Slashdot Mirror
TeraHertz Molecular Switch Arrays
Bfaber wrote in about researchers at the University of Illinois having come up with a method to produce atomic-scale TeraHertz switches. It's possible that when attached to specifically designed molecules, these puppies would act like transistors that can switch at 100 trillion times a second. Kind of throws MHz right out the window, don't it?
Raw speed does not magic make. Imagine if AMD dropped a few thousand gigahertz Athlons on the homebrew computer club back in the pre-Altair days. Do you think those hobbyists (or damn near anyone for that matter) could have created an OS with anything like the complexity of FreeBSD? No of course not. They still would have had to climb the learning curve, building information as they went.
I'm not saying Thz won't help us eventually reach brain-type computing; what I AM saying is that an ultrafast microprocessor is not going to result a priori in a "thinking machine"...
Recall that the magic of the human brain is not a single blinding fast unit; rather it is by _MAsSiVe_PaRrAlLeLiSm_ that we believe our brains do all that info-crunching.
In other words I'm saying that when we finally DO create a silicon 'brain', I'll bet blood that the researchers turn around and say "If we only knew lemmas X, Y and Z in 1987 then we could have built this thing with 68000's." It's not the raw horsepower that counts, its the COMPLEXITY and ORDERING of that horsepower.
IMHO.
mu!
Silicon is forever. I mean, who would want to hang out with babes on beaches made of anything else?
I'm not saying Thz won't help us eventually reach brain-type computing; what I AM saying is that an ultrafast microprocessor is not going to result a priori in a "thinking machine"...
Recall that the magic of the human brain is not a single blinding fast unit; rather it is by _MAsSiVe_PaRrAlLeLiSm_ that we believe our brains do all that info-crunching.
I agree that the human brain is not a blindlingly fast sequential processing unit, but I think there are several 'secrets' that are often overlooked in 'human brain as thinking machine'
1) We define the problem and the successful outputs. In other words, we humans may be terminally screwed up in how we perceive and analyze our environment or computational problems, but we will *not* accept a machine as "thinking" until it is approximately as screwed up, and in the same peculiar ways as we are.
2) We have highly specialized circuitry for most subtasks like vision, memory, verbal and nonverbal language(nuance, inflection)... we don't even understand what all the tasks are yet. This is not massively parallel processing, it's more like my kitchen (which can toast bread in the toaster, make coffee in the percolator, cook eggs on the stove, preserve food in the 'fridge, and warm a danish in the microwave, clean last nights dishes in the diswasher, and dispense me a glass of water at the same time).
3) While these circuits are complex and specialized, evolution doesn't (strictly speaking) optimize anything by any objective standard. you may argue that we 'out-competed' some other species (say neanderthals), but the very task at which we 'outcompeted' them is undefined. it might be something as trivial as being less susceptible to the Great Mastodon Flu of 50,000BC or having a slick print shop who let us get our IPO brochures out faster.
4) this brings us back to #1: we don't recognize anything as thinking that doesn't closely match our own screwed up thinking. Once upon a time, doing math was enough - but they beat us blotto at that. Then it was chess. Similarly blotto. Then it was conversation (the turing test), which *guess what* means simulating us.
Soon "thinking" will mean the ability to surf pr0n with your left hand, while flaming M$ with your right (without wondering, as a sensible Flesh-o-matic 2020 might, why you were flaming M$ instead of.. never mind)
__________
If you can go to bed, knowing you did a valuable thing today, you're very lucky. If you can't... it's not bedtime
For the record, as far as I can tell, after a little background surfing, and some BOTE calculations (similar calulations were often 'background exercises' for the student of molecular biology ) it appears they are talking about:
10 femtosecond (e-14) switching times NOT an operating speed of 100 terahertz (e14) The term "femtosecond switching" will allow you to more accurately find existing work in the field. Switching in sub-10 femtosecond range has been around for years, at this same 'bench theory' level of investigation.
This is a very interesting piece of work, but hardly a breakthrough when 2 femtosecond capacitor switching was announced in 1997 (I had my doubts then, but didn't check it out) and 2-5 femtosecond laser optical switching has probably been around even longer
You can immediately deduct 1+ order of magnitude from the risetime to get a practical operating speed (you want digital square waves, not sawtooths, right?) even when this switching speed becomes a practical reality.
You can also deduct a few orders of magnitude from the operating speed of a single switch to the operating speed of a CPU or RAM. Think about how many sequential transistor operations there are in a single RAM bit (on-chip, on-card, and system transistors)
And now, as a public service to those of you who need a refresher (we'll all need these terms soon enough)
__________
If you can go to bed, knowing you did a valuable thing today, you're very lucky. If you can't... it's not bedtime
I think the question is 'How are such rotational frequencies measured?', not 'how do they get such frequencies in the first place'.
... hydrogen's energy spectrum is WELL known).
After reading the article, I'm not sure if they actually measured such frequencies, or just presented theoretical calculations (it's mostly a basic quantum mechanics problem involving hemispherical potential wells
Eric
100THz corresponds to 10^-14 seconds (10 femtoseconds). Light can travel approximately 3um (3 millionths of a metre, or a little more than one ten-thousandth of an inch) in that time. While 'nanotech' will make some Very Small Systems Indeed, I'd say it's more likely that we'll see asynchronous subsystems within the computing devices of the next few decades, running at very high clock speeds, and communicating with each other over comparatively high-latency links (you know, down in the picosecond range... :)
The dang thing eats > 99% of my PIII CPU doing paperclip animations and futzing with those summarized/unsummarized menus. I'm running NT because my development environment crashes so often that running a ms-dos-based environment (Win98) just wouldn't cut it.
I guess the post here it way too enthusiastic, since even in the original article they never mentioned some basic facts.
Read it well : it's only theory.
And there are a few questions to be asked :
What about the stability of the memory ? What are the operating conditions ? If it has to be kept below, say 200K it will be quite difficult to use it, wouldn't it ?
What would be the needs for producing such a memory ? Because it's quite hard to find pure silicon in nature (and producing itis quite expensive I imagine)
And last but not least : how about compatibility with the actual technology ? I guess none of you is actually imagining that this will be joyfully embraced by companies that are strong on the memory/processors market.
This may actually end up being buried by marketing, because nowadays an university doesn't really have the possibility (financially) of designing a competitive chip/architecture.
So I guess I will be a bit reluctant towards its success until I see it on the commercial ads of AMD or Intel.
Reminds me of that high school teach of mine. One day, he was telling me about his 1K RAM card he got for his Altair. Came with 1-256 byte chip. He told me (as we were installing the 48K RAM card) that he thought if he ever fully populated that 1K card, he would have more RAM than he would know what to do with.
Seems that most "barriers" in the computer industry are not real; they are merely perceived.
cat
But does anyone find it strange that Moore's law is so consistently true? Why does computer hardware advance at such a steady rate? What exactly is Intel doing in their lab that allows them to make a 1GHz chip now but not a month ago? And what will allow them to make it at half the price 18 months from now? When was the last time somebody took advantage of a discovery like this?
Could it be a conspiracy to keep the power of hardware and the requirements of software in sync? If Intel did come out with a 1THz chip right now, everybody would run out and buy one and then nobody would need another chip for years. I smell a conspiracy.