Continuous Sensing on Intermittent Power

Abstract

Battery-free energy-harvesting devices have the potential to operate for decades, since they draw power fromvirtually unlimited energy sources, such as sunlight. However, ambient energy sources are volatile, and tiny harvesters can extract only weak power from them. Thus, small energy-harvesting devices operate intermittently: first, they charge their buffers then start operating, which depletes the buffered energy and causes the devices to power down, letting the harvesters to refill the energy buffers for the next operational round. Classical programming architectures assume continuous power. Therefore, frequent power failures render them useless; power failures reset the computational progress and delete volatile data. Thus, the intermittent programming and execution paradigm has emerged. Generally, there are two strategies being employed to support intermittent execution: checkpoint-based and task-based. Prior checkpoint- and task-based systems tackled mainly challenges related to enabling efficient computing on intermittent power. However, they have ignored the challenges associated with sensing, which is the primary application for intermittent systems. Therefore, from a sensing standpoint, these systems have several drawbacks.