Magnetic core memory reborn (2011) | Hacker News Reader

Magnetic core memory reborn (2011)

corememoryshield.com

104 pointsmcbits17 comments

Open articleView on HN

Besides being a really cool project, the article does a great job of explaining how core memory works. Nicely done!

Here's a core memory plane I bought a few years ago. This is a 4096 bit plane from an early-'60s Univac 418:

https://geary.smugmug.com/Computers/History/i-jb2rgF5/A

Full size close-up:

https://geary.smugmug.com/Computers/History/i-jb2rgF5/O

People may also be interested in analog counter part: Magnetic Amplifiers.

"I had always believed that first came the vacuum tube, then the transistor, period. But thanks to an old Navy tech manual sent in by a reader, I've discovered a third "lost" entity. This document, unusually passionate and well written for a military tech manual, is their promotional brochure..."

http://teslapress.com/magamp.html

https://archive.org/details/MagneticAmplifiers

https://en.wikipedia.org/wiki/Magnetic_amplifier

There is a fourth one - the electromagnetical amplifier - essentially a transducer mechanically coupled to a carbon microphone. While not relevant for logic circuits, it would come at the start of that chain.

I always thought it would be cool to make a necklace or something with ferrite cores after they'd been "written" to, so you could keep a secret message with you (albeit a short one) without suspicion.

I'm sure the TSA would be suspicious.

They'd most likely make you take it off and put it through the tunnel just because it's metal. But wouldn't that destroy the data thanks to the magnets?

A piece of equipment I worked on in the USAF had 512 bytes of core memory. If I recall correctly, it was constructed of nine planes of 512 cores each, where you had 8 data planes (bits) and one parity plane. It dated from the Vietnam war as a joint Army-Air Force project, and was being phased out in the mid-80's with a new microprocessor-based system.

I once had to replace a defective memory assembly -- the end of messages in the system were indicated with 4 N's in the first few columns ("NNNN") and one of the cores went bad so the N in that column was never recognized (the other characters in that column were munged too), and so the message never ended. Over a day with an oscilloscope to find that one bad core and convince the NCOIC that I was right... I didn't have a microscope, but to the naked eye the core looked fine. It just wasn't flipping like it should have.

Was that the 490L Overseas Autovon? I went to school for that phone switch in 1987 and was in the last class before they decommissioned it for the DOD.

It had a core memory when it was initially deployed.

This was DSTE [0] connected to AUTODIN [1]. AUTOVON [2] was still around -- I only heard of Flash Override being used once, and the recipient of the call was sent back stateside shortly thereafter... We had a Siemens phone switch at that base, probably because AT&Ts 5ESS wasn't available at contract-signing time.

At a later base I was at, they were installing the then-new X.400 system [3]. It didn't last long, as SMTP ate it's lunch. Which is funny, as I got to watch an IMP [4] get installed by the BBN [5] contractor that same year.

[0] http://www.1882nd.com/images/DSTE.jpg

[1] https://en.wikipedia.org/wiki/Automatic_Digital_Network

[2] https://en.wikipedia.org/wiki/Autovon

[3] https://en.wikipedia.org/wiki/Defense_Message_System

[4] https://en.wikipedia.org/wiki/Interface_Message_Processor

[5] https://en.wikipedia.org/wiki/BBN_Technologies

I've always wanted to do this! Maybe I'll make the rope memory version.

https://en.wikipedia.org/wiki/Core_rope_memory

That's great how there is a measurement of 2.5MB per cubic meter. I've love to see a 1 x 1 x 1 meter cube of rope memory... Imagine the person hours required to do something like that, you'd have to hire professional persian carpet weavers or something.

> Imagine the person hours required to do something like that, you'd have to hire professional persian carpet weavers or something.

Apollo Guidance Computer (the computer of the Apollo Command Module) has 36K 16-bit words of core rope memory. That is about 0.5MB. There were at least ten such computers created. So, this much core rope memory was woven for the Apollo Command Module's computers alone. See a previous discussion:

https://news.ycombinator.com/item?id=7656282

The original version of that was the Dimond ring translator for #5 crossbar.[1] The purpose of this device was to translate a line position in the switching system to the phone number for billing purposes. It was a big ROM which could be manually changed by moving wires. This allowed some phone number portability - if you moved within the same central office area, you could keep the same phone number.

It shows how hard it was to make a memory device in the 1950s. For a long time, that was the bottleneck in computing. Electronic arithmetic units date from around 1940, but memory devices that didn't suck only came in around 1965, and didn't get cheap until around 1990.

[1] http://etler.com/docs/Crossbar/articles/30-AMATranslator.pdf

That is awesome! I did a 1 bit core memory to demonstrate the principle but this is a much better example.

Some microcontrollers have Ferroelectric RAM now-- a nonvolatile memory that's basically magnetic core memory on a chip!

https://en.wikipedia.org/wiki/Ferroelectric_RAM

While it confusingly uses the "ferro" prefix, the ferroelectric effect isn't related to magnetism.

https://en.wikipedia.org/wiki/Ferroelectricity

Previously discussed at https://news.ycombinator.com/item?id=3888926.

Reply on news.ycombinator.com