“Google has shown that its Sycamore quantum computer can detect and fix computational errors, an essential step for large-scale quantum computing, but its current system generates more errors than it solves.
Error-correction is a standard feature for ordinary, or classical, computers, which store data using bits with two possible states: 0 and 1. Transmitting data with extra “parity bits” that warn if a 0 has flipped to 1, or vice versa, means such errors can be found and fixed.
In quantum computing the problem is far more complex as each quantum bit, or qubit, exists in a mixed state of 0 and 1, and any attempt to measure them directly destroys the data. One longstanding theoretical solution to this has been to cluster many physical qubits into a single “logical qubit”. Although such logical qubits have been created previously, they hadn’t been used for error correction until now.
Julian Kelly at Google AI Quantum and his colleagues have demonstrated the concept on Google’s Sycamore quantum computer, with logical qubits ranging in size from five to 21 physical qubits, and found that logical qubit error rates dropped exponentially for each additional physical qubit. The team was able to make careful measurements of the extra qubits that didn’t collapse their state but, when taken collectively, still gave enough information to deduce whether errors had occurred.
Kelly says that this means it is possible to create practical, reliable quantum computers in the future. “This is basically our first half step along the path to demonstrate that,” he says. “A viable way of getting to really large-scale, error-tolerant computers. It’s sort of a look ahead for the devices that we want to make in the future.”
The team has managed to demonstrate this solution conceptually but a vast engineering challenge remains. Adding more qubits to each logical qubit brings its own problems as each physical qubit is itself susceptible to errors. The chance of a logical qubit encountering an error rises as the number of qubits inside it increases.
There is a breakeven point in this process, known as the threshold, where the error correction features catch more problems than the increase in qubits bring. Crucially, Google’s error correction doesn’t yet meet the threshold. To do so will require less noisy physical qubits that encounter fewer errors and larger numbers of them devoted to each logical qubit. The team believes that mature quantum computers will need 1000 qubits to make each logical qubit – Sycamore currently has just 54 physical qubits.
Peter Knight at Imperial College London says Google’s research is progress towards something essential for future quantum computers. “If we couldn’t do this we’re not going to have a large scale machine,” he says. “I applaud the fact they’ve done it, simply because without this, without this advance, you will still have uncertainty about whether the roadmap towards fault tolerance was feasible. They removed those doubts.”
But he says it will be a vast engineering challenge to actually meet the threshold and build effective error correction, which would mean building a processor with many more qubits than has been demonstrated until now.
Journal reference: Nature, DOI: 10.1038/s41586-021-03588-y”