Slashdot Mirror
Michio Kaku's Dark Prediction For the End of Moore's Law
nightcats writes "An excerpt from Michio Kaku's new book appears at salon.com, in which he sees a dark economic future within the next 20 years as Moore's law is brought to an end when single-atom transistors give way to quantum states. Kaku predicts: 'Since chips are placed in a wide variety of products, this could have disastrous effects on the entire economy. As entire industries grind to a halt, millions could lose their jobs, and the economy could be thrown into turmoil.'" Exactly the way the collapse of the vacuum tube industry killed the economy, I hope.
Noone will take a disaster prophecy seriously if you can't even be bothered to pair it with some planetary alignment or ancient calendar.
Software developers are going to have to consider increasing efficiency as they make their wares more complex! And we might have to actually implement concurrency research which is under two decades old!
Who knows, we might even end up with the responsiveness of my RISC OS 2 Acorn A3000 in 1990.
Before we had transisters we didn't have them yet either.
So what? Already today the chips are just perfect for most applications. Add 20 more years of Moore's law, and we won't even need more powerful chips. You'll have the power of today's supercomputers on your cell phone. I doubt Moore's law would continue even if physically possible, because there will be no need for it.
The Tao of math: The numbers you can count are not the real numbers.
From weird analogies and a certain amount of misunderstanding things the excerpt draws strange conclusions.
a) Misunderstanding how the frequency spacing relates to required number of cycles: The correct assumption would be that if light has 10^14Hz and you restrict yourself to single-octave circuit (for the sake of simplicity: lets say 10% relative bw circuit), then you can if you "cram" ideall of modulate fast enough, 10^13bits*log2(S/N) bits per second. so probably 10^14bits/second - that is a lot.
b) limits to Moores Law: Moores law is an economic law. There is no physical limit which i see which can be reached technologically until 2020 (in mass production). There is a technological limit to what can be produced, but going in the third dimension and new materials will give opportunity to continue on the same course for a while. If you look at what physicists are currently looking at, you realize that the end of silicon/metal/oxide technology will not be the end of Moores Law or classical computing
c) "on the atomic level i cant know where the electron is". As it happens to be i work on quantum computation and i really hate to explain that: If you arrange a specific situation, then you cant know where the electron is on the atomic scale. If the statement would be as general as he makes it, it would be impossible to have different chemical configurations of the same stoichiometric mixtures. SIngle-molecule electronic/magnetic configurations. The quantum tunnel coupling in single molecule magnets between states can be designed, and i dont see a specific reason why it should be impossible to realize single molecule devices in which tunneling does not play a role
d) he does not understand FETs AFAIU
e) contrary to his opinion, very thin 2DEGs exist and i dont see a reason why upon (finding and) choosing the right layers, the confinement can be very steep in the third direction (not infinity, but also not requiring more than 50nm thickness)
The funny thing is that he forgot what already is and probably will (there *may* be ways out, like superconductors or ballistic transport but don't bet on it) really be a problem for all classical/room temperature computers: heat. While the designing smaller elements may be possible when using the right physics/technology, reducing the capacitances of lines (associated with an energy loss in the line resistance per switching) will be difficult. Once we *really* stack in the third dimension it will need a lot of clever computer scientists (and maybe mathematicians) to reduce thee needed interconnects, since otherwise stacking the third dimension wont give us anything besides memory capacity.
So to conclude: i believe until 2050 the definition of Moores law will be obsolete. but it will not break down because we are unable to make circuits smaller, but because it may be too expensive to make them smaller or powering and cooling the circuits may become impractical. We probably will have a replacement of moores law by an equivalent scaling law for power per switching.
Today's chips were perfect for most applications in the 1980s. Once WordPerfect could outrun a human in terms of spell check and could outrun even the fastest printers CPU upgrades didn't do much. Same with Lotus 1-2-3, once complex speadsheets with lots of macros could be processed faster than a human could read a spreadsheet....
But all that excess power led to the GUI. And then technologies like OLE. Which drove up requirements by orders of magnitude. But OLE hasn't really hit another generation because everything is so unstable. Imagine the next generation of applications that have data embedded from dozens of devices and hundreds of websites. I do a Quicken report which
a) contacts my banks internet connections and pulls in all the credit card transactions
b) hits each of those vendors (100+) with the credit processing number and pulls up all the items for each transaction
c) does an item lookup to figure out what sort of expenses they are and prorates out general costs, like sales tax. That's 1000s of web information requests for an annual report.
That sort of data processing we don't yet have and certainly not on cellphones. Another area is AI where systems are underpowered.
Imagine a news search engine that knows my entire browsing history. Like a Pandora across all my news choices for the last year. I search for a story and because the system knows my preferences on dozens of dimensions its able to feed me the stories that most fit my preferences. Analyzing every article every day to do simple word counts is about the limits of a massive datacenter of google. Analyzing every article every day to determine: how much scientific background is this assuming in biology, in chemistry, in mathematics; what sort of editorial biases does it have, how human interest heavy is the presentation, how respected is in the journal.... that's way beyond what we can do today.