| name | sparse-mamba-qec-decoder |
| description | Sparse Mamba Decoder (SMD) for quantum error correction — defect-centric neural decoder using state-space (Mamba) backbone. Processes only k active detection events (O(k) complexity) instead of full O(d²R) syndrome array. Reduces MWPM logical error rate by up to 49%, runs 95-467x faster than Tesseract near-MLD, achieves 24-57μs latency across d=3-9. Activation: sparse mamba decoder, SMD, QEC neural decoder, surface code Mamba, defect-centric decoding, sparse syndrome processing, quantum error correction state space model. |
| category | quantum |
Sparse Mamba Decoder for Quantum Error Correction
Defect-centric neural decoder for surface code syndromes using a Mamba (state-space model) backbone. Based on arXiv:2605.17156.
Problem Statement
Quantum error correction (QEC) requires decoders that are simultaneously accurate, fast, and scalable.
Existing limitation: State-of-the-art neural decoders process the full dense syndrome array of size O(d²R) regardless of actual error rate. At physically relevant error rates (p ~ 0.1%), fewer than 5% of syndrome entries contain active detection events — yet all existing decoders process the entire syndrome volume.
Core Innovation: Defect-Centric Sparse Processing
The Sparse Mamba Decoder (SMD) processes only the k active detection events using:
- 13-dimensional feature representation per defect — compact encoding of each detection event's spatiotemporal properties
- Mamba state-space backbone — linear-complexity sequence modeling of the defect stream
- O(k) total complexity — decouples from code distance d and number of measurement rounds R
