公众号记得加星标⭐️，第一时间看推送不会错过。 铠侠表示，该公司已开发出可高度堆叠的氧化物半导体沟道晶体管，能够支持高密度 3D DRAM。 这项技术的发展有望通过降低每 GB 的制造成本，并通过高导通电流和超低关断电流晶体管提高能源 效率，从而带来更便宜、更快的内存。 然而，这项技术需要精确的多层对准、集成到标准制造工艺中以及长期可靠性测试，所有这些都可能 需要数十年时间。 该技术在最近于旧金山举行的 IEEE 国际电子器件会议上进行了展示，演示了堆叠成八层垂直结构的 晶体管的运行情况。 垂直层由水平排列的晶体管组成，这些晶体管是通过用氧化物半导体材料 InGaZnO 取代传统的氮化 硅区域而形成的。 这种设计方案无需依赖传统的平面DRAM结构即可增加内存容量。 氧化物半导体沟道晶体管将成熟的氧化硅和氮化硅薄膜与新型 InGaZnO 材料相结合。 Kioxia 推出的 3D 存储单元结构扩大了垂直间距，使得单位体积内可以堆叠更多的存储单元。 在该工艺中形成的水平晶体管具有超过 30 微安的高导通电流。 它还表现出低于 1 阿安培的超低关断电流，从而最大限度地减少了刷新周期中的能耗。 通过降低刷新功耗，该设计 ...

公众号记得加星标⭐️，第一时间看推送不会错过。 众所周知，苹果公司热衷于垂直整合，尽可能将关键技术节点保留在公司内部，其庞大的定制芯片设 计工作或许是这种模式最恰当的例证。 当然，推理芯片的架构与用于训练人工智能模型的芯片的架构有着根本的不同，前者更加注重延迟和 吞吐量。人工智能推理芯片还采用了精度较低的基于数学的架构，例如 INT8。 因此，鉴于此背景，我们可以合理推断，苹果和博通在推进 Baltra 的整体设计过程中，可能会重点 关注这些方面。 与此同时，苹果庞大的定制芯片产品线仍在不断扩展。除了广为人知的A系列和M系列芯片外，苹果 现在还使用其自主研发的C1调制解调器芯片。此外，这家库比蒂诺巨头可能还会推出一款基于其 Apple Watch专用S系列芯片的衍生产品，用于其计划于明年发布的AI智能眼镜中。 参考链接 https://wccftech.com/apples-ai-server-chip-baltra-likely-to-be-used-primarily-for-ai-inference/ （来源 ： 编译自wccftech ） 当然，这些定制人工智能芯片的实际部署预计将在 2027 年进行 ...

Core Insights - Google is significantly increasing its orders for Tensor Processing Units (TPUs) from MediaTek, with the order volume exceeding initial plans by multiple times [2] - MediaTek's first TPU, the v7e, is set to enter risk trial production by the end of next season, and it has also secured orders for the next-generation TPU, v8e [2][3] - The collaboration with TSMC is expected to boost MediaTek's production capacity for Google projects, with a projected sevenfold increase in CoWoS capacity by 2027 [2][3] Group 1 - The demand for Google TPUs is driven by strong client needs, leading to an increase in CoWoS capacity from 10,000 to 20,000 units annually for the v7e project [3] - MediaTek's ASIC business is anticipated to contribute significantly to its revenue, with estimates suggesting that the v7e could add over two times its equity in profits by 2027 [2][3] - MediaTek's CEO expressed confidence in the growth of ASIC revenue, targeting $1 billion in cloud-related ASIC revenue by 2026 and potentially reaching several billion by 2027 [4] Group 2 - Meta is exploring a partnership with Google to utilize TPUs for its AI projects, which could challenge NVIDIA's market dominance [6][7] - If successful, Meta plans to start renting TPUs from Google Cloud in 2026 and deploy them in its data centers by 2027, marking a significant shift in its AI infrastructure [6][7] - The potential collaboration has led to a drop in NVIDIA's stock price, reflecting investor concerns about Meta's future chip orders [7] Group 3 - Analysts predict that Google plans to double its TPU production by 2028, with TSMC expected to produce 3.2 million TPUs in 2024, increasing to 5 million by 2027 and 7 million by 2028 [10] - Morgan Stanley estimates that Google could generate up to $13 billion in revenue for every 500,000 TPUs sold to external clients [9] - Google's vertical integration strategy aims to enhance its technological advantages and profitability by developing its AI hardware and software [8][11]

Luminar Technologies股价周一暴跌58%，此前这家激光雷达技术公司申请了第11章破产保护并宣布 计划出售其资产。 公众号记得加星标⭐️，第一时间看推送不会错过。 这家总部位于奥兰多的公司在约91.3%的第一留置权票据持有人和85.9%的第二留置权票据持有人的 支持下进入破产程序。Luminar将"传统债务义务和行业采用步伐"列为挑战其可持续运营能力的因 素。 作为重组的一部分，Luminar已达成协议，将其子公司Luminar Semiconductors Inc. (LSI)以1.1亿 美元现金出售给Quantum Computing Inc.。Quantum Computing的股价在公告后下跌6.8%。 Luminar首席执行官Paul Ricci表示："在过去六个月中，我们采取了有意义的步骤来推动运营纪律， 精简成本结构，并明确战略方向。经过对我们各种选择的全面审查，董事会确定法院监督下的出售程 序是最佳前进道路。" 该公司计划在破产过程中继续运营，保持向客户交付其激光雷达硬件和软件。Luminar已提交动议， 寻求授权为LSI股权和其激光雷达业务进行销售程序，预计交易将在2026年 ...

