Foundry model
The foundry model is a business strategy in the semiconductor industry wherein a specialized fabrication company, known as a pure-play foundry, manufactures integrated circuits (ICs) and other microelectronic devices exclusively for third-party clients, without designing or marketing its own semiconductor products.[1] This approach separates chip design from production, enabling fabless semiconductor firms—those that focus solely on design and innovation—to outsource manufacturing and reduce capital-intensive investments in fabrication facilities.[2] Pioneered in 1987 by Dr. Morris Chang through the establishment of Taiwan Semiconductor Manufacturing Company (TSMC) as the world's first dedicated foundry, the model emerged amid economic pressures in Taiwan and the growing complexity of semiconductor fabrication processes during the late 1980s.[2] Prior to this, most semiconductor companies operated as integrated device manufacturers (IDMs), handling design, production, and sales in-house, which limited accessibility for smaller players and slowed industry innovation.[3] The foundry model's rise was facilitated by advancements in process technology standardization and the increasing demand for custom chips in consumer electronics, computing, and telecommunications, transforming the global supply chain.[4] Key benefits include cost efficiencies through economies of scale, as foundries amortize high fixed costs across multiple clients, and accelerated time-to-market for designers by leveraging specialized expertise in advanced nodes like 3nm and below.[2] Major foundry operators today include TSMC, which holds approximately 70% market share as of Q2 2025;[5] United Microelectronics Corporation (UMC); GlobalFoundries; and Samsung Foundry, with emerging competition from Intel's evolving internal foundry services aimed at external customers.[6] This model has democratized semiconductor production, fueling the explosive growth of fabless giants like Qualcomm, NVIDIA, and Apple, and underpinning the industry's shift toward hyperscale computing and AI applications.[7]Overview
Definition and Core Concepts
The foundry model is a microelectronics business arrangement in the semiconductor industry wherein a fabrication plant, known as a foundry, specializes in manufacturing integrated circuits (ICs) exclusively based on designs submitted by external clients, such as fabless companies, without engaging in its own IC design or branded product sales.[8][1] This model enables specialization by decoupling the capital-intensive process of chip fabrication from the innovation-driven task of design, allowing multiple parties to collaborate efficiently in producing advanced semiconductors.[9] Core to the foundry model are distinctions between pure-play foundries, which dedicate operations solely to contract manufacturing without internal product development (exemplified by TSMC), and hybrid foundries, which offer fabrication services alongside their own chip production for proprietary products (such as Samsung).[1][6] Foundries play a pivotal role in the semiconductor supply chain by bridging fabless design firms—which focus exclusively on IC architecture and intellectual property, like Nvidia—with fabrication and subsequent assembly, testing, and packaging stages, thereby optimizing resource allocation across the ecosystem.[9][10] In contrast, integrated device manufacturers (IDMs), such as Intel, internalize both design and fabrication within a single entity, differing fundamentally from the outsourced approach of the foundry model.[8][9] The concept of the foundry model was pioneered in the 1980s by Morris Chang, who envisioned separating design from manufacturing to enhance efficiency and lower barriers for smaller innovators in an industry dominated by vertically integrated firms.[11] An early precursor to this shared fabrication approach was the MOSIS program, initiated in 1981 to provide multi-project wafer services for academic and research prototyping.[12]Comparison to Integrated Device Manufacturers
The integrated device manufacturer (IDM) model refers to companies that control the full semiconductor value chain, encompassing chip design, fabrication, assembly, testing, and sales of complete products. Prominent examples include Intel and Texas Instruments, which operate their own fabrication plants (fabs) to produce semiconductors exclusively or primarily for internal use, enabling tight integration between design and manufacturing processes.[13][14] In comparison, the foundry model promotes specialization by decoupling design from fabrication, allowing fabless companies—such as those focused on system-on-chip development—to innovate without investing in expensive manufacturing infrastructure, while foundries like TSMC concentrate on advancing process technologies and scaling production. IDMs achieve vertical control that facilitates proprietary optimizations and supply chain security but incur substantial capital expenditures and operational risks, particularly in funding standalone fabs for evolving nodes like 3nm, where technology shifts demand frequent, costly upgrades.