Slashdot Mirror
IBM Demonstrates High-k/Metal Gate Chips
Last summer we discussed twin announcements from Intel and IBM/AMD about a new chip manufacturing technology dubbed high-k/metal gate. Intel is using the tech to improve speed and power consumption in its 45-nm chips. IBM, along with its manufacturing partners, just demonstrated chips it says show that high-k/metal gate technology at 32 nm can result in performance gains up to 30% and power savings up to 50%, compared to 45-nm process. IBM plans to be manufacturing 32 nm parts by the end of 2009. (AMD is not using high-k/metal gate yet, but it has access to the technology by virtue of its agreements with IBM.)
and it just goes to show that silicon is not dead yet
I know that I've posted about this before, but...
Huzzah! For the first time in 25 years, the name MOSFET ( Metal -Oxide-Semiconductor Field Effect Transistor) will correctly describe the device that goes by that name!
(For those confused as to my jubilation, highly doped polysilicon replaced metal gates over 25 years ago. As a result most MOSFETS haven't actually had metal in them since.)
>> Standing on head makes smile of frown, but rest of face also upside down.
AMD earned my loyalty many times over the years, and now that it's fallen from the top of the price/performance heap, I feel bad buying another chip. This is the company who made the chip for my first computer, that made 64 bit mainstream, and made intel actually improve their products. They've done so much for the industry, it'd be a shame for them to continue taking a pounding like they have.
Also, I own some of their stock. Go team!
But Germanium-Arsonide is a much-neglected technology that could do with more investment, as it should do much better than silicon. Graphene is another technology that risks being ignored for as long as silicon is a viable option. I'd far prefer chip companies to be pushing the boundaries with materials that should offer far more extreme performance. Nonetheless, any progress is good progress.
It's a small world and it smells funny; I'd buy another if it wasn't for the money; Take back what I paid (SoM)
I've been getting my seeds from the Netherlands for some time now.
because I have an ongoing discussion about this topic, and I want to know, in what part of that spectrum (45 - 32 - 29 nm) is Intel at this moment? Are they already manufacturing chips with such technology?
high-k/metal gate technology at 32 nm can result in performance gains up to 30% and power savings up to 50%, compared to 45-nm process.
Really revolutionary announcement there...
Power consumption scales with the square of gate size. (32*32)/(45*45) = 0.51, or 50%.
Clock speed scales linearly with gate size. 32/45 = 0.71, or 29%.
Not to minimize the fact that these gates reduce leakage enough to actually get those gains, but the drop in gate size alone (all other factors equal) would give the same numbers.
Graphene is not at all nearly ready to even build reliable, well-performing transistors with it. I'm in a research group that is trying to implant a gate electrode into Silicon-carbide with a Graphene layer ontop, but that's still basic research. If it should really work with good yield and that also in an industrial process, then we can talk about Graphene-based CPUs.
And by the way: it's spelled "Arsenide"
This MAY BE what "turns AMD around" in the PC market (largely gaming/home use besides office/server work), & especially in a HUGE market (gaming) segment, home usage:
For example, w/ some GOOD reliable numbers from a reputable tech website below?
Give AMD this technology being put into their CPU's & it appears that AMD really has a legit handle on using it from IBM, per this article here??
That SHOULD give AMD a 30% "pop"/"boost" in performance CPU-wise in gaming!
E.G. -> FROM TOM's HARDWARE PAGES CPU PERFORMANCE CHARTS:
http://www23.tomshardware.com/cpu_2007.html?modelx=33&model1=1106&model2=1072&chart=424
& on a game like Quake 4, for instance, where above you can see AMD's 'best' currently (Athlon64 x 6400) vs. vs. Intel's current 'best' (Core Extreme QX9770)??? They'll be in a fight again, this is certain, by 2009 or so... they're only around 30% apart as is, now.
This COULD be AMD's trump card for winning the hand.
All in all, which for consumers is good - we as the 'end-user's' get "kick butt" gains in performance again, which is FINE by me, speaking as a user. Especially @ gaming, & this game (ID Software's Quake4 SMP's huge, no doubt about it & has tons of buyers as well as ID Software's doing so well in PC Gaming over the past 15 yrs. now), Quake 4, is just 1 example!
Heck - I know this kind of boost would help in a game like Crysis (heaviest one I have actually seen lately, I still play Doom III quite a lot in fact (I don't cheat while gaming, takes the fun out imo, plus, & I don't have a lot of time due to work, so I finish games MEGA-SLOWLY, lol)), even though graphics boards (Go NVidia 8800GTX/9800GTX) mean more today in gaming...
NO DOUBT - this can help AMD a lot, & just based on the numbers above as a single example of why/how!
Performance, to gamers (again, HUGE part of PC market)? It's EVERYTHING - if you're a gamer reading this, you know what I mean (I am one too just not "hugely avid" anymore like 10 yrs. back is all, but I still am). I mean, look @ the numbers from the URL above, & if you give AMD that 30% performance jump, no doubt, they'll be even again w/ INTEL (for starters) really!
Hey - those chart's from Tom's Hardware pages show you that INTEL's best, right now, IS ONLY ABOUT 30% faster than AMD is, now/currently. The alleged performance punch this may yield for AMD will make it interesting... for gaming, AND investing in the future, imo!
