Congestion Prediction Algorithm for Network on Chip

Abstract

Network on chip (NoC) traffic congestion was one of the important reasons for the data transmission performance degradation. In this paper, we presented a congestion judgment algorithm, which was based on neural network. The congestion control algorithm firstly used the hamming network to compute the NoC's link buffer congestion state, secondly used the competitive network to find the worst congestion node, and then adopted avoiding congested node  routing policy to improve the NoC’s transmission performance. In this paper, the congestion control algorithm can make the data stream as far as possible evenly distributed in the NoC’s nodes and links and reduce the transmission resource competition. The simulation results showed that the congestion control algorithm could achieve better network throughput and average transmission delay. DOI: http://dx.doi.org/10.11591/telkomnika.v11i12.3987