光子学具有巨大的潜力，因为光波的传播和干涉无需消耗能量，因此可以通过工程设计来实现可扩展的功能，而无需增加能耗。硅光子学在过去二十年中 得到了广泛的发展，如今已完全具备提供近乎理想的平台的能力，从而释放其巨大的潜力。事实上，硅光子学能够提供高效的高密度互连，实现高带宽和 长距离链路；能够实现低能耗的光路切换，且不受信号带宽的限制；以及能够进行光速计算的光子神经网络，从而加速人工智能计算。 在本文中，我们将回顾这些光子技术的发展趋势和进展，我们将论证，为了使这些光子技术成为人工智能时代可持续基础设施的重要组成部分，硬件和软 件以及电子和光子学需要以互补的方式进行开发。 光收发器和交换机 近年来，生成式人工智能的迅猛发展导致全球范围内超大规模人工智能集群的部署速度空前加快。随着摩尔定律的放缓，只有通过并行计算才能实现更高 的性能，因此，数据处理和/或传输性能的提升必然导致能耗的增加。也就是说，人工智能基础设施的快速增长带来了严重的能源危机。如图 1 所示，随 着数据量的指数级增长，所需的能源供应量也将呈指数级增长。从这个意义上讲，解决这一能源问题的唯一有效途径是开发一种能够将能源增长与数据增 长分离的技术。 A. ...

近年来，生成式人工智能的迅猛发展导致全球范围内超大规模人工智能集群的部署速度空前加快。随 着摩尔定律的放缓，只有通过并行计算才能实现更高的性能，因此，数据处理和/或传输性能的提升 必然导致能耗的增加。也就是说，人工智能基础设施的快速增长带来了严重的能源危机。如图 1 所 示，随着数据量的指数级增长，所需的能源供应量也将呈指数级增长。从这个意义上讲，解决这一能 源问题的唯一有效途径是开发一种能够将能源增长与数据增长分离的技术。 光子学具有巨大的潜力，因为光波的传播和干涉无需消耗能量，因此可以通过工程设计来实现可扩展 的功能，而无需增加能耗。硅光子学在过去二十年中得到了广泛的发展，如今已完全具备提供近乎理 想的平台的能力，从而释放其巨大的潜力。事实上，硅光子学能够提供高效的高密度互连，实现高带 宽和长距离链路；能够实现低能耗的光路切换，且不受信号带宽的限制；以及能够进行光速计算的光 子神经网络，从而加速人工智能计算。 在本文中，我们将回顾这些光子技术的发展趋势和进展，我们将论证，为了使这些光子技术成为人工 智能时代可持续基础设施的重要组成部分，硬件和软件以及电子和光子学需要以互补的方式进行开 发。 光收发器和交换 ...

然而，埃尔旺格坚信Tower的股价还能涨得更高。公司的大部分员工都持有股份，虽然三年前，由于中国监管机构的反对，英特尔取消了原定以50亿美元 收购Tower的计划，当时许多员工都对此感到后悔，但如今，他们越来越觉得这反而是他们遇到的最好的事情。 在英特尔 50 亿美元的收购要约失败两年后，Tower 的首席执行官 Russell Ellwanger 解释了对硅光子技术的持续投资如何将潜在的收购方变 成了竞争对手，以及该公司为何认为其人工智能驱动的增长是结构性的，而不是炒作。 "几周前，我去Tower公司的自助餐厅吃午饭，和坐在桌子另一头的两位员工聊了起来，"Tower公司首席执行官Russell Ellwanger在接受Calcalist采访时回忆 道。"我问他们：你们感觉如何？对公司有什么看法？他们的回答是：'看看股价吧。我晚上回家，全家人，甚至邻居，都在跟我聊Tower的股票。他们觉 得我是个特别的人。'" Ellwanger经营这家公司已有二十年，他用这些话描述了Tower现在的氛围。 如果有人认为以色列人工智能革命的最北端是英伟达位于约克尼姆的总部，那他很可能没有关注这家位于米格达勒哈埃梅克的芯片制造 ...

