What is quantum error correction?

How it works

Qubits, the basic units of a quantum computer, lose information easily. Decoherence, the loss of the delicate quantum states that hold data, was recognized early as an obstacle to long calculations. In 1995, Peter Shor of AT&T Bell Laboratories showed a way to reduce its effects on stored quantum information using a quantum analog of classical error-correcting codes.

The core idea is redundancy. Each "logical qubit" is encoded in a group of physical qubits. So an error in any one of them need not destroy the information, as IEEE Spectrum explained in June 2025. One widely used scheme, the surface code, takes roughly 1,000 physical qubits per logical qubit, the magazine reported.

Why the threshold matters

More qubits help only if the hardware is good enough. Enlarging a code suppresses logical errors exponentially, a Nature paper published online in December 2024 says. It says that holds only when the physical qubits' error rate is under a critical threshold. In it, Google Quantum AI and collaborators reported operating below that threshold with surface codes on Willow superconducting processors. Each increase of two in code distance, a measure of code size, cut the logical error rate by a factor of 2.14. A 101-qubit logical memory outlasted its best physical qubit by a factor of 2.4.

Where things stand in 2026

IBM has published targets for lower overhead and larger scale. The company said in June 2025 that its quantum low-density parity check (qLDPC) codes cut error-correction overhead by about 90 percent compared with other leading codes. It said it plans to deliver Starling, a fault-tolerant system with 200 logical qubits able to run 100 million operations, by 2029. IBM uses fault-tolerant to mean error corrected well enough to complete large workloads without faults. Those are IBM's claims and targets, not delivered results.

The U.S. Department of Energy announced on June 23, 2026 an initiative to develop and deploy a fault-tolerant quantum computing capability for research by 2028. It includes a competition that aims to demonstrate systems with logical qubits in the low hundreds. IEEE Spectrum reported that month that the record holders were QuEra, claiming 96 logical qubits, and Quantinuum, claiming 94. Edward Parker of the Rand Corporation told the magazine that hundreds of logical qubits by 2028 is plausible. But Parker said such qubits would still be noisy and unable to sustain very long computations.