公众号记得加星标⭐️，第一时间看推送不会错过。 随着人工智能服务器的普及，HBM（高带宽存储器）已成为半导体行业的关键参与者。业内外人士 都关注着"基础芯片"（Base Die）的演变。这种基础芯片此前采用 DRAM 工艺制造，但从 HBM4 开始，正在转向代工工艺。三星电子、SK 海力士和美光科技各自制定了不同的战略，以迎接 HBM4 时代的到来。 Base Die，HBM的"大脑" 据存储器行业28日报道，据悉，全球三大存储器公司正在大幅改进Base Die结构，为下一代HBM4的 量产做准备。这是因为，随着AI计算量呈指数级增长，Base Die的作用变得越来越重要，它不仅仅是 简单的存储器堆叠，还决定着信号处理和功率效率。 HBM 是一种存储设备，它通过使用 TSV（硅通孔）互连技术堆叠多个 DRAM 芯片来最大化带宽。 底层芯片接收来自最底层的信号并执行逻辑功能，决定了 HBM 的整体性能和稳定性。它可以被视为 HBM 的"大脑"和"信号控制中心"。 在 HBM3E 之前，该基础芯片都是采用 DRAM 工艺制造的。DRAM 制造商直接设计逻辑电路，并 在自己的 DRAM 生产线上生产。然而，基于平面 ...

Core Viewpoint - The article discusses the development of the world's first "full-band" 6G chip by Chinese researchers, which utilizes photonic technology to achieve transmission speeds exceeding 100 Gbps, laying the foundation for AI-native wireless networks [3][5]. Group 1: Chip Development - The 6G chip integrates the entire frequency spectrum from 0.5 GHz to 115 GHz into a chip the size of a fingernail, which traditionally would require nine separate radio systems [5]. - The chip measures only 11 mm x 1.7 mm and seamlessly switches between millimeter-wave and terahertz communication with low-frequency microwave bands [6]. Group 2: Technical Innovations - Researchers employed photonic-electronic integration technology to overcome the limitations of traditional wireless hardware, which typically operates within a narrow range [7]. - The system achieved 6 GHz frequency tuning in 180 microseconds, significantly faster than the blink of an eye, with a single-channel data rate exceeding 100 Gbps [7]. Group 3: Applications and Future Prospects - The chip is designed for high-demand environments, such as concerts or sports venues, where thousands of devices connect simultaneously [8]. - It establishes a hardware foundation for AI-native networks, capable of dynamically adjusting communication parameters through built-in algorithms to adapt to complex electromagnetic environments [8]. - The goal is to create plug-and-play communication modules no larger than a USB stick, which can be embedded in smartphones, base stations, drones, and IoT devices, potentially accelerating the arrival of flexible and intelligent 6G networks [8].

Core Viewpoint - The article discusses the increased sanctions by the U.S. government against South Korean chip manufacturers Samsung and SK Hynix, particularly focusing on the revocation of their authorization to receive U.S. semiconductor manufacturing equipment in China, which will impact their ability to produce chips in China [2][4]. Group 1: U.S. Government Actions - The U.S. government has revoked the authorization that allowed Samsung and SK Hynix to receive semiconductor manufacturing equipment in China, requiring them to obtain licenses for such purchases [2]. - The revocation will take effect in 120 days, and the U.S. Commerce Department plans to grant licenses for existing operations but not for capacity expansion or technology upgrades [2][5]. - Intel, despite having sold its subsidiary in Dalian, China, is also affected by the loss of authorization [2]. Group 2: Impact on Companies - SK Hynix has stated it will maintain close communication with the U.S. and South Korean governments to minimize business impacts [2]. - The changes may reduce sales for U.S. equipment manufacturers like KLA Corp, Lam Research, and Applied Materials, although these companies have not yet commented [3]. - The revocation of the "validated end-user" status for Samsung and SK Hynix will complicate the process for U.S. suppliers to ship equipment to them [5]. Group 3: Broader Industry Implications - The actions taken by the U.S. may benefit local Chinese equipment manufacturers and Micron Technology, a major U.S. competitor in the memory chip sector [5]. - The ongoing trade tensions between the U.S. and China, including a tariff truce, have significant implications for the semiconductor supply chain and broader economic relations [4]. - The article highlights that the U.S. has a backlog of thousands of license applications for exports to China, including semiconductor manufacturing equipment worth billions [5].

