Advancing Photonic Inverse Design with Interpretable Machine Learning

Lirandë Pira, Airin Antony, Nayanthara Prathap, Jamika Ann Roque, Daniel Peace, Jacquiline Romero

Photonic chip design has in recent years seen significant advancements with the adoption of inverse design methodologies largelyenabled by the increasing computational efficiency of electromagnetic solvers. However, the often black‐box nature of this optimization method presents new challenges in understanding their outputs, particularly in the case of topological inverse design implementations. This challenge is prevalent in machine learning–based optimization methods, which can suffer from the same lack of transparency. To this end, interpretability techniques address the opacity of optimization models. In this work, we apply interpretability techniques from machine learning, with the aim of gaining understanding of inverse design optimization used in designing photonic components, for the specific case of a two‐mode (de)multiplexer. We base our methodology on the widely used interpretability technique known as local interpretable model‐agnostic explanations, or LIME. As a result, LIME‐informed insights point us to more effective initial conditions, directly improving device performance. This demonstrates that interpretability methods can do more than explain models—they can actively guide and enhance the inverse‐designed photonic components. Our results demonstrate the ability of interpretable techniques to reveal underlying patterns in the inverse design process, leading to the development of better‐performing components.