Methodology

Our experimental evaluation leverages the ECCentric framework to systematically explore the QEC landscape under realistic noise models and hardware constraints.

Simulation Environment

We run stabilizer simulations using Stim, which allows for efficient, exact sampling of error propagation and syndrome triggering. Detector error models are generated from these samples and fed to decoders to estimate logical error rates.

Experimental Parameters

• Code Distances: Unless stated otherwise, we use circuits at the maximal distance supported by the topology.
• Correction Rounds: Equal to the code distance for all experiments.
• Shot Count: 1000 shots per experiment, providing sufficient statistical resolution to distinguish between error suppression and code failure regimes.

Noise Modeling

We inject noise gate-by-gate, supporting both standard models (SI1000) and realistic, device-specific models for platforms such as Infleqtion, Google Willow, IBM Flamingo, and Quantinuum Apollo. Noise sources include:

• Single- and two-qubit gate errors.
• Idle noise (constant or T1/T2-derived).
• Leakage, crosstalk, reset, and readout errors.
• Qubit shuttling and inter-chip gate errors.

Decoding Strategies

We evaluate performance using general-purpose decoders:

• MWPM (Minimum-Weight Perfect Matching): Selected for speed and efficiency.
• BP-OSD (Belief Propagation with Order Statistic Decoding): Selected for its superior accuracy across various code families, specifically using order-7 OSD and up to 10,000 iterations for convergence.