Bitcoin Forum

Title says all. What's after 28nm? Is it worth going to 14nm ( for ASIC's that is )?

knc already has 20nm.
I don't know if it is going to work out with knc since they delayed their customers very long.

I don't know where you're getting your information but they said shipping was to before the end of Q2. It's the end of Q2 and guess what? They're shipping. So they're not late.

I thought Q1, as it turns out, it is Q2.
My mistake.

So nobody's gone to 15 or 14 nm yet?

None of the commercially usable fabs are offering 14nm yet. 16 nm may be possible over the next year or so but there's a huge amount of capacity going to be taken by very large semi companies that want to build CPUs, etc.

20 nm is an option but it's a half node so it's only about 2x as good as 28 nm.

GPU manufacturers are going towards 14nm after 20nm.I think we will see the same with asics.

Both AMD and nVidia have been having trouble with this next step.  I remember the 6xxx series cards were supposed to be the same process as the 7xxx.  nVidia is taking forever to release new Maxwell gear.

It will take some time to jump down

We will see 14nm next year surely even in asics.

ASML is currently working on EUV lithography machines, (it are actually soft X-rays).
But I don't know what resolution is feasible with that.

mine with x-rays machines?

Yup...KnC confirmed to be releasing 20nm product in their next-gen processors.

This would be their current generation, the Neptune, which is already shipping.

Is that a joke?
That is to make the chips with.

It never stops, 28nm -> 20nm -> 14nm -> 10nm -> 1nm -> 0.1nm

Physics preclude the latter numbers :)  sub-12nm is already going to be a challenge.

Like I said it will be a while.   Switching from electricity to photons essentially.

I don't know about 0.1nm, that is smaller than one silicon atom.(about 0.2nm). source: wikipedia.
1nm is already increadibly hard, even if one manages to get that high of a fidelity, quantum tunneling effects play a big role.
You will then lose the properties of a 'normal' transistor.

The main benefits of photonics are not so much in feature size - if anything, right now they tend to be bigger - but in the fact that a photon can travel through a vacuum unimpeded - so if you need to get a signal from one side of a chip to the other, doing so with a photon means you use less energy than you do with nudging an electron in one end and having another get bumped out the other end with losses (and thus heat) along the way - and a high frequency signal doesn't meaningfully affect neighboring elements and pathways.

Sub-12 is really, really hard (technical term).  I think it more likely that performance increases will come from more novel techniques (whether or not it ends up standing the test of fabrication, Introducing the Vacuum Transistor: A Device Made of Nothing (http://spectrum.ieee.org/semiconductors/devices/introducing-the-vacuum-transistor-a-device-made-of-nothing) [ieee.org] is a recent example), rather than trying to shrink things down further.  As mikerbiker6 points out again, some things are just physically 'impossible'.

Yes! That was what I was thinking of! The Vacuum Transistor! If only theirs a foundry that already supports it....and I had sufficient funds  :P

thanks for the source steve

2x? think you need to look up your scaling numbers for node shrinks (especially theoretical scaling vs actual on the latest generations), also 28nm is a half node so the jump is similar as for example 32 > 22nm.

"Half node step" :-)

20nm should give around 2x- it will depend on the exact process of course but there's an increase in density and an uplift in frequency (or reduction in power consumption). We're talking about a class if design where trivial parallelism leads to almost linear speedups.

It seems that it will be worth going to 14 maybe even to 10 but for less it could get incredibly hard.