Convolutional Neural Network-based Fast and Accurate Irregular Shape Power/Ground Plane Impedance Estimation Method for High-Speed Signaling

In this paper, we propose a convolutional neural network (CNN)-based irregular shape power/ground impedance estimation method. For the given input power/ground (P/G) plane structure, the proposed method can estimate self impedance in a fast and accurate manner. Similar to the images represented in 2D matrices whose element values indicate the brightness, we represented shape and the dimension of the P/G plane in 2D matrices. Hence, CNN can be applied to estimate P/G plane impedance which is suitable to and outperformed in the image process. Once, CNN model is trained with numerous training data sets, the inference of P/G impedance can be performed in a few milliseconds (ms). For verification, we compared the estimated P/G plane impedance by the proposed method with the conventional 3-dimensional electromagnetic simulation results. As a result, we confirmed that the average error at each frequency is about 0.44% compared to 3-dimensional electromagnetic simulation. Moreover, unlike 3-dimensional electromagnetic simulation which required 10 minutes, the proposed method took 5.4 ms resulting in a 99.99 % reduction of computing time.