Intel Has 5 nm Processors in Sight

Page 5 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Toggle sidebar Toggle sidebar
Status
Not open for further replies.
New materials like graphene should allow to go even further, to the scale of 1nm. And after that the industry could stack layers of transistors and increase surface area too.
How much time could that be used? What physics allow. But for sure there's a limit and we'll hit it in this century.

I only hope that limit gives us enough computer power to do all the things that must be done.
 
[citation][nom]spentshells[/nom]Carbon nano tubes and Graphene[/citation]

Tesla's Ether physics was much more accurate. and his inverse square law of gravity explained both macro and micro objects and how they behave without need to change calculations or alter laws going from one to the other.. Tesla: a man before his time
 
[citation][nom]danwat1234[/nom]Where the heck did you hear that!?[/citation]

AMD is still on 32nm and even so they are competing with a company on 22nm. Their CPUs blows anything Intel had at 32nm out of the water. I mean utterly demolishes it. Intel might compete AMD into non-existence riding on a huge cash pile and fanatic fandom, but I think that in the end AMD would be the ones to make the best processor.

Bit sad really.
 
Um... It highly depends on the workload. As I saw, Vishera has yet to beat Sandy Bridge in gaming for one thing, along with other lightly-threaded tasks, though it still deserves merit for its substantial improvements over Zambezi. :)

Oh yeah, I have a feeling that the i7-2600K/2700K would still trump Visher generally speaking...
 
[citation][nom]blazorthon[/nom]Technically, Intel might have hit 10GHz if they stuck with Netburst or at least Netburst-derived architectures up until today, but it probably wouldn't be as practical as Sandy Bridge and Ivy Bridge.[/citation]
No they would not. Remember how hot the 3Ghz+ Pentiums got?
 
I seem to recall Intel having 10Ghz netburst technology in sight as well.

A silicon atom has a diameter of 234pm so in theory there is 4 atoms per nm. To make semi-conductors (transitors) silicon needs to be doped with 12-15% impurities. That means that one in 7 or 8 atoms must be replaced by a different molecule.

So in theory 0.5nm2 is the absolute physical limit. However, it would require a significant breakthrough in doping technology to make direct doping possible on an atomic level instead of a technique that is based on the law of averages as it is now.

At 5nm (25nm2) there's 400 atoms each transistor. Between 48 and 60 of these are doped. More or less would mean that single transistor doesn't work and the whole processor in principle doesn't work as designed either.

With current doping techniques that would mean that at 5nm, 99.999% of all processors produced will contain enough flaws that they have to be scrapped them.

Then we haven't even touched on the quantum tunnelling issue that will start to have a significant effect when distances between transistors are 6nm or less.

I remain skeptical that Intel really has a roadmap to solve the technological challenges to go to 5mn, let alone beyond.
 
Silicon to Carbon to Molly (off memory). Yes there will be issues. I know too little to know if they can get over those problems.
But just a side point
"Also, Intel said that it is planning to move from 300 mm to 450 mm production wafers, but this switch is still about five years out."
They will produce a lot of cpu's per wafer at 10nm, this should make them a lot cheaper too.
 
[citation][nom]math1337[/nom]Remember when we were supposed to have 10 GHz processors?[/citation]

yea and imagine the radiation coming out of tat frequency ....

more cores is safer.
 


~You~ won't be putting the CPU onto anything, the mobo manufacturers will be soldiering them on. It's a blatant money grab from Intel, they don't want people "upgrading" their CPU to get more performance, they would rather you buy an entirely new board.

Intel makes their money not from the CPUs but from all the licensing and additional chipset / components that go into the motherboard and supporting technology for that CPU. Think of it the same way the Inkjet printer makers or the mens razer manufacturers do (the swappable heads). Better to make money through a disposable product the consumer has to keep repurchasing.
 


Intel waffles back and forth on that issue. They will still produce socket CPU's for the time being, but they will eventually cease to offer them. It's just economics, as you shrink something smaller and smaller eventually you get to the point where your gotta integrate things. You'll start to see it in the next two years with "sockets" being for the desktop world and that will eventually shrink to enthusiast only and then server / workstation only.
 
