A Non-volatile State Retention Unit for Multi-storage Energy Management in Transient Systems

Abstract

Energy management in energy harvesting (EH) transient computing systems is challenging due to the common reliance on volatile memory (VM) elements, which require the energy management units (EMUs) of these systems to be powered at all times. Such a requirement is unattainable due to the intermittent and varying nature of EH. Additionally, these EMUs often use only one large energy storage to power the systems, which is not optimal considering the distinct energy needs of different system tasks. We addressed these issues in our recent study by proposing an EMU capable of selecting task-specific operating voltage levels and energy storage sizes at runtime while reliably retaining this information (internal EMU state) on the EMU side, thanks to the non-volatile memory (NVM) elements used. However, this solution had only two options for voltage and storage selection, preventing the system from providing the precise energy levels required by each task. Hence, this study extends these options for greater granularity in optimizing task-specific energy needs via a multi-storage EMU approach, offering an ever-efficient state retention unit (SRU) solution. We use the Signal Transition Graph (STG) method to design SRU’s control logic that handles the NVM elements for retaining the internal EMU state. The empirical measurements reveal that the actual energy overhead added by the SRU is as low as 0.1mJ to update the EMU state while the static current consumption is $\simeq3\mu$A.