Mark 1 Story : Introduction, The Baby, Manchester Mark 1, Ferranti Mark 1
Current Page : Background, CRT Store Principle, Refreshing Process, Bottom of Page

The Williams Tube
or The Williams-Kilburn Tube

Background

As a result of a trip to the U.S.A. in June 1946, Dr F.C. (Freddie) Williams started active investigation at TRE into the storage of both analog and digital information on a Cathode Ray Tube. Storage of analog information could help solve the problem of static objects cluttering the dynamic picture on a radar screen (because). Storage of digital information could solve the problem holding up the development of computers worldwide, i.e. lack of a storage mechanism that would work at electronic speeds. By November 1946 he was able to store a single bit (with the "anticipation" method), based around a standard radar CRT, and filed a provisional patent for the mechanism in December 1946.

In December 1946 Freddie Williams was appointed to a chair at the University of Manchester, and left TRE. However both he and TRE wanted the research to continue, so Tom Kilburn, who was in his group at TRE, was seconded to the University of Manchester to continue the work with Freddie Williams on digital CRT storage. A Scientific Officer from TRE was also seconded full time to help him, initially Arthur Marsh, who left after a few months, and was replaced in the summer of 1947 by Geoff Tootill.

By March 1947 Tom Kilburn had discovered a different and better method of storing information, more suited to storing a large number of bits on the same tube. By November 1947 they had succeeded in storing 2048 bits for a period of hours, having investigated a number of variations on storing a set of bits (dot-dash, dash-dot, defocus-focus, focus-defocus).

The general principle behind the storage of binary information was to plant charge in one of two different ways at an array of spots on a CRT using standard techniques. The type of charge at any spot, representing a 0 or 1, could be sensed by a metal pick-up plate on the outside of the CRT screen, thus "reading" the "value" of the spot. However, the charge dissipated very quickly, so values were preserved indefinitely by continuously reading their value and resetting the charge as appropriate to the value.

Although the phosphor on the CRT would glow at charged points, in a way that might be distinguishable (between 0 and 1), the current contents of a CRT storage tube could not be viewed directly in practice. The front of the tube would be obscured by the pick-up plate. (At first this was a wire mesh, but later it was a metal plate.) Also the CRT had to be screened from outside electrical interference, e.g. local trams or close and aggressive motor cycles, so it was enclosed in a metal box. So typically the information on a Williams-Kilburn Tube would be displayed on a separate Display Tube, which would be updated synchronously with the refresh cycle of the Williams-Kilburn CRT Store. However the opportunity would also be taken to lay the values out in a manner most convenient to the onlooker.

The memory storage system nearest to a successful realisation elsewhere was the Mercury Acoustic Delay Line store, which was chosen as the basis for the earliest active stored-program computer projects (EDVAC, the EDSAC and ACE). The most significant advantage of the Williams-Kilburn CRT Store over a Delay Line store was that the CRT store allowed fast random access to short strings of bits, e.g. 20-bit or 40-bit addressable strings corresponding to "lines" or "words" in RAM. This contrasts with the sequential access mechanisms inherent in Delay Line stores, which would store say 1024 bits in a single delay line and a smaller string could only be read when it "came round" on a 1024-bit cycle. The Williams-Kilburn CRT Store also had the practical advantages that it was made of standard components, was cheap and compact, and did not require temperature control or accurately controlled power supplies.

When in Autumn 1947 the group had successfully stored 2048 digits, they had the problem of proving that the store would operate successfully inside a computer. They could only alter bits at the rate of around 1 a second, which was 100,000 times slower than the store's capability. In the end they decided that the simplest way to test that the CRT storage system was suitable for use in computers was to build a small computer round it. So the way was clear for the construction of the SSEM, the "Baby", "to subject the system to the most searching test possible." The Baby was therefore built to show that Williams-Kilburn CRT Stores could work dynamically in a working computer, and to demonstrate the feasibility of building a much larger machine using them. That it also was the first machine to realise the stored-program computer, and prove its feasibility was a secondary consideration.

Although by 1948 standards the Williams-Kilburn CRT Store was acceptably reliable, it was not 100% so, and there was regular experimentation and development for many years from 1947 onwards to improve its reliability. The method described below is the "dot-dash" method used on early machines.

The Williams-Kilburn CRT Store was used in several models of computer including the IBM 701 and 702 computers. It was eventually superseded in new systems in about 1955 by a cheaper random access store called the "magnetic core store."

The successful operation of the Williams-Kilburn CRT Store before other storage systems was the key factor which led to the Manchester group being the first to build a stored-program computer.


The term "Williams Tube" was coined by the Americans and has been used universally since (up to 1998) to refer to the family of storage devices based on the Williams-Kilburn patents. It was never called the Williams Tube in the early papers by Williams and Kilburn.

Only the provisional patent of December 1946 was taken out in Freddie Williams' name alone; all the other patents had both names on. Tom Kilburn was able to work full time on the development that took place throughout 1947, whereas Freddie Williams, as head of a Department, had many other duties. Also it was Tom Kilburn who by March, having got the original experiment transported to Manchester and back into working order, observed that there were better methods than the "anticipation" method.

