TSMC vs Samsung Foundry: Manufacturing Compared

TSMC and Samsung Foundry are the world’s two most advanced semiconductor manufacturers, competing directly in the leading-edge chip fabrication market. TSMC leads in market share, process maturity, and yield, while Samsung offers competitive specs and more aggressive adoption of new transistor architectures. This article compares their manufacturing businesses across technology, capacity, customers, and strategy.

Background: How the Two Foundries Got Here

TSMC (Taiwan Semiconductor Manufacturing Company) was founded in 1987 as the world’s first dedicated semiconductor foundry, meaning it only manufactures chips designed by other companies. Samsung Foundry, part of Samsung Electronics, began as an internal chipmaker and opened its foundry services to external customers in 2005. TSMC’s pure-play model allowed it to focus entirely on manufacturing, while Samsung also designs its own chips (Exynos) and makes memory, giving it a different set of priorities and pressures.

Both companies have invested tens of billions of dollars annually in R&D and new fabs. TSMC’s largest customer is Apple, which relies on TSMC for its A-series and M-series processors. Samsung’s foundry clients include Qualcomm, NVIDIA, and AMD, though some of these also use TSMC for different products. The two firms are the only ones currently producing chips at the 3nm class, with Intel trailing behind.

Process Technology: Who’s Ahead in Nanometers?

The term “nanometer” (nm) in chip marketing no longer refers to a physical dimension; it’s a generational label. TSMC’s N3 (3nm) process entered mass production in late 2022, using FinFET transistors. Samsung’s 3nm GAA (Gate-All-Around) process, called SF3, started production in 2022 as well, but with a different transistor design. GAA wraps the gate around all four sides of the channel, reducing current leakage and improving performance compared to FinFET, which only covers three sides.

• TSMC N3: FinFET, mature yield, used in Apple’s A17 Pro and M3 chips.
• Samsung SF3: GAA, higher theoretical performance, but lower yield and fewer known customers.
• Next steps: TSMC plans N2 (2nm) with GAA in 2025; Samsung targets SF2 with GAA in 2025.

In practice, TSMC’s 3nm FinFET has outperformed Samsung’s 3nm GAA in early benchmarks, largely due to better yield and design infrastructure. Samsung’s GAA is a technological leap, but it has struggled to attract large orders, partly because customers are wary of unproven yield and design rules.

Manufacturing Capacity and Locations

TSMC operates most of its advanced fabs in Taiwan, with new facilities in the US (Arizona) and Japan (Kumamoto). Its capacity is concentrated in the Hsinchu and Tainan science parks. Samsung has fabs in South Korea (Giheung, Hwaseong, Pyeongtaek) and a new plant in Taylor, Texas. Both are expanding to meet demand for AI chips and automotive semiconductors.

TSMC’s total capacity is roughly 1.5 times that of Samsung Foundry for leading-edge nodes, according to industry estimates. TSMC also has a larger share of mature-node capacity, which is used for analog chips, power management, and microcontrollers. Samsung’s capacity is more vertically integrated, with memory and logic on the same campus, allowing faster turnaround for some products.

Customers and Market Share

TSMC holds about 60% of the global foundry market, while Samsung has around 11-12% (2023 data). TSMC’s customer list includes Apple, AMD, NVIDIA, Qualcomm, MediaTek, and Broadcom. Samsung’s notable foundry customers include Qualcomm (for some Snapdragon models), IBM, and a few AI startups like Groq and Cerebras.

One major difference is that Samsung competes with its own customers in the system LSI business (Exynos processors). This creates a conflict of interest: a company like Qualcomm may hesitate to give Samsung its most advanced designs if Samsung could use them for its own chips. TSMC, being a pure-play foundry, has no such conflict, which is a key reason for its dominance.

Yield, Performance, and Power Efficiency

Yield is the percentage of functional chips on a wafer. Higher yield means lower cost per chip. TSMC has consistently achieved higher yields at the same node generation than Samsung. For example, early reports suggested TSMC’s 5nm yield was above 80%, while Samsung’s 5nm was around 70-75%. At 3nm, Samsung’s GAA yield reportedly started below 50%, improving later, while TSMC’s 3nm FinFET yield was above 60% from the start.

Performance and power efficiency are measured by benchmarks like Geekbench and power consumption tests. TSMC’s N3 chips generally show better power efficiency than Samsung’s SF3 at the same clock speed. However, Samsung’s GAA has the potential to surpass FinFET once mature, as it allows more aggressive voltage scaling. For now, TSMC remains the safer choice for high-performance computing.

