Some dumb questions about the semiconductor industry

# Some dumb questions about the semiconductor industry * What do the unit economics of chip manufacturing look like *in 2022* for different node sizes? Eg. For different node sizes, how much does it cost to: * Build a fab from scratch? * Tape out a chip? * Make a wafer? * Make a single transistor? * Make a single unit of area ($/cm^2 or something) I’ve found people talking about all of this in incredibly hand-wavy terms, but nothing that lets me make head to head comparisons. * Concretely, what are things that you can do with smaller nodes that you can’t do with bigger nodes? I’ve read some allusions to “AI” — do high-end GPUs and tensor processors have a maximum node size? I know you get better power efficiency at smaller nodes (how much better?) — but where does that make or break an application? * My understanding is that there’s a pretty big break between the process for doing nodes bigger than ~25nm and smaller than that. What mechanically is the difference — is it that extreme UV needs you to create plasmas to create the UV lasers and that’s really hard?  [Web URL for this note](http://notes.benjaminreinhardt.com/Some+dumb+questions+about+the+semiconductor+industry) [Comment on this note](http://via.hypothes.is/http://notes.benjaminreinhardt.com/Some+dumb+questions+about+the+semiconductor+industry)