公众号记得加星标⭐️，第一时间看推送不会错过。 三星电子设备解决方案（DS）事业部总裁兼首席技术官（CTO）宋在赫公布了公司下一代产品路线 图。 在2月11日于首尔江南区COEX举行的"SEMICON Korea 2026"主题演讲中，三星电子社长宋载赫表 示："随着人工智能从智能体人工智能（Agent AI）向物理人工智能（Physical AI）发展，我们预计 工作负载（数据计算量）将大幅增加。三星电子正在开发能够显著降低内存带宽限制的技术。" 宋社长强调，三星电子是唯一一家涵盖存储器、晶圆代工（半导体代工制造）和封装的集成器件制造 商（IDM），并表示："我们计划展示三星半导体独有的强大协同优化能力。"他解释说，三星旨在通 过涵盖设计、工艺、存储器和封装的集成解决方案，引领先进技术的发展。 宋总裁还介绍了下一代HBM架构"cHBM"和"zHBM"的研发进展，并指出"我们正在与客户沟通"。他 分享了"三星定制HBM（cHBM）"的研发成果，并表示"我们正在研发定制HBM，通过主动采用芯片 间接口IP，确保更高的带宽"。cHBM是一种专用集成电路（ASIC），旨在通过为AI半导体客户进行 定制来最大限度地提 ...

Summary of J.P. Morgan's Memory Market Update Industry Overview - The report focuses on the **memory industry**, particularly **DRAM** and **HBM (High Bandwidth Memory)** solutions, as discussed during the **SEMICON Taiwan 2025 Memory Executive Summit** [1] Key Insights 1. **Long-term Growth Prospects**: - Positive outlook on long-term DRAM growth driven by: - Increasing custom silicon needs for next-gen HBM solutions - Enhanced power savings contributions - Development of emerging memory solutions (CXL, PIM, Z-NAND, CUBE) - New memory solutions for edge AI applications - Pathfinding to higher capacity through hybrid copper bonding [1] 2. **Custom HBM's Role**: - Transition from passive to active memory solutions, with custom HBM (cHBM) becoming integral to AI infrastructure design - Memory makers are focusing on tailored features and processing capabilities to meet diverse customer needs - Collaboration with foundries and supply chain partners is emphasized [1] 3. **Power Consumption and AI**: - AI's explosive power consumption is leading to a critical role for HBM in total cost of ownership (TCO) savings - Datacenter power consumption is projected to rise from 1.5% in 2025 to 3.0% by 2030, with AI servers expected to account for 65% of datacenter power consumption by 2030, increasing 56 times compared to 2023 [7] - Memory makers are addressing design challenges to meet power-saving requirements, with SKH highlighting the importance of higher stack and bandwidth [7] 4. **Emerging Memory Solutions**: - Memory makers are targeting the commercialization of emerging memory solutions for specific applications - Notable developments include Samsung's LPD5X-PIM and SEC's SOCAMM2, which offers a 70% smaller form factor compared to DIMM [12] 5. **AI Inferencing Growth**: - Shift from AI training to AI inferencing is expected to drive growth in edge AI hardware, with a projected 21% CAGR from 2023 to 2030, reaching a total addressable market of $50 billion [12] - Memory suppliers are preparing advanced edge AI solutions for market adoption in the next 2-3 years [12] 6. **Hybrid Bonding Solutions**: - Increasing commitment to hybrid bonding solutions in HBM, with expectations of higher layer stacking and better thermal resistance [15] - SKH has completed HBM4 development, supporting speeds of 10Gb/s+, ahead of competitors [15] 7. **Investment Recommendations**: - Memory shares have risen 27% in the last month, driven by strong AI demand and improved pricing expectations for conventional memory [15] - The memory sector's risk-reward profile is viewed favorably over the next 6-12 months, with SKH and SEC recommended as top picks [15] Additional Important Points - The report highlights the importance of R&D investment in hybrid bonding technology and the expected impact of HBM4 qualification on pricing negotiations [15] - The ongoing strength in AI demand is anticipated to maintain tight supply-demand conditions until new capacity expansions begin in 2027 [15]