Sanchez-Masis, AllanAllanSanchez-MasisRimolo-Donadio, RenatoRenatoRimolo-DonadioRoy, KallolKallolRoySulaiman, ModarModarSulaimanSchuster, ChristianChristianSchuster2024-02-062024-02-062023-124th IEEE MTT-S Latin America Microwave Conference (LAMC 2023)9798350316407https://hdl.handle.net/11420/45518Neural Networks are often used for classification problems, where the electrical system must meet certain specification or performance metrics by selecting the appropriate input parameters or features. However, in many scenarios, the full response of the system is required, for instance, in terms of S-parameters in the frequency domain. Learning this continuous system response is a non-trivial task. An efficient regression model needs to learn from the training data sampled at different frequency points. In this paper, a feed-forward neural network as a predictive S-parameter response model of multilayer interconnects is proposed. Hyperparameter optimization by genetic algorithms is employed, and it was found that the model complexity (number of trainable parameters) increases for longer maximum electrical lengths of the transmission. Therefore, it becomes increasingly difficult to derive a good prediction with long electrical lengths that covers all the frequency range of interest.eninterconnectsmachine learningneural networksregressionscattering parametersSignal integrityMLE@TUHHConference Paper10.1109/LAMC59011.2023.10375594Conference Paper