From 'Faster than Thought', B.V. Bowden, 1953
THE HARWELL ELECTRONIC DIGITAL COMPUTER
THE HARWELL ELECTRONIC DIGITAL COMPUTER is not intended to undertake enormous computations too lengthy to carry out by other means. It is intended rather to do the work of a few operators with desk machines where the work to be done is of a routine or repetitive nature. As a result, simplicity and reliability have been regarded as of more importance than speed of operation, and the machine performs the operations of addition, subtraction, multiplication and division little faster than a desk machine. Apart from speed, however, the computer provides most of the facilities found in the larger, faster machines.
Relays are well suited to the switching of a multiplicity of circuits and they are therefore used for control and for routing information. To minimize the storage capacity needed, all orders and other data are normally read off a perforated paper tape as they are needed in the calculation. Numerical and other data obtained in the course of the calculation must, however, be stored in the machine. Use of relays for this purpose is not economical and Dekatrons have therefore been used. These are cold-cathode gas discharge tubes having ten cathodes and a common anode. The discharge flows from the anode to any one cathode and may be stepped from one cathode to the next by application of suitable pulses. Each tube can therefore store one decimal digit. Each store uses 9 Dekatrons, providing 8 decimal digits and a sign. There are facilities for using up to 90 such stores, though, at the time of writing, only 40 are being used. The larger number is expected to be adequate for most purposes, since very few orders will normally be stored in the Dekatrons.
The Dekatrons are compact and use negligible power. Their use also simplifies routing, since parallel operation is relatively simple, and numbers in two stores may be added or subtracted merely by transferring the contents of one store directly into the other in a single operation.
The computer contains about 380 relays, 18 Dekatrons, 80 thermionic valves and 40 cold-cathode triodes, plus 18 relays and 90 Dekatrons per group of ten stores. The total power consumption is less than one kilowatt. It has been in use by the Computing Group at Harwell since May, 1952, and is frequently run unattended over-night and over week-ends.
CONTROL AND ROUTING
The computer works on a two-address principle; each arithmetic order consists of 5 decimal digits which specify the operation to be performed and the sending and receiving addresses. There are also orders for changing the source from which subsequent orders are to be drawn, selecting one of the ten alternative layouts for the printed results, and calling for attention. Orders can, if desired, be stored in any of the Dekatron stores, subjected to arithmetic operations if necessary, and drawn from successive addresses for use. At any point in the programme the contents of any storage location may be fed out to one or more page printers (modified teleprinters) or tape perforators. The tape from a perforator may be fed back to a tape reader at the input to the computer, thus providing a restricted form of long-term storage.
Each eight-digit number and its sign are stored in 9 Dekatrons with the decimal point fixed and the numbers stored lying between +10 and -10. Between each of the 10 cathodes of the Dekatron are two guide electrodes, and by making these two electrodes in turn negative with respect to the cathodes the discharge is transferred from one cathode to the next. This feature of requiring pulses in the correct time relationship on two guide electrodes has been used to reduce the number of routing relay contacts associated with each store. A digit is stored in a Dekatron by applying to the guide electrodes the correct number of pulses to step the discharge to the required cathode. Only the zero cathode is separately connected and the method of extracting a stored digit therefore depends on a signal generated when the discharge reaches this cathode.
The basic arithmetical operation in the computer is the transfer simultaneously of digits from one storage location to another. Every tube of the sending store can be fed with a train of ten pulses so that the discharge circulates and returns to its original position. The output cathode of each tube of the sending store is connected through a relay "shifting" network to a transfer circuit. The latter also receives a train of 10 pulses which it routes to one output until the sending tube passes its zero cathode, and then switches the remainder of the train to a second output. The number of pulses appearing at the second output is equal to the number in the sending tube, while at the first output there appears the complement on ten of this number. The complement train is fed to the receiving tube for subtraction, and the number train for addition. If the information in the sending store is not required for subsequent use the train or pulses used for circulation may be derived from the complement outputs of the transfer circuits so that each tube only circulates as far as its zero cathode.
A carry circuit is associated with each transfer circuit. If the corresponding receiving tube passes zero during a transfer, the carry circuit remembers that a carry-over is needed, and routes a pulse to the next more significant digit of the receiving store after the transfer has finished. This may initiate a further carry-over and carry pulses are generated until no more are required.
Addition and subtraction involve two storage positions and do not require the use of the accumulator. Multiplication by each digit of the multiplier in turn requires repetitive addition (or subtraction) of the multiplicand. Storage locations are needed for the multiplier and multiplicand and for the accumulation of the product. Since the product contains more digits than either multiplier or multiplicand, a long store of 15 digits is provided for use as an accumulator. Division, by the usual long-division process, uses this accumulator to contain the dividend, and normal storage locations for the divisor and for the formation of the quotient. The selection of each digit of the multiplier, and the routing of the multiplicand into the accumulator through the shift network, is controlled by sequence relays, but the repetitive transfers are controlled electronically. Multiplication of 2 eight-digit numbers takes between five and ten seconds.
Table VI illustrates the presentation of results which were obtained in a typical computation. Each line involved about five minutes' work by the computer.
1. BARNES, R. C. M., COOKE-YARBOROUGH, E. H. and THOMAS, D. G. A.
Electronic Engineering. (Aug., Sept., 1951)
2. BACON, R. C. and POLLARD, J. R. Electronic Engineering (May, 1950) 173