While I understand the convenience of compact light sources, I'm curious as to why we wouldn't just use mirrors and (on Mars) solar concentrators to light greenhouses. This is much more efficient than converting sunlight to electricity and then back to light again.
Heat buildup is a problem in the craft itself, but you can just insulate the greenhouse and regulate the amount of light you give it to let it regulate its own temperature by radiation of heat back into space. It's no coincidence that the Earth is a comfortable temperature - At our distance from the sun and with the Earth's albedo (reflectivity), energy input from sunlight and output from radiative emission are exactly matched at Terrestrial temperatures. Rig your mirrors so that you have the same average amount of energy absorbed by the greenhouse plants as by the Earth's surface, and you should be fine for temperature.
On the other hand, if you have a powerful, compact power source like a fission or fusion plant, you'd want to be able to pack your plants more densely, which means many layers and an artificial light source. However, I doubt we'll be lofting that large an amount of nuclear fuel into space any time soon.
Varying wavelengths might be a bit tough. But why not simply pack coherent waves together in a "sheet" to make a diffraction grating?
The problem is that you need a lens or the equivalent if you want to make images, which is what you need for lithography. A diffraction grating with constant spacing bends light by the same amount no matter where it strikes the surface (assuming you're starting with a parallel beam; this is just an example). This can't be used to form an image. A lens needs to bend light by different amounts depending on where on the surface it strikes. This means that the spacing of your grating has to change depending on where you are, if you're using a grating-like pattern as a lens.
The experiments discussed in the article have already demonstrated grating-like behavior, but we need something more complicated than that for lithography.
It seems that this suggests a couple of attractive long-term alternatives to photolithography. The experiments reported diffraction of argon atoms and of electrons, both of which have far shorter wavelengths than light. By using light-based "optics", you would in principle be able to either use imaged electrons to cure a photoresist, or deposit matter directly with direct masking and imaging of the matter stream, at far higher resolution than photolithography allows.
Problems with this:
You'd need a standing wave pattern with cylindrical symmetry and a wavelength that _varies_ with radius to get a diffracting "lens". Varying wavelength in-flight is impossible in vacuum. There may be workarounds.
For electron lithography, you'd be better off just using magnetic fields to focus electron images as with electron microscopes. Presumably there are problems that prevent this from being used, because it would beat the heck out of the scanning e-beam lithography that's currently used for bleeding-edge research.
Still interesting to think about, though.
Does anyone have information on why electron imaging isn't used for lithography now?
By coincidence, I happen to both be a grad student studying IC architecture and living about 20 feet from someone working on rad-hard space electronics.
It turns out that the situation isn't quite as grim as the scenario you've painted.
Solar radiation is an extremely serious problem for any computer in space. To be rad-hard, chips need to be made of silicon on sapphire, which means a $1 embedded processor suddenly costs twenty thousand dollars.
Silicon-on-insulator chips are used because they aren't vulnerable to latch-up (triggering of parasitic SCR structures formed by the many regions of doped silicon in conventional chips). However, there are other approaches to dealing with latch-up.
A common approach is to just add enough substrate contacts and apply design rules conservatively enough to ensure that latch-up currents won't be immediately destructive, and to power-cycle the chip either on a regular schedule, or when you see a huge current spike, or both. Powering down the chip turns off the SCR, and when you power up, everything's fine again.
On the flip side of this equation, SiOI is slowly becoming more common. There was a Slashdot article about IBM rolling out a SiOI process a while back; while plain silicon is still cheaper, I doubt you'd be looking at a factor of 10,000 price difference. The main problem with spacecraft electronics is that any custom chips will be fabbed in very low quantities, so you don't get the economics of devoting a wafer run to them. This is true whether they're rad-hard or not.
Nobody can afford sapphire RAM banks, and thus memories get a flipped bit per orbit, in general. The only way they keep working is that there is a "washing" process that scans memory and does ECC correction continuously.
You get noise events affecting the processor's activities too. You can get around this either by running two processors back-to-back with HA hardware to compare outputs, or by living with occasional errors and resetting the chip every so often. An expensive solution isn't necessarily needed:).
Also, using SiOI doesn't save you from these noise events. It's only useful for latch-up. An ionizing event could still cause conduction through gate oxide or do any of a number of other fun things that cause errors.
Because it's available in sapphire and is flight-proven, the microprocessor of choice for controlling satellites is the 1802.
Actually, rad-hard 386 chips have been standard for many applications for quite a while now.
Heat is a problem, too. Heat sinks don't work so well without an atmosphere to carry away heat. You have to pipe heat around with heat-pipes filled with a phase-change gas, and then radiate the heat away
Heat is indeed a problem, but you can get away with using the spacecraft structure as a passive heat sink if your electronics are low-power enough. This is a common trick, because you're on a limited power budget and want low-power electronics anyways. That way you only have to worry about craft-wide climate control (well, that and instruments that require very stable temperatures).
More directly to the question, rockets waste a lot of energy carrying oxygen, when they spend much of their trip into orbit flying through an atmosphere carrying quite adequate amounts of the stuff. If research into scramjets succeeds, the propellant requirement for launching rockets decreases radically.
I'm doubtful of this, for a couple of reasons.
Firstly, the proposed space-plane designs I've heard talked about still only used the scramjet for intermediate speeds (above Mach 5 but still well below orbital velocity). Building a scramjet that can work at 8 km/sec is going to be *really* *really* hard. If you're using rockets for the last stage of the trip, you're still saddled with most of the resulting efficiency problem (though it might be a bit easier to pull off a single-stage-to-orbit vehicle under these conditions).
Secondly, a scramjet still only works in the atmosphere. If you're travelling at orbital velocity in the atmosphere, you're dumping a *lot* of energy into your shockwave, and have a *lot* of air friction heating you up. You'll be burning a lot of fuel just to maintain velocity (not accelerate), and you'll need the heat shielding of the Gods. This means you're probably forced to use the scramjets for only a short time and intermediate velocity range, which means you're stuck with rockets for the last part of the trip.