公众号记得加星标⭐️，第一时间看推送不会错过。 在英特尔 50 亿美元收购计划失败两年后，Tower 首席执行官拉塞尔·埃尔旺格（Russell Ellwanger）解释了公司持续加大对硅光子技术的投入，如何让这位原本的潜在收购方转变为直接竞 争者，以及为何 Tower 认为自身由 AI 驱动的增长是结构性的，而非市场炒作。 "几周前，我下楼到 Tower 的员工餐厅吃午饭，和坐在桌子另一端的两名员工聊了起来，"Tower 首 席执行官拉塞尔·埃尔旺格在接受《Calcalist》采访时回忆道。"我问他们：你们感觉怎么样？你们怎 么看这家公司？他们的回答是：'看看股价吧。我每天晚上回到家，家里所有人，甚至邻居，都会跟 我聊 Tower 的股票。他们觉得我好像成了什么了不起的人。'" 埃尔旺格用这番话，描绘了如今在 Tower 内部弥漫的氛围。这位执掌公司已达二十年的 CEO 表示， 公司的整体士气正处在一个前所未有的高点。 如果有人认为以色列 AI 革命最北端的前沿阵地是位于约克尼阿姆（Yokneam）的英伟达总部，那他 显然还没有关注这家位于米格达尔哈埃梅克（Migdal HaEmek）的芯片制造商的表现。自 ...

Core Viewpoint - Recent market focus has shifted towards CPO (Co-Packaged Optics) technology, with companies like Tianfu Communication, Zhongji Xuchuang, and Xinyi Sheng showing active performance as AI computing power demands continue to surge [1] CPO Technology and Market Trends - CPO technology is seen as a key path for next-generation data center interconnects, addressing the limitations of traditional pluggable optical modules in terms of transmission speed and energy consumption [1] - Securities firms have released reports highlighting the potential of CPO and silicon photonics, with expectations that 1.6T optical modules will become a primary demand in 2024, extending the lifecycle of pluggable optical modules [1][2] - The integration of optical engines with computing chips in CPO technology raises the requirements for upstream optical chips and silicon photonics technology [2] Optical Modules and Advanced Packaging - Optical module manufacturers are crucial participants in the implementation of CPO technology, which is not expected to completely replace traditional pluggable optical modules but rather serve as a complementary or evolutionary direction in high-density, high-performance scenarios [3] - Leading optical module companies are expanding into CPO solution provision by developing core technologies such as silicon photonics and optical engines [3] Supporting Facilities and Materials - The promotion of CPO technology will drive demand for supporting facilities and specialized materials, including precision connectors and optical fiber arrays for efficient chip and optical engine coupling [4] - New packaging forms will necessitate advanced cooling solutions, potentially increasing the penetration of liquid cooling technologies [4] Company Information - Xingsen Technology has indicated that its products can be used for the packaging of CPO products, utilizing MSAP (Modified Semi-Additive Process) technology [5] - Anfu Technology has invested in Suzhou Yilong Micro Semiconductor Technology Co., which specializes in silicon photonic technology applicable to CPO optical modules, seen as a move to build a "second growth curve" [5] - Changguang Huaxin, as an IDM optical chip enterprise, is advancing its technology competitiveness in the silicon photonics field, with its 100G EML chip already in mass production and 200G EML in customer validation [5]

Dec. 8, 2025 产业洞察 根据Tr e n dFo r c e集邦咨询最新研究，随着数据中心朝大规模丛集化发展，高速互联技术成为决 定 AI 数 据 中 心 效 能 上 限 与 规 模 化 发 展 的 关 键 。 2 0 2 5 年 全 球 8 0 0G 以 上 的 光 收 发 模 块 达 2 4 0 0 万 支，2 0 2 6年预估将会达到近6 3 0 0万组，成长幅度高达2 . 6倍。 至于光收发模块当中，除了发射端的激光光源之外，同时也需要光二极管（Ph o t oDi o d e , PD） 作为光接收元件。为了要搭配更高速的激光光源传输速度，目前各家PD厂商也纷纷投入开发可 以接收2 0 0G传输信号的高速PD，包括Co h e r e n t, Ma c om, Br o a d c om,Lume n t um等厂商也推出 了2 0 0G的高速PD。 由于高速PD和EML与CW激光一样，皆采用INP(磷化铟) 基板再进行磊晶。在激光光源短缺的 情况下，激光厂商倾向将多数磊晶产能配置于生产激光光源，同时通过外包的方式将INP磊晶 交由iET-英特磊、全新等专业磊晶代工厂商协助生产。 Tr ...

