Modeling And Analysis Of Nonlinear High Speed Links Based On Machine Learning

As the increasing scale of high-speed links and the increasing complexity of structures,it is quite difficult that the modeling and analysis of high-speed links for engineer.Especially,with the data rate of high-speed links increasing,the impact of nonlinear factors becomes more and more prominent,which bring great signal integrity challenges to the analysis and design of high-speed links.Therefore,it is necessary to correctly characterize the model of nonlinear high-speed links and perform accurate link simulation.However,there is still a gap in the study of the nonlinear behaviors of the high-speed links.Including popular simulation tools of the high-speed links on the market,currently,there is no simulation platform which can flexibly adjust and deal with different kinds of nonlinear behaviors.At the same time,there is no accurate and effective method to solve the system response of nonlinear high-speed links.In the area of signal integrity,there are two major types of simulation techniques to get the system response of high-speed links: transient simulation and fast time domain simulation.First,When nonlinear factors exist in highspeed links,transient simulation can achieve accurate results,but the simulation time is proportional to the input sequence length,which is very time-consuming and cannot perform low bit error rate(BER)simulation.Then,with the nonlinear features of the high-speed links becoming more and more serious,the traditional fast time-domain simulation techniques based on linear time-invariant hypothesis cannot accurately predict the system response.Finally,it is of great challenge to generate accurate models between the input sequence and the output responses for nonlinear high-speed links.In response to the above problems,after fully researching current simulation tools and algorithms,this subject uses programming languages for matlab and python to develop,and an algorithm is proposed for modeling and analysis nonlinear high-speed links based on Machine Learning.In this paper,a nonlinear high-speed link simulation platform based on Simulink is proposed first,which can flexibly adjust the nonlinear factors in the link.Then the method based on Machine Learning is proposed to develop a time-domain model to accurately deal with the nonlinear factors in the high-speed links.A comprehensive comparison is presented through numerical simulation results to analyze the limitation of traditional fast time-domain simulation techniques and the accuracy of the Machine Learning modeling method.The final simulated results prove that the proposed Machine Learning-based method can facilitate accurate and fast transient simulation.This paper organically combines Machine Learning with high-speed link analysis,which makes full use of the powerful advantages of intelligent,self-learning,and the ability to handle nonlinear factors of Machine Learning,to complete modeling and analysis of highspeed links with severe nonlinear factors.This is not only a practical application of subject fusion and technology crossing,but also a new breakthrough and progress in the field of nonlinear high-speed links.The Machine Learning-based modeling method proposed in this subject has won the only best paper award at the 12 th International Workshop on the Electromagnetic Compatibility of Integrated Circuits on October 22,2019,which also fully proves the algorithm for modeling and analyzing nonlinear high-speed links based on Machine Learning is creative,practical and forward-looking.The modeling method proposed in this subject can be used to guide the simulation analysis and design of nonlinear high-speed links.At the same time,the research ideas in this paper can provide more opportunities and applications in the field of signal integrity.