Off and On Again: Optimising Intermittently-Powered Systems
Powering billions of Internet of Things (IoT) nodes will be a significant challenge: batteries increase cost, environmental impact and maintenance.
Energy harvesting (EH) scavenges power from sources such as light, vibration, or temperature difference. EH sensor nodes are now being deployed with a battery topped up to smooth out temporal variations in EH and ensure reliable operation.
A recent competing approach proposes nodes with no energy storage, which only operate when power is available, maintaining their state in non-volatile memory (NVM) between intermittent power cycles. Such systems either use a purely software approach (taking regular program state checkpoints in NVM) or purely hardware (with all state being in NVM).
This project proposes to explore and demonstrate a HW/SW co-design strategy to maximise the amount of useful computation per power-cycle through optimising software structures and the use of NVM.
“This project explores the use of NVM for intermittent computing systems, as the industry moves towards one trillion IoT devices, many such devices will be intermittently powered, NVM provides the promise of enabling fast and efficient switch off and then on again, also as MCUs scale below 28nm, NVM provides a promising alternative to replacing eNOR flash and disrupting on-chip SRAM.”
William Wang – Staff Research Engineer