[15][16] Foundries leverage economies of scale through multi-client production runs, achieving higher fab utilization rates (often exceeding 80%) and lower per-unit costs compared to IDMs, whose dedicated capacity can lead to underutilization during demand fluctuations. Conversely, IDMs benefit from in-house process tuning tailored to their product roadmaps, though this integration can hinder agility in adopting third-party intellectual property or responding to market-driven node transitions, such as the shift to gate-all-around transistors in sub-3nm processes. The primary clients of foundries are fabless firms, which outsource manufacturing to access cutting-edge capabilities without vertical integration.[17][18] Since the 1990s, the foundry model's growth has reshaped the industry landscape, eroding IDM dominance in high-performance logic and leading-edge technologies such as nodes at 7nm and below, amid rising demand from AI and mobile applications as of 2025. As of 2025, this shift is accelerated by AI and high-performance computing demands, driving foundry revenues to grow 17-20% year-over-year.[19][20]History
Early Developments and MOSIS
In the 1960s and 1970s, the semiconductor industry was dominated by integrated device manufacturers (IDMs) such as Fairchild, Texas Instruments, and Intel, which controlled the entire process from design to fabrication within vertically integrated operations.[21] These companies benefited from relatively lower initial fabrication facility costs—around $4 million in the late 1960s and early 1970s (equivalent to approximately $31 million in 2024 dollars)—but by the 1980s, building a state-of-the-art fab escalated to over $100 million due to increasing technological demands and equipment complexity.[21][22] This escalation severely limited access for smaller firms, universities, and startups, as the capital-intensive nature of fabs favored large incumbents and stifled broader innovation in chip design.[21] Early experiments in shared semiconductor fabrication emerged in the 1970s amid growing interest in very-large-scale integration (VLSI), with efforts at research institutions focusing on advancing wafer processing techniques and prototyping for complex circuits.[23] These initiatives, including collaborative academic projects, began to explore cost-sharing models to support VLSI design tools and foster research beyond IDM constraints.[23] Such experiments highlighted the potential of aggregating multiple designs to amortize fabrication expenses, influencing the development of standardized tools and enabling academic contributions to semiconductor progress.[12] The MOSIS (MOS Implementation System) program, launched in 1981 by the U.S. Defense Advanced Research Projects Agency (DARPA) in collaboration with the University of Southern California's Information Sciences Institute (USC/ISI), formalized these early ideas into a structured multi-project wafer service.[24] MOSIS aggregated designs from multiple users—submitted via ARPANET in CIF format—onto single wafers, contracting with commercial foundries for mask-making, fabrication, and packaging, which dramatically reduced prototyping costs from tens of thousands of dollars per wafer to as low as $258 for a full chip design.[12] This service proved essential for universities, startups, and researchers, supporting over 2,000 VLSI projects by 1983 and accelerating innovation by providing 8-10 week turnaround times.[25] Key drivers for this transition included the rapid rise in integrated circuit complexity during the 1970s, where MOS technology enabled VLSI chips with more than 10,000 transistors—reaching around 68,000 by the end of the decade (as in the Motorola 68000 microprocessor introduced in 1979)—and created a need for neutral, accessible manufacturing to decouple design from production.[26] These developments underscored the limitations of IDM exclusivity and paved the way for neutral shared fabrication, bridging to later commercial models.[12]Emergence of Dedicated Foundries
