Slashdot Mirror
Gordon Moore: Moore's Law is Dead
Golygydd Max writes "Moore's Law will not hold forever, claims Gordon Moore.
In a Techworld article, he points out the limitations of the law, in particular, the limitations as we approach the size of atoms.
He helpfully explains, however, that the law will hold for a few years yet." Still, sticking around for forty years is pretty impressive.
You know, seems to me that as long as I can remember using computers, people have been saying Moore's law can't hold out forever. And, while, I guess, logically, that has to be true, it seems to be out-living most of these predictions. A lot like Apple and FreeBSD :-)
+Pete
Score:-1, Funny
The "law" will be stretched to include multiprocessing and a multitude of other imporperly attributed leaps in technology... (this helps to solidify how much BS is so called science)
People are clever. They figure out ways to do things that seem impossible. While the physical laws of the atom will be a barrier, I have faith that we will work around them (so to speak). Perhaps getting atoms to do multiple things at once (who knows). But don't bet against a breakthough with economic gain at steak.
"Those that start by burning books, will end by burning men."
Actually, I would think this would lead to, not prevent, innovation. The engineers are more likely saying "we've reached the limit. WTF do we do now? We can't just make it smaller..."
Against stupidity, the gods themselves contend in vain.
"Moore's Law" is a theory about innovation, not a law in any way. Sure it's fun to call it a law, but it has no basis in physical phenomena, and it's breakable-- Moore himself says it should run out. Scientific laws don't expire.
Many people have used Moore's Law to loosely talk about computer power doubling every x months. Interpreted that way, Moore's law could survive quite a while longer.
Having said the above however, exponential growth always ends when it bumps into physical barriers. Otherwise the planet would be covered a thousand feet deep in dead flies (who as we all know reproduce exponentially when the environment permits.)
I guess Ray Kurzeil's predictions that computers will have the same power as the human brain by 2020 will not be met...
It can be met right now using clusters. The technology is here now. The problem is that we can't even make a machine as intelligent as a honey bee (only about 1 million neurons), what good would a system with a hundred billion neurons be other than to sit and vegetate?
But few if any of those involve exponential improvement.
You're right, it's going to lead to other innovations: we'll might start seeing expansion in a "wider" direction becoming more common than "faster" chips. (128-bit architectures, with the next step to 256 bit machines, etc.) And/or engineers will focus on different problems, perhaps something like coming up with innovative ways to dissipate on-die heat. Things like this usually lead to other breakthroughs, too. For example, the more efficiently you can get rid of heat, the more layers you could stack on the chip. Technically, the transistor density wouldn't increase, but the transistor count on a single chip could be multiplied by orders of magnitude.
John
If you can double the density of your transistors anymoore, you still can fake it, by doubling the number of cores every year, as Intel and AMD will do. Another thendy trick is to add units for hardware threads... But, if you can figure out how make several layers of cores, the density will double every year again, mixing DVD technology and CPU manufacturers projects, this is the commercial version of moore's law...
Yes, because this time the guy pointing out is the inventor of the law himself!
Very contradictory: The title is "Moore's Law is Dead" but then the article states, "He helpfully explains, however, that the law will hold for a few years yet."
I guess "Moores Law will hold for a few years" isn't as much of an attention grabber, but at least it's honest.
Ah, but by your parenthesized comment, you yourself met the conditions of Godwin's law-- and more importantly, by meeting it in your own comment, you met the second condition!
Great job, I like such a self-encapsulated prophecy :)
Allegedly real newspaper headline from 1998:
Man Struck by Lightning Faces Battery Charge
It occurs to me that following Moore's law as an "industry standard," so to speak, would be a good profit source--as a chip manufacturer, you don't want to put forth your absolute best product prematurely and then developmentally stagnate for the next ten years; you need to pace yourself and drop your products gradually, at gradually increasing quality levels. Moore's law would be a useful measuring stick against which to consistently increase quality without going in too much too fast.
N0, everybody knows that Finagle's Law is the supereme law of the Universe.