[citation][nom]buzznut[/nom]Agreed, quantum computing is where I'd be investing my R&D right now. I think they will have problems before 2019.[/citation]

The problem with quantum computing is that any vibration, you sneezing a truck driving by an earthquake someone walking across the floor or even someone knocking on your door will destroy the current calculation a quantum computer is doing. That's why they are only found in research centers and universities. They have yet to find a way to make them stable.
 
[citation][nom]blazorthon[/nom]Actually, Moore's "law" did fail. It stated that the amount of transistors that can be inexpensively placed on a chip (IIRC) should double roughly every twelve months. We're down to something like every twenty-four or thirty months right now.[/citation]

Moore's law states that over the history of computing hardware, the number of transistors on integrated circuits doubles approximately every two years.

The shorter time periods often quoted "12 - 18 months" are due to Intel executive David House, who predicted that the period for a doubling in chip performance based on a combination of the effect of more transistors and their being faster should be shorter.
 
This commentary is pretty funny. In the future this will happen and that will happen.I am sure again this just another marketing ploy of intel that is ahead of the actual science. What part of the chip is 22nm? It is the cut between the features that make up a transistor. ? Hmmm. I have to tell you that the stuff has to come alot bigger to connect to anything useful. I am pretty sure many of the transistors are alot bigger.

I quote //Jonathan Kang, CPU Designer

Those are typically the feature size. That is to say, it is the smallest "cut" that the fabrication method can make into the silicon. If one were to carve out a transistor on the surface of a silicon wafer, one has to be able to etch to a certain degree of precision. If the smallest cut you can make into the silicon is, say, 11nm, then in order to make a piece of a transistor, you'll need to cut a small square (really a dot) and then leave a space for the next cut.

This means the smallest effective "feature" that can be made is 22nm. For a typical MOSFET (field effect transistor), this size is usually the length of the gate. See the background of a MOSFET:

http://en.wikipedia.org/wiki/MOSFET

In practice, a smaller feature size means smaller transistors can be carved onto silicon. Smaller transistors require less voltage to operate and therefore, use less power. It used to be that smaller transistors also meant they were faster, but that has stopped being true since ~45nm.//

There is more. The problem is doping. They don't talk about that. How do they get each area to be a different material? Ok it is layered with all the layers and then the other layers are removed to make up the transistors. The chip is layered. The thing is, can reliably insulate the area between each transistor as they get smaller cuts?. What will be the failure rate?

The CEOs and other idiots think its just going to go down to 14nm between features. Coughs B.S. I am pretty sure they will lie either way.
 
It's more than JUST marketing... Intel has been following this development plan for a very long time. When each generation hits production the next generation is already far into development and planning in place for multiple generations beyond that. The last Intel R&D presentation I saw was in 2008 or so... and they were planning out to 2016 then (all the way from wafer size to feature size). There's a reason they're the premier chip company in the world.

The other thing that many don't realize is that when they say 22nm feature size that doesn't mean ALL features are that size... typically only a small subset of the die is actually pushing that size limit.

5nm feature size... is really amazing, and it will no doubt involve challenges... but I wouldn't bet against them.
 
@math1337 "Remember when we were supposed to have 10 GHz processors ?"
Yes, but then Intel / AMD discovered that 2 slower dual core CPUs outperformed 1 faster CPU. So the PR frequency race was pretty much abandoned and the focus shifted to real performance & power with multi-core CPUs
 


this was only related to p4 tech, obviously it was a pos, even the slower clocked single athalons were beating 3.6 ghz p4's
 
The reality is that Moore's Law was already busted long ago when chip designers decided to "cheat" by adding multiple distinct cores on the same die.

That trend, which is the standard and has been adopted by all pure play and vertically integrated companies, serves one philisophical purpose: to cheat Moore's law.

In actual fact, manufacturing multi-core per die CPUs is an incredible headache.

what happens when one of the 3 cores is bad? how do you bin that?


errors do not add, they multiply.
 
Status
Not open for further replies.

TRENDING THREADS

Latest posts

Moderators online

Share this page

COMPANY
RESOURCES
FOLLOW
Top Bottom