System on a chip (SoC) is the leading technology in the recent global world of digitization. The classical bus-based regular communication infrastructures of SoCs cannot handle the growing number of cores. Designers came up with new communication methods known as the network on chip (NoC) to solve this problem. NoC offers scalability, flexibility, modularity, and efficiency. As the number of cores of SoC keeps on increasing due to advancement in technology and shrinking size of transistors, the arrangement of cores on the NoC becomes more significant that can influence the Network on Chip’s overall performance and efficiency. Therefore, there is always a need for efficient and intelligent mapping algorithms for achieving optimal mapping solutions since existing mapping techniques either do not achieve optimal results or possess more execution time. To resolve this issue, we proposed an improved hybrid particle swarm and simulated annealing optimization algorithm iHPSA. The proposed iHPSA combines Improved Particle Swarm Optimization and Simulated Annealing for NoC mapping and incorporates a machine-learning K-means clustering algorithm to segregate tasks into clusters based on communication bandwidth. The Elbow method is used in the K-means clustering algorithm to predict the number of clusters in large applications intelligently.Comparative analysis of different performance factors shows that the proposed approach has performed better than the existing mapping algorithms. The proposed algorithm offers improved performance by reducing communication cost, latency, and power of the network. the proposed iHPSA shows an average enhancement on power reduction of 22.67%, 30.95% , 22.87%, 27.61%, 16.07%, 8.87%, 8.01% ,30% and 9% over PSO,SA,GA,ACO,BA,SCSO,BEMAP,CSO and ILP. Improved energy consumption by 5.70%, 20%, 25%, 24% and 4.09% over PSO, SA ,GA , BEMAP and CSO. iHPSA improved latency by 1.5% over PSO, 6.2% over SA, 6.67% over GA , 9.7% over BEMAP and 1.2% over CSO. The proposed mapping approach’s efficiency is evaluated with a comparative analysis of communication cost, power, energy, and latency for standard and randomly generated benchmark applications. The experimental results revealed that iHPSA outperforms other nature-inspired algorithms by offering an efficient application approach that reduces communication cost, power, energy, and latency.
Read full abstract