RSSIntel Reinvents the Transistor
By Scott M. Fulton, III | Published January 27, 2007, 12:01 AM
In a presentation to exclusively invited reporters Friday morning, Intel announced a breakthrough development in microprocessor manufacturing that may be given historical significance in decades to come: the discovery of a new molecular compound material that will replace silicon dioxide in microprocessors using 45-nm and smaller lithographies.
It is what both wide-eyed engineers and anxious executives have described as the "Holy Grail of semiconductor technology," and Friday morning Intel revealed it has developed working 45-nm processor samples running Microsoft Windows Vista, Mac OS X, Linux and other operating systems, where this material - a compound based on the element hafnium, atomic number 72, a frequently occurring impurity in zirconium typically found in fake diamonds - serves as the dielectric gate between the current source and the current drain.
With the hafnium material serving as the gate, Intel will then replace the polysilicon electrode layer with a metal electrode, the exact alloy used here also being kept secret. As a result, transistors for 45 nm semiconductors starting with Intel's Penryn family will be fabricated at half the size of those used in today's 65 nm Core 2 processors. At the same time, transistor switching power can be reduced by as much as 30%, while still obtaining a performance improvement of as much as 20%. And current leakage at the gate will be reduced by a factor of 10.
How big of a page has been turned here, really? Since Intel announces advancements several times a year any more, what makes this one substantive beyond the typical hyperboles reserved for press releases?
Intel's current processor roadmap leaps between processor technology families every two years. We saw the latest leap just last summer, from the last of the Pentium D dual-core processors at 90-nm, to the Conroe/Merom/Woodcrest series at 65-nm. But while this "high-k + metal gate" (HK+MG) development does play into Intel's planned leap to the Penryn architecture, Friday's revelation literally marks only the start of a second era in metal oxide semiconductor production.
A diagram depicting the differences between a first-generation MOS transistor and an HK+MG transistor to be used in Intel's Penryn 45 nm CPUs. (Courtesy Intel)
Since the 1960s, gates have been made using silicon dioxide (SiO2), and electrodes have been made with polysilicon. Substrates are the third major material; historically Intel has used silicon or silicon germanium (SiGe), and the industry has experimented with gallium arsenide. HK+MG replaces two of the three compounds used in semiconductor transistors since the 1960s.
"These are not laboratory devices," Intel's director of process architecture and integration, Mark Bohr, announced Friday morning. "These are not just research results. We've actually made these transistors in a fully-integrated 45-nm CMOS process flow. We have high-K metal gate PMOS transistors [positive flow] and NMOS transistors [negative flow], both providing higher performance than the previous generation and lower leakage than the previous generation. This integrated process flow also meets our reliability requirements and is manufacturable in high volume."
It may not be a feat that goes unanswered by Intel's chief competitor. Last month, AMD, along with manufacturing partner IBM, announced at a materials conference the discovery of an ultra-low-k dielectric compound that those companies will use in the development of new interconnects in processors at 65-nm and lower lithographies.
"I don't want to underestimate the feat that these companies have achieved," Jim McGregor, principal analyst at In-Stat, told BetaNews. But HK+MG replacing the silicon gate with metal, he said, "is very, very significant because there's been significant issues with some of the early high-k [materials], in terms of the crystalline structure deterioration, as well as the fact that they're dealing with different types of materials, trying to bond them together [and] work together for a very long time with very high performance."
45-nm is the transition point, McGregor believes, where semiconductor manufacturers must change the materials and processes they've been accustomed to, simply because the same old materials don't behave the same way once they're so small. As Intel explained, as a transistor is made smaller, its gate dielectric layer naturally becomes thinner. Toward 65-nm, this hasn't been a bad thing; the thinner gate has led to an increase in the gate field effect, which is desirable. But over time, after everyday use, a depletion area emerges between the gate and the electrode, in effect thickening the gate area, increasing the level of current necessary to implement a switch, and degrading the overall "on" current.
Replacing silicon dioxide with the hafnium compound lets Intel go ahead and thicken the gate layer, which in turn naturally increases the amount of "on" current. Meanwhile, the replacement of polysilicon with a metal electrode completely eliminates the depletion of electrode area over time.
