Widhalm, D., Goeschka, K. M., & Kastner, W. (2022). Undervolting on wireless sensor nodes: a critical perspective. In Proceedings of the 23rd International Conference on Distributed Computing and Networking. 23rd International Conference on Distributed Computing and Networking, India. https://doi.org/10.1145/3491003.3491018
Data analytical services ultimately rely on the availability and quality of data provided by wireless sensor networks (WSNs) consisting of sensor nodes. These sensor nodes are usually battery-powered with limited or no possibility for recharging or replacement. Thus, energy efficiency is crucial to ensure a long lifetime. Undervolting of the sensor nodes with supply voltage levels even below the recommended voltage threshold has shown promising energy-saving potentials. Undervolting, however, comes with a challenge: different components of the sensor nodes face different threshold voltages and the on-chip interaction can cause soft-faults (silent data corruption) when only some of the components are operational. Such soft-faults are hard to detect and endanger the reliability of WSNs. In this paper, we consequently contribute with an analysis of the effects of undervolting by an extensive set of practical experiments with varying supply voltages under different ambient temperatures. The novelty of our approach results from our focus on Our experiments show that the energy-saving benefits of undervolting have to be balanced with the threat of soft-faults. We specifically show that the ideal voltage level has to be derived by checking all components in use, not only the ALU like in previous approaches.
Computer Engineering and Software-Intensive Systems: 80% Automation and Robotics: 20%