新华社迪拜2月3日电 专访｜高性能计算是国家核心能力——访图灵奖获得者杰克·唐加拉 对于中国高性能计算的发展路径，唐加拉给予了积极评价。他表示，中国以系统化、大规模的方式将高 性能计算作为国家战略能力持续推进，其显著特点包括长期稳定的投入、高性能计算与国家重大任务的 紧密结合，以及对本土供应链和产业生态的高度重视。 唐加拉认为，一个国家建设高性能计算能力，关键并不在于采购最先进的设备，而在于能否构建一套真 正缩短"科研到成果"周期的综合能力体系。这要求国家在顶层设计上具备清晰的任务目标和系统性规 划，同时拥有稳定的电力保障和可靠的基础设施支撑，并在硬件不断迭代的同时，在软件、算法和方法 体系建设领域持续投入。 他指出，长期稳定的科研和工程技术团队同样至关重要，只有在系统运行、性能优化和计算科学等方面 形成持续积累，才能将峰值算力转化为真实的科研产出。此外，计算、存储、互联和输入输出等关键环 节必须协同设计，避免系统结构失衡制约整体效能。 在谈及技术发展趋势时，唐加拉强调，人工智能与高性能计算已进入深度融合阶段，两者不再是彼此独 立的技术领域。人工智能正成为科学计算中的常规工具，而高性能计算则为大规模人工智能训 ...

Core Viewpoint - Samsung Electronics' foundry business has been experiencing long-term losses, with an estimated operating loss of approximately 6 trillion KRW (about 4.1 billion USD) last year. However, with the full-scale production of the 2nm process expected this year, the operating loss is projected to decrease to around 3 trillion KRW [2]. Group 1: 2nm Process and Production Plans - Samsung Electronics plans to start mass production of its advanced 2nm foundry process in the second half of this year, with successful development progress reported in yield and performance targets [2]. - The company is conducting performance, power consumption, and area (PPA) evaluations in collaboration with major clients, ensuring that technical validation is proceeding as scheduled before mass production [2]. - The new semiconductor production at the Taylor, Texas facility marks the 30th anniversary of Samsung's semiconductor operations in the U.S., aiming for a significant leap forward [2]. Group 2: Taylor Factory and Market Competition - The Taylor factory, with an investment of 37 billion USD (approximately 247 billion RMB), is preparing to produce advanced processes of 3nm and below to meet the demands of high-performance computing (HPC) and artificial intelligence (AI) [3]. - The factory has received orders, including Tesla's autonomous driving chip "AI6," signaling a recovery for Samsung's foundry business after a challenging period [3]. - Competition in the U.S. foundry market is intensifying, with TSMC announcing its largest investment plan in history, ranging from 52 billion to 56 billion USD [3]. Group 3: Challenges and Strategic Moves - Samsung's 2nm process yield is approximately 50%, which is lower than TSMC's known yield of 70-90%, highlighting a significant technical gap that needs to be addressed [3]. - TSMC holds a dominant market share of 70% in the global foundry market, significantly outpacing Samsung Electronics, which holds only 7% [3]. - The U.S. government's support for Intel, including an investment of about 8.9 billion USD (approximately 12.3 trillion KRW), adds competitive pressure on Samsung's foundry business [4]. Group 4: Future Prospects and Collaborations - Samsung's foundry business must demonstrate its advanced process technology capabilities to regain momentum, with Tesla's order volume expected to serve as a proving ground [4]. - A contract worth 24 trillion KRW for semiconductor supply with Tesla is anticipated to enhance the reliability of Samsung's 2nm technology [4]. - Samsung is reportedly in discussions with companies like Google and AMD for collaboration on AI chip mass production based on the 2nm process [4]. Group 5: Competitive Strategy - Samsung is enhancing its competitiveness through a "turnkey strategy," offering a one-stop solution from semiconductor design to foundry, memory manufacturing, and advanced packaging, which is seen as a unique differentiator compared to TSMC [5]. - The company expects to secure 2nm process orders primarily in high-performance computing and AI applications, with a projected growth of over 130% compared to last year [5]. - The roadmap for the 1.4nm process is in development, with plans to achieve mass production by 2029 and distribute the first version of the process design kit (PDK) to clients by the second half of next year [5].

