Return to List of Reports
Highlights from
Volume 13 ---- Summer 1985 |
Contents of Highlights
| Here, the artist drew the first and last pictures of the series using a tablet connected to a computer, and the machine generated the frames in between. When seen as moving film, the metamorphosis appears continuous. |
|
| Howard Bromberg and Commodore Grace Hopper share a gleeful moment by the infamous COBOL Tombstone. (Photo: Lilian Kemp) |
|
Participants in COBOL's 25th Anniversary Celebration at The Computer Museum on May 16, 1985, surround the COBOL Tombstone. Left to right: Ron Hamm, current CODASYL Committee Chairman John L. Jones, Dr. Jan Prokop, Oliver Smoot, CODASYL Secretary Thomas Rice, current COBOL Committee Chairman Donald Nelson, Commodore Grace M. Hopper, Michael O'Connell and Howard Bromberg. Also present were Connie Phillips and Nan Wilson, the daughters of Charles A. Phillips. (Photo: Lilian Kemp)
| Figure 1. An early RCA cathode ray tube that could have been used for storage. |
|
{Illegible diagrams}
| Figure 6. A 256 digit selectron tube from the ]ohnniac at Rand. Gift of Keith Uncapher and Tom Ellis. |
|
Our approach, the Selectron Tube, was a radical departure from all the cathode ray tube attempts of the time. It utilized a purely digital selection system based on a uniform electron bombardment of "windows" created by two orthogonal sets of parallel bars. By applying appropriate voltages to the bars, the passage of electrons was stopped in all windows except a selected one. The onerous number of individual connections to each bar and its individual drive were avoided by connecting the bars inside of the tube into groups and making connections and drives only for the much smaller number of resulting groups. Such a reduction of addressing channels is possible since the passage of electrons between two bars depends on the potential of each bar. Both need to be relatively positive and equal to each other for the electrons to pass. Hence there is an "AND" gate. By appropriate connections between the bars, a row of bars, or a "picket-fence"; controls N spaces by means of only 2 N at right angles to each other, e.g., an array of 1024 x 1024, or more than a million, could be controlled by only 20 channels. The principle of selection is illustrated by figures 2- 5.
Moreover, the Selectron, in contrast to other memory tubes attempted at the time, used a radically different method for storage. It utilized discrete metal elements that were forcefully maintained at one or another of two stable potentials by a constant electron bombardment. Hence storage of information was not dependent on insulation and did not need any explicit refresh, as in other approaches. The overall electron bombardment of the matrix of bars was not stopped by the bars in the storing condition, thereby providing the "locking-in" current for every element. Only momentarily, during the selection, was that locking current interrupted. Read-out was obtained by using a part of the bombarding current of the element passing through a hole in the element, illustrated in figures 4 and 5.
The particular selectron tube design brought to practical realization had only 256 bits of storage, had a cycle time of 20 microseconds (very short in those days), and required rather extensive power-consuming circuits. (Plans made earlier for larger capacity tubes were not carried out, mostly due to the advent of core memory.)
The Selectron can be viewed as "integrated vacuum technology." We thought of applying such a technology to binary adders and multipliers. These tubes were based on the concept of many internal electrically floating electrodes. Some research was funded by the government and several tubes were partially built. However, the general concept did not seem practical because it required an exact logic predesign that did not tolerate the changes and additions that are inevitable in real life. Incidentally, the early integration of transistor semi-conductor circuits suffered from the same rigidity of design.
During the development of the Selectron, I conceived what came to be known later as the core memory. About a year after we had started to work on it, we heard that at MIT Jay Forrester had independently had the same concept. MIT was working on it for the SAGE project. From that time on we helped each other with frequent mutual visits.
Figure 7. The monster circuitry and power supplies needed to drive the selectron memory at RCA. This machine is similar to the ]ohnniac built at Rand.
The idea of the core memory is very simple. A core is made of a material that has a square hysteresis loop. When magnetized by a current pulse, it will assume one or the other of its two magnetizations, and thereby "remembers" in which direction it was magnetized. This "memory" property is a free gift of nature. The main artifice that had to be devised was the magnetization of one core among many in an array in a desired direction, without disturbing the sate of any other core. This is achieved by the coincidence of two currents, one along rows and the other along columns, whose combined effect magnetizes the core at the intersection. The currents are too weak to singly change the magnetization of a core as their magnetomotive force is below the "knee" of the hysteresis loop. Of course the critical need is for a material with a square loop. Actually I had thought of the concept long before; in fact, I cannot remember when it was not evident to me. However I did not know of any material with a "square loop."
