A hand-powered computer?

I’ve written a little bit about my desire for a portable computing device powered by keystrokes on its keyboard, in particular for note-taking on journeys where I’m not driving a car. (I haven’t driven a car since the last time I was outside Argentina, in 2008.) I’ve even done some calculations about the energy available from normal keystrokes on a keyboard, and it turns out that it’s more than an order of magnitude greater than what’s needed to run an e-Ink word processor. But I haven’t written much about why I think this kind of machine is interesting.

Some decades back, my photographer stepfather Karl boated down the Colorado River with a friend of his. Karl took photographs; his friend wrote about the trip on a portable typewriter he carried with him. For a while I owned the typewriter. It weighed about two or three kilograms, a hefty weight for most such wilderness trips, but bearable in a boat. The typewriter didn’t need batteries, since its mechanics were powered by the action of the fingers pressing its keys, and still worked when damp.

A netbook might weigh less, but lacks the other desiderata here.

Consider the case of a week-long hiking trip, with the objective of spending two or three hours each day writing, a total of 20 hours. It’s straightforward to pack enough food for such a trip: if we’re only worried about calories, each kilogram of oil is three days’ worth of calories, even without spending down our body’s fat reserves; so one or two kilograms of dry food is plenty, if you have a way to purify water. And the cost is reasonable, if not trivial. (Two kilograms of Clif bars can set you back a fair bit, and you will definitely get sick of them before the week is up. I speak from experience.)

I'm typing this on a city bus on a netbook that weighs about 1kg, about half of which is its battery. This battery might last for 4 hours of active typing use. A kilogram of batteries, then, will last you about 8 hours, far short of the 20 hours we're hoping for. 20 hours of batteries, plus the netbook would weigh 3 kilograms, as much or more than the typewriter, and cost about US$600.

Imagine, by contrast, an electronic device actually intended for use in such situations. With an e-ink screen, it's clearly visible in direct sunlight, unlike the screen of this netbook, where I struggle to read this text through the reflection of my sunlit T-shirt. It can maintain a display of a static image indefinitely without power, like a book; but it can also easily contain tens of gibibytes of maps, photographs, wildflower and edible-plant identification guides, knot-tying instructions, and so on, not to mention the entire human literary canon, from Mozi to Mark Twain, from the Bhagavad Gita to Plato. It could have a weight similar to an Amazon Swindle, perhaps 200 grams, though taking a weight hit to make the e-ink screen as robust against impacts and pressure as a paper book. It can be more waterproof than a paper book. And perhaps it can include long-distance low-bandwidth radio communications without being dependent on cellphone networks or satellites. It can include voice recording and even voice recognition.

Such a device could run from the energy of typing on it, just like the old typewriter. Or it could include a solar panel. You couldn't repair its electronics without the resources of an industrial civilization, even if it included its own blueprints, but as long as it remained working, it would be independent of that civilization. And it could remain working for a very long time indeed; you could build it without non-solid-state components except for the e-ink screen and, perhaps, parts of the keyboard.

Could you use it, additionally, for some kind of mechanical measurement and control? Historically speaking, tooling precision has been a major limitation on manufacturing; we spend a lot of effort, even today, on meeting manufacturing tolerances. A good measurement tool can dramatically speed up any number of processes. Typically mechanically exercising this kind of control requires relatively large amounts of both energy and power, but consider:

