Fast Calculation Method for Crosstalk of High-Density Printed Circuit Board by Using Machine Learning

This article introduces a novel crosstalk prediction method, defines a standard data format for printed circuit board (PCB) modeling, and establishes an efficient computation model utilizing machine learning. The crosstalk prediction is undertaken by using a bidirectional long short-term memory (Bi-LSTM) model and is enhanced with an attention mechanism. To validate the model's precision and efficiency, we compare its prediction results with the results of traditional full-wave electromagnetic simulations. The model demonstrates outstanding crosstalk prediction capabilities, achieving an accuracy exceeding 95.72%. Notably, the proposed method diminishes the full board prediction time from hours to mere seconds. To verify the model's reliability, the trained model was used to predict the crosstalk of three unseen PCBs and accurately identified the potential crosstalk in the unseen PCB. Thus, the proposed method can practically be applied to a fully automated and intelligent electromagnetic compatibility design of electronic products.