半导体行业观察
Search documents
云巨头,为何倒向英伟达?
半导体行业观察· 2026-02-19 02:46
公众号记得加星标⭐️,第一时间看推送不会错过。 当Meta Platforms与英伟达(Nvidia)达成大规模 AI 系统交易时,通常意味着该公司此前的某些开 放硬件计划已无法满足紧迫的算力需求。这与项目延期不完全是一回事,但效果是一样的。提醒一 下,这类情况我们掌握的数据并不多,而如今这家社交网络巨头、AI 模型厂商与 AI 硬件巨头英伟 达之间宣布的巨额合作,已是第三起。 这笔交易远比 Meta 上一次与英伟达的合作规模更大,对英伟达而言价值至少数百亿美元,再加上原 始设计制造商将英伟达芯片集成到 Meta 系统中所能获得的额外收益。 在前两起案例中(几乎可以确定第三起新案例也是如此),一旦 AI 算力需求足够紧迫,Meta 便愿 意放弃自家开放计算项目(OCP)的设计方案。 在超大规模云厂商与大模型厂商中,Meta 的定位略有不同:它不只是为搜索加入 AI 能力,或是打 造能与 OpenAI、Anthropic 等抗衡的通用大模型,同时还高举开源大旗(至少目前是这样)。该公 司还运营着庞大的高性能集群集群,作为旗下各类服务的推荐引擎。这些系统需要CPU 与加速器紧 密耦合,让加速器能直接访问 CPU ...
10BASE-T1S,悄然崛起
半导体行业观察· 2026-02-19 02:46
Core Viewpoint - The article discusses the emergence and significance of the 10BASE-T1S standard in the automotive and industrial sectors, highlighting its advantages over traditional communication protocols like CAN and LIN, particularly in the context of evolving vehicle architectures and increasing sensor integration [2][3][5][33]. Group 1: Background and Industry Opportunity - 10BASE-T1S is a new physical layer standard for automotive and industrial control, established by IEEE 802.3cg in February 2020, featuring a transmission rate of 10 Mbps and designed for short-distance connections of up to 25 meters [3]. - The shift towards zonal architecture in vehicles, which consolidates multiple functions into fewer controllers, necessitates a more efficient communication protocol like 10BASE-T1S to manage the increasing number of sensors and actuators without overwhelming bandwidth [5][6]. - The trend of unifying vehicle networks under Ethernet protocols is driven by the need for over-the-air updates, centralized data processing, and software upgrades, making 10BASE-T1S a suitable choice for modern automotive applications [6][10]. Group 2: Advantages of 10BASE-T1S - 10BASE-T1S supports multi-drop connections, allowing multiple devices to connect over a single pair of wires, significantly reducing wiring complexity and costs, which is crucial for electric vehicles [6][10]. - The limitations of CAN FD in terms of scalability and protocol integration are becoming apparent, prompting manufacturers to consider 10BASE-T1S for long-term platform design [6][10]. - Compared to traditional buses like CAN, RS-485, and RS-232, 10BASE-T1S offers a more integrated and efficient solution, addressing issues of protocol fragmentation and complexity in industrial applications [11][12]. Group 3: Competitive Landscape - Major chip manufacturers are actively developing 10BASE-T1S products, with strategies ranging from simplifying Ethernet integration to completely rethinking edge node architectures [12][19][20]. - Microchip and TI focus on making Ethernet as user-friendly as CAN, integrating MAC and PHY in single packages to facilitate easier adoption in low-end microcontrollers [13][14]. - ADI's E²B technology aims to centralize control by offloading software burdens from edge nodes, enhancing communication efficiency and reducing system costs [19]. - Infineon and NXP emphasize high integration and safety for complex zonal architectures, with Infineon’s BRIGHTLANE switch and NXP’s TJA1410 designed for reliability in safety-critical applications [20][26]. Group 4: Future Outlook - The adoption of 10BASE-T1S is seen as a gradual transition rather than an outright replacement of existing protocols like CAN and LIN, driven by the need for a unified communication framework in the software-defined vehicle era [33]. - The article concludes that 10BASE-T1S is a crucial component in the evolution towards a fully integrated Ethernet architecture in vehicles, addressing the challenges of protocol fragmentation and enhancing overall system efficiency [33].
