Observing TSMC's SiC Strategy ：SiC Enters the Advanced Packaging Mainstage 观察台积电的碳化硅战略：碳化硅进入先进封装的 主舞台 Original Articles By SemiVision Research (Nvidia ,TSMC, Globalwafers, Wolfspeed) SemiVision Research 原创文章（英伟达、台积电、环球晶圆、Wolfspeed） SEP 21, 2025 2025 年 9 月 21 日 ∙ PAID ∙ 付费 SEMIVISION 7 Share 分享 20 shorten the power path and reduce PDN impedance, enabling faster voltage response. ASE, through its VIPack and FOCoS-Bridge (with TSV) platforms, is optimizing PDN, signal interconnects, and thermal performance sim ...

安靠科技公司（Amkor Technology）今日与美国政府共同宣布，将在美国亚利桑那州建设一座全新的、最先进的 外包半导体先进封装与测试园区，并计划大幅扩大投资规模。 此次扩建投资包括新增洁净室空间和第二座新建封测工厂，使项目总投资额增加超过50亿美元，达到总计70亿美 元，分两期实施。奠基仪式汇聚了政府官员、业界领袖及社区代表，共同庆祝这一重要里程碑。安靠的投资将支 持强化美国半导体领导力的国家战略，并成为美国首座大规模先进封装生产基地。 项目全部完工后，园区将拥有超过75万平方英尺的洁净室空间，并创造多达3000个高质量岗位。首座生产厂房预 计于2027年年中完工，2028年初投产。这些新建设施将成为美国先进封装能力的基石，服务包括苹果（Apple） 与英伟达（NVIDIA）在内的核心客户。安靠的扩建项目获得了特朗普政府的"美国芯片法案（CHIPS for America Program）"、先进制造业投资税收抵免政策，以及州与地方政府的支持。 安靠科技总裁兼首席执行官Giel Rutten表示："此次奠基是安靠在长期增长与创新战略上的大胆一步。我们正在建 设一座能够满足客户最先进需求的工厂，它将推动 ...

Investment Rating - The report rates the semiconductor industry as "Outperform" [1] Core Insights - The dual engines of storage and computing power are driving the AI era, with Q2 revenue for the top 60 global semiconductor companies increasing by 9% quarter-on-quarter, and an expected 8% increase for Q3 [12][15] - The semiconductor sector is experiencing a strong recovery, with various segments showing significant growth in revenue and profit margins [19][24] Summary by Sections Global Semiconductor Leaders - The top 60 global semiconductor companies saw a Q2 revenue increase of 9% quarter-on-quarter, with an expected 8% increase for Q3 [15][19] - The semiconductor sales for Q2 reached approximately $179.7 billion, with a year-on-year growth of 20% [15] - The expected Q3 revenue for these companies is projected to be around $225.5 billion, reflecting a year-on-year increase of 23% [15] Semiconductor Equipment - The semiconductor equipment sector's revenue increased by 4% in Q2, primarily due to sales growth in Taiwan [2][21] - Q3 revenue is expected to decline by 1% due to a decrease in guidance from equipment manufacturers [2][21] Semiconductor Silicon Wafers - The silicon wafer sector's revenue increased by 7% in Q2, with strong demand for AI products [3][24] - Q3 revenue is expected to remain flat, as demand for AI continues while inventory digestion in other areas takes time [3][24] Packaging and Testing - The packaging and testing sector's revenue increased by 9% in Q2, with an expected 14% increase in Q3 due to strong demand for advanced packaging [4][21] Upstream Manufacturing - The wafer manufacturing sector is benefiting from strong demand for advanced processes, with Q2 revenue increasing by 16% and an expected 6% increase for Q3 [3][21] Semiconductor Design - The storage segment is expected to see a revenue increase of 11% in Q3, driven by strong AI demand and geopolitical factors [4][21] - The MPU segment is projected to grow by 15% in Q3, supported by robust data center demand [4][21] - The RF segment is expected to rebound with a 14% increase in Q3, driven by the launch of the iPhone 17 series [4][21] Investment Recommendations - The report emphasizes the importance of focusing on companies within the AI supply chain and those that are undervalued or in a turnaround situation [11][24]

