A machine learning approach for power gating the FPGA routing network
Power gating is a common approach for reducing circuit static power consumption. In FPGAs, resources that dominate static power consumption lie in the routing network. Researchers have proposed several heuristics for clustering multiplexers in the routing network into power-gating regions. In this paper, we propose a fundamentally different approach based on K-means clustering, an algorithm commonly used in machine learning. Experimental results on Titan benchmarks and Stratix-IV FPGA architecture show that our proposed clustering algorithms outperform the state of the art. For example, for 32 power-gating regions in FPGA routing switch matrices, we achieve (on average) almost 1.4× higher savings (37.48% vs. 26.94%) in the static power consumption of the FPGA routing resources at lower area overhead than the most efficient heuristic published so far.
Seifoori19 A machine learning approach for power gating the FPGA routing network.pdf
Preprint
openaccess
792.38 KB
Adobe PDF
73fef4023891ada9f4532ca24a116e4d