In summary, while I agree that scramjets will make it easier to build ground-to-orbit craft, I don't think they'll improve the cost by orders of magnitude.
Additionally, many of the costs of running space launches are because we do so few of them. If we were doing twenty a day, we'd be able to set up much more efficient production lines for the job. The propellant cost of a space shuttle launch is a tiny fraction of the mission cost.
The propellant cost will probably remain a small part of the mission cost even with mass production, at least for chemical rockets. The fuel:cargo mass ratio forced by the Isp of chemical rockets requires a big, complex rocket for relatively small amounts of cargo. Big, complex devices are expensive to build and to maintain.
Now, the shuttle is still many times more expensive than it needs to be, because 1) it's an experimental craft pushed into routine service, and 2) NASA has to bend over backwards to make sure there are *no* failures, which pretty much means dismantling and rebuilding the whole shuttle between flights to maintain it. A more reasonable benchmark would be one of the commercial satellite launch groups. They can afford the occasional launch failure, so they don't have to spend a truly insane amount of money on maintenance. These are the closest thing we have to mass-produced rockets. Costs are still very high, though.
The Blitter, yes, that can be emulated in realtime. But the Copper, it's an entirely different story.
The Copper (it stands for CoProcessor) was a very simple processor, that executed simple statements trigged by the SCANLINE POSITION of the monitor!!! With the bang-the-hardware programming model typical of the Amiga, you could do some terrific tricks without ANY main processor overhead:
Hmm. This is a really neat-looking piece of hardware, but I think I see a sneaky way of emulating this functionality quickly.
It was widely used for graphical demos. I had UAE (an Amiga Emulator) installed on my K6-2 300Mhz, and some scenes of "State of the Art" couldn't go faster than 9 fps (they were 50fps on a 7Mhz Amiga).
It should be possible to do better than *that*. Now you've got me all curious about the implementation problems:).
Do you have links handy for the UAE emulator (ideally source) and for details of the amiga hardware (ideally complete)?
Oh yes - there's also the little problem of bringing our troubles with us. The ills of society are a direct result of human nature. Geeks are human. Their kids will be human. Thus, any utopia you establish away from normal society will soon be fraught with all of the troubles it tried to escape. Examples of this on small scales and large are all around us.
But the technical argument was more fun than just pointing this out.
Every year spent on Earth, subject to the whims of beurocrats, questionable legislation and business practices far more insidious than all the religious peasents in the world could ever be, is a year spent on Earth by choice.
And you propose we leave earth how?
I suppose death works, but that leaves the destination a little uncertain.
Chemical rockets are not a viable way of migrating from Earth for Joe Geek, as the cost per person will be many, many times even a successful geek's yearly salary.
Ion and plasma rockets are not a vaible way of migrating from earth, as there is no way in heck you can get them giving 1 G of thrust.
Magnetic launch or "supergun" style gas-launchers are not a viable way of migrating from earth, unless you don't mind being squashed to jelly. The acceleration path *has* to be at least a thousand kilometres long (about 600 miles) or preferably more, if acceleration is to be something that humans can withstand. That means a self-propelled craft. A thousand-kilometre tube on the ground would send you through far too much atmosphere on the way out. A tube that turns up at the end squashes you to jelly from centripetal acceleration.
Laser launchers are almost certainly not a viable way of migrating from earth, because your launch path must be steep, and you run out of atmosphere after a few tens of kilometres. Acceleration required is far too great for humans to withstand. And powering the lasers is very expensive (efficiency is horrible).
Fisson rockets won't work, because they just heat a working fluid to chemical rocket temperatures, which means your mass efficiency is no better than chemical rockets.
Fusion rockets will most likely have acceleration characteristics comparable to ion and plasma rockets - far too low acceleration to be useful for ground-to-orbit. And fusion has been 20 years off for the past 50 years.
We're not going to be able to move large numbers of people off-planet for a long time. We'd need free power (for a laser launcher), or much better materials and free construction (for a really huge launch cannon with a muzzle outside the atmosphere), or both (for building a space elevator). Don't hold your breath.
It should also be mentioned that your average non-techie computer user is likely to feel more comfortable plugging in cable than a card. As the computer becomes a comodity item, then more and more of your stuff will be plugging in from the outside, unless you are a techie and then you will insist on putting everything inside because it takes less room and because there are less wires.
Actually, I'd suspect integration would be the order of the day. Joe User neither knows nor cares what a hard drive is, or a graphics card. The more things that are "just *there*", the more comfortable the average user will likely feel.
The average user probably won't ever upgrade any one part of the machine; they'll just dump the old machine and buy a new one, especially when computers come closer to being commodity items. The only cables needed would likely be for things that the user doesn't consider part of the computer.
so, higher radius = less gradient from head to feet.
At least, that seems to be why you'd care about RPM if gravity gradient was of any concern. (Though the coriolis effect is important, too.)
It turns out that any rotation rate below about 10 RPM involves a large enough station to make gravity gradients small, so this most likely isn't the factor that was limiting the rotation rates in the original thought experiments.
There have been speculative non-fiction books describing the construction of spinning space stations for years.
A major design constraint is that they try to keep the rotation rate as low as possible. Original designs used a rate of about 1 RPM, but later designs cited a requirement to push this to about 1/4 RPM for humans' long-term health.
I have no idea why such a low rate would be required. Neither does a doctor friend I queried.
Is there any physiological basis for needing a slow rotation rate for long-term habitation, or were these numbers pulled out of a hat by the profs designing the stations?
[My doctor friend says that as long as there isn't much of a "gravity" gradient between a person's head and feet, he couldn't thinnk of any difficulties offhand. I noted that the speed of the station's rotation should be much greater than the speed of blood moving through the body to avoid coriolis effect concerns, but the number he cited for blood velocity is low enough that this is a non-issue. Disclaimer: These were all off-the-cuff answers, and not official medical advice.]
