<< p1 | p2 | p3 | p4 | p5 | p6 | Ode to Gertie | Notes | Comments | Bibliography >>
The DRTE Computer (page 5)
THE COMPUTER
Core Memory
The design of fast, random access memories has always been a central problem in the development of computer technology. Many of the first generation computers used acoustical delay lines to store binary coded information. The acoustical delay lines, however, did not provide immediate access and their storage capacity was limited. Another storage method in first generation computers was the electrostatic tube. Binary data, in the form of a charged points, could be stored on the screen of a cathode ray tube and sensed by the charge it induced on a capacitive type element on the surface of the tube. Electrostatic storage was random but extremely sensitive to outside electrical disturbances. This form of storage was very volatile. There was also the problem of constantly regenerating the memory. It was also difficult to build big parallel memories. (Comment 15)
While working on the Whirlwind project Forrester was forced to look at alternative forms of memory for the Whirlwind computer. The electrostatic memory tubes were too expensive and too difficult to maintain.
"The tubes cost about $1,000 apiece, the Whirlwind project was spending about $32,000 a month for internal memory. Moreover, the computer was unreliable, out of order several hours a day, and unable to run programs requiring a lot of read/write memory. Whirlwind desperately needed a better form of memory." (Augarten,1984:201) (This reference is missing.)
In 1949, Forrester came up with the idea of using the magnetic properties of matter as a way on storing binary information. In particular, he realized that the hysteresis phenomenon could be utilized to produce a non-volatile memory. Hysteresis is a continuous effect in which magnetization lags behind the magnetizing force producing it. Ferrite is a material in which the hysteresis curve changes very abruptly. If a small current is passed through a coil of wire that is wrapped around a ferrite bar, the bar will remain unmagnetized. As the current is increased nothing will happen until a critical value is reached in which case the bar very abruptly becomes magnetized. By reversing the current a similar process will take place; no change in the magnetic properties of the bar until a critical current value is reached. Instead of using a bar Forrester proposed using ferrite toroids with driving wires running through the centres. Depending on the convention, the magnetic state of the toroid would correspond to either 0,1. It took Forrester 4 years to perfect the memory. It was incorporated into the Whirlwind in 1953.
The conventional geometrical arrangement of core memory in parallel machines comprised a stack of two-dimensional core planes. Each core element in the plane had X,Y coordinates which corresponded to the location of one word. The different elements in a word had the same X,Y coordinates but different Z coordinates. Each bit and word were thus uniquely addressable. Thus in a 1024 40-bit memory, the core memory would be comprised of 40 two-dimensional arrays; each array a being a 32 by 32 matrix of toroids. This produced a memory of 1024 40-bit words. Each core had 4 wires going through it; 2 drive wires, 1 read wire, and 1 inhibit wire. Read and write operations to particu7ar bits in the array occurred when currents in the wires through a particular toroid added up to the critical value needed to magnetize.
Cobbold faced two major challenges in the development of the core memory: (1) devising transistor circuits that could produce enough current to enable the core elements and yet have fast switching times; (2) overcoming the small disruptive voltages that could interfere with the read-out signals.
"Because the circuits for the core memory required very high currents for very brief periods time, it required some very fundamental investigation of the properties of transistors," (Cobbold,1985:3). Cobbold faced the problem of having to generate l/2 ampere currents in a microsecond or so. According to Cobbold "the transistors which were available at that time were really quite limited in their capability. It was necessary to find the best way of driving them, to determine which were the best transistors to use, and to devise appropriate circuits'. The outcome of this work was a monostable transistor switching circuit, which he patented with Norman Moody, which made use of complementary pair of junction transistors: pnp and a npn.
According to the patent (Cobbold&Moody,1962), the new switching circuit overcame certain limitations in the previous switching circuits used:
"The limitations involved with using either saturated or unsaturated switching circuits are firstly that unsaturated switching circuits are limited to the current output which they can sustain although the switching time is quite fast and secondly that the saturated circuits while providing an adequate current output are very slow to switch from the conducting to the non-conducting state. The present invention combines the high switching speed of the unsaturated circuit with the high current output of the saturated circuit in a transistor switching circuit using controlled saturation", (Cobbold&Moody~1962:2).
