I’ve been thinking for some years about building a personal computing environment up from zero: bootstrapping using an existing computer and existing fabrication technology, but quickly getting into self-sustaining technology, along the lines of the VPRI STEPS program or colorForth/GreenArrays, but also along the lines of Open Source Ecology.
So what is needed to, say, send an instant message to someone? What’s the smallest amount of code you could write to fully describe the design of this system?
So far, despite various experiments, the simplest way I’ve found to design a readable font is as a fixed-width pixel grid. Here’s a minimal implementation of ASCII text rendering with an included 4×6 font in 1K of DHTML, including the font itself as a PNG:
<body bgcolor="black" onload="r()">
<a href="http://canonical.org/~kragen/sw/dofonts.html">cf</a>
<img src="data:image/png;base64,
iVBORw0KGgoAAAANSUhEUgAAAEAAAAAkAQMAAAADwq7RAAAAAXNSR0IArs4c6QAAAAZQTFRFAAAA
////pdmf3QAAAQxJREFUGNMtjlFqg0AYhIcQikhJryBFgoiHWGQR2QfJIaQPfegZRCQEkZxBZCny
I9sr+LBIkZBDBPFBQs7QNXTgh2Hg+2ewpZh7Ehtsf26S89XYmvHHbgu8ebVJLJPcDlyyGACDhwzg
ifpK4VTomjYnnoSY3LuOkLmQ4XvbIGFI7kF84U6Bpyxzqb7Oyi9dELUn6Z8J6rcbp+C8oKS2lFQS
PvX1qMTyT63Sw0I0VwzUCiK3qDFqobhZhrIRD1m4QNElQs4BzPyndmCZ4xReD+yHNOptoWG3tHRB
QsZ0kfQF4WX4CJW/yJV4Rf8kmfkRHzB8m9KJ16CLIBFEwJgLFa2lx2ovmsmYfMU2Fv4AgeVrl8Qw
sFoAAAAASUVORK5CYII=
" id="font" />
<canvas width="512" height="512" id="c"></canvas>
<script>
d = document
function r() {
t = d.body.innerHTML
x = 0
y = 0
for(i in t) {
c = t.charCodeAt(i)
if(32<=c && c<128) {
fi = c-32
d.getElementById('c').getContext('2d')
.drawImage(d.images[0],fi%16*4,~~(fi/16)*6,4,6,x,y,4,6)
x += 4
}
if(x>508 || c===10) {
x = 0
y += 6
if(y>=512) break
}
}
}
</script>
Other experiments I’ve tried include Hershey-like line fonts, spline-based fonts, fonts scanned from paper, and so on, but I haven’t been able to get any of these implementations down to a kilobyte, or even nearly so. You could maybe argue that I’m unreasonably leveraging bitblit and PNG-decompression here, but I don’t think those actually make a whole lot of difference.
The current standard is an LED-backlit LCD with diffraction gratings to give color to RGB subpixels. This may not be the least demanding possible way to display pictures under computer control; it seems likely that some kind of cathode-ray tube may be simpler to construct.
Perhaps the electronically-controlled display with the least stringent demands on metallurgy and the like would be a small array of tiny carbon arc lamps or spark plugs controlled at near-megahertz frequencies, viewed through a system of spinning mirrors like the ones used in supermarket barcode scanners in order to scan the image of the arc across your field of view, and color-filtered both to eliminate harmful ultraviolet light and to improve contrast.
Suppose you use just an X-axis scanning mirror, with one spark gap per scan line, and you want minimally 200 scan lines (25 lines of readable 8-scan-line text or 33 lines with the less-readable 4×6 font above). A 16-faceted mirror spinning on an air bearing at 32000 rpm, as Michelson used in his 1877–1931 experiments to measure the speed of light, scans at 8528 frames per second. We only really need about 24 frames per second to be readable, even though that’s kind of flickery, and you could do that with 1.5 revolutions per second of a 16-faceted mirror, 90 rpm. This is an easily achievable rotational speed even if you’re turning the mirror by hand.