Core Viewpoint - The article discusses the strategic transformation of Saiwei Electronics, highlighting its shift from traditional navigation to becoming a leading player in the MEMS semiconductor industry, particularly in the context of AI and data center infrastructure [3][4][5]. Group 1: Company Transformation - Saiwei Electronics has transitioned from a traditional military navigation company to a high-tech semiconductor foundry after acquiring Silex Microsystems, a leading MEMS chip manufacturer [6][4]. - The company has established a dual business model focusing on "semiconductors + special electronics," with MEMS manufacturing now at its core [6][4]. - The "Pure-Foundry" model allows the company to avoid direct competition with clients, attracting top-tier customers globally [7][6]. Group 2: MEMS Industry Position - MEMS technology is characterized by high technical barriers and emphasizes complex microstructure design [8]. - Saiwei Electronics holds a leading position in the global MEMS market, with Silex consistently ranked first in pure MEMS foundry services [10]. - The MEMS market is projected to grow at a CAGR of 3.7%, reaching $19.2 billion by 2030, indicating strong future demand [10]. Group 3: Financial Impact of Strategic Decisions - The sale of Silex has led to a significant drop in revenue, with a 17.4% year-on-year decline in the first three quarters of 2025 [14]. - Despite the revenue drop, the net profit surged by 1438% to 1.58 billion yuan, primarily due to a one-time investment gain from the sale [16][18]. - The company maintains high R&D investment, with R&D expenses reaching 300 million yuan in the first three quarters of 2025, indicating a commitment to innovation [19][18]. Group 4: Future Outlook - The focus will shift entirely to domestic production lines, particularly the Beijing FAB3, which is one of the few capable of large-scale MEMS production in China [22][23]. - The domestic market shows significant potential due to the rapid development of sectors like 5G, autonomous driving, and biomedical, which are driving demand for high-end MEMS devices [23][22]. - The company has also made strategic acquisitions to strengthen its position in the supply chain, with substantial cash reserves enhancing its resilience against industry fluctuations [25][24].

Group 1: Global Chip Industry Developments - Microsoft and Nvidia plan to invest $15 billion in AI startup Anthropic, which will purchase $30 billion worth of computing power from Microsoft's Azure cloud platform [1] - US semiconductor manufacturer GlobalFoundries announced the acquisition of silicon photonics wafer foundry AMF, positioning itself to become the largest silicon photonics chip manufacturer by revenue [1] - Chip design giant Arm announced the integration of Nvidia's NVLink Fusion high-speed interconnect into its Neoverse platform, strengthening the collaboration between the two companies [1] Group 2: Domestic Semiconductor Market Insights - The 22nd China International Semiconductor Expo (IC China 2025) will be held from November 23 to 25 in Beijing, aligning with the "14th Five-Year Plan" to enhance technological innovation and ensure supply chain stability [1] - China Galaxy noted that the rapid development of strategic emerging industries such as digital economy, AI, and smart connected vehicles will create unprecedented application scenarios and market space for domestic chips [1] Group 3: Flash Memory Price Surge - Following significant price increases in DRAM, flash memory prices have also surged, with Flash Wafer prices rising by up to 38.46% as of November 19 [2] - Guosen Securities highlighted strong AI demand leading to supply shortages and price increases in storage chips and other components, suggesting that the market may be underestimating the sustainability of demand for domestic computing and storage capabilities [2] Group 4: Electronic Sector Performance - The electronic ETF (515260) saw early gains of over 2% before a market pullback, indicating strong buying interest despite a current decline of 0.47% [2] - Key stocks in the electronic sector, such as Huadian Co. and Wentai Technology, led gains of over 2%, while companies like Zhongwei and Beifang Huachuang experienced declines of over 3% [4] Group 5: Electronic ETF Characteristics - The electronic ETF (515260) tracks the electronic 50 index, focusing on semiconductor and consumer electronics sectors, with significant holdings in PCB, AI chips, automotive electronics, and 5G [5] - The ETF's composition includes a 44.63% weight in Apple supply chain stocks, benefiting from the anticipated strong performance of the iPhone 17 [5] - External pressures are pushing China towards achieving self-sufficiency in the semiconductor industry, with AI reshaping consumer electronics and receiving strong policy support [5]

