Return to On Line Documents
This description of the Williams tube memory is from
Computer Development (SEAC and DYSSEAC) at the Nation Bureau of Standards Washington, D.C. NBS Circular 551starting at page 96.
There is a note in the front of the book to Ken Olson (CEO of DEC)
from a Fred Boutin dated 1979 stating
and a reply by Ken saying that he passed it on to Gordon Bell
With the decision in 1950 to develop a Williams-type memory, work at the Electronic Computers Laboratory was planned along five major lines:
An early result of the memory improvement program was the construction of a full-scale experimental electrostatic memory of 48 tubes, each storing 512 binary digits of information. This equipment was completed and placed in experimental operation in SEAC by February 1951 and was one of the earliest full-scale Williams-type systems constructed in this country.
In the SEAC Williams memory,3 each tube is assigned one particular digit of the 45 binary digits making up a word that can be either an instruction or a number. (The other three of the 48 positions in the SEAC word are spares.) All of the digits of one word are assigned the same relative positions in the array of 512 stored spots on each tube. Figure 5.4 is a block diagram of the SEAC electrostatic memory system.
Action during a reading operation is as follows: The address (numerical designation of location) of a word is taken from the Address Register and translated into X and Y coordinates of its location on the tube face by the Staticizer and Deflection Generator. All deflection operations are done in parallel. In 3 usec, the deflection transients disappear, and the beam is turned on in all tubes for 0.5 usec (see fig. 5.5). The outputs from the signal plates are amplified and sensed during the STROBE period which occurs within this time. A positive output represents a binary one, and a negative output represents a binary zero. This information is transferred to the Shift Register in parallel, from which it is fed serially to the arithmetic unit or to an output device.
Since the act of reading causes all spots to be written to "zero," the "ones," must be restored. To accomplish this, at the end of the DOT pulse all tubes are provided with a sawtooth deflection pulse which displaces the beam about one spot-diameter. If the output from any tube is positive at STROBE time, the beam is held on by the DASH pulse which occurs during the TWITCH. The spot charged during reading is then discharged by secondaries from under the moving beam and a one is thus restored. If the output of the tube is negative, a zero has been sensed and rewritten during DOT so that no other action is needed. The absence of the positive output prevents the beam from being held on during the TWITCH and the spot is left charged. The total time allowed for this read-write operation is three microseconds. An additional time of 6 usec is required to allow the staticizer pulse to swing back, making a total operation time of 12 usec, 3 for deflection, 3 for read and rewrite, and 6 for flyback. Writing is done in similar fashion. The information is stored in the shift register and is used in place of the output of the amplifier to control the holding on of the beam during TWITCH.
Since the charges stored tend to leak away or are discharged by secondary electrons and stray primary electrons during reference to neighboring spots, it is necessary to regenerate (read and rewrite) each word periodically. The operation is the same as consultative reading, except that the source of the address to be rewritten is the regeneration counter rather than the address register, and no information is gated into the shift register. The regeneration counter advances consecutively through the electrostatic storage addresses as each word is regenerated. Since the minimum time to perform an operation in the arithmetic unit of SEAC is 48 usec three regenerations can be carried out between each useful consultation. If there is no consultation of the electrostatic memory at the allotted time, an additional regeneration occurs. This may happen when S EAC is using both memory systems and an address in the acoustic memory is called for, or during a long operation such as multiplication.
In operation with SEAC, over 1,500 hr of useful operation have been obtained, but it has never reached the long-term reliability of the SEAC mercury memory.
The problem of supplying storage tubes for a Williams-type system has proved more difficult than was originally anticipated. During the early planning of Williams-type memory systems in this country, standard cathode-ray tubes were expected to provide adequate data storage, and all but one of the currently operating computers which have Williams memories do use standard types of tubes. However, standard tubes which were commercially developed for other purposes have certain limitations when. they are used in high-speed memory systems. These limitations include nonuniformity of surface, which causes the blemish problem; interaction between storage locations, which causes the read-around-ratio problem; and interaction between electrodes of the tube, which causes the crosstalk problem.
Performance of the standard type tubes in operating computer memories has not been as good as has been hoped. In order to use these tubes reliably, either the number of bits per tube has had to be reduced below the number originally planned, or the ratio of references to the memory per regeneration has had to be limited. It has also been necessary to reject about four out of every five tubes received in production lots; acceptable tubes are therefore more expensive than had been expected.
Two lines of attack on the problem of improved tubes have been emphasized at the Electronic Computers Laboratory. The first approach was to try to obtain an improved tube which, although special purpose, would nevertheless be less expensive than intricate tubes proposed for other systems of electrostatic storage. The second approach was to try to circumvent the faults of available tubes. Both lines of attack have recently been quite successful.
