Metaphorically speaking, that is. In the early 1800s a number of governments and corporations spent huge sums of money to build networks of canals. They were, after all, the fastest and cheapest way to move bulk cargo around. Almost all of them went bankrupt shortly after completion. Not because of mismanagement, but because of the invention of the railroad. They could haul more cargo, faster, and could be built quicker than canals, it was only natural that railroads would come to dominate so quickly. Of course, when the canal projects were first started, nobody had a clue about railroads, and even when the first railroads were built, it wasn't obvious to people how transformative they were going to be. People even wondered if human beings could survive conditions which existed at such blistering speeds as 20 MPH or faster! Today, of course, if you were going to suggest building a canal to carry people as part of a mass-transit system, folks would think you were nuts. (Unless it was a tourist route to show off the sights of a city. For workday commutes, they're a lousy idea.) However, we're currently doing something similar to proposing building a canal, instead of a rail line, for mass transit in the 21st Century. Some years ago, IBM researchers (and forgive me for not posting a link, as I can't find one ATM, which talks about this) decided that they'd like to skip ahead to the endpoint of Moore's Law and see just how small we can build a transistor. This wouldn't be a full-fledged processor, so it would be incapable of doing much of anything, but it would tell us a lot about how small we could shrink computer chips, and the kind of technology that would be needed to build them. The end result was a transistor that was made up of 12 atoms. That's it. Below that point, they couldn't hold on to the electrons. When asked if they were going to put the technology into production, IBM said "no" because of certain limitations of silicon, and it would be too expensive to convert over existing chip fabs, though they expected that as technology progressed, by the time we got to ~2045 (when Moore's Law is projected to run out), it would be no more expensive to make such chips than it is to make conventional chips today. Superficially, this makes sense. Its like planning a trip cross-country, and deciding that instead of shelling out thousands of dollars for a plane ticket, you'll spend $50 to take the bus. You'll get where you're going, but it'll just take a little longer (a week, instead of a few hours). Assuming the horrors of riding the bus are no worse than being groped by the TSA, its merely a trade-off in time. (FYI, I'll take TSA groping over riding the bus any day. Greyhound should be nuked from orbit.) This ignores, however, what happens when you get to your destination. Say you've been offered a job on the opposite side of the country from where you now live. Not only is it your dream job (which is why you're willing to pick up stakes and move so far from home), but it pays ten times what you're making now. In that situation, you'd be foolish to take the bus over the plane, because you're dramatically cutting into your income. As the pay increase goes up, it becomes even more foolish to take so long to get to your new job. Effectively, the folks at IBM looked at an increase of 32 times their current pay, and opted to take the bus. I don't know about you, but if somebody offered me 32 times what I'm making now, not only would I be on the first plane out of here, but I'd probably blow my boss when I got there. Granted, there are some issues involved with dumping current fab technologies to building such chips, and over the 32 years between now and when we hit the end of Moore's Law, we'll figure them out. (Heat dissipation being the primary one.) However, we don't have to scrap every existing fab plant to see the benefits of jumping to the end of Moore's Law. We just have to convert one plant, and sell the chips produced to the right people (Elon Musk would be at the top of the list). Why? Because a chip produced in such a fab, would have the same dimensions as current chips, but would be 100 billion (that's 100 Carl Sagan units) times as powerful as a human brain. With such a chip (and the right software, I'm thinking Musk knows a few software engineers), we could figure out an economical way of producing such chips in mass quantities. (A lot of high end engineering software does this to a limited extent already.) Think of the benefits and how transformative of our society this would be. It would revolutionize everything. Not only would our ability to build computer chips improve, but things like nanoscale manufacturing would be possible (and cheap). Combined with adaptive software (which already exists), and just like Picard, we could say, "Computer, tea, Earl Grey, hot." and *bam!* it'd show up in a few minutes. We'd be able to do this with anything we might want (provided the laws of physics allowed it to exist, of course). I will get to what this really offers us in a bit. Granted, the issue of heat dissipation with silicon computer chips is a huge one. Had we not switched to multi-core designs a few years ago, chips would be producing as heat as a nuclear reactor by now. However, that's no longer an issue, as Stanford scientists have developed the first working computer which uses carbon nanotubes instead of silicon. This means that heat and power consumption are no longer an issue when it comes to chip design. So, you could have a high powered smartphone and not have to worry about needing a battery the size of an aircraft carrier to get you through the day, or it incinerating you when you wanted to play Candy Crush.
