Evaluation of Neural Networks to Predict Target Impedance Violations of Power Delivery Networks


An artificial neural network approach is presented to predict whether a power delivery network setup violates the target impedance. Random decoupling capacitor distributions are evaluated. It is shown that a prediction accuracy close to 90% can be achieved by choosing the right preprocessing method for the input data. The applicability towards variations of port location, target impedance and port termination is shown.

[EPEPS 2019 - 28th Conference on Electrical Performance of Electronic Packaging and Sytems]
Christian Schuster
Professor, MLE-Gründer & -Advisor