▲ 81 points
back
19 comments
This is pretty cool by itself of course. But I'm wondering, is there any reason at all to use core memory today? I know EMP/disaster resistance has been said in the past, but a small modern durable chip encased in a bunch of protection should work better right?
The only competitor core has for eternally non-volatile low tech near radiation proof infinitely rewritable electrical data storage is latching relays and each bit of latching relay is huge in comparison to core. Also latching relays unlike core has moving parts that will eventually wear out. A million cycles sounds like a long time for a latching relay to live, but as a register or memory for a computer at 0.1 Hz thats only two months or so. So if you want to use latching relays you're limited to applications where the system will never perform more than a couple hundred thousand bit flips.
The marketplace for eternally non-volatile low tech near radiation proof no moving parts infinitely rewritable data storage is roughly nil, unfortunately. Maybe a space probe to the worst parts of Jupiter or as part of a nuclear missile warhead.
"In the old days" they had to make the ferrite cores, then characterize them, then wrap them into memory units by hand, but I would imagine given enough money one could find a way to 3-d print cores around taut stainless steel wires then fire the core ceramic, automating the process. Given a very large budget of course.
Its probably cheaper in practice to ship very large amounts of flash wrapped in lead shielding.
>The only competitor core has for eternally non-volatile low tech near radiation proof infinitely rewritable electrical data storage is latching relays
except for tape
>The marketplace for eternally non-volatile low tech near radiation proof no moving parts infinitely rewritable data storage is roughly nil
incidentally google is the biggest buyer of tape in the world
>I would imagine given enough money one could find a way to 3-d print cores around taut stainless steel wires then fire the core ceramic, automating the process.
you are describing tape and it's predecessor wire recording. Rather than coat a wire in magnetic material they just used steel wire. They only use ceramic to make core because it's easier to fire it in an annulus than to punch out laminations that small
You bring up a fair point but tapes not really infinitely rewritable. You can take something like a 60's PDP-8 and write a byte of data to the same core address 333 thousand times per second for at least half a century and it'll be fine. In my head that's a 14 or 15 digit number of writes which no tape or even flash memory can withstand.
In theory much like a hard drive head floats on an air bearing over the platter it should be possible given enough development time and money to make a tape head that floats and never quite touches tape. I worked at a place in the 90s that had IBM floating tape transport where compressed air maintained some loops bouncing in channels to make sure the tape was never snapped taut and therefore stretched but I'm taking about actually floating a non-contract head which AFAIK has never commercially shipped for tape (although it would be cool to be proven wrong).
I was going to bring up how tape can throw oxide after a couple decades in storage but I suppose core could have dissimilar metal corrosion over the course of a century or maybe temperature cycles would stress and crack individual cores after a couple centuries unless kept at a constant temp. Also in theory non-archival core could exist just like non-archival tape definitely exists. So fair enough on that account.
Radiation.
There was a period in time right between core memory and solid state where something was developed (I believe by IBM) which was core memory, but layered "thin film" on either a glass or maybe polyester film substrate.
It essentially gave all the advantages of core memory, but could be printed by machine to eliminate the need for hand assembly that standard core memory required. Fabrication was done using something akin to how PCBs are made, coupled with something akin to offset printing (or something like that).
Information about this memory is very sparse; I've encountered a few references about it here and there on the internet, but mainly I've read about it in old computer books contemporary with the time period (mid-to-late 1960s to early 1970s). I also believe you can find an article or two in old issues of Popular Science magazine for the same period (google books has them all online).
Ultimately, though, solid-state won out for many good reasons; it was a case of a technology which - had it came about 5 years or more before, would have taken over the market, but its window was too small for acceptance (and probably there were bugs to work out).
As an aside: If you like this kind of thing, it is very interesting to read about the history of these kinds of shifts in technology. To read about this history of the transition to tube technology (think about how you would represent and generate a carrier wave for radio communication before tubes existed - it was done), then to transistors, then to integrated circuits, then to surface mount from thru-hole.
Then also the transition from hand-wired "rat nest" construction to printed circuits.
But the real interesting parts on all of these transitions happens between the transitions (and the overlaps) - where there exist both side-by-side, plus interesting technologies developed in the transition, anticipating things, dead ends, and in some cases ideas which won't become common until much later when other technology could "catch up" (mainly for assembly or whatnot).
Early integrated circuits, for instance, were actually tiny soldered devices in a "can" or on a ceramic substrate - of very few parts. But they were anticipated by the small modules done prior by companies like IBM for assembly and maintenance of computers (and even before that, with multi-component tube assemblies of an "entire radio in a tube" type devices). There were also such things as surface mount technology being used when nothing else would do: For instance in the Apollo Guidance Computers, early surface-mount technology was used to make it more robust to vibration, and also lighter weight. Also some early integrated circuit tech was used in them as well. All of these things wouldn't become common at the business or consumer level until years later - but the tech was needed ahead of time of course for the purpose; also an example of the space program of NASA pushing the tech envelope that trickled down to consumers.
Check it out if you haven't already and have an interest; I've found it pretty fascinating, as you can probably tell...
