iansjack wrote:But you said people were "scared" of this. I don't think they are.
I wouldn't say that necessarily. Maybe the word "scared" is a bit strong, but the semiconductor industry and the research behind it is putting quite the push behind extending it as far as they can. They have a vested interest to make it continue, and while people may not be scared about Gordon Moore's observations not holding accurate, they certainly are scared about the safety of the money they've put on that bet.
Back in ye olden days, the conventional notion was that you take a photomask of a circuit, sorta like a negative, and shoot UV light through it to knock out photoresist, which is a UV-photosensitive coating applied to the wafer. That imprinting of the pattern then allowed you to apply the various coatings to the wafer needed to make the transistors, such as N/P doping, dielectric and insulation, metallization, etc. After 30-40 exposure and treatment passes, you wiresaw and package the wafers and a new batch of x86 devices has seen their first sunlight.
This worked until the industry hit the point where they would be required to make features that are smaller than the wavelength of the light coming out of the source they were using, sort of like how the microwaves in your oven can't "see" out the holes of the mesh in the door. First it was 405nm UV lamps, then it was 248nm KrF lasers and then 193nm ArF lasers. Then we hit the point where we could no longer punch lower, so we began using specialized interference techniques when generating the photomask patterns to cause the light to cancel out and make the effective patterns. Getting to 22nm has been done with 193nm light. The electrical physics boys had joo-jooed the system and were painting really tiny lines with a much bigger paintbrush.
The issue is that even that can't continue, and the 22nm fabrication nodes really proved that. Hence, extreme ultraviolet, or EUV. EUV's been in the works for years and it's behind schedule, but it's finally starting to roll out. The premise here is that you need to imprint the wafers with 13.5nm light, which is interestingly enough a) emitted by no matter under normal conditions, and b) absorbed by >99.9% of all matter. The funny principle here is that the most efficient manner of generating EUV source for the lithographic steppers involves taking a sub-millimeter molten drop of tin in high vacuum and ramming it with 100kW of infrared laser power. The resulting plasma will emit EUV, and as of some conferences I attended in 2013, the record output was about 40W of light. Combine that with the fact that the whole situation must take place under high-vacuum pump-down so it's an order of magnitude slower, and you get a real piece-of-work solution.
All that is required in order to potentially hit 13nm or lower. I'd say the industry at large is probably afraid of some sort of end-of-Moore's-law boogeyman, seeing that they've put billions upon billions of dollars of research into technology with less than a 1% efficiency rating. When Moore's law begins to cross the threshold on the way out, a lot of people in a lot of places are going to have to sit down and think real hard about ways to survive in an industry that tends to sustain itself on the notion that "better stuff" will continue to be released year in and year out.
