Data-Driven Prediction of Temperature-Dependent Dielectric and Conductive Parameters Based on Differential Stripline Characteristics
José Enrique Hernández-Bonilla, Torben Wendt, Torsten Reuschel, Cheng Fu Yang, Christian Schuster · 2025
The design of automotive high-speed interconnects requires the use of broadband material models that consider a wide range of environmental conditions, e.g. temperature and humidity. However, even for nominal conditions these models are nontrivial to generate. This paper extends our previous work on data-driven predictors used for the estimation of dielectric and conductive parameters of high-speed material models. Parametrized frequency-domain simulations of a differential stripline at various temperatures yield a dataset of modal propagation constants, and characteristic impedances up to 25 GHz. These inputs are used to train a 1D Convolutional Neural Network to estimate the temperature-dependent conductive and dielectric model parameters for the cannoball Huray and wideband Debye models, respectively. The trained model shows good accuracy after training/testing. Additional simulations outside the training/test dataset are used for further model validation with good agreement between predictions and expected values.