Core Insights - The global High Bandwidth Memory (HBM) market is experiencing rapid growth, with shipments expected to increase from 1.5 billion gigabytes (GB) in 2023 to 5.7 billion GB by 2026, and revenues projected to rise from $4.35 billion in 2023 to $50 billion in 2026 [6][7][8]. HBM Industry Definition and Advantages - HBM is a high-performance semiconductor memory based on 3D stacking technology, offering high bandwidth and energy efficiency, primarily used in high-performance computing and networking applications [1][4]. - HBM has four main advantages over traditional DRAM: high bandwidth, high capacity, low power consumption, and small size [2][3]. HBM Industry Development Status - HBM technology is becoming a standard for AI acceleration cards (GPUs, TPUs, etc.), with its value share continuing to increase [4][6]. - The demand for HBM is driven by the needs of AI and high-performance computing, with significant growth expected in the coming years [6][10]. HBM Industry Chain - The HBM industry chain includes upstream materials (electrolytes, precursors, IC substrates) and semiconductor equipment (lithography machines, etching machines), with midstream focusing on HBM production and downstream applications in AI, data centers, and high-performance computing [8][9]. HBM Industry Competitive Landscape - The global HBM market is dominated by foreign manufacturers, with SK Hynix holding a 53% market share, followed by Samsung at 38% and Micron at 9% [14][15]. - Domestic companies in China, such as Changxin Memory, Changdian Technology, and others, are making significant progress in the HBM supply chain, aiming to increase local production capabilities [15][16]. HBM Industry Development Trends - HBM is positioned as a critical hardware component for AI and high-performance computing, with its unique 3D stacked structure providing superior bandwidth compared to traditional memory solutions [16][17]. - The future memory landscape will be heterogeneous, with HBM focusing on training scenarios, while other memory types will cater to specific workloads, creating a diverse memory ecosystem for the AI era [17].

该技术平台的核心是 4 个大型逻辑芯片块，据称基于英特尔 18A 制程工艺打造，因此集成了环 绕栅极晶体管（RibbonFET）与背面供电技术（PowerVia）。逻辑芯片块两侧配置类 HBM4 内 存堆栈与 I/O 芯片块，各组件间预计通过直接嵌入封装基板的增强型嵌入式多芯片互连桥接技术 2.5D 桥接器（EMIB-T）实现互联。英特尔对 EMIB-T 技术进行了升级，在桥接器内部增设硅 通孔（TSV），使电力与信号既能横向传输，也可纵向流通，从而最大限度提升互连密度与供电 效率。从逻辑架构来看，该平台针对通用芯片互连标准（UCIe）的芯粒间接口设计，支持 32 吉 比特每秒（GT/s）及以上的传输速率，这一接口标准似乎也被用于连接类相干高带宽内存 4 增强 版（C-HBM4E）堆栈。 如果您希望可以时常见面，欢迎标星收藏哦~ 英特尔代工事业部（Intel Foundry）于本周发布一份宣传文件，详细介绍其面向人工智能（AI） 和高性能计算（HPC）应用的前沿前后端制程解决方案，并展示了一款 "人工智能芯片测试载 体"，以此佐证公司当前具备的封装技术实力。该测试载体的表现确实亮眼 —— 其尺寸达到 8 个 ...

ASC竞赛组委会委员刘羽介绍，大赛全方位锤炼学生在硬件、中间件、软件、应用等全链条能力，塑造 兼具理论与工程实践的复合型人才。这类人才正是当前破解 "卡脖子" 技术难题、推动科技自立自强的 核心力量，与国家发展需求高度契合。 工信部电子信息司电子系统处处长金磊表示，计算是信息技术核心、数字经济基石，更是培育新质生产 力的关键领域，当前人工智能与高性能计算深度融合催生产业变革，高素质人才支撑至关重要。ASC作 为全球超算竞赛，精准契合产业与人才培育需求。期待赛事深化产学研协同，为提升科技自主创新能力 夯实人才基础。 据了解，为引导参赛者探索人工智能与科学前沿，ASC26预赛设置了"具身智能"与"引力波数值模拟"两 大应用赛题。其中，具身智能是人工智能的前沿领域，旨在构建能够感知、推理并与物理世界交互的智 能系统。人形机器人作为该领域的典型代表，需在非结构化的真实物理环境中操作各类物体，并基于对 物理规律的理解进行决策。 2026 ASC世界大学生超级计算机竞赛（ASC26）启动会日前在北京举行。本届大赛吸引了来自中国、 美国、德国等国家和地区的超300支高校队伍报名。通过预赛选拔的晋级队伍，将于2026年5月1 ...

