公众号记得加星标⭐️，第一时间看推送不会错过。 来源：内容 编译自 restofworld 。 三星电子曾是全球领先者，但在争夺人工智能供应主导权的激烈半导体竞争中却失利；该公司还失去 了工程师，留下的员工只能长时间高强度地工作以弥补空缺；工程师们表示，他们被要求伪造数据， 导致缺陷漏出，影响芯片质量。 午夜时分，在首尔郊外三星电子的半导体办公室里，芯片设计工程师韩基白（Han Ki-bak）看到一位 同事在连续数月通宵工作后晕倒了。韩基白惊呆了，动弹不得。 "因为我熬了那么多夜，我都疯了。我没有冲过去帮她，只是坐在椅子上，心想'我该怎么办？'"他在 2020年接受《restofworld》采访时回忆了这件事。 当医护人员赶到时，韩先生茫然地看着这一切。 10名三星现任和前任工程师向《restofworld》透露，由于工作时间长、工作量大以及奖金低于竞争 对手，许多三星工程师正转投韩国芯片制造商SK海力士。其中一些人则转投了美国的美光科技和英 特尔，以及长鑫存储和长江存储等中国竞争对手。他们表示，在人手短缺、本已捉襟见肘的团队中， 倦怠感正在加剧。除韩某外，所有员工都要求使用假名，因为他们担心遭到公司的报复。 ...

公众号记得加星标⭐️，第一时间看推送不会错过。 来源：内容 编译自 NRC 。 如何粉碎一块薄饼？这个问题困扰着阿姆斯特丹纳米光刻高级研究中心 (ARCNL) 的研究员迪翁·恩 格尔斯 (Dion Engels)。 这里所说的薄饼指的是被粉碎的锡滴，它在 ASML 的光刻机中每秒被引爆五万次。这会产生等离子 体，发射出极紫外 (EUV) 光：一种极紫外辐射，将高度精细的芯片图案投射到硅片或晶圆上。 集成的晶体管越多，芯片的性能就越强。得益于 EUV 光刻机，如今用于手机或人工智能数据中心的 最先进处理器上的芯片设计线间距仅为几十纳米——百万分之一毫米。 ASML 与美国 Cymer 实验室合作研究 EUV 技术长达二十年，发现粉碎锡滴可以产生更多的 EUV 光。其结果是：生产英伟达、苹果、三星或英特尔先进芯片的机器很快就会更加高效地运转。 芯片元件尺寸缩小的速度正在放缓，因为芯片设计变得越来越高：制造商正在将芯片元件粘合在一 起，并想出更智能地排列晶体管的方法，例如从底部供电。 ASML 系统的许多基础设计都诞生于阿姆斯特丹东区的科学园。ARCNL 成立于十年前，与阿姆斯特 丹大学合作成立。其推动力来自 AS ...

公众号记得加星标⭐️，第一时间看推送不会错过。 来源：内容 编译自 eetjp 。 索尼集团透露，由于2024年主要客户销售额低于预期，以及中国市场高端领域竞争加剧，导致市场份 额与上一年持平，预计2025年实现60%市场份额的目标将推迟。 索尼集团影像与传感解决方案部门（I&SS）预测2024财年的销售额将达到1.799万亿日元，同比增长 12%，营业利润将达到2611亿日元，同比增长35%，均创历史新高。销售额的增长得益于有利的汇 率、产品结构的改善以及移动设备传感器销量的增长。 索尼半导体解决方案公司（以下简称"索尼半导体"）总裁指田真司在2025年6月13日发布的视频中解 释道："图像传感器业务一直以2025年实现60%的市场份额为目标，但2024年我们未能实现超过市场 的销售额增长，市场份额保持平稳。" 具体来说，2024年的实际业绩将与上一年持平，为53%。预计 2025年将达到56%。 索尼半导体CFO高野康弘解释称，2024年公司市场份额持平主要有两个原因，一是主要客户的销售 额没有达到预期，二是中国高端领域与对手的竞争加剧，"所以我们在这个领域有点挣扎"。 同日（6月13日），索尼半导体也公 ...

