Some Core Memory
  

A Type 184A core memory module

Memory for a PDP-8 computer

Part of the Core Memory pages
by Douglas W. Jones
THE UNIVERSITY OF IOWA Department of Computer Science

Contents

Overview

thumbnail image
thumbnail image
The 4K core memory subsystem of a PDP-8 occupies two rows of DEC standard backplane segments, filling most of 6U of 19-inch rack space. That is, a 10.5 inch (268 mm) high area between rack rails with outside edges spaced 19 inches (483 mm) apart, leaving 17 3/4 inches (450 mm) of usable space between the rails). DEC's standard backplane segments were 3U high, holding 2 rows of flip-chip circuit boards.

The core stack itself is enclosed in an aluminum box 5.7 inches (145mm) high, 5.5 inches (140 mm) deep and 7 inches (178 mm) wide. The stack is mounted to the bottom of the box on 4 6-32 UNC threaded studs arranged in a square on 3.4 inch (86.4 mm) centers. The core stack sits on 4 nylon spacers 3/4 inches (19 mm) high by 3/8 inches (9.5 mm) in diameter.

The 12 core planes are mounted in 3 7/8 inch (98.4 mm) square frames with electrical contacts protruding an additional 1/8 inch (3.2 mm). There are 64 X or Y select wires terminating on each edge of each plane, plus 2 sense wires and 2 inhibit wires terminating on two of the corners. Select wires on each edge of the plane alternate so that half have their contact fingers at the top and half at the bottom, and after the core planes were stacked, the contact fingers of adjacent planes were soldered so that each X or Y select lines is in series with the correspoining line in all of the planes.

Looking at one side of the core stack, the array of contact fingers fills a square about 2 1/2 inches (64 mm) on a side, suggesting that the select wires within each core plane are spaced about 0.04 inches (1 mm) apart. In the photo of the exposed side of the core plane, the twisted pairs wired to the left side and exiting at the center right are sense wires, while the twisted pairs on the right exiting to the bottom right are the inhibit wires. The bundled wires attached at the top and bottom that exit to the rear are select wires.

thumbnail image
thumbnail image
With the side of the core box open, it was possible to completely figure out the mounting of the core and determine that it would be safe to remove it from the system. The inhibit wires exit the core at the bottom below the sense wires, terminating on a single W025 connector paddle, while the X and Y addressing wires exit on the opposite end, terminating on two W025 connector paddles.

The wires to the three W025 paddles are just long enough to allow these to be unplugged. With them unplugged, removing 6 screws from the opposite side of the backplane allows the core box to be lifted free.

With the core box perched on the edge of a workbench, removing 3 screws from the top of one side and 3 more from the bottom of the other side allows the cover over the core stack to be removed. All of the interior wiring is very neatly done, with insulating sleeves over each exposed solder joint, and with wires neatly bundled with lacing tape.

Interconnections between one pin and another on the small 8-slot backplane for the G603 boards is all done with neat point-to-point wiring. This is an archaic feature! Within a few short years, it would be common to use a printed circuit board for this kind of interconnection.

thumbnail image
thumbnail image thumbnail image

Removing the 4 nuts on the top of the core stack allows the thin perfboard cover to be removed, exposing the actual core. With the core enclosure properly sealed, the perfboard serves little purpose, but during assembly, it no-doubt offered significant protection from acidental damage to the core stack.

With the protective cover removed, it was easy to photograph the top surface of the core stack. As was common with early core memories, the core plane hangs in space, suspended by the X and Y wires. There are a total of 4 wires through each core, X, Y, sense (diagonal) and inhibit (parallel to Y).

The close up photos here show a machinists ruler resting very gently on the core plane. The graduations on the ruler are spaced 0.01 inches (0.254 mm). Measuring from the photos confirms the 0.04 inch (1 mm) spacing between the X and Y wires, an reveals that the cores are 0.035 inches (0.89 mm) in diameter, 0.006 inches (0.15 mm) thick, with a 0.025 inch (0.64 mm) hole. The wires themselves appear to be 0.005 inches (0.127 mm) in diameter, making it 36 AWG.

thumbnail image
Where later core memories used simple diode matrices to route the power to the appropriate X and Y select lines, this generation of DEC equipment used an ingenious scheme. Each of the 64 X and 64 Y select lines is driven by a diode-balun matrix. A balun is a small transformer used to balance the current through the X or Y addressing line when it is driven by an unbalanced source. DEC used toroidal ferrite-core transformers potted in red epoxy, the most prominent components in the photo here. The 64 diode-balun network on each axis are logically arranged in an 8 by 8 matrix, with 3 bits of the address decoded for the row and 3 bits decoded for the column. The 8 G603 boards directly attached to the core box each hold 16 of these diode-balun networks.

thumbnail image thumbnail image
Each of the 12 sense lines is connected directly (by a pair of plug-in connector pins) to a G007 sense amplifier board. This board has 13 transistors, 9 in individual packages plus 2 pairs in 6-pin packages. Each pair, plus one simple transistor forms a stage in a differential amplifier, and while the remaining transistors convert the output to DEC standard logic levels (0V and -3V) and gate the output with a strobe input.

