Slashdot Mirror
AMD, IBM Announce Transistor Advances
Jugalator writes: "AMD announces it has built a CMOS transistor with the highest switching speed in the semiconductor history. The transistors are manufactured with .015 micron technology and allows a twenty-fold increase in transistors per chip with a ten-fold increase in performance when compared to the transistors in use today. So far, AMD has only produced a prototype and a larger scale production is not planned for until 2009 at earliest. AMD will announce further information regarding their research in the semiconductor field at the 2001 International Electron Devices Meeting today, December 4." schongo sent in a note about IBM's double-gate transistor. This and the Intel announcement recently are all related to the International Electron Devices Meeting.
it's good to see Intel's primary competition coming up with innovations like these. Continued development on the part of AMD is really the key to keeping Intel from dominating the entire market.
I wonder what temperatures these will function under. Personally, I want to see light-based chips, due to what I hope will be a huge reduction of heat loss.
Then again, on cold winter days it's nice to have a 900MHz space heater.
Moderation: Put your hand inside the puppet head!
...will continue to be right for a while. ;) (See Moore's Law if you're unfamiliar. :)
libertarianswag.com
But with all of the recent advances in technology of this nature, I doubt that we will ever see this hit the market. With an expected public release date of 2009 (at the earliest...), I would have to think that something a bit more advance and more easily produced would be widely available and send this cool little advancement into obsolecence...
Beer is proof that God loves us and wants us to be happy. -- Benjamin Franklin
As is par for AMD, this advance is an impressive improvement over what their contendors at Intel are doing (especially lately). As is also par for AMD, though, these transistors produce a great deal of heat. One of my co-workers once worked at another semiconductor firm which experimented with a similar technology, and said that the heat generated by these things is astronomical. (That should come as no surprise to overclockers, who know that the faster you run it, the bigger your heat problems become.)
It is pretty obvious that AMD has some big heat issues. After all, Tom's Hardware was able to cook an AMD CPU and motherboard all at once just by removing the heatsink from the chip. Heat is a serious concern with these things.
However, I am optimistic that AMD can solve whatever problems there are with this technology and bring it to the consumer eventually. Hopefully that will happen before Intel uses their size and budget to crush AMD permanently.
df
ten-fold increase...not planned for until 2009 at earliest
Sounds like the rate of increasing performance is starting to drop. Isn't it supposed to double every 18 months? Shouldn't we then expect a 25 times increase between now and 2009? (2^(7years * 12months/18months))
Hope they have some other tricks to make chips faster!
And this is intresting too - "Rambus founder's Matrix unveils first 3D memory chip" (The Register), intresting stuff, it kinda reminds me of Orson Scott Card's multidimensional memory in "Children of the Mind".
All browsers' default homepage should read: Don't Panic...
what's wrong with you?
don't you know about design cycles and that they HAVE designed stuff that advances the chips of tomorrow? they did this 5 years ago,
stuff like MOCVD, e-beam lithography, etc, etc, etc.
it takes a LONG time to get transistors into products. Look at SOI and SiGe, the first SiGe HBT was fabricated in ~1970. they are only now making it into products. SOI has been in the works for 10 years, and they are still only using partially depleted channels because of clean interface issues.
Sorry, that's it, no more accurate predictions until next year!
Ok, so I successfully predicted this, what does it say? The game of one-upsmanship is to reassure investors that R&D proceeds during uncertain economic times? Though we aren't selling much, we're preparing for the future, just like our competition is? Sounds good to me.
A feeling of having made the same mistake before: Deja Foobar
Just a FYI 3.3Terahertz is a 303 femto second period. How many electrons can you move around in 303 femtoseconds? If the clock is running at 3.3Thz, (assuming this is not a clockless chip) it leaves about 102 femto seconds for a rise time measurement, and 101 femto seconds for a fall time measurement. (Looking at the pulse). Projecting further I woud expect the rise time zero to 50% point would be about 50 femtoseconds. I'm supposed to know this, but can't recall the rule of thumb for the oscilloscope bandwidth I would need to look at this, but I'm guessing I would like more than 4Terahertz... I'd really be comfortable with about 8-20Terahertz bandwidth on the scope... or somthing that could digitize the signal at 20 Terahertz or so (20THz = 50fs). I'm sure this analysis is flawed, but I just like to imagine that I could actually look at a 100fs rise time with a scope in about 5-10 years. -- Ross
Ross Youngblood
I personally don't care about another 8fps in Quake - I hit the frame rate cap a long time ago. What does matter to me is getting a render done in Brazil before the next ice age. With a 1.1GHz chip, it can take a couple of minutes for a pretty simple raytraced scene. One that starts getting complex, then you feel the need for a faster and faster processor.