Square pixels at 200 scan lines and a 4:3 aspect ratio would be 266 pixels per line; 16:9 would be 355 pixels per line; a 5×8 font at the traditional 80 columns would require 400 pixels per line. 266 pixels in 1/24 of a second is 6384 pixels per second per spark gap, or 157μs of cool-down time; 400 pixels is 104μs (9600Hz). Spark times like these are easily achievable. Neon glow discharge lamps can easily reach this régime, and they last for decades of continuous use and don’t pollute the air, unlike carbon-air arcs.
Edgerton was achieving 500ns air-gap spark times in the 1950s by using a thin layer of air as the plasma medium and cooling it with a quartz heatsink, but I think much higher speeds are achievable if the spark just needs to be visible rather than illuminating its surroundings brightly enough for photography.
A 500ns spark could potentially paint 2 million pixels in a second, or 33,333 pixels in the 16⅔ms of a 60fps frame, not quite enough for a text display by itself. An array of 20 500ns spark gaps, assuming you could fire them all continuously, could thus give you a ⅔Mpix display at 60fps, nearly the 1024×768 that we’re used to.
At the currently popular 16:9 aspect ratio, this would be about 1075×620, so you’d need a vertical or Y scanning mirror that scanned at 31×60Hz, or 1860Hz. If it had 16 facets, it would need to spin at 116¼Hz, or 6975 rpm, which is totally feasible using gears and ball bearings (not even air bearings), but a little bit tricky. Again, though, this is about four times slower than Michelson was spinning mirrors nearly a century ago.
If, instead of scanning an entire column of 20 spark gaps across the visible area in 20-pixel-high bands, you divided the visible area into 20 horizontal bands, each illuminated by a single spark gap, you could have 20 facets in your visual field at once. This allows you to use more facets on your mirror and makes the scan lines more horizontal, but requires careful alignment of the mirror’s phase with your viewpoint. It would, however, allow you to rotate your Y-scanning mirror at only 31×60Hz÷16÷20 = 5.8125Hz or 348¾rpm.
The first graphical user interface was Sketchpad, implemented at the end of the 1950s by Ivan Sutherland on the TX-2 at Lincoln Lab. The TX-2 had an attached CRT in which the CPU independently controlled the X and Y coordinates of the electron beam, so it could draw simple shapes on the screen at high resolution, even though it was only capable of drawing a few tens of thousands of points per second. Such “vector displays” survived until the 1980s, and CAD systems of the 1970s were built on “storage tubes” that could maintain the vector image, once drawn, for up to a few minutes, allowing very detailed line drawings.
You could do the same thing with a single spark gap and two mirrors, but this requires controlling the mirrors independently and actively, not just scanning them at a constant speed like a raster display. But it would permit a useful display from a single spark gap modulated even down at kilohertz speeds.
Laser shows scan lasers across the screen with mirrors controlled by galvanometers or “galvos”: electric coils that magnetically push against a spring to bring the mirror to the desired position. Typical galvos are capable of “20 kpps”, twenty thousand points per second, although the definition of this is kind of dodgy.
The Harmonic Series was a project by Brazilian artists Luisa Pereira and Manuela Donoso which substituted regular dynamic speakers (electric coils that magnetically push against a screen) for laser-show galvos, getting bandwidths of up to several kilohertz for displaying audio Lissajous patterns, just as Lissajous himself had done with the “Lissajous apparatus” of mirrors on tuning forks in 1857.
You could imagine a resonant mirror scanning system that scans in a Lissajous pattern rather than the sawtooth pattern we’re used to from raster-scan devices; you get higher resolutions around the edges, lower resolutions in the center, and a less regular scanning pattern that the computer can compensate for. However, I think the Pythagorean sum of the scanning frequencies still limits your resolution, equivalent to the scan frequency of the X-scanning mirror in a traditional raster setup. So with the 20 500ns spark gaps at ⅔Mpix at 60Hz described above, you could get by with oscillation frequencies possibly as low as 42Hz or so.
A particularly simple way to achieve these Lissajous oscillations is by using a sprung mass that is sprung with two different spring constants in X and Y; for example, a thin rod, fixed at one end, with a rectangular cross-section. (There is a Lissajous Rods exhibit in the Exploratorium demonstrating this; the Lissajous patterns described in the air by the shiny ends of the rods are quite conspicuous.)
