Post-Route Power Estimation: A Case Study of RIKEN-CGRA

In this paper, we propose a novel machine learning (ML)-based method to estimate the power consumption of Coarse-Grained Reconfigurable Array (CGRA) circuits post-routed. The proposed method allows us to predict the power consumption of post-route circuits only with the information obtained through the logic synthesis of some modules within CGRA, thereby reducing a very time-consuming phase needed to compute the power consumption of post-route CGRA circuits (i.e., placing-and-routing the overall CGRA circuit). Our experimental results show that the proposed method can estimate the power consumption of various RIKEN-CGRA circuits post-routed 6 times faster than the conventional method, with an average error of 11% (1% in the best case).