Controlling Uncertainty and Handling Variability in System-Level Dynamic Power Management

Project Summary: Variability represents diversity or heterogeneity in a well-characterized population. Fundamentally a property of Nature, variability is usually not reducible through further measurement or study. For example, different dies have different leakage power dissipations, no matter how carefully we measure them. Uncertainty represents partial ignorance or lack of perfect information about poorly-characterized phenomena or models. Fundamentally a property of the observer, uncertainty is usually reducible through further measurement or study. For example, even though an observer may not know the leakage power dissipation of every die coming out of a manufacturing plant, he or she can surely take more samples to gain additional (albeit still imperfect) information about the leakage power distribution. With the increasing levels of variability in the characteristics of nanoscale CMOS devices and VLSI interconnects and continued uncertainty in the operating conditions of VLSI circuits, achieving power efficiency and high performance in electronic systems under process, voltage, and temperature variations as well as current stress, device aging, and interconnect wear-out phenomena has become a daunting, yet vital, task. This research tackles the problem of system-level dynamic power management (DPM) in systems which are manufactured in nanoscale CMOS technologies and are operated under widely varying conditions over the lifetime of the system. Such systems are greatly affected by increasing levels of process variations typically materializing as random or systematic sources of variability in device and interconnect characteristics, and widely varying workloads and temperature fluctuations usually appearing as sources of uncertainty. At the system level this variability and uncertainty is beginning to undermine the effectiveness of traditional DPM approaches. It is thus critically important that we develop the mathematical basis and practical applications of a variability-aware, uncertainty-reducing DPM approach with the following unique features and capabilities.

Improving the Efficiency of Power Management Techniques by Using Bayesian Classification In an ISQED-08 paper, we presented a supervised learning based dynamic power management (DPM) framework for a multicore processor, where a power manager (PM) learns to predict the system performance state from some readily available input features (such as the state of service queue occupancy and the task arrival rate) and then uses this predicted state to look up the optimal power management action from a pre-computed policy lookup table. The motivation for utilizing supervised learning in the form of a Bayesian classifier is to reduce overhead of the PM which has to recurrently determine and issue voltage-frequency setting commands to each processor core in the system. Experimental results reveal that the proposed Bayesian classification based DPM technique ensures system-wide energy savings under rapidly and widely varying workloads.

Resilient Dynamic Power Management under Uncertainty In a DATE-08 paper, we presented a stochastic framework to improve the accuracy of decision making during dynamic power management, while considering manufacturing process and/or design induced uncertainties. More precisely, the uncertainties are captured by a partially observable semi-Markov decision process and the policy optimization problem is formulated as a mathematical program based on this model. Experimental results with a RISC processor in 65nm technology demonstrate the effectiveness of the technique and show that the proposed uncertainty-aware power management technique ensures system-wide energy savings under statistical circuit parameter variations.