Core Viewpoint - TSMC is recognized as the most trusted semiconductor foundry globally, built on decades of technological leadership, a rigorous business model, excellent operations, and reliability [2][5]. Group 1: Business Model and Trust - TSMC's pure foundry model is the foundation of its reputation, as it does not compete with clients and maintains strict confidentiality, ensuring clients' intellectual property is protected [2][3]. - The company has established deep trust within a vast ecosystem, making it the preferred manufacturing partner for leading chip designers like Apple, AMD, and Qualcomm [2]. Group 2: Technological Leadership - TSMC consistently leads in the production of advanced process nodes such as 7nm, 5nm, and 3nm, maintaining high yield rates, which is crucial for reliable and timely delivery of innovative products [3][5]. - The company's focus on process control and defect management has resulted in industry-leading yield rates, allowing clients to reduce costs and accelerate production ramp-up [3]. Group 3: Intellectual Property and Confidentiality - TSMC has a strong reputation for protecting intellectual property, having proven its ability to handle sensitive data securely through long-term collaborations with thousands of clients [4]. - The company's internal culture and strict access controls contribute to its reliability in safeguarding client information [4]. Group 4: Scale and Ecosystem Integration - TSMC's significant investments in fabs, equipment, and talent have created unmatched manufacturing capabilities, earning client trust [4]. - Collaborations with equipment suppliers and EDA vendors form a robust ecosystem that reduces risks and shortens time-to-market for clients [4]. Group 5: Long-term Strategic Vision - TSMC's proactive investment strategy, often initiated before demand arises, ensures sufficient capacity even during industry fluctuations [5]. - The company's capacity planning during recent global chip shortages has reinforced its image as a stable and responsible industry leader [5]. Group 6: Global Reputation and Governance - TSMC demonstrates transparency and compliance, fostering cooperation with governments and clients worldwide, despite geopolitical risks [5]. - The company's expansion into the US, Japan, and Europe reflects its commitment to supply chain resilience and global trust [5].

如果您希望可以时常见面，欢迎标星收藏哦~ 随着各家大厂持续进行资本支出、AI竞争白热化，市场目前盛传两种主流观点，一是TPU和ASIC 作为GPU替代方案崛起，使英伟达（NVIDIA）护城河遭到侵蚀；另一个是Google逆袭，透过AI 模型Gemini取得市占和主导AI搜寻，最终将击败OpenAI。 根据外媒siliconANGLE的看法，上述两种说法都严重夸涨，且不一定如市场所想。理由之一是辉 达将持续精进其GB300和后续Vera Rubin架构，可维持领先出货并不断降低成本，成为迄今最具 成本效益的平台。 至于Google则在壮大的同时面临一种「创新者困境」。报导称，由于Google搜寻业务与广告营收 高度绑定，一旦Google将广告模式转向类似聊天机器人的体验，单次搜寻查询的服务成本将暴增 至原来的100倍；再者，Google必须将商业模式转向更高度整合的购物体验，必须连结使用者与广 告主，但即便Gemini近期取得一定成功，Google仍不具备这样的信任基础。相比之下，OpenAI核 心在于强调可信资讯，而非推送广告，仍有其优势。 为何TPU难以打破英伟达AI护城河？ 目前领先AI训练走向为「高频宽 ...

