Node Shrink Economics
Node-shrink economics is the comparison between how much more a new, smaller process node costs to build and run against how much benefit, in transistor density, performance, or power, that smaller node actually delivers, and the central fact shaping this comparison today is that the cost side has kept climbing steadily while the benefit side has not. For decades a node shrink reliably cut the cost per transistor along with the die area, which is the economic engine that made Moore's Law a business proposition and not just a technical one; at the most advanced nodes today, cost per transistor has flattened or even reversed for many designs, so the decision to move to the next node is no longer an automatic win and has to be justified case by case.
Three separate cost drivers push against the next node's economics simultaneously, which is why the flattening is structural rather than a temporary blip. Photomask sets for the most advanced nodes now run into the tens of millions of dollars per design because of the sheer number of layers and the precision each one requires; the fab capital cost per unit of capacity keeps rising as EUV and multi-patterning tools replace cheaper prior-generation equipment; and chip design cost itself has grown sharply at advanced nodes due to the verification and physical-design effort required to make a design actually work at that node's tighter tolerances, so all three cost categories are rising even as the per-transistor area shrinks.
This is the direct reason chiplets and advanced packaging have grown in importance alongside pure node migration — splitting a large design into smaller chiplets lets only the portions that genuinely benefit from the newest, most expensive node actually move there, while less performance-critical functions stay on a cheaper, older node, capturing part of the density benefit without paying the full leading-edge cost across the entire design.
| Cost driver | Historical nodes | Leading-edge nodes today |
|---|---|---|
| Photomask set cost | Modest, scaled gradually | Tens of millions of dollars per design |
| Fab capital cost per capacity unit | Rising gradually with each generation | Rising sharply with EUV/multi-patterning |
| Design and verification effort | Manageable with existing tools and teams | Major cost center in its own right |
| Net effect on cost per transistor | Reliably falling | Flat or rising for many designs |
Node-shrink economics is the design-level counterpart to the fab utilization economics covered in the previous row — both describe how the enormous fixed cost of leading-edge semiconductor manufacturing has to be justified against actual output, whether measured in wafers run or transistors delivered per dollar.