Process Node Naming
A process node name like "7nm" or "3nm" stopped describing any single physical dimension on the chip decades ago — it's now a marketing-era label each foundry assigns to a new process generation, and the actual transistor geometry has to be looked up separately.
This disconnect solves nothing — it's a naming problem, not an engineering one, and it exists because the industry kept the old naming convention long after it stopped describing anything physical. In the planar-transistor era, a node name like "90nm" genuinely tracked the minimum gate length a process could print, so smaller numbers meant smaller, faster, denser transistors in a fairly direct way. Once FinFETs and other 3D transistor structures took over, gate length stopped being the dimension that limited density or speed, but foundries kept assigning ever-smaller node numbers anyway — partly because customers expected the familiar cadence, and partly because there was never a single successor metric everyone agreed to switch to instead.
That gap is exactly why comparing node names across foundries is misleading, and why engineers look at real specs instead. TSMC's "7nm," Samsung's "7nm," and Intel's old "10nm" (later renamed Intel 7) were never built to the same transistor density or performance despite the similar-sounding labels — each foundry's marketing name is internally consistent across its own generations but not calibrated against any other company's naming at all, which is why a serious comparison between two processes uses gate pitch, metal pitch, and transistor density instead of trusting the node name to mean the same thing twice.
| What people assume | What's actually true |
|---|---|
| "7nm" means a 7 nanometer gate length | Gate length hasn't tracked the node name since the FinFET era |
| Same node number across foundries means the same process | Each foundry's node name is its own internal label |
| A smaller node number always means a better process | Real density and performance require comparing actual specs |
| Node names are a standardized measurement | Node names are a marketing-era naming convention |
| Real metric | What it actually measures | Why it replaced the node name |
|---|---|---|
| Gate pitch | Distance between adjacent transistor gates | Tracks real logic density directly |
| Metal pitch | Minimum spacing of interconnect wiring | Often the true density-limiting dimension |
| Transistor density | Transistors per square millimeter | The actual apples-to-apples density comparison |
| Contacted poly pitch (CPP) | Gate pitch including the contact | A standard density-related figure of merit across the industry |
Engineers who need a real comparison between processes skip the marketing node name entirely and go straight to these physical numbers, because that's the only way to compare processes from different foundries fairly. Transistor density, measured in millions of transistors per square millimeter, and metrics like gate pitch and contacted poly pitch give a foundry-independent way to size up how a process actually compares to another — it's slower to look up than quoting a node name, but it's the only comparison that actually means what it claims to mean.
Read the node name as a product-line label, not a measurement: it still tells you roughly where a process sits in a foundry's own generational lineup, but the moment you're comparing across foundries, or asking what a node number actually buys you in density or performance, the node name itself stops being useful and the real physical specs are the only honest answer.