情報処理学会 第78回全国大会講演要旨

Because of the power wall and the limitation of instruction-level parallelism techniques, the microprocessor industry has shifted to integrating multiple processor cores on a chip. Many-core processors have now become main stream. There already exist some processors with up to 100 cores on the market. And the core count is expected to grow up to thousands of cores. Unfortunately, conventional software simulators scale poorly, and thus can not provide a practical simulation time for designing such large-scale systems. In this study, we use FPGAs, which have been adopted in accelerating many computational tasks such as data processing in data centers, for fast and accurately emulating systems with hundreds to thousands of cores. We describe the techniques to overcome the capacity constraints of FPGAs while leveraging their high degree of parallelism to improve the simulation speed.