Summary of Key Points from the Conference Call Industry and Company Overview - The discussion centers around the semiconductor industry, particularly focusing on advanced packaging technologies and the role of Silicon Carbide (SiC) in AI chip design and manufacturing. Key players mentioned include TSMC, NVIDIA, AMD, Google, and AWS, with a specific emphasis on TSMC's strategies and innovations in packaging solutions [1][2][3]. Core Insights and Arguments 1. **Challenges in AI Chip Design**: The increasing complexity and power demands of AI chips have led to significant challenges in power delivery networks (PDNs) and thermal management. Traditional methods are becoming inadequate, with single GPUs now requiring over 1000A of current [5][19]. 2. **Innovative Solutions**: Companies like Marvell and ASE are proposing solutions such as Package-Integrated Voltage Regulators (PIVR) and optimized PDN platforms to address these challenges. TSMC is also innovating with its CoWoS-L platform, which integrates embedded voltage regulators and advanced thermal management techniques [7][10][11]. 3. **Emergence of SiC**: SiC is highlighted as a critical material for AI chip and system design due to its superior properties, including high thermal conductivity and mechanical strength. It is increasingly being viewed as essential for advanced packaging and heterogeneous integration [13][14][16]. 4. **Market Demand**: The demand for ultra-large-scale GPUs and ASICs is driven by generative AI and large-scale model training, with power consumption often exceeding 1 kW. This has exposed bottlenecks in thermal management and power delivery [19][20]. 5. **Bottlenecks Identified**: The exponential growth in AI computing has revealed three critical bottlenecks: thermal challenges, power delivery bottlenecks, and electro-optical integration demands. TSMC is actively addressing these through its 3DFabric strategy and various packaging solutions [22][28][30][32]. Additional Important Content 1. **SiC's Role in Advanced Packaging**: SiC is positioned as a hybrid integration enabler, linking power delivery, thermal dissipation, and optical interconnects. Its unique properties make it suitable for high-voltage integrated circuits (HVICs) and optical interposers [40][44]. 2. **Competitive Landscape**: TSMC's exploration of SiC as an interposer material could provide a competitive edge in thermal management and electro-optical integration, especially compared to Intel and Samsung, who are also advancing their own technologies [45][46]. 3. **Challenges Ahead**: The successful commercialization of SiC in advanced packaging faces challenges such as defect density control in large-size wafers, process compatibility, and cost structure improvements [53][54]. 4. **Future Directions**: The integration of SiC into TSMC's platforms like COUPE and CoWoS-Next could reshape the AI semiconductor supply chain, establishing new industrial advantages in the AI and high-performance computing (HPC) era [44][97]. This summary encapsulates the critical insights and developments discussed in the conference call, emphasizing the strategic importance of SiC in the evolving semiconductor landscape.

法人指出，颖崴向来聚焦高频高速、高端市场，在5G手机APU、服务器、HPC、GPU、CPU等高复杂 度产品线测试界面皆已布局。今年以来，垂直探针卡产品营收占比达双位数，MEMS产能持续提升，为 今年运营增长重要动能之一。为满足客户需求，颖崴稳定提高探针自制率，探针产能今年目标从去年 300万提升至450万针。法人预期，颖崴随AI相关大客户新品上市，加上新产能持续开出，下半年营收 有望逐季向上，并优于去年同期。 旺硅来自国际大客户需求强劲，旗下高端探针卡接单大爆满且稼动率满载，公司连续两年大扩产，去年 底垂直式探针卡（VPC）月产能达约90万针、MEMS探针卡月产能约40万针，今年第二季度VPC月产能 已突破120万针，MEMS更目标在今年底挑战100万针规模。 精测推动更多HPC、AI、车用等新专案，且在北美客户市占率持续扩大，美系GPU客户，精测主要供应 晶圆测试、成品测试所需测试载板、PCB等零组件，近期都通过新一代产品验证。 半导体技术持续突破，推动芯片性能与能耗表现进入另一阶段，在这波技术革新中，相关测试与封装扮 演不可或缺角色，其中探针卡是确保芯片品质与性能核心关键，测试界面"千金股"颖崴、旺硅、精 ...

