Intel identifies silicon dioxide successor for chips, due in 2007

[Asher]

This is some pretty amazing tech news that no one seems to be reporting...
http://www.siliconstrategies.com/art...questid=184485

Intel: ready with shift to high-k at the gate by 2007

By David Lammers, EE Times
Silicon Strategies
11/05/2003, 12:10 AM ET

AUSTIN, Texas -- Faced with intolerable levels of wasted power in its microprocessors, Intel Corp. said that in 2007 it will make the switch to a high-k gate insulator, reducing current leakage at the gate by at least 100 times.

At the 45-nm process node, Intel will make a double switch, replacing the tried-and-true silicon dioxide with an unidentified high-k insulator. At the same time, it will move from doped polysilicon for the gate electrode to two different metals, one for the NMOS and another for the PMOS transistors.

Nathan Brookwood, principal analyst at Insight64 (Saratoga, Calif.), said "this is a huge deal. The leakage problem has been threatening to slow down Moore's Law, and until now no one has been able to find a high-k dielectric that would not slow transistor switching speeds."

As silicon dioxide is thinned to about five atomic layers in order to turn on the transistor more quickly, electrons have tunneled through the oxide layer, causing wasted power during the "on" state. While the switch to a high-k material will reduce power leakage while the transistors are switching, it does little to improve power dissipation while the transistors are in the "off" state.

Yale University professor T.P. Ma, an authority on the gate oxide issue, said "it is encouraging that Intel has picked the materials it plans to use. This is a big shift. Of course, the industry still has to wonder what those materials are. But this announcement means the race is on" among the major semiconductor suppliers to switch to a high-k oxide.

Sunlin Chou, senior vice president and general manager of the technology and manufacturing group at Intel, said "we believe this is the first convincing demonstration that the industry can use these new materials. This gives us greater confidence that we can keep scaling transistors."

The search for a replacement for silicon dioxide has faced "serious roadblocks," Chou said. The high-k oxides, such as hafnium oxide, zirconium oxide, and others, have resulted in mobility degradation for carriers in the channel below the gate oxide. Moreover, there have been serious problems setting the threshold voltage, particularly for the PMOS transistors.

Intel and others have found that polysilicon is largely incompatible with the various high-k materials. That has forced a dual switch, replacing polysilicon as the gate electrode as silicon dioxide is phased out in favor of a high-k material. In order to tune the work function of the NMOS and PMOS electrodes, Intel will use two different metals, which Chou and others declined to identify.

Intel fellow Robert Chau, who led the gate oxide investigation at Intel's process development laboratory in Hillsboro, Ore., is scheduled to present an invited paper at the International Gate Insulator Workshop in Tokyo on Thursday that stops short of identifying which high-k gate oxide or gate electrode metals Intel will use. Intel said only that the high-k material has 60 percent greater capacitance than silicon dioxide.

The paper describes NMOS and PMOS transistors with physical gate lengths of 80-nm and an electrical oxide thickness of 1.4-nm (14 Angstroms), as measured at inversion. The threshold voltages of the devices were within Intel's targets, and drive currents and off currents were among the best reported to date, according to Intel.

Ken David, director of components research at the manufacturing group, said it is possible that Intel could switch gears at the 45-nm node, using a tri-gate transistor with silicon dioxide, or possibly with a high-k material, as the gate insulator. But if Intel sticks with the conventional planar transistors, it is committed to using the high-k gate insulator and metal gates that it has identified, he said.

Rather than grow the insulation layer as is commonly done with silicon oxide, Intel will use atomic layer deposition (ALD) equipment to deposit the high-k materials, he said.

--additional reporting by Rick Merritt

The jist of this is, for those of you non-tech inclined is as follows:
Current chips are wasting extraordinary amounts of power (mostly generating heat, which is why chips only get hotter). Each time the chips move to a new generation (ie, from 180nm transistors to 130nm, 130nm to 90nm, 90nm to 65nm, etc) the "leakage" problem gets exponentially worse. This is why upcoming 90nm chips like Intel's Prescott dissipate up to 100W of heat, compared to 30W just few years ago.

