Slashdot Mirror
Reinventing The Transistor For Molecular Computing
unnique writes "MIT's Technology Review, has an article on HP's research into finding a new way to make transistors smaller, and further stretching Moore's law." The article has some nice illustrations of the nano-componentry they're working on, too.
It seems like we're constantly hearing this same type of story over and over again but never hearing about any substantial results...Be it diamonds, gel, or nano-technology. What does Gun-Young's research mean to me, the almighty consumer? Nothing but a few more years of speculation before anything actually happens.
on 200 gigabit nanotube memory cubes.
I am not so sure I want my chips to be living organisms. On the otherhand I am certain that the choice between faster organic computer and slower inorganic computer would be a no-brainer. I'm just rooting for the inorganics right now. Thought then there is ice-nine goo and all that to be concerned about which is not much better than a computer virus destroying all life forms.
A 'puter [not including DNA synths which incidentaly should be cautiously defended since they are potential hacking targets to 3li4e geno-hackers] passing a virus directly to a human (or some other animal) becomes a probability when the computer has a DNA factory as part of its makeup.
Amplification seems like a reasonable quick solution to hard problems of routing traveling salesmen, but make sure you don't get any of it on you.
-- Each tock of the Planck clock is a new world and here we are still life. --
The basic computing element will of course keep getting smaller and faster, until it reaches certain physical limits which cannot be exceeded. At this point, a new paradigm will be invented to provide the way beyond the limits.
How small can something be? It can be down to the molecular level. How fast can something go? Up to the speed of light. So eventually the fastest "transistor" will be composed of individual molecules, with changing states caused and communicated by light (photons).
Electricity was stated in the article as "the way" that information will be input and extracted from tiny transistor, but I think this paradigm will change! Once you get to a certain speed and smallness, electricity loses its ability to transmit information. This happens due to sluggish time response properties of the medium (capacitance and inductance and other jazz) and wave interference and delay of the electrical wave of electrons flowing.
Once a wavelength (directly related to frequency) becomes a certain fraction of the distance it has to travel, the electrical path becomes a "transmission line" instead of a "lumped element." Basically you are trying to send waves of electricity (1's and 0's) down the line too fast for the physical capabilities of the medium. So that's one more thing that complicates the process of making computers smaller and faster--getting the information out and transmitting it to other components.
That's why I was mentioning a new paradigm...because I was thinking of reading Isaac Asimov's stories that mentioned his ultimate computer, Multivac, which filled up miles and miles of space underground. He extrapolated the ideas that made the cutting edge computers of his time into what he thought the future's computer would be like--namely, huge. But of course he couldn't predict the advent of the transistor and later the microprocessor which changed everything and made everything shrink instead of getting bigger....by the way--some parts in computers, like the connectors and traces, are already becoming speed bottlenecks for some of the reasons mentioned...
Putting "ry" on the end of the word doesn't make it a plural, even in your two cases.
Well yes, that was my point after all.
Likewise, componentry is used in the fabrication of components.
OK, that sounds plausible at least. Now, are you able to back up your claim by providing some links where "componentry" is used in this sense, rather than in the "I think it's a more marketable word than components" sense? My random sampling of Google hits seems to favor the latter.
Williams reseach is good stuff, and essential. Something for the ./ crowd to realize, however: the wires forming the cross bar are still macroscopic. Although the switching component is based on molecules, the interconnects are large, and essentially will limit the density of transistors per area.
This means that Moore's law will still hold, unless the interconnects are molecules as well.
IAAMEE
A huge element of Si technology's success is the way lithography allows mass production. The problem with molecular schemes is that they involve pieces that have to be added to the substrate. William's approach of using crossbars as the basic element gets around this problem somewhat. But Si + lithography is still going to be a more robust technology.
There is also the problem that molecules are delicate objects. You simply can't make millions of molecular switches and expect them all to work. With Si all the switches work often enough that you can make chips. Williams plans on using fault tolerant architectures to get around this problem.
So, HP's program isn't as crazy as a lot of stuff I see at conferences. But it is still far fetched, and I think it will fail because it is competing with Si VLSI instead of aiming for some niche.
