当前位置:首页 > 报告详情

DC25_SLIDES_Track1_ReducingChipsRespinRate_Zamek.pdf

上传人: S** 编号:1240879 2026-05-16 16页 1.72MB

1、Information Classification:GeneralWelcome to Reducing Chips Respin Rate by Fixing Common Decoupling Methods Flaws1ILIYA ZAMEKInformation Classification:GeneralSemiconductor industry losses on chips re-tape outsThe rate of chip re-tape out is increasing The growing complexity of semiconductor devices

2、,especially high-consumption HPC,GPU,and AI/ML processors lead to an increased rate of chips re-tape out(respin),according to the Wilson Research&Siemenssurvey 1.TSMC,with its 61%market share manufactured 11895 various products in 2023 2,3.Therefore,11895/0.61=19500 products were manufactured in 202

3、3 worldwide.The total number of additional respins made in the global semiconductor industry in 2023 is 19500*1.35=26,325 respins.The cost of a single chip respin as nods topology shrinks,is also increasing due to:-the increasing mask cost 4,5;-the increased expenses on design validation and verific

4、ation,which can exceed high mask costs.For example,the recent AI/ML processor development project has over 300 people and over$2 billion invested 6-8.Assuming an average cost of$3-$5 million for all necessary expenses related to a single respin.The worldwide semiconductor industry incurs losses(resp

5、in cost)*(#of respins)ranging between$80 and$130 billion yearly.The analysis of reasons for chip re-tape outs and developing methods to reduce the respin rate has become critically important.First silicon success 2020202232%24%76%of the new chips require between 2 and 7 respins in 2023 1.From data 1

6、 we estimated that a total of 2.35 tape-outs,or more than 1.35 additional re-tape outs were required per device.Information Classification:GeneralASIC Flaws leading to the chips respin The Categories of the ASIC Flaws IC flaws are shown in Fig.2,not normalized.After normalizing all categories based

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. **芯片重投片率激增**:2023年全球半导体行业芯片重投片次数达26,325次,造成年损失800-1300亿美元,主要因HPC/GPU/AI处理器复杂度提升及掩模成本上升(单次重投成本300-500万美元)。 2. **电源完整性(PI)问题**:14%的芯片缺陷归因于功耗和IRdrop,实际因误判可能更高。传统去耦方法(如目标阻抗法)追求低PDN阻抗,导致ESR过低引发电压谐振,是重投主因。 3. **新方法验证**:提出“频率双倍跟随法”(FTF)测量片上PDN性能,并通过“临界阻抗法”优化去耦设计,实验显示可降低电压波动(如从255mV降至14mV),减少芯片失效。 4. **关键改进**:增加片上静电容(Cq)和提升PDN去耦环ESR,可抑制瞬态电压,显著降低重投片率。
**芯片重投成本高?** 半导体行业每年因芯片重投(respin)损失高达800-1300亿美元,主要原因是什么?如何通过改进电源完整性(PI)设计降低这一成本?* **低阻抗PDN的陷阱?** 传统“目标阻抗法”为何会导致芯片电压瞬态问题?新的“临界阻抗法”如何有效减少芯片重投率?* **FTF方法如何应用?** 文中提到的“频率双重跟随法”(FTF)如何通过测量电源诱导抖动(JITTERPI)诊断PDN问题?实际案例中效果如何?*
客服
商务合作
小程序
服务号
折叠