ARM,失宠了
半导体行业观察· 2026-02-19 02:46
公众号记得加星标⭐️,第一时间看推送不会错过。 英伟达本周已出售所持 ARM 的最后剩余股份,与几年前曾试图收购该公司的情景已相去甚远。 英伟达与 ARM 的合作,在当代 AI 基础设施建设中至关重要 —— 正是凭借 ARM 的 CPU 架构,英 伟达才得以推出 Grace Hopper、Blackwell 等系列重磅产品。更重要的是,ARM 还将在英伟达即将 推出的Vera CPU中扮演关键角色,这类处理器的重要性正在急剧提升。 据彭博社报道,根据最新提交给美国 SEC 的文件,英伟达已出售其持有的 ARM 剩余股份,价值约 1.4 亿美元。耐人寻味的是,此举恰好发生在ARM 在未来 AI 竞赛中的地位开始受到质疑的节点。 很多人尚未意识到:CPU 近期正迎来空前重要的地位提升。原因在于推理 workload,尤其是智能体 (agentic)相关负载—— 这类场景的重心正从 GPU 计算转向更依赖 CPU 的任务,例如工具调用、 API 请求、内存查找与调度逻辑。 这一转向已非常明显:英特尔、AMD 均表示,超大规模云厂商对其数据中心 CPU 需求暴增,背后 正是 CPU 整体市场(TAM)的高速扩张。与此 ...
全球最快ADC芯片,发布!
半导体行业观察· 2026-02-19 02:46
公众号记得加星标⭐️,第一时间看推送不会错过。 本周在2026 年 IEEE 国际固态电路会议(ISSCC 2026)上,先进半导体技术研发中心imec发布一款 7 位、175GS/s 模数转换器(ADC)。该产品兼具创纪录的小尺寸(250×250 平方微米)、低转换 能耗,以及目前公开报道中最快的采样速率之一。 基于上述特性,imec 这款 ADC 可满足由人工智能与云计算驱动的数据中心日益增长的吞吐量与处 理需求,同时避免了超高速采样率下常见的芯片面积与功耗大幅激增问题。 在人工智能与云计算应用的驱动下,数据中心光通信网络需要持续升级,以支撑不断提升的吞吐量与 处理需求。但当采样速率超过 100GS/s 时,作为光收发器核心组件的有线通信 ADC 等基础器件往 往会出现面积增大、互联线路更长的问题,并引入寄生效应与能量损耗。 在2024 年 IEEE 国际固态电路会议(ISSCC)上,imec 已通过一项突破性技术应对这一挑战:推出 大规模时间交织斜坡型 ADC 架构,其芯片面积仅为传统方案的一半以下,且具备业界领先的能效表 现。 基于本届 ISSCC 发布的 ADC 成果,imec 正基于3 纳米工艺开 ...
化合物半导体,日益重要
半导体行业观察· 2026-02-19 02:46
公众号记得加星标⭐️,第一时间看推送不会错过。 随着各行业转向使用替代材料来提供比硅更强大的功率、更快的速度和更高的效率,化合物半导体 正逐渐成为主流。 与几十年来一直是芯片制造主要材料的硅不同,化合物半导体是由 SiC、GaN、GaAs 或 InP 等材 料构成的,这些材料作为关键组件,使高性能成为可能。 SiC 和 GaN 在电力电子领域处于领先地位,GaAs 和 GaN 广泛用于射频系统,InP 和 GaAs 是光 子学和激光器的关键,而 GaN 和 GaAs 仍然是照明和显示技术(如微型 LED)的关键。 化合物半导体器件的日益普及在衬底层面得到了清晰的体现,该领域的投资正在不断扩大。Yole Group在其最新报告中预测,2025年至2031年间,整个市场将以14%的复合年增长率增长,规模将 从13亿美元增长至28亿美元,增长超过一倍。 外延晶圆(外延片)是价值链中的另一个关键环节。开放式外延片市场展现出强劲的增长势头,外 延片厂商与包括集成器件制造商 (IDM)、无晶圆厂和代工厂在内的不同参与者紧密合作。正如化合 物半导体报告深入分析的那样,该市场预计将与衬底市场同步增长,从 2025 年的 11 ...
