Interpretable geometry sensitivity for inverse design of integrated photonics

As an increasingly powerful technique in integrated photonics, inverse design uses optimization algorithms to automatically create compact, high-performance photonic structures, often yielding non-intuitive layouts far more compact than conventional designs [J. S. Jensen and O. Sigmund, Laser Photonics Rev. 5, 308 (2011) and S. Molesky et al., Nat. Photonics 12, 659 (2018)]. While adjoint-based inverse design is a prominent optimization method, the resulting free-form layouts are difficult to interpret or diagnose under fabrication variability [C. M. Lalau-Keraly et al., Opt. Express 21, 21693 (2013); L. Su et al., arXiv:1910.04829 [physics.app-ph] (2019); and A. Y. Piggott et al., Nat. Photonics 9, 374 (2015)]. We present an experimentally validated interpretability workflow that produces pixel-level sensitivity maps directly on the binary mask of an inverse-designed device. Using wavelength-division demultiplexers at 1310/1550 nm as examples, we train a lightweight convolutional surrogate to regress figures of merit and apply integrated gradients to attribute predicted transmission to individual pixels. We demonstrate that high-attribution hotspots correspond to physically meaningful substructures, such as splitter hubs and high-curvature edges. Experimental results show that controlled perturbations in these high-sensitivity regions result in up to an ∼11× higher excess insertion loss compared to perturbations in non-sensitive regions, consistent with full-wave simulations. This approach adds a practical explainability layer to existing pipelines, offering a clear pathway toward foundry-compatible design-rule checking, targeted quality control, localized metrology, and fabrication-aware constraint allocation without modifying the underlying electromagnetic solver.