Core Viewpoint - MRDIMM is a new memory module architecture designed for high-performance computing and data-intensive applications, achieving a data transfer rate of 8800 MT/s, which is nearly a 40% increase in peak bandwidth compared to standard DDR5 RDIMM [1][2] Group 1: MRDIMM Technology Overview - MRDIMM is suitable for high-performance computing and artificial intelligence applications, addressing the bandwidth decline per core due to the increasing number of processor cores [2] - The first generation of MRDIMM supports a data transfer rate of 8800 MT/s, with the third generation expected to reach 14000 MT/s [2] - The architecture of MRDIMM includes 1 MRCD (Memory Register Clock Driver) and 10 MDB (Memory Data Buffer) chips, indicating significant growth potential in the market [3] Group 2: Market Demand Projections - By 2030, the demand for MRCD and MDB is projected to increase significantly, with estimates under pessimistic, neutral, and optimistic scenarios suggesting demand for MRCD could reach 274 million, 456 million, and 593 million units respectively, while MDB demand could reach 2.736 billion, 4.560 billion, and 5.928 billion units [4] - The penetration rates for MRDIMM are estimated to be 30%, 50%, and 65% under pessimistic, neutral, and optimistic scenarios respectively [4] Group 3: Industry Implications - The adoption of MRDIMM technology is expected to drive demand for MRCD and MDB chips, particularly as it becomes compatible with Intel's sixth-generation CPUs [1][3] - Key companies in the market include 澜起科技 (Lanke Technology) and 聚辰股份 (Juchen Technology), which are involved in the production of MRCD/MDB chips and DDR5 memory modules respectively [1]

Investment Rating - The report maintains a "Positive" investment rating for the industry [1]. Core Insights - The transition of server-level CPU memory modules towards MRDIMM technology is highlighted, with a focus on MRCD and MDB chips as key growth areas. MRDIMM is suitable for high-performance computing and AI applications, achieving data transfer rates of up to 8800MT/s, with future generations expected to reach 14000MT/s [3][5][11][12]. - The report anticipates significant demand growth for MRCD and MDB chips, projecting that by 2030, the demand could reach 5.93 billion and 59.28 billion units respectively under optimistic scenarios [5][20][22]. Summary by Sections 1. Development of Server CPU Memory Modules Towards MRDIMM - The report discusses the evolution of memory modules in servers, emphasizing the importance of stability, error correction, and low power consumption as data storage and processing loads increase [8][9]. - Current mainstream memory module technologies include UDIMM, RDIMM, and LRDIMM, with MRDIMM emerging as a new architecture designed to enhance memory bandwidth and performance [10][11]. 2. Core Incremental Growth from MRCD and MDB Chips - MRDIMM utilizes a "1+10" architecture, consisting of one MRCD and ten MDB chips, which are crucial for its performance enhancements. The technology is already compatible with Intel's sixth-generation CPUs, and major memory manufacturers have begun producing MRDIMM products [5][17]. - The report outlines potential penetration rates for MRDIMM by 2030, estimating demand for MRCD and MDB chips under various scenarios [20][22]. 3. Demand Projections for MRCD and MDB Chips - Under optimistic assumptions, the report forecasts that by 2030, the demand for MRCD and MDB chips could reach 5.93 billion and 59.28 billion units respectively, driven by the increasing adoption of MRDIMM technology in server environments [20][23].

