The "2nm" in a modern process node is not a literal measurement — node names decoupled from physical transistor size after roughly the 22nm generation. "2nm" instead signals a manufacturing generation built on Gate-All-Around transistors, with TSMC's N2, Samsung's SF2, and Intel's 18A all reaching mass production in Q4 2025.
What Does the "nm" in a Process Node Actually Mean? Why Isn't 2nm Literally 2 Nanometers?
Process node names have not corresponded to any single physical transistor dimension since roughly the 22nm generation, and the number in a node name today functions mainly as a marketing label for a manufacturing generation rather than an actual nanometer measurementCITE:E1. Before that point, "Nnm" described a real physical feature such as gate length; after it, each foundry's naming scheme diverged from any shared physical standardCITE:E1.
Does a Smaller Number Always Mean Better? What "2nm" Really Signals
A smaller node number generally points to higher transistor density and better power efficiency — either faster performance at the same power draw, or the same performance at lower powerCITE:E2. But because node naming is decoupled from physical size, a "2nm" chip from one foundry is not guaranteed to match another foundry's density or efficiency; the name alone cannot answer that, only measured density and power data canCITE:E2.
Is 2nm an Architecture Turning Point: From FinFET to GAA?
The 2nm generation also marks a structural break in transistor design, replacing FinFET — the architecture the industry has used for roughly a decade — with Gate-All-Around (GAA) transistorsCITE:E4. Each foundry brands its own GAA implementation: TSMC calls its version Nanosheet, Samsung Electronics calls its version MBCFET, and Intel calls its version RibbonFET, all aimed at tighter current control and lower leakage than FinFETCITE:E4.
Why Is Advanced Process Manufacturing So Expensive and Hard? EUV Lithography and a Three-Way Oligopoly
Every advanced node generation depends on Extreme Ultraviolet Lithography (EUV) equipment, with a single EUV machine often costing hundreds of millions of dollarsCITE:E6. That cost barrier leaves only three companies — TSMC, Samsung Electronics, and Intel — currently operating at mass-production scale on the leading edge, while Japan's national foundry consortium Rapidus began 2nm trial production in 2025 as an emerging challengerCITE:E6.
2025's Three-Way Convergence at 2nm: TSMC N2, Samsung SF2, Intel 18A
TSMC's N2, Samsung Electronics' SF2, and Intel's 18A all enter mass production in the fourth quarter of 2025, converging on the same "2nm generation" label at effectively the same timeCITE:E3.
| Company | Node Name | Transistor Architecture | Mass Production Timing |
|---|
| TSMC | N2 | Nanosheet (GAA) | Q4 2025 |
| Samsung Electronics | SF2 | MBCFET (GAA) | Q4 2025 |
| Intel | 18A | RibbonFET (GAA) | Q4 2025 |
Timing data above reflects foundry process node trackingCITE:E3; architecture naming reflects each company's GAA implementationCITE:E4.
What's Next? Backside Power Delivery and 3D Structures
The next differentiation point beyond 2nm is backside power delivery: Intel's 18A introduces a backside power network called PowerVia, and TSMC's A16 (1.6nm), planned for the fourth quarter of 2026, will route power delivery on the back of the wafer separately from signal wiring on the frontCITE:E5. Separating power and signal layers this way is aimed at further gains in efficiency and density beyond what planar scaling alone can deliverCITE:E5.
What This Means
Taken together, the evidence points to a naming system that stopped tracking physical geometry after the 22nm generation, yet three foundries — TSMC, Samsung Electronics, and Intel — still converge on the same "2nm" label in the same quarter, each using its own branded version of the same underlying GAA architecture shift away from FinFET. That convergence is bounded by a cost barrier: EUV lithography's hundreds-of-millions-of-dollars machine cost is why only three companies operate at mass-production scale, with Rapidus's 2025 trial production the one visible attempt to join them. The next competitive marker already flagged — backside power delivery in Intel's 18A and TSMC's planned Q4 2026 A16 — suggests that as node names lose meaning as a size indicator, structural techniques like backside power delivery become the more concrete signal of generational progress.