APK
P.S.=> That's IF the numbers mean anything, & usually, they're helpful (for performance seekers AND investors in tech stocks too)... from an investment perspective, AMD stockholders will do well imo @ least, since AMD typically charges less "pound-for-pound/dollar-for-dollar" than INTEL typically has over time, which imo means AMD will get ontop again FINANCIALLY, due to volume sales (just as powerful/good a product, for less... a sure WINNING solution for sales, which of course = $)... apk
Comment removed based on user account deletion
Help prevent lameness, use a filter.
Well, that's some trolling you've got going there! As an AMD stockholder I take no pleasure in saying you're full of shit, an idiot, or both. "Only 30% apart?" First, 30% would be a huge difference between two current-gen CPUs of similar price point, but the chart you linked shows the Intel CPU taking 22s to do what the similarly-priced AMD chip does in 54s. 54 is 145% more than 22. 22 is 58% less than 54. However you look at it, it's a hell of a lot worse than "30%" for AMD.
everything in moderation
Yes, it does get more expensive. Inflation's the obvious one. Another obvious one is that the cost of employees, in the higher skill brackets, rises faster than inflation and also increases the total costs. However, you've also got the fact that to be competitive, GaAs would need to do as well or better than Si. Being better than it is now won't significantly change the marketplace. Say that gap can be bridged right now by an investment of $x, but that the gap will widen by $y per year, on average, where y includes the extra time involved and the fact that the target will be moving over that time. Then, in twenty years time, the total initial cost (before any worthwhile returns are seen) would be $(x + 20y). This raises the question of how large y is, relative to x. This depends on the relative amount of research being done, of course, which in the case of Si is gigantic, which means that the target you have to catch and pass is moving very rapidly.
If research progressed linearly, then a year's development in Si would be equal to a year's development in GaAs by the same number of people at the same level of investment. It's not linear. For a start, the tools don't exist for GaAs yet, so you've got to design and build those as well. You've got to train skilled researchers, which isn't cheap, in numbers comparable to the competition (Si), where the competition has an ever-increasing number of skilled workers automatically from Universities. Then, there's the rate at which understanding is built. Understanding is built on prior experience, and there is far less prior experience with GaAs. Understanding won't transfer, because different materials have very different properties.
Finally, there's the ability to do something with that experience. There are two factors here. Si fabrication plants are numerous, and competitiveness is a function of production quantity. If you cannot make the product cost-effective, you won't compete. The second factor is quality control. We're able to produce Si wafers with so few defects that you can get extremely high yields of extremely sophisticated VLSI chips. Motherboard and expansion card manufacturers are going to buy the cheapest chips that perform as required, and won't switch from what already works unless there is a competitive advantage of sufficient size (marketplace inertia). That means you need yields that are equal or better, preferably better. Then there's the problem that Si is cheap in comparison, which means that to have a comparably-priced product, the savings elsewhere must equal or exceed the difference in cost of materials.
Since the price of the product being sold must recoup the entire cost of developing it plus the entire cost of producing the means to produce it plus the entire cost of then producing it plus recover the amount not earned through not investing in what already existed, divided over the units of the product you expect to actually sell by the time investors will expect you to show a positive balance, and since the marketplace inertia will increase as the amount of Si within the marketplace increases relative to other technologies, it is a trivial deduction that you would have to ramp up GaAs production facilities far in excess of all the existing Si production facilities in order to mass-produce enough to reduce the value of each chip sufficiently.
All this assumes, as you say, that knowledge that isn't being utilized remains constant. It doesn't. Knowledge isn't useful without experience and practiced skills. The further ahead Si becomes, the smaller the
It's a small world and it smells funny; I'd buy another if it wasn't for the money; Take back what I paid (SoM)
...may I ask, is it a rant to say that progress is non-linear and finite in any field? Did your great-to-the-nth-degree grandfather invest too heavily in the flint mines to notice that stone was close to the limits of what could be done with the medium and copper was the way to go? (Followed by bronze, followed by iron.) Every era has a dawn and a dusk. It's not an anti-this or a pro-that, it's simply the way technology functions. We don't use CISC chips any more, they're RISC or hybrid. Steam-powered trains and cars were consigned to the history books a long time ago, and I'm damn sure you're not using magnetic core for RAM. Everything that is born will die. Silicon isn't dead yet, but it's an old-age pensioner as technology goes. It might very well maintain to Moore's Law for another 20+ years, but it can't maintain it forever. Nothing can. Eventually, the rate of progress will become slow enough that the cost of progress exceeds the returns obtained from that progress. History shows that it is highly desirable for something to be able to replace a technology that hits that point. Stagnant technology is generally not a good thing. Why you're so up in arms about me not liking stagnation, and why you are wilfully confusing that with some absurd notion of not liking silicon, is completely beyond my comprehension. My posts would have applied to any of the decayed and replaced technologies of yester-year, in any society of any culture. Nothing lasts forever and you've got to think enough in advance to allow for that. Entire empires have thought otherwise. The ruins and relics of such deluded thinking make for great tourism.
It's a small world and it smells funny; I'd buy another if it wasn't for the money; Take back what I paid (SoM)
To replace an established technology, the newcomer has to be both suitable and mature. That point did not come for any competing technology you named yet. It will, when the time comes, but, like with a Greyhound bus, you will not make it arrive faster by ranting on slashdot.