Core Viewpoint - The recent surge in DDR4 DRAM spot prices, with increases of nearly 8% in a single day, indicates a significant market shift, driven by supply constraints and increased demand from OEM/ODM manufacturers [1][4]. Price Trends - DDR4 8Gb (1G×8) 3200 spot price rose from an average of $2.73 on May 30 to $3.775, marking a 38.27% increase in just half a month [2]. - Since March 31, the DDR4 8Gb price has skyrocketed from $1.63 to nearly $3.775, representing an increase of approximately 132% [3]. - The DDR4 16Gb (1G×16) 3200 spot price increased from $6.1 on May 30 to $8.2, a rise of 34.42% in June alone, and over 107% since March 31 [3]. Market Dynamics - The current DDR4 spot prices have reached levels comparable to Q1 2022, where major suppliers like Nanya Technology and Winbond reported substantial profits [4]. - With major manufacturers like Samsung and Micron halting DDR4 supply, concerns over future supply shortages have led to aggressive purchasing behavior in the market [4]. - Nanya Technology is reportedly expanding its DDR4 production capacity, indicating a recognition of the significant market opportunity for DDR4 [5]. Supplier Insights - Nanya Technology and Winbond are positioned to benefit from the rising DDR4 prices, with DDR3 and DDR4 accounting for over 80% of Nanya's revenue [5]. - Memory IC design firms like Etron Technology and Gigabyte Technology are also beginning to receive customer orders, suggesting a positive outlook for their operations amid the DDR4 price surge [5].

Core Viewpoint - The article discusses the evolution of smartphones into intelligent, context-aware devices through the integration of agentic AI, which enhances their capabilities beyond simple communication tools [1][3]. Hardware Challenges - The main challenge in implementing agentic AI on smartphones lies in hardware limitations, including battery life, processing power, and memory constraints [3]. - To meet the demands of agentic AI, significant upgrades in smartphone hardware components such as processors, memory, storage, battery, sensors, and thermal management are necessary [3]. Memory System Requirements - There is a growing demand for advanced memory subsystems to efficiently deliver data for AI applications on devices [5]. - The current standard LPDDR5X offers speeds up to approximately 10.7 Gbps, while the upcoming LPDDR6 standard promises faster bandwidth (14.4 Gbps+) and improved power efficiency [5]. Key Technologies for Memory Solutions 1. **Memory Processing (PIM) Architecture**: PIM integrates computing functions directly into memory, significantly reducing latency and power consumption, although standardization is still developing [7]. 2. **Wide I/O Interfaces and Advanced Packaging**: These methods enhance bandwidth and assist in thermal management, potentially allowing OEMs to offload DRAM to separate packages for AI-intensive workloads [11]. Importance of Quantization and Collaboration - Techniques like quantization are crucial for introducing GenAI into smartphones by reducing memory and computational demands while maintaining accuracy [15]. - Collaboration among SoC designers, memory and storage suppliers, OEMs, OS developers, and AI researchers is essential to optimize hardware and software for edge AI [16]. Industry Call to Action - The article emphasizes the need for a collective vision and investment in next-generation memory, storage, and packaging technologies to keep pace with rapid AI advancements and unlock the transformative potential of agentic AI in smartphones [17].

Core Viewpoint - The competition between TSMC and Samsung in the production of advanced 2nm chips is intensifying, with TSMC currently having a higher yield rate, which poses a challenge for Samsung in attracting orders [1][2][3]. Group 1: TSMC's 2nm Progress - TSMC has begun receiving orders for its 2nm process and is expected to start production in H2 2023 at its facilities in Hsinchu and Kaohsiung [1][3]. - The 2nm process utilizes Gate-All-Around (GAA) architecture, which is anticipated to improve performance by 10% to 15% and reduce power consumption by 25% to 30% compared to the 3nm process [1][3]. - Major clients for TSMC's 2nm chips include Apple, Nvidia, AMD, Qualcomm, and MediaTek, with MediaTek's CEO hinting at a product design completion by September 2023 [1][3]. Group 2: Samsung's 2nm Challenges - Samsung aims to start producing 2nm chips in H2 2023, likely for its upcoming Galaxy S26 flagship device, but its yield rate is reportedly around 40%, significantly lower than TSMC's 60% [2][3]. - Samsung is leveraging its previous experience with GAA architecture to improve its 2nm yield rates and attract orders from major tech companies [2]. Group 3: Production Capacity and Future Plans - TSMC's 2nm process is entering trial production, with plans to ramp up to a monthly capacity of 30,000 wafers by Q4 2025, and potentially expand to 120,000 to 130,000 wafers per month by 2027 [4]. - TSMC is investing over 1.5 trillion NTD to expand its facilities in Hsinchu and Kaohsiung, aiming to create the world's largest semiconductor manufacturing hub [4].

