研究実績の概要 |
We performed studies on High-speed ultra-low-power simulation methodology for material science. The main goal of the study is to establish technologies for high-speed computation, high-speed network and high-speed memory systems to accelerate simulations in material science area. Our focus is the use of FPGA (field programmable gate array) for accelerating computation, network operation and memory access acceleration. During 2010 ~ 2014 periods, we have constructed two computation accelerator, two network accelerator and several other FPGA board to perform network and memory emulation. As the results of these studies, we have established basic architecture for the FPGA based platform for simulation in wide application area including material science. We also adopt genetic algorithm (GA) to optimize heterogeneous networks that maximize performance and power efficiency. In addition to such hardware systems, we also developed HPC ruby system that consists of static HPC ruby compiler, type inference system and runtime systems for application programmers. Another important contribution of this project is establishing a new educational unit, “Computing Science Educational Course(学際理学講座 計算物理)” as a joint activity between the department of computer science, Graduate School of Information Science and Technology(情報理工学系研究科コンピュータ科学専攻) and the department of physics, Graduate School of Science(理学系研究科物理学専攻) at the University of Tokyo. This educational course is a result of collaboration in this project.
|