Reflectionless Design for Differential-Via Transitions Using Neural Network-Based Approach

Based on the concept of impedance matching, this paper presents a novel design method that can efficiently achieve the least reflection noise for differentia-via transitions. The relation between the effective via-impedance and the physical dimension is well approximated by a three-layer neural network. In use of the full-wave simulation, the time-domain reflectometry waveform is also acquired to confirm the accuracy of the proposed approach.