Documentation

According to the PDP-8 Price List (F-87, Digital Equipment Corporation, Aug. 1, 1967), DEC sold the Type 184 4K memory module, without parity, for $7500.

Chapter 4 of the PDP-8 Maintenance Manual (F-87 2/66 Digital Equipment Corporation, 1966) gives the overall architecture of the core memory, while detailed schematics are given on later pages:

The Introduction to the Maintenance Manual states that the core memory has a 1.5 microsecond cycle time. It also states that an add instruction takes 6 microseconds. Since an add requires fetching an instruction and loading an operand from memory, and since reading core is destructive, the manual must be defining cycle time as the time taken to read and then refresh the core, that is, a read cycle followed by a write cycle.

Provenance

This core stack comes from a PDP-8 serial number 85, ordered by the University of Iowa psychology department in 1965 and delivered in December or January of 1966.

A paper label on the backplane framework to which the core box is mounted says DIGITAL EQUIPMENT CORPORATION CORE MEMORY 08-291-1000 Type 184A. The DEC sales brochure, PDP-8, A High Speed Digital Computer (F-81, 3/65, Copyright 1965, Digital Equipment Corporation) lists the Type 184A and 184B memory modules on page 16. The difference is that the 184A module does not include a parity bit on each word. The 184A we have is integrated into the structure of the computer, but this module was also available as a separate assembly which, if used with the Type 183 memory extension control, could be added to a PDP-8 in order to expand the machine in 4K increments up to 32K.

The embossed aluminum label on top of the core box says FERROXCUBE MEMORY EQUIPMENT, SERIAL: 51-238-71. Ferroxcube is, to this day, a well known manufacturer of ferrite cores for the electronics industry, and in the days of core memory, it supplied core memory on an OEM basis to a number of computer manufacturers.

thumbnail image
Each of the 12 sense lines is connected directly (by a pair of plug-in connector In disassembling the core memory, we found the numbers B51-238-19 stamped on the plate holding the connectors for the G603 memory selection matrix boards, and C51-238-16 stamped on the base plate under the center of the core stack, visible in highly distorted form in the side view of the core stack. These images are straightened out here.

There were probably more stamps on other parts. It seems fair to speculate that the prefix B and C indicates the part within the assembly, while the 51-238 is a batch number, and the final suffix is a serial number within the batch. If this is the case, the core assembly workers at Feroxcube made no attempt to match the final digits, they just picked up parts at random when doing the final assembly of core boxes. In this case, the serial number embossed on the aluminum sticker either followed the actual core stack and not the enclosure, or it was also randomly affixed at the end of the assembly process.

A smaller paper label on the top of the core box says ACCEPTED DATE 10/25/5; presumably, this indicates that DEC received this core memory ran acceptance tests on it on Oct. 25, 1965.

A paper label on the backplane framework to which the core box is mounted says DIGITAL EQUIPMENT CORPORATION CORE MEMORY 08-291-1000 Type 184A. The DEC sales brochure, PDP-8, A High Speed Digital Computer (F-81, 3/65, Copyright 1965, Digital Equipment Corporation) lists the Type 184A and 184B memory modules on page 16. The difference is that the 184A module does not include a parity bit on each word. The 184A we have is integrated into the structure of the computer, but this module was also available as a separate assembly which, if used with the Type 183 memory extension control, could be added to a PDP-8 in order to expand the machine in 4K increments up to 32K.

Scratched into the fiberglass shims around the top edges of the core stack, although very difficult to read, are the following notices:

These are apparently instructions to the person wiring the X and Y addressing lines from the core stack to the G603 card-edge connectors. The sequence of colors in the cables match these instructions, if you read each side of the core stack from left to right while facing that side.

Condition

The (now retired) technician for the psychology department who maintained this machine indicated that it worked when last used, which was some time before 1980, perhaps as late as 1978. There is no reason to believe that it cannot be put back in service, but as of this writing, something is wrong. A number of bad diodes have been replaced, but not all of the boards have been checked. The best explanation of the bad diodes is that they failed due to thermal cycling or corrosion while the machine was stored in various attics and closets after it was retired.