Intel announces 'game-changing' 3D transistors, demos Ivy Bridge

By on May 4, 2011, 7:16 PM

Intel has announced a significant breakthrough in chip design that opens the door to smaller, faster, more power-efficient processors. First disclosed by Intel in 2002, the company's 3D tri-gate transistors will replace the traditional two-dimensional transistors in the chipmaker's upcoming 22nm Ivy Bridge processors -- the successor to Sandy Bridge.

Unlike conventional planar transistors that lay flat, tri-gate transistors use a three-dimensional fin that stands vertically from the silicon substrate. Orienting the transistors upright presents several benefits. For starters, Intel can cram more transistors into less space, which will be incredibly valuable as fabrication tech shrinks to 22nm and beyond.

In addition to doubling transistor density, the new design exposes three sides of the transistor to a gate instead of one. Transistors carry an electrical signal while gates control that flow by turning the current on and off. By increasing the surface area of the transistor that is touching the gate, it gains additional control over the flow of electricity.

The video above does an excellent job of breaking things down. All you really need to know is that the company expects 22nm chips with tri-gate transistors to reduce power consumption by more than 50% with 37% more performance than Intel's existing 32nm technology. New 22nm tri-gate wafers should be cheaper to produce, as well.

"The performance gains and power savings of Intel's unique 3-D Tri-Gate transistors are like nothing we've seen before," said Intel's Mark Bohr. "This milestone is going further than simply keeping up with Moore's Law…We believe this breakthrough will extend Intel's lead even further over the rest of the semiconductor industry," Bohr continued.

During the same event, Intel demonstrated its new Ivy Bridge processors in action. The company showed three different form factors performing separate actions: a notebook playing an HD video, a web server launching a webpage, and a desktop playing a racing game. Nothing too exciting, but it's a sneak peek at Ivy Bridge nonetheless.




User Comments: 40

Got something to say? Post a comment
Guest said:

Nice. When they come out lets see the performance benchmarks for good evidence.

Tekkaraiden Tekkaraiden said:

Fascinating, best use of 3d technology to date.

Mizzou Mizzou said:

If I'm following this correctly, we're going from one to three connections per gate which means eight possible states per transistor instead of two. If so, that would have far reaching implications for computer design principles which have up to this point been based on two state boolean algebra principles. Need to let this one sink in for a while, it's game changing to say the least.

Win7Dev said:

I hope that prices go down since "New 22nm tri-gate wafers should be cheaper to produce".

Staff
Steve Steve said:

If I'm following this correctly, Intel has developed a shrink ray? That will win them the coolest tech of 2011 award for sure.

Guest said:

And that shrink ray runs off there new 22nm 3D transistor chip!

mpsteel said:

Tekkaraiden said:

Fascinating, best use of 3d technology to date.

I didn't know the third dimension was technology?

p.s its not a TV

ramonsterns said:

mpsteel said:

Tekkaraiden said:

Fascinating, best use of 3d technology to date.

I didn't know the third dimension was technology?

p.s its not a TV

Does this mean I won't need glasses to use it?

red1776 red1776, Omnipotent Ruler of the Universe, said:

And that shrink ray runs off there new 22nm 3D transistor chip!

wait! how can you run the shrink ray off the 3D chip if you need the shrink ray to invent the ....oh you minx!

Guest said:

@Mizzou

You do not understand that correctly. They just called it tri-gate. It does not mean there are three gates on the transistor.

red1776 red1776, Omnipotent Ruler of the Universe, said:

Its been widely theorized that around 14nm is the shrink limit to transistors while maintaining a reliable signal. It would appear that before this we were close the end of Moore's law. Any educated guesses at how much/longer this will extend it?

TeamworkGuy2 said:

cool, but if this 3D CPU discovery was copyrighted by Intel, which any logically company would do, what does this mean for AMD..?

Tekkaraiden Tekkaraiden said:

Yes I know I was trying to make a funny, sounded funnier before I typed it.

dividebyzero dividebyzero, trainee n00b, said:

Will be interesting what kind of thermal limits are imposed by the increased transistor density.

