SimWorks Finite Difference Solutions
The Eigenmode Expansion (EME) solver divides a long waveguide structure into multiple cells along the propagation direction, solves for the eigenmodes of the cross-section in each cell, and then computes the coupling of these modes between adjacent cells to obtain the complete optical characteristics of the entire device.
This update brings two major enhancements to SimWorks' Slurm cluster resource management: support for file transfer via shared folders, and the ability to submit Slurm jobs directly from a local machine (Linux only).
Adds a new rectilinear grid dataset type rectilineardataset, providing more flexibility for data storage and processing.
Focus On
SimWorks Finite Difference Solutions
With the rapid development of high-performance computing technology, computational electromagnetics has made significant progress, making it possible to numerically solve complex electromagnetic field problems using computers. SimWorks timely introduced Finite Difference Solutions, a software with an intuitive interface that creates virtual experiments, reproduces complex micro-nano optoelectronic phenomena, predicts unknown optoelectronic behaviors, analyzes and optimizes complex structures or materials, and provides users with a complete professional numerical solution for optoelectronic problems.
Learn MoreFDTD
SimWorks FDTD is a powerful tool for researchers and engineers to handle various micro-nano optoelectronic problems.
FDFD
SimWorks FDFD is a powerful tool for analyzing the spectrum of resonant cavities and metal antennas.
FDE
SimWorks FDE is a powerful tool for solving large-scale integrated planar optical waveguides, long-distance transmission devices, and various new fiber optic problems.
Cloud Computing
Pay-per-use, private deployment, providing 24/7 uninterrupted service
Parallel Mode
Supports various parallel solutions including GPU, MPI, OpenMPI, CUDA
System
Compatible with three major operating systems: Windows, Linux, and macOS
Computing Speed
Simulation computing speed is about 50% higher than other mainstream products in the industry
Computational Accuracy
The computational accuracy of the product solver is completely consistent with world-class software levels
Software Features
Software Functionality/Performance
SimWorks deeply applies GPU acceleration technology, fully leveraging the performance advantages of hardware and significantly enhancing the simulation speed. In typical optical simulation scenarios, accelerated computing solutions based on mainstream GPU architectures demonstrate performance far exceeding that of traditional CPUs. SimWorks provides comparative test data of CPU and GPU simulation speeds, which can help users evaluate and select the computing resource configuration that suits their own needs, achieving the optimal balance between performance and cost. In addition, SimWorks also supports multi-GPU parallel computing, allowing multiple Gpus to be scheduled to work collaboratively in a single simulation task. Compared with single-card computing, multi-card parallelism not only doubles the computing speed but also expands the video memory capacity, supporting larger-scale simulation. Whether it is a single-machine multi-card configuration or a multi-machine multi-card cluster, SimWorks can fully utilize multi-GPU resources to achieve more efficient computing and meet the needs of simulation tasks of different scales.The following comparison chart illustrates the acceleration effect of a single GPU over a CPU server (Chart 1), as well as the further speed improvement of multi-GPU parallelism over a single GPU (Chart 2).
When the FDTD solver runs on NVIDIA GPUs, it now supports half-precision (FP16) computation. Compared to single-precision (FP32), FP16 significantly reduces memory usage and improves computational efficiency. Arithmetic FP16 operations only supported on arch greater than 5.3 (NVIDIA GeForce GTX 10 series (Pascal architecture) or newer GPUs). The numerical range and precision of FP16 are much smaller than FP32, so do not use it in projects with large energy or nonlinearity. For more details, please refer to Advanced. For example, on an NVIDIA Tesla V100 GPU, where the theoretical FP16 performance is twice that of FP32, using FP16 in actual FDTD simulations achieves approximately 35% faster computation (Chart 3) and 30% lower memory usage (Chart 4) compared to FP32. The detailed comparison is shown below.
