R&D: 15 nm Circuits Have Wires Just 150 Atoms Apart

[qubit]

The process of microchip miniaturization continues apace, with researchers at McGill University and Sandia National Laboratories having build a circuit that has two wires that are separated by a mere 150 atoms having built them at the incredibly tiny 15 nm (nanometres) level. Circuits of this size should bring all the usual benefits of reduced power and heat, along with greater functionality due to the larger transistor density. Of course, the challenges of working at this level are great. Dan Olds, an analyst at The Gabriel Consulting Group said of this research: "This kind of research also uncovers other potential problems arising from ever smaller shrinks. Getting to 15nm or 16nm will mean smaller and more powerful devices that are more energy efficient. But when we're talking about such a small scale, designing chips that can be mass-produced with decent yields is quite a challenge. There will also be challenges for the design of devices that will use these processors." Of course, the benefits are significant and therefore well worth the challenge: "Devices based on 15nm processes will pack more performance and functionality into much smaller form factors. Functions that used to take two or more chips will be accomplished by one transistor-jammed processor." Of course, Intel, the world leader in chip manufacturing is also researching manufacturing at this process node, but there isn't much information available about it at this time.

View at TechPowerUp Main Site

 

[radrok]

Now THIS is small just think about the benefits of such fab node applied to CPUs... We can already reach insane frequencies on air with the 32nm, 22nm is almost ready to show us new rooftops! Exciting times are ahead for sure

 

[dogchainx]

Its 2011. I want my damn 10GHz air-cooled CPU.

 

[Benetanegia]

Its 2011. I want my damn 10GHz air-cooled CPU.

Wow, that article is hilarious in 2011. And the comments? Bwahahahahaha.

My favourite comment is the first one:

If 10 GHz is the best that Intel can do by 2011, AMD or somebody else is going to eat their lunch. Intel better pick up the pace if they want to remain dominant.

BTW you want your 10 Ghz CPU? I want my 128 Ghz one! Or whatever Rob had taken, he damn sure was high when he wrote that.

 

[EpicShweetness]

Its 2011. I want my damn 10GHz air-cooled CPU.

If you think about it though a Intel 3960X with 3.3GHZ across all 6 cores is a combined 19.8GHZ. So maybe they were thinking along the lines of a single core. Which became outdated with 64bit processing and the limitations of silicon.

 

[Trackr]

Its 2011. I want my damn 10GHz air-cooled CPU.

That was before Dual-Cores existed.

You should have compared an i7-3960x to a Pentium 4 1.0Ghz.

With architecture, cache size, etc etc. improvements, plus 6 cores and 3.3x core clock..

It should be equivalent to a 150Ghz Pentium 4, in terms of performance.

I'd say Moore's Law is being proven correct.

 

[NC37]

Not necessarily before dual cores existed. GPUs were going multicore back then. But on the CPU front, yeah, we had dual CPU machines but that was it.

If they never hit the wall on speed, literally we prolly wouldn't have multicore in CPUs. Multicore systems would possibly exist in servers and even then it would be mostly multi sockets. 64bit, may or may not have happened as fast. Without AMD I'd say it would have happened later. Intel slapped 64 together in a hurry to confront AMD after it was clear things were moving to 64bit. If there is no pressure I don't see Intel in a hurry to innovate.

Same with any company. We saw the same with nVidia till ATI got their act together. They sat on the G92 for years. Same with DX10.1 and Tesselators. ATI did it, NV pressured devs not to use it and then finally when ATI starts coming back, they whip out the tech with DX11.

 

[Fx]

If you think about it though a Intel 3960X with 3.3GHZ across all 6 cores is a combined 19.8GHZ. So maybe they were thinking along the lines of a single core. Which became outdated with 64bit processing and the limitations of silicon.

except that no one knows how to write code that is optimized to proficiently use the cores to perform at anywhere near that combined level of performance

 

[micropage7]

15nm = more complicated
more complicated could mean more fail percentages
15nm could mean better performance with lower power consumption
its kinda hard to imagine. we just had 32nm just few years ago then we hit 15nm (although maybe it need time before wide applied)

 

[de.das.dude]

Wow, that article is hilarious in 2011. And the comments? Bwahahahahaha.

