-based error correction for quantum computers. Several companies, including IBM and Alice & Bob, use low density parity check code, which has been around since the 1990s, mostly to help improve communications.

Carbon code is different, says Svore. “We do not consider the Carbon code to be an LDPC code,” she says.

Technically, Carbon code is a stabilizer code of the Calderbank-Shor-Steane variety, which is a quantum error-correction code. “It encodes two logical qubits among 12 physical qubits,” she says.

In addition, the announcement demonstrates a full error-correction cycle, not just parts of the cycle, as was the case in previous experiments, says Sam Lucero, chief quantum analyst at Omdia. “Fault-tolerant quantum computers are not just a theoretical possibility but have a strong chance to be realized in the real world,” he says.

Putting quantum advancement in perspective

Some quantum-computing experts doubt that the breakthrough is as big as Microsoft and Quantinuum are painting it to be. “Microsoft itself says it needs to improve the fidelity by at least three orders of magnitude,” says Omdia’s Lucero.

And the experiment only showed Clifford gates, he says. Clifford gates only support some types of computations. This means that the logical qubits Microsoft demonstrated aren’t enough for a full universal computer, he says. “Non-Clifford gate functionality will have to be added at some point.”

And four logical qubits is a ways away from the 100 needed for scientific value, he adds.

On a positive note, this means that current encryption methods are still safe. “It appears that about 2,000 logical qubits will be needed to run Shor’s algorithm sufficiently to break AES 256-bit encryption,” Lucero says.

Microsoft may have also cherry-picked results to get the attention-grabbing headline, says David Shaw, chief analyst at Global Quantum Intelligence. “They allow themselves to discard runs that didn’t work,” he says. “So you kind of have to squint to see that there was any error improvement. It was impressive, but you had to squint.”

In other words, we don’t yet have a Sputnik moment for quantum computing, he says. “We’ve seen a rocket engine ignite. A static burn test, rather than getting your stuff into orbit.”

It’s not immediately clear how Microsoft’s approach could be used in a scalable way to suppress errors, or how it would apply to universal quantum gates.

“Yes, it is a good milestone,” Shaw says. “The debate that’s there in the field is how soon can we build large-scale fault-tolerant systems. This doesn’t really change the debate. A four-logical-qubit machine would have scientific interest, and maybe niche applications, but they’re not likely to be general-purpose, widely usable applications.”

Other companies have built quantum computers with more qubits. But quantum computing today is so nascent that there are multiple and radically different approaches to building the physical qubits.

Microsoft’s approach is heavily dependent on Quantinuum’s quantum-computing hardware, says Baptiste Royer, professor at the University of Sherbrooke, so it’s not immediately likely that other companies will jump on this same technology. “But they might be inspired by the theory behind this,” he says.

According to Royer, the latest announcement is the result of a series of small improvements – in addition to the error-correction codes, there were also improvements in the hardware, in calibration, in fabrication, in precision, as well as new measurement protocols.

As a result, there are no significant, immediate benefits for enterprises looking for practical quantum computing, especially with the smaller number of logical qubits.

“In terms of researchers looking at error correction, this gives them a playground to test things,” Royer says. “For researchers, this is exciting. In terms of the general public, I don’t see any immediate practical impacts – but it will bring quantum computing closer and reduce the time it will take to get to the practical things.”