Core Viewpoint - The article discusses the transformative Ultra Ethernet (UE) 1.0 standard, which defines a high-performance Ethernet protocol for artificial intelligence (AI) and high-performance computing (HPC) systems, emphasizing its innovative Ultra Ethernet Transport (UET) layer designed for reliable, high-speed communication in large-scale systems [2][4]. Group 1: Overview of Ultra Ethernet - Ultra Ethernet (UE) aims to standardize high-performance networking for AI and HPC, addressing limitations of existing protocols like InfiniBand and RoCE [4][8]. - The development of UE involved collaboration among major tech companies, leading to the formation of the Ultra Ethernet Consortium (UEC) in July 2023, with over 100 member companies by the end of 2024 [9][10]. - UE is designed to be compatible with existing Ethernet infrastructure, allowing for easy deployment and scalability in data centers [10][11]. Group 2: Technical Innovations - The UET layer allows for hardware-accelerated communication, significantly improving computational efficiency by a factor of 1000 for every bit of data transmitted [2][7]. - UE introduces a connectionless API and supports various topologies, including traditional fat tree and optimized structures, to meet the scalability needs of future AI systems [10][12]. - The protocol supports multiple delivery modes, including reliable unordered delivery and reliable ordered delivery, catering to different application requirements [49][50]. Group 3: Addressing Limitations of Existing Protocols - Previous protocols like RoCE faced challenges such as head-of-line blocking and congestion issues, which UE aims to resolve through innovative congestion management and packet delivery mechanisms [6][10]. - UE's design allows for packet spraying, which distributes packets across multiple paths to avoid traffic polarization and improve bandwidth utilization [22][21]. - The UET layer is built to operate seamlessly over existing Ethernet networks, ensuring compatibility while enhancing performance [14][27]. Group 4: Application and Use Cases - UE is applicable in various network types, including local networks connecting CPUs to accelerators, backend networks for high-performance connections, and frontend networks for traditional data center operations [12][13]. - The standard provides three configuration profiles (HPC, AI Full, and AI Base) to support different functionalities and complexities in implementation [24][25]. - The architecture of UE is designed to facilitate efficient communication in large-scale systems, making it suitable for modern AI workloads and HPC applications [28][29].

Core Viewpoint - CANG.US announced the acquisition of a 50 MW Bitcoin mining facility in Georgia for $19.5 million, marking its entry into the Bitcoin mining and energy infrastructure sector, and initiating a new chapter in diversification [1] Group 1: Acquisition Details - The acquisition represents a critical first step for CANG.US in expanding its self-operated mining facilities [1] - The facility will allocate 30 MW for self-mining operations and 20 MW for third-party hosting services [1] Group 2: Strategic Goals - The acquisition aims to enhance operational efficiency and financial resilience for the company [1] - CANG.US is actively planning to expand energy supply into high-performance computing (HPC) applications, broadening the long-term development potential beyond Bitcoin mining [1] - The company seeks to leverage its internal operational and technological advantages as it develops its energy strategy [1]

