High-speed Memory Signal Integrity Compliance using the CNN

This paper proposes a methodology to evaluate the signal integrity of PCB’s signal waveforms using deep learning. The presented method includes the convolutional neural network (CNN) model which can classify automatically the result utilizing images of the high-speed signal waveform measured in the memory circuit. The conventional method is necessary to understand the standard and make an effort to define it to make sure the resulting waveform is evaluated, however, this method can judge pass/fail only with the images of the signal, so it has the advantage to reduce the time (20%) of data processing. In this paper, high-speed signal waveform data of the LPDDR bus were analyzed using the Altair PollEx simulation tool, and the resulting waveform was processed in Python language for the training. The result showed that compliant waveforms satisfying the signal integrity criteria were found within Epoch4 with high accuracy, which validates the effectiveness of the proposed methodology.