公众号记得加星标⭐️，第一时间看推送不会错过。 Unified Bus 的协议文档终于发布了。协议最初的设计大多数是四五年前的工作了，我也有两年 多没有继续做网络互联方面的工作，但今天读到这本 500 多页的文档，还是倍感亲切。 与大多数协议文档一样，UB 文档介绍了 Unified Bus 协议的大量细节，但很少涉及它设计背后 的思考。作为曾在早期参与 UB 项目的一名小兵，介绍一些我个人的思考。今天产品化的 UB 可 能与我们当年的设计有诸多不同，因此不要把本文作为权威指南。当成段子看就行了。 为什么要做 UB 要理解 Unified Bus (UB) 诞生的必然性，我们必须回到一个计算机体系结构中的根本性矛盾：总线 （Bus）与网络（Network）的割裂。 长久以来，计算机世界被这两种截然不同的互联范式划分为一个个孤岛。 在孤岛内部（例如一台服务器或一个机箱内），我们使用总线技术，如 PCIe 或 NVLink。它们是为 紧耦合系统设计的，设备间共享着统一的物理地址空间，通信延迟可以做到纳秒级，带宽极高。这是 性能的天堂，但这个天堂的疆域极其有限——总线的物理距离和可连接的设备数量都受到严格限制。 在 ...

Core Viewpoint - DDR4 memory prices are rising due to reduced supply rather than increased demand, as major DRAM manufacturers shift focus to DDR5 production [2][4]. Group 1: DDR4 and DDR5 Transition - DDR4 memory is losing its role as a market trend indicator due to the accelerated transition to DDR5 in the PC and server markets [2]. - Major DRAM manufacturers, including Samsung, SK Hynix, and Micron, are continuously reducing DDR4 production as the market increasingly supports DDR5 [2][3]. - Intel's latest server processors have completely stopped supporting DDR4, further pushing the market towards DDR5 adoption [2]. Group 2: Performance and Demand Drivers - The explosive growth in demand for artificial intelligence (AI) and high-performance computing (HPC) is driving the transition to DDR5, which offers 1.5 to 2 times the transmission speed and 30% better energy efficiency compared to DDR4 [3]. - Market research firm TrendForce predicts that DDR5 will account for over 50% of all PC and server DRAM shipments in the second half of this year [3]. Group 3: Market Dynamics and Future Outlook - The current price increase of DDR4 is attributed to supply reduction rather than demand increase, indicating that the market is not recovering [4]. - The global memory semiconductor market is expected to maintain stable growth through 2026, with DRAM and NAND flash performing well due to balanced supply and demand [6]. - The demand for DRAM is projected to grow by 19.3% in 2025, slightly above the industry production growth rate of 18.1% [6]. Group 4: Investment in Next-Generation DRAM - Major memory manufacturers are accelerating investments in 1c DRAM, with Samsung and SK Hynix planning to ramp up production lines [7][8]. - Micron is also expected to speed up investments in 1c DRAM, supported by significant subsidies from the Japanese government [9].

据市场监管总局消息，因高通公司收购Autotalks公司未依法申报经营者集中，涉嫌违反《中华人民共和国反垄断法》，市场监管总局依法对高通公司开展 立案调查。 2024年3月，美国联邦贸易委员会（FTC）曾表示担忧，认为此次收购可能削弱V2X芯片市场的竞争活力。欧洲方面，15个成员国联名要求欧盟委员会启 动深入审查，认为交易可能导致高通在V2X技术标准制定中的话语权过度集中。 受此影响，高通于2024年3月底宣布终止首次收购计划。但经过约一年的谈判与方案调整，双方最终于2025年6月重新完成交易，Autotalks团队正式并入高 通汽车业务部门。据外媒估计，最终交易金额约为8,000万至9,000万美元，低于最初传闻的3.5亿至4亿美元。 尽管双方早在2023年达成初步收购协议，但交易推进过程并不顺利。由于涉及汽车通信安全、无线频谱使用及产业标准竞争等敏感领域，该交易先后受到 欧盟、英国、美国等多个监管机构的重点关注。 据了解，Autotalks是一家总部位于以色列的无晶圆半导体公司，成立于2008年，长期专注于V2X通信芯片的研发与商用落地。其核心产品能够支持DSRC （专用短程通信）和C-V2X（基于蜂窝的 ...

