半導體 Semiconductor

CoWos Three Types MindMap (Chillyin illustrated, 2025)

參考文獻 reference:

Taiwan Semiconductor Manufacturing Company. (n.d.). CoWoS®. TSMC 3DFabric™. Retrieved Feb 11, 2025, from https://3dfabric.tsmc.com/english/dedicatedFoundry/technology/cowos.htm

https://3dfabric.tsmc.com/chinese/dedicatedFoundry/technology/cowos.htm 

#Chillyin的半導體觀察筆記 🌹 

#ChillyinSEMI 

Integrator: Chilly Chiou 《CHILLYIN LTD.》

 CoWoS 三種類型簡述及比較  

簡言之

#CoWoSS 性能最佳但成本最高

#CoWoSR 成本最低但性能較低

#CoWoSL 則兼顧性能和成本，並提供更高的 HBM 堆疊數量和尺寸彈性

台積電的 CoWoS

#先進封裝 技術主要有三種類型：CoWoS-S、CoWoS-R 和 CoWoS-L，它們的主要區別在於

#中介層 使用的材料。

1. CoWoS-S 

中介層：矽中介層 (Silicon Interposer)

特性：性能最佳，但成本最高。

矽中介層利用

#矽穿孔技術 ，提供高密度互連和良好的散熱性能。

適合 AI 伺服器晶片、高效能運算 (HPC) 產品，例如 NVIDIA Hopper H100、H200 和 AMD MI300。

缺點：矽材料較脆弱，尺寸放大後良率難以提升。

2. CoWoS-R 

中介層：重分佈層 (RDL Interposer)

特性：成本較 CoWoS-S 低，封裝尺寸彈性較高。

RDL 中介層由

#聚合物和銅線 構成，具備較高的彈性，允許封裝尺寸進一步擴展。

適合網通設備、邊緣 AI 等產品。

缺點：性能不如 CoWoS-S。

3. CoWoS-L 

中介層：重分佈層 + 局部矽互聯 (LSI)

特性：結合 CoWoS-S 和 CoWoS-R 的優點，成本介於兩者之間。

局部區域使用矽中介層 (LSI) 進行高速互連，其他區域使用

#重佈線層 ，提供高度靈活的整合能力。

可堆疊的 HBM 數量較 CoWoS-S 多，最多可達 12 顆。

適合 NVIDIA Blackwell 系列晶片，例如 B100、B200、B300 和 GB200。 

 CoWoS - A Brief Introduction and ComparisonOfCoWos Three Types 

In short, CoWoS-S offers the best performance but comes at the highest cost. CoWoS-R is the most cost-effective but has lower performance. CoWoS-L balances performance and cost while providing higher HBM stacking capacity and size flexibility.

TSMC's CoWoS advanced packaging technology mainly comprises three types: CoWoS-S, CoWoS-R, and CoWoS-L. Their main difference lies in the materials used for the interposer. 

●CoWoS-S 

Interposer: Silicon Interposer 

Features: Best performance, but highest cost.

Utilizes silicon vias to provide high-density #interconnect and good thermal performance. 

Suitable for AI server chips and high-performance computing (HPC) products, such as NVIDIA Hopper H100, H200, and AMD MI300. 

Drawbacks: Silicon material is relatively fragile, making it difficult to improve yield when scaling up the size.

●CoWoS-R 

Interposer: Redistribution Layer (#RDL Interposer)

Features: Lower cost and higher package size flexibility compared to CoWoS-S.

The RDL interposer is composed of #polymers and #copper wires, offering high flexibility and allowing for further expansion of package size.

Suitable for network communication equipment and edge AI products.

Drawbacks: Performance is not as good as CoWoS-S.

●CoWoS-L 

Interposer: Redistribution Layer + Local Silicon Interconnect (#LSI)

Features: Combines the advantages of CoWoS-S and CoWoS-R, with the cost falling between the two.

Uses LSI for high-speed interconnect in local areas and redistribution layers in other areas, providing highly flexible integration capabilities.

Offers higher HBM stacking capacity than CoWoS-S, up to 12 chips.

