Signal Integrity Analysis of Electronic Circuits Based on Machine Learning

With the rapid development of wireless communication technology and high-tech, the signal transmission frequency of high-speed electronic lines and wireless communication systems continues to increase, and intelligent electronic equipment and base stations are developing toward miniaturization and densification. Signal is more susceptible to interference in the transmission process, which leads to signal integrity problems, so signal integrity analysis is very important in the design of high-speed electronic lines and wireless communication systems. However, the traditional signal integrity simulation is slow and costly, which prolongs the design cycle of the system. Although the existing signal integrity analysis algorithms based on machine learning can greatly improve the simulation speed, they still face the problems of high demand for labeled data and poor applicability. To solve the problems of high cost and long time of traditional signal integrity simulation, existing signal integrity simulation algorithms based on machine learning require a large amount of labeled data and poor applicability. In this paper, a signal integrity simulation algorithm based on active learning and transfer learning and a circuit automatic parameter optimization algorithm based on deep reinforcement learning are proposed, aiming at a signal integrity simulation algorithm with low tagging cost, strong applicability and high simulation accuracy and a high-performance automatic parameter optimization algorithm for electronic circuits. In this paper, a signal integrity simulation algorithm based on active learning is proposed. Compared with other active learning algorithms applied to regression tasks, this algorithm models data annotation problems in active learning based on Markov decision process, and then learns data annotation strategies with the help of deep reinforcement learning algorithms. On this basis, in order to further reduce the cost of data annotation and improve the simulation performance of neural networks, this paper proposes a signal integrity simulation algorithm based on generative reinforcement active learning, which uses generative adversarial networks to enhance the newly constructed annotated data and generate data with the same distribution for training neural networks. Simulation results show that the algorithm can reduce the data volume by nearly 50%, greatly reduce the cost of data annotation, and ensure the performance of simulation accuracy.