24 ms·
I'm ... questioning utility here, though if you want to pursue this, I guess that's OK. I have a few general concerns: For any human-oriented computing based
by dredmorbius 5y ago
I'm ... questioning utility here, though if you want to pursue this, I guess that's OK.
I have a few general concerns:
For any human-oriented computing based on accessing remote or general-distribution content, I think you'll find that as compute power gets cheaper, the response will be for payloads and tasks to become ever more complex. What used to be communicated in plain-text email, talk(1), or write(1) payloads is now wrapped in apps or JSON or HTML+CSS+JS, where the package-to-payload ratios often run 100:1 or worse. PDF documents become ever more complex and heavy. Your ultra-low-power system simply won't keep up.
I'd resurrected a 2005-era system for general use over the past few years. It was great for pretty much anything except surfing the Web. Firefox + 2-3 tabs would routinely lock the system. Terminal-mode browsers were far more viable, though much of the Web is now inaccessible to those.
So that's the first problem.
The second was a realisation I'd had a few years back that with the falling cost of SoCs and the like, a period where a full computing system would be available at less than a $1/unit price point (and falling from there at the rate of an order of magnitude every 3 Moore's cycles or so, that is, about every 6--8 years), was close.
The major problems at that point are power and comms.
Current-generation logistics tracking, particularly for orientation, shock, and temperature excursions, are based on physical detectors. These tend to change state if tolerances are exceeded and require direct observation to detect that shipping violated spec, but now when.
With a $0.10 adhesive-affixed data logger having an integrated WiFi or SIM capability, days, weeks, or months of data logging at 1s -- 1m resolution are possible. A one second resolution costs you ~2 MB storage per data channel byte month, uncompressed. And it should compress fearsome well.
Surveillance devices don't need e-ink displays ... but might be all the more interesting if they offered these.
As for solar power: the Max Lumi has a surface area of roughly 1/16 m^2 (the size of a sheet of A4 paper). With a case that has an integrated solar cell, that ought to be good for 6 W power, or 48 WH given 8 hours of irradiance per day.
(Mind: you can't use the device whilst it's charging, but ... let's roll with that.)
At a power draw of 2000-9000 mW, or 2-9W, the cell by itself could ... mostly keep up with the draw of the device itself. And if you had a battery to store that energy ... The Max Lumi with battery masses at about 500g.
My thinking is that you're far better off having your power source off-device. An external solar panel can be much larger than the device, and pull in far more energy. Again, that can be stored in various systems, and yes, batteries are a mess, but they're a pretty decent mess. For the forseable future you're sacrificing a lot of capability by excluding them.
- kragen 5y agoI mostly agree, except that in theory an A4 at 21% efficiency is 13 W, not 6. But those monocrystalline panels produce 0 in the library, on the subway, on the bus, etc.
- dredmorbius 5y agoIn practice, about 10% of total incident energy is likely, based on spacing factor, orientation, shading, etc. 20--35% is optimal, but unlikely to be consistently achieved. PV charging cover does seem like an option. A deployable array, of whatever size is suitable, would be a more flexible and IMO better option. The notion of a computing device which could function on ambient energy is an interesting / terrifying one, so I follow you there. Low-end chips (apparently Zilog -- the Z-80 from Sinclair fame -- is one of these and popular in low-energy applications) might offer this. With sufficiently rudimentary text formatting at least a general information device might be possible. The original Unix ... seems like a reasonably decent candidate for this.
- kragen 5y agoEfficiency depends a lot on illuminance and cell type as well. To hit 35% you need super pricey multijunction PV panels for outer space. The amorphous panels that work best in dim indoor light are only 10% efficient at best. Conventional monocrystalline is 21%. As you point out, shading and orientation can pull you down from those peak numbers. I really liked the Z80 idea myself, because CP/M is close to the minimum system where self-hosted development is a reasonable thing to do, and there are full free-software toolchains for it. But as it turns out, according to the datasheet, the Z80 is super power hungry, like 500 mW when it's running. Two orders of magnitude worse than the STM32L, and three orders of magnitude worse than the Ambiq chips! And there doesn't seem to be a low-power clone. Also, the Z80 isn't very good at running C, even if it's not quite as bad as the 8080 or the 6502. Unix doesn't demand much more than CP/M, but fork() does require some sort of MMU, and writing a large system in an unsafe language probably requires some kind of memory protection for fault isolation in practice (at least an MPU). There's lots of available chips that can do milliwatt computing with the oomph of a SPARC-20 or PowerMac, but none of them have MMUs, not even PDP-11-style segmentation registers, and they also have less RAM. You'd have to compensate for the lower RAM by using secondary storage more aggressively; fortunately, secondary storage is four orders of magnitude lower latency now than it was in SPARC-20 or PowerMac days, but in the submilliwatt regime you're probably limited to I/O rates that are slower than those machines could typically manage. If you're interested, more detailed notes can be found in Dernocua in notes/energy-autonomous-computing.html. Along with, of course, lots of marginally relevant crap. You may be interested in the discussion about Oberon that started the other day if you missed it: https://news.ycombinator.com/item?id=30467122 https://news.ycombinator.com/item?id=30467122