Scalable Surface-Code Decoders with Parallelization in Time.
summary
Claim: a sandwich decoding scheme splits the 3D surface-code syndrome graph into overlapping time windows with buffer zones and runs inner decoders in parallel (minimum-weight perfect matching or union-find), which makes decoding throughput scale with the number of parallel windows and become independent of syndrome-generation rate, removing the exponential backlog from sequential logical feedforward. Implemented in circuit-level depolarizing-noise memory simulations, it keeps logical error rates and threshold almost equal to offline decoding when buffer ≈ d/2, so it works with constant overhead.
AI-written summary · about
chapters (11)
- 0:00Classical processing in quantum fault tolerance
- 3:43Exponential backlog in teleportation
- 9:27Parallelization in time for backlog
- 13:54Repetition code and decoder basics
- 21:03Sandwich decoding with parallel windows
- 26:32Extending sandwich decoding to surface codes
- 27:59Summary and future directions
- 30:31Q&A: Decoder reconciliation
- 36:20Q&A: X-Z decoding correlation
- 39:15Q&A: Avoiding backlog without decoders
- 40:01Q&A: Fusion Blossom differences