Customizing Stim to Take Nonuniform Qubit Noise into Account and Analyzing DEM/Tanner Graph in Surface‑Code Simulations

Fault‑tolerant quantum computing (FTQC) is essential for practical quantum computation with quantum devices including large number of qubits. In real devices, physical error rates significantly vary from qubit to qubit, and multiple error mechanisms coexist. Operating quantum error‑correcting codes under such device‑specific nonuniform noise remains a key challenge. We address the challenge by customizing the stabilizer‑circuit simulator Stim [1] to incorporate device structure and assign error probabilities per qubit/gate regarding reset, single- /two-qubit gates, idling, shuttling, and measurements, enabling to incorporate hardware‑realistic nonuniform noise. To clarify the impact of device specific errors in decoding process, we focus on the Detector Error Model (DEM) provided by Stim, which is a graph representation composed of nodes being the parity of measurement outcomes and edges representing abstract errors identified by the measurement outcomes. To analyze the errors on specific qubits in decoding, we provide Tanner style DEM visualizations, showing detectors (nodes), errors (edges), and links back to original circuit faults. This facilitates how nonuniform edge weights influence minimum‑weight perfect matching (MWPM) decoding [2] and where failure comes. As an example, in a distance‑3 surface code with initialization errors, we find that data qubits covered only by a single weight‑4 stabilizer impact logical‑error rates most highly. This customization reveals which location leads to a logical error rate and how the code behaves in this setting, providing a first step toward proposing code‑operation strategies that explicitly account for device inhomogeneity.