波動信号処理研究室│日本大学理工学部電気工学科 大貫研究室

JAPANESEENGLISH

研究紹介

コンピュータシミュレーション技術の高速化と高精度化

世界最速のスーパーコンピュータで利用されるCell Broadband Engine(Cell/B.E.)や,安価かつ超高速並列演算性能が注目されているGraphics Processing Unit (GPU)を用いて,電磁界シミュレーションの高速化を検討している. 端点を有する形状に対して,高精度な電磁界解析が可能な数値計算法を開発している.開発手法は有効桁数8桁程度と信頼性が高く,計算精度の予測と計算時間の最適化が可能である.

コンピュータシミュレーション技術の高速化と高精度化

Fast Electromagnetic Simulations using Ultra High-Speed Parallel Processors

We have studied potential capabilities of hardware acceleration for electromagnetic simulations using ultra high-speed parallel processors, such as a Cell Broadband Engine (Cell/B.E.) processor and a Graphics Processing Unit (GPU). The Cell/B.E. is one of the many-core CPUs. It consists of one power processing element and eight synergistic processing elements. This architecture is suitable for wide variety of programming models. GPUs are specialized processors to deal with 3D graphics rendering substituted for conventional CPUs. They have many computing cores which can run simultaneously. Figure 1 shows the hardware acceleration of the method of moments (MoM) using the Cell/B.E. processor. The operation can be about 150 times faster than that using a conventional CPU. Figure 2 shows the computational time of an electromagnetic simulation performed by the finite-difference time-domain (FDTD) method using a GPU. To distribute computational processes into many cores, the operation can be about 10 times faster than that using a CPU.

コンピュータシミュレーション技術の高速化と高精度化
PAGETOP