Summary of Key Points from Conference Call Records Industry Overview - The global semiconductor industry is expected to approach $1 trillion by 2026, with significant growth in the storage sector. The hardware segment is projected to outperform the software segment, with raw materials, storage, and semiconductor equipment showing strong performance, while consumer electronics brands are expected to be the most affected, with Apple being relatively less impacted [1][4]. Smartphone Market - Global smartphone shipments are forecasted to decline by 6.7% in 2026, with Apple and Samsung remaining stable, while Chinese brands are expected to drop by 14% due to storage shortages affecting brands like Huawei, Honor, Xiaomi, OPPO, and vivo [1][9]. Capital Expenditure Trends - Microsoft and Meta are projected to increase their capital expenditures significantly in 2026, with Meta's spending rising from $70 billion to $120 billion, and Microsoft also showing over 40% growth. However, internet companies are only expected to see a 15% revenue increase, leading to tighter cash flows [1][10]. ASML and Equipment Market - ASML reported a record high in EUV order revenue, with over 100% year-on-year growth, driven by major clients like TSMC and storage manufacturers concerned about supply shortages [1][11]. Lam Research predicts a WFE market growth of over 20% in 2026, significantly higher than the 10% forecast by SEMI, driven by TSMC's 2nm transition and DRAM manufacturers' shift from HBM3 to HBM4 [1][13]. Company-Specific Insights - **Apple**: The company reported strong performance with iPhone 17 demand exceeding supply, and a 23% quarter-on-quarter revenue increase. However, concerns remain regarding storage shortages and SOC supply issues, which may impact margins [2][8]. - **Samsung**: The company saw a significant improvement in operating profit, with a rise from 14% to 21%. The semiconductor division's profit doubled to 16 trillion KRW, primarily due to DRAM and NAND price increases [1][17]. - **Hynix**: The company is expected to launch a new storage product, HBF, in late 2026, which will enhance AI inference capabilities. Hynix has a leading advantage in HBM technology [1][3][18]. - **Intel**: The target price for Intel has been raised to $71.5 based on significant advancements in process technology and successful execution of the IDM 2.0 strategy, attracting investments from both government and private sectors [1][23]. Market Valuation Changes - The valuation method for the storage industry has shifted from PB to PE, reflecting market recognition of stable profitability and growth potential for companies like Micron and Hynix. Hynix is currently valued at 9.4 times PE, with its target price raised based on strong financial performance [1][19]. HBM Technology Impact - The development of HBM technology is significantly impacting the storage market, enhancing performance and addressing heat issues. Samsung is leading in HBM4 technology, while Micron is also advancing its HBM4E plans to meet future demand [1][20]. Supply and Demand Dynamics - The global DRAM market is dominated by Micron, Samsung, and Hynix, with limited new capacity expansion expected, leading to a tight supply situation. The NAND market is similarly constrained, with major players controlling pricing and capacity [1][21]. Investment Recommendations - Hynix is identified as a preferred investment target due to its technological advantages, particularly in HBM technology, while Micron is also performing well but is slightly less favorable in comparison [1][22]. This summary encapsulates the critical insights and projections from the conference call records, providing a comprehensive overview of the semiconductor industry and key players.

Core Viewpoint - The PCB industry is experiencing significant growth driven by increasing demand for AI servers and high-performance computing, with leading companies like Shengyi Electronics and Shenghong Technology planning substantial capacity expansions to meet this demand [1][2]. Group 1: Company Performance - Shengyi Electronics expects a net profit of 1.431 billion to 1.513 billion yuan for 2025, representing a year-on-year increase of 331.03% to 355.88% [1]. - Shengyi Electronics plans to raise up to 2.6 billion yuan for projects including an AI computing HDI production base with an annual capacity of 167,200 square meters and a smart manufacturing high-layer circuit board project with a capacity of 700,000 square meters [1]. - Shengyi Technology anticipates a net profit of 3.25 billion to 3.45 billion yuan for 2025, reflecting a year-on-year growth of 87% to 98% [3]. Group 2: Industry Trends - The PCB sector is witnessing a surge in demand for high-end products such as HDI, HLC, and SLP, prompting major players to announce new financing and expansion plans [2]. - The growth in the PCB market is also positively impacting upstream materials, with Shengyi Technology reporting increased sales and revenue from copper-clad laminates [3]. - The industry is responding to global market demands for high-performance materials, particularly in AI, cloud computing, 6G communication, and smart automotive electronics [3].

公众号记得加星标⭐️，第一时间看推送不会错过。 这一封装方案与台积电目前的技术路线存在显著差异（后文将进一步阐述）。简言之，该技术概念印 证了一个趋势 —— 下一代高性能人工智能处理器将采用多芯片粒架构，而英特尔代工事业部已具备 相应的制造能力。 （来源： tomshardware ） 该技术平台的核心是 4 个大型逻辑芯片块，据称基于英特尔 18A 制程工艺打造，因此集成了环绕栅 极晶体管（RibbonFET）与背面供电技术（PowerVia）。逻辑芯片块两侧配置类 HBM4 内存堆栈与 I/O 芯片块，各组件间预计通过直接嵌入封装基板的增强型嵌入式多芯片互连桥接技术 2.5D 桥接器 （EMIB-T）实现互联。英特尔对 EMIB-T 技术进行了升级，在桥接器内部增设硅通孔（TSV），使 电力与信号既能横向传输，也可纵向流通，从而最大限度提升互连密度与供电效率。从逻辑架构来 看，该平台针对通用芯片互连标准（UCIe）的芯粒间接口设计，支持 32 吉比特每秒（GT/s）及以 上的传输速率，这一接口标准似乎也被用于连接类相干高带宽内存 4 增强版（C-HBM4E）堆栈。 这款测试载体还提前披露了英特尔向垂直整合 ...