当前位置:首页 > 报告详情

DC25_SLIDES_Track14_ImpedanceProfilePredictionAndClassification_Hassab.pdf

上传人: S** 编号:1240899 2026-05-16 81页 6.96MB

1、Information Classification:GeneralWelcome to ConferenceJanuary 2830,2025 Santa Clara Convention Center1ExpoJanuary 2930,2025 Information Classification:GeneralImpedance Profile Prediction and Classification for PCB-based PDN Decoupling Using AutoencodersYoucef Hassab,Hamburg University of Technology

2、(TUHH),Germany.Jan Severin Heling,TUHH,Germany.Morten Schierholz,TUHH,Germany.Ihsan Erdin,Celestica Inc.,Canada.Jayaprakash Balachandran,d-Matrix Corp.,USA.Christian Schuster,TUHH,Germany.2Information Classification:GeneralImageImage SPEAKERSYoucef HassabPhD.Candidate,Hamburg University of Technolog

3、yyoucef.hassabtuhh.de|www.tet.tuhh.de/en/staffReceived the B.S.degree in General Engineering Sciences and M.S.degree in Electrical Engineering from Hamburg University of Technology(TUHH),Hamburg,Germany in 2019 and 2022,respectively.In November 2022,he joined the Institute of Electromagnetic Theory

4、at the TUHH to pursue a PhD degree focusing on ML for efficient EMC.Jan Severin HelingM.S.Student,Hamburg University of Technologyjan.hesslingtuhh.de|www.tet.tuhh.de/en/staffJan Severin Hessling received his B.S.degree in Electrical Engineering from TUHH in 2022.He is pursuing the M.S.degree in Elec

5、trical Engineering at the Institute of Elec tromagnetic Theory at TUHH,working on machine learning tools in the context of PCB design.3Information Classification:GeneralImage Co-AuthorsMorten SchierholzPhD.Candidate,TUHHmorten.schierholztuhh.de|www.tet.tuhh.de/en/staffIhsan ErdinSI SME,Celestica Inc

6、.Ihsan.erdinieee.org4ImageImage Jayaprakash BalachandranSenior Technical Leader,d-Matrix Inc.jayapd-matrix.aiChristian SchusterFull Professor,TUHHschustertuhh.de|www.tet.tuhh.de/en/staffInformation Classification:GeneralImpedance Profile Prediction and Classification for PCB-based PDN Decoupling Usi

word格式文档无特别注明外均可编辑修改,预览文件经过压缩,下载原文更清晰!
三个皮匠报告文库所有资源均是客户上传分享,仅供网友学习交流,未经上传用户书面授权,请勿作商用。
1. **研究目标**:基于自编码器(AE)和神经网络(ANN)的PCB电源分配网络(PDN)阻抗预测与分类,加速PDN设计流程。 2. **核心方法**: - **阻抗预测**:通过ANN-AE模型从PCB参数(如尺寸、介电常数)预测全频段阻抗曲线(1MHz-1GHz),预测时间<30ms。 - **迁移学习**:利用预训练AE减少新案例数据需求,提升数据效率。 - **阻抗分类**:利用AE潜在空间的正则性识别相似阻抗 profile,复用去耦策略。 3. **数据与验证**: - 4种PCB案例(3,000–18,000样本),采用部分波(PB)工具生成数据,与全波仿真结果一致。 - 预测误差随案例复杂度增加,但整体行为预测准确(如共振峰位置)。 4. **应用价值**:替代耗时电磁仿真,通过相似案例复用设计策略,显著缩短PDN迭代周期。
**AI如何加速PCB设计?** **阻抗预测如何提升效率?** **相似设计如何复用策略?**
客服
商务合作
小程序
服务号
折叠