Researchers have developed a new decoder for quantum error correction (QEC) that significantly improves the trade-off between accuracy and latency. This advancement is crucial for large-scale quantum computation, which requires fault-tolerant systems capable of extracting and decoding error syndromes in real time. The decoder, termed coset ensemble decoding, optimizes the Union-Find (UF) decoding approach by explicitly exploiting logically equivalent cosets, allowing for a more accurate approximation of coset-level maximum-likelihood decoding.

The proposed method uses ensemble forest exploration to generate multiple coset-consistent candidates and aggregates them to enhance accuracy. Furthermore, computational and memory complexity have been reduced through reverse-order elimination and lossless graph compression, while maintaining decoding fidelity. At the hardware level, the team designed a domain-specific architecture that temporally reuses resources, avoiding the code-distance-proportional resource growth characteristic of prior spatial architectures. Optimizations such as multi-bank memory hashing and hierarchical ID mapping have been implemented to mitigate pipeline stalls and memory conflicts under highly concurrent access patterns.

In tests with a circuit-level depolarizing noise model, this new co-designed approach outperforms decoders based on Minimum-Weight Perfect Matching (MWPM) and Union-Find (UF) in terms of accuracy-latency trade-off. Additionally, it reduces FPGA LUT (Look-Up Tables) consumption by up to 8.2 times compared to resources reported by previous UF decoders. The ability to tune the number of candidates offers additional flexibility, allowing decoding performance to be tailored to the specific requirements of different fault-tolerant workloads in quantum computing.