Thermo-Optic Fault Detection Benchmark for Silicon Microring Resonators

This software package provides a reproducible benchmark framework for evaluating sequential change detection algorithms applied to thermo-optic fault monitoring in silicon microring resonators.

The framework includes: (1) a physically calibrated first-order compact thermal model (k₀ = 72.9 pm/°C for Si SOI; k₀ = 11.7 pm/mW for 4H-SiC, calibrated to Sun et al. arXiv:2506.15035); (2) a realistic colored-noise simulation (Gaussian + 1/f + random-walk drift); (3) implementations of three primary sequential change detection methods — fixed threshold, CUSUM, and the W-formula — with an additional windowed GLR detector evaluated separately via a window-size sweep (see the manuscript's Discussion section for why GLR is not part of the main comparison); and (4) four synthetic fault scenarios — heater efficiency loss, thermal time constant increase, thermal drift, and contact degradation.

Detectors are compared via full Delay–False Alarm Rate operating frontiers obtained by threshold sweep, rather than at single operating points. Simulation-based results are complemented by a real-signal validation using a genuine oscilloscope transient (see experimental_validation_corrected.md and data/STEP-20DB-V0.4-*.csv), provided by Q. Li (Carnegie Mellon University).

The W = Q·D − T formulation is an empirical heuristic; the variable definitions are domain-specific adaptations. No theoretical derivation from first principles is claimed.

Repository structure

thermal_model_v3.py                    Core simulation, detectors, Monte Carlo sweep
digital_twin_sic.py                    SiC digital-twin model
run_real_detectors.py                  Runs the three detectors on the real transient
seven_detectors.py                     Earlier detector-count exploration (superseded; kept for history)
glr_sweep_exact.py                     Windowed GLR window-size sweep
experimental_validation_corrected.md   Write-up of the real-transient validation
microring_manuscript_v4_corrected.md   Corrected manuscript text
data/                                  Real oscilloscope data + digitized figure data
figures/                                Generated result figures
tests/test_detectors.py                Test suite (pytest)

Installation and usage

See INSTALL.md for setup instructions and CONTRIBUTING.md for how to propose changes.

Dataset attribution

Raw oscilloscope data kindly provided by Q. Li (Carnegie Mellon University), corresponding to Fig. 2(e) of Sun et al., arXiv:2506.15035.

Keywords

microring resonator, silicon photonics, sequential change detection, CUSUM, GLR, thermo-optic, fault detection, digital twin, health monitoring

Related work

This benchmark's simulation results are software-only; the accompanying manuscript places these results relative to established practice (industrial digital diagnostic monitoring, and the sequential-detection literature) — see the manuscript's Discussion section.

License

MIT License (see LICENSE).

Citation

If you use this software, please cite the Zenodo deposit: DOI: 10.5281/zenodo.21782606