Trained Model for Prediction of Congestion at the Router Level using SNN

This paper proposes a unique method for predicting NoC traffic congestion in WiNoC by utilizing Spiking Neural Networks (SNNs) to reduce its effect on the overall NoC throughput. The most recent computational model that imitates the actions of biological neuron systems is called a Spiking Neural Network (SNN). In this paper, We explored converting the digital throughput values of the router to spike trains and we also explored the idea of encoding spike trains using latency encoding and used a trained machine-learning model based on SNN to predict the dynamic VTH value for LIF. The main novelty is the use of SNNs to forecast traffic hotspots by identifying temporal patterns of the numerical parameters of NoC router buffers using our Python program. The Leaky Integrate and Fire (LIF) neuron model has been used for the prediction of congestion in WiNoC, our approach intends to optimize data flow and show the efficiency of SNN towards on-chip communication systems.