The "k" in the term "high-k dielectrics" refers to a material's relative ability to hold an electric charge. In electrodynamics, it's actually written with the Greek lower-case letter kappa. The "kappa," if you will, of silicon dioxide is about 3.9. That number means nothing until you compare it to that of hafnium compounds. Intel has not revealed whether the compound it discovered is based on hafnium silicate or hafnium oxide. The "kappa" for pure hafnium silicates has been observed in a range between 15 and 25, while pure hafnium oxides have been observed at 40. Intel is likely using an "impure" compound for the sake of structural integrity; laboratory tests on hafnium oxides yielded observed "kappa" of over 16. So it's quite possible that Intel's new compound may be as much as four times more efficient than silicon dioxide at holding a charge.
Next: Why Intel's enthused about high-k and AMD about low-k
"...the exact alloy used here also being kept secret"
for about one week.
Score: 0
|Great article. Very informative.
One point. Intel is enthusiastic, not "enthused." Please. This is not a word.
Thanks.
Score: 0
|The verb enthuse is not well accepted. Its use in the sentence The majority leader enthused over his party's gains was rejected by 76 percent of the Usage Panel in the late 1960s, and its status remains unfavorable: the same sentence was rejected by 65 percent of the Usage Panel in 1997. This lack of enthusiasm for enthuse is often attributed to its status as a back-formation; such words often meet with disapproval on their first appearance and only gradually become accepted over time. But other back-formations such as diagnose (a back-formation from diagnosis that was first recorded in 1861) and donate (first cited in 1785 as a back-formation from donation) are considered unimpeachable English words. Since enthuse dates from 1827, something more significant may be overriding the erosion of popular resistance. Unlike enthusiasm, which denotes an internal emotional state, enthuse denotes either the external expression of emotion, as in She enthused over attending the awards ceremony, or the inducement of enthusiasm by an external source, as in He was so enthused about the diet pills that he agreed to provide a testimonial. Possibly, some people's distaste for this emphasis on external emotional display and manipulation is the source of unease that is manifested by a distaste for the word itself.
Score: 0
|What he said...
Score: 0
|Nice.. I liked the update at the end. So IBM wanted to keep the discovery as a secret for a little longer - no big deal. Besides, once the CPU's go to mass production, it's VERY easy to bribe some worker to steal one for reverse engineering the exact materials used.
I just wonder why there was no mention of octa-core CPUs. It seems the real revolution will be having all popular software rewritten for parallel processing and smarter use of idle CPU cycles (for background virus checking of EVERY file with LATEST defs, backup, compression, indexing, preloading fav websites while making sure content is fresh the moment u click on tab, pre-drawing latest emails you've received incl attachment-preplaying for ZERO wait-time - not even HD access, whatever). Just smarter use of already avail resources, that's all...
Score: 0
|" ... So IBM wanted to keep the discovery as a secret for a little longer - no big deal. Besides, once the CPU's go to mass production, it's VERY easy to bribe some worker to steal one for reverse engineering the exact materials used. ..."
I know about IBM horning in on everything patentable ala "Copper bottom" chips, etc. to support their almost totally misleading ad campaigns. As for reverse engineering the Intel "mystery metal" ... there is not eough time left in my life to re-invent someone elses wheel.
" ... there was no mention of octa-core CPUs ...(?)"
As for the transitions to high numbered CPU cores on a single die, the jump would be from quad-core (4 processors) to hexidecimal-core (16 processors) or hexi-core(?) The math works better, the traces match more easily to the individual PUs ... as does the gross topography, the core layout on the die and die layout on the board(s). The next "multi-core level" after that is to 64 processor units, which has all manner of thermal questions like hot areas on the die, etc. (The prefered gross, macro topography for multiple hexi-core PUs will be a round computing device about the size of a coffee can with hexi-core "blades" feeding a central, axial glass bus ... my next paper/article will explain the details. Copyrights & patents are non-IBM.)
Score: 0
|So if the jump is from 4 to 16, that probably means we won't be able to buy those 16-core CPUs in the next 3 years or so? Just cuz the jump is that high? And that is why there was no mention of the "next" multi-core CPU ver?
BTW re the re-inventing the wheel - well, if someone else's intel about wheels can make your wheels go 10% faster, you'd use that knowledge even in 2 years WHILE continuing work on your own wheel using your existing tools and knowledge. The race, after all, is a long term race.
Score: 0
|To do all that, we'll still probably have to wait for common RAM configurations to be in the double digit gigabyte range.
Score: 0
|Actually, I think hexicore would be six, and I think 8-core processors will be next. You are assuming they are going to be going by square powers of 2, when they could go by square and have 9, like the Cell; or they could just go by powers of 2 like their current quad core. IIRC, the Core Quad is just two Core Duos( i.e. a dual-dual core), not a "pure" quad core.