The problem of disruptive voltages stems from the shape of the hysteresis curve. The typical hysteresis curve is shown in Fig. 13. This kind of curve produces a continuum of magnetic field strengths from zero to a certain maximum value as the current varies continuously between two values. If one is going to identify magnetic field strength with binary numbers then it would ideal if a material could be found with a square hysteresis curve (see Fig. l4). Ferrite cores approach this square type curve. The hysteresis curve describing the cores in the DRTE computer memory is shown in Fig.15.
 |
 |
|
|
Fig. 14 |
 |
|
|
|
The small departure of the curve in Fig. 15 from the ideal rectangular curve creates a problem in the detection of the voltage pulses in the read cycle.
"The voltages induced in the read-out winding by the half-excited cores on the selected lines did present a problem, as these 'disturb' voltages are additive and may mask out the desired read-out signal, particularly in large arrays. Direct pick-up from the exciting windings also adds to the disturbs". (Rajchman,1961:108)
The most common solution to this problem was to thread the read wire diagonally through the core elements in . In this manner the read wire passes in alternate directions through consecutive cores in any row or column. Thus the small disturb voltages would tend to cancel out instead of reinforcing each other. Rajchman(l96l) observed that "cancellation is, in general, not a sufficient, remedy to the problem of the disturb voltages..."(Rajchman,1961:108).
In the DRTE computer core memory the induced read voltages were a little over 1OO millivolts. But the disturb voltages were about 30millivolts. Cobbold realized that the ability of the conventional diagonal threading of the read-wire to cancel out the disturb voltages would depend on the exact positioning and uniformity of the read windings. In practice this kind of precision wiring was difficult to achieve. Cobbold concluded that if the read-wire was a right angles to the drive wires, cancellation of the disturb voltages from the cores onto the read-wire would occur for both drive-wires. Describing the wiring arrangement in a 1961 U.S. Patent Cobbold wrote:
"This configuration of magnetic core memory provides advantages over the known structure in that all conductors are threaded through the core in a rectangular system and no conductors are threaded through the core at angles other than 90 degrees with respect to other conductors. When this structure is incorporated into a computer memory utilizing planes composed of rows and columns of these cores, there is no single conductor threading each core of every plane. Since the read-out conductor is at right angles to the two coincident current conductors no voltage due to magnetic induction of the coincident current impulses is induced in the read-out winding and one of the large sources of unwanted signals in a computer memory is avoided". (Cobbold 1961:2,55-68) (see Fig. 16)
 |
|
|
This wiring arrangement had other important benefits. Firstly, it made the wiring process of the core much simpler. Wiring the read wire diagonally through all the core elements was a very time consuming task. Cobbold hoped that his method of wiring would make the assembly of core memories more automatic. It was this practical problem of threading the core memory that seems to have led Cobbold to come up with the novel wiring arrangement.
"There were a number of difficulties associated with making core memories and one of them was the large amount of work that had to go into threading all the cores with wires. It was while I was pondering some of these difficulties that I came up with a new way of wiring up a core which would avoid having any zig-zag wires going through it".(Cobbold,1985:4)
This system of wiring up the core also had the added advantage of simplifying the maintenance of the memory:
"The replacement of a single damaged core is made practicable in a memory according to the invention since it is only necessary to remove the two pairs of conductors threading through a given core and that core will drop out of the memory plane. With known memories it has also been necessary to break the read-out wire at the defective core and resolder it on insertion of a new core. Since magnetic core memories are normally conducted in a very compact form with a typical density ~f 10 cores per inch along the sides of the memory, it will be seen that breaking and resoldering of the read were in the centre of a plane is such a difficult procedure that it has proven less expensive to replace a whole memory plane rather than repair a damaged core." (Cobbold,1961:2,71-3,12)
Because of the nature of the wiring in the DRTE core memory , the geometrical location of a word in memory differed from the usual arrangement where each bit of the word was located on a different core plane. Instead a single word was stored in a single line of cores on the same plane. In this way access to any word required that only one row of core elements in a given plane receive simultaneous coincident current pulses.
As it turned out Cobbold's novel memory system found application in the Prince Albert Data Processing System. The Prince Albert System had, according to Cobbold, memory requirements that resembled those of the DRTE computer. In addition to the DRTE core memory, two core memories of 1024 18-bit words were made for the Prince Albert station according to Cobbold's design.
 |
|
|
 |
|
|
 |
|
|
<< p1 | p2 | p3 | p4 | p5 | p6 | Ode to Gertie | Notes | Comments | Bibliography >>