Suitable for NVIDIA Blackwell series chips, such as B100, B200, B300, and GB200. 

晶片戰爭下的科技競逐 (Chip War: Tech Race & Reshaping) MindMap

參考文獻 reference:

Ko, Justin (2025). TSMC, SMIC, and the Global Chip War.

Hübner, Jörg (2025). Why Is the Chip Industry So Special? In Business and Policy Challenges of Global Uncertainty (pp. 235-251).

Ryu, Yongwook (2025). Chips on the Deck: US-China Rivalry and Reorganizing the Supply Chains of Semiconductors. In N. Hung Son & N. Thi Lan Anh (Eds.), The South China Sea: The Geo-political Epicenter of the Indo-Pacific? (Chapter 8). Springer, Singapore.

 晶片戰爭下的科技競逐 

當前全球科技發展的核心，無疑聚焦於半導體產業。這場被喻為「晶片戰爭」的競賽，不僅是技術實力的較量，更深刻地重塑了全球供應鏈格局，並與複雜的地緣政治緊密交織。綜合近期多份研究報告，我們可以清晰地看到幾個關鍵的科技發展趨勢、重點討論議題以及被特別強調的觀點。

一、 技術自主與「蠻力突破」：非對稱的創新路徑

在美國對中國半導體產業實施嚴格出口管制，特別是限制先進製程設備（如ASML的EUV光刻機）的背景下，中國的晶片製造商如中芯國際（SMIC）正試圖以非傳統方式突圍 。Justin Ko在其研究中指出，中芯國際在2023年據報導成功為華為Mate 60生產7奈米晶片（麒麟9000s），並可能正利用舊有的DUV（深紫外光刻）工具，透過自對準四重圖案化（SAQP）等「蠻力」手段，嘗試生產5奈米晶片 。

這一趨勢的重點在於：

• 規避制裁的創新驅動力： 出口限制反而可能激發了在現有技術框架下尋求極限突破的動機 。若非EUV設備取得受限，企業不太可能投入巨大資源去優化效率較低、速度較慢的生產技術。

• 供應鏈的「紅色」自主化： 中芯的技術突破，即使在效率和成本上不及主流EUV方案，其地緣政治意義重大。它增強了中國本土晶片消費者（如華為、阿里巴巴）使用國產晶片的意願，以降低對外國供應商（如NVIDIA、TSMC）的依賴，並減輕未來遭受更嚴厲制裁的風險 。這將直接推動一個更能抵禦外部限制的「紅色晶片供應鏈」的形成 。

• 非對稱競爭： 這種「蠻力突破」雖然在技術指標上可能追趕，但在生產效率、成本和良率方面，與台積電等採用最新EUV技術的領導者相比，仍存在巨大差距。然而，其戰略價值在於展示技術潛力和部分自主能力。

二、 全球供應鏈的專業化、集中化及其脆弱性

Jörg Hübner在其研究摘要中強調，微晶片是幾乎所有用電產品的基礎成分，不僅革新了資訊處理和通訊，也在再生能源系統和穩定電網中扮演不可或缺的角色。先進晶片的生產具有高度軍事意義，支持基於AI的先進武器系統，從而產生地緣政治影響。

Yongwook Ryu也指出，半導體產業擁有全球最複雜、高度整合的供應鏈之一，不同任務由不同地區的不同參與者執行 。這種高度專業化和全球價值鏈上的整合，使得關鍵的價值鏈活動集中在特定地理區域：

• 設計與IP： 美國和英國在電子設計自動化（EDA）工具和核心IP方面佔據主導 。

• 製造設備： 美國、日本和歐洲是半導體製造設備的主要供應商 。ASML的EUV光刻機是先進製程的瓶頸技術 。

• 晶圓代工： 最先進的製造設施集中在東亞，特別是台灣（台積電）和韓國（三星）。台積電在全球晶圓代工市場長期佔據超過50%的份額，甚至接近67% 。

• 封裝測試（OSAT）： 中國在後端封測領域佔有約20%的市場份額，但此環節相對容易被替代 。

這種高度集中和專業化的結果是，全球晶片價值鏈變得高度耦合且缺乏彈性，難以承受重大干擾，正如COVID-19疫情期間所見證的那樣。Hübner的摘要指出，這促使美國、歐洲以及東亞的市場領導者紛紛試圖提升本地產能。

