Design of Reflectionless Vias Using Neural Network-Based Approach

As the data rate is significantly increased, the effect of via discontinuity on the signal integrity of printed circuit boards has become more prominent and nonnegligible. In this paper, the artificial neural network approach to efficient design of reflection-less via structures with single-ended interconnections is presented. Full-wave analysis is employed to characterize the via structures, from which the empirical representation is derived using a neural network method with the physical parameters as inputs and the electrical characterization as the output. With the accurate and fast neural network models of desired via structures, one can rapidly determine the solution space of the feasible geometries with minimized reflection noise. The proposed approach is finally validated by the time-domain simulation results and time-domain reflection experiments.