LoLiPoP-IoT: Advancing the Energy-Efficient Internet of Things

This paper presents a portion of recent research outcomes from the LoLiPoP-IoT Chips JU project, which focuses on developing sustainable, long-life IoT platforms by integrating advanced energy harvesting, intelligent energy management strategies, and low-power HW/SW co-design techniques to optimize battery longevity with the intention of reducing the economic and ecological impacts of frequent battery replacements. The main objective of this research is to investigate how integrated energy harvesting, adaptive power management, and efficient data-processing techniques can significantly extend battery lifetime while maintaining performance and usability in real IoT deployments.

Unlike many existing studies that address isolated aspects of low-power IoT design, this work provides a comprehensive and practical approach that combines energy harvesting dimensioning, including simulation of the deployment environment, real HW power profiling, adaptive energy planning algorithms, predictive maintenance modeling, and their deployment on resource-constrained devices. The holistic integration of available technologies with newly designed approaches, such as dynamic energy scheduling, enables improvements in the overall IoT experience and a more sustainable usage.

Experimental results demonstrate several outcomes. The proposed dynamic energy planning framework, particularly the “Slope” algorithm, can extend battery lifetime by up to five times compared to baseline operation. If full energy autonomy is required, the photovoltaic panel area can be reduced by approximately 77 %. Our developed simulation toolkit enables accurate estimation of energy consumption and optimal sizing of photovoltaic harvesters, while predictive maintenance models based on statistical model checking enable forecasting fault probabilities of factory equipment based on collected data. Furthermore, we conducted experiments to confirm that optimized machine-learning models can achieve high accuracy with reduced memory footprint and inference time on embedded IoT platforms.