Core Insights - The article discusses the advancements and market trends in the CPO (Co-Packaged Optics) technology within the optical communication industry, highlighting its growing importance for AI and data center applications [2][5]. Group 1: CPO Technology Developments - Nvidia announced the adoption of single-channel 200G CPO technology in its InfiniBand and Ethernet switches in March [2]. - Meta's testing in September demonstrated the reliability of Broadcom's previous CPO products, leading to Broadcom's launch of its third-generation single-channel 200G CPO product in October [2]. - At the TEF conference in December, Nvidia reported that AI clusters based on CPO switches showed a tenfold improvement in reliability compared to systems using pluggable optical modules, translating to a fivefold increase in cluster uptime [2]. Group 2: Market Trends and Predictions - Ciena's acquisition of Nubis Communications and Marvell's acquisition of Celestial AI indicate major companies' focus on CPO technology, with expectations of more M&A activities in this space by early 2026 [5]. - Currently, CPO applications are limited to Scale-Out network design switches, with the next challenge being to extend Scale-Up interconnects beyond single racks, aiming to scale GPU clusters from 128-144 chips to 500-1000 chips for accelerated AI training [5]. - Amazon is using AEC to interconnect Trainium accelerators across two racks, but this may not scale well to more racks, while Huawei employs 800G pluggable LPO optical modules in its vertical Scale-Up network [5]. Group 3: Future Market Outlook - LightCounting has raised its market forecast for CPO, which now includes 1.6T and 3.2T ports for Scale-Up scenarios with transmission distances under 50 meters [5]. - Broadcom and Nvidia are expected to launch integrated CPO Scale-Up switches, GPUs, or XPUs by 2026, with shipments starting in 2027 [8]. - By 2030, the market size for CPO engines covering both Scale-Up and Scale-Out scenarios is projected to reach $10 billion, with nearly 100 million CPO ports expected to be shipped [8].

如果您希望可以时常见面，欢迎标星收藏哦~ 12月24日，外交部发言人林剑主持例行记者会。 法新社记者提问，美国贸易代表调查结果称"中国实行了不公平的措施，企图主导半导体产业"，美 国拟自2027年起对中国半导体产业加征关税。中方对此有何评论？ 林剑表示，中方坚决反对美方滥施关税、无理打压中国产业。美方的做法扰乱全球产供链稳定，阻 碍各国半导体产业的发展，损人害己。 "我们敦促美方尽快纠正错误做法，以两国元首达成的重要共识为引领，在平等、尊重、互惠的基 础上，通过对话解决各自关切，妥善管控分歧，维护中美关系稳定健康可持续发展。如果美方一意 孤行，中方必将坚决采取相应措施，维护自身正当权益。"林剑说。 美国对华芯片关税，官宣延期 美国特朗普政府周二在联邦公报上表示，美国将于2027年6月提高对中国半导体进口的关税，具体 税率至少提前一个月确定。 但与此同时，根据美国贸易代表办公室提交的文件，从中国进口半导体的初始关税税率将在 18 个 月内为零。 作为一年前启动的一项调查的一部分，该机构发现中国在该行业从事不公平贸易行为。 该办公室在文件中表示："几十年来，中国一直以半导体行业为目标，力图在该行业占据主导地 位， ...

如果您希望可以时常见面，欢迎标星收藏哦~ 分红拆成4次领，台积电避开工程师领完就跳槽 没想到二十几年后，留在台积电的年薪已经300万了，去中国和鉴之后回台湾，幸运的回到半导体 业，但大叹非常不划算，年资都没有了，但跟他一起去中国的，有的回不来了，有的去了中芯半导 体。 这位主管也透露，台积电除了月薪，每一季的分红最吸引人，邻居一个工程师有次薪资单不小心掉 在门外没发现，偷看一眼，「哇，这一季就领50万元，一年领个四次还得了。」为什么一季发一 次呢？以前台积电也是过农历年前发一笔年终奖金，结果很多人过完年就跳槽了，人力一下子短 缺，对制造产生危机，后来改一季发一次，人力波动才能分摊到每一季。 台积电是今年台湾最赚钱的公司，今年前三季税前净利1.4兆元，已经超过去年全年总合，根据台 积电去年年报，董事长魏哲家及前董事长刘德音在今年7月还被美国半导体产业协会（SIA）选为 2025年罗伯特‧诺伊斯奖（Robert N. Noyce Award）共同获奖者。 身为全台湾唯一同时和美国总统川普及总统赖清德握过手的CEO、魏哲家每年替台积电创造那么 多价值，台积电董事会一年给他的薪水多少？根据台积电2024年年报，他一年 ...