Tom Kilburn contributed most of the work and much of the new ideas in 1947, and produced the key document at the end of 1947 that was influential in other groups adopting the Williams-Kilburn CRT Store (most notably IBM).

So in retrospect a more considered coining (if less convenient) is arguably the "Williams-Kilburn Tube". Throughout these pages the preferred usage is "Williams-Kilburn CRT Store" to refer to the general mechanism and "Williams-Kilburn Tube" to refer to an individual Cathode Ray Tube (plus the additional pick-up plate and circuitry).


The Williams-Kilburn Tube Principle

This description of the dot-dash method for a Williams-Kilburn CRT Store, is an edited version of one provided by Prof. D.B.G. Edwards.

The principle of the CRT store can be summarised as follows. Bits of information are stored on the insulating screen of a CRT as small areas of electronic charge in an array of isolated dot/dashes. If a metal pick-up plate is placed close to the outer surface of the CRT screen then voltage pulses will be detected when the charge is changed, due to the capacitive coupling. These changes occur successively in time when the storage areas in the array are bombarded by the CRT electron beam in a regular time sequence in each full screen refresh cycle. Their nature is now described.

Suitable X and Y deflection voltages and a short repetitive bright-up signal Z applied to the CRT grid will cause an array of dots to be generated on the screen at a set of fixed positions. If a particular fixed dot position is to store a 0, then the bright-up signal Z is applied precisely to that position, and the screen acquires a positive voltage when it is bombarded with a high energy electron beam due to the fact that the secondary emission ratio of the screen material is greater than 1. Provided that the dot position is bombarded again before the positive potential has had a chance to decay significantly, then there will be no change of charge at the dot position, and the screen holds a 0 in this position.

If a dot position is to hold a 1, the bright-up signal is lengthened so that the dot is extended into a dash. Now the initial positive potential in the precise dot position attracts secondary electrons from the bombardment occurring close to it as the beam moves along the dash, and is rapidly lost. This means that when that dot position is bombarded again on the next refresh cycle it will have to acquire a positive potential and that change of charge from the storage area will provide a positive signal at the pick-up plate during the dot period. So a 1 has been stored and detected, and if it is to be preserved then the Z bright-up signal must be extended to cause a dash to be rewritten.

In practice a transient signal is observed on the pick-up plate when the electron beam is turned on for a dot position and again when it is turned off. These signals are due to the cloud of secondary electrons introduced in the vicinity of the screen when the electron beam is switched and is negative going at switch-on and positive going at switch-off. Therefore both a "screen size" pick-up plate and "dot size" pick-up plate are used, to detect whether the signals are whole-screen related rather than single dot related. If they are whole-screen related then there is no signal due to potential change at the dot position, so the value is a 0, but if they are dot related the potential has changed, so a 1 was stored previously and the beam must be extended along the dash.

Normal leakage from the screen would destroy the storage charges in tenths of a second but refresh rates significantly faster than this read the information and replenish the charges so that information can be retained indefinitely. It also means that long term drifts in the voltage supply, which cause slight movement of the array positions, are not critical to the storage operation.

The Refreshing process

The refresh scan for a 32/40 bit word on the Manchester Mark 1 machines was typically around 300/400 microseconds. Reading from and writing to a Main Store CRT was interleaved with refreshing scans, with each line of each CRT store being refreshed in turn cyclically. Operations not requiring Main Store access would be overlapped with refresh scans.

So for example on the Baby there was a simple "four beat rhythm" for a standard instruction:

  1. Refresh the next line (in the cycle) of Main Store; add 1 to the address of the current instruction
  2. Fetch the next instruction from Main Store
  3. Refresh the next line (in the cycle) of Main Store; decode the instruction
  4. Read/Write from/to Main Store as required; complete the instruction
Each beat took the same time (hence the official 1.2 millisecond instruction time). The complete 32 word Main Store would therefore be refreshed every 16 instructions, well before there was any significant discharge.

Note that where there was a 32-bit operation taking place in beat 4, i.e. A = A - S, the operation would still be completed within the beat, since the 32 bits were read from store serially and the combination with the accumulator value could take place serially at the same time.

On later Mark 1 machines the pattern of beats got increasingly more complex and irregular, but the same principle applied, i.e. of interleaving refresh scans with Main Store read/writes and overlapping refresh scans with other activity. Of course where there were a number of Main Store CRTs, each of the corresponding lines on each CRT would be refreshed simultaneously at each refresh scan.



Mark 1 Story : Introduction, The Baby, Manchester Mark 1, Ferranti Mark 1
Useful Links : Display Tube, F.C. Williams, Home Page, Picture Gallery, Mark 1 Literature
Context : 50th Anniversary pages (The Mark 1 story, Celebrations, Virtual Museum)
        at : the School of Computer Science, The University of Manchester
Maintainer : Brian Napper; last updated November 25th 1998

Copyright The University of Manchester 1998, 1999