Power Reduction of Functional Units considering Temperature and Process Variations

Deepa Kannan, Aviral Shrivastava, Sarvesh Bharadwaj, Sarma Vrudhula,

VLSI 2008: Proceedings of the 21st International Conference on VLSI Design

Abstract: Continuous technology scaling has resulted in an increase in both, the power density as well as the variation in device dimensions (process variations) of the manufactured processors. Both power density and process variations have a significant impact on the leakage power. As processors are fabricated in nano scale technology, leakage power has become a dominant factor in total power consumption due to its exponential dependence on temperature and device dimensions. Thus even small variations in temperature and device dimensions lead to significant variations in the leakage power and in turn the total power consumption across dies. Therefore, power optimization techniques should be sensitive to the variation in leakage power due to both temperature as well as process variations. Operation to Functional Units Binding Mechanism (OFBM) is the mechanism to dynamically issue operations to Functional Units (FUs) in superscalar processors. Existing OFBMs are unaware of the process variations and the variations in temperature, and are therefore unable to effectively reduce the leakage and the variation in leakage of FUs. In this paper, we propose a Leakage-Aware OFBM (LAOFBM), which is both temperature and process variation aware. Our experimental results demostrate that LA-OFBM reduces the mean and standard deviation of the total energy consumption of ALUs by 18%, and 46% respectively, as compared to the traditional OFBM, without any performance penalty.