While Moore’s Law may be coming to an end, why did this trend happen for as long as it did?
What bothers me about the regular doubling of transistor density every 18 months is that it implies dependency. That you can’t get to 16 nanometers without first doing 32 nanometers, and so forth. It almost implies that you need the computational power of a 32 nanometer IC before you can start working on the 16.
This doesn’t really follow. I don’t understand why the semiconductor manufacturers couldn’t figure out what the highest frequency of light they could ever practically use (extreme UV) and jump right to that.
Nothing ever works that way.
You need 1st generation tools to make second generation ones.
Semiconductor manufacturing is a hugely complicated process - it requires mask making (which in turn requires CAD tools), imaging equipment, exposure and developing equipment, etching systems, and all kinds of wet and dry chemistry.
I used be the engineer on an electron-beam mask making system. We could do 1 micron lines and spaces on a good day. The fab guys were still on 10 micron design rules, because their projection aligners and wet chemistry were the limiting factor. Once they move to step-and-repeat, they got down to the 5 micron range. Even though everyone knew that sub-micron was going to happen, there were no tools that could do it. The Mask Shop could maybe make the masks, but the step-and-repeat tools weren’t good enough (no bright enough light sources).
The only way the technology has gotten this far is by some exceedingly cleaver CAD tools that pre-distort the mask, so the the diffraction of the light as it passes through the mask actually sharpens the image. When I was in the business, everyone thought that we’d be using x-rays at this size design rules.
Thanks, beowulff. That’s a good explanation of why the improvements happen in small increments, and not in big jumps. But I’m also curious about the timing. The OP is pointing out the amazing coincidence that each increment takes about the same amount of time. Take a look at this chart in Wikipedia, which shows how incredibly steady the advances have been over forty years. There’s gotta be something behind it.
Do keep in mind that log plots make things look a lot more even than they are. The number of transistors could be 75% or 150% the expectation and still be visually not that far off the line.
It’s not spontaneous. A lot of the major companies in the industry are coordinating with each other to follow a predetermined roadmap/timeline for when they will move to the next increase in semiconductor density.
[QUOTE=Wikipedia]
As the industry evolved, each of these operations were typically performed by specialized machines built by a variety of commercial companies. This specialization may potentially make it difficult for the industry to advance, since in many cases it does no good for one company to introduce a new product if the other needed steps are not available around the same time. A technology roadmap can help this by giving an idea when a certain capability will be needed. Then each supplier can target this date for their piece of the puzzle.
[/QUOTE]
