what is a foundry node
A foundry node — often written as "5nm" or "3nm" — is the name a foundry gives to one generation of its chip manufacturing process, and despite the number sounding like a precise physical measurement, it has become primarily a marketing label: two different foundries' "5nm" processes can differ meaningfully in actual transistor density and performance, because there's no single, enforced industry definition tying the number to a specific measurement anymore.
```flowchart
{
"rows": [
{ "type": "nodes", "items": [
{ "title": "Decades ago: node name = a real transistor dimension", "sub": "e.g. gate length in nanometers", "tone": "neutral" }
]},
{ "type": "arrow" },
{ "type": "group", "title": "Transistor shapes changed; old measurement stopped applying cleanly", "items": [
{ "title": "New 3D transistor structures", "sub": "no single dimension maps to \"node size\" anymore", "tone": "orange" }
]},
{ "type": "arrow" },
{ "type": "nodes", "items": [
{ "title": "Today: node name is mostly a marketing generation label", "sub": "compare real specs, not just the number", "tone": "green" }
]}
]
}
```
**Node names originally described a real, physical transistor dimension, which is why the naming convention stuck around even after that link broke.** In earlier decades of chipmaking, a process node's number genuinely corresponded to a specific measurable feature, like the length of a transistor's gate — a smaller number meant a genuinely smaller, denser transistor by that specific measurement, and this correspondence made comparing "90nm" to "65nm" meaningful across different manufacturers.
```svg
```
**The naming split from physical reality once transistor shapes stopped being simple, flat structures.** As transistors moved to more complex three-dimensional shapes to keep shrinking effectively, no single number could honestly capture the transistor's overall size the way "gate length" once did — but the marketing convention of naming each generation with a shrinking number had become so entrenched across the industry that foundries kept using it, now largely as a competitive label rather than a strict physical specification.
| What node names used to mean | What they mean today |
|---|---|
| A specific, measurable transistor dimension | Mostly a marketing generation label |
| Directly comparable across different foundries | Not reliably comparable between foundries |
| Smaller number = objectively smaller feature | Smaller number = newer generation, roughly denser |
| One clear industry-wide standard | Each foundry defines its own naming internally |
**Because node names aren't standardized, meaningful comparisons require looking past the number to real specifications.** Metrics like actual transistor density (transistors per square millimeter), power consumption at a given performance level, and independent third-party analysis of manufactured chips give a far more honest picture of how two "same node" processes from different foundries actually compare, rather than assuming identical numbers mean identical technology.
**Choosing a foundry node is a major decision for any chip design team, balancing cost, performance, and availability rather than simply picking the smallest number.** A newer, smaller node typically offers better performance and efficiency but costs dramatically more to design for and may have limited manufacturing capacity available; many chip designs deliberately use an older, cheaper, more mature node when the performance gain from the newest node isn't worth its higher cost and lower initial yield.
Read the foundry node name through a generation-label lens, not a ruler: the number tells you roughly how new and how advanced a process is relative to that same foundry's own older processes, but comparing the actual number across two different foundries tells you far less than it appears to — real comparisons need real specifications, not just matching digits.