Samsung Foundry 3nm Woes: Impact on Future Plans

Updated 6 November 2026 · Standard formula, shown below

Samsung Foundry’s struggles with its 3nm gate-all-around (GAA) process have raised real questions about its ability to win major customers and keep pace with TSMC. Yes, the yield and performance issues at 3nm have forced Samsung to adjust its roadmap, delay volume production for key clients, and rethink how it pitches its next-generation 2nm technology. This article explains what went wrong, how it affects Samsung’s foundry business, and what it means for the broader chip industry.

Background: How Samsung Got to 3nm

Samsung was the first foundry to ship a 3nm chip using gate-all-around (GAA) transistors, a structural change from the FinFET design used by TSMC and Intel at that node. GAA wraps the gate around all four sides of the channel, reducing current leakage and improving drive current. Samsung introduced this as its 3GAE (gate-all-around early) process in mid-2022, with a more refined 3GAP (gate-all-around plus) version planned later.

The early move was a calculated risk: Samsung hoped to leapfrog TSMC, which planned to stick with FinFET at 3nm and only adopt GAA at 2nm. By being first, Samsung aimed to attract high-performance computing and mobile customers who wanted the power and area benefits of GAA. The first known product on 3GAE was a cryptocurrency mining chip from a Chinese design firm, not a flagship smartphone processor, which hinted early on that the process was not yet ready for mainstream high-volume parts.

The Problems: Low Yields and Performance Shortfalls

The core issue at Samsung’s 3nm has been low yield, meaning the percentage of working chips per wafer is too low to make large orders economical. Industry reports, though unconfirmed by Samsung, have cited yields in the 10-20% range for the early 3GAE process, far below the 60-70% that a mature process typically needs for profitable volume production.

Beyond yield, the 3nm GAA transistors did not deliver the expected performance leap. Samsung’s own Exynos 2500 processor, initially planned for its Galaxy S25 series, reportedly missed performance and thermal targets, forcing Samsung to use Qualcomm’s Snapdragon 8 Gen 3 for the global release. The Exynos 2500 was later used in some regional models of the Galaxy S25, but with a lower clock speed and fewer active cores, indicating the 3nm chip did not meet original design goals.

These problems are not just about one chip. They create a cascading effect: low yields mean higher costs per chip, which deters large customers like Qualcomm, NVIDIA, and AMD from placing orders. Without volume, Samsung cannot amortize the development cost or improve the process through manufacturing learning, creating a vicious cycle.

Impact on Samsung’s Foundry Roadmap

Samsung’s public roadmap had planned a 2nm GAA process (SF2) for mass production in 2025, with a more advanced SF2P in 2026. However, the 3nm struggles have forced delays and a strategic pivot. In late 2024, Samsung reorganized its foundry division, bringing in a new leader and reportedly shifting resources to focus on the 2nm node while scaling back on 3nm improvements.

The company now aims to make SF2 its first truly high-volume GAA node, skipping further refinement of 3GAP for most external customers. This is a significant change: originally, 3GAP was supposed to be the process that would attract major orders. Instead, Samsung is betting that a more mature 2nm process, with a longer development time, will finally deliver the yields and performance that customers expect. The risk is that TSMC, with its N2 process also scheduled for 2025, will have a similar GAA offering with a proven track record, making it harder for Samsung to catch up.

How It Affects Customers and the Market

The immediate impact of Samsung’s 3nm issues is that most major chip designers have not committed to the node. Qualcomm, Samsung’s largest foundry customer, has used TSMC for its flagship Snapdragon chips since the 4nm generation. Samsung’s own Exynos processors are now a secondary player, used mainly in lower-volume regional models. NVIDIA and AMD, both of whom use TSMC for their leading-edge GPUs and CPUs, have shown no public interest in Samsung 3nm.

For the broader market, this means TSMC’s dominance at the leading edge is even more pronounced. TSMC holds roughly 90% of the global market for chips at 5nm and below, and Samsung’s 3nm failure only strengthens that position. This concentration creates risks for the industry: if TSMC’s factories are disrupted by a natural disaster or geopolitical event, there is no viable second source for the most advanced chips.

On the positive side, Samsung’s struggles have pushed it to be more transparent about its process technology and to work more closely with customers on design co-optimization. The company has also increased investment in its US fab in Taylor, Texas, which is expected to produce 2nm chips, aiming to offer a geographically diverse alternative to TSMC’s Taiwan-based fabs.

Comparison: Samsung 3nm vs. TSMC 3nm vs. Intel 18A

To understand the stakes, compare the three leading-edge processes as of early 2025:

Samsung 3GAE/3GAP: GAA transistors, but low yields and performance issues; used mainly in Samsung’s own Exynos and a few small orders. No major external customers announced.
TSMC N3 (3nm): FinFET transistors, in high-volume production since late 2022. Used by Apple (A17 Pro, M3), NVIDIA (RTX 40 series), AMD (Ryzen 7000), and Qualcomm (Snapdragon 8 Gen 3). Yields are mature and costs are competitive.
Intel 18A (1.8nm): Intel’s first GAA process (called RibbonFET), scheduled for production in 2025. Intel plans to use it for its own products and offer it as a foundry service. It is unproven in volume, but Intel has secured Microsoft as a customer.

