An efficient and effective performance estimation method for DSE

Design Space Exploration (DSE) is a critical step in the chip design. The tradeoffs and interactions among parameters are traditionally evaluated by simulating or synthesizing a variety of designs which is intractable. The predictive modeling techniques have been applied to predict the design performance for DSE. For the system-on-a-chip (SoC) DSE cases, however, it is difficult to achieve high accuracy with previous methods due to their limitations. In this paper, we proposed a new estimation method based on Regression Random Forests (RRF) to build accurate and reliable prediction models. Our method can significantly improve the prediction accuracy and accelerate the procedure. In addition, due to the comprehensible tree model, RRF could guide the SoC design by ranking the parameters' importance. The experimental results show that RRF can reduce relative errors by about 60%, which demonstrate the effectiveness of our method.