Fault-Tolerant Quantum Computing
Fault-Tolerant Thresholds & Surface-Code Decoding · Quantum Computing
Cross the fault-tolerance threshold — decode the surface code fast enough that adding qubits drives the logical error rate down, not up.
📋 The problem
Qubits are noisy; useful quantum computing needs error correction. Below a threshold physical error rate, the surface code suppresses logical errors exponentially with code distance — if a decoder can keep up in real time.
🧗 Why it's a grand challenge
Decoding must be accurate AND fast enough to run every cycle; correlated noise, leakage and crosstalk break idealized models; demonstrating below-threshold scaling is hard.
🧮 Governing model
p_L ≈ A·(p/p_th)^{(d+1)/2}; Λ = p_L(d) / p_L(d+2)Logical error rate p_L of a distance-d surface code under a decoder: p_L ≈ A·(p/p_th)^{⌊(d+1)/2⌋}; below threshold p<p_th, p_L falls exponentially with d.
Current best: Surface-code below-threshold demonstrations (Google Willow) + ML decoders
🧭 Possible approaches
- Neural and Union-Find real-time decoders
- Noise-adaptive decoding from measured syndromes
- Co-design of codes and decoders for biased noise
🎯 Build the benchmark
From distance-3…25 syndrome streams across error rates, minimize per-cycle logical error and show suppression Λ > 1 within the cycle-latency budget.
Metric: logical_error_rate — logical error rate per cycle (lower better)
Datasets to start from: Surface-code syndrome-measurement corpus, Repetition-code error chains
🤖 Build an AI agent to solve it
An agent that learns a hardware-specific decoder from calibration data and tunes the code to the device's noise.
Once a benchmark exists, an AI4Science agent can iterate solutions against it — every verified solution earns PWM.
This is a frontier framing page — an open problem, not yet benchmarked or verified, unlike PWM's mature computational-imaging benchmarks.