So, effectively, we have the generalities worked out about how to build the ultimate computer, and once we're able to get the kinks figured out about chip production (and carbon nanotubes could be used as 99% efficient solar cells, so we'll have lots of energy), we could have the most powerful computers imaginable along with Star Trek-style replicators. It gets better. Remember when the original iPhone was introduced, and how amazing it seemed, only to be quickly surpassed, not only by Android phones, but by the explosion of various apps that now are pushing smartphones to Tricorder territory? Imagine what will happen when the 2045 model computer chips hit the market. Then, realize that you're underestimating what they can do. And not only are you underestimating the capabilities, but so are the people who're experts on such things. Case in point, the guys at Planetary Resources. Their goal is to find asteroids rich in rare earths and other precious metals, have self-replicating robots mine them, and then send the ore to Earth. This will crash the metals market, and make it possible to use those materials in areas in which its not presently economically feasible to do so, while at the same time, preventing the environmental damage which comes from mining operations. A noble idea, and one that, along with replicator-style technology, will make money (at least as we know it), obsolete. It does not, however, encapsulate what will be possible to do, by any significant degree. Musk and the others connected with Planetary Resources have admitted that once they recoup their startup costs, the production of materials mined from asteroids will essentially be free. Musk has rather famously said that he wants to retire on Mars, and that many people will be able to buy tickets to Mars to retire to, once they've sold everything they own on Earth. Of course, guys like Musk tend not to retire (look at Warren Buffett, still going strong at 83), but also who's going to buy your crap, if money is effectively worthless? Musk hasn't given too many details as to how he expects it all to work out, but given the kinds of things he has said, its not too difficult to get a rough idea of what his plan is: Self-replicating robots would mine asteroids for fuel and materials, these would be used to build the ships which would take colonists to Mars, where they would land, and find self-replicating robots had built a retirement community and had it waiting for them. (I can't help but think of the retirement communities of the future as imagined during the 1950s as I type this.) Your costs would basically be buying land on Mars, and paying for the fuel to get you into orbit. A good vision, and with luck, one that I might live to see (assuming things like anti-aging treatments don't happen, if they do happen, then I'll most likely live to see it). Does it, however, accurately reflect what will happen? I mean, after all, we've been promised flying cars for decades, and they've never shown up. Why should we expect that Musk can pull of his vision? Then there's the zillions of other things we've been promised to have in the 21st Century that haven't shown up as of yet. In the case of the flying car, there's really been two large issues preventing us from having them. One is the amount of energy that such a car would require. Planes take a lot of fuel, and those capable of VTOL, require even more. The second is driver training. Take the vision of the flying car so often presented in the '50s where dad drives through the sky just like the '50s-era dad drove on the shiny new interstates. Now, suppose somebody in 1959 had figured out how to build a cheap, economical flying car, that used as much gas as the average car from that era and was just as easy to drive as a contemporary car. Things would have been fine, until cellphones and texting came along, at which point, the carnage caused by distracted drivers crashing into one another and then falling to Earth, would have been horrific. We're now getting to the point where planes can practically fly themselves (as can UAVs), so if we lick the fuel problem, we could actually have flying cars if we wanted them.