如果您希望可以时常见面，欢迎标星收藏哦~ Rapidus 参加了 2025 年日本半导体展（2025 年 12 月 17 日至 19 日），并展示了在北海道千岁 市的研发制造中心 IIM 生产的 2nm GAA（全环栅极）晶体管和 600mm 方形重分布层 (RDL) 中 介层面板的原型。 2025年6月16日，Rapidus首次将晶圆装载到位于北海道千岁市的研发制造中心IIM的前端试验线 上。同月28日，首个2nm GAA晶体管原型完成并确认运行正常。这是IIM制造的实际原型首次在 展会上展出，展位吸引了众多参观者。 点这里加关注，锁定更多原创内容 *免责声明：文章内容系作者个人观点，半导体芯闻转载仅为了传达一种不同的观点，不代表半导体芯闻对该 观点赞同或支持，如果有任何异议，欢迎联系我们。 10万亿，投向半导体 芯片巨头，市值大跌 黄仁勋：HBM是个技术奇迹 Jim Keller：RISC-V一定会胜出 全球市值最高的10家芯片公司 推荐阅读 除了展示实际原型外，展位还利用显示屏解释了2nm工艺和GAA结构的优势。与2008年左右的尖 端工艺40nm平面晶体管相比，2nm GAA晶体管的功耗可以降低约二十 ...

Core Viewpoint - The article discusses the evolution of smart automotive communication systems, emphasizing the importance of Time-Sensitive Networking (TSN) as a critical infrastructure for intelligent driving, highlighting the dominance of international giants in this field and the emergence of domestic players like Beijing Chiprise Semiconductor Technology Co., Ltd. [2][6] Group 1: Smart Automotive Communication - Traditional automotive communication networks are limited in bandwidth and speed, while smart vehicles require high-speed, reliable data transmission to support multiple sensors and real-time processing [4][6] - TSN technology enhances traditional Ethernet to ensure deterministic and reliable data transmission, crucial for safety in smart driving scenarios [6][10] Group 2: Competitive Differentiation - Chiprise Semiconductor aims to create value through differentiated innovation rather than simple replacements, focusing on the unique demands of the Chinese market [8][10] - The company is developing next-generation vehicle communication technologies, including the SV37XX series TS-PON chips, which support high-speed data transmission and various interface protocols [8][10] Group 3: Future Trends in Communication - The trend of "optical fiber replacing copper" is seen as inevitable for future vehicle communication, driven by the increasing demand for high bandwidth and low latency in automotive applications [10][12] - The company believes that as the automotive industry matures, the challenges of integrating fiber optics in vehicles will be addressed, leading to faster adoption of this technology [10][12] Group 4: Strategic Vision - Chiprise Semiconductor's strategy extends beyond the automotive sector, aiming to provide high-speed, time-sensitive communication solutions for various intelligent applications, including industrial automation and robotics [12][14] - The automotive industry is viewed as a foundational market due to its scale and stringent quality standards, which can help develop technologies applicable to other sectors [12][14] Group 5: Industry Outlook - China's automotive production is projected to reach 32 to 33 million units this year, with a growing demand for next-generation intelligent semiconductors [14][15] - The Chinese automotive chip industry is expected to undergo significant transformation, moving from reliance on imports to developing autonomous capabilities and eventually exporting advanced technologies [14][15]

当AI 应用逐步从「训练」转向「推理」，对于高速资料储存与即时存取的需求大幅提升，必须仰 赖HBM 等记忆体，将资料持续供应给GPU 进行运算。 即便通用记忆体的效能不及HBM，市场对高效能通用记忆体的需求仍快速成长。在推理初期阶 段，工作负载多由GDDR7、LPDDR5X 等通用DRAM 处理，而HBM 则主要保留给更密集的推理 任务。 NVIDIA 在以推理为主的AI 加速器中采用GDDR7，便是一项代表性案例。 记忆体业者也计划透过开发AI 导向的高效能产品，延续「记忆体为核心」的产业趋势。举例来 说，记忆体内运算（Processing-In-Memory，PIM）技术，让记忆体可承担部分原本由GPU 执行 的运算工作。报导也指出，垂直通道电晶体（VCT）DRAM 与3D DRAM 等新技术，透过在更小 面积中储存更多资料来提升密度，预期将陆续导入市场。 如果您希望可以时常见面，欢迎标星收藏哦~ 随着人工智能需求快速攀升，记忆体价格大幅上涨，半导体产业的获利结构正出现变化。根据《韩 国经济新闻》报导，三星电子的记忆体部门，以及SK 海力士，预期在2025 年第四季的毛利率将 超越台积电，这将是自2018 ...

