A thermal-aware mapping algorithm for 3D Mesh Network-on-Chip architecture

In this paper, we present a thermal-aware mapping algorithm based on genetic algorithm that automatically maps IP cores onto 3D Mesh Network-on-Chip (NoC) architecture, which mainly takes the temperature deviation into account. In order to verify the efficiency of the algorithm, we applied our algorithm on various multimedia benchmark applications, and evaluated the performance and temperature with Nirgam and HotSpot. The result shows that, the deviation of temperature between nodes is reduced by 35% on average in our algorithm compared with random mapping, and the peak temperature is reduced by 13% on average, such that the elimination of hotspots and equilibrium of temperature are achieved with better performance.