I thought I’d look into this mention of MOSAIC, from the nearby thread General Purpose Systems Simulator (1960s):

(Diagram from PO EEJ - see more below.)

Ed Thelen has permission to put this book by Simon Lavington online:

Early British Computers

and in there, in Chapter 10, we find some MOSAIC details.

As orientation:

…at the Post Office’s Dollis Hill research laboratory both Flowers and Coombs, ex-Bletchley men, had the necessary interest and expertise to implement a stored-program computer, and indeed the immediate post-war intention was for the Post Office to do just that…

And here’s MOSAIC:

… MOSAIC (Ministry of Supply Automatic Integrator and Computer) project, which was implemented between 1947 and 1954 by a Post Office team led by Dr A. W. M. Coombes. NPL contributed to the mathematical specification and the All-Power Transformer Co. helped with the manufacture and assembly. Parts of the MOSAIC project are still secret, it having been used for processing radar tracking data in experiments on aircraft. … The computer itself had storage for 1024 40-bit words in mercury delay lines, involving nearly a ton of triple- distilled mercury. The physical layout of the delay-line tanks was copied from the scheme used at Cambridge University for the EDSAC. The man who had helped to build COLOSSUS was not afraid of thermionic valves: Coombs used 6000 of them in MOSAIC which, together with 2000 germanium semiconductor diodes, gave a total power dissipation of 60 kilowatts. Although not the earliest of early British computers, MOSAIC was arguably the largest.

The MOSAIC was housed at Malvern and, whilst it was no doubt very valuable in defence circles, its design had little influence on the mainstream of computer development. In view of the price of mercury, perhaps this was just as well.

Here’s an earlier version of a photo found in the book:

Fig. 10.1 Some central units of the MOSAIC computer, built by the Post Office for a secret Ministry of Supply defence project at Malvern. MOSAIC was working in about 1953 and in terms of electronic components was the largest early British computer.

Nearby, for interest, a map:

Fig. 5.1 Map showing the location of some centres of early computer activity. The National Physical Laboratory is at Teddington, the Telecommunications Research Establishment at Malvern, and the wartime Code and Cipher School at Bletchley.

Lavington supplies some references which we can chase:

- Coombs, A. W. M. An electronic digital computer, Parts 1-4. Post Office EEJ, 48, 114, 137, & 212, July & October 1955 & January 1956; 49, 18 & 126, April & July 1956. (Summary information also given in Automatic Digital Computation, pp. 38-42, see reference 54 below.)
- Coombs, A. W. M. MOSAIC - the Ministry of Supply Automatic Computer. In Automatic Digital Computation, (Proceedings of a symposium held at NPL, March 1953), pp. 38-42. HMSO, 1954.

Helpfully, issues of the Post Office EEJ are online at WorldRadioHistory, so we can check those out (1950s editions all here - fairly slow to load!)

In passing, in the Oct '56 anniversary edition we find a discussion of Dollis Hill research, and in the subsection on Automatic Switching and Signalling, Electronic Switching we find

The possibilities offered by what is now described as “electronics” were appreciated from wartime work which brought Post Office engineers into contact with engineers and scientists in other fields. Artillery predictors, digital computers, code converters and numerous other applications clearly established the versatility of electronic devices. Perhaps more important still, mathematicians had their attention directed to problems then being solved by electronics, and they made notable contributions to the basic concepts underlying the design of automatic machines.

The post-war research program included the construction of a large digital computer, subsequently christened “Mosaic,” and also an investigation into the possibilities of electronic switching in the telephone system. The Mosaic computer uses ultrasonic pulses in mercury as the memory organ and vacuum valves as switching elements.

The basic switching operations are timed at intervals of about 1 µs, which is about 1,000 times faster than can be achieved with relays. The large memory capacities and high switching speeds suggested new kinds of register-translators (or directors).

Now to the four-part series “Mosaic — An Electronic Digital Computer” by Coombs (each part is only a few pages):

This article, to be published in four parts, describes a high-speed electronic computer designed at the Post Office Reseurch Station for the Ministry of Supply. Part 1 explains the need for such a machine and goes on to give a general description of two of the chief components, the Store and the Arithmetic Unit. Part 2 will complete the outline by describing the other two chief components, the Control Unit and the Input-Output mechanism. Part 3 will be devoted mainly to the use of the machine, particularly as regards Programming. The electronic techniques employed will be discussed in Part 4.

Part 1 —The Store and Arithmetic Units

The number of separate arithmetical operations in even a small calculation may be very large when each operation is limited to a simple addition or subtraction. Digital machines must therefore be designed to carry out any one such operation at very high speed, and in fact Mosaic can add together two numbers, each the binary equivalent of 12 decimal digits (2^40) in 70µs. Clearly, to use the machine time efficiently, it must be possible to feed the instructions in at a comparable rate, and since this is well outside the scope of any human operator, it follows that the whole programme of instructions must be fed into the machine in advance and held in some form of store which will allow the machine to draw on it at high speed as the computation advances. In addition to storing the instruction programme, the store must hold the data figures relating to the problem being solved, interim results (the sort of figure one scribbles in the margin) and possibly the numerical values of such physical constants as 𝝿 and

𝑒. It must also act as the intermediary between the essentially low-speed input and output gear and the high-speed computing circuits. It is therefore a vital part of any digital computer.

