Ivor Catt Archive | Dinosaur Computers - 48 ELECTRONICS WORLD June 2003 p50

continued from previous page]

Thus, my public campaign for the array processor began with my article in New Scientist in 1969 entitled Dinosaur among the Data? I argued that computer applications of the future would need an array of processors and that the single processor (von Neumann) machines like those of twenty (now fifty) years earlier would not meet our needs. Next came my March 1989 article in Electronics World + Wireless World entitled 'The Kernel Machine', where I discussed the many applications which needed a large array of processors. The latest publication in the list is by Nigel Cook in Electronics World January 2003, Where he argues that Air Traffic Control must use an array processor. A tangential article was 'The New Bureaucracy', also published in Wireless World December 1982. (See yellow box).

Liquid cooling

The reason why the Von Neumann computer has minimal problems with heat extraction is because, with only one processor, it does so little computing. Computing consumes power because every time a logic line is raised to the value 1 and then discharged back to the value 0, some electric charge has been dumped down through five volts, thus losing energy. In contrast, memory, which is what virtually the whole of a Von Neumann computer does, does not in principle consume power, especially a DRAM.
A million processor Kernel Machine consumes a helluva lot of power while working, that is, while computing. A liquid is chosen from the wide range available which boils at the fastest switching temperature for the transistors used. Using boiling liquid on the surface of the chip gives a thermal clamp, so that all components operate at the single boiling temperature. This means that the larger logic swings previously needed to tolerate temperature variations across the machine are not needed. The resulting lower logic swing leads to lower power dissipation. The advantage is startling because power dissipation is more than proportional to voltages used, in the same way as the power dissipated in a resistor depends on the square of the voltage across it. Weirather and Go did the research for me in Motorola R&D to prove conclusively that direct chip contact liquid cooling is perfectly viable, and published a paper on it. Since then, liquid cooling has been totally ignored. This is partly because the USA went lunatic. Liquid cooling is unsuitable slopping around half way to the moon, as is the high power and high speed, which liquid cooling makes possible. There is not much to compute on the way to the moon anyway. However, the core reason for blocking liquid cooling is that we dare not depart from usage of the Ancients, and the way they designed and built computers in 1950. They didn't use on-chip (or for them on-valve) liquid cooling, so why should we?

The New Bureaucracy
Summery of article first published in Wireless World December 1982

It is helpful to discuss the reason why professionals in the computer industry seemed happy to freeze the computer's architecture for half a century. The "antimanufacturing-industry" psychosis discussed by Martini Wiener in his book "English culture and the decline of the industrial spirit"8, expanded with a vengeance into hostility to hi-tec. My early peers in the computer industry soon migrated into paper, that is, into programming, away from degrading hardware. They then ganged up with humanities, which is anti-science and has traditionally governed Britain. Lacking technical, or hardware competence, it was helpful for both parties that the digital computer's architecture be frozen, making it a controllable, standard device for bureaucratic manipulation (by technology-free programmers and by technology-free managers and politicians), so this is what happened. The architecture of today's much-vaunted digital computer is identical to the von Neumann computer of 1950. Its architecture makes no concessions whatsoever to the radically different technologies now used in its make-up. (I have made this assertion more than once in the past and not been contradicted, so by now we should all take it as fact.) This fact is concealed by the total lack of appreciation of, or discussion of, this fact. The result is that no computercrat less than 30 years old knows that the architecture of his machine has been frozen for half a century, in defiance of increasingly pressing financial and performance imperatives. Generally, they have floated further and further away from the machine proper on a cloud of ever higher-level software packages. It is as though the horse and carriage had been miniaturised and constructed out of aluminium, or some other material more modern than wood, drawn by specially bred dwarf horses, or poodles, while nobody noticed that something irrational and uncommercial was happening. In such a case, the vested interest obstructing progress would have been skilled (dwarf?) drivers of horse drawn carriages, not programmers and politicians.

GEC's contribution to Britain's Defence

The British government set up the National Research Development Corporation (NRDC) to support new invention and new hi-tec industry. On March 1, 1973, the NRDC wrote that Catt Spiral "could be of fundamental importance in the design, construction and operation of future digital processors and stores" (i.e. computers) - quoted in The Spectator in 197410. This interest by our government, plus government funding, caused me to leave teaching remedial English and return to the electronics profession I had abandoned. I needed to learn the latest technical buzz-words, some of which change every few years.

I was hired by GEC in Borehamwood, and given the task of designing and building the prototype Fast Fourier Transform (FFT) for Nimrod, the British airborne early warning equivalent of AWACS. Both Nimrod and Tornado had four or eight of these massive boxes of electronics to do the job in parallel in order to achieve the required speed. I told my boss Ian Deegan that "Property la", the Catt WSI invention which followed Catt Spiral and was the subject of a government funded project at Brunel University, was ideal for the FFT job, and he agreed. The problem was, how to infiltrate technical innovation into a military system funded by the MoD and designed and developed in Britain.

"Property 1 a", the WSI approach, had massive advantages over the approach we were using with conventional chips, interconnected on boards, interconnected on the back plane in a box measuring two feet by one foot by one foot. "Property 1a" would be self repairing and so more

continued >>>