Generalized modeling approach for waveguide meshes enables reliable auto-configuration of complex applications.

Abstract

As a highly promising reconfigurable platform for optical computing, waveguide meshes have been extensively studied for various applications. However, most existing methods are limited to optical routing, while others rely on unrealistic assumptions or impose constraints that lead to incorrect configuration and performance deviation. These limitations have hindered the widespread adoption and commercialization of waveguide meshes. In this work, we propose a reliable and general modeling approach for waveguide meshes by accelerating the modeling of the tunable basic unit (TBU) to quadratic time. This TBU model is reusable and can be assembled to model arbitrary mesh geometries composed of identical TBUs. Unlike existing models, our approach accurately incorporates device dispersion, inter-component reflections, and losses, enabling high-precision configuration and eliminating parasitic circuits and phase fluctuations. The mesh model supports the automatic configuration of complex applications by integrating optimization algorithms with carefully designed error functions tailored to target functionalities. We demonstrate its effectiveness in configuring multiple functions, including optical routing, phase control, unitary matrices, complex filtering, and wavelength-division multiplexing (WDM). Compared to current techniques, our method exhibits stronger generalization capability and higher reliability, unlocking the broader potential of waveguide meshes. Furthermore, when integrated with an adjoint gradient method, it is expected to achieve even higher efficiency in automatic configuration.