arXiv:2610.13879
GPU-Based Parallel BP Decoding for Quantum LDPC Memory
Ming Wang, Yanbo Zhao, Ang Li, Frank Mueller
Abstract
Quantum low-density parity-check (qLDPC) decoding requires reliable corrections and low latency over long syndrome histories. We extend BP-SF, an oscillation-guided belief-propagation (BP) decoder with independent syndrome-flip trials, by introducing windowed and streaming operations. Our CUDA implementation combines disordered-memory retries, sliding windows, correlation-aware graphs, and layered scheduling. Circuit-level simulations test fixed-basis memory and complete logical Pauli-frame recovery. On the [[144,12,12]] bivariate bicycle code at physical error rate p = 0.002, parallel BP-SF reduces 99.9th-percentile call latency by a factor of 13.15 relative to an eight-chain Relay-BP GPU baseline on 100,000 matched samples for which every tested method's initial BP stage fails within 80 iterations. Windowed BP-SF maintains a flat logical error rate and nearly constant per-round compute time through 512 rounds, reaching a 17.1-fold speedup over monolithic decoding. In 512-round full-frame streaming decoding with 2.97-ms syndrome arrivals, layered BP-SF on a transformed joint graph with 20 retries reduces 99.9th-percentile correction-release age from 166.29 to 36.25 ms relative to the joint graph with flooding updates and 100 retries. These results connect parallel BP post-processing to lower decoding costs and earlier correction release in streaming quantum memory.
AI Summary
The paper asks how parallel belief-propagation decoding can provide accurate, low-latency corrections over long qLDPC syndrome histories. It extends the previously introduced oscillation-guided syndrome-flipping decoder, BP-SF, with CUDA-parallel, posterior-initialized disordered-memory BP retries, sliding windows, correlated-error graphs, and layered updates. Evaluation uses circuit-level stochastic-Pauli simulations, GPU call timings, and measured service traces replayed against FIFO syndrome arrivals. On the $[[144,12,12]]$ bivariate bicycle code at $p=0.002$, BP-SF achieves a logical error rate of $4.38\times10^{-6}$ per round in twelve-round fixed-basis memory. On 100,000 matched samples conditioned on every tested decoder’s initial BP stage failing, its mean and 99.9th-percentile call latencies are 1.081 and 7.039 ms, versus 6.796 and 92.556 ms for eight-chain Relay-BP. Seven-round cores with three-round buffers maintain approximately $10^{-5}$ logical error per round and 33–36 microseconds of compute per round through 512 rounds, giving a $17.1\times$ speedup over monolithic decoding at that length. For twelve-round full-Pauli-frame recovery, layered BP-SF on a GARI-transformed graph with 20 retries has overlapping accuracy intervals with joint-graph flooding using 100 retries, while reducing mean latency from 31.304 to 1.624 ms. In 512-round streaming replay with 2.97-ms arrivals, their 99.9th-percentile correction-release ages are 36.25 and 166.29 ms. Limitations include substantially worse accuracy than Relay-BP on the $[[288,12,18]]$ code, ideal boundaries in full-frame tests, and unresolved generalization across code families and long-run queue stability. The headline Relay-BP timing comparison is conditional, not representative of unfiltered calls.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.