Still, some of the stuff they predicted is impossible simply because the laws of physics prevent it from happening. Or its been supplanted by a technology which no one could have foreseen at the time. For example, people once thought that you'd have a printer hooked up to your TV, and this would spit out your morning paper, so it'd be ready for you at breakfast. Now, of course, many of us get our morning news via our computer, smartphone, or tablet. Who needs to print it out? However, if you take just the things which Musk has said that he wants to do: mine asteroids using self-replicating robots, build colonies on Mars with similar robots, and you combine it with what we know is coming in computer technology (even if a computer with a 2045 chip in it isn't sentient, it'll be able to apply "brute force" to solve tasks far faster than we can today) there are some huge possibilities for the application of the technology even Musk isn't talking about. If your effective cost to design and build something using 2045 computers, 3D printers, fully automated robotic systems, and whatever else we come up with by then, is zero, then why dick around with something like colonizing Mars? Building a generation ship (even if its powered by throwing nukes out the back), costs you as much building a rocket to take you to Mars (i.e. nothing). You don't even have to do the design work, you just give the basic parameters of what you want to your computer and it figures out the rest. Granted, you do run the risk of launching your generational ship, and then having someone invent warp drive a couple of years later, thus obsoleting your work, but its not like they couldn't pick you up on their way. Still, you're pushing some pretty big boundaries as it is, and unlike the 2045 computers, its not like we've got the broad details for warp drive already figured out. I bring all of this up, not to slam Musk and others like him, but merely to raise awareness of what is possible, if we're willing to fully reach for what we can see. Software tends to lag behind what hardware can do (the first X-64 chips were out before there was any software which could really take advantage of them), and if we continue on our current track, it will be some time after the 2045 chips are introduced before they can be fully utilized. If, however, we decide that we're going to invest the bulk of our resources into building the 2045 chips as soon as possible, we'll have the chips before 2045, and odds are, we'll have the software that can fully exploit those chips long before we would if we waited. Imagine, then, not signing up to retire on Mars, but to travel to a planet in some other star system. Perhaps medical technology will allow you to live long enough to reach the destination, but even if it doesn't, you will have embarked on a voyage far greater than humanity has ever known. In less than 100 years, we would have gone from men landing on the Moon, to heading towards the stars. That's as remarkable as if Neil Armstrong had planted his boot on the Moon in 1592, instead of 1969. All told, the cost to do this, would probably be less than the total amount (adjusted for inflation) spent on the Apollo program. Something to think about, no?
Hang on, you're criticizing IBM for not immediately jumping to that level of chip development? The issue with Moores law isn't knowing what we want to build, it's knowing how to build it that is the hard part. If any of the large tech companies could jump to that sort of technology now they would do so without a moments thought regardless of the cost, the cost would be outweighed by overnight destroying their competitors.
...and, while this was interesting reading (until it became TL,D), you're kind of shooting your metaphor in the foot. In the 1800s, people went nuts building canals because that was what we knew and we put a shit-ton of money into it. Who is to say that by the time IBM sank a shit-ton of money into figuring out how to build these processors some smart Indian in a lab wouldn't figure out a way to harness unicorn farts for an entirely new (and cheaper) method of quantum computing?
I think Bailey and Volpone make some good points on the specifics, but I largely agree with the general idea that we are on the precipice of massively accelerating change and the end of resource constraint.
Well, yes, that's the theory, and if it worked that way in the real world, then Tesla never would have gotten out of the gate with their electric car, because the big automakers would have already beaten them to the punch. That's not what happened however. Nor is it a unique case. Who owns the largest search engine in the world? Google, not IBM, not Apple, not Microsoft. All companies hit a point where they adopt the attitude of, "We'll just keep on doing what we're doing, and people will continue to buy our product." this is when some small, upstart company comes in and totally overthrows the markets, as Apple did with the iPod and the iPhone. Had IBM said, "Ok, we've got no idea of what the chip design of a 2045 processor will look like, but we do know how small the transistors for it will be, let's just modify our current processors to use those transistors." there still would be a massive bump in performance, because just cramming those transistors that much closer together would give you a speed boost. One which could be used to do another leapfrog in technology. The same argument could be made about Tesla building electric cars. Indeed, Honda and Toyota are promising to come out with hydrogen fuel cell vehicles at the same price as their current cars cost at roughly the same time Tesla's low-price models are due to hit the market. Fuel cells have several advantages over electrics (ATM, at any rate, some of the advantages will be going away sooner or later, like recharge times and battery costs), so shouldn't Tesla just throw in the towel? Of course not. Look at the result of the end point of building the ultimate processor, however: Money is worthless. I have tried and tried to think of a way that we can still have money in a society where Trek-style replicators exist. I can't think of any way in which money will have meaning. The two most likely scenarios aren't really all that enticing. The first is something like Bitcoin, where your net worth is determined by how much processing power you have available to you, but