Machine Learning-Assisted Adaptive Task Scheduling for Low-Power Heterogeneous RISC-V SoCs

Modern System-on-Chip (SoC) designs increasingly use heterogeneous processors in which different cores provide different levels of performance and energy efficiency. Low-power RISC-V SoCs are suitable for embedded, Internet of Things (IoT), and edgecomputing applications because the open instruction-set architecture can support diverse processor implementations. However, assigning tasks to heterogeneous cores remains challenging because workloads change dynamically and each core has different performance and energy characteristics. This research proposes a Machine Learning-Assisted Adaptive Task Scheduling framework that monitors task characteristics, processor utilization, execution time, and estimated energy consumption to assist dynamic processor assignment. Lightweight machine-learning models, such as decision-tree classifiers or reinforcement-learning agents, can be used to predict suitable task-to-core assignments. The proposed approach is evaluated using performance, energy consumption, task completion time, processor utilization, and deadline-miss rate. The study aims to determine whether machine learning can assist heterogeneous RISC-V SoCs in balancing computational performance and energy efficiency.