On of the major problems is the presence of minute areas of low secondary emission ratio on the storage surface of the tubes. These spots are variously called blemishes, flaws, or (in England) phonies. Storage is impossible on these areas, and they must be avoided or the tube must be discarded. Unfortunately, the dodging of blemishes is difficult when many tubes are operated in parallel. In this case, computer operation may become unreliable unless very careful engineering prevents motion of the spot over the face of the tube.
Blemishes appear in about 80 percent of the tubes produced commercially, largely because extreme cleanliness is required to produce blemish-free surfaces. Blemish-free surfaces have been successfully produced in this and other laboratories when care is taken to maintain a high degree of cleanliness. Manufacturers of cathode-ray tubes interested in good surface quality are cooperating in the work of improving the tubes, as shown in table 1.
TABLE 1. Work at commercial tube plants
|Manufacturer||.||Success||Direction of effort||Remarks|
|A--------||Smaller spot size; improved spot profile; improved deflection defocusing characteristics.||A likely design was produced.||Further testing of this model.||Design originated at NBS Computer Laboratory.|
|B--------||Smaller spot size; better beam profile.||Tube with half spot size, twice storage capacity of 5UP11.||Possible further testing of this gun in 3-in. bulb.||Finished sample tube delivered one week after initial discussion.|
|C-------||Over-all improved storage tube.||Final sample tubes were blemish-free at 2KV. Read around ratio was very high compared to that of standard CRT used.||Small-scale of production of further improved tubes, some refinements to be considered.||Work done under BuShips contract. Testing done at many cooperating government laboratories.|
|D--------||Experimental tubes for surface studies.||Study produced. No exceptional surfaces. Good bulb design for gun-signal plate shielding.||Small production with standard phosphor for tests in SEAC.||For experimental use only. Tube design originated at NBS Computer Laboratory.|
|E--------||Fine spot, prefocused gun.||-----------------------||-----------------------||In early stages. Work being done under BuShips contract.|
The investigation of tube blemishes undertaken at NBS was first directed to the determination of the nature of the blemishes so that some control might be attempted and to correlating the location of these blemishes with the location of physical blemishes in the phosphor. Specially coated plates were checked in a demountable system under a microscope for this purpose. Correlation was low. An attempt to determine the size of a blemish was made by using India ink markings on a glass plate. It was found that a positive blemish (defect in an area coated with ink) or a negative blemish (spot of ink on a preponderantly glass area) could be identified when one dimension was equal or greater than the beam diameter.
The first method of finding blemishes was to sweep over a rectangle on the face of the tube with a Lissajous pattern produced by sinusoidal X and Y deflection voltages and to apply the same deflection voltages to a monitor tube. The output at the signal plate of the storage tube was amplified and applied to the grid of the monitor. As long as there was no change in the storage surface, the brightness of the monitor was constant However, when a blemish was struck by the electron beam, it caused a change in brightness on the monitor, and its relative position was marked by a bright or dark spot. The system located the blemishes but did not tell whether they would affect storage. - This test was very sensitive but only qualitative.
A test called the Line-of-Dots test was developed at this Laboratory. It was not as sensitive but was able to determine whether a blemish would affect storage. Results of this test correlate very well with the performance of tubes in actual operation, and the test is now used as a specification test by a commercial manufacturer of a developmental Williams tube.
The Line-of-Dots test operates as follows: The beam is turned on and swept across the horizontal axis of the tube under test during the first half of the cycle. During the second half-cycle the beam is swept across the same line, but the beam is turned on only during 10 very short periods. During each one of these short periods the screen on the front of the tube picks up a positive-going signal which is very similar to the usual "dash" signal. These periods are nonsynchronous with the sweep, so that the entire line is interrogated during successive sweeps. When displayed on the vertical axis of a synchronized monitor, the envelope of the dash signals will show a notch or dip at a point where there is a blemish. The block diagram of this equipment is shown in figure 5.6.
Because blemishes afflicted 80 percent of the tubes, attempts were made to remove them. The use of infrared, ultraviolet, and electron bombardment failed to affect them. Rapping with a soft mallet helped to remove some. The best tool seemed to be a spark coil such as that used for checking vacuum leaks. The high-voltage brush discharge across the face of the tube charged the blemish and the surface alike, and the electrostatic repulsion removed many of the blemishes. This method was refined to the extent that 80 percent of a batch of tubes which were treated in this manner were acceptable.