三、 地緣政治對決下的供應鏈重組與「友岸外包」

美中之間的戰略競爭已從貿易戰轉向科技競爭，半導體領域首當其衝 。Ryu在其研究中詳細闡述了美國試圖利用其在晶片生產前端（軟體、EDA、IP）的市場主導地位以及其科技盟友網絡，迫使中國與全球晶片供應鏈「脫鉤」或顯著降低其地位 。

美國的半導體政策目標有二：

1. 建立本土安全供應鏈： 透過《晶片法案》(Chips for America Act) 和《美國晶圓代工法案》(American Foundries Act) 等產業政策，提供稅收優惠和資金支持，鼓勵在美國本土投資設廠，提升本土製造能力 。美國的製造產能已從1990年的37%下降到2010年的13%，預計到2030年可能進一步降至10% 。

2. 限制中國的技術進步： 將華為、中芯國際等中國企業列入實體清單，限制其獲取美國技術和先進設備（如EUV）。出口管制範圍也從最初的邏輯晶片擴展到記憶體晶片，並從先進製程延伸至中階製程 。

與此同時，美國積極推動與歐洲和亞洲盟友的合作，即所謂的「友岸外包」(friend-shoring)，加強與韓國、台灣、日本、荷蘭、德國等的夥伴關係 。台積電和三星應美國政府要求，已分別承諾在亞利桑那州和德州建廠 。

四、 領導者與企業文化：塑造產業格局的無形力量

Ko的著作特別強調了企業領導者及其背景對公司發展軌跡的深遠影響。他詳細比較了台積電創辦人張忠謀 (Morris Chang) 和中芯國際創辦人張汝京 (Richard Chang) 的相似背景（均出生於江南，赴美接受工程教育，曾任職於德州儀器）以及他們各自的經營理念和人生抉擇 。

• 張忠謀與台積電的「純晶圓代工」模式： 張忠謀開創並堅守「純晶圓代工」模式，不與客戶競爭晶片設計，贏得了全球設計公司的信任 。其嚴格的管理（如對保密的高度重視）和對技術創新的持續投入（如支持林本堅的浸潤式光刻技術和胡正明的FinFET技術商業化）是台積電成功的關鍵 。台積電的反週期投資策略也使其能在經濟下行時擴大產能，鞏固領先地位 。

• 張汝京的「紅色供應鏈」夢想與SMIC的挑戰： 張汝京則帶有更強的家國情懷，致力於在中國大陸建立自主的半導體產業 。他以極度注重成本控制和優先扶持本土供應商著稱 。儘管SMIC因與台積電的專利訴訟遭遇重挫（張汝京因此辭職），但其後在政府支持和後續領導者（如蔣尚義、邱慈雲、梁孟松等同樣具有台灣背景和美國教育經歷的工程師）的努力下，SMIC依然是中國大陸半導體自主化最重要的希望 。

五、 未來的競賽：技術創新、效率與地緣政治的角力

Ryu總結道，美中之間的晶片競賽將日益激烈，其動態類似於軍備競賽，雙方都關注相對收益，使之成為一場零和遊戲 。美國憑藉其盟友網絡在重組供應鏈方面具有成本優勢，而中國則面臨被孤立的風險，儘管其會努力尋求自主並分化美國的同盟 。

最終，誰能主導晶片領域，不僅取決於技術的先進性，更取決於創新能力、生產效率以及在複雜地緣政治環境中的運籌帷幄 。各國政府必須與企業密切協商，制定長期的研發和產業策略 。正如張忠謀所言，在其任內從未遇到的「友岸外包」、「境內製造」等地緣政治挑戰，已成為當前領導者必須面對的現實 。而張汝京的經歷則顯示，即便面臨制裁和訴訟，強烈的使命感和政府支持也能催生出堅韌的追趕者。

這場晶片戰爭的結果，將深遠影響全球科技版圖和國際力量的平衡。

 The Tech Race in the Chip War 

The core of current global technological development undoubtedly revolves around the semiconductor industry. This competition, often dubbed the "Chip War," is not merely a contest of technical prowess but is also profoundly reshaping global supply chain structures and is intricately intertwined with complex geopolitical dynamics. Synthesizing insights from several recent research reports, we can clearly identify key technological development trends, focal points of discussion, and particularly emphasized perspectives.

