中国半导体行业协会封装分会会刊

中国电子学会电子制造与封装技术分会会刊

导航

电子与封装 ›› 2025, Vol. 25 ›› Issue (1): 010307 . doi: 10.16257/j.cnki.1681-1070.2025.0012

• 电路与系统 • 上一篇    下一篇

基于神经网络的PCB电源分配网络阻抗预测方法

段克盼,贾小云,韩东辰,蒋建伟,杨振英,郭宇   

  1. 陕西科技大学电子信息与人工智能学院,西安
  • 收稿日期:2024-09-18 出版日期:2025-01-22 发布日期:2025-01-22
  • 作者简介:段克盼(1998—),女,硕士研究生,主要研究方向为电源完整性、智能信息处理;

Impedance Prediction Method of PCB Power Distribution Network Based on Neural Network

DUAN Kepan, JIA Xiaoyun, HAN Dongchen, JIANG Jianwei, YANG Zhenying, GUO Yu   

  1. Schoolof Electronic Information and Artificial Intelligence, Shaanxi University of Science and Technology, Xi'an 710000, China
  • Received:2024-09-18 Online:2025-01-22 Published:2025-01-22

摘要: 针对传统电源分配网络(PDN)建模及计算方法存在的局限性和高计算资源消耗问题,提出了一种基于深度学习的PDN阻抗预测方法(URPNet)。该方法在融合PCB不规则形状、多叠层信息及多种电容端口位置信息的基础上,采用U型编解码结构和残差单元来处理特征,通过多层感知机(MLP)及全连接(FC)层对特征进行解码和重构,从而提升网络的特征处理能力。实验结果显示,URPNet模型的决定系数R2达到0.999,均方根误差为0.431,相较于现有深度学习方法,URPNet在通用性较强的同时预测结果更准确。此外其计算速度快,能够在不到1 s的时间内完成预测,可以有效应对PDN设计中的挑战。

关键词: 电源分配网络, 目标阻抗, 电源完整性, 神经网络

Abstract: Aiming at the limitations of modelling and calculation methods and high consumption of computing resources of traditional power distribution network (PDN), a PDN impedance prediction method (URPNet) based on deep learning is proposed. Based on the fusion of PCB irregular shape, multi-layer information and position information of various capacitive ports, U-shaped codec structure and residual units are used to process features, and multi-layer perceptron (MLP) and fully connected (FC) layers are used to decode and reconstruct features, thereby improving the feature processing capability of the network. Experimental results show that the determination coefficient R2 of URPNet model reaches 0.999, and the root mean square error is 0.431. Compared with existing deep learning methods, URPNet has strong universality and more accurate prediction results. In addition, the calculation speed is fast, and the prediction can be completed in less than 1 s, which can effectively meet the challenges in PDN design.

Key words: power distribution network, target impedance, power integrity, neural network

中图分类号: