近年、生活の様々な場面で高速な認識システムが必要とされている.そこで、我々はDDI（Direct Data Implementation）と呼ぶ手法を提案している.DDIは既知のパターンを直接論理回路として埋め込むことでサンプルパターンと同数の並列度を有するパターン認識専用回路を構成できる手法である.本報告では、DDIの消費電力評価を行った.FPGA（Field Programmable Gate Amay）を用いたシステムは周辺回路が存在するため、FPGA単体の消費電力を正確に測定することは難しい.そこで本報告では、専用の測定基板を作成し基本的な回路でFPGA単体の消費電力が測定できることを確認し、DDIの消費電力測定を行った.その結果、DDIは従来の逐次処理によるパターン認識回路に比べてパターン認識処理に必要なエネルギーが約280分の1であることがわかった.%Recently, high-speed recognition systems are required in various situations in life. In order to reply the requirements, we have proposed a novel architecture called "Direct Data Implementation (DDI)". DDI is composed of dedicated circuits that have the same parallelism as the sample patterns by embedding known patterns directly to logic circuits. In this paper, we evaluated the power consumption of DDI. Systems using FPGAs have the peripheral circuits. Therefore, it is difficult to measure the power consumption of FPGAs accurately. In this paper, we have developed a dedicated measurement board and showed that the board can measure power consumption of the basis circuits using FPGAs precisely. We measured the power consumption of DDI implemented in the FPGAs on the board. As a result, we have found that DDI only requires about 1/280 of the energy that is required in the traditional sequential pattern recognition circuit.
展开▼
