Martin Banks, Personal Computer World 06/85 - checked

June 1985

Seven years ago I met a man called Ivor Catt, a man who was trying very hard to make a great many waves, and seemed to be creating a large number of enemies along the way.

Ivor, you see, had a dream. He had worked in and around the semiconductor industry for some time, and before that had worked with Ferranti on computer systems when computer systems really were 'clever stuff'. Ivor could genuinely remember working on machines that were the archetype; they had less power than a Sinclair ZX81 and really were as big as a house. He had seen how the computer had come to be organised along the lines established by John Von Neumann, the man who developed the first stored-program computer system.

Ivor's dream was that the so-called Von Neumann architecture, the classic way in which the processor controls all the activities of the memory, peripherals, and so on, was bunk: there were other ways of doing things, he felt. Ways that were quicker, more convenient and, given the majority of tasks that computers are used for, more logical than a centralised single processor having to 'housekeep' an entire computer system.

Like all dreamers, Ivor was laughed at. Most of the semiconductor establishment of the day were refusing to have anything to do with him. 'After all, the world is full of crazy guys who know they have the answer' would be a suitable paraphrase of their collective views on Ivor's dream.

What Ivor needed, of course, was money. This would have shown once and for all whether his dream could be made into a working reality. He managed to get enough in the way of research grants to keep the idea alive, with small projects running at places like Middlesex Polytechnic. He kept body and soul together in those days by running a course on digital electronics, plus some consultancy work.

To the semiconductor industry in general, however, as well as the mainstream computer industry, he was something of a pariah. But that was way back when.

There was a certain wry grin on my face when the Inmos Transputer appeared, for here were some of Ivor's ideas about the structure of computers appearing, in a different form it is true, but without Ivor.

It was with some interest I heard that Ivor had suddenly surfaced again, and not just surfaced like a piece of driftwood. He had risen meaningfully in the harbour of Sinclair Research. There, if anywhere, Ivor could find a home for outlandish ideas, a home where many such ideas have been shown to be perfectly practicable with enough applied courage.

Ivor's dream is about to see some reality as an add-on for Sinclair's QL. If it works, it could point the way to a radical change in the way we all think about computers and their peripheral systems.

I rather liked the idea of Ivor's dream the first time I heard him expound upon it. Certainly, the criticisms he had of Von Neumann's architecture made considerable sense. Imagine for just a second what this architectural structure actually involves. All actions in a computer are controlled by a single processor in a Von Neumann system. It is the processor which goes to memory and receives a program instruction; it's the processor which then goes to memory to get a byte or word of data; it's the processor which then executes the instruction; and it's the processor which then sends that processed data back to memory. Every single action is controlled and executed by the one single processor. It is a fundamental tenet of Ivor's dream that this structure is inherently slow. It can only be speeded up by increasing the speed of the processor, and there must be a finite limit to that capability.

Although such a structure is good at number-crunching applications, the majority of computers are used to performing mundane data manipulations of the 'add-this-data-to-that-record-in-the-file-named-xxx' variety. All very boring and, what is more, all involving intensive processor l/O activity. Essentially, Ivor's dream said: 'Do the processing where the data is - in memory.'

Stage one of a plan to produce such a system (that is, proving the technology) would seem to be what Sinclair is about to launch. Ivor's dream is coming alive as a 'solid-state Winchester disk', an add-on system for the QL which will provide half a megabyte of storage which is available at an access time of under 100 microseconds. This may not appear to be a lot of data, but it will be on tap to the machine at least 100 times faster than anything that can be achieved with a real electro-mechanical system.

It has to be said that this first product will not see a full culmination of Ivor's true dream. It will be a high-speed secondary data storage system that should give the QL some respectable clout as a computer, and which could overcome the drawbacks of the often criticised Microdrives. It will, however, utilise the technology he has propounded for so many years.

This is wafer-scale integration. Instead of cutting up the devices on a semiconductor wafer into individual chips and packaging them for sale, leave them on the wafer and, if the right combinations of chip are put together, connect them up to make a working sub-system or system. The problem with this is that the defects inherent in semiconductor manufacture mean that it would be almost impossible to get a full, working wafer. Ivor Catt found a way round that all those years ago. He called it the Content-Addressable Memory, and the basic structure, which is now to be used in the Sinclair 'Winchester', is also the vehicle by which the full dream of processing in memory can be achieved.

The physical structure of the Sinclair wafer is a collection of serial registers, each with a small amount of switching logic added. The logic dynamically links these registers together in chains, finding workable registers by successively injecting test patterns into neighbours on the wafer. This means that the system is inherently fault-tolerant and can find its way round any defective wafer elements.

This technology is being used first of all to produce what is, in effect, a cheap RAM disk. It will be cheap because one wafer will be easier to produce than lots of separated memory chips. Add a bit more logic, however, and there is the basis of a pipelined processing system. Here, probably, lies the basis of Sinclair's aspirations for fifth-generation computing.

end