If a small quantum computer makes a small number of errors, will a large quantum computer make even more errors, making it completely useless? No, say researchers at Google who have made a key breakthrough in error correction for quantum devices, setting out a theoretical path to creating machines that are useful and practical.
Ordinary computers store data as bits that are either a 0 or 1, but errors can cause the bit to 鈥渇lip鈥� to the wrong value, which is why devices from smartphones to supercomputers have built-in error correction. Quantum computers are even more prone to errors, but the problem is also more complex for them because each quantum bit, or qubit, exists in a mixed state of 0 and 1, and any attempt to directly check its value destroys the fragile quantum state.
To solve this, quantum computers can be made to group a number of physical qubits to act as a single 鈥渓ogical鈥� qubit, allowing for cross-checking of values while preserving their quantum state. In this scenario, instead of directly checking the value of a qubit, a technique called surface code correction lets you observe the relative properties of the physical qubits in different ways, which can give hints that an error has occurred.
Advertisement
But this introduces a further problem, as Google discovered in 2021, since adding more physical qubits unhelpfully introduced even more errors than they solved. The goal was to find a way of reducing errors faster than they were introduced, a breakeven point known as the threshold.
Google now says it has crossed that threshold. In experiments, Google researchers managed to pass the breakeven point with both a 72 qubit processor and a 105 qubit processor, each acting as a single logical qubit. In one test, the logical qubit accurately preserved quantum information for more than twice as long as any of the physical qubits it was made up of could do individually. In their paper on the work, the researchers claimed an error rate of just 0.143 per cent per cycle of the machine; better than previous quantum machines but still high enough to cause problems with large calculations.
The firm declined to speak to 快猫短视频, but in the paper the researchers say that although previous efforts by other teams had 鈥渄emonstrated different features of quantum error correction鈥�, they fell short of a definitive display, which Google now claims to have made. 鈥淲ith below-threshold surface codes, we have demonstrated processor performance that can scale in principle, but which we must now scale in practice,鈥� writes the team.
But the team also warns that significant hurdles remain. Improving its processors to the point of only making one error per million operations would be 鈥渞esource intensive in practice鈥� and require 1457 physical qubits for each logical qubit 鈥� which is more than contained in even the largest quantum computers currently in existence.
Google is now in a race to scale its quantum computers as rivals attempt to do the same. One key question is whether Google鈥檚 chosen architecture for quantum computing, using superconducting qubits, will prove to be the correct way forward. Teams using an alternative approach based on trapped ions have previously demonstrated error correction, but only for a dozen or so physical qubits.
at the University of Texas at Austin says that the work pushes quantum computing closer towards practicality and is a vindication for Google鈥檚 choice of quantum technology over competitors. 鈥淚鈥檇 characterise this as another significant step forward in realising error correction, and as representing a collective effort of Google鈥檚 experimental group for the past couple years,鈥� he says. 鈥淭rapped ions and neutral atoms pushed ahead of superconducting qubits lately in terms of fidelity, so the ball has really been in the court of Google, IBM and the other superconducting groups to answer that challenge, and this could also be seen in that light.鈥�
at the University of Glasgow says the current generation of quantum computers can do nothing useful that couldn鈥檛 be done with the computer power found in a smartphone, but that Google鈥檚 latest work shows that the machines can at least be scaled up in a practical way in the future.
鈥淭hey really showed that by putting qubits together, you reduce the error rate. You actually get better and better and better,鈥� he says. 鈥淚s it useful [yet]? Probably not. You would need tens of thousands, millions [of error-corrected qubits] to build anything useful. But it鈥檚 a great demonstration of the concept.鈥�
arXiv