that doesn't have much going for it, since you'll be able to increase your processing power by dedicating a replicator (or two) to building yourself more machines. We'll then end up in a situation like we have with the crazy investments on Wall St. where money is spending most of its time being shuffled around, instead of being invested into things that actually produce meaningful items. I don't know about you, but if I've got a choice between investing my money into giving me a fat bank account that I can dangle in front of people, or I can put it into building myself a rocket and going to freakin' Mars, then I know where I'm sticking my cash. The other potential scenario is in identity. Right now, one of the things that extreme geeks are doing is signing their mail with a cryptographic key in order to prove that the email comes from them. They register this key with a central service, and you gain status by the number of people you have who attach their keys to yours (and when you've exchanged keys with people, you both can send encrypted mail to one another). So, potentially, having a large number of people attached to your key could be seen as having a fat bank account. But what would you use it for? I can't think of anything. So, say IBM spends a couple billion dollars to knock out a few 2045 era processors and a couple of weeks after the first chips roll off the fab line, some guys in a lab in India, China, Vietnam, or wherever, manage to build the same kind of chip for a few pennies. IBM's screwed, right? Wrong. Once those chips hit the general public, then money becomes worthless, and IBM's investment is no greater of a loss than that anyone else has experienced. We're all simultaneously broke and incredibly wealthy. Our money's no good, but we don't need money to get what we want, so who cares? More importantly, we have to start shifting our thinking now to the end point, and not just what's going to happen next week or next year. Presently, if you're a terrorist nutjob and you want a nuke, you either have to steal one (not an easy job) or convince enough people with money and ability to build you a covert nuclear weapons program (an even harder task). When we've got replicators, the nutter doesn't need to convince anyone to build him the nuke. He just tells his replicator to do it. What happens then? Do the nutters decide that they'll simply pack up and move to another planet where they can be the mad dictator they've always wanted to be? Or do they decide to take us all out with them in an orgy of destruction? I don't know, but I don't really want to stick around to find out, and 30 odd years isn't enough time for humans to get past their stupidity to the point where the unhinged among us are no longer interested in shoving their views down other people's throat at the point of a gun.
I think that IBM should try to build those transistors. But I'm not sure why you think we are going to have replicators. I think we are a long way off before we can figure out how to take matter and mold it into other matter.
Something else to consider is that IBM or any other of tptb has no real incentive in the present to usher in this future in which they become irrelevant. IBM is arguably better off now, at the top of a disparity pyramid, than it would be 10 years from now, as an equal to everyone who bought the replicator jr. So why invent an affordable version of this future?
Except that we still need 1: raw materials to make what we want (no asteroids are being mined yet) and 2: space to store it in.
Technically, we don't need asteroids to mine, we just need an environmentally friendly method of mining the materials we need. Instead of the suitcase sized robots this article describes, imagine nanobots doing the same things. Nanobots would be even more efficient at sifting out the trace elements which can be found in soil. So, you build a bigger house. If you don't have a house, you have your nanobots build you an island in a lake or off-shore and then put a house on it.
Oh, and that's assuming you're some kind of perfectionist who has to have, for example, his replicated Ferrari out of the original type of materials. If you're someone who wants, again going with the Ferrari example, something that looks like a classic Ferrari, but want to have things like reliability, airbags, or even the option to have the car drive itself, you won't mind having it made out of carbon fiber, when the version it was based on was built before the introduction of carbon fiber, or having it all electric, or what have you.
And just as an FYI, I fucked up the links and just went back and fixed all of them, so they should be working now.
There are MANY ENORMOUS TECHNOLOGICAL HURDLES between building a single transistor in a lab and being able to cost-effectively mass produce chips with 100 billion of them. Knowing what the future looks like--particularly with transistors--isn't that hard: you can't go wrong betting on smaller, faster, and lower-power. But it isn't the destination that's difficult; it's figuring out how to get there. Aside from that, there's a really big reason why we don't just skip to the end: it would be an enormously expensive endeavor, so much so that failure would easily bankrupt the largest company you can think of. Think about it: the large semiconductor manufacturers today spend a BILLION dollars or more on each state-of-the-art chip fab. And that's just to fulfill the logical progression from previous generations. Transistors are still much, much larger than a dozen atoms. Atomic scale transistors are likely going to require a whole new manufacturing paradigm; photolithography is not going to cut it. Revolution is much more difficult and expensive than evolution. (I won't even go into the issues with providing I/O to a processor built with atomic-scale processors; remember, to keep a processor operating, you've got to be able to continuously feed it data from the outside world. This means inventing new computer architectures, etc.) I know the future of space travel is a ship that can take off vertically, attain orbital velocities without staging, carry passengers and crew safely and reliably anywhere in the world in an hour or two, be so cost-effective as to be commonplace, and be immediately reusable after it lands. But how do I invent the 100 years worth of stuff that will be needed to make that happen?
