Arrive at IAMM

Flash Quantum Surface-Code Decoding on GPU Thread-Block Clusters Room: 147 Abstract: Quantum error correction (QEC) is essential for reliable quantum computing because physical qubits are highly susceptible to noise and cannot simply be copied or directly measured without disturbing their quantum information. The surface code addresses this challenge by encoding logical information across many physical qubits and repeatedly measuring error syndromes using ancillary qubits. A classical decoder must then process these syndrome measurements and determine appropriate corrections in real time, creating a demanding tradeoff between decoding speed and accuracy. In this work, we develop a neural-network-based surface-code decoder optimized for GPU thread-block clusters. Rather than repeatedly transferring data between the GPU and external memory, the entire decoding computation remains on-chip during each decoding step. Eliminating this memory-access bottleneck enables roughly an order-of-magnitude reduction in latency compared with conventional implementations of the same neural network. The resulting speed allows a sufficiently large neural network to operate within the real-time decoding budget while achieving a lower logical error rate than a leading matching-based decoder. Because each decoder instance occupies only a small fraction of an NVIDIA H100 GPU, a single commodity data-center GPU can decode many logical qubits simultaneously. These results demonstrate a practical path toward accurate, low-latency, real-time quantum error correction using modern GPU hardware.

Research Meeting

Research Meeting

Research Meeting

Research Meeting

Research Meeting

Research meeting