A Structural Customization Approach for Low Power Embedded Systems Design

Power consumption is a critical design issue in embedded systems. This paper presents a structural customization approach, targeting two basic design structures for a multiple choice function that is common in an embedded system: tree structure and chain structure. A theoretical comparison between these two structures is performed. We find that the chain structure has a lower area cost and offers more customization opportunities, for improving performance and power efficiency, than the tree structure. We then propose two algorithms to reduce power consumption of the chain design for different performance constraints. A case study on ALU (Arithmetic and Logic Unit) customization has been carried out, Our experiments show that up to 50 percent of ALU power consumption can be saved with our customization approach.