CS转行EE,可行吗
半导体行业观察· 2026-02-19 02:46
公众号记得加星标⭐️,第一时间看推送不会错过。 为了解决芯片行业的人才短缺问题,人们正在开发和测试各种新方法,从更广泛地部署人工智能工具 到对工程专业毕业生进行核心学习领域以外的交叉培训。 在人工智能领域,新型工具能够帮助工程师更高效地设计和验证半导体硬件,从而弥补部分不足。大 型语言模型和自然语言智能体人工智能工具可以被训练成定制化的助手。随着需要更先进的芯片来驱 动用于辅助芯片设计的人工智能,这项技术将持续发展、演变和融合,形成一个循环。 与此同时,学术界正在尝试各种不同的方法来填补人才缺口,包括缩短培训时间、加强培训和交叉培 训,以及利用机器学习工具、大型语言模型、多智能体人工智能和混合专家人工智能来训练软件工程 师,使其能够胜任硬件工程师的工作。教会软件工程师如何设计硬件或许可行,但这并非易事。 丛教授和他的团队发表了许多关于如何帮助软件工程师设计硬件的技术论文,他在演讲中得出了这样 的结论:"其中一个重要信息是,精心设计的AI/ML工具绝对可以帮助芯片设计。我向大家展示了我 们如何将图神经网络(GNN)与逻辑学习模型(LLM)、捕获设计层次结构、程序转换、从FPGA到 ASIC的任务转移以及利用专家 ...
闪存芯片,越来越缺
半导体行业观察· 2026-02-18 01:13
公众号记得加星标⭐️,第一时间看推送不会错过。 如果你想涉足DRAM和闪存市场,最好做好承受过山车般波动的准备。因为这些行业的繁荣与萧条周 期,真可谓惊心动魄。 正当GenAI市场在2022年11月迎来ChatGPT主流化的关键时刻时,此前近三年受新冠疫情推动的个 人和数据中心基础设施建设热潮不仅没有消退,而且随着整个IT行业需求的萎缩,DRAM和闪存的价 格也下跌了一半甚至更多。内存和闪存厂商遭受重创,库存积压严重。 如今推动 GenAI 繁荣发展的超大规模数据中心运营商、云构建商和模型构建商可能都希望自己有时 光机,因为现在对 DRAM 和闪存的需求远远超过了供应,价格再次飙升至天价。 在DRAM方面,全球超过一半的服务器需要配备数百GB的HBM堆叠式内存,以满足数百万台服务器 的需求。为了制造八个高HBM3内存堆叠,每个最终能正常工作的堆叠都需要消耗三个DRAM芯片, 因为堆叠过程中经常会出现问题,而且无法将废弃堆叠中的内存拆解并重新利用。堆叠越高,获得正 常工作的堆叠就越困难,良率也越低,而且随着每一代新的HBM的出现,其固有的良率也在不断降 低。因此,尽管HBM需求旺盛,但它会消耗大量的芯片,而这些 ...
不止GPU,Meta扫货英伟达CPU
半导体行业观察· 2026-02-18 01:13
Core Viewpoint - Nvidia and Meta are expanding their long-term partnership, with Nvidia providing millions of Blackwell and Rubin GPUs, CPUs, and networking products for AI model training and operation in Meta's data centers [2][5]. Group 1: Partnership Details - The partnership involves the deployment of Nvidia's CPUs and millions of GPUs in Meta's data centers, leveraging Nvidia's cloud partners for additional resources [2][5]. - Nvidia's CEO emphasized Meta's unique capability to deploy AI at scale, combining cutting-edge research with industrial-grade infrastructure for billions of users [2][5]. - Meta is launching its first large-scale Grace CPU servers and plans to introduce the Vera CPU system by 2027, which will not include GPUs [2][5][6]. Group 2: Market Implications - This move may pose challenges for Intel and AMD, who have dominated the CPU server market for decades [3]. - Meta will also utilize Nvidia's confidential computing technology in its WhatsApp application for private data processing [3]. - Concerns have arisen regarding AI-related stocks, with Meta's stock down 3.3% and Microsoft's stock down over 17% since January 1, 2023 [3][4]. Group 3: Competitive Landscape - Nvidia's stock has decreased by over 1% this year, while AMD's stock has dropped over 5% [4]. - Analysts believe Nvidia is unlikely to lose its lead in AI due to the versatility of GPUs compared to more specialized chips like Google's TPU and Amazon's Trainium [4][6]. - The collaboration represents the first large-scale deployment of Nvidia Grace, enhancing performance through joint design and software optimization [6].
