computer science//quantum computing//quantum error correction

Quantum error correction is the set of methods that encode one protected **logical qubit** in many physical qubits and repeatedly detect and undo errors in them, so that a computation can run far longer than any single qubit stays coherent. It is the precondition for every quantum algorithm that would matter commercially: those need billions of operations, and physical qubits fail about once every thousand.

Quantum error correction is the set of methods that encode one protected logical qubit in many physical qubits and repeatedly detect and undo errors in them, so that a computation can run far longer than any single qubit stays coherent. It is the precondition for every quantum algorithm that would matter commercially: those need billions of operations, and physical qubits fail about once every thousand.

The trick has to work under two quantum restrictions. An unknown state cannot be copied, so the classical recipe of keeping three copies and voting is out; and measuring the data qubits would destroy the superposition. Instead the code entangles the data with extra measurement qubits that check parities between neighbours (are these two the same?) without revealing the values themselves. The pattern of failed checks, the syndrome, tells a classical decoder which error most likely happened, and the correction is applied or tracked in software. This cycle repeats every microsecond or so, for the whole computation, which makes the decoder a hard real-time system.

Below the threshold, more qubits mean fewer errors; above it, more qubits mean more.

Adding qubits adds places for errors to happen, so a code only helps when each physical qubit is good enough. Under that threshold, each step up in code size (its distance) divides the logical error rate by a roughly constant factor.

The surface code is the leading scheme for superconducting chips: data qubits on a square grid with measurement qubits between them, needing only nearest-neighbour connections. A code of distance ddd uses about 2d22d^22d2 physical qubits and corrects up to (d−1)/2(d-1)/2(d−1)/2 errors per round.

Google's Willow chip (December 2024) showed the threshold behaviour for the first time at useful scale: logical error rates fell by about half from distance 3 to 5 to 7, and the distance-7 logical qubit (101 physical qubits) outlived its best physical qubit by more than a factor of two. A memory that holds a state is still far from a fault-tolerant computer that runs algorithms.

The overhead sets the size of useful machines. Estimates for breaking RSA-2048 with Shor's algorithm fell from about 20 million physical qubits (2019) to under a million running for a week (Gidney, 2025), with error rates and speeds no current machine reaches.

The engineering reading resembles the redundancy of fault-tolerant control, with a heavier bill: where a flight computer triplicates, a logical qubit may take a thousand physical ones plus the cooling, wiring and real-time decoding that each of them needs (decoherence).