Energy-efficient processor system design.

摘要

Motivated by the pervasive use of general-purpose processors in portable electronics devices, energy-efficient processor system design is presented as a critical enabler for smaller, more powerful, and longer-running devices. A decade of research has demonstrated that extremely energy-efficient ASIC and custom DSP design is achievable, but the energy-efficiency of general-purpose processors has severely lagged behind. Thus, despite only performing a small fraction of the computation in these portable devices, the processor system contributes a significant, if not dominant, fraction of the device's total energy consumption. This thesis introduces and demonstrates a top-down processor system design methodology for dramatically reducing energy consumption, while maintaining the desired level of performance.; By understanding the fundamental usage requirements of a processor in portable devices, combined with analytical models for energy consumption and performance, energy-efficiency metrics are derived. A key design technique derived from these metrics, dynamic voltage scaling, is then described for achieving the single-largest increase in energy-efficiency. The metrics are further utilized in developing an overall energy-conscious design flow methodology, and more specifically, energy-efficient architectural and circuit design methodologies to additionally improve system energy-efficiency.; The design and measured results are reported on a prototype processor system, which successfully demonstrates the design techniques and methodologies presented in this thesis, and the potential improvement in processor system energy-efficiency. The system consists of four custom chips: a microprocessor, an SRAM, a voltage converter, and an I/O interface chip. On top of this system runs a real-time operating system, which executes software programs typically found in portable devices, to demonstrate a complete embedded processor system. This prototype system's energy-efficiency was quantified, and demonstrated to be more than order of magnitude higher than the most energy-efficient processor system available today.