Musk招聘工程师,开建超级晶圆厂
半导体行业观察· 2026-02-18 01:13
公众号记得加星标⭐️,第一时间看推送不会错过。 特斯拉首席执行官埃隆·马斯克在社交媒体上发布了一则招聘启事,寻求在韩国发展半导体行业的人 才。 "我正在努力弄清楚的事情之一是——我们如何才能生产出足够的芯片?"马斯克周四在特斯拉年度股 东大会上说道。 特斯拉目前依赖于代工芯片制造商台积电。以及三星电子。马斯克表示,他也在考虑与美国芯片公司 英特尔合作,以生产其芯片设计。他还表示,正在考虑与英特尔合作。 "但即使我们从供应商那里推断出芯片生产的最佳情况,仍然不够,"他说。 特斯拉可能需要建造一座"巨型"芯片制造厂,马斯克称之为"特斯拉超级芯片制造厂"。"我想不出还 有什么其他办法能达到我们所需的芯片产量。" 微芯片是几乎所有现代技术的大脑,包括从智能手机等消费电子产品到大型数据中心的一切,随着人 工智能的蓬勃发展,对微芯片的需求也急剧增长。 16日(当地时间),CEO马斯克在其X(原Twitter)账号上转发了特斯拉韩国的招聘公告,并附上了 几个太极旗表情符号。他在评论中写道:"如果你住在韩国,想从事半导体设计、制造或软件方面的 工作,那就加入特斯拉吧。" 此前,特斯拉韩国公司发布了"人工智能(AI)芯片设计工 ...
芯片设备,销量大增
半导体行业观察· 2026-02-18 01:13
Core Viewpoint - The global chip tool manufacturers are expected to achieve double-digit revenue growth for the first time in three quarters, driven by increasing demand for AI computing capabilities [2]. Group 1: Revenue Growth Expectations - Nine major chip tool suppliers based in Japan, the US, and Europe are projected to see a 16% year-over-year increase in quarterly revenue, surpassing the previous quarter's growth of 8% [2]. - The expected net profit for the quarter is anticipated to grow by 20%, marking the eighth consecutive quarter of double-digit profit growth, following a 26% increase in the prior three months [2]. Group 2: Market Dynamics - The primary driver of this growth is the capacity expansion by chip manufacturers, with TSMC announcing plans to increase capital investment to a historical high by 2026, and SK Hynix and Samsung also planning to boost investments this year [2]. - ASML expects a revenue increase of approximately 10% for the January to March quarter, with a notable improvement in market outlook as stated by CEO Christoph Hock [3]. Group 3: Regional Sales and Market Trends - Tokyo Electron has raised its six-month new equipment sales forecast in response to accelerated delivery demands, with many clients opting to retrofit existing equipment to maintain operations [4]. - The global market for front-end equipment is expected to grow by 15% to 20% this year, with DRAM chip tools being a significant growth driver [4]. - Despite ongoing US-China tensions, sales of chip manufacturing tools to China have increased, with total revenue from eight companies in the Chinese market rising by 8% to $10.2 billion, accounting for over 30% of total sales [4]. Group 4: Challenges and Regulatory Environment - ASML's sales to China have grown by about 60%, primarily due to the sale of older generation lithography equipment not subject to export restrictions [5]. - The Chinese market's share in ASML's revenue is expected to decline from 33% in 2025 to around 20% [5]. - The US government has raised concerns regarding chip tool manufacturers profiting from sales to China, leading to legal actions such as Applied Materials agreeing to pay a $252 million fine for illegal exports [5].