The operations, apart from multiplication, take place in precisely the time taken by a 40-digit number to pass a given point—that is, one minor cycle, or about 70µs—but the multiplier requires about 6 ms to produce its answer. Therefore, when a multiplication is called for, it is arranged that the multiplier is ‘‘sealed off” from the rest of the machine while the operation is carried out; the other functions of the machine can continue to be employed, and the result of the multiplication, the product, is placed in a temporary store of its own.

Part 2 —The Control and Input-Output Units

It is always difficult to describe a sequence of events forming a closed ring, where each part of the sequence depends on what has happened immediately before, and there is no beginning. The Control unit involves two such closed rings—the Current Instruction Staticiser with its circuit, and the main Control itself.

With all its complex functions, the Control unit uses only about 500 valves of the 7,000 in the machine, and conveniently occupies one complete rack.

Part 3(a) —The Art of Programming

Part 3(a) deals with the machine as a whole and Part 3(b) will deal with programming in general and will give examples of programming for particular problems. Diagrams will still be schematic, for it will be in Part 4 only that the major problem of translating the theoretical ideal machine into practical circuitry will be considered.

Now-familiar concepts here, about binary arithmetic:

when a multiplication is called for, it must also be specified whether the numbers are to be regarded as “signed” or “unsigned.” Addition and subtraction may require suppression of the "carry” digit after the 40th digit, after the 80th digit (double-length arithmetic) or not at all, and this facility involves four possible function instructions for each.

Part 3(b) —The Art of Programming (Continued)

It is convenient at this stage to draw up a chart of the procedure that is to be adopted. For a short computation, it may be a few written lines indicating the sequence of major operations to be performed; for a long computation it will probably be a Block Diagram, each block covering a whole group of operations, with connecting links showing how the machine is to move through the diagram. In a large programme, each block may be a complete problem in itself (for instance, a long division) and may be used several times; such a block is called a “Sub-routine” and the programme to cover it may well be already available to the programmer, it having been worked out for some other problem.

It is now necessary to produce a Flow Diagram. This is a more detailed chart showing the path of the computation in machine language—that is to say, it is the problem broken down-into simple additions, subtractions, multiplications and comparisons and nothing else.

Finally, the flow diagram is put into "machine order”— that is, the instructions are arranged in the order the delay lines will hold them in order to give maximum possible speed, which implies that as many instructions as possible are to be available at the exact moment when they are required. This operation usually requires fairly complicated juggling with characteristics, timing numbers and "next instruction sources.”

There’s a footnote:

It is good practice to make subroutines as general as possible. The “cossin” routine can be used whether cosine or sine or both are required, and will indirectly provide tan 𝜃.

Here’s a time-space tradeoff:

The seven constants 1/2! to 1/14! could be worked out by the machine, but it is more convenient in this case to hold them ready and complete in the main store.

There is a school of thought which regards the valve as the weakest link in an electronic machine—as a short-lived and unreliable structure, to be used sparingly.

The designers of Mosaic took the directly opposite Viewpoint, that circuit troubles were likely to be much more difficult to deal with than valve troubles, and that the complex circuit was therefore the real danger.

… the whole argument breaks down if valves are really unreliable components; a machine with very safe circuits is still useless if it can never work more than a few minutes without a valve failure. The policy of valve multiplication must be accompanied by special precautionary measures. Thus, in Mosaic, the standard valve for general purposes is the CV138 miniature pentode, which hr.s probably been the subject of more research to make it reliable than any other valve, and its characteristics are therefore well known. About 5,000 of the 7,000 valves in the machine are CV138.

…

…preliminary ageing (to show up potential mechanical faults) and, after that, great care in handling. For instance, withdrawal of valves for general routine testing was forbidden; “leave well alone” was to be the rule. Further, it was arranged that heater voltage was applied and withdrawn slowly, over a full minute, under motor control.

… circuit designs using only about two-thirds of the theoretical anode current

The provision of a synchronous clock to a large machine was done by making a standing wave in a resonant composite cable:

In Mosaic, the total length of cable required was about 4𝝿/3 radians (200 yd), and this was conveniently obtained with two composite cables, each of two sections, 𝝿/3 radians per section, joined at the far end to make a ring main…

One of the problems assoccaited with mercury-line storage is that the pulse velocity down lines is variable with temperature. It is, however, vital that a delayed pulse emerging from a line shall be synchronizable by the correct K-pulse of the correct P-digit of the correct minor cycle — that is, that the delay of a long storage line shall be 638⅔ pulse periods for Mosaic, within very close tolerance. Therefore, either both temperature and Master Clock frequency must be constant, or else both must vary together according to some rule which will maintain the required equality; Mosaic uses the latter scheme, with one long delay line (No. 0) used as control. It is still necessary that all the lines be at the same temperature,

In conclusion:

Perhaps it would be true to say that present interest is largely on machines for commercial use. Mosaic has made its distinctive contribution to the art; it is at present the only 4-address machine in use in this country, and its K-pulse technique, with standing-wave control, resolves very satisfactorily the problem of combined high speed and large size, a problem which was bound to arise sooner or later.