草榴社区

GPU-Accelerated FullWAVE FDTD

One of the most exciting updates in the 2024.09 release is the introduction of GPU acceleration for FullWAVE FDTD. Utilizing the parallel processing capabilities of NVIDIA GPUs, the new feature dramatically speeds up simulations, with benchmarks showing 8 GPUs yielding 91 times faster performance compared to traditional 24-core CPU computations. To show parallel efficiency, we normalize the curve with a speedup factor as shown in Figure 1. This shows that the use of multiple GPUs scales well. With 8 GPUs, the speed is 6.5 times faster than our single-GPU simulation. This acceleration is particularly beneficial for large-scale simulations, enabling users to achieve results more quickly and efficiently. 

The acceleration in FullWAVE FDTD is supported by NVIDIA GPUs with the CUDA 12.3 or higher versions, providing a seamless integration for users with compatible hardware. This improvement not only enhances productivity but also opens new possibilities for research and development in photonics.

Improvements to BeamPROP BPM

The Beam Propagation Method (BPM) in BeamPROP has also seen significant improvements. The new release introduces automated power renormalization for full-vector BPM in high-index contrast structures, such as silicon photonics, to ensure accurate power calculations.

The automated renormalization process simplifies the workflow, allowing users to achieve stable and reliable results with minimal manual intervention. By adjusting the scheme parameters to suppress numerical instabilities, the updated BPM solver automatically renormalizes the power and eliminates artificial numerical loss to provide accurate results in a wide range of complex photonic structures. 

Improvements to MetaOptic Designer

MetaOptic Designer now includes a new feature for far-field optimization. This addition allows users to map the far-field screen to a near field at a focal point using a thin lens model. The far-field optimization is particularly useful for applications requiring precise control over the far-field pattern, such as LIDAR and holographic displays.

Users can specify design parameters, including target screen distance and metalens size, to fine-tune their designs. This flexibility enables more accurate and efficient design processes for metasurfaces and metalenses. Additionally, the improvements in controlling the simulation grid size and domain enhance the accuracy and flexibility of simulations, allowing for better optimization of complex optical elements.

Improvements to DiffractMOD RCWA

DiffractMOD RCWA now supports non-orthogonal simulation domains, providing more efficient simulations for hexagonal and other non-orthogonal lattice structures. This update allows users to define custom lattice vectors and angles, improving the accuracy and convergence of simulations for complex periodic structures.