Semi-Supervised Learning Based on Hybrid Neural Network for the Signal Integrity Analysis

Abstract

The signal integrity analysis of high-speed circuit channels becomes a challenging task, with the development of integrated circuit technology. To solve this problem, we proposed a fast-training semi-supervised learning method based on hybrid neural network (HNN) to predict the eye-diagram metrics. Compared with the existing methods, the proposed method only requires a small amount of training data with labels, the proposed method can automatically generate the labels for the unlabeled data with a small amount of labeled data with HNN based semi-supervised learning. To this end, the proposed method can save a great amount of time, which will be a more realistic solution for the practical application. Compared with existing machine learning-based methods, the proposed method requires 50% less labeled data for training with 32.29% and 20.73% accuracy improving on deep neural network (DNN) and co-training-style semi-supervised regression (COREG) methods, receptively.