Bubble Memory!
It was started by Bell Labs although toward the end IBM was one of the last groups pushing it.
In a way bubble memory is so simple. A strong magnetic field can "push around" magnetic domains on a piece of tape or whatever without flipping the bits if you get it just right. The two fundamental tricks that make it an engineering project that was successful for awhile are that its possible to build a thin film and some other components so stable and linear that you can push thousands, even millions of magnetic domains around without screwing them up, and superficially you would feel that electromagnetic stuff is all speed of light but moving magnetic domains can happen at just the right speed that you can read out a bubble memory using 1970s TTL logic, not too fast and not too slow. It was quite temperature sensitive like all magnetic things.
It died as a technology because it only scaled over a factor of a thousand or so in an exciting computational era of changes much faster than now, where an order of magnitude only lasted a year or so at most.
It couldn't have been done five years earlier, there was a very narrow window for it between TTL type logic finally being fast enough to read the bits as they flew by vs just a few years later simple static ram and eprom would store the same amount faster and simpler.
It was kind of like a no-moving parts delay line that relied the speed of magnetic domain flipping being predictable repeatable and a useful speed.
Bitsavers has some fun reads on ancient hardware technology. You can read IBM manuals as the technology was advancing, pretty interesting stuff. DEC also.
There are plenty of rad hard memory solutions out there that have modern interfaces and take up no space relative to the size of core memory. Core memory is a thing of the past and shouldn't ever really be part of anyone's modern designs. You can get rad hard RAM, EEPROM, and NV Flash today for space and military applications, for a price [1]. If you want disaster-proof offline backup, a below ground water-proof storage locker bolted to the foundation with a faraday cage inside is more than enough. Dig deeper if you want it to survive a nuke.
I really like these kinds of practical demonstrations of how technology worked. Its such a shame that an up-to-date version could need scanning microscopes to see visually
I know that this is entirely off-topic but I wish Google made their services a lot more visible than they are. For example I completely forgot about sites.google.com until I saw this link and the service has been there since ~2009.
I've never seen "live" core memory. Anyone know how it stands up to physical shock? For example, if someone accidentally bumped the rack, could it cause some of the bits to flip?
I used core memory in machines during the 1970s (primarily Data General Novas). There was no problem with physical shock, but we did get errors during strong thunderstorms.
Back then, when I was in highschool, I was told that the density of core memory was limited by the dexterity of the Filipina workers who threaded the wires through the tiny magnetic donuts.
For some engineering scaling reasons small things tend to be pretty tough in proportion to their size; supposedly the core arrays in the space shuttle Challenger survived ocean impact and were readable once recovered.
core was relatively sensitive to temperature and the currents had to be varied to match the temp for optimum performance. Which can be worked around as OP did by not optimizing for minimum power or minimum error rate. Also the more cores the sense wire ran thru the more temperature sensitive the core became.
Actually, some of the later versions of core memory were very difficult to see with a "naked eye"; almost grains-of-sand kind of thing.
I own a couple of examples of core memory; one is a very old core assembly (I swear the frame is made of bakelite). It's highly damaged, missing several frames, and some of the existing "outside" frame have damage (but the inner frames are "ok"). Those cores are readily visible, and each measure a few millimeters in diameter (the wires of the grids are very thin magnet wire). There are around 10 or so layers in this core assembly (and I'm not sure how many were originally there).
My other core assembly is a 16K board for some kind of a minicomputer or mainframe, made in the early 1970s. It even has a few surface mount components on it, plenty of thru-hole components, a few ICs (mainly line drivers), and the core memory module (which is actually protected by a metal cover - I've never opened it, so I don't know what it looks like underneath). My wife and I have it hanging in our dining area, next to our periodic table, stirling engine, and Tesla coil (yep). The whole board measures about 16 x 16 inches, and includes some metal frame stiffeners soldered on as well. The actual core assembly measures about 6 x 6 inches.
In either of these, physical shock wouldn't do much (other things would break long before the core assemblies). That said, in actual construction they were typically protected in some manner; either in a case of some sort (in the case of the larger unit I have) or behind a metal plate (as in the board I have). This was mainly done to prevent fingers from poking where they shouldn't go (as well as prevent issues with tool slippage or dropping).
The value is stored magnetically, so physical shock is not a problem, but other environmental conditions might be a problem.
I used to be an electronic technician (305x4) in the Air Force and spent 3 years on one of the old 490L Overseas AutoVon phone switches right before the DOD phased them out in 1987. This telephone system originally had a ferrite core memory and it was NOT non-volatile. Every few milliseconds the system had to run through each bit reading them and then re-setting them, as the magnetic field collapsed over time. The system used 50 volts.
Obligatory quote from "Real Programmers Don't Use Pascal":
Your typical Real Programmer knew the entire bootstrap loader by memory in hex, and toggled it in whenever it got destroyed by his program. (Back then, memory was memory -- it didn't go away when the power went off. Today, memory either forgets things when you don't want it to, or remembers things long after they're better forgotten.)
Brilliant work!