A comprehensive study and holistic review of empowering network-on-chip application mapping through machine learning techniques

This study investigates machine learning (ML) techniques for optimizing Network-on-Chip (NoC) application mapping, focusing on supervised learning, unsupervised learning, reinforcement learning, and neural networks. Through a comparative analysis of recent research, the study reveals that supervised learning methods, like artificial neural networks (ANNs), enhance core vulnerability prediction and runtime mapping. Unsupervised learning techniques improve NoC mapping via multi-label models, while reinforcement learning approaches, including actor-critic frameworks, reduce communication costs and power consumption. Scenario-aware strategies adapt mapping processes to varying operational contexts. Despite these advancements, challenges such as computational overhead, data quality, and model interpretability persist. Future research should focus on scalable ML algorithms, improving data quality, and enhancing model transparency. This study underscores the significant potential of ML to advance NoC application mapping and highlights the need for ongoing innovation to address existing challenges.