I've been thinking about building my own (unmanned) ornithopter ever since seeing UTIAS's prototypes flying on Discovery Channel a few years back.
To eliminate vibration in most of the craft, you can use two pairs of wings arranged dragonfly-style. Diagonally opposite wings would move in one direction, and the other diagonally opposite pair would move in the other direction 180 degrees out of phase.
The center of mass of the unit stays in one place, and the forces of the wings on the air are symmetrical, so vibration is only in the engine.
Your thrust would still "vibrate" at twice the wings' flapping frequency, but a shock absorber should take care of that. It's vibrating up and down as the wings flap that's the big problem, and using two pairs of wings solves this problem.
As for this being an insurmountable design challenge - it isn't. The mechanics of ornithopters and of bird and insect flight have been well-understood for quite a while now. It's just a materials and engineering issue, and we have enough of a handle on both to build ornithopters.
The real reason why you don't see bird-planes flapping across the sky - and won't in the future - is that using flapping wings is only a benefit for slow-moving craft, and existing slow-moving craft are already adequately efficient (actually, a helicopter might even be _more_ efficient than an ornithopter).
[For anyone wondering, the efficiency gain of an ornithopter comes from it moving a larger mass of air more slowly to generate thrust; same reason a propeller's more efficient than a jet turbine, and a helicopter's blades are more efficient than an airplane's propeller. You're just limited to a slower speed, due to several concerns.]
There's no provincial election this month (or year even), so our wonderful budget cutting premier Mike Harris isn't in danger of being dumped.
There IS a minor by-election to fill a single seat this month though - so maybe this wouldn't have been quite so loudly trumpeted without it.
Ok, I was misled by the voting card I received in the mail, the party reps phoning around, and the forest of Liberal, NDP, and occasional PC signs on lawns for kilometres around my home in Toronto.
Guess I'm in the lucky seat's riding.
Has the date for the next provincial election been set yet? Harris's term should be pretty close to "up".
FWIW, there's a provincial election in Ontario this month. "Chainsaw Mike" has taken flak for massive cuts he's made to health care and other services in the province over the course of his two terms.
He may honestly be doing this to give Ontario residents better health care.
Or he may be doing this in a (hopefully futile) attempt to make us forget his past unpopular acts.
So it seems like it can do everything a DSP can vis a vis hardware-level parallelism, plus some things that are hard for SIMD-style DSPs. Which is faster, 25 SISD processors sharing an address space, or one SIMD processor which can operate on 25-ary vectors? Multiple SISD cores seem to win, I think, though they may be harder to program for.
I don't see how you support this conclusion.
I agree that multiple SISD cores are more _flexible_. That isn't the issue.
If you're doing, say, a dot product on multiple SISD cores, you have to deal with several instructions dedicated to control flow.
If you're doing the same dot product on a SIMD core, you don't have any loop overhead.
Heck, even if you're just doing a MAC operation, the example holds. Which would make best use of processor capabilities - issuing a multiply and then an add, or issuing one operation with data flow built into the hardware?
I'll place my speed bets on SIMD and hardwired data flow, thank you.
You may want to look again at trinary.cc; especially notice the accuracy of how a three-based counting system handles numbers bigger than 17.
I've read your page, and I'm having trouble seeing what you're getting at with this.
If you're arguing that you can express more numbers with fewer digits, I would argue that this is of marginal use. You still have to manipulate your numbers, and as far as I can tell from your schematics, trinary arithmetic and logic circuits are not smaller (in silicon area) than binary circuits that could handle a comparable number of possible values.
You might be able to save a bit of chip area by requiring fewer routing traces, but not very much, as you'd still require a similar number of control lines, and the number of data lines is reduced by only a factor of ln(2)/ln(3) = 0.63.
Trinary calculations are more like nature, and the calculations of fractions are more accurate than binary. The problem with roundoff errors you mentioned is offset by the fact that you don't have to roundoff so much, because trinary is better with moving larger numbers than binary.
This is pretty much the same effect as you noted in your previous point - you can represent more values with fewer digits (just with a different mapping than with integers). However, I again point out that this doesn't buy you much - the logic to _manipulate_ numbers to a given precision isn't any smaller, and might even be larger (I'd have to spend a little while mucking about with schematics to give you an exact transistor count for each).
I didn't say it was the Windows of hardware design. I think it's fascinating, and something that real geeks might be interested in.
It most certainly is fascinating. I've bookmarked your page, and might even try soldering together a trinary CPU together for fun at some point. However, I still do take issue with your claim that it's "more efficient" than binary, as I feel that the metric you're using to assess efficiency isn't a very realistic one (as discussed above).
Thanks for the link, though. Your site was an interesting read.
Running on a general-purpose device, even a parallel one, won't get you these benefits.
Uhm, a DSP is a general-purpose parallelized device. It's not like their adders and multipliers are any faster then those in modern processors. The only real difference between a DSP and a CPU is that DSPs have typically embraced parallelism in hardware (VLIW, multiple cores on a chip, pipelining) to a greater extent than contemporary CPUs, at the necessary expense of backwards compatibility with earlier models.
We seem to be using the term "general-purpose" differently.
The 25x chip is an array of more-or-less independent cores, optimized for a general-purpose instruction set and geared towards SISD integer instructions, with each processor running its own instruction stream with its own control flow.
A generic DSP chip is a single-instruction-stream core optimized to do things like dot products, multiply-accumulates FFTs, and/or various other signal-processing-specific operations in parallel (SIMD-style) in hardware.
Ask a DSP processor to play chess, and it'll crawl.
Ask it to perform frequency-domain feature extraction or to do geometry transformations for rendering, and it'll work at blinding speed, because much of the control flow and register shuffling and memory shuffling that you'd have for these tasks with a general-purpose chip doesn't have to be performed.
A DSP's instruction set and chip hardware are geared towards a narrow class of applications (signal processing), and as a result it does these tasks (and pretty much only these tasks) extremely well.