公众号记得加星标⭐️，第一时间看推送不会错过。 玻璃中介层支持嵌入基板的芯粒与直接堆叠于顶部的芯粒（chiplets）之间的3D堆叠，这是硅中 介层无法实现的。在本研究中，我们通过关键系统级指标（包括面积、线长、信号完整性、电源 完整性和热完整性）论证了玻璃中介层相较于硅中介层在这种堆叠方式下的优势。我们利用芯粒 和中介层的GDS版图设计以及签核仿真实现了这一目标。 实验表明，玻璃中介层相比硅中介层可实现2.6倍的面积优化、21倍的线长缩短、全芯片功耗降 低17.72%、信号完整性提升64.7%、电源完整性改善10倍，但温度会升高15%。 引言 如今及未来，提升高复杂度系统良率的一个可行方法是将系统划分为"芯粒"。这些芯粒需集成以构成 完整系统。根据物理结构，芯粒集成有两种类型：2.5D中介层集成和3D堆叠集成。2.5D集成因允许 在中介层上集成多个现成芯粒或复用不同技术节点的知识产权 (IP)（异构集成），成为颇具吸引力的 选择。在2.5D集成中，芯粒以倒装芯片方式并排置于中介层封装顶部，如图1 (a) 所示。此外，它们 通过再分布层 (RDL) 连接，RDL是无源中介层基板上的金属层，用于提供芯粒间的横向 ...

Core Viewpoint - Nvidia's latest generation of graphics cards, particularly the RTX 5060, is underwhelming in performance and specifications, resembling older models like the RTX 5050 rather than meeting the expectations for a 60 series product [1][3][5]. Group 1: Product Specifications - The RTX 5060 has a core configuration and memory bandwidth that aligns more closely with the RTX 5050, indicating a significant downgrade in expected performance for a 60 series card [5][10]. - Historical data shows that the core count for the RTX 5060 is only 18% of the flagship model, which is significantly lower than the average for 60 series GPUs, which typically range around 36% [10][12]. - The RTX 5060's specifications are comparable to older models like the GTX 750 and GTX 1050, suggesting a regression in performance standards for the 50 series [10][12]. Group 2: Pricing and Value - The pricing strategy for the RTX 5060 reflects a trend where Nvidia is charging higher prices for lower-performing products, with the RTX 5060 priced at 15% of the flagship GPU, which is a deviation from historical pricing norms [18][21]. - The RTX 5060 is positioned as the second cheapest 60 series model in seven generations, yet it offers performance that is more akin to a 50 series card, leading to perceptions of value erosion among consumers [21][35]. - The increase in production costs has not been matched by corresponding improvements in GPU specifications, leading to higher prices for lower performance [52][57]. Group 3: VRAM and Memory Bandwidth - The RTX 5060 features only 8GB of VRAM, which is 25% of the flagship model's capacity, falling short of historical averages where 60 series GPUs typically offered around 41% of the flagship's VRAM [32][31]. - The memory bandwidth for the RTX 5060 is significantly lower than that of previous generations, indicating a trend of reduced performance capabilities in newer models [23][27]. - The overall trend shows that Nvidia's GPUs are not keeping pace with the memory requirements of modern gaming, particularly when compared to the specifications of contemporary gaming consoles [48][40]. Group 4: Market Dynamics and Competition - Nvidia's strategy appears to be driven by a desire to maximize profits from AI GPU sales, leading to a compromise in the quality and performance of gaming GPUs [51][57]. - The lack of meaningful competition in the GPU market has allowed Nvidia to maintain high profit margins while passing increased production costs onto consumers [57][58]. - The introduction of AMD's Radeon RX 9000 series may begin to exert pricing pressure on Nvidia, potentially benefiting consumers in the gaming GPU market [58].

