System-level power optimization

Analysis

Why This Paper Matters

This tutorial, published in 2000, remains a foundational reference for system-level power optimization. At a time when energy efficiency was becoming critical for portable and embedded systems, Benini and De Micheli provided a structured survey that unified hardware and software perspectives. The paper's organization around design phases—conceptualization, implementation, and runtime—offered a practical roadmap for engineers and researchers. Its high citation count (403) reflects its lasting influence on the field of energy-efficient computing.

Technical Contributions

The paper's key innovations include:

• Taxonomy of energy consumers: Identifies computation, communication, and storage as the three major hardware constituents consuming energy, and reviews reduction methods for each.
• Software energy models: Introduces models for analyzing the energy cost of software, bridging the gap between hardware and software design.
• Design phase framework: Structures techniques around three phases: conceptualization and modeling (e.g., power estimation), design and implementation (e.g., low-power synthesis), and runtime management (e.g., dynamic voltage scaling).
• Cross-layer approach: Emphasizes co-design of hardware and software for holistic energy savings.

Results

As a tutorial survey, the paper does not present new experimental results. Instead, it synthesizes existing techniques such as power gating, clock gating, and compiler optimizations. The value lies in its comprehensive categorization and clear exposition, which enabled subsequent work to build on these foundations.

Significance

This paper helped establish system-level power optimization as a distinct research area. By integrating hardware and software considerations, it influenced the development of energy-aware design tools and methodologies. Its framework remains relevant for modern challenges like IoT and mobile computing, where energy efficiency is paramount. The survey's emphasis on runtime management anticipated later advances in adaptive and dynamic power management.