DB-KBNN based Approach for PSIJ Analysis with a Comparative Study of Energy Models

An efficient hybrid neural network method for handling multiple noise sources, including the power supply noise, input data noise, and the ground bounce noise in a power integrity analysis is presented. The proposed hybrid model, combination of deep belief and knowledge-based neural networks, provides reasonable accuracy for PSIJ response using training data from semi-analytical models as well as a circuit simulator. Also, a comparative study of different energy models in generating an optimal training data set is presented.