"I think he was truly surprised at how little I cared about how big a market the Mac had" - Linus on Jobs
I have no idea why they keep posting articals like this to slashdot. Chip design has always been about slow improvement, by the time this actualy hits the streets, it just won't be that impressive.
autopr0n is like, down and stuff.
Anybody have any idea where this puts the projected number of transistors per cm2?
Get busy living or get busy dying. Carpe diem.
Uh, I don't see why this is considered revolutionary. More's law states that chip density doubles every 18-24 months.
:P
Well, 2009 is in 8 years, or 4 doublings if you're going by the 24 month rule. Top of the line chips now are minted at 130 nanometers. Double once, and you get 65, double again and you get 32.5, and double the final time and you get 16 nanometers... and the AMD transistor is 15. Going by the 18 month rule and you get a bit more then 5 doublings.
In other words, while its great that they haven't hit the wall yet, this is really all they're telling us. CPU speed has been improving predictably for decades and this is no exception.
If they'd announced that these transistors were going to be used q1 2002 in new Athlons it might actually have been news
autopr0n is like, down and stuff.
who thinks, "Good, now we can overclock our "CMOS" and get to a less than 5 second boot time."
Only down side I see is needing a CMOS heat sync and a 2 1/4inch, 5K rpm fans to keep the thing cool.
Think about it, seriously for a minute:
We have heat syncs and fans on; processors, chipsets, powersupplies, memory (sometimes), case fans, hard drives (on occasions)...what is missing?
Oh, yeah, we need some of this action on the CMOS!!!
OOoooKaaayyyy, works for me.
Cheers,
Moose.
Have you read the moderator guidelines? Well, have you, PUNK? (and I want a Karma: Gnarly option)
I never knew 'x fold' was supposed to denote an exponential increase. In my experience when most people say it they mean increase via multiplication. Maybe the word has changed meaning (as words do). Although thinking about it makes sense.
autopr0n is like, down and stuff.
You will invent a transistor that reflects the size of your penis
...in bed.
Liquid Nitrogen is CHEAP! Shouldn't take too much to design a case cooled with the stuff! And as an added bonus, the cloud of vapor your system releases when you fire up a spreadsheet recalc will be WAY COOL!
I'm trying to teach myself to set people on fire with my mind... Is it hot in here?
What's wrong with you guys? Aren't you engineers?
Then come on! I want a spacecraft to reach Mars in 1 second, and I want it for tomorrow!
And please, don't give me that pathetic excuse about the fundamental limits and the speed of light and all of that bullshit, find some breakthrough and do it now, or you think I pay you for the neat innovations that will be useful in 15 years from now, if they ever get to be useful at all?
I believe AMD / IBM should develop this to the point it could be released say in 3 or 4 years. Right now the chip market is moving to the 0.13-micron mark. Imagine an Athlon with a 0.015 micron setup. AMD would have such a advantage over Intel at that point that it would not be funny. If they don't implement this until 2009, then Intel might beat them to the punch by developing there own 0.015-micron technology, or just pay IBM cash for the designs.
I can see it now, a dual AMD Athlon system running 10 GHz per processor at a 0.015-micron chip scale. Now that's speed
my 2 cents plus 2 more
Issues which contribute to delay bringing to market:
Expensing research which contributed to current technology (if it took x million dollars of research to get this advancement, then that cost has to be paid, and it's not all at once)
Current technology must be affordable. Ok, if you're the FBI scanning billions of emails or any other deep pocketed government department (NASA, DoE, etc.) You can buy it, but you don't buy a lot of them.
