Quantum Error Correction Below the Surface Code Threshold
Google Willow: distance-5 and distance-7 surface codes operate below threshold — logical error suppression factor Λ = 2.14 per 2 distance steps; distance-7 achieves 0.143% error per cycle; logical memory exceeds physical qubit lifetime by 2.4x
Quantum Error Correction Below the Surface Code Threshold
Abstract
The Google Quantum AI team demonstrates surface code quantum memory operating below the critical error threshold on their Willow processor. Distance-7 and distance-5 codes both achieve exponential logical error suppression as code distance grows: a logical error suppression factor of Λ = 2.14 ± 0.02 when distance increases by 2. The distance-7 code (101 physical qubits) reaches 0.143% ± 0.003% logical error per correction cycle — and logical memory exceeds physical qubit lifetime by a factor of 2.4 ± 0.3. This is the foundational "beyond break-even in a surface code" result that anchors the 2026 lattice-surgery follow-on work.
Key Contributions
- Below-threshold operation — both distance-5 and distance-7 surface codes suppress logical errors faster than physical error rates introduce them.
- Exponential suppression — Λ = 2.14 ± 0.02 per 2 distance steps (double distance, halve error).
- Distance-7 @ 101 qubits: 0.143% error/cycle — lowest reported surface code logical error rate at this scale.
- Logical memory > physical memory — 2.4× longer coherence in the protected logical qubit.
- Real-time decoding at distance-5 over 1 million cycles — decoder latency 63 μs on average.
- Scaled to distance-29 repetition codes — to probe performance limits.
Methodology
- Surface code quantum memory on Google's 101-qubit Willow processor.
- Real-time decoder integrated with the control stack.
- Cycle time 1.1 μs.
- Scaling experiments from distance-3 through distance-7 in surface codes; up to distance-29 in repetition codes (simpler, scales further).
Results
| Metric | Value |
|---|---|
| Logical error suppression factor Λ | 2.14 ± 0.02 |
| Distance-7 error per cycle | 0.143% ± 0.003% |
| Logical vs physical memory ratio | 2.4 ± 0.3 |
| Cycle time | 1.1 μs |
| Decoder latency @ distance-5 | 63 μs average |
| Real-time decoding duration | 1 million cycles |
| Maximum distance tested (repetition) | 29 |
| Distance-7 physical qubits | 101 |
Limitations
- Rare correlated error events (~1 per hour, i.e., every 3×10⁹ cycles) limit logical performance at the largest distances tested.
- Below-threshold demonstration is on surface code memory, not logical computation — lattice surgery (covered in Besedin 2026) is the next step.
- Decoder latency (63 μs avg) may become a bottleneck as code distance grows further.
- Results scale-limited to single logical qubit; multi-logical-qubit systems face additional challenges.
Why This Matters
This is the paper that changed the conversation from "can QEC work?" to "how fast can QEC scale?" Every 2026 QEC result — Quantinuum's iceberg codes, IBM's qLDPC decoding, the Besedin lattice-surgery paper — exists on the ground this paper laid. The exponential error-suppression factor Λ is the critical quantity: if Λ > 1 (it is, at 2.14), adding qubits makes logical errors drop exponentially, and fault tolerance is a matter of engineering rather than physics.
Full Content
Abstract and benchmarks extracted from arxiv preprint 2408.13687. Published in Nature (s41586-024-08449-y). 251 co-authors is standard for Google Quantum AI papers — effectively the entire Willow team.
Source: Quantum error correction below the surface code threshold, Acharya et al., Google Quantum AI, ArXiv Aug 24 2024. Nature version: s41586-024-08449-y.