semi·news
Headlines要闻 / Research研究 / /
Thursday, November 19, 2026 2026年11月19日 星期四

AI Infrastructure Reshapes Packaging and Memory Demand AI基础设施重塑封装与存储需求

Microsoft's EMIB commitment gives Intel a named advanced-packaging customer, while DDR6 previews a new bandwidth step. HBM price forecasts and falling consumer NAND prices point to divergent conditions across the memory market. 微软采用EMIB,为英特尔先进封装业务带来明确客户,DDR6规格预览则展示了下一阶段的带宽提升方向。HBM涨价预测与消费级NAND价格回落,显示存储市场不同领域的供需正在分化。

AI & Accelerators AI与加速器

Bull Shows Diamond Rapids Package in Sequana Platform at SC26 Bull在SC26展示Sequana平台中的Diamond Rapids封装

ServeTheHome · 7h ago ServeTheHome · 7小时前

Bull displayed Intel's Xeon Diamond Rapids LGA9324 package in its Sequana XH31600 platform at SC26, ServeTheHome reports. The demonstration gives a concrete view of the processor's platform integration as server vendors prepare their next systems. A package display does not establish production availability or measured application performance. 据ServeTheHome报道,Bull在SC26上展示了Sequana XH31600平台中的英特尔Xeon Diamond Rapids LGA9324封装。这让外界能够具体观察服务器厂商为下一代系统准备的处理器集成方案。封装展示本身尚不能说明量产供货时间或实测应用性能。

Foundry & Manufacturing 晶圆代工与制造

Microsoft Commits to Intel EMIB for Azure Custom Silicon 微软为Azure自研芯片采用英特尔EMIB封装

Wccftech · 6h ago Wccftech · 6小时前

Microsoft became the first publicly announced customer for Intel Foundry's EMIB advanced packaging for its custom silicon, according to Wccftech. The expanded collaboration also includes a multiyear Xeon capacity commitment covering Diamond Rapids. The agreement gives Intel a named customer for its packaging services; that commitment is distinct from a wafer-fabrication award. 据Wccftech报道,微软成为首家公开宣布为自研芯片采用Intel Foundry EMIB先进封装的客户。双方扩大的合作还包括覆盖Diamond Rapids的多年Xeon产能协议。这为英特尔封装服务带来明确客户,但封装合作与晶圆制造订单属于不同业务。

Memory & Storage 存储器与存储系统

JEDEC Previews DDR6 at 12.8–14.4 GT/s JEDEC预览DDR6规格,速率为12.8至14.4 GT/s

Wccftech · 6h ago Wccftech · 6小时前

JEDEC previewed DDR6 specifications starting at 12.8 GT/s and extending to 14.4 GT/s, Wccftech reports. The preview covers UDIMM, RDIMM and CAMM designs, placing the next memory generation across desktop, server and compact-system formats. DDR6 remains in development, so these specifications describe a roadmap rather than current shipping-module availability. 据Wccftech报道,JEDEC预览DDR6规格,起始速率为12.8 GT/s,并可扩展至14.4 GT/s。预览涉及UDIMM、RDIMM和CAMM设计,使下一代内存覆盖桌面、服务器及紧凑型系统等形态。DDR6仍在开发中,这些规格反映的是技术路线,并不表示相应模组已经上市供货。

Consumer NAND Prices Fall 20%, With Further Declines Forecast 消费级NAND价格下跌20%,明年初或继续回落

Wccftech · 6h ago Wccftech · 6小时前

Consumer NAND spot prices have fallen 20%, and a further 5–10% decline is expected in early 2027, Wccftech reports. Lower component prices could ease costs for SSD, laptop and smartphone makers after a period of pressure on consumer storage. Spot-market declines do not necessarily pass immediately into contract prices or retail SSD prices. 据Wccftech报道,消费级NAND现货价格已下跌20%,预计2027年初还将下降5%至10%。在消费级存储经历价格压力之后,元件价格回落有望缓解SSD、笔记本电脑和智能手机厂商的成本。现货市场降价未必会立即传导至合约价或SSD零售价。

