Multilayer Modeling and Design of Energy Managed Microsystems

• Main Author: Zarrabi, Houman
• Format: Others
• Published: 2011
• Online Access: http://spectrum.library.concordia.ca/7371/1/Zarrabi_PhD_S2011.pdf; Zarrabi, Houman <http://spectrum.library.concordia.ca/view/creators/Zarrabi=3AHouman=3A=3A.html> (2011) Multilayer Modeling and Design of Energy Managed Microsystems. PhD thesis, Concordia University.

Aggressive energy reduction is one of the key technological challenges that all segments of the semiconductor industry have encountered in the past few years. In addition, the notion of environmental awareness and designing “green” products is yet another major driver for ultra low energy design of electronic systems. Energy management is one of the unique solutions that can address the simultaneous requirements of high-performance, (ultra) low energy and greenness in many classes of computing systems; including high-performance, embedded and wireless. These considerations motivate the focus of this dissertation on the energy efficiency improvement of Energy Managed Microsystems (EMM or EM2). The aim is to maximize the energy efficiency and/or the operational lifetime of these systems. In this thesis we propose solutions that are applicable to many classes of computing systems including high-performance and mobile computing systems. These solutions contribute to make such technologies “greener”. The proposed solutions are multilayer, since they belong to, and may be applicable to, multiple design abstraction layers. The proposed solutions are orthogonal to each other, and if deployed simultaneously in a vertical system integration approach, when possible, the net benefit may be as large as the multiplication of the individual benefits. At high-level, this thesis initially focuses on the modeling and design of interconnections for EM2. For this purpose, a design flow has been proposed for interconnections in EM2. This flow allows designing interconnects with minimum energy requirements that meet all the considered performance objectives, in all specified system operating states. Later, models for energy performance estimation of EM2 are proposed. By energy performance, we refer to the improvements of energy savings of the computing platforms, obtained when some enhancements are applied to those platforms. These models are based on the components of the application profile. The adopted method is inspired by Amdahl’s law, which is driven by the fact that ‘energy’ is ‘additive’, as ‘time’ is ‘additive’. These models can be used for the design space exploration of EM2. The proposed models are high-level and therefore they are easy to use and show fair accuracy, 9.1% error on average, when compared to the results of the implemented benchmarks. Finally, models to estimate energy consumption of EM2 according to their “activity” are proposed. By “activity” we mean the rate at which EM2 perform a set of predefined application functions. Good estimations of energy requirements are very useful when designing and managing the EM2 activity, in order to extend their battery lifetime. The study of the proposed models on some Wireless Sensor Network (WSN) application benchmark confirms a fair accuracy for the energy estimation models, 3% error on average on the considered benchmarks.