Again, even simply building a conventional processor with the smallest possible transistor would be a huge boon. Operations which now take hours would only take minutes. Sure, if you're going to convert an entire fab over to a new process, a billion dollars (or more) is about what its going to cost. However, that's not necessary to see the benefits from such a chip. Every chip maker in the world has a prototype fab maker that tests their new versions of chips. Switching one of these over to make a few hundred chips, isn't a huge outlay. I've only played with mid-range engineering software, and the amount of things that are automated in it are impressive. Change one item in it, and not only can it propagate that change through the entire drawing (as well as related files), but it can find where those changes will cause problems and alert you to them, while suggesting solutions. It'll also analyze the materials, as well as the stresses and loads put on them, and tell you if you can use different materials or different thicknesses of materials, to achieve the same results. The high-end stuff is said to have even greater automation capabilities and better precision at figuring out potential problems. Granted, I'm assuming that chip makers have comparable software to Solidworks and Pro-Engineer, but I don't think that's too much of a stretch. So, in theory, they should be able to use those chips to figure out faster ways of making 2045 processors. Even if they can only manage to bump it up to 2025 or 2035 models, the gains would be incredible. IIRC, that has been an automated process for some time now, as its gotten too complex for ordinary humans to figure out. Even if it hasn't, that's not an issue, if you just adapt existing designs to smaller transistors. Which, again, will yield huge payoffs in terms of speed. There are people working on that problem: http://100yss.org For the moment, let's ignore that, and let's also ignore the idea that we could, at a very high price, come up with the 2045 processor in the near future (were we willing to ignore the high cost of producing such a chip and the cost of developing the software which could take advantage of that improved processor). Instead, let's take current chip design, and software, and make that our limitation, but have the transistors only require 12 atoms. For simplicity's sake, we'll say that this cuts computational time by 4. (So, if an operation with a conventional chip takes 4 hours now, it'll only take 1 hour with our improved chip.) That's a huge benefit, and allows us to work out a great many more things in far less time than we're currently able to do. So, why do I think that folks like IBM don't try to make such chips? Complacency. They know they'll get there eventually at "X" cost, so they see no reason to skip ahead for "X+Y" cost, and it makes as much sense to do that, as it does to spend the money on a manned space program which will yield lots and lots of benefits (most of which cannot be predicted at this point in time) months or years later after the initial investment. Politicians can only see to the next election, and corporations can only see to the next earnings report. This is why the biggest innovations tend to only come from startups or companies desperate to stay in business. (The former is more likely than the latter, of course.) Current projections are that chips produced in 2017 (roughly 4 times current processing power) will have the same processing power (and much greater speed, of course) as the human brain. Even if we're unable to create an AI system that allows a PC to think like a human, the speed bump alone will give huge dividends. You and I want to get there now, because we can see all the benefits this will give us, corporations don't want to get there now, because they're more worried about the cost to travel that road, than the fruits to be found at the end of it.