Core Points - The U.S. government has granted NVIDIA an export license to ship advanced AI GPUs worth hundreds of billions to the UAE, marking the beginning of a significant bilateral technology cooperation [1] - The agreement allows the UAE to purchase up to 500,000 advanced NVIDIA processors annually, with a commitment to invest $1.4 trillion in the U.S. over the next decade [1] - The deal is seen as a shift in U.S. policy, moving from restricting AI chip exports to establishing a bilateral framework with allies [2] Group 1 - The initial shipment of AI accelerators will not be delivered to G42, an AI company in Abu Dhabi, but will be operated by U.S. companies with data centers in the UAE [1] - The U.S. aims to counter China's influence in the region by embedding American cloud service providers and hardware into UAE infrastructure [2] - Future authorizations will depend on the progress of UAE's investments, potentially serving as a model for similar agreements with other allies seeking advanced AI hardware [3] Group 2 - The agreement includes a provision for G42 to eventually receive 20% of the AI processors in the UAE [1] - The U.S. Commerce Secretary confirmed that only approved U.S. operators managing data centers will be eligible to run these systems [2] - Critics argue that the agreement lacks sufficient safeguards to ensure the AI accelerators operate in reviewed environments, especially given the UAE's economic ties with Beijing [3]

Core Viewpoint - The article discusses the transformation of Yunhan Chip City from traditional offline distribution to an online model, addressing the long-tail demand in the semiconductor distribution industry and highlighting its successful integration of digital platforms, data, and supply chain networks to create a scalable business model [1][5][24]. Group 1: Industry Structure and Challenges - The semiconductor distribution industry faces a paradox of concentrated upstream manufacturers and a fragmented downstream market, leading to challenges in meeting the long-tail demand effectively [1]. - Traditional supply chain models struggle with small orders, where the cost structure makes fulfilling orders below 5000 yuan unprofitable, causing difficulties for small and medium enterprises in procurement [1][4]. Group 2: Yunhan Chip City's Transformation - Yunhan Chip City began its transition to an online distribution model in 2011, aiming to reduce operational costs and improve efficiency through digitalization [1][2]. - The company has developed a procurement platform that allows customers to search, compare, and order electronic components online, significantly lowering operational costs [5]. Group 3: Data and Supply Chain Integration - Yunhan Chip City has built a comprehensive database with over 40 million product specifications and 93 million parameter relationships, which helps reduce the time for engineers to select components from 15-30 days to 1-2 days [8][10]. - The company collaborates with over 2500 suppliers to create a coordinated supply network, enabling faster assembly of components and reducing procurement costs for small tech enterprises [8][15]. Group 4: Quality Control and Competitive Advantage - The company has established a quality control system with over 100,000 reference samples to ensure that no counterfeit products reach customers, supported by a certified testing center [12][14]. - Yunhan Chip City has invested in data governance and technology to enhance its data quality, which serves as a competitive advantage in the market [10][11]. Group 5: Market Trends and Future Outlook - The global semiconductor market is expected to grow, with a projected sales figure of $627.6 billion in 2024, translating to a 19.12% year-on-year increase [15]. - Yunhan Chip City has seen significant revenue growth, with a 17.82% increase in revenue and a 41.17% increase in net profit in the first half of 2025, outperforming industry averages [15]. - The company is strategically positioned to capitalize on the accelerating domestic semiconductor market and the trend towards small-batch, rapid iteration manufacturing [19][20].