This comparison shows that Samsung’s 3nm is not just behind TSMC; it is also behind Intel’s upcoming 18A, which promises GAA with a more conventional development timeline. Samsung’s only advantage is that it has already shipped a GAA product, but that product is not competitive in performance or cost.

Common Questions and Misconceptions

Misconception: “Samsung 3nm is a failure, so Samsung will exit foundry.” That is unlikely. Samsung is one of only three companies (with TSMC and Intel) capable of leading-edge logic manufacturing. Exiting would cede a strategic technology to competitors and undermine Samsung’s own semiconductor division, which is its most profitable business. Instead, Samsung is doubling down on 2nm, hoping to correct the mistakes of 3nm.

Misconception: “Lower yield only affects price, not performance.” Yield and performance are linked. A process with low yield often has high variability, meaning some chips run faster than others. To sell a chip with a guaranteed clock speed, the manufacturer must either test each chip and bin them (selling slower ones at lower prices) or lower the overall clock target. Samsung did both for the Exynos 2500, which is why the chip in the Galaxy S25 has a lower maximum frequency than originally planned.

Question: “Why didn’t Samsung just copy TSMC’s FinFET approach?” Because Samsung had already committed to GAA as a differentiator. Switching back to FinFET at 3nm would have meant redesigning the process from scratch, adding years of delay. The company chose to push through GAA, hoping the long-term benefits would outweigh the short-term pain. That bet has not paid off yet.

Future Outlook: Can Samsung Recover at 2nm?

Samsung’s future plans hinge on the success of its SF2 process. The company has stated that SF2 will offer a 25% performance improvement or 45% power reduction over its 3nm process, with a 17% area reduction. These are aggressive targets, but they are similar to what TSMC promises for its N2 node. The key difference is execution.

Samsung has taken several steps to improve its chances:

Appointed a new foundry head in late 2024, signaling a fresh approach.
Created a dedicated “foundry business control tower” to improve communication between design and manufacturing teams.
Invested heavily in the Taylor, Texas fab, which is expected to produce 2nm chips for US customers, offering a geopolitical hedge.
Reportedly secured a pre-order from a major AI chip company for 2nm, though the identity has not been confirmed.

However, the same problems that plagued 3nm could recur at 2nm. GAA is still a relatively new technology, and scaling to 2nm brings additional challenges like backside power delivery (which Samsung plans to introduce at SF2P). TSMC will also be using GAA at N2, but with the advantage of a proven track record in high-volume manufacturing. Samsung must not only hit its performance targets but also achieve yields above 60% within the first year of production, something it failed to do at 3nm.

For customers, the message is to remain cautious. Samsung’s 2nm may be a viable second source in 2026-2027, but it is unlikely to displace TSMC as the primary foundry for most leading-edge designs. The most realistic outcome is that Samsung becomes a niche player for specific applications (e.g., automotive, AI accelerators) where its GAA technology offers unique advantages, while TSMC continues to dominate the mainstream.

Frequently Asked Questions

Q: Is Samsung’s 3nm process actually used in any phone? Yes, the Exynos 2500, used in the Galaxy S25 series sold in some regions, is manufactured on Samsung’s 3nm GAA process. However, it has a lower clock speed and fewer active cores than originally planned, reflecting the process’s performance limitations.

Q: Why is yield so critical in chip manufacturing? Yield is the percentage of functional chips per wafer. A low yield means fewer sellable chips per wafer, driving up the cost per chip. For example, if a wafer costs $10,000 and yields 100 chips, each costs $100. If yield drops to 50 chips, the cost doubles to $200. For a flagship phone processor, that extra cost either eats into profit or raises the phone’s price.

Q: What is GAA and why is it better than FinFET? Gate-all-around (GAA) wraps the transistor’s gate around all four sides of the channel, providing better electrostatic control than FinFET, which wraps the gate around three sides. This allows for lower leakage current and potentially higher performance at smaller sizes. However, GAA is more complex to manufacture, which is why Samsung’s early adoption caused problems.

Q: Will Samsung stop making its own Exynos chips? No, but it may reduce their use. Samsung has used Exynos in fewer models each year due to performance gaps. However, Exynos is still important for Samsung’s internal supply chain and for negotiating with Qualcomm. The company is reportedly developing a new Exynos chip on 2nm for its 2026 flagship, but it may again be limited to certain regions.

Q: How does this affect consumers? In the short term, it means fewer choices for high-end Android phones, as most will use Qualcomm chips made by TSMC. In the long term, if Samsung fails at 2nm, the foundry market could become a monopoly, potentially leading to higher chip prices and less innovation. If Samsung succeeds, consumers could see more competitive pricing and performance.

Method: the standard formula for this calculation, rounded as each result states. See how the calculators work or report an error.