Core Viewpoint - The transformation of the CHIPS Act from a subsidy program to a government equity investment model signifies a major shift in the U.S. semiconductor industry strategy, reflecting a move from "market repair" to "national control" [2][4]. Group 1: Origin of the CHIPS Act - The CHIPS Act was born out of deep anxiety over the decline of U.S. semiconductor manufacturing capabilities, with the U.S. share of global semiconductor production dropping from 40% in 1990 to just 12% by 2020 [4]. - The COVID-19 pandemic exacerbated the semiconductor shortage, leading to significant losses for automakers and revealing critical weaknesses in the U.S. semiconductor supply chain [4]. - The CHIPS Act authorized $52.7 billion for semiconductor manufacturing incentives, aiming to increase U.S. production of advanced chips to 20% by 2030, attracting global semiconductor companies with a total investment commitment of $388 billion [4]. Group 2: Intel's Situation - Intel, the largest beneficiary of the CHIPS Act, received $7.86 billion in subsidies but faced significant operational challenges, including a net loss of $1.654 billion in Q2 2024 and a market cap decline of over 60% [6]. - The company is lagging behind competitors like TSMC and Samsung in advanced process technology, leading to delays in new factory constructions and a restructuring of its leadership [6][7]. - The U.S. government is negotiating to acquire a 10% stake in Intel, marking a shift from support to direct government control, raising legal and ethical concerns [7]. Group 3: TSMC's Challenges - TSMC received $6.6 billion in subsidies for building advanced chip manufacturing facilities in Arizona but faced cultural clashes and labor issues that delayed project timelines [8][9]. - The company had to increase local employee ratios to 85% due to union pressures, which extended the timeline for production ramp-up and increased costs [8]. - TSMC's executives discussed the possibility of returning subsidies if forced to accept government equity, highlighting the tension between government control and corporate autonomy [9]. Group 4: Samsung's Restrictions - Samsung received $4.75 billion in subsidies for a facility in Texas but encountered significant technical challenges, delaying production and leading to workforce reductions [10][11]. - The company faced strict limitations on expanding its production capabilities in China, which could hinder its global competitiveness [10]. - Samsung's subsidy amount was reduced from $6.4 billion to $4.75 billion, signaling the unpredictable nature of government support based on political considerations [11]. Group 5: Micron's Position - Micron, the only U.S.-based memory manufacturer, received $6.1 billion in funding to build new factories but faces challenges in entering the high-bandwidth memory market, where it is significantly behind competitors [12]. - The company is not required to offer equity to the government, which alleviates some control risks but may lead to over-reliance on government support [12]. Group 6: Traditional Manufacturers' Struggles - Texas Instruments received $1.6 billion for new factories but has not garnered the attention that larger projects have, despite the critical role of traditional chips in various industries [13]. - GlobalFoundries, another traditional manufacturer, received $1.5 billion but still faces significant funding challenges and must rely on self-financing for expansion [14]. Group 7: Research Institutions' Dilemma - The National Semiconductor Technology Advancement Center (NATCAST) was allocated $7.4 billion for research but recently had its funding canceled, jeopardizing its operations and future projects [16][17]. - The cancellation of funds highlights the fragility of research institutions that depend on public funding, raising concerns about the sustainability of semiconductor research in the U.S. [17]. Group 8: Overall Assessment of the CHIPS Act - The CHIPS Act's failure is attributed to a fundamental misunderstanding of market dynamics and the complexities of a globalized industry, leading to ineffective resource allocation and a lack of long-term solutions [19][20]. - The act has not only failed to reshape the global supply chain but has also accelerated fragmentation in the industry, increasing costs and complicating global innovation [20].

公众号记得加星标⭐️，第一时间看推送不会错过。 来源：本文综合整理自网络信息。 芯片制造商 Marvell Technology 周四预测第三季度营收将低于华尔街预期，因为经济不确定性和关税担 忧影响了客户支出和整体需求。 该公司生产用于支持人工智能工作负载的定制芯片，其股价在盘后交易中下跌了 8% 以上。 由于华尔街对人工智能寄予厚望，芯片制造商几乎不会失望，因此面临着投资者的严格审查。 Stifel 分析师托尔·斯万伯格 (Tore Svanberg) 表示，由于其他 AI 硬件公司业绩强劲，Marvell 的预测令 人失望。 Marvell 支持 AI 部署，满足超大规模计算的需求。随着企业不断推进技术战略并扩展 AI 工作负载， genAI 的广泛应用正在推动定制芯片的需求增长。 "我们的定制业务表现良好，与第一财年相比，下半年仍有望实现增长。然而，我们预计定制业务的增长 将呈现非线性，第四季度的增长将显著强于第三季度。"首席执行官马特·墨菲在财报发布后的电话会议 上表示。 持续的通货膨胀和经济不确定性导致客户推迟购买，从而导致 Marvell 的汽车、工业和运营商基础设施 终端市场需求疲软。 芯片巨 ...

