Improving Energy Efficiency in Sensing Subsystems via Near-Threshold Computing and Device Aging

Near-threshold computing (NTC) has been shown to achieve maximal energy gains while maintaining performance metrics more favorable than subthreshold operation of circuits. Sensor subsystems are perfect candidates for NTC since low energy is often a crucial sensor system design desideratum and speed is rarely required. However, process variation affects threshold voltages (Vth), thus forcing higher-than and lowerthan desired supply voltage (Vdd) and ground voltage (Vgnd), respectively, and results in higher switching energy. We propose to use device aging to slow those gates on the lower tail of the Vth distribution in order to increase Vgnd and reduce overall energy consumption. We present a methodology for sensor subsystem aging and observe a reduction in energy consumption by 50% over traditional settings when applying NTC and a further 15% reduction by incorporating our sensor subsystem aging algorithms. Furthermore, we apply our developed approach for device aging at the slice-level granularity to minimize energy expenditure in FPGA systems. Through application of our techniques to simulations of FPGA designs we achieve a factor of up to 6× over traditional operation.