Reliable Energy-Aware Task Execution for Battery-free Wireless Sensors

Battery-free sensors operating under intermittent energy supply face intrinsic challenges in guaranteeing reliable task execution and efficient use of harvested energy. Existing intermittent computing approaches mitigate power failures but still rely on speculative execution or oversized energy buffers, leading to energy waste and limited responsiveness. This paper introduces an energy-admissible execution model for batteryless sensors, in which tasks are dispatched only when the available stored energy is sufficient to guarantee their completion. The approach is realized through a lightweight, energy-aware scheduler and a hardware–software co-design with a dynamically reconfigurable energy storage architecture, CapDYN, improving the fraction of harvested energy that contributes to successful task completion. The proposed model is implemented and evaluated on a battery-free sensing node for precision agriculture. The LTspice evaluation compares CapDYN against monolithic and ablation baselines using both an application-derived task set and a synthetic linear task set. CapDYN eliminates task failures, improves successful task completions by up to 41.67%, and reaches up to 97.17% and 97.46% effective efficiency on the two task sets, respectively. The results also show that CapDYN improves task completion timing when the workload can exploit intermediate storage stages, while the advantage decreases when most tasks map to the highest storage levels.