Howdy, Stranger!
It looks like you're new here. If you want to get involved, click one of these buttons!
Quick Links
So this is how Moore's Law dies
Moore's Law asserts that the number of transistors on a processor doubles every two years. This is driven by being able to do a full node die shrink every two years, so that each transistor takes half as much space as before. That means you get twice as many transistors in the same die size.
So, how are we doing on those die shrinks of late?
Intel processors:
First 32 nm product: Arrandale/Clarkdale (Core i3, etc.), January 2010
Early expectation of first 22 nm product: Ivy Bridge, January 2012
Prognosis: Not happening. Rumors say it has been greatly delayed twice so far, and currently looks set for mid-2012. Further delays are still possible.
AMD processors:
First 45 nm product: Deneb (Phenom II), January 2009
Expectation of first 32 nm product: Llano (A-series), January 2011
Prognosis: Didn't happen. Llano was repeatedly delayed and ended up launching in June 2011. Five months later, yields are still problematic, and the volume ramp-up has been maddeningly slow.
Video cards:
First 40 nm product: RV740 (Radeon HD 4770), April 2009
Expectation of first 28 nm product: Southern Islands (Radeon HD 7000 series), early to mid 2011
Prognosis: Didn't happen. Still hasn't happened. AMD switched to a different process node at TSMC that would be ready earlier but isn't really meant for video cards, and may be able to get cards out in January 2012. Nvidia cards should be coming later, and possibly much later.
Rumors say that Global Foundries' 28 nm process node was so badly delayed that Wichita/Krishna wouldn't be ready until mid-2012. AMD has reportedly cancelled the products entirely as a result. That probably also kills the rumors of AMD building 28 nm video cards at Global Foundries.
That's three different process node jumps at three different fabs. All missed the two year cycle, and missed it badly. This comes on the heels of TSMC cancelling its 32 nm process node entirely, which is the first time that has happened since sometime around, well, ever.
If this is the wave of the future, then we could very well see Moore's Law collapse amid ever longer cycles to get to the new process node.
Comments
Well the quantum instabilities had to put an end to it eventually, we knew that all along. The kind of tech we've been using so far will have to be replaced by things like quantum tunneling, bioneural circuits and other experimental stuff we haven't even though bout yet. The jump in speed will be incredible though.
My Guild Wars 2 First Beta Weekend "reviewette" : http://www.mmorpg.com/discussion2.cfm/post/4944570/thread/349125#4944570
We knew this would happen, you can't just keep dividing by 2 indefinitely. At some point (soon) it will be easier to develop a new breakthrough than it will be to continue the die shrink process. Once it happens, the floodgates will open back up.
Call me when we have Quantum Computing =D
Moore's law will die untill we move to new types of computing. I.E. Quantum computing/DNA computing. I remember reading Michio Kaku's Physics of the Impossible around 4 years ago, and he laid out the death of Moore's law.
Once we get Silicon down into the 10-12 nanometer range, there is no way to track the electrons, and this will result in the chip short circuiting.
I wouldn't rush to any conclusions just yet.
Our current CMOS technology is running on 45 years now, and we really haven't come up with anything better yet, we've just kept shrinking the traces. It was a huge leap over older TTL logic, but wasn't widely used for several years. Consider the jump from mechanical gears and springs to relays, then to vacuum tubes, to TTL, and we've been using CMOS for a really long time now.
Intel's planar double-gate could change things up a good bit (and other companies have similar technologies, such as FlexFET and FinFET). SOI opened up a lot of new opportunities 10-12 years ago (when we were more or less stuck under 1Ghz). The jump from aluminum traces helped a lot. A lot of other physical changes (the use of Germanium, for instance), has opened up the flood gates for another few years of advancement each time that we think we have stalled out.
Just looking at the advancements in silicon manufacturing technology, the small leaps have come from chemistry, but the huge leaps have come when we have totally changed technologies all together. True, we are getting to the point where we are so small that atomic size and quantum effects start to actually effect operations, but I don't think that it's dead yet. After all, we still use TTL in many IC's, even though it's not been seriously used in CPU's for 30 years or more.
So who knows - A lot of people have speculated that Quantum is the next big advancement: and there was not too far back announced a working quantum computer (although in the very basic sense of the word "computer") - we probably will never abandon CMOS technology all together, but maybe it does hit a brick wall when we get too small. Some people have placed big bets on biological computing - living cells operating massively in parallel. And maybe distributed computing is the key: we have some much underutilized computing power already in existence, it just needs to have a high enough bandwidth connection to bridge it all together...
I propose that Moore's law was effectively done back about 6-7 years ago, when we had to shift to multi-core CPU's. Once you get to the point where you can't really increase computing power via transistor count alone, and have to resort to just cloning your cores to get more performance, I think you've hit a brick wall anyway. I think the next big advancements in computing come from bandwidth: we have so much computing power already, we just aren't effectively using it. Most computing devices sold anymore are mobile and ultra-low power low performance (think of all the smart phones, tablets, etc) - they just need connectivity to a server farm and they have all the horsepower they need, because they can outsource it.
And how would you extend Moore's law to a quantum device anyway, or to network bandwidth... How would it have applied to vacuum tube devices? To relay-driven devices? To a cuckoo clock and wrist watches? Moore's law is neat, but if we are witnessing the death of it, I say The king is dead, Long live the King - whatever new technology comes up, I hope it has as long a useful life and provides as wondrous and dizzying rate of advancement as TTL and CMOS has provided us. We need to come up with a new standard, although it's been amazing that Moore was able to make that prediction as uncanny as he did (even he didn't think it would last past the 1970's), it really only works as long as we cling to the transistor and the idea that our computers need to be bound to an internal central processor.
You pointed out a new doorway for hackers, advertisers, and even more income for developers alike.
I struggle not with life, money, emotions, and world, but against old mindsets and selves to be proven obsolete in a age and time of rapid changes. Go create fun, so you can have fun.
It's not quite as dead as we think yet.
It will go as fast as companies can make it go, we have the capability to do it at the speed and time that Moore's Law describes but wether or not it is financially viable is the question.
http://www.macrumors.com/2011/09/13/intel-previews-low-power-haswell-processors-for-2013/
[edit]
Better link