BEEF: A Hybrid Integrated Photonic Simulation Framework

Abstract

A Python framework is presented that retains first-principles electromagnetic modeling while extending simulation to devices whose length and geometric complexity place them beyond the practical reach of either method alone. The Bridged Engine for eigenmode expansion (EME) and finite-difference time-domain (FDTD) (BEEF) partitions a device into sections which are simulated using either EME or FDTD independently developed Maxwell solvers depending on what best suits each section's geometry, and cascades the resulting responses. BEEF calibrates the EME propagation constants against FDTD-measured phase accumulation, providing a mode identity verification diagnostic across solver interfaces. For a 2x2 multimode-interference splitter, the hybrid result agrees with full-device FDTD to within 0.59% (bar) and 0.76% (cross) across the C-band; a tapered variant agrees to within 0.87% and 2.01%, respectively. For a directional coupler, effective-index calibration reduces cut-position dependence 167-fold. A circuit-level cascade reproduces the measured Fabry-Perot resonance lineshape and free spectral range while exploring the coupler-length and arm-length design plane through post-processing of four cached simulations. Finally, we use BEEF to simulate a 167.3 um-long 4x4 MMI-based 90-degree optical hybrid that is computationally impractical for full-device FDTD, with a reduced-length variant confirming close agreement in transmission and relative phase.