The emergence of dedicated foundries marked a pivotal shift in the semiconductor industry, building on earlier prototyping efforts like the MOSIS program, which facilitated access to fabrication for researchers and helped validate the outsourced manufacturing model in the 1980s.[24] In 1987, Morris Chang founded Taiwan Semiconductor Manufacturing Company (TSMC) in Hsinchu, Taiwan, establishing the world's first dedicated pure-play foundry with initial capital of $220 million, half provided by the Taiwanese government to support national industrial development.[11] TSMC initially focused on complementary metal-oxide-semiconductor (CMOS) processes, starting with 1-micron technology to produce logic and memory chips for external customers without designing its own products.[27] This model separated fabrication from design, enabling fabless companies to innovate without owning expensive facilities. The 1990s saw accelerated growth in the foundry sector, coinciding with the rise of fabless semiconductor firms, exemplified by Qualcomm's founding in 1985 and a broader boom in the early 1990s driven by demand for specialized chips in computing and communications.[28] United Microelectronics Corporation (UMC), established in Taiwan in 1980 as an integrated device manufacturer, transitioned to a pure-play foundry model in 1995 by spinning off its design operations and focusing exclusively on contract manufacturing.[29] Similarly, Chartered Semiconductor Manufacturing was founded in Singapore in 1987 as a joint venture backed by the government to diversify the economy, rapidly expanding capacity to serve global clients in logic and analog processes.[30] Key milestones in the 2000s further solidified the foundry model's global footprint. Samsung Electronics entered the dedicated foundry business in 2004, leveraging its existing fabrication expertise to offer services beyond its internal needs, targeting mobile and consumer electronics markets.[31] In 2009, Advanced Micro Devices (AMD) spun off its manufacturing operations to create GlobalFoundries, backed by investment from Abu Dhabi, allowing AMD to adopt a fabless strategy while providing a U.S.-based foundry option for various process nodes.[32] China's Semiconductor Manufacturing International Corporation (SMIC) was established in 2000 in Shanghai with foreign expertise and state support to advance the country's semiconductor capabilities.[33] This period witnessed a geographic shift from U.S. and Taiwan dominance to broader Asian expansion, fueled by lower costs, government incentives, and proximity to design hubs in the region.[34] By 2010, dedicated foundries had grown significantly, capturing an increasing share of global logic integrated circuit production and reflecting their role in enabling scalable manufacturing for diverse applications.[35]Business Model and Operations
Manufacturing Processes
The manufacturing processes in semiconductor foundries encompass a highly precise sequence of steps to transform raw silicon into functional integrated circuits, divided primarily into front-end and back-end phases. The front-end process, also known as wafer fabrication, begins with a polished silicon wafer—typically 300 mm in diameter—and involves creating the intricate transistor structures and interconnects through repeated cycles of material deposition, patterning, and modification. Key steps include deposition, where thin films of insulating, conducting, or semiconducting materials are layered onto the wafer using techniques like chemical vapor deposition (CVD); photoresist coating, applying a light-sensitive polymer to prepare for patterning; lithography, which projects circuit designs onto the resist using ultraviolet or extreme ultraviolet (EUV) light through a photomask; etching, selectively removing material to define features via wet chemical or dry plasma methods; and ion implantation (doping), bombarding the wafer with ions to alter electrical properties and form transistors. These steps are iterated dozens of times to build multilayer structures, with each wafer potentially yielding thousands of dies after rigorous inline inspections for defects.[36][37] The back-end process follows wafer fabrication and focuses on transforming the completed wafers into usable chips, including dicing the wafer into individual dies, assembly onto substrates with wire bonding or flip-chip methods for electrical connections, testing for functionality and performance under various conditions, and final packaging to protect the die while enabling integration into electronic systems, such as encasing it with a heat spreader and lid. This phase ensures reliability and prepares chips for shipment, often taking weeks to complete due to the need for high-precision handling to avoid contamination or damage.[36][37] Foundries employ advanced technologies to push the limits of transistor density and performance, particularly through shrinking process nodes to 3 nm and 2 nm by 2025, which rely on EUV lithography systems operating at a 13.5 nm wavelength for resolutions as fine as 8 nm in high-numerical-aperture (NA) tools. These EUV systems, such as ASML's NXE for 3 nm high-volume manufacturing and EXE platforms slated for 2 nm logic and memory nodes starting in 2025–2026, enable the precise patterning required for complex architectures like gate-all-around transistors. Yield optimization is critical throughout, with foundries targeting over 90% good die yields for mature nodes (e.g., 28 nm and above) through process control, defect detection via AI-driven analytics, and statistical modeling to minimize variability from sources like contamination or misalignment.