Many researchers believe that the only way to build unambiguously useful quantum computers is to enable them to correct their own errors. A breakthrough in December from researchers at Google Quantum AI charted one path towards making this a practical reality. Their approach, however, may already be in danger of becoming outdated.
A big factor preventing quantum computers from living up to their promise – solving seemingly intractable problems in materials science, chemistry, logistics and many other fields – is that they constantly make errors. And as they get larger and gain more computing power, these accrue even more.
Earlier this month, however, researchers at Google Quantum AI demonstrated that this doesn’t have to be the case. Using a quantum computing processor called Willow, they showed that computations can be corrected by grouping qubits – the basic building blocks of any quantum computer – into so-called logical qubits, which can be made larger without negatively affecting performance.
Advertisement
The mathematical recipe the team used to group qubits is called the surface code, and it has long been the front-runner among quantum error-correction approaches. Here, qubits are arranged in two interwoven square grids, one consisting of data qubits, which are used for computation, and one comprising ancillary qubits, which can detect errors in their data qubit neighbours.
In 2023, researchers at IBM introduced a competitor – the QLDPC code – where each qubit is connected to six others, and all seven monitor each other for errors. This approach cuts down the overall number of necessary qubits, with IBM’s team estimating that it could use only 288 qubits to achieve the same level of error correction that the surface code delivers with 4000 qubits. Quantum computers have only recently broken the 1000-qubit barrier, and no 4000-qubit machines exist yet – researchers are racing to build one in the next few years, but using QLDPC may allow them to make quantum computers useful without jumping over that engineering hurdle.
“To me, that lower qubit overhead is hard to compete with,” says at the start-up Horizon Quantum.
IBM researchers have also been designing their quantum computing chips to be suitable for the connections that the QLDPC code requires. Adding new connections between qubits is tricky because it creates new ways for qubits to be disturbed and new routes for their information to be lost. But the IBM team found that the redesign didn’t change the reliability of logical operations that their chips can support, said IBM’s at the in Santa Clara, California, on 10 December. “Not all logical qubits are created equal,” he said, underscoring that while they all protect calculations from errors, some are harder to make.
at Global Quantum Intelligence, a UK business intelligence company, said at the conference that advances in the QLDPC code and similar mathematical recipes have turned error correction into a “a hotbed of innovation”.
Which approach will win out, enabling a useful quantum computer first, remains to be seen. “Maybe someone somewhere is working on a type of surface code that is really great, but right now there is competition [to the surface code],” says at US-based quantum computing start-up QuEra.
Another ingredient that could tip the balance in favour of one approach is the physical makeup of the qubits – the way qubits are made constrains which codes they can be used for. Both the IBM and Google Quantum AI teams use qubits made from tiny superconducting circuits that are connected in a relatively fixed way. But others use ultracold atoms, which can be shuttled around and constantly re-connected to one another.
Boger says this may restrict the two companies to using codes that best match their hardware instead of codes that simply deliver the lowest error rates most efficiently. His team previously used a quantum computer with qubits made from extremely cold atoms that had one of the largest numbers of logical qubits yet, and he says the group is interested in trying different codes to reconfigure the logical qubits and make them more useful.
However, implementing the kind of connectivity that codes like QLDPC require may be a non-trivial engineering challenge for many existing quantum computers, says Shaw. For him, this is a reason not to write off the surface code just yet. Additionally, the details of how to use logical qubits created with QLDPC for full calculations is still less clear than for the surface code, which has been studied for more than two decades.
“The surface code is well understood, with a well-studied theoretical framework. It offers a balance between performance and required qubit connectivity, making it well-suited for superconducting qubits, and thus well-suited for Willow,” says at Google Quantum AI. But his team is also exploring other quantum error-correction codes too, he says.
Ěý