Core Insights - The semiconductor industry is experiencing growth in EDA and silicon IP revenues, with a projected increase of 8.6% in Q2 2025, reaching $5.089 billion, compared to $4.685 billion in Q2 2024 [2] - The growth in EDA revenue is primarily driven by the CAE category, which saw a significant increase of 17.2% [2][4] - Despite a recovery in the Chinese market after five quarters of decline, growth remains modest, while India and Taiwan show strong performance in the IP sector [2][3] EDA Revenue Breakdown - Total EDA revenue for Q2 2025 reached $3.0614 billion, up 8.4% from $2.825 billion in Q2 2024 [3] - CAE revenue increased to $1.929 billion, reflecting a 17.2% growth [4] - PCB & MCM revenue grew by 7.8% to $430.5 million, while IC Physical Design & Verification revenue decreased by 9.9% to $701.9 million [4] Regional Performance - The Americas experienced a growth rate of 14.1%, while EMEA and Japan saw increases of 10.7% and 11.8%, respectively [3][5] - The Asia-Pacific region, primarily driven by China, had a lower growth rate of 7.2% [5] Employment Trends - The total number of employees in the tracked companies rose to 72,529 in Q2 2025, marking a 14.8% increase from 63,188 in Q2 2024 [5] Challenges in Semiconductor IP Sales - The semiconductor IP market faces challenges, particularly in the analog IP sector, which lacks standardization and requires custom design and verification [6][8] - Documentation and support for semiconductor IP are often inadequate, leading to integration difficulties for customers [7][8] - Successful semiconductor IP sales require a collaborative approach, focusing on high-quality, well-documented solutions rather than merely increasing the quantity of offerings [9]

Core Viewpoint - The article discusses the increasing importance of location verification in the semiconductor supply chain due to heightened export controls and concerns over AI chip smuggling and counterfeiting. It highlights the shift from traditional manual tracking methods to more advanced technological solutions for monitoring chip movements from manufacturing to end-users [1][14]. Summary by Sections Location Verification Methods - Various methods for tracking chip locations include built-in GPS technology, ping-based systems, and geofencing to restrict chip behavior within certain boundaries [5][6]. - These methods aim to enhance supply chain oversight with minimal effort, addressing the complexities of tracking millions of chips across multiple sources [1][4]. Challenges and Innovations - Current tracking technologies, while effective, may be intrusive and susceptible to deception, prompting the need for innovative solutions to mitigate negative impacts [2][3]. - The ping-based technology can determine location through round-trip time measurements, but verifying the accuracy of these pings remains a challenge [4]. Economic and Political Considerations - Different chips and systems present varying economic risks and political boundaries, leading to diverse solutions based on accuracy requirements and operational costs [4]. - The U.S. government is considering location verification as part of a broader strategy for AI and semiconductor security, reflecting current policy trends that balance technology export support with national security [14]. Cost Implications - Implementing location tracking can significantly increase the cost of chips, as seen in the example where a microcontroller's price rose from $2 to $5 due to added security features [8]. - The integration of tracking technologies necessitates additional infrastructure and software, which can further drive up costs and impact battery life [7][8]. Security and Privacy Concerns - While verification methods do not inherently create new attack surfaces, geofencing could introduce vulnerabilities if not managed properly [9][10]. - The article raises concerns about the implications of location tracking on privacy and the potential for misuse, especially in sensitive applications [15][16]. Future Outlook - The article concludes that while location verification presents a promising avenue for enhancing chip security, it also raises significant privacy issues and requires careful consideration of alternative delivery methods [14].

