当前位置:首页 > 报告详情

DC25_SLIDES_Track10_PanelPowerDeliveryStrategiesFor_Dannan.pdf

上传人: S** 编号:1240860 2026-05-16 33页 3.30MB

1、Information Classification:GeneralWelcome to ConferenceJanuary 2830,2025Santa Clara Convention Center1ExpoJanuary 2930,2025 Information Classification:GeneralPower Delivery Strategies for AI and Data Center ASICs:Horizontal vs.VerticalBenjamin Dannan,Signal Edge SolutionsSteve Sandler,PicotestBrian

2、Hostetler,Hewlett Packard EnterpriseHoule Gan,GoogleTeng Long,University of Cambridge2Information Classification:GeneralPANELISTSSteve SandlerManaging Director,P| Steve Sandler has been involved with power system engineering for more than 40 years.Steve is the founder of PICOTEST.com,a company speci

3、alizing in power integrity solutions,including measurement products,services,and training.Steve frequently writes articles and books related to power supply and PDN performance,and his latest book,Power Integrity Using ADS,was published by Faraday Press in 2019.3Brian HostetlerPower Integrity Engine

4、er,Hewlett Packard E|Brian Hostetler is a senior electrical engineer at Hewlett Packard Enterprise,specializing in power integrity.He is responsible for the design of power delivery networks for Cray supercomputer systems.He holds degrees from Indiana State University and Purdue University.Benjamin

5、Dannan(moderator)Chief Technologist,Signal Edge SolutionsbenSignalEdgeS|SignalEdgeSBenjamin Dannan is a Chief Technologist and an experienced signal and power integrity(SI/PI)design engineer,advancing high-performance ASICs,chiplets,and high-speed digital designs.He is a Keysight ADS Certified Exper

6、t with numerous publications on SI/PI-related topics and has twice received the prestigious DesignCon best paper award.Information Classification:GeneralPANELISTSHoule GanTech Lead Manager,G| Houle Gan is a Tech Lead Manager with Google,Sunnyvale,CA,USA,where he leads the power delivery solutions of

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. **会议主题**:2025年1月28-30日圣克拉拉会议中心,聚焦AI与数据中心ASIC的供电策略(横向vs垂直)。 2. **核心数据**: - AI芯片功耗激增:Nvidia DGX B200(8×B200 GPU)整机约14.3kW,单机架需60kW供电。 - 供电需求预测:2028年AI将占数据中心功耗19%(高盛数据),ChatGPT单次查询耗电为谷歌搜索的10倍。 3. **技术对比**: - **横向供电**:传统方案,但PCB损耗大(电流平方增长),需厚铜层及主动冷却。 - **垂直供电**:Google已部署1kA模块(损耗降低7%),剑桥大学推出54V-0.8V/450A微型VR(67×16×3.9mm),动态响应<80mV。 4. **挑战**:垂直供电需解决散热、控制环路稳定性(需>32mF电容)及成本问题,未来目标>2kA。
AI供电新趋势? 垂直供电可行吗? 供电如何突破极限?
客服
商务合作
小程序
服务号
折叠