[38][39] To accommodate diverse client needs, foundries implement adaptations like multi-project wafer (MPW) services, where multiple customer designs are aggregated onto a single wafer or reticle set, sharing fabrication costs and reducing prototyping expenses by up to 90% for small-volume runs. This approach is particularly valuable for startups and research, allowing rapid validation without full-wafer commitment. Complementing this, process design kits (PDKs) provided by foundries serve as comprehensive libraries of device models, layout rules, and verification tools tailored to their specific fabrication processes, ensuring client designs are compatible and manufacturable while integrating with electronic design automation (EDA) software for simulation and optimization.[40][41] The equipment ecosystem underpinning these processes features specialized tools from key suppliers, with ASML holding a monopoly on EUV lithography scanners essential for advanced nodes and Applied Materials providing dominant solutions for deposition, etching, and other thin-film processes. All operations occur in ultra-clean environments adhering to Class 1 cleanroom standards (ISO 3 equivalent), which limit airborne particles to no more than 1,000 per cubic meter at ≥0.1 µm size, achieved through high-efficiency particulate air (HEPA) filtration, positive pressure, and stringent gowning protocols to prevent defects from even a single dust particle.[42][43]Design Support and Customer Services
Foundries provide essential design support through Process Design Kits (PDKs), which are comprehensive sets of files and models that describe a specific manufacturing process for integration with electronic design automation (EDA) tools from vendors such as Cadence and Synopsys.[44] These PDKs include device models, design rules, and layout parameters, enabling fabless companies to simulate and verify chip designs accurately before fabrication. For instance, Tower Semiconductor's PDKs support major EDA flows and incorporate analytical techniques for optimization and reliability analysis.[45] Additionally, foundries offer reference designs for intellectual property (IP) cores, such as analog and mixed-signal blocks, to accelerate integration and reduce time-to-market. Beyond PDKs, foundries deliver a range of verification and prototyping services, including design rule checks (DRC), layout-versus-schematic (LVS) verification, and circuit simulations tailored to their processes. These services ensure compliance with manufacturing constraints and help identify potential issues early in the design cycle. Prototyping is facilitated through multi-project wafer (MPW) runs, where multiple customer designs share a single wafer to lower costs—often reducing expenses by up to 90% compared to full-wafer production. TSMC's CyberShuttle program, operational since 1998, has supported thousands of devices via MPW shuttles, providing rapid turnaround for initial silicon validation.[46] Similarly, GlobalFoundries' GlobalShuttle aggregates projects for efficient prototyping of differentiated designs.[40] Foundries foster fabless ecosystems through collaborative platforms that integrate design tools, IP, and supply chain partners. TSMC's Open Innovation Platform (OIP), for example, encompasses EDA alliances with over 20 vendors, IP partnerships, and value chain support to minimize design barriers and enable advanced node adoption.[47] Samsung's SAFE program similarly promotes SoC innovation by certifying components and offering joint design assistance. For capacity allocation, foundries secure volume production via long-term agreements (LTAs), which guarantee dedicated fab capacity in exchange for committed volumes; GlobalFoundries reported over 40 such LTAs by 2023, stabilizing supply for customers. Rapid prototyping options, like MPW shuttles, cater to startups and early-stage projects, contrasting with LTAs focused on high-volume scaling. Customization services distinguish foundries by offering specialty processes alongside leading-edge logic nodes. For radio frequency (RF) and power applications, foundries provide tailored process options, such as Tower Semiconductor's modular platforms on 200mm or 300mm wafers, which integrate high-voltage and RF capabilities for analog, power management, and sensor ICs. These specialty flows prioritize performance in niche markets like automotive and industrial, while leading-edge logic emphasizes scaling for compute-intensive designs, allowing customers to select processes aligned with application needs.Major Companies and Market Dynamics
