previous | contents | next

Chapter 7

The Manchester Mark 11

S. H. Lavington

Upon arrival at Manchester in December 1946, Williams and Kilburn set about perfecting a digital store, at first using the commercially available type CV1131 12-inch diameter cathode ray tubes [Kilburn, 1948; Williams and Kilburn, 1949]. The principle of a two-state electrostatic store can be visualised from the following simple experiment. Start with a focussed CRT beam and turn the beam current on (thus producing a charged "dot") and off again repeatedly. Negative voltage pulses will be induced by capacitive coupling in a pick-up plate placed close to the. outer surface of the CRT screen. Now move the beam whilst it is on so as to write a "dash" on the screen, then move the beam back whilst the current is off, and then switch on the current again. This time a positive voltage pulse is induced. With dots and dashes representing logical 0's and l's, readable as negative and positive voltage signals, a binary storage system is available. Other representations such as a "focus/defocus" system were also used. Now although the electrostatic charge leaks away in about 0.2 seconds, automatic refreshing (re-writing) of the information in less than 0.2 seconds is a simple matter electronically. (cf a modern MOS solid-state store.) Since the refresh rate is rapid, long term drifts in electrode supply voltage, etc are not critical and a robust store can be made from standard components. In contrast, the mercury acoustic delay-line stores chosen by other workers had to be constructed to close physical tolerences. The biggest advantage of the CRT store was that it allowed random access whereas other contemporary systems were sequential.

By the Autum of 1947 the Manchester group had successfully stored 2048 digits for a period of hours [Kilburn, 1948] and the way was clear for the construction of a prototype computer "to subject the system to the most searching tests possible" [Williams et al., 1951]. Kilburn took the initiative with the logical design. The "baby machine," as it was called, had a specification which may be expressed in modern terminology as follows:

32-bit word length

Serial binary arithmetic using two's complement integers

Single-address format order code

Main store: 32 words, extendable to 8192 words, random access

Computing speed: 1.2 milliseconds per instruction.

The instruction format had three bits assigned to the function field, 13 bits to the address field and the remaining 16 bits were unused. The main store consisted of a single CV 1131 Williams Tube, with each 32-digit line occupying about 10 cms on the screen and being scanned in 272 microseconds. A complete "beat" of 306 microseconds consisted of 32 X 8.5 microsecond digit periods plus a four digit fly-back time. The rhythm of the whole processor was synchronised to this store beat. There was notional provision for extending up to 256 Williams tubes to yield a total storage capacity of8192 words. The arithmetic unit was based on a serial subtractor and the logic employed EF5O pentode tubes, used widely for wartime applications. Using this technology, flip-flops (bistable circuits) were extremely costly and temporary storage throughout the central machine was implemented with Williams Tubes wherever possible. Thus the accumulator and control register (instruction counter) were Williams Tubes. One incidental advantage of the use of CRT's was that the contents of main store, accumulator and control register could be viewed on a monitor CRT during or after a computation-so providing a simple output mechanism. Input for the prototype was via a 32-position keyboard and operators control switches.

The machine first ran a program in June 1948 [Williams and Kilburn, 1948] and as far as can be ascertained it was therefore the world's first stored-program computer. A complete diagram of the prototype Mark 1 is given in Williams et al. [1951] and Fig. 1 is a simplified version showing the main flow of information. The Williams Tube which implemented the control register was also used to hold the present instruction (P1) itself subsequent to its being read out of main store. Either the value of control or the value of this P1 could be fed from the "control" Williams Tube to

1Excerpted from S. H. Lavington, A History of Manchester Computers, NCC Publications, Manchester, England, 1975, pp. 7-10. Editor's note: Further discussion of the Mark 1 can be found in the above publication.

previous | contents | next