The main problem is we're reaching the end of life for current materials, they're just too inefficient. So semiconductor manufacturers have been looking for a successor for silicon dioxide to allow for continued growth in speed, shrinking in size, and lower power and heat requirements at the same time.

The fact that Intel has identified and found materials that are usable as a replacement, and 100 times better at reducing gate leakage than silicon dioxide, is a huge deal. No one else has found materials that'd work, and Intel's keeping theirs a trade secret.

[Asher]

CNet now has an article on it: http://news.com.com/2100-1006_3-5102...l?tag=nefd_top

Intel mulls metal over silicon for future chips
Last modified: November 4, 2003, 9:00 PM PST
By Michael Kanellos
Staff Writer, CNET News.com

Moore's Law is alive and well, but Intel is changing its basic semiconductor recipes to make sure it stays that way.

The Santa Clara, Calif.-based chipmaker is looking at revamping two fundamental elements of its transistors--the transistor gate and the gate dielectric--so its chips will continue to increase in speed and performance.

Currently, the gate, which controls whether a transistor is on or off, is made of silicon atoms while the gate dielectric, an insulating layer below the gate, is made of silicon dioxide. By making both out of metal, Intel will be able to clamp down on electricity leakage and other looming problems that could put a lid on improvement. In experiments, the new transistors are setting records on certain parameters, according to the company.

"We'd love to continue with silicon dioxide, but we can't do it because of leakage," said Ken David, director of components research in Intel's Technology and Manufacturing Group. "People keep running into these fundamental roadblocks."

Chips with metallic gates and metallic gate dielectrics (also called high-k dielectrics) may appear in 2007 with the 45-nanometer manufacturing process.

Semiconductor design is currently undergoing a major overhaul and prompting engineers and designers to incorporate new structures or materials into chips at a more rapid rate than ever before. "The way the industry has approached it is change one material at a time," David said. Now, semiconductor designers are being asked to incorporate two or more novel concepts every two years.

Some of these technologies are already coming to the fore. Intel has just started to make processors with strained silicon, a design convention that lets electrons move more rapidly, while IBM has already released a dual-core processor.

Other ideas on the drawing board include multiple gate transistors, controlling transistor voltage, replacing wires inside chips with optical fiber and carbon nanotubes.

The changes are largely necessary because of the unsavory consequences of news:link id="1014887">Moore’s Law, the famous dictum that states that the number of transistors on a chip double every two years.

Transistor count can be doubled because engineers can shrink the size of their transistors. Shrinkage, though, has made heat a major problem because millions of circuits are now crammed into small spaces where only a few hundred thousand transistors may have existed years before.

With the gate dielectric, thinness is an issue. The gate dielectric on chips coming out of Intel’s fabs next year will only be four to five atoms thick, David said. Thinning it further will cause additional leakage, or unintentional energy dissipation. Leakage can drain batteries and increase internal computer heat because more energy than should be necessary is required to animate these chips.

By switching to metal, leakage goes down because the chemical and physical properties of metal prevent electricity from escaping. With less leakage, chips can provide comparable performance on far less electricity, or it can run at a higher speed at the same energy levels.

As an added bonus, the gate dielectric layer can actually be thicker, which makes it easier to manufacture, but it will perform like a very thin traditional gate dielectric.

AMD has reported similar results in its published experiments.

Switching to metal gates and high-k gate dielectrics also eliminates phonon scattering. Increasingly, the atoms inside transistors are vibrating. Incorporating high-k gate dielectrics alone does not solve the problem.

"This slows down the mobility of electrons," David said. "The metal gate seems to act like a sink for this phenomenon."

David, however, declined to identify what metals Intel is experimenting with. AMD is working with nickel in its metal gates.

Intel is presenting the results of its research into metal gates at the International Gate Insulator Workshop taking place this week in Tokyo.

