نبذة مختصرة : IoT systems can operate efficiently in scenarios with limited or sporadic power availability by utilizing intermittent power sources, such as energy harvesting. Typical implementations are based on fixed-size super-capacitors as energy storage. However, this limits the possible implementations. Large capacitors take more time to charge, resulting in extended off-time following a power failure. Small capacitors charge faster but provide a shorter active time, leading to more frequent power failures. This paper presents a fully intermittent, machine learning-based, low-power smart camera for monitoring applications integrating a two-stage energy harvester. A small first stage supports data acquisition and analysis. The second bigger energy storage is activated only for data streaming to support the integrated sub-GHz Low Earth Orbit transmission radio. This dual-stage energy storage strategy ensures both system reactivity and the capacity to sustain energy-intensive data transmission. The simulation highlights how the fully intermittent pipeline (i.e., for both the neural accelerator and the ARM core) makes the system resilient to power fluctuation and increases the throughput of processed images in ultra-low light conditions by up to 13%.
Processing Request
No Comments.