Core Viewpoint - Intel's AI strategy is shifting focus towards ASIC and edge AI, aiming to regain competitiveness in the AI sector where it lags behind Nvidia and AMD [2][6]. Group 1: AI Strategy and Market Position - Intel has historically dominated general computing and server-level computing but has struggled in the AI domain, as acknowledged by former CEO Pat Gelsinger [2]. - The company is working on a power-optimized GPU for inference as part of its edge AI strategy, with products like Meteor Lake, Lunar Lake, and the upcoming Panther Lake series [2]. - Intel plans to expand its edge product line with Crescent Island, a processor focused on inference, featuring LPDDR5X memory [2]. Group 2: ASIC Business Development - Intel has established a dedicated ASIC department under Srini Iyengar, aiming to provide customized chips for specific workloads, similar to Google's TPU and Amazon's Trainium [3][5]. - The ASIC business is expected to play a crucial role in Intel's operations, potentially opening new revenue streams and addressing past missteps in the AI market [5][6]. - The company aims to offer a "one-stop" solution for customers seeking custom AI chips, leveraging its chip technology expertise and internal foundry services [6]. Group 3: Competitive Landscape and Future Prospects - Intel's ASIC business is positioned to compete with companies like Broadcom and Marvell, focusing on custom network ASIC chips for network-intensive workloads [4][6]. - The centralized engineering group (CEG) is expected to reduce costs by integrating design services with manufacturing and packaging [6]. - If executed effectively, the custom chip business could become a significant revenue source for Intel, enhancing its position in the AI supply chain [7].

Core Insights - The article discusses the significant restructuring of the semiconductor industry in the United States over the past three years, driven by initiatives like the CHIPS Act and state-level competition to attract wafer fabs and R&D centers [2][11]. Group 1: California Semiconductor Ecosystem - California is home to the largest integrated cluster of design, software, IP, and equipment, primarily located in the San Jose-Santa Clara-San Diego corridor, which focuses on GPU, AI, mobile communications, and server SoCs [5][9]. - Major companies in California include NVIDIA, AMD, Broadcom, Qualcomm, and many others, making it a vital hub for chip innovation [9]. - California also leads in semiconductor equipment and materials, with companies like ASML, Lam Research, and Applied Materials contributing to a comprehensive advanced manufacturing ecosystem [10]. Group 2: Arizona as a New Wafer Manufacturing Center - Arizona has become a focal point for wafer manufacturing, with TSMC and Intel establishing significant operations, supported by favorable environmental policies and infrastructure [11][13]. - The state boasts a complete semiconductor ecosystem, including advanced processes, OSAT capabilities, and a robust materials supply chain [14]. - Arizona is positioned to become a major player in U.S. wafer manufacturing, akin to Taiwan's Hsinchu Science Park, aiming to mitigate geopolitical risks in global supply chains [14]. Group 3: Texas as a Center for IDM and Automotive Electronics - Texas is recognized as the largest center for Integrated Device Manufacturers (IDM), microcontroller units (MCU), and automotive electronics in the U.S. [15]. - Key players include Texas Instruments, Samsung, and NXP, with a strong focus on automotive and power semiconductor applications [17]. - The state is evolving into a new growth hub for semiconductors, driven by the convergence of automotive, power, and AI technologies [18]. Group 4: Northeast Research Corridor - The Northeast region, encompassing New York, Massachusetts, and New Jersey, is a leading area for semiconductor research, hosting prestigious institutions like MIT and Harvard [19][20]. - Companies such as IBM and GlobalFoundries are pivotal in advancing semiconductor technology and manufacturing capabilities in this corridor [20]. - The region's ecosystem is further strengthened by a network of universities and research institutions, fostering innovation in materials, EDA, and photonics [24][27].

公众号记得加星标⭐️，第一时间看推送不会错过。 如今，MEMS器件已成为我们日常生活中不可或缺的一部分，尽管它们的存在常常被人忽略，其应用 范 围 从 耳 机 中 的 麦 克 风 到 用 于 工 业 天 线 稳 定 的 惯 性 传 感 器 ， 无 所 不 包 。 根 据 Yole Group 发 布 的 《2025年MEMS产业现状报告》，预计到2030年，MEMS器件市场规模将达到350亿件，营收将达到 192亿美元，2024年至2030年的复合年增长率（CAGR）为3.7%。 在这种结构中，水和灰尘被防止滞留在膜片和背板之间，从而实现降噪音频采集，信噪比范围为 68- 75 dB(A)。 值此年末之际，正是回顾MEMS领域创新发展历程的绝佳时机。这个成熟的行业非但没有放慢脚步， 反而不断展现出新的创新层面、技术改进和竞争优势。从尖端的麦克风架构到新一代定时器件和性能 卓越的汽车传感器，MEMS技术正悄然而坚定地重塑着我们日常生活中所依赖的系统。 芯片之外是什么？设计选择、工艺创新和成本动态如何推动MEMS前沿价值的提升？这些进步又揭示 了该行业未来的发展方向？ 尽管MEMS产业已经相当成熟，但技术仍在不断 ...