Improvement in the manufacturing process makes it possible, practical and affordable (yet, more R&D which you don't often hear about which must be expensed)
Today's technology pretty much meets todays needs. 99% of the market would actually be just fine with a 500 Mhz P3 or Athlon system, with 20 Gig HD and 128 Meg of ram. Launch a 100GHz CPU and people wouldn't have the need, though if it cost marginally more than the current crop, of course they'd buy it, but only gamers (no, not in FPS, but in behind the scenes complexity) and engineers would see benefit.
Raw materials, supply chain, etc. Sometimes all the materials, meeting purity/quantity aren't there and you have to wait for them to catch up.
A feeling of having made the same mistake before: Deja Foobar
This press report is overrated. I have created a car that can travel 10000 miles powered only by a bowl of salsa. The secret is an engine that is 4000 times faster than what is currently available, by using a technique called "warp drive." I expect to produce these vehicles in mass in 2061.
- Zephram
Semiconductor companies are constantly announcing "Breakthroughs". In my eyes, I don't see them as major breakthroughs, simply an evolutionary step. McDonalds constantly fixes it's processes, changes recipies, etc.. but they don't run a commercial unless they come up with a new sandwich. Develop new materials besides Si, and SiO2 and I might just buy some stock!
While it's wonderful that they can create a 300fs inverter, you also have to consider that they have yet to prove that they can actually mass-produce these structures to get adequate yields. This is not a trivial operation. Bell Labs, IBM, Intel, and AMD have all announced ultra-small and/or ultra-fast transistor structures, but they all admit that they are far from mass-producing them on a wafer/die.
Also - the rest of the componentry in a computer or other electronic structure, and how it will all communicate all of these calculations, will also be a problem. Already, integrated circuit I/O circuits are having trouble transporting data back and forth on a PCB.
ALSO, consider that the photolithography tools that are supposed to support the next generation of smaller structures are already off-track. 157nm lithography tools have been delayed due to development and financial difficulties. See SiliconStrategies.com). My personal guess is that the vertical MOSFETs will be the winners in the short term because, until they get other factors in line, they will have to make do with what they've got, though *again* the additional processing required for the wafer will impact yields, so it will be an expensive technology to implement either way.
And an AC has to point out basic math:
130 to 65 is four times the density. Likewise with 65 to 32 and 32 to 16. Of course, you could also go by the twelve-month rule.
This is great, we have transistors hat are faster than anything else humans have developed so far.
But, this really doesn't give us any leap in abilities. What about massive parallel processing? What is holding the human race back from creating a chip that is basically 16 or 32 seperate but equal processors?
Linear processing is fine for things like calculators and basic tasks, When do we get our hands on some real leaps in processing?
Does anyone know of any links that point to research in this?
Do not look at laser with remaining good eye.
When a light beam is used to modify another light beam, it is usually made with some materials which reacts to lights.
The material will be heated by the light, so it will generate heats..
Light doesn't interact with light directly, so a light-based chip would be really a light - non-linear material - light chip.
Usually the interesting effect in those materials used to modulate light are only a "second order" effect, which means that you have to use quite intense lights to have something usefull.
Intense light --> heat.
It's quite in line with the Moore's law, which states that the number of transistors per chip will double every 18 to 24 months, and therefore expects this number to be 2^(8*12/24) = 16 to 2^(8*12/18) = 40 times larger in 8 years. AMD promises 20-fold increase, which is within this range.
Uh, I don't see why this is considered revolutionary. More's law states that chip density doubles every 18-24 months.
First, it's Moore's Law. Second, calling it a law is ridiculous, because it's entirely dependent on continuing R&D, as well as bringing such R&D into production. Taking it for granted that you'll have your 40GHz CPU in 9 years is really quite naive.
Personally, I can't wait until Moore's Law fails (either by falling short from or totally surpassing the prediction), so that people stop using it to degrade the really quite amazing research and amount of work that goes on in order to bring such results.
Thanks,
Mike.
-
...For Optical processors, when we wont need any transistors. I really don't think that a 0.15 micron will be anything competetive in 7 years. If they already have the micron figured out, you'd think they'd be manufacturing them? why not release it?