Core Insights - The evolution of integrated circuits (IC) and system-on-chip (SoC) technologies is leading to a diverse combination of processing elements that optimize current workloads while preparing for future demands [2] - Programmable components like FPGAs and DSPs are becoming essential as they allow for hardware upgrades without the need for expensive new devices, addressing the rapid pace of technological advancements [2][3] - The integration of AI into DSPs is enhancing their adaptability and performance, particularly in handling the increasing complexity of analog signals in modern SoCs [6][7] Group 1: Programmable and Reconfigurable Technologies - The shift from simple choices between ASIC, FPGA, and DSP to a combination of various processor types reflects the need for flexibility in design [2] - FPGAs offer high programmability, allowing complete design changes, while other components provide limited configurability [5] - The emergence of CGRA technology offers a balance between FPGA and GPU capabilities, providing flexibility and efficiency for specific tasks [4] Group 2: AI and DSP Integration - AI is playing a crucial role in addressing the challenges posed by increasing analog content in SoCs, enabling dynamic calibration and optimization of DSPs [6][7] - DSPs are evolving to handle not just numerical computations but also analog signal processing, which increases their complexity and parallelism [6] - The integration of AI engines within FPGAs allows for synchronized execution of DSP tasks, enhancing overall system performance [9][12] Group 3: Chiplets and eFPGA Solutions - Chiplets provide a flexible solution for rapidly changing application scenarios, allowing for easy upgrades without redesigning the entire SoC [13] - Embedded FPGAs (eFPGAs) offer reconfigurability but may increase area costs, necessitating careful design considerations [14] - The market is experiencing volatility, prompting companies to accelerate chip development cycles to adapt to evolving requirements [15] Group 4: Future Trends and Software-Defined Products - The trend towards software-defined and AI-driven products necessitates hardware that can support software changes, highlighting the importance of programmability in semiconductor design [16] - Companies are increasingly investing in compiler technologies to ensure that software development aligns with hardware capabilities, facilitating rapid updates and feature enhancements [16]

如果您希望可以时常见面，欢迎标星收藏哦~ 以前跟随为主，现在慢慢走到无人区，可参考的地方越来越少了。米硅已全面着手研发4*224G ASP(CDR)系列产品，4*448G ASP 系列已在蓝图。一边定义下一代产品，一边把技术边界再往前 推，前面没有参照物，就自己成为坐标。 关于米硅科技 技术原理对比： CDR(时钟数据恢复)通过模拟电路提取时钟信号，主要解决抖动问题。核心组件包括鉴相器、锁相 环(PLL)和频率检测器，像传统相机的光学防抖，用物理方式稳定信号。 ODSP(光学数字信号处理)采用ADC将模拟信号数字化，通过数学算法补偿光纤损耗、色散和非线 性效应 >。相当于用Photoshop的AI算法逐像素修复，能处理PAM4/相干调制等复杂信号。 算力中心短距连接 米硅科技的4x112G PAM4 CDR核心技术除了400G、800G光模块可替代oDSP方案打破海外垄断 同时成本功耗降50%-70%, 同时也可用于400G、800G AEC方案，在3-7米铜缆传输场景替代传统 DSP方案，成本功耗可降低50-70% 新的挑战，新的规划 米硅科技首款自研4X112G ASP(CDR)电芯片收发套片ms8904 ...