Journal Title - IEEE Transactions on Circuits and SystemsAuthors - Arm: Anand Savanth, James Myers, David Flynn. University of Southampton: Alex Weddell, Bashir Al-Hashimi
Wireless sensor systems are expected to represent a large proportion of IoT devices. Powering a trillion wireless sensor nodes each using a small lithium coin cell would require approximately 109,000 metric tons of Lithium - nearly three times the annual worldwide production. This rather daunting number has motivated Arm Research, together with colleagues in the Arm-ECS Research Centre at the University of Southampton, to investigate compute systems that harvest energy from their surrounding environment.
Traditional energy harvesting systems feature two voltage converters. One converter is used to match the voltage produced by the energy source (e.g. solar cell, piezo electric) with the energy storage (battery, capacitor). Another converter is used to step between the energy storage and the system to be powered.
Conventional power conversion in energy harvesting systems 
The traditional system topology results in multiple conversion losses and also loses harvested energy in scenarios where the harvested energy is below the threshold needed to charge the energy storage device (for example when light levels are too low).
In this paper, Arm researcher Anand Savanth, with colleagues from Arm and the University of Southampton, explores an innovative approach to voltage conversion having a highly efficient integrated voltage converter with a novel selective direct mode of operation.
The voltage converter is integrated on-die with the compute system with no off-chip passives and achieves an 80% conversion efficiency for dim indoor lighting conditions - an improvement of 20% over previous proposals.
Comparison with prior-works for low-light efficiency (1µW)
The second contribution of the paper is a direct mode of operation which powers the compute system direct from the energy harvesting source. This mode allows energy from the solar cell to be effectively utilised in conditions where the ambient light is sufficient to power the compute system but insufficient to charge the storage cell.
The direct mode of operation is useful for two reasons. Firstly, by designing for sub-threshold operation, the compute system can operate at a lower voltage than the minimum voltage needed to charge the energy storage. Secondly, the output of an energy harvester will typically fluctuate through the day - for example with ambient light levels.
In Arm Research this concept has been tagged as "use it or lose it" and is shown conceptually in the graph below; the yellow portions of the graph represent conditions where the energy supplied by the harvester are below that needed to charge the storage device but are sufficient to power the compute system.
Conceptual representation of energy harvesting through converter-less operation of CPU system
In testing, the direct operation mode provided the system with 30% additional compute cycles during lower ambient light conditions experienced over a period of several days. Read on to find out more about how this mode was able to achieve this.
Read the whole paper