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 techpowerup 。 Rapidus能否实现日本半导体复兴？ 日本正押下重注，重振其长期衰退的先进芯片制造业。一家名为 Rapidus 的新政府-民间合资企业正致力 于实现一个雄心勃勃的目标：到 2027 年实现 2 纳米逻辑半导体的量产。 20世纪80年代，日本芯片制造商在全球市场占据了主导地位。然而，随着市场的发展以及东亚地区新竞 争对手的出现，日本在先进逻辑芯片制造领域的竞争中逐渐落败。如今，日本已落后世界领先水平多达 二十年。 日本 Rapidus 已成功流片 2 纳米 GAA 测试芯片，计划于 2027 年实现量产。 在 Hot Chips 2025 大会演讲中，该公司概述了新 IIM-1 代工厂的目标（这些芯片将在该工厂生产），并 介绍了其针对习惯于其他晶圆厂的客户（如台积电和潜在的英特尔）的宣传。Rapidus 2 纳米 GAA 测试 芯片采用 ASML 的 EUV 工具制造，该节点已达到最初设定的所有所需电气特性。Rapidus 首席执行官 指出，随着 2027 年的量产，其 IIM-1 晶圆厂每月将生产约 25,00 ...

公众号记得加星标⭐️，第一时间看推送不会错过。 在全球智能芯片技术突破与AI算力爆发双轮驱动下，"机器人+AI"的深度融合正加速重构全球 机器人产业格局——从家庭服务到工业制造，从商用场景到特种作业，机器人的"智能移动"能 力正以前所未有的速度渗透至千行百业，推动产业边界持续拓展。在行业变革关键节点，深耕 机器人核心芯片与算法技术十余年的珠海一微半导体股份有限公司，8月26日在横琴举办的"机 器芯·万象新"品牌焕新盛典暨全场景智能机器人技术平台新品发布会上，宣布"一微半导体"品 牌焕新为"一微科技"，旨在携手产业链上下游伙伴，共筑智能移动机器人技术平台发展新生 态。 作为国内较早聚焦机器人核心芯片研发的企业，一微科技的技术积淀与市场地位在行业内早有共识。 活动现场，一微科技总裁姜新桥在回顾企业十一年发展脉络时透露，自成立以来，一微始终以"用芯 定义机器人"为核心理念，以自主研发的机器人主控芯片为技术底座，持续向下深耕传感器、算法等 底层创新，同步布局家用、商用及工业级全场景应用，逐步成长为机器人专用芯片领域市场份额领先 的企业。其技术积累不仅支撑了下游产品智能化水平的跃升，更推动了"智能移动"能力从单一功能向 ...

Core Viewpoint - Recent stock market prosperity has increased attention on chip companies, but two prominent chip firms have issued warnings regarding their stock performance and market risks [2][8]. Group 1: Company Performance and Predictions - Zhongke Hanwuji Technology Co., Ltd. forecasts an annual revenue of between 500 million to 700 million yuan for 2025, highlighting that this is a preliminary estimate and not a commitment to investors [2][5]. - The company's stock price increased by 133.86% from July 28, 2025, to August 28, 2025, significantly outpacing most peers and major indices [4][5]. - The company's rolling price-to-earnings (P/E) ratio is 5117.75 times, and the price-to-book (P/B) ratio is 113.98 times, both substantially higher than the industry averages of 88.97 times and 5.95 times, respectively [5][11]. Group 2: Market Risks and Stock Volatility - The company operates on a Fabless model, relying on various suppliers, which poses risks to supply chain stability, especially since some subsidiaries are on the "entity list" [3][12]. - Dongxin Semiconductor Co., Ltd. has experienced significant stock price fluctuations, with a cumulative increase of 207.85% from July 29, 2025, to August 28, 2025, and an average turnover rate of 11.77%, indicating potential market overheating [9][11]. - Dongxin acknowledges that its stock valuation is excessively high, with a rolling P/E ratio reported as negative, contrasting with the industry average of 53.35 [9][11]. Group 3: Product Development and Market Competition - Dongxin has no new product release plans, and recent information circulating about new products is deemed misleading [2][10]. - The company faces several risks related to its investment in Shanghai Lishuan Technology Co., Ltd., including industrialization progress, market competition, and reliance on a single product, the "7G100" GPU [12][13].

