Assessing the potential of artificial intelligence and machine learning for thermal management in electronic devices

The object of this study is the potential of artificial intelligence (AI) and machine learning (ML) techniques for thermal management in electronic devices. One of the most problematic aspects identified is the challenge of ensuring performance, reliability, and energy efficiency across diverse systems, including semiconductors, data centers, and consumer electronics. In the course of the research, the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) methodology was used to systematically analyze 150 studies. These studies employed various approaches, such as predictive modeling, optimization algorithms, and real-time control systems. Our findings indicate that AI-driven thermal management can reduce energy consumption by up to 81.81 %, depending on the cooling method and optimization. Reinforcement learning-based HVAC control achieves 17.4 % energy savings, while ML-driven power management in manycore systems reduces energy use by 30 % and lowers peak chip temperatures by 17 °C. Neural network-based thermal forecasting achieves <1 % error, improving prediction accuracy. Additionally, LSTM models for thermal prognosis achieve a 3.45 % relative prediction error, outperforming traditional regression methods. These results highlight the potential of AI in optimizing thermal behavior across data centers, smart buildings, and manycore chip architectures. Key limitations were also identified, including limited data availability, challenges in model interpretability, and integration with legacy systems. The study provides a roadmap for scalable AI-driven thermal management. Emerging trends such as physics-informed ML models and the integration of cooling technologies promise innovation. Compared to conventional methods, these advancements deliver clear benefits in sustainability and adaptability.