Improved Mapping Optimization Algorithm based on Generative Adversarial Networks (IMOAG)

To solve the NP-difficult problem of Network-on-Chip (NoC) mapping, researchers currently use heuristic search algorithms represented by Genetic Algorithm (GA) for optimization. However, as the scale of the problem increases, the performance of these algorithms will deteriorate rapidly, and the global optimal solution cannot be obtained. Therefore, this paper proposes an improved mapping optimization algorithm based on Generative Adversarial Networks (GANs), which combines GA with GANs to overcome the former's shortcomings. The simulation results show that compared to GA without GANs post-processing, in MMSE-IRC and SVD examples of the wireless communication system, the optimal solution obtained by trained GANs is closer to the true optimal solution, and their power consumption is reduced by 24.49% and 17.54% respectively, while the delay is reduced by 10.89% and 8.77% respectively, and it performs better in terms of distribution and convergence.