(Note to purists: I'm considering a stream of VLIW instructions to be a stream of "single" instructions for purposes of this thread, because you don't have independent control flow in the multiple instructions per clock being executed. This is an arbitrary terminology distinction on my part.)
One of the things that can never get enough power is compression. Right now the next generation of image and video compression looks like JPEG 2000 and Motion JPEG 2000. I have tested it and it seems like a miracle compression, it consistently works at least 4x better than jpeg compression. Apply that to video and you have something incredible but very VERY expensive. He said it was a solution looking for a problem, and there it is.
Actually, you'd almost certainly be better off just building dedicated hardware optimized for JPEG/MPEG compression, like you already see in the hardware decoder cards for DVD. These either implement common computation-intensive parts of the CODEC algorithms in hardware, or use DSPs to implement them in firmware using hardware that's geared towards signal processing.
Running on a general-purpose device, even a parallel one, won't get you these benefits.
A trinary [trinary.cc] computer system is based on architecture which is much more efficient than binary, especially for moving large numbers around.
Actually, to the best of my knowledge, trinary systems (and other systems with a radix other than 2) are not vastly more efficient than binary.
Trinary logic is more complex to design than binary and requires more transistors. There's no substantial design time or area saving.
If you're doing math, you'll also have larger roundoff errors using a radix larger than 2.
Back in the olden days, computers were built to work in base 10 or base 16, or to work with multiple logic levels per line, but for these and other reasons, they finally converged on base 2 with two-level signalling.
From the article: We've been trying for a century, and we still don't fully understand black holes," said Dr. Andrew Strominger. And then he goes on to conclude that we need to make some.
If they're going to do something which at least sounds dangerous, I would really like it if they could say, "Nothing can possibly go wrong", not, "Our understanding is incomplete."
Actually, there's a pretty ironclad argument for this being safe - the same one that comes up every time the press starts fearmongering about more powerful accelerators:
Cosmic rays with energies far higher than will be produced by any accelerator any time soon have been striking the earth and the moon for billions of years. If high-energy collisioins could cause catastrophy, they would have already, because they've been happening in our neighbourhood for quite a while.
The fact that nothing around here has been sucked into a black hole yet leads us to conclude that if micro-black-holes can be formed, they don't do much.
Our current models of black holes suggest that micro-holes would evapourate in a burst of Hawking radiation almost as soon as they're formed. The smaller the hole, the more intense the Hawking radiation (and so the faster it loses mass).
When they say an object "moves along a geodesic", it's shorthand for saying that the geodesic (and the object's worldline in spacetime) runs along the time axis of the graph (actually that the object's frame's time axis runs along the geodesic, but I digress). This is a linguistic short form; nothing more.
No it isn't.
Yes it is, according to every textbook on physics I've ever read that discusses relativity.
Ask any of the professors you've been talking to whether this is what they meant. You'll get a resounding "yes".
Saying that "time is an invariant" is like saying "position is an invariant" on this graph-paper sketch. It isn't - you can look at different positions in space and time on the graph. Thus, I find your statement confusing.
Not at all. When I say that time is invariant, I am talking about coordinate time, not the time axis that is graphed as an axis on a diagram.
Um, a "coordinate" is *defined* as a position with respect to an axis. One's coordinate, in space or time, is one's location when projected on to the appropriate space or time axis.
Check any math textbook for this one.
Your entire premise seems to be based on the meaning of these two terms ("motion through time" and "coordinate") being garbled.
Nothing moves in spacetime, and nobody's saying it does.
You're kidding me? More than 3/4 of the emails I get from physicists (even professors) insist that there is motion in spacetime and that bodies are moving along their geodesics. They use this to support their notion that gravity is not a force because objects are following a straight path in spacetime.
When they say an object "moves along a geodesic", it's shorthand for saying that the geodesic (and the object's worldline in spacetime) runs along the time axis of the graph (actually that the object's frame's time axis runs along the geodesic, but I digress). This is a linguistic short form; nothing more.
A "geodesic" is just the shortest path between two given points (in space and time) on the spacetime "surface" that you're graphing world-lines on. A freely-falling object's worldline will follow a geodesic. General relativity models spacetime as a curved surface, which means that geodesics won't always follow the rules of euclidean geometry (which applies only to flat surfaces). A nifty consequence of this is that, by relating curvature of the hypothetical "spacetime" surface to the mass distribution in space and time, they can arrange things so that the geodesics around a large mass's worldline will curve around the mass's worldline. Freely-falling objects (following geodesics) thus orbit the mass.
It turns out that this is an extremely useful model of the universe, because it predicts the paths of objects in space and time more accurately than Newtonian gravity, and explains many other effects that had previously been mysterious.
Does this clear up what physicists mean when they say that objects "follow paths" through spacetime?
You can vary the time coodinate in your mind as much as you want but, in reality, a time coordinate cannot change It is invariant.
Um, no. It's a coordinate on your graphing surface, like any other coordinate.
As an example, draw a parabola on a sheet of graph paper. Declare one axis space, and one time, and you've just mapped out the worldline of an accelerating object in the spacetime of a one-dimensional universe.
You can consider different points in time - they're just points on the graph with different "x" values.
You can consider different points in space - they're just points on the graph with different "y" values.
Saying that "time is an invariant" is like saying "position is an invariant" on this graph-paper sketch. It isn't - you can look at different positions in space and time on the graph. Thus, I find your statement confusing.
Certainly but you misinterpreted my argument. Why is there no motion in spacetime? Because for something to move in spacetime (or in time) it would need to have a variable temporal coordinate.
Um, no.
Nothing moves in spacetime, and nobody's saying it does.
Things move in *space*. The motion (velocity) of an object in space at any given time is defined as the derivative of its position (in space, relative to some arbitrary frame of reference) with respect to the time axis on the graph of its position vs. time (its worldline in spacetime).
