Logical Qubit Error Correction
Logical Qubit Error Correction
Quantum error correction (QEC) encodes many noisy physical qubits into a smaller number of protected logical qubits that tolerate hardware errors. Two thresholds matter: below threshold (logical errors suppress exponentially as code distance grows — Google's Willow proved this for surface codes in 2024) and beyond break-even (error-corrected operations beat raw hardware — Quantinuum and IBM crossed this in 2025–2026 with different architectures).
Four anchor results now define the field:
-
Google Willow (superconducting, Aug 2024) — distance-5 and distance-7 surface codes operate below threshold. Logical error suppression factor Λ = 2.14 per 2 distance steps — doubling distance halves error. Distance-7 code (101 qubits) achieves 0.143% ± 0.003% logical error per cycle; logical memory exceeds physical lifetime by 2.4×.
-
Besedin et al., Nature Physics (Jan 2026) — first lattice surgery on superconducting qubits, merging two distance-3 repetition-code qubits via fault-tolerant circuits. Memory is passive protection; lattice surgery is the compute primitive — the way to run logical circuits on surface-code-protected qubits.
-
IBM Nighthawk + qLDPC (Nov 2025) — real-time qLDPC error decoding in <480 ns, 10× faster than prior art. Paired with the 120-qubit Nighthawk processor and experimental Loon.
-
Quantinuum iceberg codes (March 2026) — 94 logical qubits from only 98 physical qubits on a trapped-ion processor. Logical gate error ~1 in 10,000; 95% GHZ fidelity across all 94. Crossed beyond break-even with radically lower overhead than surface codes.
Together, these establish that QEC is no longer asking "can it work?" but "which architecture scales first?"
Key Claims
Google Willow — Below-Threshold Surface Codes
- Λ = 2.14 ± 0.02 error suppression per 2 distance steps — exponential scaling confirmed. Evidence: strong (Willow)
- Distance-7 code: 0.143% ± 0.003% error per cycle on 101 physical qubits. Evidence: strong (Willow)
- Logical memory 2.4× physical lifetime — first clear below-threshold operation. Evidence: strong (Willow)
- Real-time decoding at distance-5 over 1M cycles — decoder latency 63 μs average. Evidence: strong (Willow)
Besedin et al. — Lattice Surgery on Superconducting
- First superconducting lattice surgery — demonstrated between two distance-3 repetition-code qubits. Evidence: strong (Besedin)
- Fault-tolerant ZZ observable — decoded ZZ improvement vs non-encoded circuit. Evidence: strong (Besedin)
- Logical Bloch sphere tomography — validates across non-cardinal states. Evidence: moderate (Besedin)
Quantinuum Iceberg Codes
- 94 logical qubits from 98 physical qubits. Evidence: strong (Quantinuum)
- Logical gate error ~1 in 10,000. Evidence: strong (Quantinuum)
- 95% GHZ fidelity across 94 logical qubits. Evidence: strong (Quantinuum)
IBM qLDPC Codes
- qLDPC decoding <480 ns — 10× prior art. Evidence: strong (IBM)
- 100×+ cost reduction in error mitigation. Evidence: strong (IBM)
Three Code Families Compared
| Property | Surface (Willow) | Iceberg (Quantinuum) | qLDPC (IBM) |
|---|---|---|---|
| Architecture | Superconducting | Trapped-ion | Superconducting |
| Paradigm | Memory protection | Detection + correction | Dense parity checks |
| Overhead | ~100 physical → 1 logical @ d=7 | 98 physical → 94 logical | Moderate, dense |
| Demonstrated scale | d=7 memory + d=29 repetition | 94 logical qubits | Architectural + Loon |
| Compute primitive | Lattice surgery (Besedin 2026) | GHZ state, MCMR | Dynamic circuits |
| Key caveat | Large overhead | Postselection reliance | Custom decoding HW |
Benchmarks & Data
- Willow: Λ=2.14; distance-7 @ 0.143% error/cycle; 1.1 μs cycle time; 63 μs decoder latency; 2.4× logical/physical memory.
- Quantinuum: 94 logical / 98 physical; 1e-4 logical gate error; 95% GHZ fidelity; 30% 2Q gate error reduction.
- IBM: 120 qubits (Nighthawk), 5,000 2Q gates at launch, 218 couplers, qLDPC <480ns, 100×+ error-mitigation cost reduction.
- Besedin: distance-3 repetition codes, first superconducting lattice surgery, decoded ZZ improvement over non-encoded baseline.
The Logical Compute Stack
- Physical qubits — hardware, error-prone.
- Below-threshold memory — Willow; errors suppress exponentially with distance.
- Lattice surgery / fault-tolerant gates — Besedin; operate logically while protected.
- Logical algorithms — arbitrary circuits on logical qubits → this is fault tolerance.
- Quantum advantage — a useful logical algorithm outruns classical.
Willow nailed layer 2; Besedin demonstrated the first rung of layer 3 on superconducting hardware; Quantinuum demonstrated layer 2+3 on trapped-ion at scale; IBM's qLDPC decoder is the engineering substrate for all of it.
Open Questions
- Does iceberg-code postselection scale, or does it collapse at very large circuit depths?
- Can qLDPC decoding at <480ns run on commodity hardware, or only custom ASICs?
- Which architecture reaches fully fault-tolerant (not just break-even) first?
- Are trapped-ion error rates intrinsically easier to correct than superconducting, or is this a transient lead?
- What's the minimum code distance needed for useful near-term applications?
- Can Besedin-style lattice surgery scale to distance-5, 7, 9 on superconducting?
Related Concepts
- Quantum Fault Tolerance Roadmap — the milestone ladder these codes climb
- Neutral Atom Quantum Computing — the third architecture entering the QEC race
Backlinks
Pages that reference this concept:
Changelog
- 2026-04-17 (initial) — Compiled from Quantinuum (Mar 2026) + IBM (Nov 2025).
- 2026-04-17 (update) — Added Google Willow (below-threshold, Λ=2.14) and Besedin lattice surgery (superconducting, Nature Physics Jan 2026). Reframed from 2-architecture bet to 3-family comparison (surface/iceberg/qLDPC). Added logical compute stack diagram.