A Dynamic Voltage and Frequency Scaling Approach for Energy- Efficient Embedded Systems

Abstract

Battery-powered embedded systems must use energy sparingly, yet they often run their processor at full speed even when the workload does not require it, wasting power. Dynamic voltage and frequency scaling reduces energy by lowering the processor's voltage and clock frequency when full performance is unnecessary, exploiting the strongly nonlinear relationship between voltage and power. This paper presents a workload-aware dynamic voltage and frequency scaling approach for energy-efficient embedded systems that adjusts the operating point to the current load while ensuring that task deadlines are still met. The approach was evaluated on an embedded system-on-chip against a no-scaling baseline and a static scaling policy across low, medium, and high workloads, measuring energy consumption and deadline compliance. The dynamic approach reduced energy consumption by about thirty percent relative to no scaling, and more than the static policy, while meeting almost all deadlines. The savings were largest at low workloads, where the processor could run well below full speed. The results show that workload-aware dynamic scaling is an effective means of conserving energy in embedded systems without compromising real-time behaviour. KEYWORDS: Dynamic voltage and frequency scaling, energy efficiency,embedded systems, low-power design, real-time, power management