[Urban Ranger]

If you give up SiO₂, you might as well switch to GaAs or GaN.

[Lefty Scaevola]

Gallenium Vally does not have the same ring to it as Silicon Valley.

[Paul Hanson]

You'd get used to it.

[chequita guevara]

When mass produced diamond wafers are available (in the next few years), we can call it Diamond Valley. That has a much better ring.

[Asher]

I'd reckon diamond wafers would be a bit pricey.

[yavoon]

and not happening in the next few years either.

[Q Classic]

gotta love /.

[chequita guevara]

Originally posted by Asher
I'd reckon diamond wafers would be a bit pricey.

Not when they can be mass produced. You should read your Wired magazine more. Two American firms have learn how to produce gem quality stones for dollars a carat. The second firm is trying to build diamond wafers specifically for use in circuitry, and despite yavoon's claim, they're not that far from acheiving the level they want.

[alva]

I thought 'living' computers were going to be the next big thing...?

[The Viceroy]

UR, my thought exactly. I did a report on this problem back in 1996 for my degree .. so this problem has been known about for many many years.

As Asher points out, this problem is becoming more and more pronouced .. its truelly amazing we've gone as far as we have with current technology.

[mrmitchell]

Good news.

Except that they need to speed up release. 2007? That's four years away! By that time, processors will be putting out enough heat to cook a steak.

I'd reckon diamond wafers would be a bit pricey.

Artificially produced. They're just yet another string of carbon. It's already done in small amounts.

[Whaleboy]

So what kinda speed are we talking about by then?

[alva]

By that time, processors will be putting out enough heat to cook a steak.

I say a website a while ago who exactly did this, 'cept it was an egg

[Wraith]

--"Except that they need to speed up release. 2007?"

That is the big problem. Intel's already hitting the limits of watt/sq mm that can be handled with air-cooling. They're going to have to do something interim to be able to make it to the 45 node without resorting to Peltiers or watercooling.

Or drop the P4 line and try to ramp up the PMs...

Wraith
Chaotic Amorals have more fun

[Urban Ranger]

che,

They way they make artificial diamonds now is not the way wafers are made. Futhermore, making small diamonds is one thing, making large 12" wafers is quite another. Even 6 inchers are tough.


The Viceroy,

AFAIK, the advantage of using silicon is it can naturally form an insulator layer. Otherwise, its electric characteristics aren't all that great. If they are going to go the great length of depositing an external insulator layer, there really is no point of using silicon. Much better off using a far better material. That way, they can have much faster switching speed at the current 90nm process.


Wraith,

The thing I don't understand is, who needs a PC that fast?

[Asher]

Originally posted by Urban Ranger
The thing I don't understand is, who needs a PC that fast?

Off the top of my head...

• Gamers
  
• Desktop publishers
  
• Graphic artists
  
• Photographers
  
• Musicians/sound mixers
  
• People who go for multimedia desktops (real-time MPEG 4 encoding, etc)
  
• Scientists
  
• People building clusters
  
• 3D animators
  
• People who use CAD
  
• Software Developers
  

Not everybody needs the power of modern high-end computers, but a lot of people need that power and then some.

I'm totally oblivious to any point you were trying to make, though I doubt you thought enough about it to try to make a point.

[Skanky Burns]

Who doesn't want their PC to be as fast as possible?

[Urban Ranger]

Originally posted by Asher

• Gamers
  
• Desktop publishers
  
• Graphic artists
  
• Photographers
  
• Musicians/sound mixers
  
• People who go for multimedia desktops (real-time MPEG 4 encoding, etc)
  
• Scientists
  
• People building clusters
  
• 3D animators
  
• People who use CAD
  
• Software Developers
  

Gamers? No, most games are not computationally expensive. Graphic artists? No. Photographers, no. Musicians, doubly no. MPEG4 encoding, possibly, but you are much better of using a DSP. Scientists use supercomputers. The whole point of building PC clusters is each node does not have to be expensive. 3D animators, maybe. CAD is not computationally expensive, either. And what the heck do developers need fast computers? They much take coffee breaks!