Working 22nm silicon in May would auger well for the usual January (2012) launch period for Intel's mainstream platform.

Its been widely theorized that around 14nm is the shrink limit to transistors while maintaining a reliable signal. It would appear that before this we were close the end of Moore's law. Any educated guesses at how much/longer this will extend it?

Probably depends on:

1. If the next technology is ready for commercial use, and

2. How aggressive the competition between Intel, AMD and ARM is, and

3. Whether x86 is still the dominant ISA (or whether parallelization and GPGPU is a player in the future of client computing)

Fab building/refurbishment and lithigraphy (and other process) tools will require huge monetary investment afaia once foundries hit 16nm-11nm, so it probably stands to reason that unless nanotubes/graphene/whatever are ready to go by 2015 then the process nodes slow as return on investment and new etching tools/fabs reaches equilibrium.

red1776 red1776, Omnipotent Ruler of the Universe, said:

cool, but if this 3D CPU discovery was copyrighted by Intel, which any logically company would do, what does this mean for AMD..?

It means that AMD comes up with the same thing under a slightly different process, under a different name, like Intel did when AMD came out with Hyper-Transport. Then Intel sues AMD over the course of the next decade.

Or..."CPU" and Intel become interchangeable and low end CPU's start at $200. Then AMD CPU division gets thrown on the scrap heap right on top of Cyrix/IBM.

Just being a pessimist I guess....I think Intel is more afraid of the litigation from being a 'monopoly' than AMD actually surviving.

Guest said:

37% and i thought SNB was an improvement. <grumble> if these aren't socket 1155 I will stand on that tiny little !"£$%

dividebyzero dividebyzero, trainee n00b, said:

It means that AMD comes up with the same thing under a slightly different process, under a different name, like Intel did when AMD came out with Hyper-Transport.

Unlikely.

The big drawback from selling Global Foundries to the guys from Abu Dhabi is that AMD now is at the tender mercies of whatever GloFo come up with (or don't). While AMD is a customer of GloFo, it isn't the only one -and possibly not even the most lucrative one.

AMD is fabless- they go where GloFo says they'll go.

Another way to look at is (if you're a cynic that is):

GloFo don't keep pace with Intel -> Intel retains process lead -> Intel retains or increases marketshare -> AMD's stock falls -> Mubadala/ATIC (GloFo's owners/AMD shareholder) snap up enough of the remaining AMD shares at decreased value and launch takeover

EDIT-

@Red1776

AMD won't go under. Their cross licensing x86-64 with Intel and IP will keep them viable. The only thing likely up for debate is whose pockets AMD's profits are likely to end up in.

red1776 red1776, Omnipotent Ruler of the Universe, said:

Unlikely.

The big drawback from selling Global Foundries to the guys from Abu Dhabi is that AMD now is at the tender mercies of whatever GloFo come up with (or don't). While AMD is a customer of GloFo, it isn't the only one -and possibly not even the most lucrative one.

AMD is fabless- they go where GloFo says they'll go.

Another way to look at is (if you're a cynic that is):

GloFo don't keep pace with Intel -> Intel retains process lead -> Intel retains or increases marketshare -> AMD's stock falls -> Mubadala/ATIC (GloFo's owners/AMD shareholder) snap up enough of the remaining AMD shares at decreased value and launch takeover

....well....at least they wont be fabless anymore!

Grim

so what? let them have the ultra low end to keep non-monopoly status? or does Intel get busted up as well?

The big drawback from selling Global Foundries to the guys from Abu Dhabi is that AMD now is at the tender mercies of whatever GloFo come up with

This may be naive, but doesn't AMD design the architecture and go to Glo-Flo with schematics and say "manufacture this"?

dividebyzero dividebyzero, trainee n00b, said:

See my edit above.

Where AMD (or Intel itself) ends up in the market is likely dependant upon whether x86 is still needed down the track. If RISC/parallel processing makes the leap then it's not beyond the realms of possibility that virtually anyone able to stump up for an ARM licence or similar could be a player in future. Quite the opposite of the virtual hegemony we have now.