Core Viewpoint - AMD is set to launch its next-generation Instinct MI450 graphics accelerator using 2nm process technology, aiming to compete directly with NVIDIA's Rubin chip series [1][3][6] Group 1: Product Specifications - The Instinct MI450 will utilize TSMC's 2nm technology for its accelerator core chip (XCD), while the active interposer chip (AID) and media interface chip (MID) will be manufactured using TSMC's 3nm process [1][5] - MI450 is expected to deliver similar FP4/FP8 performance to NVIDIA's Rubin, but with 1.5 times the memory and bandwidth [3][5] - The MI400 series will integrate up to 432 GB of HBM4 memory, significantly enhancing bandwidth capabilities [7] Group 2: Market Positioning - AMD's data center operations head, Forrest Norrod, claims that MI450 will be the best solution for training, inference, distributed inference, and reinforcement learning, indicating confidence in competing with NVIDIA [6] - The MI450 is anticipated to be AMD's "Milan moment," referencing the significant leap made by AMD's EPYC server CPU series [6] - AMD plans to expand its rack options with the upcoming Helios system, which is expected to match the top configurations of NVIDIA's Rubin series [7] Group 3: Launch Timeline - The MI450 is projected to be released in 2026, coinciding with NVIDIA's launch of its next-generation Rubin series [7]

Core Insights - RISC-V's market penetration has reached 25%, indicating significant growth in the chip market, particularly in IoT and automotive applications [3][4] - SHD Group's upcoming forecast suggests that RISC-V chip shipments could exceed 21 billion by 2031, generating over $2 billion in revenue [3] - Meta's acquisition of Rivos, a startup developing RISC-V processors, highlights the competitive landscape for advanced AI chips and the potential for RISC-V to challenge established players like Nvidia [4][9] RISC-V Market Growth - RISC-V's market penetration is a substantial leap, with predictions of reaching 25% of the semiconductor market by 2030, equating to 17 billion chip shipments [3] - The growth is attributed to RISC-V's applications in edge AI deployments, providing localized data centers for smaller communities [3] Meta's Acquisition of Rivos - Meta's acquisition of Rivos, which has developed RISC-V processors compatible with CUDA, signifies a strategic move to enhance its AI capabilities [4][7] - Rivos aims to reduce the cost and complexity of migrating applications to its SoC, leveraging its compatibility with existing CUDA software [6][9] - The acquisition is expected to accelerate Meta's development of proprietary chips for AI workloads, with significant R&D investment [7][8] European RISC-V Developments - The eProcessor project in Europe has successfully developed the first out-of-order RISC-V processor, marking a significant milestone in the semiconductor industry [12][14] - This processor, built on a 22nm process, enhances computational performance and efficiency for various applications, including AI and automotive [12][15] - The project aligns with the EU's semiconductor innovation strategy, aiming to boost Europe's position in the global chip market [14][17] Future Implications - The success of the eProcessor project underscores the potential for RISC-V to provide alternatives to traditional closed architectures, fostering innovation in high-performance computing and embedded systems [16][17] - Continued investment and strategic support are crucial for the eProcessor project to compete with industry leaders like AMD and Intel [17]

公众号记得加星标⭐️，第一时间看推送不会错过。 来源 ：本文编译自SEMI 。 细分市场增长 预计逻辑与微电子领域将引领设备扩张，2026 年至 2028 年期间总投资额将达到 1750 亿美元。预计 代工厂将成为这一增长的主要驱动力，而这主要得益于 2 纳米以下工艺的产能扩张。关键推动因素包 括环绕栅极 (GAA) 架构和背面供电等先进技术，这些技术对于提升芯片性能和功耗效率，以应对日 益苛刻的人工智能工作负载至关重要。更先进的 1.4 纳米工艺技术预计将于 2028-2029 年实现量 产。此外，人工智能性能的提升预计将推动边缘设备（包括汽车电子、物联网应用和机器人技术）的 大幅增长。除了先进工艺之外，所有节点和各种电子设备的需求预计将大幅增长，从而推动成熟工艺 设备的投资。 预计内存细分领域将位居第二，三年支出总额达 1360 亿美元，标志着该细分领域开启新一轮增长周 期。预计 2026 年至 2028 年，DRAM 相关设备投资将超过 790 亿美元，其中 3D NAND 投资将达 到 560 亿美元。人工智能训练和推理推动了各类内存需求的全面增长。人工智能训练需要更大的数 据传输带宽和极低的延迟，显 ...