公众号记得加星标⭐️，第一时间看推送不会错过。 来源：内容来自technews。 博通（Broadcom）凭借客制化AI 芯片（AI ASIC）业务异军突起，股价在过去两个月狂飙70%，市 值突破1 兆美元大关，超越沃尔玛和特斯拉，跃升美国第7 大最有价值的上市公司。 这间相对低调的半导体巨头，成功的故事始于几年前的一个「副业」。 帮Google制造芯片，无心插柳却迎来大爆发 当时，博通开始协助Google制造AI芯片，这项业务虽然有趣但并不特别赚钱。 然而，AI热潮彻底改变了游戏规则！ 分析师们普遍认为，博通现在已掌握了包含Google母公司Alphabet、Meta，以及TikTok母公司字节 跳动等3大客户。 根据摩根士丹利估计，ASIC芯片市场规模将从去年的120亿美元激增至2027年的约300亿美元。而博 通也释出乐观预期，最新财测显示，第3季来自AI处理与网路芯片的营收将达51亿美元，较去年同期 成长60%，占整体预估营收的三分之一。 然而，博通6月5日公布第2季财报，尽管营收150亿美元、每股盈余1.58美元均优于预期，股价却意 外下跌超过3%。 MarketWatch分析指出，投资人可能对公司 ...

Core Viewpoint - The semiconductor industry is increasingly viewed as a critical aspect of national security by various governments, highlighting the complex global supply chain and the geopolitical tensions surrounding advanced chip production [2][3]. Group 1: Semiconductor Supply Chain Complexity - The semiconductor supply chain involves a highly specialized ecosystem, including advanced software for chip design, silicon wafers, complex manufacturing equipment, and decades of R&D investment [1]. - The global reliance on Taiwan's TSMC, which produces 70% to 90% of the world's advanced transistors, has raised concerns about supply chain vulnerabilities amid geopolitical tensions [2]. Group 2: Geopolitical Impacts and National Strategies - Since the U.S. implemented export controls against mainland China in October 2022, semiconductors have become a strategic focus, with countries aiming to strengthen domestic control over semiconductor production [3]. - The competition for advanced chips, particularly GPUs for AI applications, has surged, exemplified by Nvidia's market capitalization doubling from January 2023 to January 2024 [2]. Group 3: Comparative Strengths of Countries/Regions - No single country can fully control the advanced semiconductor supply chain; the U.S. excels in chip design and tools but lags in manufacturing, while mainland China leads in economic resources and packaging [5]. - The U.S., Japan, Taiwan, and South Korea maintain dominance in key supply chain nodes such as advanced manufacturing and chip design, but high costs and technological barriers hinder new entrants [6]. Group 4: Investment Trends and Future Outlook - Leading semiconductor nations are investing heavily to maintain their competitive edge, with Japan committing over $11 billion to support domestic startups and South Korea planning to build the world's largest semiconductor cluster by 2047 [9]. - The U.S. semiconductor advantage is challenged by export controls, which have impacted equipment manufacturers more severely than chip designers, as seen in Nvidia's projected $5.5 billion loss due to export restrictions [10][11]. Group 5: Emerging Markets and Regional Developments - India aims to become a key player in the semiconductor value chain, leveraging its market size and labor force, despite lagging in critical infrastructure compared to leading nations [12]. - Germany is positioning itself as a major semiconductor player in the EU, with plans to double its market share in chip production by 2030, although recent delays in factory construction raise concerns [13]. Group 6: Regional Strategies and Competitive Advantages - Singapore is capitalizing on its geographical advantages and skilled workforce to maintain a strong position in the semiconductor market, focusing on chip design and advanced packaging [14].