FDTDX

What FDTDX is

FDTDX is a free, open-source Python package for simulating and designing three-dimensional photonic nanostructures with the finite-difference time-domain method. Built on JAX, it provides automatic differentiation and gradient-based optimization, with GPU-accelerated simulations and multi-GPU scaling capabilities. The project says it can handle simulations with billions of grid cells and uses time-reversibility in Maxwell’s equations to compute gradients efficiently. Its API supports positioning and sizing objects with absolute or relative coordinates, importing GDS polygons, and working with GDS layer stacks and ports. FDTDX is available under the MIT license for research, development, and commercial use. Installation is documented with `pip install fdtdx`, and JAX acceleration options are listed for NVIDIA GPUs, AMD GPUs, and TPUs. Documentation includes interactive Jupyter quickstarts, physics tutorials, component examples, best-practice guidance, and an API reference. Community support is available through GitHub issues and discussions. The project’s FAQ describes it as relatively new and notes that bugs may occur despite testing. It is listed for Linux and self-hosted use.