Prediction of Power Supply Induced Jitter via Deep Belief and Knowledge-Based Neural Networks

In this paper, a hybrid neural network approach using deep belief and knowledge-based neural networks is developed to handle multiple noise sources including supply noise, ground bounce noise and input data noise while evaluating power supply induced jitter (PSIJ). The proposed hybrid approach provides reasonable accuracy for jitter response using training data from semi-analytical models as well as a circuit simulator. The proposed model can handle different types of inputs without needing to update the network's parameters for each case while ensuring faster convergence.