[Urban Ranger]

Originally posted by Skanky Burns
Who doesn't want their PC to be as fast as possible?

About 90% of people who use computers.

[Kuciwalker]

Gamers? No, most games are not computationally expensive.

They would be, if they could be.

[Asher]

Originally posted by Urban Ranger
Gamers? No, most games are not computationally expensive. Graphic artists? No. Photographers, no. Musicians, doubly no. MPEG4 encoding, possibly, but you are much better of using a DSP. Scientists use supercomputers. The whole point of building PC clusters is each node does not have to be expensive. 3D animators, maybe. CAD is not computationally expensive, either. And what the heck do developers need fast computers? They much take coffee breaks!

What world do you live in?

You're out of your mind...

I would make another judgement of your intellect right here but this last post of yours does it for me.

[Asher]

Originally posted by Urban Ranger
About 90% of people who use computers.

My, what astounding reason.

So you would assert that 90% of people don't buy into the MHz myth? You would assert that 90% of people wouldn't care if they got a 1GHz vs 3GHz computer?



/me waves to UR out in left field.

[Q Classic]

if people didn't give a damn about mhz/ghz... we wouldn't have 3ghz chips today.

[Az]

precisely, q^3.

As a chemist, I am very much interested in what exactly are they using.

[Wraith]

--"They way they make artificial diamonds now is not the way wafers are made. "

It's a different material, why would the process be the same? They do seem to be quite close to making usable diamond wafers. The question is how long it will take to get that to full production in a factory.

--"The thing I don't understand is, who needs a PC that fast?"

Come, now, UR. You know that, no matter how fast the computer, Windows will always bloat sufficently to bog it down

--"CAD is not computationally expensive, either."

I can see you've never worked on a major CAD-based design. You might be surprised at what kind of requirements graphic artists can have, too (especially for more memory).

In any case, it's just like it's been for a while. Basic tasks won't need cutting-edge hardware. But the more power available, the more things people are going to try to do with it.

Wraith
Shift to the left! Shift to the right! Pop up! Push down! Byte! Byte! Byte!

[Tiberius]

I work in IT support and believe it or not, 90% (or maybe 75%) of people don't care for gigahertz. Of course, they care to have a new computer (because it's fancy to have one!), but all in all they care to have a comp fast enough to do its job. I have 486 computers working with MS-DOS !!!, and they do what they are supposing to do just fine.

Those who care about MHz and GHz are gamers, graphic artists, CAD developers, etc. I care for GHz for my home computer, because I like games. For office comps, I couldn't are less about speed. I care about stability however.

Edit: that being said, there will be always need for more computational power. I am dreaming about a StarTrek like holo-room.

[Q Classic]

to a point, you're right. people don't really care about ghz. why? because it's really quite abstract to them. memory means both HD space and RAM to the average non-techie.
gigahertz to them is just a measure of how new and "good" a computer is; they have no real idea what the difference is between an Athlon or a Pentium 4.

the fact is, though, that higher numbers sell. they always have. Joe User might not know which one's a better performer, but he'll probably think an Athlon 2800+ is better than an Pentium 4 2.5GHz chip--and if it's cheaper, he'll buy the 2800+, simply because its "better".

the real effect of this is that it pushes the chip companies to develop faster and faster chips, just to make their product the "best". it has very little to do with real performance.

if it did, you wouldn't've had AMD rebranding the Athlon 1.1GHz as an Athlon 1700+.

[Tiberius]

Originally posted by Q Cubed
the fact is, though, that higher numbers sell.

Absolutely right. Which doesn't mean that a 3 GHz machine running MS Word, Internet Explorer and a mail client isn't a total waste, in terms of spent money.

There always will be domains with an avid need for more power, but for an average user the current comp speed is already too much (and no, Asher is not an average user )