All of the above describes different contortions to get around the possible slowness of a spark-gap light source, which is itself much faster than an incandescent light. But even normal vacuum tubes can turn on and off in the submicrosecond regime; you could use a vacuum tube with a phosphor-coated anode to modulate light output with deep-submicrosecond times. There are a number of electron-excited phosphors with single-digit nanosecond decay times, hundreds of times faster than an air-gap flash, including CdS:In and ZnO:Ga.
(If you’re using a vector display to draw relatively static images, a long phosphorescence time might be an advantage rather than a disadvantage.)
Vacuum tubes and exotic phosphors, though, may be hard to fabricate; and if you can fabricate them, it may be more practical to just fabricate a CRT rather than some kind of electromechanical display.
Another possibility, since we’re already taking for granted semiconductor fabrication, is LEDs. These are semiconductor diodes made of exotic large-bandgap semiconductors, and, with carefully designed driver circuitry, even regular LEDs can have response times measured in single-digit nanoseconds. And LEDs are much, much more durable than spark gaps. They were first reported in 1927, using point-contact diodes with low-purity silicon carbide, but red LEDs were first published in 1962 in gallium arsenide phosphide; they became commercially available in quantity in 1968. Other red and green LED chemistries include AlGaAs, AlGaInP, and GaP; InGaN and GaN additionally provide green, blue, and violet LEDs, but these were not possible to manufacture until 1994, and the discovery won a Nobel Prize in physics.
Once you, hypothetically, can make LEDs, making semiconductor lasers is relatively straightforward. This has the advantage that you can scan them over a large surface using mirrors to produce a high-resolution projection display.
Also, lasers in general have the ability to produce shorter light pulses than other light sources: once a population inversion forms, stimulated emission can deplete it, and so a wave packet generated by a mirrorless laser can be very short indeed, the so-called “femtosecond laser”, when the laser has a large tunable bandwidth. You should be able to trigger the generation of a pulse by dumping a bunch more energy into a barely-subcritical laser, which is more or less how erbium-doped fiber amplifiers work. (This is my vague understanding of this shit; don’t bet on it.) So it should be possible to
One possible way to rapidly scan temporally-modulated LED light over an area would be to use a transparent Lissajous rod elastically vibrating; it could be made of, for example, polished glass or acrylic, using total internal reflection to guide the light to its vibrating tip.
LCDs do not emit light, but they can switch on and off in tens of nanoseconds using current techniques. I don’t know if LCDs are less demanding to fabricate than LEDs, but I suspect so, even though they were developed later.
Above, I’ve mentioned the benefits of slowness a couple of times: if your image lasts a long time, you can use even very slow methods for producing it. The Tektronix 4014 storage tube was one example: used for high-resolution CAD drawings despite being connected to the computer over a low-speed serial link. But even a regular long-persistence phosphor would work reasonably well, without the storage-tube aspect; in fact, if you’re using mirrors to steer a beam of light rather than an electron beam, you could draw the image on a phosphor screen of, for example, copper-activated zinc sulfide, which is sphalerite/wurtzite, the main ore of zinc in nature.
But you can go slower still. A lot of Unix was developed on Teletypes running at 110 baud, which worked out to 11 characters per second; a 4×6 dot-matrix printer at this speed needs to run its 6 hammers at 44Hz, which is quite reasonable. And if you go vector instead of raster, you get a pen plotter.
Both current and prehistoric metallurgy depends mostly on iron, aluminum, copper, zinc, tin, and lead; nuclear energy production depends on uranium or thorium. Semiconductor fabrication also depends on silicon, plus trace amounts of dopants, such as boron and arsenic, as well as processing chemicals such as hydrofluoric acid and hydrogen peroxide. Refractory materials are essential to many processes listed.
Small amounts of iron, lead, and especially copper are found “native” on Earth, while aluminum, zinc, tin, and thorium are always found in compounds. Nevertheless, for bootstrapping, it’s probably necessary to be able to refine these elements from soil.
Earth’s crust is about 28% silicon., and it’s in almost all soils, so anywhere you go, you have silicon. Although you can smelt metallurgical-grade silicon from silica in an arc furnace with carbon at 1900°, smelting it for semiconductors is hell, since you need to get impurities down to <1ppb in order to control its semiconducting properties, using either CVD from trichlorosilane or repeated zone melting. Even less-demanding photovoltaic silicon needs <1ppm impurities.
