DAN: Deep Neural Network-Based Application Mapping for Optimized Network-on-Chip Design

Many core Network-on-Chip (NoC) technologies are increasingly utilized to improve the performance of high-performance computing systems. One method of achieving this is to deploy multiple applications simultaneously on the NoC. Deploying applications in NoCs involves application mapping, a challenging design process with NP-hard characteristics. While various approaches have been explored for addressing the application mapping problem, most efforts have concentrated on mathematical and metaheuristic approaches. This study introduces an application mapping approach using a Deep Neural Network for mesh NoC. Communication costs and computation time are considered to assess the effectiveness of this approach. The results indicate that our approach can reduce communication costs within 5% of the best cost calculated by the mathematical formulation-based approaches with very few computational resources and execution times reported in the literature. The findings demonstrate that the proposed strategy efficiently resolves application mapping issues without excessive complexity or computational expenses, and it is also scalable for large-size NoCs.