My favourite comment is the first one:





BTW you want your 10 Ghz CPU? I want my 128 Ghz one! Or whatever Rob had taken, he damn sure was high when he wrote that.

intel more fail than meets the eye.

 

[radrok]

I don't think we will see this lithography soon because Intel will probably dominate the x86 market without a strong AMD that forces them to innovate... Unless ARM proves itself a threat if it enters this business seriously or Intel itself will focus more on developing a SoC to challenge ARM on its field.
Of course Ivy Bridge is at the door and as I said exciting times are ahead, even though the situation of the desktop CPUs will probably hit a slowdown

 

[XoR]

my crappy i3 (from beginning of 2010) is 15x faster than processors the had in 2000 (~PIII 1GHz) and more than 6x faster in single core perfrormance. Intels 10GHz figure is because they had high hopes for NetBurst and it is understandable because they were able to double clock speed with this arch.

 

[_JP_]

If you think about it though a Intel 3960X with 3.3GHZ across all 6 cores is a combined 19.8GHZ. So maybe they were thinking along the lines of a single core. Which became outdated with 64bit processing and the limitations of silicon.

I remember stores here trying to sell HT chips and early dual-cores with that line of thought. A dual-core at 2.2GHz doesn't equal a single-core at 4.4GHz. It never worked that way.

That was before Dual-Cores existed.

You should have compared an i7-3960x to a Pentium 4 1.0Ghz.

With architecture, cache size, etc etc. improvements, plus 6 cores and 3.3x core clock..

It should be equivalent to a 150Ghz Pentium 4, in terms of performance.

I'd say Moore's Law is being proven correct.

First of all, there is no such thing as a 1GHz Pentium 4. Unless you downclock it.
Second, clock-by-clock comparison, with a difference of generations that big, is a dumb thing to do in terms of application performance. Of course there will be a huge difference, but it only makes sense to compare it with same generation processors. It makes much more sense to compare processing performance using a standard benchmark to evaluate a single unit of measurement. LINPACK/LAPACK are good methods. It will still show a huge difference, but it's a measurement.

except that no one knows how to write code that is optimized to profeciently use the cores to perform at anywhere near that combined level of performance

Since a 19.8GHz processor doesn't exist (with x86 architecture) we can only guess. But there are programs written for multi-core processors.

intel more fail than meets the eye.

You're becoming an annoying fanboi as each day goes by.

my crappy i3 (from beginning of 2010) is 15x faster than processors the had in 2000 (~PIII 1GHz) and more than 6x faster in single core perfrormance. Intels 10GHz figure is because they had high hopes for NetBurst and it is understandable because they were able to double clock speed with this arch.

That doesn't make sense. Off the top of head assumption is off the top of head.
They can be faster overall to complete a task, but that's related to more stuff (architecture, cache, RAM-CPU bus speed, NB improvements) rather than just clock speed. I think even single core performance is higher than only six times that of a Pentium 3 at 1GHz. Like all manufacturers, high hopes for a new product is a completely normal behavior.

 

[Edgarstrong]

Its 2011. I want my damn 10GHz air-cooled CPU.

Thanks for the link

I found the following comment pretty amusing:

"If 10 GHz is the best that Intel can do by 2011, AMD or somebody else is going to eat their lunch. Intel better pick up the pace if they want to remain dominant."

 

[Lipton]

"Of course, Intel, the world leader in chip manufacturing is also researching manufacturing at this process node, but there isn't much information available about it at this time."

They're going 14 nm, not 15 nm. The Intel slide referred to is old (over a year).

http://www.intel.com/content/www/us/en/silicon-innovations/intel-tick-tock-model-general.html

http://www.anandtech.com/show/4345/...-architecture-group-14nm-airmont-atom-in-2014

 

[FordGT90Concept]

I remember stores here trying to sell HT chips and early dual-cores with that line of thought. A dual-core at 2.2GHz doesn't equal a single-core at 4.4GHz. It never worked that way.

With symmetrical multithreading, it can get pretty close (maybe a 10% overhead cost). Problem is, there aren't many workloads substantial enough to bother.