Intel Developer Forum 22nm Facts

HardOCP News

[H] News
Joined
Dec 31, 1969
Messages
0
Intel, in conjunction with the world’s first demonstration of chips built on 22nm process technology, has posted a fact sheet today (.PDF) covering the highlights of the Intel’s upcoming 22nm high-k metal gate process technology:

SRAMs are used as test vehicles to demonstrate technology performance, process yield and chip reliability prior to ramping processors and other logic chips that will use the given manufacturing process.

Intel is now in full development mode on 22nm and on pace to continue the company’s “tick- tock” model into the next generation.

The 22nm test circuits include both SRAM memory and logic circuits to be used on 22nm microprocessors.

SRAM cells of 0.108 and 0.092 square microns function in an array totaling 364 million bits. The 0.108 square micron cell is optimized for low voltage operation. The .092 square micron cell is optimized for high density and is the smallest SRAM cell in working circuits reported to date.

The test chip packs 2.9 billion transistors, at approximately double the density of the previous 32nm generation, in an area as small as a fingernail.

The 22nm dimensions are patterned with exposure tools using light with a wavelength of 193nm, a remarkable testament to the ingenuity of Intel’s lithography engineers.

This 22nm technology continues to deliver the promise of Moore's Law: smaller transistors, improved performance/watt and lower cost per transistor.
 
22nm is already pretty crazy, 10nm will be very good there will be some very advanced processors and memory available at that time, but I wonder how long this will take to get released, and what will we do after that point? If there is say, a 10nm wall, and we get there around 2020 or something... What the hell are we going to do? We've come a long way in the past 20-30 years but think about the rest of humanity... Some innovation required I guess.
 
Once they hit 10nm they will have nowhere else to go imho.

You're hitting the limits of what silicon is physically capable of doing. When you have a trace that is separated by a single atom over to another... yeah. That's pretty much a short. 10nm is effectively the smallest you may be able to build with the materials used now. There's also the thermal limits and speed limits... We're getting close to hitting all three now with Silicon.

I don't think it'll be the end of things if we manage to hit 10nm and other materials don't end up working as well. We've already seen why that is: Multiple cores. The thought that we're going to "need" 8 or 16 cores as a low end system does irk me a little, though. By now, I still think it's safe to say we effectively hit a plateau for performance and specs. A lot of systems are still "only" dualcore around 2.0Ghz and more than enough for a good 80-90% of people out there.

Diminishing returns. We're gonna have to get used to it until something fundamentally new becomes the way things are done. Until then, it's more cores.
 
That's the theorized CMOS barrier. However, there are alternatives like optoelectronic processing, and graphene based computing :)

No need to fear, Intel is all about R&D.

Oh, I'm not worried about it and I foresee Graphene as the next logical step to chip making. the High-K technology has really come through for Intel and frankly I think it saved their bacon.
 
Diminishing returns. We're gonna have to get used to it until something fundamentally new becomes the way things are done. Until then, it's more cores.

QFT. We've apparently hit the limit for reliable maximum clock speeds, and pretty soon it will be how small we can physically make the transistors. Multicore will pick up the slack until another breakthrough is made which either allows for higher clock speeds or even smaller transistors.
 
You're hitting the limits of what silicon is physically capable of doing. When you have a trace that is separated by a single atom over to another... yeah. That's pretty much a short. 10nm is effectively the smallest you may be able to build with the materials used now. There's also the thermal limits and speed limits... We're getting close to hitting all three now with Silicon.

I don't think it'll be the end of things if we manage to hit 10nm and other materials don't end up working as well. We've already seen why that is: Multiple cores. The thought that we're going to "need" 8 or 16 cores as a low end system does irk me a little, though. By now, I still think it's safe to say we effectively hit a plateau for performance and specs. A lot of systems are still "only" dualcore around 2.0Ghz and more than enough for a good 80-90% of people out there.

Diminishing returns. We're gonna have to get used to it until something fundamentally new becomes the way things are done. Until then, it's more cores.

When you can no longer expand sideways you reach for the sky:)

Can't remember the pics but there has been 8+ layer stacked chips prototyped for some time. Obvious issue with cooling the middle layer will have to be solved first of course.
 
That's alotta transistors.

I figure, with some cache on each and all the interconnect logic, you could put

16,000 Motorola 68000 chips

... on that die, running at 4Ghz that would be 8 MILLION times faster than the original Mac/Atari/Amiga computers. Nothing like a 16K core chip...wooohooo. :eek::p:cool:
 
Well at 10nm getting smaller is not going to be easy, I guess the next direction follows along the GPU path of massively parallel architecture with hundreds to thousands of cores in small bunches splitting a single task into a hundred pieces.

I could see a system where you have say a 64 bit computer that has 64 core hubs and maybe 8/16/32/64 cores that do individual I/O for shared calculations. If I have read it correctly this seems to be about where nvidia is heading with the gt300 GPGPU core.
 
22nm is already pretty crazy, 10nm will be very good there will be some very advanced processors and memory available at that time, but I wonder how long this will take to get released, and what will we do after that point? If there is say, a 10nm wall, and we get there around 2020 or something... What the hell are we going to do? We've come a long way in the past 20-30 years but think about the rest of humanity... Some innovation required I guess.

You make it sound like humanity is coming to an end... :)
 
time to start to start investing in a pay day by filing for patents on things like making chips from graphene or using fiber optics instead of electrical circuitry. :D

I'm sure Intel has had this barrier in their sites for a long time and have a good idea about where they will go from their. The question is who has the better solution. AMD, Intel or someone else?
 
The future of CPU's are ridiculously small (measured in micrometers) cubes with a discrete core on each side and interconnections to other cores a/o a primary interface board in the center of all 6 sides. These will be able to be connected in near limitless numbers and will be the ultimate in multi-threading power. Thousands of cores will run at the same time on a typical PC, and super-computers will have millions of cores. The cubes and cores will be enclosed in a cube filled with dielectric liquid and will include an internal system to pump the liquid through the cube and out to an attached cooling plate that can have an HSF attached to it.
 
Interesting stuff again. Hopefully they will stay on track shrinking those transistors. ;)
 
The future of CPU's are ridiculously small (measured in micrometers) cubes with a discrete core on each side and interconnections to other cores a/o a primary interface board in the center of all 6 sides. These will be able to be connected in near limitless numbers and will be the ultimate in multi-threading power. Thousands of cores will run at the same time on a typical PC, and super-computers will have millions of cores. The cubes and cores will be enclosed in a cube filled with dielectric liquid and will include an internal system to pump the liquid through the cube and out to an attached cooling plate that can have an HSF attached to it.

yes, I believe you are correct, 3D data array is the way forward as transistors are not going to get a hell of a lot smaller.
 
Don't forget, you can only go so wide (cores) before you run into heat and power issues. A 3D chip is going to have even worse (melted core) heat issues. You can still make them wide, but you're going to hit a practical power limit (already getting close with video cards.) Either we change things radically (optical, quantum, other?) or we get comfortable with our existing computing abilities in the next 10 years. Not to mention the whole multithreading issue. Maybe software optimization will be the next big thing.
 
Back
Top