"And we have seen and do testify that the Father sent the Son to be the Savior of the World"
1 John 4:14
With Cyrix gone and AMD kind of falling behind Intel and now resorting to similar tactics that Cyrix used before they went under, the marketing of their CPUs at a higher apparent clock speed than they are actually running, does anyone think AMD will be around in 2009 to capitalize on their research?
Recently, everyone has been saying that Moore's law won't hold out much longer. And they've been saying it ever since he postulated it. I don't think it'll stop any time soon.
.18-micron process, we have chips with 42 million transistors running at 2 gigahertz.
The 386's, when matured, were built on a 1-micron process, had 275,000 transistors, and ran at 33 MHz. Now, on a
So, by shrinking the size of the transistors to 1/6th of their size, we got 153 times the packing ability, and 60 times the frequency. And these transistors that they're talking about are only 1/10th the size of the current "high-tech" transistors. That means that we could pack over 100 times more transistors on a package, and run them 100 times faster. Not bad. But I suppose that they'll need a safety device to shut them down if the flow of coolant ever stops. : )
steve
Oh, you're not stuck, you're just unable to let go of the onion rings.
First, Moore's Law refers to the doubling of the density, not the speed of the chip. So your "40GHz" example doesn't fly.
Second, it is commonly referred to as a law, it wasn't the original poster's idea.
Third, once we get past silicon for chips, I think Moore's Law is out the window.
My beliefs do not require that you agree with them.
I've actually seen it spelled both ways:
/., oye, I'm really asking for it, aren't I?
:) no offense, whomever you are.
Heatsync and/or Heatsink. Which is correct? Dunno, doan really care because the "sync" has stuck with me as far as heatsyncs go.
Any Non AC's care to reply and set me on the correct spelling? Spelling question on
C'mon dudes and dudettes, it is "sync or swim time".
Oh, and to the First AC who replied...yeah, maybe both sarcasm and ignorance is applicable.
Maybe it is me, but sarcasm, ignorance or both has never stopped anyone from posting on slashdot, has it?
But, I suppose, that is usually why the default is to filter out AC's...
Anywho... 'If I'm right, I'm right. If I am wrong, then show me I'm wrong' is the way I see it.
Have you read the moderator guidelines? Well, have you, PUNK? (and I want a Karma: Gnarly option)
I can't wait to see this stuff lithographed on the Silicon-on-arsenide (sp?) technology.
something clever
While suffering from overheating and architecture flaws as old as the 1970s, the x86 might last into the 2010's or 2020's, but it requires more and more expensive processes, innovations in the constructions of transistors and other envelope-pushing procedures. Meanwhile other architectures are enjoying higher instructions-per-clock, far less power consumption and heat production, and greater vectorization.
By the 50's, the x86 by Intel/AMD/whoever else will be a memory. The "other" major platform seems to have less of a problem with switching to new architectures every few years, whenever it becomes practical. Will Wintel users be lucky enough to have "Moore-Compliant" emulation of the Pentium-XIXX and the next CPU they're forced into?
"Look at me, I invented the stove!" -- Ben Franklin
Intel's new terahertz transistor promises smaller, faster and lower-power. AMD's new transistor is smaller and faster, but did nothing much about the power. :(
Wouldn't it be just too much to hope for, that they both try to incorporate similar ideas from each other into their newest products? Imagine, much smaller, much faster *and* lower power consumption. To me, that sounds like a great idea. 'S a pity that our "compete at all costs" system won't really allow this to happen, though *sigh*
Lemon curry?
Why is it that whenever I buy a computer, something like this happens where it will be obsolete in a few years? Nothing like buy a $2k paperweight!
She sat at the window watching the evening invade the avenue.
Except that feature size is linear while area is square. Naively then (assuming components/area is nearly exactly proportional to square of feature size), to double the components, feature size has to decrease by a factor of square-root of 2. Assuming 4 'doublings' in 8 years, that gives a feature size around 30nm or so. (5 gives about 24nm.) So 15 is a little ahead of the curve by that measurement. Of course, component count _isn't_ exactly proportional to the square of the feature size, so they'll probably be right around par.
And who said two wrongs don't make a right?
Those who fail to understand communication protocols, are doomed to repeat them over port 80.
