Am 08.03.2012 12:44, schrieb Invisible:

> I can understand how something /complicated/ - like, say, a processor -
> would take time. Inventing a new feature takes a lot of R&D. But just
> making the transistors smaller? I don't really see why they have to
> shrink in tiny little baby steps. Why not just go the whole hog straight
> away?

Because maybe - just maybe - they have to /find out/ how to make that 
next "tiny little baby step"? After all, they're creating structure 
sizes nowadays where naive classic photolitpgraphy with ~400 nm UV light 
would fail utterly. Remember that old rule that you can't etch 
structures smaller than the wavelength of the light you're using?

Some of the "baby steps" involved:

- Replacing the ~400 nm Hg vapor lamps with ~250 nm KrF excimer lasers. 
Obviously you need to figure out how to build an excimer laser for this 
feat. You may also need to develop new optical materials (in case the 
old ones absorb the new wavelength) and/or new photoresists (in case the 
old ones are insensitive to the new wavelength).

- Replacing the ~250 nm KrF excimer lasers with ~200 nm ArF excimer 
lasers. Obviously you need to extrapolate your excimer laser design to 
use ArF, and again you may need new optical materials and/or photoresists.

- Replacing the liquid etching agent with plasma, to reduce undercutting 
of the photoresist. Obviously you need to figure out that plasma etching 
will have these benefits in the first place, and then find ways to 
handle the plasma in the fab.

- Exposing the photoresist in liquid rather than in air or vacuum, to 
reduce wavelength accordingly. Obviously you need to find a suitable 
liquid with high purity, and figure out how to handle it in the fab.

- Designing masks in such a way that they no longer directly correspond 
to the desired structures, but rather cause interference patterns of the 
desired shape. Obviously you need to develop suitable algorithms to 
generate the masks from the desired structures, and you probably need 
quite some computing power to pull this off.

Other steps may involve improvements of photoresists and application 
techniques (you want a thin yet uniform coating, which adheres well to 
the wafer) plus suitable developer, etching and resist stripping agents; 
improvements in wafer cleaning processes; reduction of mechanical 
tolerances; and other some such. Much of this goes into overcoming 
problems that hadn't been anticipated before.


Last not least, there's also the question of cost: Using cutting-edge 
technology for memory ICs would result in prohibitively high per-bit 
costs; such production lines are therefore almost exclusively used for 
high-margin products such as high-performance CPUs or GPUs, until the 
next generation of cutting-edge fab technology becomes available. The 
older production lines might then be used for another generation to 
manufacture low-cost CPUs and GPUs, and finally for stuff where 
per-transistor costs are more important than die size, such as memory ICs.

Post a reply to this message