Statistical analysis and inverse design of programmable photonic integrated circuits using a Python-interfaced simulation framework

As photonic integrated circuits (PICs) scale toward high-density programmable architectures, system-level performance is increasingly governed by the accumulation of component-level loss and fabrication variability across large interferometric meshes. Therefore, efficient simulation workflows that connect verified component models to circuit-scale statistical behavior are essential. Here we present a Python-integrated automation framework for Ansys Lumerical INTERCONNECT that programmatically generates and analyzes arbitrarily large Clements meshes. Using Monte Carlo simulations, we quantify how insertion loss and coupling ratio variations impact transmission and extract an empirical scaling law for system loss that depends on both mesh size and per-component loss. We further observe a robust concentration of optical power in the central output ports for nominal 50:50 coupler settings, consistent with discrete diffraction and random-walk transport in the mesh. Finally, treating the simulator as a black-box forward model, we demonstrate inverse configuration using the Covariance Matrix Adaptation Evolution Strategy (CMA-ES), optimizing 8 × 8 and 12 × 12 meshes to match arbitrary target output-power distributions with a mean squared error of ∼ 1%. This framework enables scalable tolerance analysis and automated configuration for next-generation programmable photonics.