AMD announces it has built a CMOS transistor
this is a good effort, considering there is no such thing as a CMOS transistor.... CMOS refers to two transistors hooked up so the output can swing fully high and fully low. making this out of one transistor should be a lot bigger news than this article gives justice to...
stupid chip manufacturers don't know what they are talking about...
There are only two things in this world that smell like fish. And one of them's fish...
Imagine a beowulf cluster of those! ;P
BlackGriffen
Heat is proportional to power consumed, and power is proportional to voltage squared and frequency. The problem isn't heat though, exactly, it's heat density. A hot-plate makes about 10watts/cm^2. A P-III makes about 30. If current trends continue (faster switching speeds, smaller transistors, with only slight decreases in voltage diffential), then a 15nm process will produce more heat/area then a jet engine.
:)
Yes, heat is a problem
Expect airconditioning to be a standard feature on new computers soon.
Those who fail to understand communication protocols, are doomed to repeat them over port 80.
Corning spun off their Pyrex cookware and glass dishes, etc., to form WorldKitchen, Inc.. So they don't know as much about glass cookware as they know about optical fiber.
Ahhh... how soon they forget.
The first real wave of RISC CPUs did shatter Moore's law. Performance jumped from 4 MIPS to 12 MIPS. Prices dropped from $70,000 to $10,000. It was truly cool.
Of corse in the following years CISC have mostly died (seen any new ones launched? I have seen a lot die), except the x86, and to a lesser extent the 370/390/whatever-z-or-x-name-they-have-now, and the x86 has even caught up to the RISCs, and is almost all cases passed them (it's amazing what you can do with 10x the R&D budget...).
I think there have been a few places where there was underperform, like from the 386 to the 486 maybe, or from the 486 to the P1 (and I think that was due to the length of time between the 486 and the P1), but I forget exactly when mostly because nobody kept pointing it out like Sun did with "RISC is better, eat our dust DEC".
If you look at the SRAM market you will find similar events.
Still as a long term thing Moore's "Law" is amazingly accurate. I think because it functions as a goal for R&D managers. If they can't keep up with the "Law" they tell their bosses that they need more money or the company will die, if they have managed to keep up with it they don't fight as hard for the budget (or they do, but they have less ammo to fight with). I know R&D budget isn't everything, but it is a powerful force.
I will still take Windows 2009 5 mins to boot.
"The word "genius" isn't applicable in football. A genius is a guy like Norman Einstein," - Joe Theisman
all those claims may sound good now, but if it will take /at least/ 8 years before production begins, I wonder by that time if we could have made enough advances to already reach that performance level.
it's very unusual for companies to research on things they don't plan to produce within 5 years...
First, Moore's Law refers to the doubling of the density, not the speed of the chip. So your "40GHz" example doesn't fly.
I understand that Moore's law as he stated it applies to the number of semiconductors on a chip, but also that the general geek public accepts Moore's Law to apply to the speed of a processor, which remarkably seems to also match the 18-24 month rule.
Second, it is commonly referred to as a law, it wasn't the original poster's idea.
I haven't blamed the original poster for coining the term, but I can see how it appears so since I said it after the slightly anal spelling correction.
Third, once we get past silicon for chips, I think Moore's Law is out the window.
I agree, and getting away from silicon is just one variable that can make the whole idea useless. Others include quantum computing devices coming into play, 3D chip design, and clockless chips (referring to the speed-doubling correlary to this "law").
Thanks,
Mike.
-
That IS pretty cool, but what I consider to be the "shattering" of a given law is when the majority no longer considers it valid, and that's unfortunately not the case.
I blame the marketing department, which creates the consumer demand by telling us that buying a 2GHz Pentium IV will make the internet faster.
No offense to Mr. Moore, who probably never anticipated his comment to be taken so seriously or adhered to, but it'd be nice if the focus was taken away from how many clocks per second or transistors there are, but what is done with each tick (or per recent news whether the ticks are necessary at all), and whether each transistor is used efficiently (FPGA's, anyone?). I hope to see a market where results from open, approved benchmarking methods based on real criteria are the main selling point. Guess I shouldn't hold my breath, eh? :o)
-
and this uses how much power?
Take a look, jerk off, and you'll feel a lot better afterwards.
autopr0n is like, down and stuff.