FDTDX: High-Performance Open-Source FDTD Simulation with Automatic Differentiation

The ability to precisely manipulate the propagation of light is an increasingly critical factor in many modern applications.Aside from analytical solutions for simple geometries, the majority of practical problems require robust numerical methods such as the finite-difference time-domain (FDTD) method for discretizing Maxwell's equations in both space and time.We present FDTDX, an efficient implementation of the FDTD method with GPU acceleration through the JAX framework.It provides a simple user interface for specifying a simulation scene as well as a suite of tools for inverse design.Using the time reversibility of Maxwell's equations, gradients for optimizing the geometry of a design can be computated efficiently within FDTDX.