I. Technological Self-Reliance and "Brute Force Breakthroughs": Asymmetric Paths to Innovation

Against the backdrop of strict U.S. export controls on China's semiconductor industry, particularly restrictions on advanced process equipment like ASML's EUV lithography tools, Chinese chip manufacturers such as Semiconductor Manufacturing International Corporation (SMIC) are attempting to achieve breakthroughs through unconventional means. Justin Ko notes in his research that SMIC reportedly succeeded in producing a 7-nanometer chip, the Kirin 9000s, for Huawei's flagship Mate 60 phone in 2023, and may now be on the verge of producing 5nm chips using older deep ultraviolet (DUV) tools through processes like Self-Aligned Quadruple Patterning (SAQP) – a "brute force" approach.

Key aspects of this trend include:

• Innovation Driven by Sanctions Circumvention: Export restrictions may have ironically spurred motivation to seek extreme breakthroughs within existing technological frameworks. If EUV tools were readily available, it's unlikely a company would invest so much effort in optimizing inefficient and slower production techniques.

• "Red" Autonomous Supply Chain: SMIC's technological advancements, even if less efficient and cost-effective than mainstream EUV solutions, carry significant geopolitical weight. They increase the likelihood of Chinese domestic chip consumers like Huawei or Alibaba using domestically produced chips and DUV-based techniques to reduce dependence on foreign suppliers like Nvidia and TSMC, thereby mitigating their risk of future sanctions. This will directly contribute to the creation of a self-reliant "red chip supply chain" more resilient to external restrictions.

• Asymmetric Competition: While these "brute force breakthroughs" might catch up in terms of technical specifications, they still face a significant gap in production efficiency, cost, and yield compared to leaders like TSMC, which use the latest EUV technology. However, their strategic value lies in demonstrating technological potential and a degree of self-sufficiency.

II. Specialization, Concentration, and Vulnerability of Global Supply Chains

Jörg Hübner, in his research abstract, emphasizes that microchips are fundamental components in almost all electrically powered products, revolutionizing information processing and communication, and serving as indispensable parts in renewable energy systems and stable power grids. The production of leading-edge chips is of high military importance, supporting advanced weapon systems using AI, thus having geopolitical implications.

Yongwook Ryu also points out that the semiconductor industry possesses one of the most complex and highly integrated supply chains globally, with different tasks performed by various actors in different locales. This high degree of specialization and integration along the global value chain has led to the concentration of essential activities in specific geographical regions:

• Design & IP: The U.S. and UK dominate in Electronic Design Automation (EDA) tools and core IPs.

• Manufacturing Equipment: The U.S., Japan, and Europe are major suppliers of semiconductor manufacturing equipment. ASML's EUV lithography machines are a bottleneck technology for advanced processes.

• Foundry Services: The most advanced manufacturing facilities are concentrated in East Asia, particularly Taiwan (TSMC) and South Korea (Samsung). TSMC has long held over 50% of the global foundry market share, reaching nearly 67%.

• Assembly, Testing, and Packaging (OSAT): China holds about 20% of the market share in the back-end OSAT segment, but this segment is relatively easier to replace.

The result of this high concentration and specialization is that the global chip value chain has become tightly coupled and less flexible, making it vulnerable to major disruptions, as witnessed during the COVID-19 pandemic. Hübner's abstract notes that this has prompted various governments from the U.S. and Europe, as well as incumbent market leaders in East Asia, to attempt to advance their local production capacities.

