So "40nm" doesn't mean that one transistor is 40nm big, or do you just mean that there are different ways to achieve that size and it doesn't say anything about the technical details (leakage, Tsetup,..)
xtor's are actually 2-dimensional, they have a length and a width. The
minimum length is actually the dimension which can be the smallest of the two in any given transistor but it can actually be smaller than the value denoted by the node label itself.
In memory (dram, flash, etc) the node label is intended to correlate with two times the minimum metal pitch allowed by design rules for the node.
Two questions, then:
1.) What was the "1000nm" Nehalem shown before? Perhaps it was in IDF, a supposedly working Nehalem chip, except that it was several several times much larger than a real Nehalem, and was supposedly built with "1000nm". If the numbers don't mean anything, then what makes it 1000nm?
2.) If the numbers don't mean anything, then what's with the progression? why 130nm, then 90nm, then 65nm, then 45nm or whatever... who dicated those numbers? If it were all arbitrary, why not go down 10nm at a time? Why not 130nm->120nm->110nm, etc? Why do they insist on a "Number" quantifier affixed to the "nanometer" unit measure if it is not really measuring anything?
Thanks, IDC.
The "1000nm" nehalem show-piece at IDF was just a mock-up diagram of the diemap of an actual nehalem scaled up, printed out on a printer and glued to cardstock.
It was merely a marketing gimmick to give some physical scale to the magnitude of improvements that have been generated by successive node scaling since the 1um days.
The progression exists owing to legacy, decades ago the node label was indicative of the minimum feature-size of a component in the device (like poly width, or channel length).
The progression now is a matter of convenience, the entire industry has grown accustomed to thinking about and referring to each successive node with the label that was preconceived by the ITRS roadmap over a decade ago.
Kind of like how each year the new model of any given car is labeled by the "model year" but that doesn't mean the car is manufactured only in that year or that it is only sold in that year, etc. All that a "model year" tells you is that it is older or newer than another model year vehicle (successive numerology is conserved) and armed with the model year and make/model of the vehicle you can go to a database and from there actually learn about the specs of the vehicle.
Same with node labels. It tells you nothing about the node other than a smaller node label is intended to imply things can be smaller than the same things in a larger numbered node. Some things (very few actually) scale in proportion to the scaling of node labels, but even that isn't as an exact a science as the node label progression would imply.
Now that I think about it, if the numbers are all arbitrary, why does TSMC have to skip its supposed 32nm node to go straight to 28nm? If the numbers don't mean anything, why don't they go to the 28nm process that they are going to now and just call it 32nm, thereby not have to "cancel" anything at all (as far as the public knows).
The node label is interwoven throughout all manner of internal projects, as well as collaborations with industry partners, etc. Think of it as a project codename like Nehalem or Istanbul...if you changed everything about the project but for some reason elected to recycle the same codename for a new project you can imagine the confusion and risk of misunderstandings that could generate across all the business entities involved. Cleaner and easier to just adopt a new codename, or node label as it were in this case.