There was a time at the beginning of the 90's when HP and Intel figured that state-of-the-art microprocessors were about to reach a practical performance limit and needed to be replaced by an entirely new microprocessor architecture. Microprocessors at that time were mostly RISC (reduced instruction set) processors. The general idea of RISC is that the processor only supports a very limited amount of instructions (a reduced set), but executes these instructions with astonishing speed. The new architecture HP and Intel had come up with was called EPIC - explicit parallel instruction computing. The idea with the new architecture was that instead of having the microprocessor execute one simple instruction after another, it would be beneficial to have the microprocessor execute many instructions at the same time. The only real problem with this architecture was figuring out exactly which instructions should be executed at once. This was not something the EPIC microprocessor would ever be able to achieve all of its own, so the architecture required the development of new software compilers to optimize and parallelize programs before they were executed by the processor. After pouring many billions of dollars into the project, HP and Intel came to realize that it was much harder than they ever expected to create such a special compiler. But the whole idea of EPIC was centered around this special compiler, without having this compiler the processor was essentially useless. It took both companies many more billions of dollars to bring the EPIC processor -- now called Itanium -- to market in 2001. It was a total disaster. The Itanium was not only expensive, but its performance was not even on par with RISC processors of that time. In the decade that it took to create the Itanium, classic RISC processors had advanced by orders of magnitude, and it turned out to be impossible to build an Itanium compiler that would be able to utilize the theoretical superiority of the EPIC architecture. It had become the single most expensive project in the history of microchip computing. And this was only a new microprocessor architecture, it was not even based on new production methods. Imagine what would have happened if the Itanium project had required the development of new production methods and factories. Imagine what would have happened if HP and Intel had bet every dollar they had on this extremely promising new architecture. Both companies would have gone bankrupt in the early 2000's. Sure, atomic-scale processors sound like a great idea. But we don't know what pitfalls await us on the road to actually implementing the technologies that would be required or yet to be invented in order to produce such a computer. Progress happens incrementally, and technologies need time to mature. The best ideas may turn out to be impractial or impossible in the long run. And, keep this in mind: when even a "simple" architectural change takes two of the largest microprocessor companies almost a decade to implement, then an enormous technological change like atomic-scale microprocessors could easily take even the largest companies with the brightest and most intelligent people even longer. So we may end up with 2045 in any case.
Again, no one knows how to mass produce a transistor that small. The existing paradigm doesn't work. Let me see if I can put this into perspective. We have processes now that can produce 14nm features (maybe even a little smaller) in semiconductors. Very tiny. Big compared to atoms, but very tiny. These features are so small, they have to be created with light of a similar wavelength: X-RAYS. We're able to etch millions or billions of these features at a time onto silicon using x-rays. You have to be able to do millions or billions because that's what a modern microprocessor requires. So, we can lay down a billion transistors by exposing silicon with x-rays through a mask. The engineers at IBM manipulated individual atoms (with a scanning tunneling microscope) to create their transistor. That process doesn't scale to make millions of them. You can't (practically) build a processor by individually placing atoms. Photolithography is out because you can't etch to the atomic level; even if you could, you'd need very, very high energy gamma rays to resolve features that size, and, aside from the obvious safety issues, gamma rays are nearly impossible to focus. So, without some kind of process to make lots of transistors at once, it just wouldn't be practical, no matter what size the operation was. Solve that and, sure, atomic scale transistors become feasible...and you become fabulously wealthy. For most any computing task you can name that will be done with a single processor at some future date, you can do it now with enough multiple processors or with more computing time. If you tell me "Make this chip and it will have the power of a thousand processors," I'll say that "currently, it's easier and cheaper to use a thousand processors." This is not an issue to be hand-waved away. I/O bottlenecks are a significant problem with current processors. There is no simple "you just do X..." solution. And who knows? Perhaps in 100 years, they will succeed. But even if you knew what the technology of the future did, it still doesn't mean you can build it today. If there were some (realistic) way for a semiconductor company to jump ahead, I think they would take it. After all, AMD fights tooth-and-nail against Intel and proudly trumpets any performance benchmark they win, even if it's a small amount. The "there's fabulous technology out there we could be using except people/companies/politicians are too invested in the current situation" is, ultimately, a conspiracy theory. If someone could do it, someone would be doing it. Again, conspiracy theory. Today, companies invest billions for a marginal advantage over their competitors; there is little complacency. If someone came up with a new idea to make current transistors, say, twice as fast or twice as efficient, they would become very, very wealthy. Orders of magnitude faster? That would be a path to owning the ENTIRE semiconductor business. If I was a semiconductor company executive, I wouldn't suppress such a technology; I'd try to have as many shares of that stock as I could! As for AI, the problem is not processing power. Sufficient processing power is all but inevitable. The problem is designing software that actually thinks. And software that is creative in the way humans are? Decades away, if not centuries. (And I would LOVE to be proven wrong.)
No problem. I'll just find a way to invent a computer that's decades ahead of its time and have it do all the work for me.
Negative. We do NOT use X-Rays at this time. X-ray is still in development and every year the anticipated deployment of X-Ray litho gets pushed out another year. We're still using ArF (193nm) Excimer Laser light on immersion scanners.
I was hoping @Chaos Descending would show up to provide some industry expertise. So, in one sentence, he manages to repudiate everything? Impressive!
I stand corrected. We're not to x-rays yet. But that means we're even further away from atomic scale fearures than I imagined.
