Autonomous hardware/software partitioning and voltage/frequency scaling for low-power embedded systems

Warp processing is a recent computing technology capable of autonomously partitioning the critical kernels within an executing software application to hardware circuits implemented within an on-chip FPGA. While previous performance-driven warp processing has been shown to provide significant performance improvements over software only execution, the dynamic performance improvement of warp processors may be lost for certain application domains, such as real-time systems. Alternatively, as power consumption continue to become a dominant design constraint, we present and thoroughly analyze a low-power warp processing methodology that leverages voltage and/or frequency scaling to substantially reduce power consumption without any performance degradationall without requiring designer effort beyond the initial software development.