Whether RISC has what it takes to satisfy the market is another matter. That depend on a concerted act of will from people that aren't Intel or AMD.

Jurassic4096 said:

x86 isn't going anywhere.

Lokalaskurar Lokalaskurar said:

TeamworkGuy2 said:

cool, but if this 3D CPU discovery was copyrighted by Intel, which any logically company would do, what does this mean for AMD..?

Don't worry, AMD will think of something, just like the good ol' days. Besides, they still have some access to some of Intel's patents, so they are not out of this yet,

red1776 red1776, Omnipotent Ruler of the Universe, said:

So with that kind of voltage control,voltage reduction, and near zero leakage. Could these things be 5+Ghz right out of the gate as well? (pun intended)

Am I understanding this correctly?

dividebyzero dividebyzero, trainee n00b, said:

As I understand it, Intel seem to be playing for clocks not to dissimilar to Sandy Bridge and a much lower TDP, rather than similar (~95w) and huge core frequency. This I think would make sense if Intel are looking at 8+ core parts (consumer desktop),ease binning requirements and scaling for entry level/mainstream quad core (I don't think there are supposed to be dual core IB's). Keeping clockrate in check to a degree will also allow Intel to align similar clocked parts -and their friendlier TDP's for the mobile sectors, not to mention keeping some differentiation between IB and Sandy Bridge-E -which is now probably odds-on to receive consumer versions of the 8 core Xeon E5's already publicised.

So, I guess a lot will come down to how well IB comes out of the stove- both at stock, and turbo mode.

red1776 red1776, Omnipotent Ruler of the Universe, said:

I didn't ask that correctly.

would the aforementioned attributes of the tri-gate afford huge OC's? or does the qualities of the surrounding materials keep the frequency capabilities to about what they are now? I am trying to get a handle on what impact this is going to have. In other words , is this the big breakthrough that has been talked about for years, or do they keep working on carbon nanotubes?

dividebyzero dividebyzero, trainee n00b, said:

Ah.

Yes

Faster switching transistors @ lower voltage = bigger clocks.

From Anand:

[link]

And the Intel pdf (here).

How far Intel go is likely dependant upon how hard they have to go. Since approximately 20% of 2600K's are 5.0-5.7GHz capable outright, and could have been effectively binned at 4+GHz stock, it isn't unreasonable to assume IB is capable of that and then some.

Guest said:

Can just use the shrink ray on transistors/die directly? That'll get the process size down quick...

Jurassic4096 said:

So much for that Bulldozer upgrade.

Jurassic4096 said:

red1776 said:

I didn't ask that correctly.

would the aforementioned attributes of the tri-gate afford huge OC's? or does the qualities of the surrounding materials keep the frequency capabilities to about what they are now? I am trying to get a handle on what impact this is going to have. In other words , is this the big breakthrough that has been talked about for years, or do they keep working on carbon nanotubes?

Imagine going from Conroe straight to Nehalem, and some, with Penryn never being in the picture. With this new technology, Intel is now ~3 years ahead of AMD.

stewi0001 stewi0001 said:

so what should I do guys for my computer overhaul AMD or intel?

Arris Arris said:

The video does give a nice dumbed down overview. But then you can't help think "A decade to think of that...?".

CamaroMullet said:

stewi0001 said:

so what should I do guys for my computer overhaul AMD or intel?

Guess that depends when you want to build it. Sandybridge's are pretty fing awesome right now. I wouldn't get anything until next year. AMD if you are on a super budget.

Guest said:

WOW and to think my AMD 965 still crushes everything i throw at it. This should be intresting and massive overkill so count me in.

red1776 red1776, Omnipotent Ruler of the Universe, said:

Imagine going from Conroe straight to Nehalem, and some, with Penryn never being in the picture. With this new technology, Intel is now ~3 years ahead of AMD.
It would be like handing the Waltons' an iPhone.

dividebyzero dividebyzero, trainee n00b, said:

....And a cell tower for the mountain.

They'd get proficient fairly quickly if they texted "Goodnight" to each other at the end of every show.

princeton princeton said:

Guest said:

WOW and to think my AMD 965 still crushes everything i throw at it. This should be intresting and massive overkill so count me in.