Score: 0
|Still no answer Re: " ... With the hafnium material serving as the gate, Intel will then replace the polysilicon electrode layer with a metal electrode, the exact alloy used here also being kept secret ..."
...
The Mystery Metal ID quiz Post #23
The tally thus far:
2 or 3 votes:
" "probably a common metal" or " something really embarrassingly non cool sounding like aluminum or steel ".
1 or 2 votes:
" unobtanium " or something so wildly off the known physical universe charts that it will astound us all.
1 vote (mine):
Silver or a high percentage Silver alloy (>=90% Silver)
1 vote:
" that stuff I keep finding in my navel " or some organic compound of a high impedence nature.
Your vote goes here:
Score: 0
|Small fragments of brick.
It stops the wind and rain getting to me in my house...
Surely it can stop at bit of electricity.
Score: 0
|i think intel is using ruthenium for the pmos gate, having perhaps solved the problem associated with front-end-of-line (feol) advanced metal gate etch, notably ruthenium removal.
learning man
Score: 0
|Where the hell are the 4 and 5 Ghz processors? My notebook is three years old and it's a 3Ghz P4. You can barely get a high end notebook with a 2Ghz processor these days.
I dont want to hear about "todays 2Ghz are as fast as 3Ghz Pentiums from 3 years ago" It just aint so.
Score: 0
|Yes, sure it aint so, because a 2 GHz Core 2 class CPU today is way faster than a 3 GHz P4 three years ago. Even lower clocked single core Athlon 64s beats those P4s.
Do your research or buy actual products before making such an ignorant comment.
Score: 0
|"I dont want to hear about "todays 2Ghz are as fast as 3Ghz Pentiums from 3 years ago" It just aint so."
You are either trolling or you have no idea what you are talking about.
Score: 0
|"todays 2Ghz are as fast as 3Ghz Pentiums from 3 years ago"
No, you're quite right.
They're miles faster.
If you tested one and judged it on how fast it booted, maybe you should be checking the hard drive rpm.
Score: 0
|Its true however..... The chips are completely different. If you look at say a 486dx133 (133 Mhz CPU) and a 90Mhz Pentium 1 its the pentium destroys it. Mhz and Ghz isnt the only thing that makes a chip better or faster.
The more Ghz you have the hotter the chip is going to run. You get at a 5 ghz chip you better have a air conditioner ontop of your chip or its going to smoke the moment you turn the computer on.
Score: 0
|More like liquid nitrogen in your 'water coolant'.
http://tomshardware.co.uk/2003/12/30/5_ghz_project/
Score: 0
|After you pass 3 GHz you begin to get diminishing returns. The whole reason that Intel has stopped stressing names like "Pentium 4 3.06 Ghz" is because people get the impression that frequency is everything. That is why AMD always used numbers other then frequency to identify their products.
Score: 0
|Hafnium found as impurities in fake diamonds?
It might run cooler and faster, but
will it be cheap?
Sand we do have a lot...
Score: 0
|The follow-up to Windows Vista will need a quad-core speed-of-light processor due to a new widget system and built in 2-way firewall, that replaces the 1-way firewall introduced in Windows XP.
Microsoft executives are worried about how to artificially raise the hardware requirements for on further generations of Windows due to the speed of Intel's new processors.
Score: 0
|So we go all the way to the atom.
Then I guess we just keep increasing clock speeds and researching parallel processing?
If the speed of light is as fast as you can go and
electricity flows at that speed, then this is an issue of switching frequency rather than speed of source.
Of course, for the average Windows user, the machines are getting so fast that new and old are hard to tell apart. The rushes of speed we used to feel a decade ago aren't happening anymore for me.
Score: 0
|This should not affect AMD, the way I read it it's just a different manufacturing proccess, will it realy allow them to make a better cpu cheaper... I dont use intel cause of price.. so as long as AMD can make a cheap CPU for endusers they'll alway's sell cpu's to the consumer market who dont need a 8ghz CPU!
Score: 0
|Re: " ... With the hafnium material serving as the gate, Intel will then replace the polysilicon electrode layer with a metal electrode, the exact alloy used here also being kept secret ..."
I have started a thread/topic at DIYAudio.com to try to discover the identity of the Mystery Metal. My guess is it is Silver or a Silver alloy like Sterling Silver (90% Silver/8% Copper). I would bet as well that following metals commodity prices would likewise give additional clues at to the Mystery Metal ID.
