Rapid energy estimation of computations on FPGA based soft processors

• Published in: IEEE International SOC Conference, 2004. Proceedings pp. 285 - 288
• Main Authors: Jingzhao Ou, Prasanna, V.K.
• Format: Conference Proceeding
• Language: English
• Published: Piscataway NJ IEEE 2004
• Subjects: Analytical models; Application software; Application specific integrated circuits; Applied sciences; Circuit properties; Design. Technologies. Operation analysis. Testing; Digital circuits; Electric, optical and optoelectronic circuits; Electronic circuits; Electronics; Embedded software; Energy dissipation; Estimation error; Exact sciences and technology; Field programmable gate arrays; Hardware; Integrated circuits; Integrated circuits by function (including memories and processors); Process design; Semiconductor electronics. Microelectronics. Optoelectronics. Solid state devices; Software performance; Performance evaluation; Estimation error; State of the art; Processor; Error estimation; Fast Fourier transformation; Estimator; Field programmable gate array; System on a chip; Energy dissipation; Energetic efficiency; Integrated circuit; Software
• ISBN: 9780780384453; 0780384458
• DOI: 10.1109/SOCC.2004.1362437

FPGA based soft processors are an attractive option for implementing embedded applications. As energy efficiency has become a key performance metric, techniques that can quickly and accurately obtain the energy performance of these soft processors are needed. While low-level simulation based on traditional FPGA design flow is too time consuming for obtaining such energy performance, we propose a methodology based on instruction level energy profiling. We first analyze the energy dissipation of various instructions. An energy estimator is built using this information. To illustrate the effectiveness of our approach, the energy performance of several FFT and matrix multiplication software programs running on a state-of-the-art soft processor is evaluated using the estimator. Compared with the results obtained through low-level simulation, an average estimation error of 5.9% is observed in our experiments.