Core Viewpoint - Broadcom's CEO Hock Tan indicated that silicon photonics technology will not have an immediate impact on data centers, suggesting that the transition to this technology will take time and require two waves of innovation before it becomes essential [2]. Group 1: Silicon Photonics Technology - The CEO believes that the adoption of silicon photonics technology will occur in the future, but it is not imminent. The first wave involves expanding copper interconnect technology for rack-level systems, followed by a second wave of pluggable optical devices that combine electronic and photonic components [2]. - Tan emphasized that the transition to silicon photonics is inevitable but will require time, stating that Broadcom is preparing for this shift through necessary R&D efforts [2]. Group 2: AI Hardware Demand - Broadcom is currently focused on producing AI hardware to fulfill a backlog of orders valued at $73 billion, with over $50 billion coming from large-scale customers ordering custom AI accelerators, referred to as "XPUs" [3]. - The CEO expressed confidence in the sustained demand for AI hardware, countering claims of a cooling market, and highlighted a recent $1 billion XPU order from a new customer and an additional $11 billion order from Anthropic [3]. Group 3: Financial Performance - In the fourth quarter, Broadcom's semiconductor business revenue reached $11 billion, a 35% year-over-year increase, with $6.7 billion attributed to AI products, marking a tenfold growth in AI hardware revenue in less than three years [4]. - The company's infrastructure software business generated $6.9 billion in revenue, up 19% year-over-year, primarily driven by VMware software sales, which are expected to maintain low double-digit growth until 2026 [4]. - The CEO projected that the semiconductor business will continue to drive rapid growth, with first-quarter revenue expectations for fiscal year 2026 set at $19.1 billion, reflecting a 28% increase [4]. Group 4: Stock Market Reaction - Following the earnings announcement, Broadcom's stock price initially rose by approximately 3% but quickly fell by 5% [5].

公众号记得加星标⭐️，第一时间看推送不会错过。 相比之下，半导体行业正经历着历史性的上涨，这主要得益于全球人工智能（AI）的蓬勃发展。随着 对人工智能服务器和数据中心投资的增加，半导体需求也随之激增，这预示着超级周期的到来。 截至 11 月，芯片出口额达到创纪录的 1526 亿美元，占该国出口总额的 28.3%，几乎是 2000 年代 初的三倍。 出口日益集中于单一行业，这引发了人们的担忧：由需求和价格飙升推动的半导体超级周期，从长远 来看可能无法持续。 专家警告称，在这种不平衡的出口结构下，半导体市场的任何降温都可能损害韩国的经济稳定。由于 半导体对全球IT周期和人工智能相关投资的变化高度敏感，其下滑可能产生广泛的影响，同时对出 口、经济增长、就业和财政状况造成压力。 周日公布的数据显示，尽管韩国出口总额有所增长，但韩国出口增长越来越依赖半导体，导致出口结 构两极分化加剧。 尽管该国今年的出口额有望首次超过 7000 亿美元，但少数行业的强劲表现掩盖了更广泛的出口基础 的普遍疲软，许多行业正陷入长期低迷。 据韩国贸易、工业和资源部统计，今年1月至11月期间，韩国15种旗舰出口商品中只有5种实现了同 比增长， ...

公众号记得加星标⭐️，第一时间看推送不会错过。 主流PC市场正面临内存供需长期失衡的局面，据报道这种情况将持续到2028年。SK海力士的内部分 析显示，"商品级"DRAM的增长将十分有限，无法满足市场需求。我们已经知道DRAM价格高得离 谱，但这种情况似乎已经失控，使得大众难以买到价格合理的PC。 用户@BullsLab分享了据称是SK海力士内部分析的截图。该分析预测，除高带宽内存（HBM）和 SOCAMM模块外，普通DRAM的增长至少在2028年之前仍将受到限制。这是因为主要的内存制造商 已经将重心转移到满足AI服务器的需求上，而面向消费市场的产能出现显著增长的可能性仍然很低。 据报道，现有供应商的库存已降至历史低位，这进一步加剧了分配压力。报告显示，SK海力士等内 存制造商采取了保守的产能扩张策略，更注重维持盈利能力，而非向市场大量投放新的DRAM产品。 服务器DRAM的需求几乎呈指数级增长，预计明年增长速度将更快。 据全球最大的存储器制造商之一称，DRAM 供应紧张的局面预计将持续到 2028 年。 据估计，服务器份额将从2025年的38%飙升至2030年的53%。由于人工智能的蓬勃发展，云服务提 供商 ...