(re: http://www.diyaudio.com/...tid=1117705#post1117705 )
Score: 0
|AMD ran into the same wall intel did--they were basking in the sun because their technology bested the competition. Well, if they had anticipated this Intel comeback earlier, they would have been scrambling to get new enhancements rather than selling buttloads of current technology cpus.
AMD learned the hard way, but honestly it may be too late for them. Intel's not taking their current lead for granted and is speeding through innovative designs despite their lead this time, something AMD needed to do a year ago. Problem is AMD does not have the nearly unlimited resources to make a huge comeback like Intel did. If Intel keeps up the pace, they will certainly spell certain death for AMD, though the death of AMD may not occurr for many years to come.
Score: 0
|AMD is going to die? That cant be good.
Score: 0
|AMD has been around since 1969 and making chips for the PC since the 8086, I really don't see them dying. Even if they don't take the lead again any time soon there is room for more than one player in the CPU business (not to mention they just bought ATI). Look at VIA, practically no one uses their processors and they're still making them. AMD has a very strong following of fans, they'll be back in the lead again eventually. This is coming from an Intel fan by the way.
Score: 0
|Neat!
Score: 0
|SH*T. C'mon AMD act fast! I'm getting ready to puke knowing I'm gonna have to switch back to Intel CPU's any time now!! :( :(
Score: 0
|Why?
Score: 0
|Why?
Because Intel's new generation of chips significantly outperforms AMD at nearly every level.
If I had seen this coming, I would have held off on building an AMD machine.
Score: 0
|Great article, very interesting for someone studying microelectronics and such.
Score: 0
|I guess this is is a big deal when implementing something. But AMD is no big deal either.
Score: 0
|Well this news is more than 3 years old, but I guess they have a working prototype now.
http://www.dvhardware.net/article2075.html
AMD wasn't too worried back then...
http://www.theregister.c..._intel_highk_transistor/
Score: 0
|I enjoyed this article very much, bringing back to me the days of a decade or more ago when computer news was indeed engineering news. I don't personally care about performance or costs or timetables as much as I do about evolution and revolution in the industries. This article was well written and covered some of the lesser known techincal aspects admirably, thanks for a good read!
Score: 0
|My pleasure, Braden, thank you very much!
-SF3
Score: 0
|Article has too much on the electrical engineering details and not enough computer science.
1) What clock speeds are we talking about here?
2) How much would new chips made with this technology cost when they start mass production?
3) When would this occur?
Score: 0
|Clock speeds haven't been relevant for a long time now.
Score: 0
|So you want less facts and more guess-work from the article?
Score: 0
|"One performance improvement consumers should expect to see from the Penryn family is an old familiar one - something we were once told we might not be thinking about any more: higher clock speed.
"For Penryn, we're going to deliver some micro-architectural features that will improve performance, and separate from that and orthogonal to that, we're going to deliver some clock rate improvements," Intel's Smith told reporters. "You can expect some frequency increase within the same thermal envelopes; we know that is possible, and that's our plan for these.""
sounds like they're about to be relevant again.
Score: 0
|Fair questions. There are some unanswered questions at this point: We don't know what the new clock speeds will be, probably because tests on how much the accelerator pedal can be pushed have not been run yet. Intel's CPU tests were only conducted in the past few days; we do know a Penryn design can run Vista, Mac OS X, and Linux (don't know what distro).
How much would these chips cost? Earlier in the week, during Intel's conference call and before the materials discovery announcement, Intel's executives said their goal was to introduce the 45 nm generation at about the same price point as the 65 nm generation. My feeling is, that price point will elevate just a bit - chock that up to inflation.
I should add: Some of the real estate savings that Intel will gain by cutting the transistor size in half, will likely be partly consumed by an expanded L2 cache.
When? Intel has already announced its intention to begin production of Penryn chips in time for introduction to the mass market in the second half of 2007. So essentially now. There's not going to be a five-minute rest while these guys are getting retooled. Now, AMD has been working to phase in its 65 nm generation, and that job will be finished this year. But with IBM's announcement (see the update) that it's found an HK+MG formula as well, and it'll share that with AMD and the STI coalition (Cell), then we'll see HK+MG chips from AMD probably a year after Intel's - not 2010 as Intel's Mark Bohr was hoping.
-SF3
Score: 0
|