Download sony vegas or Maya and you'll be retracting that statement.

dividebyzero dividebyzero, trainee n00b, said:

The video does give a nice dumbed down overview. But then you can't help think "A decade to think of that...?".

I'll add some lateral thinking into the equation.

Intel and Micron are already in the process of incorporating FinFET (so-called 3-D transistor) into NAND

High density memory (RAM) using TSV now seems a reality (i.e. Samsung along with Intel et al)

Intel has been exploring how to get ultra-low voltage DDR incorporated into CPU's for 4+ years.

GPGPU (Larrabee/Knights Corner/Ferry), APU's (AMD's Fusion) and CPU's with IGP are all memory dependant -if not memory constrained. It doesn't seem unreasonable to assume that Intel is looking at adding DDR (or GDDR) directly to CPU package. I would also assume that the processor having access to RAM on-die is sure going to add to both bandwidth availability and a huge decrease in latency.

I would think that having access to 1-2Gb of high speed RAM has the potential to be somewhat of a game changer for CPU's with (and probably without) integrated graphics- especially in a mobile/ultraportable package.A next-gen ultra low voltage Atom seems to be high on Intel's list of things-to-do for the mobile space. Adding in a reasonable IGP an integral memory stack seems like a fairly neat solution if power usage/heat generation can be kept low enough.

Maybe all this tech is unconnected.....but I'm thinking that it's probably not. Just a matter of if I'm joining up the dots correctly. If so, then this 3D/FinFET tech allows for a reasonable shot at a much more complete SoC (system on a chip).

Food for thought ?

red1776 red1776, Omnipotent Ruler of the Universe, said:

Download sony vegas or Maya and you'll be retracting that statement.

Right...

or try a little rendering in 3DSMax while your at it.

Maybe all this tech is unconnected.....but I'm thinking that it's probably not. Just a matter of if I'm joining up the dots correctly. If so, then this 3D/FinFET tech allows for a reasonable shot at a much more complete SoC (system on a chip).

Food for thought ?

That is where this is heading. What it appears that it has solved is the signaling problem below the (up until now) theoretical limit of 16nm transistors. Intel announced a roadmap that spelled out 14nm (process 1274) for 2013, and 10nm (process 1272) for 2015. with ability to add as many 'stabilizing' or 'control' fins through the gate, and as you say access to 2Gb 's of memory 'on die', does this mean that the on die memory will be the 'active' memory and off chip memory will be 'storage' until the process gets refined and we see 4,8,16,32GB's on die?

would it also be a fair assumption that with these running at sub 1.0V even the substrate/depletion area will be able to shrink with it?This seems to be one of the rare game changing breakthroughs that has no downside or comprimise

dividebyzero dividebyzero, trainee n00b, said:

...access to 2Gb 's of memory 'on die', does this mean that the on die memory will be the 'active' memory and off chip memory will be 'storage' until the process gets refined and we see 4,8,16,32GB's on die?

The memory will certainly be active. IMO (and from what I've read) the prime candidate is GDDR to feed the graphics. Having the vRAM on package has a lot of plus side (latency, dedicated graphics memory), and I sincerely doubt that the options for volatile memory on-die wouldn't include system RAM.

Storage RAM on die would take some considerable time - it doesn't however discount an on(mother)board implementation of Intel's Turbo Memory (Braidwood) which was shelved (albeit as an add-in card/motherboard slot)

would it also be a fair assumption that with these running at sub 1.0V even the substrate/depletion area will be able to shrink with it?This seems to be one of the rare game changing breakthroughs that has no downside or comprimise

I think that is well underway from the articles I've seen. Hand-in-glove as it were.

bavon said:

Where can I buy the shrink-ray?

Trillionsin Trillionsin said:

COOOOOOOL!!!!!

Load all comments...

Add New Comment

TechSpot Members
Login or sign up for free,
it takes about 30 seconds.
You may also...
Get complete access to the TechSpot community. Join thousands of technology enthusiasts that contribute and share knowledge in our forum. Get a private inbox, upload your own photo gallery and more.