Si technology is damned good, and trying to compete with it has been a losing game for decades now. (e.g. GaAs and Josephson junction computers). "Novel" technologies pay off when used for an application for which Si is unsuitable (optics with GaAs, magnetic field detection with Josephson junctions).
However, I will eat my hat if in 20 years (10 years after Moore's 'law' bottoms out) VLSI is done in anythin other than Si.
Who said it has to stay two-dimensional - bring on the processor cube!
Even better, make the whole thing flexible like a piece of plastic, and fold it on itself, it could be 2 feet long, and just accordion into a little box - which a fan blows thru to dissipate heat.
Whats wrong with an underground cavern filled for miles of computer? If that computer has vast sections of itself running at 2.4+ Gigahertz, and Disk arrays the size of chevy's....wow.
Then again, there probably is something like that buried in china somewhere...I've always wondered what they do with all those parts they recycle...thats what its for people! Vast, Vast computing...with our scraps!
Who is this that even the wind and the waves obey Him? Surely this computer must submit also!
You know, alot of people talk about the death of moore's law, but uh, has anyone ever considered the possibility that moore's law might keep going and going and going ad infinitum?
It isn't impossible. Theoretically when you get down to quantum computers where your using atomic mater itself your almost at the smallest possible size for computation, until you break down the individual peices of the positrons and electrons into quarks and gluons which could possibly be used for calculation, then you think about creating an artificial black hole and stuffing ever more matter into a singularity and you could calculate the universe from something the size of the head of a pin (especially if you adhere to the multiverse theory, which states there are infinite realities). If there are infinite realities, we could litterally collapse our own reality, and possibly others nearby into a singularity for calculation, and just keep on going and going and going.
Truly as we begin to see the emergence of quantum computers we start to head towards these paths for higher and higher calculations, instead of knowing a universe around us, abit at a time. We could know it all at once, in all it's enormousness. We could then know and create others (computation being equivilant according to babbage, a computer simulating a reality perfectly is in fact a new reality as our reality is nothing but mathematical laws anyhow).
While I know moore's law can fail us at any time now being a theory and not a fact. Dismissing it as most do so casually after it has perservered time and time again for so many decades running is really getting to be rather ridiculous.
Moore's Law seems as good as Hooke's Law to me.
Hooke's Law for a spring: Force on a spring is proportional to the distance stretched from equilibrium. Until its stretched so far that the law doesn't work any more...
Special Relativity: The person in the other queue thinks yours is moving faster.
This goes to the heart of Moore's Law. Moore's Law isn't about transistor size per se. Rather, it's about the number of components that can be built on an integrated circuit at minimum cost.
In his original paper, Moore examines the effects the defect density (the number of defects in the silicon per unit area) and the size of the chip have on the economics of chip production. As you make larger and larger chips, you can put more and more transistors on them. However, the wafers have unavoidable defects in them; a physically larger chip is therefore more likely to contain one or more of the fatal defects, and be worthless.
Moore's key insight (and one that is usually overlooked) was that at any given level of technology (i.e., lithography or transistor size) there is an economically optimum number of components (almost exclusively transistors, today) per chip--that is, a number of components that minimizes the manufacturing cost per component (see the first figure of his paper). If the chip is too small, you spend too much time handling and packaging too many chips, driving up costs; if the chip is too big, the yield is low due to the wafer defects, and costs are driven up again. Crucially, Moore noted that this economically optimum number of transistors increases markedly over time, as integration technology improves; this led to his more famous second figure, showing the base 2 log of the number of components per integrated function growing without bound over time (and doubling every year, a slope that has since been reduced to doubling every 18-24 months). What is unstated in the figure itself is that this represents the economically optimum number of components per integrated fuction.
So the short answer to your question is that a chip 3 inches on a side could be made, but the yield would be so low, due to the unavoidable defects in the silicon wafer itself, that it would be fabulously expensive. It would be cheaper to make several smaller chips perform the same function, which is what is done today, if you stop to think of how many different chips are in the average PC.
Moore's paper is a marvel of prognostication; he notes in it, among many other keen insights:
He soon got his "flexible techniques for the engineering of large functions" by the invention of the microprocessor; the use of automated design techniques for digital circuits is, of course, now commonplace.