摘要 Abstract 

由於現代多核心系統的高複雜互連，晶片網路（NoC）技術有效解決通信問題，但高功率密度導致熱設計挑戰。透過預測溫度，主動動態熱管理（PDTM）能減少性能影響。本研究提出基於自適應機器學習的PDTM，結合自適應單層感知器（ASLP）和強化學習技術，有效降低溫度預測誤差0.2%至78.0%，提升系統性能2.4%至43.0%，並獲得東亞首篇IEEE TVLSI最佳論文獎。

Due to the highly complex interconnections in modern multi-core systems, Network-on-Chip (NoC) technology effectively addresses communication issues. However, high power density poses thermal design challenges. By predicting temperature, Proactive Dynamic Thermal Management (PDTM) minimizes performance impact. This study introduces a machine learning-based PDTM, combining Adaptive Single Layer Perceptron (ASLP) and reinforcement learning techniques, effectively reducing temperature prediction errors by 0.2% to 78.0% and improving system performance by 2.4% to 43.0%. It also received the first IEEE TVLSI Best Paper Award in the East Asia region.