Your statement about a "variable temporal coordinate" doesn't make much sense. All I'm doing when I "vary" the time coordinate is look at different points of the world-line, which is most certainly possible.
Slashdot Mirror
While I understand the convenience of compact light sources, I'm curious as to why we wouldn't just use mirrors and (on Mars) solar concentrators to light greenhouses. This is much more efficient than converting sunlight to electricity and then back to light again.
Heat buildup is a problem in the craft itself, but you can just insulate the greenhouse and regulate the amount of light you give it to let it regulate its own temperature by radiation of heat back into space. It's no coincidence that the Earth is a comfortable temperature - At our distance from the sun and with the Earth's albedo (reflectivity), energy input from sunlight and output from radiative emission are exactly matched at Terrestrial temperatures. Rig your mirrors so that you have the same average amount of energy absorbed by the greenhouse plants as by the Earth's surface, and you should be fine for temperature.
On the other hand, if you have a powerful, compact power source like a fission or fusion plant, you'd want to be able to pack your plants more densely, which means many layers and an artificial light source. However, I doubt we'll be lofting that large an amount of nuclear fuel into space any time soon.
Varying wavelengths might be a bit tough. But why not simply pack coherent waves together in a "sheet" to make a diffraction grating?
The problem is that you need a lens or the equivalent if you want to make images, which is what you need for lithography. A diffraction grating with constant spacing bends light by the same amount no matter where it strikes the surface (assuming you're starting with a parallel beam; this is just an example). This can't be used to form an image. A lens needs to bend light by different amounts depending on where on the surface it strikes. This means that the spacing of your grating has to change depending on where you are, if you're using a grating-like pattern as a lens.
The experiments discussed in the article have already demonstrated grating-like behavior, but we need something more complicated than that for lithography.
Problems with this:
Still interesting to think about, though.
Does anyone have information on why electron imaging isn't used for lithography now?
Wow, someone who knows first-hand.
:). Ask my father for first-hand details.
Actually only second-hand
I'm afraid I don't know the answer to either of your other questions.
By coincidence, I happen to both be a grad student studying IC architecture and living about 20 feet from someone working on rad-hard space electronics.
:).
It turns out that the situation isn't quite as grim as the scenario you've painted.
Solar radiation is an extremely serious problem for any computer in space. To be rad-hard, chips need to be made of silicon on sapphire, which means a $1 embedded processor suddenly costs twenty thousand dollars.
Silicon-on-insulator chips are used because they aren't vulnerable to latch-up (triggering of parasitic SCR structures formed by the many regions of doped silicon in conventional chips). However, there are other approaches to dealing with latch-up.
A common approach is to just add enough substrate contacts and apply design rules conservatively enough to ensure that latch-up currents won't be immediately destructive, and to power-cycle the chip either on a regular schedule, or when you see a huge current spike, or both. Powering down the chip turns off the SCR, and when you power up, everything's fine again.
On the flip side of this equation, SiOI is slowly becoming more common. There was a Slashdot article about IBM rolling out a SiOI process a while back; while plain silicon is still cheaper, I doubt you'd be looking at a factor of 10,000 price difference. The main problem with spacecraft electronics is that any custom chips will be fabbed in very low quantities, so you don't get the economics of devoting a wafer run to them. This is true whether they're rad-hard or not.
Nobody can afford sapphire RAM banks, and thus memories get a flipped bit per orbit, in general. The only way they keep working is that there is a "washing" process that scans memory and does ECC correction continuously.
You get noise events affecting the processor's activities too. You can get around this either by running two processors back-to-back with HA hardware to compare outputs, or by living with occasional errors and resetting the chip every so often. An expensive solution isn't necessarily needed
Also, using SiOI doesn't save you from these noise events. It's only useful for latch-up. An ionizing event could still cause conduction through gate oxide or do any of a number of other fun things that cause errors.
Because it's available in sapphire and is flight-proven, the microprocessor of choice for controlling satellites is the 1802.
Actually, rad-hard 386 chips have been standard for many applications for quite a while now.
Heat is a problem, too. Heat sinks don't work so well without an atmosphere to carry away heat. You have to pipe heat around with heat-pipes filled with a phase-change gas, and then radiate the heat away
Heat is indeed a problem, but you can get away with using the spacecraft structure as a passive heat sink if your electronics are low-power enough. This is a common trick, because you're on a limited power budget and want low-power electronics anyways. That way you only have to worry about craft-wide climate control (well, that and instruments that require very stable temperatures).
It's an interesting field, in any event.
More directly to the question, rockets waste a lot of energy carrying oxygen, when they spend much of their trip into orbit flying through an atmosphere carrying quite adequate amounts of the stuff. If research into scramjets succeeds, the propellant requirement for launching rockets decreases radically.
I'm doubtful of this, for a couple of reasons.
Firstly, the proposed space-plane designs I've heard talked about still only used the scramjet for intermediate speeds (above Mach 5 but still well below orbital velocity). Building a scramjet that can work at 8 km/sec is going to be *really* *really* hard. If you're using rockets for the last stage of the trip, you're still saddled with most of the resulting efficiency problem (though it might be a bit easier to pull off a single-stage-to-orbit vehicle under these conditions).
Secondly, a scramjet still only works in the atmosphere. If you're travelling at orbital velocity in the atmosphere, you're dumping a *lot* of energy into your shockwave, and have a *lot* of air friction heating you up. You'll be burning a lot of fuel just to maintain velocity (not accelerate), and you'll need the heat shielding of the Gods. This means you're probably forced to use the scramjets for only a short time and intermediate velocity range, which means you're stuck with rockets for the last part of the trip.
In summary, while I agree that scramjets will make it easier to build ground-to-orbit craft, I don't think they'll improve the cost by orders of magnitude.
Additionally, many of the costs of running space launches are because we do so few of them. If we were doing twenty a day, we'd be able to set up much more efficient production lines for the job. The propellant cost of a space shuttle launch is a tiny fraction of the mission cost.
