CROSS-LAYER CUSTOMIZATION PLATFORM FOR LOW-POWER AND REAL-TIME EMBEDDED APPLICATIONS
MetadataShow full item record
Modern embedded applications have become increasingly complex and diverse in their functionalities and requirements. Data processing, communication and multimedia signal processing, real-time control and various other functionalities can often need to be implemented on the same System-on-Chip(SOC) platform. The significant power constraints and real-time guarantee requirements of these applications have become significant obstacles for the traditional embedded system design methodologies. The general-purpose computing microarchitectures of these platforms are designed to achieve good performance on average, which is far from optimal for any particular application. The system must always assume worst-case scenarios, which results in significant power inefficiencies and resource under-utilization. This dissertation introduces a cross-layer application-customizable embedded platform, which dynamically exploits application information and fine-tunes system components at system software and hardware layers. This is achieved with the close cooperation and seamless integration of the compiler, the operating system, and the hardware architecture. The compiler is responsible for extracting application regularities through static and profile-based analysis. The relevant application knowledge is propagated and utilized at run-time across the system layers through the judiciously introduced reconfigurability at both OS and hardware layers. The introduced framework comprehensively covers the fundamental subsystems of memory management and multi-tasking execution control.