Leading Foundry Companies
Taiwan Semiconductor Manufacturing Company (TSMC), founded in 1987 and headquartered in Taiwan, stands as the world's leading dedicated semiconductor foundry, pioneering the pure-play model that separates design from manufacturing.[48] By 2025, TSMC has solidified its dominance in advanced process nodes, with its 2nm (N2) technology entering risk production and becoming available to customers in the second half of the year, enabling high-density, power-efficient chips for next-generation applications. Major clients such as Apple and Nvidia rely on TSMC for cutting-edge fabrication, particularly for AI accelerators and graphics processing units, with approximately 60% of its market focus dedicated to AI and high-performance computing (HPC) segments that drive demand for these nodes.[49][50] Samsung Foundry, the external manufacturing arm of South Korea's Samsung Electronics, operates a hybrid model that balances internal production for Samsung's devices with services for third-party clients, a strategy emphasized since the early 2000s when it expanded dedicated foundry capabilities around 2004.[51] Specializing in integrated logic with memory technologies, Samsung has advanced its offerings through innovations like the 3nm gate-all-around (GAA) process, which enhances transistor density and efficiency for mobile, automotive, and server applications.[51] This hybrid approach allows Samsung to leverage synergies between its memory division—dominant in DRAM and NAND—and logic foundry services, positioning it as a key alternative to pure-play foundries for clients seeking embedded memory solutions. Among other established players, GlobalFoundries, founded in 2009 through a partnership between AMD and Abu Dhabi's Mubadala Investment Company and based in the United States, focuses on specialty and mature process nodes (above 12nm), catering to automotive, IoT, and RF applications where reliability and customization outweigh cutting-edge scaling.[52][53] United Microelectronics Corporation (UMC), established in 1980 as Taiwan's first semiconductor firm, emphasizes cost-competitive production in mature and specialty technologies (28nm and above), serving consumer electronics and power management markets with efficient, high-volume fabrication.[54][55] Semiconductor Manufacturing International Corporation (SMIC), China's largest foundry founded in 2000, has made notable progress in advanced nodes despite U.S. export sanctions, achieving 7nm-class production for logic chips used in smartphones and AI edge devices.[56][57][58] Emerging contenders include Intel Foundry Services (IFS), launched in 2021 as part of Intel's strategic pivot under its IDM 2.0 model to offer external manufacturing alongside internal needs, targeting foundry customers with Intel's 18A (1.8nm) and subsequent nodes for AI, cloud, and edge computing.[59] In Japan, Rapidus Corporation, established in 2022 with government and industry backing, is developing a 2nm-class foundry to revitalize domestic advanced manufacturing, achieving initial prototyping of GAA transistors in 2025 and aiming for mass production by 2027 through collaborations with IBM and ASML.[3][60]Market Share and Revenue Trends
The global semiconductor foundry market demonstrated robust growth in 2025, with quarterly revenues reaching a record $41.7 billion in Q2, marking a 14.6% increase from the previous quarter.[5] This surge was primarily fueled by surging demand for advanced nodes driven by artificial intelligence (AI) applications and 5G infrastructure expansions.[61] In Q3 2025, TSMC reported a 6% quarter-over-quarter revenue increase to $33.1 billion, highlighting continued momentum from AI demand.[62] Annual revenues for 2025 are projected to reach approximately $175 billion, with the market expected to surpass $200 billion by 2030 at a compound annual growth rate (CAGR) of around 5-7%, continuing to be propelled by AI, high-performance computing, and emerging technologies like electric vehicles (EVs).[63][64] Market share in 2025 remained heavily concentrated among a few leading players, with Taiwan Semiconductor Manufacturing Company (TSMC) dominating at approximately 70-71% of the pure-play foundry segment in Q2.[5][65] Samsung Electronics held about 7-8%, followed by United Microelectronics Corporation (UMC) at around 6%, Semiconductor Manufacturing International Corporation (SMIC) at 5%, and GlobalFoundries at 5%, while smaller firms like Tower Semiconductor and Vanguard International Semiconductor (VIS) collectively accounted for less than 5%.[65][66]| Company | Market Share (Q2 2025) |
|---|---|
| TSMC | 70-71% |
| Samsung | 7-8% |
| UMC | ~6% |
| SMIC | 5% |
| GlobalFoundries | 5% |
| Others | <5% |