Another aspect of the storage surface problem is the aging of the phosphor during use. A tube which has been in use for some time frequently develops a set of blemishes corresponding to the raster position. These raster burns usually do not affect storage seriously until the tube has had a reasonable life. Los Alamos, however, reports that in MANIAC it is necessary to replace tubes at an undesirable rate because of raster burns. Although our studies of this phenomenon are not complete, it does appear that the difficulty can be prevented by baking the phosphor at higher temperatures than are normally used. This indication has been tentatively confirmed by British work.
The second major difficulty encountered in Williams-type storage is the obliteration of the stored charge at one spot by the splash of the secondary electrons emitted from neighboring spots. The number of consultations of neighboring spots which may be made before regeneration is required is called "read-around ratio," "splash number," or "repetitive consultation number." Low readaround ratio may be improved by one or more of the following methods: by reducing the total number of spots in the tube and separating them further, by systematically forcing sufficient regenerations per useful consultation of the memory, by taking the ratio into account during coding of problems, or by improving the intrinsic properties of the tube in this respect.
Low read-around ratio appears to be mostly a function of gun design. Stray electrons not in the focussed bean, poor current distribution in the beam, and large beam diameter contribute to this effect. It was felt that a relatively small effort to obtain smaller spot size and a more rectangular current profile night result in considerably improved read-around ratios. Indeed, one company (manufacturer B, table 1) provided the laboratory with an improved tube a week after the first consultation. This tube, called the 51CUP11, had a gun in which the beam was heavily masked. It would store 1,024 bits in SEAL, whereas a 5UP11 usually would store only 512. A sample lot of these tubes was produced, but they were unsatisfactory for other reasons, and no more were ordered. However, the original tube was used in SEAC for about 1 1/2 years alongside VP and 5UP types.
Later the Bureau of Ships approached Manufacturer C to do development work on storage tubes. A contract was let that provided for the production of several trial designs to be followed by small lot production of the best of these designs. In cooperation with the National Bureau of Standards, which was acting as technical consultant on the contract, the Argonne National Laboratories, the Institute for Advanced Studies, and a commercial laboratory tested these trial tubes. The purpose was twofold; first, to determine the best design of the developmental lot for use in the final sample production, and second, to determine how well the various tests used in the different laboratories agreed among themselves and with operation in a computer. The agreement among the tests in arranging the tubes in order of quality was quite good, considering the variety of test conditions. This order also agreed well with their performance in SEAC.
A second lot of tubes, all alike, are now in process of being tested in digital computers at various laboratories. These computers include ORDVAC at the Ballistics Research Laboratory, Aberdeen Proving Ground; ILLIAC at the University of Illinois; AVIDAC at Argonne National Laboratories; the Institute for Advanced Study's computer; and SEAC. Results to date are very encouraging. On the basis of these results, a conference held at the National Bureau of Standards led to agreement on specifications for a computer tube similar to this lot, and small-scale developmental production of the tube has begun.
This improved tube will have a gun in which the beam is heavily masked. Shielding is carefully designed to decrease the number of stray electrons. Distortion of the beam because of deflection is reduced by reducing the beam diameter in the deflection region and shaping the deflection plates. A storage surface of magnesium tungstate is used because of its good secondary-emission properties and because better control of screening in manufacture is expected. A 3 in. diameter was agreed upon because it was felt little could be gained by going to a larger tube, and considerations of space requirements and of safety favor as small a tube as possible. The specifications are such that no significant blemishes should appear at 2,000-v accelerating voltage, and the read-around ratio should be at least 256 in a field of 1,024 spots.
In the meantime, also under a Bureau of Ships Contract, another company (Manufacturer E) is developing a fine-spot gun with a magnetic fixed focus. The elimination of any electrode or electrode voltage is desirable because it reduces the number of parameters affecting the operation of the gun.
The third major fault with standard cathode-ray tubes when used as storage devices is the crosstalk among elements of the tube and between the signal plate and tube electrodes. For example, in the 3KP-5UP gun a deflection lead serves as a support for the grid cylinder, and the sharply rising grid pulse is capacitively coupled more to this deflection plate than to its mate. If the unbalanced coupled signal is large enough, the spot is in motion during writing or reading, which tends to spoil "dots." Again, the deflection plates in a 3-in. tube are rather near the signal plate. As these are pulsed with rapidly rising pulses of the order of 100 v and the normal signal level is about a millivolt, it is not hard to couple in signals that might cause the amplifier to block or ring so that false data are stored. A study was made of these effects, and it was found that grounding a conductive coating painted on the outside of the tube gave a substantial improvement in tubes without third anodes. A third anode gave the best shielding between gun and signal plate.
3 A. W. Holt and W. W. Davis, Computer memory uses conventional C-R tubes,
Electronics (December 1953).
Return to On Line Documents
Return to On Line Documents