TSMC, the Taiwan Semiconductor Manufacturing Company, is the world’s largest pure-play semiconductor foundry: a manufacturer that fabricates integrated circuits designed by other companies and sells no chips under its own brand. In semiconductors and computer hardware, TSMC is the reference point for leading-edge process technology, producing processors, graphics chips and accelerators for many of the best-known chip designers.
Founded in 1987 in Hsinchu, Taiwan, by Morris Chang, the company turned the foundry business model into the backbone of the modern chip industry. Its scale, manufacturing yields and advanced packaging capacity give it a central and strategically sensitive role in global electronics supply chains.
How the TSMC foundry model works
The TSMC model separates chip design from chip manufacturing. A customer, typically a fabless company, designs a circuit using electronic design automation tools and TSMC’s process design kit, the set of device models, design rules and verified library cells for a given node. TSMC then manufactures the wafers, and the customer or a partner handles testing, packaging and sales.
Because TSMC does not compete with its customers in the chip market, it can serve rivals simultaneously while protecting each design. This neutrality principle is a core commercial asset. Revenue is concentrated in high-performance computing and smartphones, with automotive, networking and consumer electronics forming smaller segments.
The following table contrasts the main ways chips reach the market and where TSMC fits.
| Model | Designs chips | Manufactures chips | Typical examples |
|---|---|---|---|
| Pure-play foundry | No | Yes, for customers | TSMC, GlobalFoundries, UMC |
| Fabless company | Yes | No, outsources to foundries | Nvidia, AMD, Qualcomm |
| Integrated device manufacturer | Yes | Yes, mainly its own designs | Texas Instruments, Micron |
| Hybrid IDM with foundry arm | Yes | Yes, own and external designs | Samsung, Intel |
TSMC process nodes and technology roadmap
TSMC sells manufacturing capacity in process nodes, named generations such as N7, N5, N3 and N2. A node name is a marketing label rather than a physical dimension: no feature on an N3 chip measures 3 nanometers. What matters is the combination of transistor density, performance and power efficiency relative to the previous generation.
Key transitions include the adoption of EUV lithography in production at N7+ and N5, which reduced the number of multi-patterning steps, and the move from FinFET transistors to gate-all-around nanosheet transistors at N2. Each family also has variants tuned for performance, cost or low power.
| Node family | Transistor type | Approximate volume start | Notable traits |
|---|---|---|---|
| N7 | FinFET | 2018 | First volume 7 nm class; N7+ added EUV layers |
| N5 | FinFET | 2020 | Extensive EUV use; large density gain over N7 |
| N3 | FinFET | Late 2022 | Highest-density FinFET node; several variants |
| N2 | Nanosheet gate-all-around | 2025 | First TSMC node with GAA transistors |
| Mature nodes | Planar and FinFET | Ongoing | 28 nm to 0.18 um for analog, MCU and automotive |
Leading nodes attract headlines, but mature nodes remain a large share of wafer volume. Power management chips, microcontrollers, image sensors and radio-frequency parts often do not benefit from extreme scaling and are cheaper on older, fully depreciated lines.
Advanced packaging and chiplets at TSMC
As transistor scaling becomes slower and costlier, advanced packaging has become a second pillar of TSMC’s offering. Instead of one large monolithic die, designers combine several dies, or chiplets, with high-bandwidth memory in a single package. TSMC groups these technologies under the 3DFabric name.
| Technology | Approach | Typical use |
|---|---|---|
| CoWoS | Dies placed side by side on a silicon interposer | AI accelerators with HBM stacks |
| InFO | Integrated fan-out wafer-level packaging | Smartphone application processors |
| SoIC | Direct die-to-die stacking with hybrid bonding | Vertical 3D integration of logic and cache |
Demand for AI hardware turned CoWoS capacity into a bottleneck after 2023, at times limiting accelerator shipments more than wafer supply. This illustrates a broader point: leadership now depends on packaging as well as lithography.
TSMC compared with competing foundries
TSMC holds well over half of global foundry revenue and a much higher share at the most advanced nodes, although exact figures vary by quarter and by how the market is defined. Its competitive advantage rests on high yields early in a node’s life, predictable roadmaps and a broad ecosystem of design partners.
| Company | Leading-edge position | Main strengths | Main challenges |
|---|---|---|---|
| TSMC | Volume leader at 3 nm class and N2 | Yields, ecosystem, packaging | Geographic concentration, cost |
| Samsung Foundry | Early GAA adopter at 3 nm class | Memory and logic under one roof | Yield consistency, customer trust |
| Intel Foundry | Pursuing 18A with backside power | US manufacturing, packaging know-how | Building an external customer base |
| SMIC | Constrained by export controls | Domestic Chinese demand | Limited access to EUV tools |
Strategic risks and common misconceptions about TSMC
TSMC’s importance creates concentration risk for the whole electronics industry. Most of its advanced capacity is in Taiwan, which exposes customers to geopolitical tension, earthquakes, water shortages and power supply constraints. In response, TSMC has been building overseas fabs in Arizona, in Kumamoto, Japan, and in Dresden, Germany, though leading-edge production remains centered in Taiwan.
- TSMC does not design the chips it makes: architecture and intellectual property belong to customers.
- Node names are not comparable across companies: an Intel or Samsung node of similar name may differ in density.
- Overseas fabs are not instant substitutes: they typically trail the newest Taiwan nodes and cost more to operate.
- Wafer prices rise at each node: leading-edge wafers can cost several times more than mature-node wafers.
- Equipment comes from suppliers: TSMC depends on ASML for EUV scanners and on many materials vendors.
A non-technical view
Think of TSMC as an extremely advanced print shop for computer chips. Companies such as phone makers and graphics card brands draw up the blueprint of a chip, then send it to TSMC, which turns the blueprint into millions of physical chips. TSMC builds the chips but does not sell its own.
Because it is very good at making the smallest and fastest chips, much of the world relies on it. That is why a problem at TSMC can affect phones, laptops and cars everywhere.
A concrete example
A fabless startup designs an AI inference chip for data centers. Its engineers use TSMC’s process design kit for an N5-class node, verify timing and power, and submit the final layout through a multi-project wafer shuttle, which shares one wafer run among several customers to cut prototype costs.
The first silicon returns after roughly three to four months. After two design fixes, the startup orders production wafers and books CoWoS packaging to place its compute die beside two HBM stacks. The company ships a competitive accelerator without owning a single fab, while its main constraint becomes securing enough packaging capacity rather than designing the chip itself.