Adaptive Voltage/Frequency Scaling in RISC-V SoC Platforms for Energy-Harvesting IoT Networks

Abstract

IoT nodes based on energy harvesting have extremely fluctuating and intermittent power levels, rendering the traditional dynamic voltage and frequency scaling (DVFS) techniques to long-term and reliable computation unsustainable. Convincing DVFS-based systems are mainly workload-based systems, and cannot take this into consideration as it would result in an unstable state, failure of tasks or collapse of an energy buffer when used in harvesting-powered systems. The present paper proposes an adaptive voltage/frequency scaling system in harvesting mechanisms of RISC-V system-on-chip (SoC) platform to operate the IoT autonomously without consuming energy. The architecture proposed closely integrates the power management unit, runtime energy monitoring and multi-level voltage frequency control to specify dynamically the matching between the need in computation and potentially available harvested energy. In comparison to fixed frequency or traditional DVFS strategies, the innovative strategy provides real-time scaling with respect to the storage buffer conditioning and variations in input power, and thus allows the service against the intermittent supply of power. An evaluation of hardware-realistic more specifically includes representative workloads in sensing and processing and communication taking place under different harvesting profiles. The results of the experiments show that they reach up to 3540 percent reduction in average energy per task and the rate of task completion is greatly increased in comparison to the traditional DVFS methods. The temporal behaviour analysis also validates constant capacitor voltage regulation and adjustive switching of frequency in transitory events of energy. The proposed system offers a scalable and architecture level implementation of long-life self-powered IoT nodes based on programmable RISC-V systems, and is a feasible bridge to energy-neutral embedded computing.