core memory

Dec 05, 2009 49 Replies

Larkin

=20

=20

Even though i am nearly a week behind, i gotta step up to the plate.

The most common form of core memory is called 3D. It is made up of bit planes (the third dimension). The cores have a very square B-H curve. In the bit planes there X and Y direction drive wires. Coincident current through the two wires produces enough field to make the addressed core switch direction of magnetization. The current pulse times are usually no longer than enough to ensure reliable operation. No other cores in the plane receive enough drive current induced flux to change state. Thus we need to add a sense wire to detect the core flux state changes. The sense wires are typically done diagonally to take advantage of the slightly larger aperture in that direction. =20

Since the mechanism to readout the value of the core forces it to either the "0" or "1" state (a design property) any information to remain in the address must be rewritten. Thus another wire called inhibit which is wired in parallel with either the X or the Y drive lines and is driven with an opposing current flow which prevents changing the state of the core on the writeback half cycle. =20

This allows three typical cycles read-writeback, read(erase)-write, and read-alter-write. These are the external interface definitions and read-alter-write may not be implemented.

I hope that have not omitted too much.

Larkin

machine=20

shipped=20

google=20

magnetisation=20

half=20

core=20

means=20

=20

you=20

restore=20

Pretty good page. Would you like to discuss drive electronics? Nothing bigger than a TO-5 needed. 74hc138 decoders are fair game. =46lat-pack packages ere fair game.

easy=20

next=20

a=20

=20

=20

Those were done by machine.

saved

Gotta love that mumetal.

TI made, for a while, integrated core stack drivers and sense amps. Ironic that as ICs got good enough to drive and sense cores, ICs got good enough to kill cores.

Anybody remember 1K drams? I did a color graphic video generator, for oil/gas/product pipeline controls, using the strange 1K AMS parts. TI made a driver for them, too. They were differential read/write (5 volt write, millivolts read) and the memory cell was basically two capacitors and two connect fets.

John

I _almost_ used some DRAMs once, except by the time I got the Z80 board almost done, they were up to 64K. The Z80 had a "refresh" output, so using them would have been within the scope of my abilities at the time. ;-)

Ufortunately, the project got interrupted and placed on the back burner, where it's been ever since. )-;

Cheers! Rich

On a sunny day (Fri, 11 Dec 2009 15:02:20 -0800) it happened Rich Grise wrote in :

I did a 265 kB RAMDISK with DRAM for the Z80. ftp://panteltje.com/pub/z80/ It was I/O mapped :-) The refresh was done by having it cycle through 512 address rows when not addressed. When addressed 512 bytes were read or written at any time, 'sector size'. It had write protest logic too, before writing a sector you had to do I/O to the unlock address. The procedure was: write to unlock address, write sector address low write sector address high write 512 bytes data. The idea was to read a complete single sided floppy (204800 Bytes = 40 tracks of

10 sectors of 512 bytes) into this RAMDISK, and then work from the RAMDISK. It was faster then anything at that time when doing that. Ran a C compiler on it, almost faster then the PC these days. 4 MHz clock.

Are you referring to the 1103 ?

Quite awful chip.

The next generation 4 K dynamic or 1 K static were much easier to use.

Perhaps this is just a naming convention, but isn't read-alter-write the same as read-modify-write naming convention used by some other manufacturers ?

After all, in current DRAMs, when the RAS signal is asserted, the data is read from all columns to a common area to be written back to the memory cells in the same row. When the CAS drops, only the interesting data is selected in the output multiplexer.

No, I used an AMS part, 6002 I think. It had essentially two capacitors and two fets per cell, possibly structured as a

4-transistor dynamic latch with two access transistors. Either way, one wrote to it by applying a differential 5-volt level to a pair of pins, charging the addressed cell. Read connected to the same cell, and you got back a small differential signal. Actually, the chip didn't care whether you were reading or writing... it just connected you to the pair of caps. Read was destructive, like core, so after a read you had to recharge the caps then disconnect. It was noisy, like core.

I've heard similarly bad stories about the 1103.

Interesting: one of the founders of AMS was J. Larkin:

formatting link

John

It is really just a difference in names.

In the oldest versions. Then they changed things and you could get four=20 successive address more quickly. This was called EDO DRAM about 20 years= =20 ago. Since then things have gone further in that direction. Gotta keep=20 the caches fed so that they can feed the core.

Join the Discussion

Have something to add? Share your thoughts — no account required.

Didn't find your answer?

Ask the community — no account required