I've always wanted to build (distinct) mechanical computers out of the following kinds of elements:
1. Spur-gear differential; and,
2. Shishi-odoshi.
Both of these are saturating mechanical devices that can be used to build NAND gates; the latter, I think, would be very pleasing, if exceedingly slow.
For the spur-gear differential, you'd need to up-scale the output by a factor of 2 (since the output is half-speed), and use a locking wedge to build a one-way gear out of one of the spur-gear differentials. However, it has the nice property that the logic is made entirely out of a single element: the spur-gear differential.
Similarly, for the shishi-odoshi: you're going to have to do a bit of analysis (drilling a hole in the bottom part of the bamboo ladle), to figure out the in-flow and out-flow to build the basic AND gate, and then balancing out the NOT gate, to build your basic NAND. This is, obviously, very finicky; but, I supposed, that'd be quite a bit of the charm of a Zen computer garden?
I built some logic gates using water and a 3D printed "seesaw" that tilts to the left or right: https://byronknoll.blogspot.com/2022/06/water-computer.html
A shishi-odoshi ALU would be amazing to see…and hear too.
I love that idea.
I wanna run my neural net on shishi-odoshi.
Has any computer been built out of spur-gear differentials? Like maybe some sort of adder circuit, not necessarily a full instruction executing computer. The only uses I could find was what seems to me like the differentials being part of some sort of analogue computer.
Spur gear differentials are naturally adders (with carry!); so, traditionally they've only ever been used for analogue logic. They're overly complicated for digital logic: you need two spur gears to build a single gate (NAND) to perform a single binary operation. If you want any sort of reasonable lash characteristics you're going to need ~60 teeth. At that point, two 60 teeth spur gears give you a 3600-valued adder. That'd take something like 300+ spur gears in binary: it just doesn't make any damn sense.
I think the last time I looked at this, if I used the cast spur gears available I needed a staged approach to "start" the computer and a 1100 hp motor to run it.
the shishi-odoshu seems like the more promising avenue. The key question in mechanical computing is never designing gates, its designing power amplifiers.
I just smile hearing the term "Millihertz Computer". I'd love it if building and designing mechanical and analog computers grew as a hobby/educational activity as I find them both fascinating and somehow satisfying.
Also, this 1950s Naval Training film explaining the fundamentals of how mechanical fire control computers work to solve complex problems is excellent. https://www.youtube.com/watch?v=s1i-dnAH9Y4
About 8 years ago I visited TU Chemnitz and they had a lab making similar things to this. It wasn't clear to me what the goal was, but it was very cool nonetheless.
I was incredibly surprised to find that this actually is a computer. Normally when you hear about a "computer" constructed in an unusual medium, it turns out to just be a binary adder or an analogue computer. I've learned to expect disappointment.
What a gem of a site Thank you for sharing