当前位置:首页 > 报告详情

DCON26_SLIDES_Track01_DistributedCapacitorCharacterizationforAdvancedPackaging_143_86.pdf

上传人: 明**** 编号:1195465 2026-04-16 29页 2.44MB

1、Welcome to ConferenceFebruary 2426,2026Santa Clara Convention CenterExpoFebruary 2526,2026 1Distributed Capacitor Characterization for Advanced PackagingSivaseetharaman Pandi,(Amazon)Uday Raj Shakelli,(Amazon)Sivaseetharaman Pandi,(Amazon),Uday Raj Shakelli(Amazon)Gustavo Blando(Amazon),Istvan Novak

2、(SAMTEC)2ImageImage SPEAKERSSivaseetharaman PandiSenior SIPI Engineer,AmazonSUday Raj ShakelliSenior SIPI Engineer,A3ImageImage SPEAKERSGustavo BlandoSIPI Manager,AIstvan NovakPrincipal Signal and Power Integrity Engineer,Samtec inc.Istvan.N 4Abstract5 Distributed capacitors play crucial role in pow

3、er integrity for modern GPUs High current delivery with minimal voltage ripple is essential Comprehensive characterization methodology presented Includes measurement techniques,modeling,and simulation correlation Temperature and bias voltage dependencies characterized Modern AI Workloads Demand Extr

4、eme Power AI accelerators(GPUs,TPUs,ASICs)require 400-700W+power Peak currents exceed 1000 amperes Rapid current transients occur in nanosecond timescales High di/dt events create severe voltage droops PDN noise threatens system reliability The Challenge:Traditional power delivery solutions struggle

5、 to meet these demandsIntroduction-The AI Power Challenge6 Rapid di/dt Events AI accelerators experience rapid current transients during training/inference cycles Transient timescales:10s of Nanoseconds Frequency content Extends into tens of MHz Current swings:Hundreds of amperes Creates severe volt

6、age droops and PDN noise Threatens system reliability and performance Impact:Voltage instability can cause computational errors or system failures Challenge:Power delivery network must respond within nanosecondsCurrent Transient Challenges7 Power Delivery Network Coverage Frequency Domains:On-die ca

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. **会议与主题**:2026年2月24-26日圣克拉拉会议中心,主题为先进封装分布式电容器表征(Amazon & Samtec)。 2. **AI功率挑战**:现代AI加速器需400-700W功率,峰值电流超1000A,纳秒级di/dt导致电压跌落,传统PDN无法满足10-100MHz关键频段需求。 3. **DSC技术优势**:分布式硅电容器(DSC)电容密度100-500nF/mm²,ESL仅20-80pH,ESR 20-100mΩ,谐振频率~45MHz,填补MLCC与片上电容间的频 gap。 4. **表征方法**:采用VNA两端口并联法(5Hz-3GHz),结合温度(25-110°C)和偏压(0-5V)测试,电容随温度升高5%,偏压呈抛物线关系(C(V)=C₀+3.455V+6.284V²)。 5. **模型与验证**:XcitePI与HFSS仿真显示,C4焊盘电感(46pH@1GHz)是主要寄生参数,单元模型包含C1(电压/温度依赖)、R1(漏电阻)、R2(ESR)、L1(ESL)。
**DSC如何解决AI供电挑战?** **DSC与传统电容有何不同?** **DSC的温度依赖性如何?**
客服
商务合作
小程序
服务号
折叠