Earth’s crust is about 8% aluminum, which, too, is everywhere; but it’s difficult to smelt because aluminum dioxide (sapphire) is so stable. The Hall-Héroult process, orders of magnitude cheaper than the alternative smelting processes, revolutionized the production of aluminum. This involves dissolving corundum (aluminum oxide) in cryolite (sodium aluminum fluoride) and calcium fluoride at 980° in order to be able to electrolyze it.
Earth’s crust is about 5% iron. People have been smelting iron from hematite and magnetite using charcoal in bloomery furnaces for millennia, but blast furnaces are much more efficient and produce much better iron.
Earth’s crust is about 60 ppm zinc, mostly as sphalerite, which can be purified by froth flotation; roasting sphalerite in air at at least 690° produces SO₂ and ZnO, and the ZnO can be reduced to Zn either with C (which was done accidentally for centuries, producing Zn vapor) or by electrowinning from water-soluble ZnSO₄ (low-current-density process: 10% H₂SO₄, aluminum cathodes, 275–325 A/m², 30–35°). Historically this smelting was done in hermetic clay retorts to prevent the zinc vapor from escaping.
Earth’s crust is about 50 ppm copper.
Earth’s crust is about 14 ppm lead.
Earth’s crust is about 6 ppm thorium, but you don’t need very much thorium, because burning thorium in a nuclear reactor produces about 79 TJ/kg of energy.
Earth’s crust is about 2 ppm tin, which makes it hard to extract.
One possible answer to the problem of extracting various elements from the complex mix of mineral compounds in soil is, rather than using individual reactions customized for each separate ore, simply vaporize the whole shebang into a plasma, mix with oxygen, and then distill out the various oxides. Even tungsten, tantalum, rhenium, and osmium boil below 6000°, and every other element below 5000°; everything melts far below this temperature, with tantalum hafnium carbide melting at 3942°.
So, if you can heat soil up to 6000°, you can probably get them to mix with a large volume of preheated oxygen and oxidize them fully. You should be able to start the process with a plasma arc torch (regular plasma arc welding torches reach 28000°), then continue heating with inductively-coupled plasma heating. Various companies are currently developing lower-temperature variants of this process under the name “plasma arc recycling” or “plasma gasification”; Westinghouse’s process runs at only 3000°.
As an alternative to inductive heating of the plasma, you can illuminate it; well-ionized plasmas absorb light fairly well. Sunlight can be focused on the plasma using non-imaging optics, which have been demonstrated to reach 84000 suns, 59 megawatts per square meter.
Another alternative means of heating the plasma after initial ionization might include mixing carbon or hydrogen fuel into the feedstock. In theory, hydrogen ought to be able to raise the temperature by about 2800° in the limit of 100% hydrogen.
The mix of oxygen and fairly inert simple elemental oxides coming out of the torch then needs to be distilled. Each oxide has its own characteristic condensation temperature, so you should be able to condense thoria in one chamber, silicon dioxide in another, aluminum oxide in another, zinc oxide in a fourth, and so on. You probably need to do this under low pressure to keep the plasma from cooling down too much before depositing its condensates.
How much energy does this need? Heat capacity of plasmas is complicated but generally around 100 kJ/(kmol·K) in the temperature range we’re considering, so heating by 6000 K should require about 600MJ/kmol. If we figure that the majority of the plasma is oxygen, then each mole is 16 grams, so we need up to about 40MJ/kg to heat the plasma, some of which we may be able to recover as we cool it back down.
Supposing we mix oxygen with the vaporized soil in a 4:1 ratio by mass, we’re spending 200 MJ per kg of soil, or 33 TJ per kg of thorium, assuming that our soil is only averagely rich in thorium. That’s less than half of the energy that thorium will produce in a reactor, which still leaves us precariously net positive, but suggests that some monazite deposits (about 10% thorium, i.e. 100 000 ppm, five orders of magnitude above average) would go a long way. (It’s also possible that the modal thorium concentration is well below the mean, which would mean that most points on the globe had thorium concentrations too low for purification by this method to be a net energy producer.)