A high-speed electrostatic store was the heart of several early computers, including the computer at the Institute for Advanced Studies in Princeton. Professor F. C. Williams and Dr. T. Kilburn, who invented this type of store, described it in Proc.I.E.E. 96, Pt.III, 40 (March, 1949). This simple account is an abridged version of the description in Bowden [ref 3].
The system makes use of an ordinary cathode ray tube, which has a pick-up plate mounted outside it, immediately in front of the fluorescent screen. When a spot on the screen is bombarded with a beam of electrons that has been accelerated by a potential between 1,000 and 2,000 V., more electrons will be emitted by the spot than fall upon it from the primary beam. The spot will become positively charged. There is a small capacity between the spot on the screen and the collecting plate so that the result of bombarding an uncharged spot will be to produce a small positive pulse on the collector plate, which can be amplified to any size required.
If we bombard the spot a second time, it will still be at equilibrium potential, so that no positive pulse will be induced in the pick-up plate. We can therefore tell, by bombarding a spot on the tube, if it has recently been bombarded or not. It is possible to write information on to a series of spots on the surface of the tube and subsequently to read it out again. Unfortunately, in the act of reading from the fluorescent screen we have to write all over it, and erase everything written there. Moreover, the insulation of a fluorescent screen is such that any charge distribution which has been built up on it will decay in a few seconds. A more sophisticated method of using the cathode ray tube is required.
Suppose that immediately after bombarding one spot, A, we move the beam a small distance, about equal to a spot diameter, and bombard an adjacent spot, B. Slow secondary electrons will be emitted by B. Some of them will be attracted to A, which is the most positively charged point nearby, and after a short time they will remove the positive charge on A completely. If we now bombard A for a second time a positive signal will appear in the amplifier. On the other hand if the spot B had not been bombarded since the first bombardment of A, A would still be at its equilibrium potential, and no positive signal would be obtained by bombarding it.
Suppose now that we arrange a mosaic of pairs of spots, which we will call AlBl, A2B2, A3B3, etc. The individual spots in a pair are to be very close together, but adjacent pairs are far enough apart to ensure that secondary electrons from one pair do not affect any other pair on the screen. When we bombard Al we get a signal which tells us whether, during the last cycle, Bl was bombarded or not; moreover this signal occurs a short time before we are ready to bombard Bl for a second time. We have time, therefore, to turn on the electron beam before the time comes to change the voltage on the deflector plates and direct the beam to Bl. We can arrange to bombard Bl if it was bombarded during the last cycle, and to refrain if we refrained last time. By doing this we renew the charge pattern stored on AlBl and in the process we read off what was stored there.
The charge pattern will leak slowly away unless it is renewed once or twice a second. The operation of the machine must ensure the regeneration of all parts of the store whether or not they happen to be needed in the computation. In the case of the SILLIAC this was called the RAR (Read Around Ratio) and its implications are described in Appendix 4 of the Programming Manual.
The great advantage of this type of "memory" is that, by suitably controlling the deflector plates of the cathode ray tube, it is possible to direct the beam almost instantaneously to any part of the screen in which we happen to be interested: random access memory.