Adaptive Transient Computing for Power-Neutral Embedded Devices

Energy harvesting has emerged as a promising technology for small electronic devices to extend the battery run time and thereby enabling an increased autonomous operation. However, frequent charge and discharge cycles cause aging effects in the battery, which results in a loss of capacity and life time. Power-neutral transient computing systems avoid energy buffers by powering the load by the harvester directly. Usually, the power outputs of energy harvesters rely on arbitrary and transient environmental excitations. The resulting power losses are handled by checkpointing, where the volatile system state is backed up using non-volatile memories. The timely detection of upcoming power losses is essential for a reliable checkpointing process. Early detections allow a proactive power loss handling, which is important to ensure the finalization of atomic operations. However, common voltage threshold-based methods only allow short-term power loss detections since they do not adapt to the dynamics of the harvester. In this paper we propose a new methodology that allows an early power loss detection by exploiting physical characteristics of the harvester. The proposed approach points out new opportunities for transiently-powered devices, as it allows an adaptive and harvester-aware computing. We show how it facilitates a proactive scheduling that is used to ensure a successful finalization of atomic operations.