Towards Power Characterization of FPGA Architectures To Enable Open-Source Power Estimation Using Micro-Benchmarks
Abstract
While in the past decade there has been significant progress in open-source synthesis and verification tools and flows, one piece is still missing in the open-source design automation ecosystem: a tool to estimate the power consumption of a design on specific target technologies. We discuss a work-in-progress method to characterize target technologies using generic micro-benchmarks, whose results can be used to establish power models of these target technologies. These models can further be used to predict the power consumption of a design in a given use case scenario (which is currently out of scope). We demonstrate our characterization method on the publicly documented Lattice iCE40 FPGA technology, and discuss two approaches to generating micro-benchmarks which consume power in the target device: simple lookup table (LUT) instantiation, and a more sophisticated instantiation of ring oscillators. We study three approaches to stimulate the implemented micro-benchmarks in hardware: Verilog testbenches, micro-controller testbenches, and pseudo-random linear-feedback-shift-register-(LFSR)-based testing. We measure the power consumption of the stimulated target devices. Our ultimate goal is to automate power measurements for technology characterization; Currently, we manually measure the consumed power at three shunt resistors using an oscilloscope. Preliminary results indicate that we are able to induce variable power consumption in target devices; However, the sensitivity of the power characterization is still too low to build expressive power estimation models.
Cite
@article{arxiv.2304.05326,
title = {Towards Power Characterization of FPGA Architectures To Enable Open-Source Power Estimation Using Micro-Benchmarks},
author = {Stefan Riesenberger and Christian Krieg},
journal= {arXiv preprint arXiv:2304.05326},
year = {2023}
}
Comments
Presented at the 3rd Workshop on Open-Source Design Automation (OSDA), 2023 (arXiv:2303.18024)