基于机器学习的PCB串扰预测

PCB crosstalk prediction based on machine learning

摘要:

随着电子系统中逻辑和时钟频率的迅速提高以及信号边沿的不断变抖，串扰成为印刷电路板(PCB)设计人员必须关心的问题。高速电路仿真软件帮助设计人员降低了一定的设计成本，但对串扰的仿真预测仍需花费大量时间。为提高PCB串扰预测的效率，提出一种用于描述PCB的统一数据结构，全面分析了PCB产生串扰的因素，选用自然语言处理(NLP)模型构建了用于PCB串扰预测的系统，成功将PCB串扰预测的时间降至秒级，并拥有73.2%的准确率。

Abstract:

With the rapid improvement of clock frequency in electronic system, crosstalk has become one of the problems that Printed Circuit Board(PCB) designers must concern. Although the design cost has been cut to a certain degree, it still takes a lot of time to simulate the crosstalk on PCB even with the help of high-speed circuit simulation software. Aiming to improve the efficiency of PCB crosstalk prediction, a new data structure is proposed to describe PCBs. The factors that cause crosstalk on PCB are comprehensively analyzed, and a PCB crosstalk prediction system is built by using Natural Language Processing(NLP), which reduces the time for crosstalk prediction to the magnitude of seconds and achieves 73.2% accuracy.