The propellant cost will probably remain a small part of the mission cost even with mass production, at least for chemical rockets. The fuel:cargo mass ratio forced by the Isp of chemical rockets requires a big, complex rocket for relatively small amounts of cargo. Big, complex devices are expensive to build and to maintain.
Now, the shuttle is still many times more expensive than it needs to be, because 1) it's an experimental craft pushed into routine service, and 2) NASA has to bend over backwards to make sure there are *no* failures, which pretty much means dismantling and rebuilding the whole shuttle between flights to maintain it. A more reasonable benchmark would be one of the commercial satellite launch groups. They can afford the occasional launch failure, so they don't have to spend a truly insane amount of money on maintenance. These are the closest thing we have to mass-produced rockets. Costs are still very high, though.
The Blitter, yes, that can be emulated in realtime. But the Copper, it's an entirely different story.
:).
The Copper (it stands for CoProcessor) was a very simple processor, that executed simple statements trigged by the SCANLINE POSITION of the monitor!!! With the bang-the-hardware programming model typical of the Amiga, you could do some terrific tricks without ANY main processor overhead:
Hmm. This is a really neat-looking piece of hardware, but I think I see a sneaky way of emulating this functionality quickly.
It was widely used for graphical demos. I had UAE (an Amiga Emulator) installed on my K6-2 300Mhz, and some scenes of "State of the Art" couldn't go faster than 9 fps (they were 50fps on a 7Mhz Amiga).
It should be possible to do better than *that*. Now you've got me all curious about the implementation problems
Do you have links handy for the UAE emulator (ideally source) and for details of the amiga hardware (ideally complete)?
Oh yes - there's also the little problem of bringing our troubles with us. The ills of society are a direct result of human nature. Geeks are human. Their kids will be human. Thus, any utopia you establish away from normal society will soon be fraught with all of the troubles it tried to escape. Examples of this on small scales and large are all around us.
But the technical argument was more fun than just pointing this out.
Every year spent on Earth, subject to the whims of beurocrats, questionable legislation and business practices far more insidious than all the religious peasents in the world could ever be, is a year spent on Earth by choice.
And you propose we leave earth how?
I suppose death works, but that leaves the destination a little uncertain.
Chemical rockets are not a viable way of migrating from Earth for Joe Geek, as the cost per person will be many, many times even a successful geek's yearly salary.
Ion and plasma rockets are not a vaible way of migrating from earth, as there is no way in heck you can get them giving 1 G of thrust.
Magnetic launch or "supergun" style gas-launchers are not a viable way of migrating from earth, unless you don't mind being squashed to jelly. The acceleration path *has* to be at least a thousand kilometres long (about 600 miles) or preferably more, if acceleration is to be something that humans can withstand. That means a self-propelled craft. A thousand-kilometre tube on the ground would send you through far too much atmosphere on the way out. A tube that turns up at the end squashes you to jelly from centripetal acceleration.
Laser launchers are almost certainly not a viable way of migrating from earth, because your launch path must be steep, and you run out of atmosphere after a few tens of kilometres. Acceleration required is far too great for humans to withstand. And powering the lasers is very expensive (efficiency is horrible).
Fisson rockets won't work, because they just heat a working fluid to chemical rocket temperatures, which means your mass efficiency is no better than chemical rockets.
Fusion rockets will most likely have acceleration characteristics comparable to ion and plasma rockets - far too low acceleration to be useful for ground-to-orbit. And fusion has been 20 years off for the past 50 years.
We're not going to be able to move large numbers of people off-planet for a long time. We'd need free power (for a laser launcher), or much better materials and free construction (for a really huge launch cannon with a muzzle outside the atmosphere), or both (for building a space elevator). Don't hold your breath.
It should also be mentioned that your average non-techie computer user is likely to feel more comfortable plugging in cable than a card. As the computer becomes a comodity item, then more and more of your stuff will be plugging in from the outside, unless you are a techie and then you will insist on putting everything inside because it takes less room and because there are less wires.
Actually, I'd suspect integration would be the order of the day. Joe User neither knows nor cares what a hard drive is, or a graphics card. The more things that are "just *there*", the more comfortable the average user will likely feel.
The average user probably won't ever upgrade any one part of the machine; they'll just dump the old machine and buy a new one, especially when computers come closer to being commodity items. The only cables needed would likely be for things that the user doesn't consider part of the computer.
so, higher radius = less gradient from head to feet.
At least, that seems to be why you'd care about RPM if gravity gradient was of any concern. (Though the coriolis effect is important, too.)
It turns out that any rotation rate below about 10 RPM involves a large enough station to make gravity gradients small, so this most likely isn't the factor that was limiting the rotation rates in the original thought experiments.
There have been speculative non-fiction books describing the construction of spinning space stations for years.
A major design constraint is that they try to keep the rotation rate as low as possible. Original designs used a rate of about 1 RPM, but later designs cited a requirement to push this to about 1/4 RPM for humans' long-term health.
I have no idea why such a low rate would be required. Neither does a doctor friend I queried.
Is there any physiological basis for needing a slow rotation rate for long-term habitation, or were these numbers pulled out of a hat by the profs designing the stations?
[My doctor friend says that as long as there isn't much of a "gravity" gradient between a person's head and feet, he couldn't thinnk of any difficulties offhand. I noted that the speed of the station's rotation should be much greater than the speed of blood moving through the body to avoid coriolis effect concerns, but the number he cited for blood velocity is low enough that this is a non-issue. Disclaimer: These were all off-the-cuff answers, and not official medical advice.]
What strikes me with this project, is that the thing still has to go to 55Mph to take off.
:)
Most bird take off instantly with just flapping.
Ever see a swan take off?
The larger birds have to be running at a decent clip before they leave the ground.
An ornithopter is a *damned* big bird.
I've been thinking about building my own (unmanned) ornithopter ever since seeing UTIAS's prototypes flying on Discovery Channel a few years back.