• An inkjet print head that you manually roll across a surface --- the output equivalent of the old hand scanner. If you print barcodes or OCRable text, this can serve as a means of permanent data storage and interchange.
• As you pull a wire through the device, it measures the wire and applies a brake at certain points in order to kink the wire at precise positions.
• Similarly, an electronic planing device could minutely adjust the angle of a blade as it's being rolled across a wood surface in order to minutely contour the wood surface.
• On a larger scale, if the device has some means of measuring its position (lasers? computer vision coregistration, like a flying optical mouse?) it could activate shoe-mounted effectors so that you imprint an image on the ground as you walk around.
• DLP projection or lasers, which can be pretty low-energy if it’s dark, can project precise reference points onto the environment --- useful both for construction and for carpentry.
• For communication over tens or hundreds of kilometers, a heliograph can be controlled with almost arbitrarily small energy input, and can be received by a human being without needing any kind of computer.
• With some focusing optics, perhaps the heliograph approach can also be used to make controllable shapes. A heavier-weight version of this approach has been used to spectacular effect to make a solar-powered 3-D printer that works by melting desert sand into quartz glass, using a giant Fresnel lens. Perhaps an array of lightweight digitally-controlled shutters in front of the Fresnel could produce the Fourier transform of an image to project around the focal point of the lens, thus burning it into whatever material is present. (What are the best photosensitive materials to use for this kind of thing? Natural materials tend not to be photosensitive at low energies, and such materials need careful handling and darkroom processing. All materials are heat-sensitive at some point --- quartz sand is sort of a worst case there --- but maybe materials that are sensitive to very small amounts of heat are better. For example, sugar melts at a relatively low temperature, and is very good at sticking to other things; finely powdered sugar mixed with some kind of "susceptor" to absorb light could work.)

Solar

My netbook battery holds a few hundred kJ (now down to 135 kJ, over 200 kJ when it was newer). My cellphone battery claims to be 4.8 Wh, which would be 17 kJ. A 10kJ battery powering 200-pJ instructions can power 50 trillion instructions. There’s a 250-watt-peak solar panel for sale right now on MercadoLibre of size 1640 mm × 992 mm, which is 154 μW/mm²; a 100 mm × 150 mm solar panel of that efficiency would be 2.3 watts peak, and take 72 minutes to charge such a 10kJ battery, assuming 100% charging efficiency. Looking at it a different way, that solar panel has a power of 12 billion 200pJ instructions per second.

One such 200-or-so-pJ-per-instruction machine is the Silicon Labs EFM8SB20F16G-A-QFN24, a 25MHz 8051, with a bit over 4 kiB of RAM, costing US$1.11. The EFM8 Sleepy Bee line is designed for special power efficiency especially in sleep mode, using 300 nA with the RTC running, and working down to 1.8 volts. 10 kJ / (300 nA × 1.8 V) ≈ 587 years. It uses 170 μA / MHz, so at the full 24.5 MHz speed of its internal oscillator, it uses about 5 mA or 9 mW, so it could run 13 days at full speed on that 10kJ battery. Reaching the 2.3 W that the above solar panel can supply for opportunistic computation would require 256 of these chips, costing US$284, and you could run them in sleep mode most or all of the time; if one of them is running at full speed, the power usage of the other 255 in sleep mode would be insignificant by comparison.

Capacitors

At this point it starts to look like a capacitor-powered machine would be a good idea, eliminating the short lifetime of batteries, which is only a few years. Individual 7.3 mm × 6.1 mm tantalum capacitors like the AVX TAJV157M025#NJ can hold 47 mJ, which would be 230 million 200pJ instructions, nearly 10 seconds at full speed, or a day in sleep. However, this capacitor is bigger and costlier than the microcontroller package itself, and it’s apparently hard to get. The apparently equivalent Vishay-Sprague 597D157X9025F2T is US$8.91 in quantity 1, at 7.6 mm × 6.0 mm; the 597D227X0025M2T, with 220μF, is US$9.55. At 2.3W, charging these 69 millijoules would take 30ms.

The Kemet T491D107K016AT is a 100μF 16V MnO₂ tantalum capacitor which costs only US$1.49. That’s 13 millijoules, 8.6mJ/$, which is slightly more cost-effective than the larger Vishay part, which is only 7.2mJ/$. Also, the Kemet part is enormously more popular. You could quite reasonably put six of them in parallel to slightly exceed the Vishay part’s energy capacity, without resorting to the higher voltage.

What does the circuitry to charge and discharge these big capacitor arrays look like? I’m not sure.

Discharging it continuously, you might use something like the AOZ1280CI buck regulator, which PWMs anywhere from 3 to 26 volts down to whatever voltage you like, regulated by a feedback voltage divider, at 80% to 90% efficiency. This particular part is overkill — it can handle six watts of output power continuously.