A Flexible Neural Network-Based Tool for Package Second Level Interconnect Modeling

This paper introduces a neural network (NN)-based practical design tool for quick assessment of package second level interconnects (SLIs) at the earlier design stages. The study addresses the well-known computational cost problem of data generation and training processes of NN implementation by proposing a flexible model approach, where the SLI geometry is divided into several building blocks, for which a separate NN model was trained. The NNs take geometrical parameters as inputs and return the complex S-parameter matrices as outputs. The electrical performance of the entire SLI geometry is obtained by cascading the S-paramaters of the building blocks.