To eliminate vibration in most of the craft, you can use two pairs of wings arranged dragonfly-style. Diagonally opposite wings would move in one direction, and the other diagonally opposite pair would move in the other direction 180 degrees out of phase.
The center of mass of the unit stays in one place, and the forces of the wings on the air are symmetrical, so vibration is only in the engine.
Your thrust would still "vibrate" at twice the wings' flapping frequency, but a shock absorber should take care of that. It's vibrating up and down as the wings flap that's the big problem, and using two pairs of wings solves this problem.
As for this being an insurmountable design challenge - it isn't. The mechanics of ornithopters and of bird and insect flight have been well-understood for quite a while now. It's just a materials and engineering issue, and we have enough of a handle on both to build ornithopters.
The real reason why you don't see bird-planes flapping across the sky - and won't in the future - is that using flapping wings is only a benefit for slow-moving craft, and existing slow-moving craft are already adequately efficient (actually, a helicopter might even be _more_ efficient than an ornithopter).
[For anyone wondering, the efficiency gain of an ornithopter comes from it moving a larger mass of air more slowly to generate thrust; same reason a propeller's more efficient than a jet turbine, and a helicopter's blades are more efficient than an airplane's propeller. You're just limited to a slower speed, due to several concerns.]
There's no provincial election this month (or year even), so our wonderful budget cutting premier Mike Harris isn't in danger of being dumped.
There IS a minor by-election to fill a single seat this month though - so maybe this wouldn't have been quite so loudly trumpeted without it.
Ok, I was misled by the voting card I received in the mail, the party reps phoning around, and the forest of Liberal, NDP, and occasional PC signs on lawns for kilometres around my home in Toronto.
Guess I'm in the lucky seat's riding.
Has the date for the next provincial election been set yet? Harris's term should be pretty close to "up".
FWIW, there's a provincial election in Ontario this month. "Chainsaw Mike" has taken flak for massive cuts he's made to health care and other services in the province over the course of his two terms.
He may honestly be doing this to give Ontario residents better health care.
Or he may be doing this in a (hopefully futile) attempt to make us forget his past unpopular acts.
So it seems like it can do everything a DSP can vis a vis hardware-level parallelism, plus some things that are hard for SIMD-style DSPs. Which is faster, 25 SISD processors sharing an address space, or one SIMD processor which can operate on 25-ary vectors? Multiple SISD cores seem to win, I think, though they may be harder to program for.
I don't see how you support this conclusion.
I agree that multiple SISD cores are more _flexible_. That isn't the issue.
If you're doing, say, a dot product on multiple SISD cores, you have to deal with several instructions dedicated to control flow.
If you're doing the same dot product on a SIMD core, you don't have any loop overhead.
Heck, even if you're just doing a MAC operation, the example holds. Which would make best use of processor capabilities - issuing a multiply and then an add, or issuing one operation with data flow built into the hardware?
I'll place my speed bets on SIMD and hardwired data flow, thank you.
You may want to look again at trinary.cc; especially notice the accuracy of how a three-based counting system handles numbers bigger than 17.
I've read your page, and I'm having trouble seeing what you're getting at with this.
If you're arguing that you can express more numbers with fewer digits, I would argue that this is of marginal use. You still have to manipulate your numbers, and as far as I can tell from your schematics, trinary arithmetic and logic circuits are not smaller (in silicon area) than binary circuits that could handle a comparable number of possible values.
You might be able to save a bit of chip area by requiring fewer routing traces, but not very much, as you'd still require a similar number of control lines, and the number of data lines is reduced by only a factor of ln(2)/ln(3) = 0.63.
Trinary calculations are more like nature, and the calculations of fractions are more accurate than binary. The problem with roundoff errors you mentioned is offset by the fact that you don't have to roundoff so much, because trinary is better with moving larger numbers than binary.
This is pretty much the same effect as you noted in your previous point - you can represent more values with fewer digits (just with a different mapping than with integers). However, I again point out that this doesn't buy you much - the logic to _manipulate_ numbers to a given precision isn't any smaller, and might even be larger (I'd have to spend a little while mucking about with schematics to give you an exact transistor count for each).
I didn't say it was the Windows of hardware design. I think it's fascinating, and something that real geeks might be interested in.
It most certainly is fascinating. I've bookmarked your page, and might even try soldering together a trinary CPU together for fun at some point. However, I still do take issue with your claim that it's "more efficient" than binary, as I feel that the metric you're using to assess efficiency isn't a very realistic one (as discussed above).
Thanks for the link, though. Your site was an interesting read.
Running on a general-purpose device, even a parallel one, won't get you these benefits.
Uhm, a DSP is a general-purpose parallelized device. It's not like their adders and multipliers are any faster then those in modern processors. The only real difference between a DSP and a CPU is that DSPs have typically embraced parallelism in hardware (VLIW, multiple cores on a chip, pipelining) to a greater extent than contemporary CPUs, at the necessary expense of backwards compatibility with earlier models.
We seem to be using the term "general-purpose" differently.
The 25x chip is an array of more-or-less independent cores, optimized for a general-purpose instruction set and geared towards SISD integer instructions, with each processor running its own instruction stream with its own control flow.
A generic DSP chip is a single-instruction-stream core optimized to do things like dot products, multiply-accumulates FFTs, and/or various other signal-processing-specific operations in parallel (SIMD-style) in hardware.
Ask a DSP processor to play chess, and it'll crawl.
Ask it to perform frequency-domain feature extraction or to do geometry transformations for rendering, and it'll work at blinding speed, because much of the control flow and register shuffling and memory shuffling that you'd have for these tasks with a general-purpose chip doesn't have to be performed.
A DSP's instruction set and chip hardware are geared towards a narrow class of applications (signal processing), and as a result it does these tasks (and pretty much only these tasks) extremely well.