Costs and Pricing

Wafer prices are not publicly disclosed, but industry analysts estimate that TSMC’s 3nm wafers cost around $20,000 each, while Samsung’s 3nm wafers are slightly cheaper, around $18,000, to attract customers. However, lower yield can offset the price advantage, making the effective cost per good chip higher for Samsung. Design costs also matter: TSMC’s PDK (Process Design Kit) and IP ecosystem are more mature, reducing design time and risk.

For a typical SoC like a smartphone processor, the die cost at 3nm is roughly $50-80 per chip, depending on die size and yield. TSMC’s higher yield often makes its total cost competitive despite higher wafer prices. Customers also factor in long-term supply security; TSMC has a track record of ramping multiple nodes on time, while Samsung has missed some deadlines.

Common Problems and Misconceptions

A common misconception is that a smaller node number always means better performance. In reality, a well-optimized 5nm chip can beat a poorly implemented 3nm chip. Another issue is “node naming inflation” – Samsung’s 8nm was actually similar to TSMC’s 10nm in density, so comparing numbers across companies is misleading.

• Thermal issues: Samsung’s 5nm and 4nm processes had higher power density, leading to overheating in some Snapdragon 888 and 8 Gen 1 chips.
• Yield problems: Samsung’s GAA at 3nm has been slow to ramp, causing some customers to move to TSMC.
• Supply chain risk: TSMC’s concentration in Taiwan is a geopolitical risk that Samsung could exploit, but Samsung has its own risks due to US-China tensions.

How to Choose Between Them

For a chip designer, choosing a foundry depends on several factors:

1. Performance needs: If you need the absolute best power efficiency and can tolerate higher cost, TSMC is the safer bet.
2. Cost sensitivity: Samsung may offer lower wafer prices, but factor in yield and design support costs.
3. Capacity availability: TSMC is often booked out months in advance; Samsung may have more open capacity.
4. IP and design tools: TSMC has a larger ecosystem of third-party IP blocks, reducing design time.
5. Political risk: Some companies prefer Samsung to diversify away from Taiwan, even if it means slightly lower performance.

In practice, most high-volume AI and smartphone chips go to TSMC, while Samsung wins some niche orders and its own Exynos chips. The choice is rarely purely technical; it involves supply chain strategy and long-term partnerships.

Frequently Asked Questions

Is Samsung foundry actually behind TSMC?

Yes, in terms of market share, yield, and customer trust, Samsung is significantly behind. TSMC has a ~60% share vs Samsung’s ~12%, and its processes are generally more mature. However, Samsung’s GAA technology is more advanced in design, so the gap could narrow if Samsung improves yield.

Why does Apple only use TSMC?

Apple uses TSMC because of its superior yield, performance, and reliable supply. TSMC is a pure-play foundry, so it doesn’t compete with Apple in the device market. Samsung, by contrast, makes its own phones and chips, creating a conflict of interest. Apple also co-develops processes with TSMC, giving it early access.

What does GAA mean and why is it important?

GAA stands for Gate-All-Around, a transistor design where the gate wraps around the channel on all four sides, reducing leakage and improving control. It allows further scaling beyond FinFET, which TSMC will adopt at 2nm. Samsung adopted GAA at 3nm, but the early yield issues show that design innovation alone isn’t enough.

Can Samsung catch up with TSMC?

It is possible but difficult. Samsung would need to improve yield, win back major customers, and prove its GAA process at scale. It also faces the conflict of interest issue, which is structural. Some analysts believe Samsung may leapfrog TSMC at 2nm, but most expect TSMC to maintain its lead through 2026.

Are there other foundries besides TSMC and Samsung?

Yes, Intel Foundry Services (IFS) is trying to enter the market, and there are smaller players like GlobalFoundries (which stopped advanced nodes) and UMC. However, for leading-edge chips below 7nm, only TSMC and Samsung are viable for high-volume production. Intel’s 18A process is promising but unproven.

How does the US-CHina conflict affect these foundries?

Export controls limit what advanced chips can be sold to China, affecting both TSMC and Samsung. Both are building fabs in the US to hedge against geopolitical risk. TSMC’s Arizona fab is set to produce 4nm chips in 2024, while Samsung’s Taylor fab will make 4nm and 2nm chips. China’s SMIC is far behind, but it is investing heavily in mature nodes.

Related

• Stephen King TV Adaptations Ranked by Book Accuracy
• WordPress Disk Space: A Practical Sizing Guide
• WordPress Fix: Could Not Insert Attachment into the Database