The Selectron

Jan Rajchman    (RCA Laboratories, Princeton)

This is an attempt to write a concise summary of the Selectron based on this lecture. The description does not follow the order or balance of the lecture.

Structure   A Selectron tube consists of a large cylindrical vacuum tube with a thermionic cathode down the axis and a dielectric forming the curved surface. In detail it comprises 7 structures lying on the surface of (imaginary) concentric cylinders around the cathode, in order from cathode outwards :

1. The first of two grids to control the electrons emitted by the cathode
2. The second grid
3. A set of M + 1 "circular" wires/conductors equally spaced along the length of its cylinder
4. A set of N "straight" wires/conductors parallel to the cathode equally spaced around its cylinder
5. A collector
6. The dielectric
7. A conducting surface around the outside of the dielectric -- the "signal plate"
Of these 7 structures only the dielectric and the signal plate are in contact. Structures 1, 2 and 5 are presumably anchored at the ends of the cylinder and controlled from there. The wires will be discussed further below.

Bit Positions   Considering the two sets of wires (or alternative form of conductor), together they form a rectangular mesh bent around to form a cylinder. If you imagine a uniform emission of 'light' from the cathode at right angles to it, then the wire mesh would form a shadow enclosing M * N small rectangular areas on the dielectric. Let us call each such area a "bit position". The Selectron is designed to be able to control the potential at each such bit position, to the potential of either the cathode or the collector, independent of the state of any other bit position. The state of each position remains stable so long as the Selectron is switched on. The state can be reset and read.

Therefore the Selectron tube provides the capability of holding M * N bits of information indefinitely, with the ability to read and write individual bit values at will (in the order of a few microseconds).

Structure Element Functions   The job of structures 1 and 2, the grids, is to control the emission of electrons and give the effect of uniform 'light' at right angles to the cathode. The job of structures 3 and 4, the wires, is to select which bit positions can be reached by the streams of electrons at any one time. In practice only one bit position is selected at any one time. The default between selections is to leave all positions selected; then the flow of electrons from the cathode will maintain the current state of stable potentials for each bit position.

Selecting a Bit   Consider now the selection process. For a given bit position there are a unique pair of adjacent circular wires and a unique pair of adjacent straight wires that define its position. So long as both pairs of wires are positive with respect to the cathode then its stream of electrons can get through to it. If any wire of the four is negative, they cannot. So a bit position can be selected individually by arranging that only the two appropriate wires from each set are energised and all the rest are not.

Writing a Bit   A bit position is written by selecting that bit position and then applying a sudden potential change to the signal plate, positive or negative depending on which value is to be set, and then controlling the various structures appropriately until equilibrium is restored (with just this bit reset and the other bits with unaltered state).

Reading a Bit   A bit position is read by selecting that bit position and then applying a (different!) sudden potential (which does not alter its potential state). By observing the resultant currents in the signal plate it is possible to distinguish whether the potential of the bit position was at cathode or collector potential.

There is a conflict between the requirements of the read process and the write process which is difficult to resolve.

Wire Groups   Finally let us go back to the wires and their ends! The proposed capacity of a Selectron tube is 4096 bits, requiring M = N = 64, i.e. 64 straight wires and 65 circular wires. In principle one end of each straight wire can be drawn through one end of the cylinder to the controlling circuits outside. Each 'circular' wire has a lead attached at some point that is also drawn through the surface of the tube to the outside circuits. This leaves 129 leads emanating from a Selectron tube to handle structures (3) and (4) alone. However a clever plan was devised (for which a number of alternatives are discussed) to reduce the number of external wires by arranging that within a structure the wires were connected in groups, in such a way that it was always possible to select a given bit position uniquely by activating an appropriate pair of groups within each structure. Although there might now be many more bit positions with one, two or three of their boundary wires energised, only one would have all four. In this way the number of external leads was significantly reduced, from the total number of internal wires to the total number of internal groups, presumably at the cost of increased complexity within the tube due to the connection of wires into groups, and extra clever circuitry outside.

Progress Report   The talk (on August 23rd 1946) concluded with a progress report saying that the problems of selection and the technology of construction were completely solved, but difficulties remained still with the stability of the storage, particularly with respect to undesired interaction between bit positions.

The 1985 reprint records that in answers to questions the size of the Selectron Tube was given as 5 inches high and 3 inches in diameter, with a planned production of 200 tubes by the end of 1946. In the event the tubes were still not available in the spring of 1948 (or even according to Max Newman by the late summer of 1948), and the IAS computer, for which it had been specifically intended, had to be switched to using a Williams-Kilburn CRT store. (Note that it used the store in a parallel fashion, with a 40-bit word stored in a particular position on a set of 40 tubes, so that it could be accessed in around 1/40th the time.)

The plans for the tube itself were scaled down from providing 4096 bits per tube to 256, with the detail of the design significantly different from that described in this lecture. Ironically, the revised Selectron was used in one of the many machines closely based on the IAS machine, the JOHNNIAC (1953), in preference to the Williams-Kilburn CRTs being used on the others, because of its much greater reliability.

Note that the lecture is 7,500 words long and goes into a lot of detail on alternative ways of grouping the wires (3,000 words) and on describing the electrical/electrostatic mechanisms.