Research work · 2026
LATTE: A Decoding Architecture for Quantum Computing with Temporal and Spatial Scalability
ACM Transactions on Quantum Computing
Abstract
Quantum error correction allows noisy quantum devices to emulate ideal quantum computers with reasonable resource overhead. As a crucial component, decoding architectures must scale with rapid advances in quantum hardware. We introduce LATTE, an FPGA-CPU hybrid decoding architecture addressing key requirements for scalability, especially in lattice surgery— L atency, A ccuracy, T hroughput and T ransmission Bandwidth, in an E clectic manner. LATTE combines (1) a fully streaming, asynchronous CPU-based block decoding system that enables temporal and spatial parallelism, and (2) a lightweight neural local decoding unit on FPGA, which transparently reduces syndrome transmission bandwidth and accelerates decoding. LATTE is compatible with various base decoders; integrated with MWPM, it achieves comparable accuracy and real-time throughput with reduced bandwidth and computational resources. Under circuit-level noise p = 0.001, LATTE achieves over \(\mathbf {90\%} \) reduction in syndrome transmission bandwidth and a 6.4 × speedup on average in single-block decoding. For streaming decoding: (1) LATTE achieves constant low latency ( 16 × – 20 × speedup over existing methods) in arbitrarily long quantum memory experiments, with near-optimal computational resources—only 2 threads are sufficient to decode surface codes up to distance 17. (2) LATTE reduces latency in multi-patch measurement experiments through highly parallelized decoding. These combined efforts ensure sufficient scalability for large-scale fault-tolerant quantum computing.
DOI: 10.1145/3858417