AI News AI资讯 2h ago Updated 2h ago 更新于 2小时前 48

Samsung Unveils Next-Gen AI Storage Roadmap Featuring zHBM and 400+ Layer V10 NAND Technology 三星发布下一代AI存储路线图,展示zHBM和400层以上V10 NAND技术

Samsung unveiled its next-generation AI storage roadmap at the 2026 Future Memory & Storage (FMS) conference in Santa Clara, introducing zHBM and zNAND-O concept products zHBM vertically stacks HBM directly above AI accelerators to shorten data transmission distance, targeting ~8x performance over HBM5 and over 10x memory density via advanced wafer bonding Samsung also launched a 400+ layer V10 BV-NAND utilizing wafer bonding technology for improved AI infrastructure storage The announcements si 三星在2026年FMS大会上发布面向AI基础设施的下一代存储技术路线图,首次展示zHBM和zNAND-O概念产品 zHBM通过将HBM垂直堆叠于AI加速器上方缩短数据传输距离,性能预计达到HBM5的约8倍 采用新型晶圆键合技术实现超过HBM5 10倍的内存密度,同时提升能效表现 同步推出400层以上V10 BV-NAND,延续NAND闪存技术演进路线

72
Hot 热度
65
Quality 质量
68
Impact 影响力

Analysis 深度分析

TL;DR

  • Samsung unveiled its next-generation AI storage roadmap at the 2026 Future Memory & Storage (FMS) conference in Santa Clara, introducing zHBM and zNAND-O concept products
  • zHBM vertically stacks HBM directly above AI accelerators to shorten data transmission distance, targeting ~8x performance over HBM5 and over 10x memory density via advanced wafer bonding
  • Samsung also launched a 400+ layer V10 BV-NAND utilizing wafer bonding technology for improved AI infrastructure storage
  • The announcements signal Samsung's aggressive push to address the growing memory bottleneck in AI training and inference workloads

Why It Matters

Samsung's zHBM architecture represents a fundamental shift in how memory is integrated with AI accelerators, potentially reshaping the hardware design landscape for next-generation AI systems. The dramatic performance and density claims could accelerate AI model training timelines and reduce operational costs, making this highly relevant for cloud providers, AI labs, and chip designers investing in infrastructure planning.

Technical Details

  • zHBM Architecture: Vertically stacks HBM directly on top of AI accelerators, eliminating traditional interposer-based connections and significantly reducing data transmission latency between processor and memory
  • Performance Targets: zHBM with next-generation interface systems aims for approximately 8x the performance of HBM5, while achieving over 10x the memory density compared to HBM5
  • Wafer Bonding Technology: Samsung employs advanced wafer bonding techniques to enable both the zHBM stacking approach and the 400+ layer V10 BV-NAND, improving energy efficiency alongside density gains
  • zNAND-O Concept: A new NAND storage concept product was also introduced, though specific technical specifications were not detailed in the announcement
  • Event Context: Unveiled at the 2026 Future Memory & Storage (FMS) conference held in Santa Clara, California on August 4, 2026

Industry Insight

  • The memory wall remains the critical bottleneck for AI scaling; Samsung's zHBM approach of 3D-stacking memory directly on compute could become a dominant architecture paradigm, pressuring HBM competitors like SK Hynix and Micron to accelerate their own next-gen roadmaps
  • AI infrastructure planners should monitor Samsung's mass production timelines for zHBM, as the 8x performance claim could enable significant cost-per-token reductions for large-scale training runs
  • The 400+ layer NAND advancement signals that storage density improvements will continue to outpace traditional scaling, benefiting data-intensive AI applications and reducing total cost of ownership for AI data centers

TL;DR

  • 三星在2026年FMS大会上发布面向AI基础设施的下一代存储技术路线图,首次展示zHBM和zNAND-O概念产品
  • zHBM通过将HBM垂直堆叠于AI加速器上方缩短数据传输距离,性能预计达到HBM5的约8倍
  • 采用新型晶圆键合技术实现超过HBM5 10倍的内存密度,同时提升能效表现
  • 同步推出400层以上V10 BV-NAND,延续NAND闪存技术演进路线

为什么值得看

三星作为全球存储芯片龙头,其技术路线直接定义AI算力基础设施的存储上限。zHBM提出的"存储-计算一体化"架构为突破AI内存墙提供了新方向,对AI芯片厂商和数据中心建设者具有重要参考价值。

技术解析

  • zHBM架构创新:将HBM垂直堆叠于AI加速器正上方,通过缩短处理器与内存间物理距离降低数据传输延迟,搭载下一代接口系统实现约8倍于HBM5的性能提升。
  • 晶圆键合技术:采用新型键合工艺实现超过HBM5十倍以上的内存密度,在提升存储容量的同时优化能效比,解决AI大模型训练中的内存瓶颈问题。
  • V10 BV-NAND:400层以上堆叠NAND闪存采用晶圆级键合技术,延续三星在3D NAND领域的技术领先优势,为AI数据存储提供高容量解决方案。
  • zNAND-O概念产品:作为新一代NAND存储概念产品亮相,具体规格尚未披露,代表三星在存储介质层面的长期技术储备。

行业启示

  • AI存储架构正从"分离式"向"存算一体"演进,zHBM的垂直堆叠方案预示未来AI芯片设计将更深度整合存储层级,打破传统冯·诺依曼架构瓶颈。
  • 存储性能已成为制约AI算力释放的关键因素,三星技术路线表明存储厂商正从被动配套转向主动定义AI基础设施标准,产业链话语权重新分配。
  • 400层以上NAND与zHBM双线并进,反映AI基础设施对"高带宽+大容量"存储的双重需求,建议关注存储技术突破对AI训练成本下降的潜在影响。

Disclaimer: The above content is generated by AI and is for reference only. 免责声明:以上内容由 AI 生成,仅供参考。

Chip 芯片 Product Launch 产品发布 Training 训练 Inference 推理