The 2nm Chip Race: TSMC, Samsung, and Intel Go All In
The race to 2nm process technology is the most expensive and technically challenging competition in semiconductor history. Three companies — TSMC, Samsung, and Intel — are each betting tens of billions of dollars on their ability to manufacture transistors at this scale. Here's where each stands, what "2nm" actually means in 2026, and why this node is particularly significant.
What 2nm Actually Means (And Doesn't Mean)
First, let's clear up a persistent misconception: the "2nm" label doesn't mean any feature on the chip is actually 2 nanometers wide. Modern node names are marketing designations that roughly correspond to transistor density and performance improvements relative to previous generations.
At TSMC's N2 node, the minimum metal pitch is approximately 28nm, and the gate pitch (the center-to-center distance between transistors) is about 48nm. The actual gate length — the shortest dimension of the transistor channel — is somewhere around 5-12nm. The "2nm" name indicates that this node delivers density and performance roughly equivalent to what a theoretical 2nm scaling of traditional dimensions would provide.
What makes 2nm genuinely special isn't the name — it's the transistor architecture. All three manufacturers are switching from FinFET to gate-all-around (GAA) nanosheet transistors at this node (though Samsung introduced GAA at 3nm, getting a head start with mixed results).
TSMC N2: The Industry Benchmark
TSMC's N2 process entered risk production in late 2024, with volume production expected in the second half of 2025. Early customer chips should appear in products by late 2025 or early 2026.
Key specifications:
Related reading: Power Semiconductor Devices: SiC MOSFETs, GaN HEMTs, and Wid.
- GAA nanosheet transistors with adjustable sheet width for performance/power tuning
- 10-15% speed improvement at the same power versus N3E
- 25-30% power reduction at the same speed versus N3E
- Transistor density approximately 300 million transistors per mm² (up from ~250M at N3)
- Backside power delivery planned for N2P variant (not the initial N2)
TSMC's approach is characteristically conservative. They waited until GAA nanosheet technology was thoroughly validated before deploying it, while Samsung jumped to GAA at 3nm. The tradeoff: Samsung had GAA experience earlier, but TSMC's implementation at 2nm is expected to have better yield and performance from the start.
Apple is widely reported to be the first N2 customer, likely for the A19 or M5 chip generation. Other early adopters include AMD, NVIDIA, and Qualcomm.
Samsung 2nm (SF2): Redemption Arc?
Samsung's foundry division has had a rough stretch. Their 3nm GAA process (SF3) launched with yields that disappointed major customers, and Qualcomm famously split its Snapdragon orders between Samsung and TSMC, with the TSMC-made variants consistently outperforming. Samsung needs their 2nm node to restore confidence.
Samsung's SF2 uses second-generation GAA nanosheets with improvements to channel strain engineering and contact resistance. They're targeting:
- 12% performance improvement at same power versus SF3
- 25% power reduction at same speed versus SF3
- MBCFET (Multi-Bridge Channel FET) architecture with optimized sheet spacing
Samsung has the advantage of GAA manufacturing experience from their 3nm generation. The disadvantage is reputation — several potential customers have moved to TSMC after yield and performance issues. Winning them back requires flawless execution at 2nm.
Related reading: ASML and the EUV Monopoly: Why One Company Controls Advanced.
Volume production is targeted for 2026. Samsung is reportedly offering aggressive pricing to secure early customers, but price alone won't win back tier-1 chip designers if yield and performance don't meet expectations.
Intel 20A and 18A: The Comeback Attempt
Intel's node naming scheme is different (because of course it's). Intel 20A is roughly equivalent to other companies' 2nm. It introduces two technologies simultaneously: RibbonFET (Intel's name for GAA nanosheets) and PowerVia (backside power delivery).
Introducing both technologies at the same time is risky — each individually would be a challenging transition, and combining them multiplies the integration challenges. But Intel argues that PowerVia is essential for getting the full benefit of GAA nanosheets, since separating power routing from signal routing on opposite sides of the transistor layer gives more room for signal wires.
Intel 18A (roughly 1.8nm equivalent) follows quickly, with additional performance and density improvements. Intel plans to use 18A for their own products and offer it to external foundry customers through Intel Foundry Services (IFS). Microsoft has reportedly committed to using 18A for custom chips.
The success or failure of Intel's foundry ambitions essentially rides on 18A. If yield and performance are competitive with TSMC's N2, Intel becomes a credible alternative foundry. If not, their multi-billion-dollar foundry investment becomes much harder to justify.
See also: Neuromorphic Chips Guide: Brain-Inspired Computing with Inte.
The GAA Transition: Why It Matters
Gate-all-around nanosheet transistors represent the biggest change in transistor architecture since FinFET was introduced at 22nm over a decade ago. In a FinFET, the gate wraps around the channel on three sides. In a GAA nanosheet, the gate completely surrounds the channel on all four sides, providing better electrostatic control.
The practical benefit: GAA nanosheets can be made wider or narrower to trade off performance and power. A wide nanosheet drives more current (faster switching) but consumes more power. A narrow nanosheet is more efficient. Designers can tune this within the same process, using wide sheets for performance-critical paths and narrow sheets for power-sensitive areas.
The manufacturing challenge: stacking multiple nanosheets with precise thickness (typically 5-7nm each), spacing, and gate material deposition requires new equipment and process steps. The inner spacer formation — creating insulating barriers between the gate material and source/drain contacts — is particularly tricky and has been a yield limiter in early production.
Cost and Economics
A 2nm chip is expensive. Wafer costs are estimated at $25,000-30,000 per wafer (300mm), up from roughly $20,000 at 3nm. Design costs for a complex 2nm SoC can exceed $500 million, including EDA tools, IP licensing, mask sets, and engineering. Only products that ship in very high volume (smartphones, data center chips) or command very high prices (AI accelerators) can justify 2nm design costs.
This economic reality means 2nm will be used for a relatively small number of chip designs, even as older nodes (7nm, 14nm, 28nm) continue to serve the vast majority of semiconductor applications. The trailing edge isn't going away — it's actually growing, driven by automotive, industrial, and IoT demand.