WHAT’S IN A NANOMETER?
What originally started out as a rough scale of transistor density and process technology capabilities has morphed into a marketing term as much as anything. Intel has (rightly) noted in the past that its 14nm technology was “better” than many of the competing solutions. TSMC also ended up on the wrong side of the fence with its 16nm node, which it later renamed to 12nm just to look better (with some minor refinements).
Looking forward, there are numerous manufacturing nodes coming from a variety of companies. Intel has SuperFIN and Enhanced SuperFIN 10nm nodes that are supposed to compete with the 7nm nodes from Samsung and TSMC. The problem is determining which 7nm node we’re actually talking about.
Samsung’s 10nm node lands between Intel’s 14nm and 10nm nodes. It’s also shipping 7LPP and second-generation 7LPP, while a special 8N (enhanced 10nm) is currently in use for Nvidia’s RTX Ampere GPUs. 5nm 5LPE already went through the risk production back in 2018, and retail products should presumably arrive soonish.
TSMC has the most going on, with N7FF, N7P, N7FF+, N6, and N5 nodes all in production. The various nodes are optimized for power or performance, and N7FF+, N6, and N5 all support EUV (Extreme Ultraviolet) layers. Apple’s M1 and A14 chips use TSMC’s N5 node.
Within any company, lower node numbers are pretty much universally better, but trying to compare TSMC N7 to Intel Enhanced SuperFIN ends up being far more nuanced. Ultimately, it comes down to density, clock speed, and power characteristics, not to mention yields—as Intel learned, a 10nm node with terrible yields isn’t very useful (RIP Cannon Lake).