Report Forecasts Latest HBM Prices at Up to 2.7 Times Current Levels 报道预测新一代HBM明年价格最高达当前的2.7倍

Seoul Economic Daily · 3h ago Seoul Economic Daily · 3小时前

Seoul Economic Daily reports that prices for the latest HBM could reach as much as 2.7 times current levels next year. If realized, the increase would raise the memory bill for AI accelerators and complicate system cost planning. The figure is a forecast for the latest HBM, rather than an observed increase across all memory products. 《首尔经济日报》报道称,新一代HBM明年的价格可能最高达到当前水平的2.7倍。若预测兑现,AI加速器的存储成本将上升,系统成本规划也会更加复杂。这一数字是针对新一代HBM的预测,并非所有存储产品已经发生的普遍涨价。

Silicon Motion Demonstrates Gen5 and Gen6 Enterprise SSD Platforms 慧荣展示PCIe Gen5与Gen6企业级SSD平台

Embedded.com · 3h ago Embedded.com · 3小时前

Silicon Motion is demonstrating PCIe Gen5 and Gen6 MonTitan enterprise SSD controllers and reference platforms at SC26 in Chicago. The systems target AI factories, agentic AI and large-scale inference, bringing storage-controller development into the same infrastructure discussion as processors and memory. The supplied report describes demonstrations, without establishing shipment volumes or customer deployments. 慧荣在芝加哥SC26上展示PCIe Gen5和Gen6 MonTitan企业级SSD控制器及参考平台。这些方案面向AI工厂、智能体AI和大规模推理,使存储控制器开发成为处理器与内存之外的另一项基础设施议题。所提供的报道介绍了现场演示,尚未给出出货规模或客户部署情况。

Equipment & Materials 设备与材料

Lam Examines Atomic Layer Etch for Three-Dimensional Structures 泛林探讨三维结构制造中的原子层刻蚀

Lam Research Newsroom · 5h ago Lam Research Newsroom · 5小时前

Lam Research published a technical discussion of atomic layer etch for three-dimensional semiconductor structures. The subject addresses the need for controlled material removal as device geometry becomes more complex. This is a process-technology discussion, with no new tool order or quantified manufacturing result established by the supplied excerpt. 泛林发布技术文章,讨论面向三维半导体结构的原子层刻蚀。随着器件几何形状日益复杂,如何精确控制材料去除成为重要工艺问题。这是一篇工艺技术讨论,所提供的摘要未涉及新增设备订单或量化制造成果。

Policy & Export Controls 政策与出口管制

Commerce Rule Targets Remote Access to Advanced Semiconductors 美国商务部新规针对先进半导体远程访问

Morgan Lewis · 6h ago Morgan Lewis · 6小时前

Morgan Lewis reports that a new Commerce Department interim final rule targets remote access to advanced semiconductors. The development puts access to computing resources, alongside physical chip shipments, at the center of export-control scrutiny. The supplied excerpt does not identify the detailed thresholds, exceptions or effective dates. 据Morgan Lewis报道,美国商务部一项新的临时最终规则针对先进半导体的远程访问。这意味着除芯片实物出货外,计算资源的访问方式也成为出口管制审查的重点。所提供的摘要尚未列出具体门槛、例外情形或生效日期。

Challengers & New Silicon 新势力与新硅

SambaNova Plans to Triple SambaStack Deployment at TACC SambaNova扩大TACC合作,SambaStack部署规模增至3倍

SambaNova · 8h ago SambaNova · 8小时前

SambaNova announced an expanded partnership with the Texas Advanced Computing Center that triples its SambaStack deployment for scientific AI. The named installation provides a concrete expansion signal for an alternative AI computing platform in research infrastructure. A threefold deployment increase does not by itself establish a threefold application-performance gain. SambaNova宣布扩大与得克萨斯先进计算中心(TACC)的合作,将面向科学AI的SambaStack部署规模增至3倍。这项有明确机构参与的部署扩展,为替代性AI计算平台进入科研基础设施提供了具体信号。部署规模增至3倍,并不意味着应用性能也同步提升至3倍。

Newsletter 邮件订阅

Daily semiconductor briefing. 每日半导体简报。