Rethink the Role of Neural Decoders in Quantum Error Correction

Quantum error correction (QEC) is essential for enabling quantum advantages, with decoding as a central algorithmic primitive. Owing to its importance and intrinsic difficulty, substantial effort has been made to QEC decoder design, among which neural decoders have recently emerged as a promising data-driven paradigm. Despite this progress, practical deployment remains hindered by a fundamental accuracy–latency tradeoff, often on the microsecond timescale. To address this challenge, here we revisit neural decoders for surface-code decoding under explicit accuracy–latency constraints, considering code distances up to $d=9$ (161 physical qubits). We unify and redesign representative neural decoders into five architectural paradigms and develop an end-to-end compression pipeline to evaluate their deployability and performance on FPGA hardware. Through systematic experiments, we reveal several previously underexplored insights: (i) near-term decoding performance is driven more by data scale than architectural complexity; (ii) appropriate inductive bias is essential for achieving high decoding accuracy; and (iii) INT4 quantization is a prerequisite for meeting microsecond-scale latency requirements on FPGAs. Together, these findings provide concrete guidance toward scalable and real-time neural QEC decoding.

Quantum computers are powerful but highly fragile. Keeping them stable is like diagnosing a patient: we cannot perform an "autopsy" to find the illness, as this destroys the quantum state. Instead, we must identify errors without "killing" the patient. Furthermore, just as humans need regular checkups, a quantum state's fleeting lifespan forces this "doctor" to complete a full checkup in under a millionth of a second (a microsecond). AI is a promising tool for this task, but researchers face fundamental questions: Do more complex AI models perform better? Is more training data the secret? And most importantly, can AI actually run fast enough to be practically useful? In this paper, we systematically answer these questions to clarify how to build a practical "AI doctor." First, we show that complex models are unnecessary; simple AI designs perform just as well. Second, data is king—the volume of training data matters far more than model selection. Finally, by applying extreme lightweighting and compression techniques, we enabled the AI to make decisions within the strict microsecond deadline. Our research proves that AI can achieve real-time error correction for quantum systems larger than the best currently available superconducting hardware, offering a clear path toward practical quantum computing.