Reducing Control Power in CGRAs with Token Flow

Autonomous wireless devices for healthcare monitoring become a reality only if such devices are embedded with enough intelligent and processing capabilities to minimize the amount of data being transferred through the wireless network. Also, the embedded processing capability must be made energy efficient so that the device can operate on scavenged energy or very limited battery power. This paper reports the development of a real-time EEG (Electroencephalography) application based on DWT (Discrete Wavelet Transform) and its mapping and optimization on an Application Specific Instruction set Processor (ASIP). It shows the drastic energy reduction that can be achieved by crossoptimization of the algorithm and the ASIP architecture. Our results indicate that such cross-optimization can reduced the dynamic energy by more than 80%.