Core Viewpoint - Riot Platforms, Inc. has experienced a significant stock price increase of 42% year-to-date, outperforming Bitcoin and CoinShares Bitcoin mining ETF, both of which have risen by 20% during the same period. However, JPMorgan has downgraded its stock rating from "Overweight" to "Neutral" while raising the target price from $14 to $15, citing a "slightly elevated" valuation as the reason for the downgrade [1][1][1] Company Summary - Riot Platforms focuses on developing and operating large-scale data centers, with its core facilities located in Rockdale and Corsicana, Texas. The company employs a vertical integration strategy that combines mining operations with proprietary engineering and power procurement capabilities to produce Bitcoin as its primary product [1][1][1] Industry Context - The future stock price increase for Riot Platforms is contingent upon two main catalysts: either achieving high-performance computing (HPC) collaboration agreements or a sustained rise in Bitcoin prices. Analysts acknowledge the potential for Riot to diversify its revenue streams beyond cryptocurrency mining, but express caution due to uncertainties in the transformation timeline [1][1][1]

Core Viewpoint - The article emphasizes the growing demand for high-end printed circuit boards (PCBs) driven by advancements in AI, high-performance computing (HPC), and smart devices, highlighting the significant market opportunities for companies like Huadian Co., Ltd. [2][3][36] Group 1: Industry Demand and Trends - High-end PCBs are increasingly sought after due to the requirements of AI and HPC, which necessitate advanced server technologies and high-density interconnects, leading to rising values for these components [3][4] - The transition of optical modules from 800G to 1.6T demands improved PCB manufacturing processes, such as mSAP technology, which presents high technical challenges but also high added value [3] - The automotive and wearable device sectors are also driving demand for lighter, more efficient, and faster PCBs, indicating a shift towards high-end products in these markets [4] Group 2: Company Performance - Huadian Co., Ltd. is positioned as a leading player in the domestic PCB market, with a strong performance forecast for the first half of 2025, projecting a net profit of 1.65 billion to 1.75 billion yuan, representing a year-on-year growth of 45% to 53% [6][8] - The company reported a revenue increase of 49% in 2024, with net profit growth of 71%, driven primarily by high-end PCB products [8][9] Group 3: Strategic Advantages - Huadian Co., Ltd. has successfully integrated into the supply chains of major international tech companies, with 86.5% of its revenue coming from overseas in 2024, showcasing its strong market position [17] - The company has a robust R&D investment strategy, increasing from 320 million yuan in 2019 to 790 million yuan in 2024, focusing on advanced technologies such as HDI and high-frequency materials [23][25] - The gross margin for PCB products has improved from 30.4% in 2019 to 35.9% in 2024, reflecting the company's ability to enhance profitability through technological advancements [26] Group 4: Future Outlook - The company plans to invest 4.3 billion yuan in expanding its high-end PCB production capacity, indicating confidence in sustained industry demand [33] - The global automotive PCB market is expected to exceed 30 billion USD by 2025, with Huadian Co., Ltd. already achieving breakthroughs in high-end automotive products [35] - The anticipated compound annual growth rate (CAGR) of 40.2% for AI server and HPC-related PCBs from 2023 to 2028 presents significant growth opportunities for the company [34]

Group 1 - TSMC reported a record revenue of $30.07 billion in the second quarter, with AI-related revenue exceeding $10 billion for the first time in a single quarter, indicating strong growth potential for the year [1] - The company expects AI accelerator contributions to revenue to double compared to last year, projecting AI-related revenue to reach approximately NT$434.1 billion in 2024 and NT$868.3 billion in 2025 [1] - In the second quarter, revenue from A-chip manufacturing and advanced packaging was approximately $8.78 billion, a year-on-year increase of 3.67 times, while high-performance computing (HPC) chip revenue was $9.26 billion, a year-on-year increase of 9.8% [1] Group 2 - TSMC's chairman emphasized that despite external factors like tariffs and currency fluctuations, there has been no change in customer behavior, with continued strong demand for AI [2] - The company raised its revenue growth forecast for the year to approximately 30% due to strong demand for advanced processes and growth in HPC platforms [2] - The rapid development of AI applications is expected to drive long-term demand, with significant growth in the processing of text tokens for large language models and sovereign AI needs [2]