A New Method for Predictive Checkpointing in Transiently-Powered IoT Sensor Devices with Thermal Energy Harvesting

Abstract

Energy harvesting is a key technology towards a more sustainable Internet of Things (IoT), as it facilitates to operate IoT sensor devices without the need of batteries. However, energy harvesting-based power sources are transient, which is why such IoT devices need to be prepared for unpredictable power losses. This can be ensured by a power loss detection method that triggers a backup procedure just before a power loss. In this procedure, the volatile system data is saved in a nonvolatile memory (NVM), from which it is restored once the power returns (checkpointing). State-of-the-art detection methods monitor the supply voltage to detect upcoming power losses by a dropping voltage level. However, these methods detect power losses very late and thus leave only limited amounts of residual energy for the backup procedure. Therefore, it is very difficult to guarantee a reliable checkpointing process on energy-intensive flash memory, which is the state-of-the-art NVM technology on common IoT sensor devices.In this paper, we propose a new method for thermal energy harvesters, i.e., thermoelectric generators (TEGs), which allows a significantly earlier detection of potential power losses than related work. The core concept is to monitor the exciting quantity of the harvester, i.e., the temperature gradient across the surfaces of the TEG, to predict the future trend of the power output once this gradient starts to decline. This method exploits the thermal capacity of the TEG, which leads to a delayed temperature penetration from the surface into its internal, power generating structure. Our results show, that the shortfall below critical power output levels can be predicted several seconds in advance.