CDC core memory One of the first computers that I programmed was the CDC 7600 at the CERN laboratory in Geneva. This machine was considered a supercomputer and had two front-ends, a CDC 6400 and a CDC 6600. When you wrote the control language for your program, on punched cards of course, you had to know whether the code was small enough to fit into SCM or "small core memory", or whether parts of it would overflow into LCM, "large core memory". At the time, this restriction was just an incomprehensible inconvenience to a physicist like myself. Nowadays I find that magnetic core memory is one of the most fascinating aspects of computers from the 1960's and 1970's.

The basic concept of the technology is fairly simple: a small ring of magnetic material can be magnetised in two "directions", clockwise and anticlockwise looking sideways at the ring such that appears as a circle. If the two magnetic states represent 0 and 1 in the binary system, the ring can store a binary digit. All that is needed is a way to write and read the magnetic field. Many rings can be combined to make kilowords of memory. As a bonus, each ring remains magnetised without external power, unlike modern semiconductor memories.

IBM core memory This short explanation hides a complex technology. To make the memory fast, and keep power consumption low, the rings had to be small. However a minimum of three wires, and in some systems up to five, had to be threaded through the hole in the middle. Wiring up thousands of rings into a memory 'plane' was a difficult job, largely done by hand. Drive electronics costs had to be kept to a minimum, so an X-Y system was used which relied on two co-incident drive pulses flipping the ring to the desired state, but the neighbouring rings which received only a single pulse were not flipped. This required careful calibration and testing, with possible bad spots in the memory. The resulting memory was expensive and difficult to maintain, and the smaller the rings, the greater the cost. That is why the CDC machines had a limited amount of fast and expensive SCM, but a much larger amount of the slower and cheaper LCM.

I have a small number of core memory planes as shown on the right. The upper image shows a plane that I was told came from a CDC 6400 machine. However, almost identical planes were used in several CDC models, so I cannot be certain. The lower image shows a plane from an IBM 360 model 30. This has a totally different wiring system from the CDC memory. These are both pure memory planes with no drive electronics mounted on them.






Valid XHTML 1.0 Strict