Skip to content
Quantum.world

The Signal

FaultScope Maps Quantum Error Hotspots in QEC Simulations

FaultScope estimates how individual noisy components affect logical error rates in quantum error correction circuits. Its authors report that calculating all component sensitivities added 0.33% to 0.97% runtime in repetition-code tests.

FaultScope Maps Quantum Error Hotspots in QEC Simulations

We introduce FaultScope, a classical simulator for quantum error correction circuits, implemented in Rust with a Python interface.

Yuguo Shao, Zhenyu Chen, Ningfeng Wang, Song Cheng, Zhaohui Wei, Zi-Wen Liu and Zhengwei Liu

The FaultScope paper appeared on arXiv on October 30, 2026, describing software designed to show which noisy circuit components contribute most to logical failures. The tool estimates those contributions alongside the logical error rate, giving researchers a way to prioritize candidate sources of noise in simulations. Its authors report that the extra calculations added less than 1% to runtime in their repetition-code tests.

From error counts to component sensitivity

A component’s error frequency alone does not tell researchers how much it threatens a logical qubit. A common fault may be corrected successfully, while a less frequent event may leave the decoder with the wrong logical correction. FaultScope instead estimates how the logical failure rate changes as the error probability assigned to each event changes.

The paper calls that quantity sensitivity. For a given event, its estimate compares the chance of logical failure in sampled shots where the event occurs with the chance in shots where it does not, averaging over the other sampled errors. Researchers can combine event sensitivities for a physical component and group results by qubit, gate, syndrome-extraction round or operation to build a hotspot map.

One batch of shots supplies the map

FaultScope computes sensitivities from the same sampled and decoded shots used to estimate the logical error rate. It does not require changes to the decoder or differentiation through the decoder, the authors write. The software supports two simulation routes: forward simulation of Clifford circuits and direct sampling from detector error models.

The implementation is written in Rust and provides a Python interface. The paper describes the package as open source and says its benchmark scripts will be included in the project repository. The method is designed to add diagnostic information to a QEC simulation rather than substitute for the decoder or for measurements of a physical device.

CPU tests reach 1,024 qubits and code distance 100

For random Clifford circuits with 64 layers, the researchers compared six CPU simulators using one million shots per circuit. On the largest tested circuits, with 1,024 qubits, FaultScope’s median runtime was 2.931 seconds, against 5.182 seconds for Stim; across three matched instances, the median speedup was 1.70 times. Those timings include circuit preparation and processing of the sampled outcomes, while excluding software installation and warm-up.

A second benchmark simulated rotated surface-code memories with circuit-level Pauli noise at probability 10-3. FaultScope completed all 12 tested code distances, from 2 to 100; it was the only simulator in the comparison to finish the distance-100 case, taking 2,043.1 seconds. That circuit contained 19,999 qubits, 1,009,900 measurements per shot and 3.96 million CNOT operations. The benchmark ran on CPU backends, and the authors exclude timeouts and errors from numerical runtime comparisons.

To test attribution overhead, the team used repetition-code circuits with 7 to 2,047 error locations and 100,000 shots per evaluation. Computing every sensitivity took 0.33% to 0.97% longer than estimating the logical error rate without recording error events. The study’s own qualification is important: when logical failures are rare, a fixed shot budget may not resolve sensitivity differences well enough to choose between noise-reduction targets.

Sources

  1. FaultScope: A software package for fast error attribution in quantum error correction
  2. arXiv (full text)

Related stories

Explore

More articles