Moore's law will be dead soon. Again

GPU maker NVIDIA declared it dead already, even Intel predicted its ultimate demise and now a futurologist is putting another nail in the coffin of Moore’s Law: Michio Kaku said that 10 years from now the aforementioned law will flat out, forcing the tech industry to find viable alternatives to the now-standard microchips made of silicon.

Kaku, an American theoretical physicist known for his work on the string theory and his role as a “science communicator”, recently addressed the issues that the microprocessors industry will face in the upcoming future: soon CPUs will not be able to double their performance every two years or so as predicted by the practical rule thought up by Intel co-founder Gordon E. Moore in 1965.

In fact, Kaku explained, Moore’s Law is already slowing down: Intel is now “tweaking” its silicon-based technology with tri-gate three-dimensional transistors (Ivy Bridge), but when the shrinking of technology nodes will come to five nanometers and beyond the laws of physics will do justice of the so-called “Moore’s Law”.

After that, Kaku suggested, there is only uncertainty: in what the scientist called the “post-silicon” era, the industry will probably tweak current technologies (optical chips, three-dimensional designs, extreme multi-core architectures) to further evolve microchips but there will not be a doubled performance per biennium anymore.

At the end of the Moore’s Law cycle, Kaku foresaw molecular (organic-based) chips and quantum computers as possible candidates to inherit the reign of silicon. In this latter case, however, humanity will have to wait until the late 21st century to see the issues of this still immature technology overcome.

Report a problem with article
Previous Story

Windows Phone dev conference slated for June, Apollo around the corner?

Next Story

RIM: Physical keyboards for some Blackberry 10 phones

14 Comments

Commenting is disabled on this article.

This is noticeable now because processors haven't really gained in the MHZ department, instead they've gained extra cores, extra cache, improved thermals, smaller nano technology is bound to come to its limit as the article states.
I especially like the idea of optical chips..

netsendjoe said,
This is noticeable now because processors haven't really gained in the MHZ department, instead they've gained extra cores, extra cache, improved thermals, smaller nano technology is bound to come to its limit as the article states.
I especially like the idea of optical chips..

The number of MHz doesn't really have anything to do with Moore's law. Moore's law states that the number of transistors on a chip will double ever 18-24 months. That's the "smaller nano technology" thing you mentioned in your post. Having a chip with more MHz doesn't necessarily mean that the chip will be faster.

soon CPUs will not be able to double their performance every two years or so as predicted by the practical rule thought up by Intel co-founder Gordon E. Moore in 1965.

That's not Moore's law, Moore's law is that transistor density will double every 18 - 24 months

Sraf said,

That's not Moore's law, Moore's law is that transistor density will double every 18 - 24 months


Which, in practical terms, means the same thing.

The King of GnG said,
Well, "going wrong" after almost 50 years means you made a very strong prediction, imho :-P

Going strong maybe? It's not a law it is an assumption of physics. Something will be invented/re-invented to cope.

Moores law is prediction rather than law which was proposed in the 60's.
So yes predictions can go wrong and shouldn't be a surprise.

still1 said,
Moores law is prediction rather than law which was proposed in the 60's.
So yes predictions can go wrong and shouldn't be a surprise.


THIS, in addition to my post further down

Indeed. I would take that even one step further, it's just a speculation. The lowest size we can possibly reach with silicon transistor is ~5nm (about 9 Si atoms in a square) and we are dangerously close to that limit (23nm i think?). Every 2x smaller we go, there are exponentially more quantum effect and instability we will need to take into account.

However: Moore's law, or Moore's speculation, should be slightly modified. Instead of Number of transistor per *square* mm doubles every 18 months, it should be the number of transistor per *cubic* mm doubles every 18 months. The next big thing is going z-axis for transistors. Modern CPU only has about 4~5 layers, we are aiming for thousands if not millions of layers.

Quantum computer is still a long way from us, it is highly efficient for probabilistic computing, but it is not optimized for binary calculation. 3D Core seems like the rewarding and easily reachable for us right now, there are a lot of research going in this direction right now, I for one, work in a lab building three dimensional processors.

still1 said,
Moores law is prediction rather than law which was proposed in the 60's.
So yes predictions can go wrong and shouldn't be a surprise.

The law of universal gravitation or the ideal gas law or Ohm's Law are all good guidelines, but of course do not always sufficiently predict what will occur under certain conditions. A law simply describes how things tend to behave. I don't see the problem with calling it Moore's Law.

noPCtoday said,
The next big thing is going z-axis for transistors. Modern CPU only has about 4~5 layers, we are aiming for thousands if not millions of layers.

Modern CPUs are not 3d. They're still planar cmos. The only 3d chips out there are proof of concept. Maybe two layers.

The layers you're thinking of are in Polysilicon and metal. They're interconnects. There is only one transistor layer. Stacking actual transistors is still a nightmare. Still some issues to be resolved, there. Thermally, it's hell. We have enough trouble as it is cooling one layer. Plus, we don't really know how to fab them. We can dice a wafer and stack them, but your expense for making the chip is now multiplied, just to start, by the number of layers. Binning also becomes a nightmare. Also, they're a real PITA to actually design. We don't know how, quite yet.

Edited by Mordkanin, May 3 2012, 12:29pm :