Current development of extremely low-power computing devices and efficient energy harvesters led to the creation of computing systems that are powered by intermittently available harvested energy, e.g., solar, piezoelectric, and radio-frequency (RF). Such computing systems go through power-on and off phases due to the lack of adequate harvesting energy. These systems are known as Intermittent Computing Systems. While existing works on intermittent computing systems concentrate preliminary on the lower level goals, e.g., execution progress and memory consistency, the potential of such systems under timing constraints is yet to be explored. Some applications of intermittent systems with timing constraints include monitoring wildlife, health, infrastructure and environmental conditions, pedestrian safety, indoor localization and occupancy detection. We focus on the timely-response and learning capability of intermittent systems by (1) developing unified frameworks that integrate harvesting and real-time systems, (2) engineering machine learning algorithms providing learning capabilities to this intermittent systems with timing-constraints, and (3) designing system framework for life-long learning.
Publications
PhD Forum Abstract: Scheduling Tasks on Intermittently Powered Systems, IPSN ‘20
WiP: Time-Aware Deep Intelligence on Batteryless Systems, RTAS ‘19
Poster Abstract: On-Device Training from Sensor Data on Batteryless Platforms, IPSN ‘19