Thermal-Aware AI-Based Power Management and DVFS Optimization for High-Performance VLSI Systems
Keywords:
Thermal-aware power management, dynamic voltage and frequency scaling (DVFS), artificial intelligence (AI), reinforcement learning (RL), VLSI systems, energy efficiency, thermal optimization, power consumption, adaptive control, high-performance computing.Abstract
The growing power density in high-performance VLSI systems has exacerbated thermal issues, directly affecting system reliability, performance and energy efficiency. A common method of power management is the Dynamic Voltage and Frequency Scaling (DVFS), but traditional, heuristic-driven DVFS methods cannot flexibly respond to changing workloads and thermal conditions. The paper will introduce a thermal-aware AI-controlled power management framework that combines the decision-making component of machine learning with the adaptive control of DVFS. The proposed model is a dynamically-adjusting voltage and frequency model of the actual workload and temperature feedbacks. An optimization framework is integrated with a mathematical model of power and thermal behavior and a controller based on reinforcement learning is used to reach optimal trade-offs amongst power consumption, performance and thermal constraints. Experimental analysis shows that there are major advancements in energy usage and thermal control over conventional DVFS techniques, attaining as much as 28 percent energy savings and 15 percent of temperature characteristics with no decline in performance. The proposed framework is suitable for next-generation high-performance and energy-constrained VLSI systems.
