Hacker News Reader

Comment by hugh3 | Hacker News Reader

The ongoing miniaturization of silicon technology has to end fairly soon, as the atomic scale is approaching rapidly. The 11nm node is scheduled for 2015. The silicon-silicon bond length is 0.235 nm. There's no longer plenty of room at the bottom.

The question is whether, when silicon runs out, something else can take over, or whether we just put our hands up and say "OK, that's it, that's about as fast as our computers are ever gonna get".

There's the possibility that alternative technologies will exist but that they'll be much more expensive. So consumer electronics will be based on silicon, and outrageously expensive supercomputers will use graphene or diamond or any one of a number of currently speculative technologies. So far, though, none of 'em is even close to being able to compete with silicon for any price.

View on HN

> The ongoing miniaturization of silicon technology has to end fairly soon, as the atomic scale is approaching rapidly.

Again: The issue is not the material science or the physics, it is the economy of scale.

Even if miniturization stopped tomorrow, there would be bigger wafers, or 3D stacking, or wafer-scale integration, or (insert favorite technology here.)

Those economics are very tough to compete with. The net effect is that other materials are relegated to markets where there is an honest differentiator. For example: III-V's are much superior lasers compared to Si-based technologies. (Look up "Direct Bandgap" as to why.)

> So consumer electronics will be based on silicon, and outrageously expensive supercomputers will use graphene or diamond or any one of a number of currently speculative technologies.

Outrageously expensive supercomputers are massively parallel. For massively parallel loads, the fastest chips are the cheapest chips. If the new stuff cannot break into consumer space, there is no market for it above consumer space.

Presuming there will never be a market for a very fast, very small computer is making a lot of assumptions about, for example, space travel.

Actually, I think diamond is getting much cheaper. It's becoming much more integral for satellites and extremely small electronics because of their susceptibility to radiation. http://news.vanderbilt.edu/2011/08/nanodiamond/

The first thing I'd wonder is what the susceptibility to radiation is for spintronics. The paper "Spintronics: a new paradigm for electronics for the new millennium" says that it is as hardened as the silicon. So then I'd question how hard the graphene is and found a couple of papers about that: Unzipping and folding of graphene by swift heavy ions, Effect of electron-beam irradiation on graphene field effect devices.

Researchers out there realize the problems popping up with silicon and are working to find the solutions. It's a really cool time to be in material sciences, physics and/or electrical engineering.

> Actually, I think diamond is getting much cheaper.

Yeah, but good luck doping it n-type.

> The question is whether, when silicon runs out, something else can take over, or whether we just put our hands up and say "OK, that's it, that's about as fast as our computers are ever gonna get".

The single-core speed trend has already started to level-out, which is why we went to multi-cores.

There are some relevant slides in the talk by http://www.stanford.edu/class/ee380/Abstracts/111005.html

My money's on chip stacking! At least until other processes reach the efficiency of silicon, anyway.

Chip stacking tends to be too expensive. Think of it this way: take the full output from a factory creating chips. Now if you stack just one chip on top of another, you have effectively cut your production in half. It also means that your chips will have to be at least twice as expensive (probably more since demand will be greater since the chips are now more scarce).

I think there is some interesting research work going into chip stacking but it wouldn't be accessible for public consumption until it's cost effective.

Reply on news.ycombinator.com