The Lost Cause
In 1962, when I was about to
leave Ferranti Ltd. in Manchester England, to take my family to the U.S.A., the
company held an in-house conference to discuss the implications of the coming
of integrated circuits.
Previously, after graduating in
Engineering from Cambridge in 1959, I had been working on the logic design of
the first transistorised computer, Sirius. It had 2,000 logic gates and 40,000
bits of memory (= 5,000B), and sold for £25,000. I also attended a course in
programming the earlier, valve computer, called Pegasus.
Ken Johnson (KCJ) pointed out
that if one component was added to a (static R.A.M.) memory bit, which
comprised about ten components, then a column of words in memory could be
searched in parallel. I was dumbstruck. This meant that the whole world of
digital computers changed. The Content Addressable memory (also called
Associative memory) was upon us. We could ask a memory to deliver to us words
with a particular characteristic, without reading them out of memory one word
at a time as we had had to do with previous memory technologies, for instance
the thenn fashionable magnetic core memory. This
would massively speed up the digital computer. Obviously, since the new
technology for memory was the same as the new technology for processing, we
would, later on, be able to instruct all words in memory with a certain
characteristic, to be modified in a prescribed way, in parallel, without even
reading them out from memory (parallel processing). This would further speed up
the digital computer.
I departed for Los Angeles and my
new job, in Ampex, with high expectations. At the
time, I predicted that digital electronics was set to take 10% of G.D.P.
In the event, the world stuck to
Von Neumann, machines with only one processor, and processing within memory was
taboo for the next 40 years. Deviation from Von Neumann, will probably be
banned for another 40 years, until at least the year 2040. The implications for
digital electronics are disastrous, limiting it to much less than 10% of
G.D.P., today make it impossible numeroous
applications impossible, for instance the simulation of global warming, a feat
easy to accomplish with the Kernel Machine (see this website), with its one
million processors (or even better, using a special, larger Kernel machine for
the particular application of global warming) 10 to 100 million processors,
working in parallel. This would be a machine mostly working as S.I.M.D. with
one processor with its own memory dedicated to each square mile of the earth's
surface. (Successful delivery into the marketplace of the "Catt
Spiral" memory machine proved the viability of the approach.)
For a list of applications which
are frustrated by this worldwide commitment to only one processor, see E.
Galea, Supercomputers and the need for speed, New Scientist,
12nov88, p50, or I. Catt, The Kernel Logic Machine, Electronics and
Wireless World, mar89, p154.
Ivor Catt 5jan01