(Note to purists: I'm considering a stream of VLIW instructions to be a stream of "single" instructions for purposes of this thread, because you don't have independent control flow in the multiple instructions per clock being executed. This is an arbitrary terminology distinction on my part.)
One of the things that can never get enough power is compression. Right now the next generation of image and video compression looks like JPEG 2000 and Motion JPEG 2000. I have tested it and it seems like a miracle compression, it consistently works at least 4x better than jpeg compression. Apply that to video and you have something incredible but very VERY expensive. He said it was a solution looking for a problem, and there it is.
Actually, you'd almost certainly be better off just building dedicated hardware optimized for JPEG/MPEG compression, like you already see in the hardware decoder cards for DVD. These either implement common computation-intensive parts of the CODEC algorithms in hardware, or use DSPs to implement them in firmware using hardware that's geared towards signal processing.
Running on a general-purpose device, even a parallel one, won't get you these benefits.
A trinary [trinary.cc] computer system is based on architecture which is much more efficient than binary, especially for moving large numbers around.
Actually, to the best of my knowledge, trinary systems (and other systems with a radix other than 2) are not vastly more efficient than binary.
Trinary logic is more complex to design than binary and requires more transistors. There's no substantial design time or area saving.
If you're doing math, you'll also have larger roundoff errors using a radix larger than 2.
Back in the olden days, computers were built to work in base 10 or base 16, or to work with multiple logic levels per line, but for these and other reasons, they finally converged on base 2 with two-level signalling.
From the article: We've been trying for a century, and we still don't fully understand black holes," said Dr. Andrew Strominger. And then he goes on to conclude that we need to make some.
If they're going to do something which at least sounds dangerous, I would really like it if they could say, "Nothing can possibly go wrong", not, "Our understanding is incomplete."
Actually, there's a pretty ironclad argument for this being safe - the same one that comes up every time the press starts fearmongering about more powerful accelerators:
Cosmic rays with energies far higher than will be produced by any accelerator any time soon have been striking the earth and the moon for billions of years. If high-energy collisioins could cause catastrophy, they would have already, because they've been happening in our neighbourhood for quite a while.
The fact that nothing around here has been sucked into a black hole yet leads us to conclude that if micro-black-holes can be formed, they don't do much.
Our current models of black holes suggest that micro-holes would evapourate in a burst of Hawking radiation almost as soon as they're formed. The smaller the hole, the more intense the Hawking radiation (and so the faster it loses mass).
When they say an object "moves along a geodesic", it's shorthand for saying that the geodesic (and the object's worldline in spacetime) runs along the time axis of the graph (actually that the object's frame's time axis runs along the geodesic, but I digress). This is a linguistic short form; nothing more.
No it isn't.
Yes it is, according to every textbook on physics I've ever read that discusses relativity.
Ask any of the professors you've been talking to whether this is what they meant. You'll get a resounding "yes".
Saying that "time is an invariant" is like saying "position is an invariant" on this graph-paper sketch. It isn't - you can look at different positions in space and time on the graph. Thus, I find your statement confusing.
Not at all. When I say that time is invariant, I am talking about coordinate time, not the time axis that is graphed as an axis on a diagram.
Um, a "coordinate" is *defined* as a position with respect to an axis. One's coordinate, in space or time, is one's location when projected on to the appropriate space or time axis.
Check any math textbook for this one.
Your entire premise seems to be based on the meaning of these two terms ("motion through time" and "coordinate") being garbled.
Nothing moves in spacetime, and nobody's saying it does.
You're kidding me? More than 3/4 of the emails I get from physicists (even professors) insist that there is motion in spacetime and that bodies are moving along their geodesics. They use this to support their notion that gravity is not a force because objects are following a straight path in spacetime.
When they say an object "moves along a geodesic", it's shorthand for saying that the geodesic (and the object's worldline in spacetime) runs along the time axis of the graph (actually that the object's frame's time axis runs along the geodesic, but I digress). This is a linguistic short form; nothing more.
A "geodesic" is just the shortest path between two given points (in space and time) on the spacetime "surface" that you're graphing world-lines on. A freely-falling object's worldline will follow a geodesic. General relativity models spacetime as a curved surface, which means that geodesics won't always follow the rules of euclidean geometry (which applies only to flat surfaces). A nifty consequence of this is that, by relating curvature of the hypothetical "spacetime" surface to the mass distribution in space and time, they can arrange things so that the geodesics around a large mass's worldline will curve around the mass's worldline. Freely-falling objects (following geodesics) thus orbit the mass.
It turns out that this is an extremely useful model of the universe, because it predicts the paths of objects in space and time more accurately than Newtonian gravity, and explains many other effects that had previously been mysterious.
Does this clear up what physicists mean when they say that objects "follow paths" through spacetime?
You can vary the time coodinate in your mind as much as you want but, in reality, a time coordinate cannot change It is invariant.
Um, no. It's a coordinate on your graphing surface, like any other coordinate.
As an example, draw a parabola on a sheet of graph paper. Declare one axis space, and one time, and you've just mapped out the worldline of an accelerating object in the spacetime of a one-dimensional universe.
You can consider different points in time - they're just points on the graph with different "x" values.
You can consider different points in space - they're just points on the graph with different "y" values.
Saying that "time is an invariant" is like saying "position is an invariant" on this graph-paper sketch. It isn't - you can look at different positions in space and time on the graph. Thus, I find your statement confusing.
Certainly but you misinterpreted my argument. Why is there no motion in spacetime? Because for something to move in spacetime (or in time) it would need to have a variable temporal coordinate.
Um, no.
Nothing moves in spacetime, and nobody's saying it does.
Things move in *space*. The motion (velocity) of an object in space at any given time is defined as the derivative of its position (in space, relative to some arbitrary frame of reference) with respect to the time axis on the graph of its position vs. time (its worldline in spacetime).
Your statement about a "variable temporal coordinate" doesn't make much sense. All I'm doing when I "vary" the time coordinate is look at different points of the world-line, which is most certainly possible.