Every Microjoule Counts: Zero-Failure Task Execution in Batteryless Sensors

Existing intermittent computing approaches - such as checkpointing and atomic task models - struggle to guarantee task completion, often leading to energy waste and reduced Quality of Service. A common workaround is to oversize energy buffers to support worst-case task demands, which increases system latency and leads to inefficient hardware/energy utilization. This work introduces the first sensor node architecture that guarantees zero task failures and improves energy efficiency without relying on oversized storage. The system enforces a runtime energy-admissibility condition: tasks are executed only when the available stored energy exceeds their known cost. A hardware-software co-design, including a configurable storage capacitor array, dynamically manages energy storage and enables safe task dispatch. The proposed approach is validated through both simulation and real-world prototyping. Results show a 62.5% increase in task throughput, zero task failure, and increased Effective Energy Utilization from 69.55% to 96.26% compared to a baseline system.