
三星计划在其2nm工艺上构建HBM5的基础芯片,目标速度提升50%。内存不再仅仅追求容量,它正在追求逻辑节点经济。
通俗易懂 - 三星将在其2nm GAA工艺节点(逻辑类节点)上构建下一代HBM5内存的基础晶圆,以实现比HBM4E高出50%以上的速度提升。
ETNews报道(2026年7月27日),三星正在将HBM5基础晶圆推向2nm门极全包围(GAA)工艺。三星代工业务的副总裁Koo Ja-geum通过公司新闻室表示,HBM5预计需要比其直接前身HBM4E提高50%以上的速度,基础晶圆的节点选择既受速度驱动,也受功耗效率驱动。Tech in Asia证实了与HBM4E相比50%以上的框架。HBM5是HBM4E的继任者(HBM4E本身是HBM4的进一步发展),目前仍处于SK海力士/三星/美光的爬坡阶段。
基础晶圆是位于堆叠DRAM晶圆下方的逻辑层,负责协调与GPU的接口。历史上,它是基于落后的节点构建的,因为DRAM经济主导。将基础晶圆移至2nm GAA——与领先的逻辑芯片使用的相同节点——表明了两点:(1) HBM接口带宽现在是瓶颈,而不是容量;(2) 三星愿意在内存上花费逻辑节点晶圆,这重塑了代工厂的晶圆分配。
本文由人工智能撰写,并经人工编辑审核。
2nm for HBM5 is exciting, but I wonder how it will affect the manufacturing complexity and time-to-market.
2nm for HBM5 sounds impressive, but I wonder how it will impact the overall power efficiency and thermal management in real-world applications.
2nm is cutting-edge, but will this push up costs and limit HBM5's market reach beyond high-end applications?
Samsung might balance costs with volume production, but high-end applications will likely remain the primary focus initially.
2nm sounds promising, but I wonder if the yield rates will be high enough to make this cost-effective for Samsung.
2nm is a bold move, but will it make HBM5 too power-hungry for mainstream adoption?
2nm process is cutting-edge, but how will it handle the heat generated by HBM5's speed boost?
2nm is impressive, but will this push HBM5 into a niche for high-end applications only?
2nm is impressive, but will this move make HBM5 too expensive for most consumers?
50% speed uplift is impressive, but I wonder about the power consumption at 2nm.
50% speed boost is great, but will this make HBM5 more expensive? Cost is a big factor for many applications.
Capex mémoire : la course aux HBM/DRAM