Intel predicts Moore's Law to apply down to 7nm

[HEXUS]

And possibly further. Says pilot 10nm programme is on track.

Read more.

[shaithis]

Every time they predict Moores coming to an end, it continues

Very funny considering it was a "finger in the air" idea to begin with.

Maybe we will see stacked CPUs next, that'll keep it going for a while......

[Kumagawa]

So Intel when are you going to show working 10nm chips? even just SRAM prototypes?

16/14nm prototypes were first shown off in 2009 with retail chips in 2014.

Yet when ever Intel's investors bring up 10nm at investment meetings Intel's all talk and no show.

[AETAaAS]

I'm wondering is there a physical reason it cannot get smaller? Nanometer has been the measurement for a while now and seeing as we have been closing in one the smallest full number, is it theoretically possible for there to be pico or femtometer lithography in the future?

[Gerrard]

The size of an atom is at least 0.1nm, so I guess in theory that's as small as you could go?

[Fraz]

Indeed. You definitely can't make transistors out of less than one silicon atom...

[Peter Parker]

Indeed. You definitely can't make transistors out of less than one silicon atom...

Who knows? We've managed to do some crazy stuff in the last 50 years. I'm loath to predict what can happen. I've read about extreme ultraviolet lithography being possible for down to 5nm at least.

I used to argue that <5nm was impossible based on the size of a silicon atom. In Intel's 14nm process (which is just a marketing term), the distance between fins on the FinFETs is something like 45nm. Individual features in each transistor are smaller, and some layers of the chip are probably less than 10nm. For sake of argument, let's pretend those tiny fins are 10nm wide. The crystal lattice size of silicon is about 5.43 Angstroms, or 0.543nm. That's the length of one of these cubes. I make that fin to be about 18 crystal cells wide, or about 36 atoms wide!

So at 6-7nm ... we're talking about features 15-20 atoms across. The problem now is accuracy and reliability. Sure, in a research lab I can make a single transistor at this scale. One out of 100 might work, but all the rest have weird defects. Companies did this 10 years ago. Making a chip where all 10-50 billion transistors function reliably to the same spec is insanely tough. One atom out of place is a 5% error, which is just too much.

But Intel have $$$ for R&D. Perhaps they can't do 15 atoms, but they might get 25 and call it 10nm. And they might have another clever shape (2.5D/3D chip "stacking") like the FinFET and call it 7nm. And perhaps another material works better than silicon - III/V semiconductors like Gallium/Indium with Phosphor/Arsenic. Or perhaps they invent a chip that doesn't need every calculation to be 100% accurate! Can you use 5% extra transistors for error checking if you get a 15-20% process shrink.

At some point the atomic limits really do kick in and something will have to be done. Molecular computing, graphene, electron spin and other quantum "stuff" that I just don't understand ... it's all being researched right now. I'm fairly sure a single atom transistor has been demonstrated. Perhaps someone can figure out how to superimpose multiple quantum states and have a "dual" transistor on one atom?!

But what would you do with it? Crysis 5?

[Fraz]

Ok, perhaps I should have said: You definitely can't make *silicon* transistors out of less than one silicon atom. By definition, it wouldn't be silicon.

That's not to say there aren't many avenues for improving the performance of computers. In the short term I expect we'll crack optical computing or something graphene based, given that it's everyone's favourite material right now. And then in the longer term, you'd hope we'd get a quantum computer together that wasn't basically just a fancy analog computer.

[Corky34]

I didn't think it was to do with how small you can make a transistor, but more about what happens to electrons on such a small scale.

EDIT: Just done some quick research and it seems once you start going under 10nm you experience quantum tunnelling, not a road block in it's self, but things start getting rather strange.

[edzieba]

The size of an atom is at least 0.1nm, so I guess in theory that's as small as you could go?

That's feature size. Chips aren't perfect 2D flat planes, they're 3d structures. If you;re clever about how you arrange things vertically, you might be able to end up with features that undercut each other, with the feature size (measured from above) ending up blow that of a single atom.
Of course, at that sort of scale you're performing picoassembly and have all kinds of weird quantum effects to consider in addition to the geometry.

[mazty]

The size of an atom is at least 0.1nm, so I guess in theory that's as small as you could go?

No - that's the size of a helium atom, so unless you want to try and turn that into a transistor, you are going to be limited to the size of silicon, which is much larger. Last time they investigated this ( a few years back) 4nm was the complete limit. Granted different materials may be discovered, however realistically speaking, from 2020 on-wards they may have to completely redesign CPU's a.k.a. spend billions on bringing quantum computing up to speed.

[mazty]

Indeed. You definitely can't make transistors out of less than one silicon atom...

Who knows? We've managed to do some crazy stuff in the last 50 years. I'm loath to predict what can happen. I've read about extreme ultraviolet lithography being possible for down to 5nm at least.

I used to argue that <5nm was impossible based on the size of a silicon atom. In Intel's 14nm process (which is just a marketing term), the distance between fins on the FinFETs is something like 45nm. Individual features in each transistor are smaller, and some layers of the chip are probably less than 10nm. For sake of argument, let's pretend those tiny fins are 10nm wide. The crystal lattice size of silicon is about 5.43 Angstroms, or 0.543nm. That's the length of one of these cubes. I make that fin to be about 18 crystal cells wide, or about 36 atoms wide!

So at 6-7nm ... we're talking about features 15-20 atoms across. The problem now is accuracy and reliability. Sure, in a research lab I can make a single transistor at this scale. One out of 100 might work, but all the rest have weird defects. Companies did this 10 years ago. Making a chip where all 10-50 billion transistors function reliably to the same spec is insanely tough. One atom out of place is a 5% error, which is just too much.

But Intel have $$$ for R&D. Perhaps they can't do 15 atoms, but they might get 25 and call it 10nm. And they might have another clever shape (2.5D/3D chip "stacking") like the FinFET and call it 7nm. And perhaps another material works better than silicon - III/V semiconductors like Gallium/Indium with Phosphor/Arsenic. Or perhaps they invent a chip that doesn't need every calculation to be 100% accurate! Can you use 5% extra transistors for error checking if you get a 15-20% process shrink.

At some point the atomic limits really do kick in and something will have to be done. Molecular computing, graphene, electron spin and other quantum "stuff" that I just don't understand ... it's all being researched right now. I'm fairly sure a single atom transistor has been demonstrated. Perhaps someone can figure out how to superimpose multiple quantum states and have a "dual" transistor on one atom?!

But what would you do with it? Crysis 5?

But the smaller you go, you need to factor in the quantum effects that end up playing havoc with the system. It may be much more feasible to redesign CPU's rather than create a gigantic error correction array.

[DanceswithUnix]

Looks like Samsung are doing OK as well: http://www.fudzilla.com/news/process...ff-10nm-finfet