To my great amazement one day, I was reading a technical journal and I found that the Germans had developed a square loop material that was used in magnetic amplifiers for submarines. ARMCO Corporation in Philadelphia acquired the patent rights and were manufacturing the material, which consisted of a very thin ribbon of permalloy. This very delicate ribbon was "wrapped" around a ceramic bobbin. Each such bobbin could serve as an element of the core memory MIT had also discovered the ARMCO bobbins and we both used them in early experiments. They were about $10 each, relatively bulky and delicate. It seemed evident that ferrites would be preferable. Ferrites are made of metal oxides, are insulators, produce no eddy currents, and were and are widely used for high frequency transformers and television yokes. In these applications, any hysteresis produces great losses and is carefully avoided. I approached experts on ferrites at RCA and asked them whether the hysteresis they so carefully avoided could instead be greatly accentuated and I was very surprised that in less than six months they produced excellent square hysteresis materials. We immediately proceded to model tiny cores from those materials. Incidental MIT approached other material experts and also obtained good materials at approximately the same time.
| Figure 10. Detail of an early RCA memory. Note the use of decimal numbers, chosen because of the craze for decimal machines prelevant at the time. |
|
As is well known, the core memory became the standard and was a key in the development of computers. It was surprising that the memory, which by its very operation requires many elements, should be made by discrete elements assembled into arrays. Why not an "integrated" fabrication of some sort whereby all magnetic elements and their linking conductors are made by some overall integrated technique that made the whole array at once. Thus, from the very beginning there was an issue of "integration" versus "automation" (as cores became gradually made by automated presses, were tested automatically and assembled semiautomatically). For example, RCA and Bell Labs made ferrite plates with an array of holes, each threaded by metalized coatings on the plates. Many groups worked on plated wires, which could be made by a continuous process. However, the cores continued to be made by improved methods and, by and large, provided better operation at lower cost, and thus prevailed against all other magnetic memory approaches. In a sense, automation won against integration.
All the efforts at integration were not lost, however. In experimenting with apertured ferrite plates, we invented the transfluxor, a core with two holes, i.e. a relay with no moving parts. The transfluxor was used in some of the early satellites and for foolproof controls in the New York subway. Ironically, the Russians read our papers and used these devices in many industrial controls as they were very slow in developing transistors. Such magnetic logic circuits might be the basis of computers (in fact Univac had a design) if the transistor had not been invented.
A brief mention should be made of our early attempts at integration on a grand scale: planes with half a million bits. These utilized the cryotron, a superconductive switch invented by Dudley Buck at MIT, and made by thin film evaporation techniques. Interestingly enough, our main problems turned out not to be with the indispensable operation at liquid helium temperature, but rather with the problems of imperfections that seem inevitable with such large and dense arrays. It is these imperfection problems that plague present day large capacity chips, and that are being solved by sophisticated error correcting methods and extreme care in fabrication.
The modern development of integrated circuits is of course one of the present day wonders. Memory chips with a million or more bits are being manufactured at very low cost. The integrated circuit memory chips have given us a solution to the memory that is better by orders of magnitude than any previous technology. In fact, it is very difficult to imagine a better technology. The chip is a triumph of fabrication of geometries at the micron, and soon submicron, scale. Operation is obtained by deliberate geometrical shaping and deliberate synthesis of materials, and is all human artifact, not based on some fortuitous natural property, as that of the square hysteresis of some magnetic material.
In the early days, when any workable random access memory was a great achievement, von Neumann thought that a forty thousand bit capacity would be sufficient, provided there was a sufficiently large serial mechanical memory to back it, i.e., tape, drums and later discs. I was always convinced that there is essentially no limit to the need for capacity in the random access memory, and thought that there was no fundamental need for a hierarchy of memories but merely a practical recognition that such hierarchies provide indispensable storage capacity. Today, large capacity chips provide enough memory so that some personal computer systems need nothing additional (HP). This trend will continue into larger computers, particularly when non-volatile techniques are further developed. In the meantime, greater capacity in random access memories are being sought for image storage and manipulation, as well as for many, if not most, tasks sought by artificial intelligence. I believe that semi-conductor technology will provide ever greater, capacities for these uses. Though nature stores in DNA at densities orders of magnitude greater no reasonable proposal has yet been made to exploit such molecular storage for a random access memory or even for a memory that is accessed in some more sophisticated way, such as through the stored contents. Most inventions of men are imaginative intellectual constructs that more often try to defy nature rather than to imitate it.
Figure 11. Cores held on a strand of human hair.
| The first sculpture was a fairly primitive, pterodactyl-looking bird escaping from a cage. The headline proclaimed, "You're free. Honeywell's 'Liberator' lets you switch to the H-200 without reprogramming." |
|
The second sculpture was a racehorse. The headline was: "The Honeywell 200 is off and running."
The dragon on display at the Museum was used with the slogan, "Honeywell's new computers introduce a little magic to banking." Walking around the case, the visitor can see how the components are attached to the wire mesh frame.
After use within the ads, the popular animals were often given as awards to employees and customers. We have heard that the pride of lions lie in rest in Phoenix and a six-foot span eagle is in Washington, D.C. The Museum would like to play Noah and at least compile a listing-one by one-of the locations of the animals with a guarantee that we would take any in and preserve them for posterity
| The fish. |
|
| The fox has a styrofoam base and can be identified as one of the later sculptures because of the use of integrated circuits for the legs. |
|
Return to List of Reports