Core Insights - GlobalFoundries has announced the acquisition of Advanced Micro Foundry in Singapore, marking a significant step in its innovation strategy and leadership in silicon photonics [1] - The acquisition will enable GlobalFoundries to integrate manufacturing assets, intellectual property, and expertise from Advanced Micro Foundry, enhancing its silicon photonics technology portfolio and production capabilities in Singapore [1] - Following the acquisition, GlobalFoundries will become the largest pure-play silicon photonics chip foundry globally based on revenue [1] Industry Context - The driving force behind this acquisition is the urgent demand for data transmission efficiency in artificial intelligence infrastructure, as the computational power required is growing exponentially [2] - AI workloads are increasing computational demands by four to five times annually, while data transmission efficiency has not kept pace, leading to significant idle time for high-end accelerators in data centers [2] Technology Overview - Silicon photonics technology, which uses lasers instead of electronic signals for data transmission, is seen as a solution to high energy consumption and signal delay issues [3] - This technology can achieve data transfer speeds that are 100 times faster than traditional methods, making it a key component for next-generation semiconductor technology [3] Company Strategy - Advanced Micro Foundry has over 15 years of process accumulation in silicon photonics, with a focus on specialized processes and significant patents [4] - GlobalFoundries plans to leverage its large-scale manufacturing experience to transition these technologies from a 200mm platform to a 300mm platform, which is crucial for the future of co-packaged optics (CPO) technology [4] Competitive Landscape - GlobalFoundries, established in 2009, has shifted its focus from advanced processes to mature processes and specialty technologies, including silicon photonics [5] - As of Q2 2023, GlobalFoundries holds a 3.9% market share, ranking fifth among global foundries, while TSMC leads with a 70.2% share [5] - The silicon photonics foundry market is becoming competitive, with major players like TSMC and Intel also investing in this technology [5] Growth Plans - GlobalFoundries launched its Fotonix silicon photonics platform in March 2022, integrating 300mm photonic characteristics with RF-CMOS technology [6] - The company aims to expand its production capacity, targeting a total output of 2 million 12-inch wafers by 2024, with plans to increase this to 3 million [6] - An investment of €1.1 billion is planned to expand its facility in Dresden, Germany, with expectations to boost annual production capacity to over 1 million wafers by the end of 2028 [6]

Core Insights - GlobalFoundries (GF) has acquired Advanced Micro Foundry (AMF), a Singapore-based silicon photonics chip manufacturer, with the financial details undisclosed [1][3] - GF plans to establish a silicon photonics R&D Center of Excellence in Singapore in collaboration with A*STAR, focusing on next-generation materials for 400Gbps ultra-high-speed data transmission [1][3] Company Expansion - The acquisition will enhance GF's silicon photonics technology portfolio, capacity, and R&D capabilities in Singapore, complementing its existing technologies in the U.S. and opening new market opportunities in data centers and communication technologies [3] - By integrating AMF's manufacturing assets, intellectual property, and expertise, GF aims to become the highest-revenue pure-play silicon photonics foundry globally [3] Market Demand and Applications - Silicon photonics is a rapidly growing field, widely used in AI data centers and quantum computers, combining traditional chip manufacturing with optical data transmission technologies [5] - GF plans to leverage AMF's 200mm platform to meet demands in long-distance optical communication, computing, lidar, and sensing, with plans to expand to a 300mm platform as market demand grows [3][6] Strategic Vision - GF's CEO Tim Breen emphasized the increasing need for faster, more precise, and energy-efficient data transmission, which is critical for AI data centers and advanced telecom networks [6] - The acquisition of AMF allows GF to offer a more comprehensive and differentiated roadmap for pluggable transceivers and co-packaged optical devices, while accelerating the development of photonics technology in adjacent markets like automotive and quantum computing [6]