A neural network based method for predicting PCB glass weave induced skew

This paper proposes the use of a neural network based tool to predict the skew factor of PCB laminate differential channel designs. A multitude of differential stripline design scenarios are 3D modelled, each with a different expected within differential pair skew factor. The modelled data is used to train a neural network. The neural network is tested using an unseen set of design data in order to evaluate the goodness of its predictions. Preliminary results show this machine learned technique to be a viable way to predict PCB glass weave skew without the need to resort to intensive 3D modelling. This method has potential to shorten design cycles and simplify analysis while still achieving good simulation accuracy.