III. Supply Chain Reorganization and "Friend-Shoring" Amidst Geopolitical Rivalry

The strategic rivalry between the U.S. and China has shifted from a trade war to technological competition, with the semiconductor sector at the forefront. Ryu, in his research, details how the U.S. is attempting to leverage its market dominance in the front-end of chip production (software, EDA, IP) and its extensive network of tech-capable allies to force China to "decouple" from global chip supply chains or significantly reduce its role.

U.S. semiconductor policy has two main goals:

1. Establishing a Secure Domestic Supply Chain: Through industrial policies like the Chips for America Act and the American Foundries Act, the U.S. provides tax incentives and funding to encourage investment in domestic manufacturing facilities, aiming to boost its own production capacity. U.S. manufacturing capacity declined from 37% of global capacity in 1990 to 13% in 2010, and is projected to fall further to 10% by 2030.

2. Restricting China's Technological Advancement: Placing Chinese companies like Huawei and SMIC on entity lists limits their access to American technology and advanced equipment (like EUV). Export controls have expanded from logic chips to memory chips and from advanced to mid-tier processes.

Concurrently, the U.S. is actively promoting "friend-shoring," strengthening partnerships with allies in Europe and Asia, including South Korea, Taiwan, Japan, the Netherlands, and Germany. TSMC and Samsung, at the U.S. government's request, have committed to building fabs in Arizona and Texas, respectively.

IV. Leadership and Corporate Culture: Intangible Forces Shaping the Industry

Ko's work particularly highlights the profound impact of corporate leaders and their backgrounds on their companies' trajectories. He draws detailed comparisons between TSMC founder Morris Chang and SMIC founder Richard Chang, noting their similar backgrounds (both born in the Jiangnan region, U.S.-educated engineers, former Texas Instruments employees) yet distinct business philosophies and life choices.

• Morris Chang and TSMC's "Pure-Play Foundry" Model: Morris Chang pioneered and adhered to the pure-play foundry model, refraining from chip design to avoid competing with customers, thereby earning the trust of global design houses. His stringent management (e.g., extreme emphasis on confidentiality) and continuous investment in technological innovation (such as supporting Burn-Jeng Lin's immersion lithography and Chenming Hu's FinFET commercialization) were key to TSMC's success. TSMC's counter-cyclical investment strategy also allowed it to expand capacity during economic downturns, solidifying its lead.

• Richard Chang's "Red Supply Chain" Dream and SMIC's Challenges: Richard Chang was driven by a stronger sense of patriotism, dedicating himself to building an indigenous semiconductor industry in Mainland China. He was known for his intense focus on cost control and prioritizing domestic suppliers. Although SMIC faced a major setback due to a patent lawsuit with TSMC (which led to Richard Chang's resignation), with government support and the efforts of subsequent leaders (many of whom, like Jiang Shangzhou, Chiu Tzi-yun, and Liang Mong-song, also had Taiwanese backgrounds and U.S. educations), SMIC remains Mainland China's best hope for semiconductor self-sufficiency.

V. The Future Contest: A Tug-of-War of Innovation, Efficiency, and Geopolitics

Ryu concludes that the U.S.-China chip competition will likely intensify, taking on dynamics similar to an arms race, with both powers concerned about relative gains, turning it into a zero-sum game. The U.S. has a cost advantage in reorganizing supply chains due to its network of allies, while China faces increasing isolation, though it will strive for self-sufficiency and attempt to create wedges in U.S. alliances.

Ultimately, dominance in the chip sector will depend not only on technological advancement but also on innovation capabilities, production efficiency, and strategic maneuvering in a complex geopolitical environment. National governments must work closely with corporations to formulate long-term R&D and industrial strategies. As Morris Chang noted, geopolitical challenges like "friend-shoring" and "onshoring," which he never encountered during his tenure, are now realities that current leaders must navigate. Richard Chang's experiences, on the other hand, show that even in the face of sanctions and lawsuits, a strong sense of mission and government backing can foster resilient contenders.

The outcome of this chip war will profoundly influence the global technological landscape and the international balance of power.