This scientific calculator has a "memory" location and supports entry of expressions with parentheses (up to five levels of nesting), unless it's in "STAT" mode, in which it can calculate the standard deviation and average of a set of scalars, but not calculate with parentheses or respect order of operations. From this bizarre interaction, I infer that its memory can only hold five or so numbers beyond what's on the screen; each number might be up to 40 bits. So I infer a total memory of something over 200 bits.
I was excited to find it, though, because it does more complicated calculations than anything I'd found so far (the Brick Game, the FM radio, and the large and little watches) and also has a much larger keyboard, 42 keys instead of two, four, six, or nine. One of my concerns with personal computing devices had been that a keyboard large enough to support good usability and self-expression (as opposed to selecting items from menus) might cost too much to keep the price below my US$10 target. This device demonstrates that that's not a problem, at least for large production runs.
As a scientific calculator, it seems to support pretty much the standard complement of basic functions: circular and hyperbolic functions and their inverses, arithmetic in various bases, sexagesimal, powers, roots, order of operations, natural and Briggsian logarithms, and arithmetic, plus the standard-deviation/mean feature. It takes half a second, or a little less, to compute a cosine.
However, it's crappy in ways going beyond the limits of its mathematical features: the batteries are unnecessariily difficult to replace, the keyboard is mushy, the adhesive holding on the keyboard template is bubbling and falling off, and the "FOLD BACK CASE" is fragile — in fact, it was already broken when I bought it.
The whole thing is fabricated on a single-sided stiff PCB, with a flexible plastic ribbon cable connecting to the LCD (presumably glued on?). Most of the area is occupied by the keyboard; each key is two interlocking capital E's of copper in a little square devoid of solder mask.
There's a single chip covered with a blob of epoxy, connected to power, all the LCD traces (there are 30 by my count), and all the keyboard traces (which I can't count, but estimate at 14). Rather than going with a double-sided layout with plated-through vias, they did this very interesting thing where they laid down a second layer of traces on top of the first, with apparently some insulating substance in between. The width of the traces in this second layer suggests that it may be composed of some substance less conductive than copper.
Some of the keyboard traces are connected to some of the LCD traces, which suggests to me that they were trying to minimize the number of connections to the chip (I count only 38 traces that go under the epoxy, rather than 46 as you'd expect) by time-sharing some of the connections between keyboard duty and LCD duty.
The keyboard proper, in the sense of the keys you actually push, consists of two injection-molded pieces of rubber which fit through holes in the front of the case. Each key is a raised region of this rubber, with a black conductive dot on the back that presses against the PCB. Some marks from the milling of the molds are visible.
It seems likely that most of the unit cost comes from the PCB, which is about 60x90mm, around 8 square inches. There are literally no electronic components other than the chip and the LCD (and the batteries), not so much as a capacitor or clock crystal. Presumably the chip has an RC oscillator onboard for a clock. Most of the rest of the calculator is injection-molded plastic, even the keyboard.
Clearly you can build a battery-powered device with a mushy QWERTY keyboard and a 7-segment LCD and sell it for under $7 at retail. With these manufacturing techniques, I suspect you'd need market volumes at least in the thousands, probably in the tens of thousands, to amortize the cost of making the injection-molding molds.
Rebraining a calculator like this one (by drilling out or chipping off the chip, then soldering in a smarter replacement) could be a labor-intensive but effective way to build prototype devices with large keyboards. It probably wouldn't scale up for production runs without a supply of much cheaper calculators.