Around age 12 or so, the urge to build my own computer was becoming irresistible. I was on a rather limited budget back then, and the only logic elements which were available at a low enough cost for me to play with were relays. Relay logic is a fascinating area, but the relays that I experimented with were removed from military surplus electronic equipment of WWII vintage which used to be present in huge quantities in a military surplus store in Edmonton in 1963-1965 (I used to think there were huge amounts of equipment, but when you're half the height you are now, everything looks bigger).
All a relay consists of is a coil of wire wound around an iron core which, when energized with a current, closes or opens various metal contacts which are attached to a metal plate which is pulled into contact by the magnetic field of the energized relay. At this time I obtained my power from dry cells as a number of spectacular shorts resulted in my parents forbidding me to use 110 VAC as my power source (besides, the shorts had fried, respectively, the expensive silicon high current diode and selenium rectifiers which I was using to produce DC for my relay logic. The smell of a burning selenium rectifier is an unforgettable experience). Obiously my projects had to have fairly low current requirements for me to be allowed to continue with my electrical experiments.
One of the requirements of a digital von-Neuman type machine is that it have some form of storage. Volatile storage was out since I didn't have the ability to maintain data for more than about 15 minutes with relays and dry cells. Then, one day while playing around with paper clips, I suddenly realized that I could make a flip flop out of 3 paperclips and 3 thumbtacks. Two paperclips were wound with magnet wire which made them into electromagnets. The third paperclip was bent into a shape so that it was free to move between the two electromagnets and be attracted by either of the magnets, but because of its shape, once it was in contact with the magnet, it would remain in contact once the current was removed. Passing a current through the other paperclip electromagnet would attract the swinging paperclip to it, and thus a bistable element was born. Readout of the bistable element was accomplished by using the presence or absence of swinging paperclip contact with the metal core of a paperclip magnet to read out the state of the flipflop (FF).
This is something that is much better appreciated in pictures rather than words, and here is a drawing of what the circuit looked like (I don't have any of my constructions any more).
This paperclip bistable storage element worked, although the electrical contact between swinging center paperclip and the magnet frames was sometimes a bit tenuous. I was quite proud of this one bit storage element, and I think that I got 8 bits of a proposed 64 bit memory completed before I grew bored with the task. Also, there were problems in switching times for these mechanical memory elements which ranged from under 100 msec to close to 1 second. I could have clocked relay logic with an internal combustion engine distributor, but getting the motor speed slow enough was a problem, and the power requirements quickly exceeded that of the D cells I was using for my experiments.
It is strange looking at this childhood attempt at building a computer from a perspective 35 years in the future. Given constraints that I would have to use paperclip storage elements and relay logic, I could build a working digital computer, but it would take a major disaster that wiped out world semiconductor manufacturing capability to make me actually complete such an item. The power consumption would be huge and even trivial computational tasks would require garage sized assemblages of components. I much prefer using my 900 MHz Athlon processor, but if Moores law holds, in 20 years I'll consider this Athlon to be as quaint as the paperclip logic of 1965.