Picture a vast hall where thousands of janitors are sweeping the floor at the same time. Each one knows only their own patch and sweeps debris toward the nearest wall, without asking a manager. It used to be thought that such an approach was impossible for a quantum computer: too many errors, and you need a central brain that sees the whole picture. But now physicists have shown that this brain can be done away with.

A quantum computer stores information in qubits, which are extremely fragile: the slightest noise from outside corrupts the data. To fix this, error-correcting codes are used, such as the surface code. It arranges qubits on a two-dimensional lattice, like tiles on a floor. When an error occurs, a defect appears on the lattice: a violated parity check. Usually these defects are collected by a powerful classical computer, but it becomes a bottleneck: the more qubits, the faster it has to work.

The authors of the work built a cellular automaton: a simple rule by which each node communicates only with its neighbors. Defects move toward the corner of their cluster, where they disappear. For more complex errors, a stack of several layers of such an automaton is used: defects rise up the stack, like bubbles in soda, and annihilate there. All of this works continuously, even while the computer is performing operations: initializing states, entangling qubits, or reading out the result.

Previously, local decoders could only store quantum information, not perform computations. Now it has been shown that both can be done. Moreover, the density of classical resources is constant: no matter how many qubits there are, each node needs only a fixed set of bits. This opens the way to quantum computers that correct errors directly on the chip at low temperatures, without sending data outside.

Let's return to the hall with the janitors. Now each janitor doesn't just sweep, but also knows that debris must be carried to the corner, and does so without orders. A quantum computer built on this principle could become simpler and more reliable, and therefore closer to real-world use.