Cypherpunks: Civil Liberty Through Complex Mathematics. Those who live by the sword die by the arrow.
However, most common math does not involve some physical matter that shrinks exponentially. It's really the exponential part that is impressive. Exponential growth over a couple of year is not such a big deal, but 40 years is huge. The 1965's chip had 60 devices (transistors + resistors) and today's chip have 1,700,000,000 transistors... if that's not impressive growth, I don't know what is.
After 3 days without programming, life becomes meaningless
- The Tao of Programming
Silicon is usually etched as a single-sided, flat medium. Of course, the wafer has two sides (doubling the usable surface area, if you can get rid of the extra heat fast enough), and space is three-dimensional, which means that transistors don't need to take real-estate on the wafer itself.
Finally, and this is what would eliminate the upper limit problem, you'd need an N-state transistor. In other words, one that could handle N-state discrete signals, rather than binary signals. Then, you can fold as many binary transistors as you like into a single physical device.
Of course, Intel being Intel, the sun will have long since faded into a white dwarf, long before we see any of these - or any other technology for saving Moore's Law over the long term - put into practice.
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)
Why is the size of atoms a limitation to the computational speed?
There are many different bottlenecks in a system besides the main CPU and even for the CPU there are sub-atomic particles that can make a difference. For example photons have many possible quantum states which span through dimensions we don't even understand yet.
I believe that the law that he is speaking of fails in the Newtonian physics arena but there is a lot more to information processing. Look at a human brain for example. Do you think that the human brain is slower then the speed of a CPU in 3 years from now?
Ever thought that maybe Moore has something to do with why CPUs don't get faster quicker? The industry is clocked at the speed defined by Moore's law. Overclockers have proved again and again that Moore's law is not really a law but a rule of thumb.
First, Moore's Law is about transistor density. If you use these nano-crossbar thingies instead of transistors, Moore's Law no longer applies. Second, even if you allow that crossbar nano-whatsits are the equivalent of transistors in terms of Moore's Law, it still can't extend out to "near infinity", as there is an easily calculable finite limit to how small you can make a mechanical device.
If a job's not worth doing, it's not worth doing right.
If you look beyond size and clock speed, there are lots of innovations in modern CPUs that are only present because we can put such a large number of transistors on a chip. Branch prediction, instruction reordering, etc all take up large amounts of space, and only increases in transistor density allowed them to be feasibly implemented in real-world commodity chips. Plus, there have been many advances in fabrication technology and material science made as byproducts of living up to Moore's predictions, like strained silicon or silicon-on-insulator or low-k dielectric.
Considering that a long term corporate plan is about... 3 months, it makes sense. Moore was saying that there are like 10-20 years left of density doubling. That is way beyond how far ahead CEOs look, so it is out of sight to him.
I'll never make that mistake again, reading the experts' opinions. - Feynman
Moore's law has stuck around for forty years in the same way that my pet hampster lived for ten years. It died but got replaced by something similar with the same name and nobody noticed.
In theory of science, a theory is a hypothesis that has been been strengthened through many experiments, and never been falsified. The concept of law doesn't exist in the theory of science. It's just an unfortunate fact of history that some well-established simple-to-state theories have been known as "laws".
In everyday language, a law is something drafted by legislators, and used in courts. A theory is either (a:) a rough guess, or (b:) something that scientists come up with to explain things that would be hard to understand without them.
Thus, yes, perhaps Moore's law should be called Moore's theory, since theory meets both everyday language standards and scientific language standards better. But that isn't restricted to Moore's law. E.g. Newton's laws should also be called Newton's theories if you follow this argument.
On the other hand, it's hard to change things that works. When someone speaks of Newton's laws or Moore's law, the listener know exactly which law(s) the speaker intended. If you keep renaming stuff, it hinders understanding. So, in response to your question, I would say NO.
Enough of the linguistic perspective. What you probably wanted to say, was that Moore's law is not a scientific theory we can put the same faith in, as e.g. Newton's laws. That could be true. If that's the case, then it would be better to rename Moore's law to something like e.g. Moore's observation.