Slashdot Mirror
A Greener Chip Manufacturing Process
gardenermike writes "A new chip manufacturing process has been developed which uses UV light instead of high temperatures to prepare the silicon. This could lead to cheaper chips and greener factories if it pans out. Apparently the main problem is defects in the material, which are currently 'ironed out' as a side-effect of the extreme temperatures used."
They're not exactly arbitrary, but they're not physically imposed (like quantum rules or something) either. They're basically just more-or-less a constant ratio from one down to the next.
The semiconductor companies get together and publish a roadmap called ITRS that says we should all try to get to X nm by 20xx, and here are the challenges, etc.
Now someday we're going to get to one of these technology nodes, as they're called, and find out there really is a fundamental phyiscal limitation that keeps us from going any smaller but we haven't got there yet. (Finally, my sig is directly relevant to what I'm posting!)
...the future crusty old bastards are already drinking the Kool-Aid.
Semi-arbitrary. Each whole generation is half the feature size of the generation before, starting from 0.650 micron (650nm). In between are "half" generations counting down from 1000micron.
Half generations: 1000, 500, 250, 130, 65, 32, 16
Whole generations: 650, 350, 180, 90, 45, 22
The precise digits are chosen for convenience and actual processes vary a bit up and down for a given technology node. Each node requires new equipment. By moving from node to node together, manufacturers share some of the cost of development. Still, odd ball nodes do exist. DRAM's are often manufactured at intermediate dimentions and 150nm is used by some foundaries.
Many fabless chip makers will skip half generations. I know a lot of manufacturers went straight from 350 to 180. Still, the choice to skip or not is mostly economic. If a node lands durring a recession, fabless chip makers are likely to hold off until the node that follows. The fabs don't really have a choice. They have to produce each generation in sequence, at least at small scale, or they will not have the technological base to start work on the nodes that follow.
The sizes are governed by 2 factors, the wavelength of the UV light used, currently 193nm, transitioning to the 157nm for the 45nm chips, and the diffraction gratings/refraction of immersion fluid/polarization of said fluid, etc. used. Due to the physics behind this (i wont bore you with the long equations, because i dont want to do them again) there's basically certain points at which these effects add up to the greatest possible resolution/intensity/etc. Any more in depth and i'd have to dig up my lithography text, and i dont really want to :)
drunk chemists
Each is the previous divided by the square root of 2.
The reason for this is that if you decrease the feature size by the square root of 2 on each side, the feature shrinks to half size (since they are 2D features).
You can see this by squaring them all
180^2 = 32,400
130^2 = 16,900
90^2 = 8,100
65^2 = 4,225
45^2 = 2,025
32^2 = 1,024
22^2 = 484
16^2 = 256
See how each is half the size of the previous?
I guess doubling the number of things (transistors) makes sense to humans. It sure makes it easy to calculate how many dice you can fit on a wafer after a shrink. It'll be about twice as many as before.
You see it in hard drive sizes:
8^2 = 64
5.25^2 = 28
3.5^2 = 12.25
2.5^2 = 6.25
1.8^2 = 3.24
1.3^2 = 1.69 (missing for some reason)
1^2 = 1
You also see it in f-stops on cameras, where each f-stop lets in twice the light of the previous
1/22^2 = 1/484
1/16^2 = 1/256
1/11^2 = 1/121
1/8^2 = 1/64
1/5.5^2 = 1/30
1/4^2 = 1/16
1/2.7^2 = 1/8
1/2^2 = 1/4
1/1.4^2 = 1/2
1/1^2 = 1
All are arbitrary, being for the convenience of humans doing math in their heads.
http://lkml.org/lkml/2005/8/20/95
The inter-metal-layer oxide that is begin replaced by low-K dielectrics is/was not formed by oxidation but by chemical vapour deposition. This oxide process is targetted at gate or isolation dielectrics. High-K dielectrics are in the roadmap to replace oxide at the gate dielectric, but development is much slower than people thought 5-6 years ago. In fact, it may never happen before we switch to different materials altogether.