Core Viewpoint - The current state of the RF front-end industry is characterized by a competitive environment that is both challenging and necessary for rational development, as companies face varying degrees of losses and must navigate through market pressures to avoid resource misallocation [1][2]. Competition Landscape - The ODM market and certain Cat1 markets are experiencing intense competition driven by low procurement standards, leading to a situation described as "blood flowing in the streets" [2]. - In contrast, the brand client market is orderly and conducive to rapid industry iteration, with major smartphone manufacturers selecting a limited number of domestic RF front-end suppliers based on comprehensive evaluations rather than just price [2][3]. Market Size and Growth Potential - The global consumer RF front-end market is approximately 1200 billion, with Apple and Google accounting for about half of this market [3]. - The domestic RF front-end market is currently under 200 billion, indicating significant growth potential, as it is expected to double in size [3][4]. Profitability and Business Strategy - A healthy profit margin for the RF front-end industry is estimated to be between 20% and 30%, as evidenced by the financial reports of leading companies like Zhaoshengwei and Weijie Chuangxin [2][3]. - Companies are advised to be cautious in their operational strategies, particularly regarding capacity expansion, to avoid oversupply and intensified competition [2][6]. Opportunities for Domestic Companies - Domestic RF front-end companies need to focus on high-end modules to maintain growth, as the mid-to-high-end module market is currently dominated by Qualcomm and Qorvo [5]. - There are significant opportunities in high-performance modules, Sub6G modules, and automotive-related RF front-ends, which require companies to enhance product development and differentiation [5][6]. Collaboration and Industry Health - Companies are encouraged to strengthen collaboration across the supply chain to avoid excessive capacity building and to ensure a healthy industry ecosystem [6]. - Smaller RF front-end companies should consider differentiated development strategies and manage cash flow effectively to avoid unnecessary losses [6].

Core Viewpoint - The article argues for a paradigm shift from traditional memory hierarchies to specialized memory architectures that leverage application-specific access patterns, proposing two new memory categories: Long-term RAM (LtRAM) and Short-term RAM (StRAM) [2][4][45]. Group 1: Current Memory Landscape - SRAM and DRAM have reached fundamental physical limitations, halting their scalable development, which has made memory a major bottleneck in performance, power consumption, and cost for modern computing systems [4][10]. - DRAM accounts for over 50% of server hardware costs, highlighting the economic impact of memory limitations [4][10]. - The rise of memory-intensive workloads, particularly in artificial intelligence, exacerbates the challenges posed by the stagnation of SRAM and DRAM [4][10]. Group 2: Proposed Memory Categories - LtRAM is designed for persistent, read-intensive data with long lifecycles, while StRAM is optimized for transient data that is frequently accessed and has short lifecycles [12][26]. - These categories allow for tailored performance optimizations based on specific workload requirements, addressing the mismatch between current memory technologies and application needs [12][26]. Group 3: Emerging Memory Technologies - New memory technologies such as RRAM, MRAM, and FeRAM offer different trade-offs in density, durability, and energy consumption, making them suitable for various applications but not direct replacements for SRAM or DRAM [16][21]. - RRAM can achieve density up to 10 times that of advanced HBM4 configurations, indicating significant scalability advantages [20][21]. Group 4: Workload Analysis and Memory Access Patterns - Analyzing memory access patterns is crucial for identifying opportunities for specialization, as seen in workloads like large language model inference, which is read-intensive and requires high bandwidth [28][30]. - Server applications and machine learning workloads exhibit diverse memory access patterns that can benefit from specialized memory technologies [29][31]. Group 5: System Design Challenges - The introduction of LtRAM and StRAM presents new research challenges, including how to expose memory characteristics to software without increasing complexity [35][37]. - Data placement strategies must adapt to heterogeneous memory systems, requiring fine-grained analysis of data lifecycles and access patterns [38][39]. Group 6: Power Consumption and Efficiency - Memory specialization can lead to significant power savings by aligning storage unit characteristics with workload demands, thus reducing static power and data movement costs [41][43]. - The increasing power density in data centers necessitates innovative cooling solutions and power management strategies to support high-performance computing [43][44].