IBM Creates Working "Racetrack Memory"

holy_calamity writes "IBM has created the first working 'racetrack memory' device — a technology we've discussed as it's been touted as the future of memory. It works by writing bits using the magnetic domains inside a very thin wire. Those domain can be shunted along this 'racetrack' and past read heads."

Sounds like...

[Gordonjcp]

... bubble memory. Welcome to 1968.

Re:

[thered2001]

Or a wire recorder...welcome to 1938.

Re:

[synesis]

Or a mercury delay line storage.

Re:Sounds like...

[hey!]

Ah, you beat me to it. The mercury delay lines were readily available because they had been developed for radar systems in WW2.

CRT based memory was also, in a sense, a product of radar. If you've seen early radar depictions from old movies, you had this kind of linear cursor started at the center of a round CRT tube and went to the edge. The end swept around the perimeter of the display, and when a line crossed a "blip", it would be refreshed. Over the next couple of seconds the blip would fade and the sweeping line would refresh the blip in a slightly different place. The persistence of phosphors on the screen were a kind of short term memory, so it's not surprising that engineers familiar with radar hit on the idea of making CRT storage units.

Random access is not the only memory model ever used in computers, nor is it the only one that will ever be used in the future. This is one of the reasons CS students are taught to regard polynomial time differences between classes of algorithms as relatively unimportant in a theoretical sense, although they are obviously important in a practical sense.

Re:

[DrSkwid]

Packard Bell did machines with mercury delays & CRT storage in the 1960s

http://research.swtch.com/2008/04/computing-history-at-bell-labs.html [swtch.com]

Re:

[Gordonjcp]

Ah, no. Mercury delay line storage actually used acoustic waves in a tube of mercury to store pulses, which were regenerated by amplifiers and sent back into the tube by the actuator. This thing with the loop of wire and the magnetic bits moving while the wire stays still sounds much more like the idea of magnetic bubbles in bubble memory.

Re:

[Angst Badger]

I was wondering if anyone else noticed that. The more things change...

Re:

[Windows Breaker G4]

Kinda like the diesel engine running on vegetable oil and the electric car coming back?

Re:

[meatmanek]

Actually, sounds more like Twistor [wikipedia.org] memory.

1964: Outer Limits - Demon with a Glass Hand

[jerryasher]

Demon with a Glass Hand [wikipedia.org] by Harlan Ellison tells the story of a man who can only remember the past ten days struggling to determine who he is, why he has a glass hand with missing fingers, and why an alien race is trying to kill him.

In the denouement a very important copper wire with very high density storage is central.

44 year old spoilers at the link.

Re:

[ChrisLeif]

Mercury delay line. Welcome to the 1950's. http://en.wikipedia.org/wiki/Delay_line_memory [wikipedia.org]

Re:

[BountyX]

I swear slashdot has the funniest sigs...

Re:

[PPH]

Can disco be far behind?

wire spool recordings

[Tungbo]

perfect for passing information inside POW camps.

"I see Nuthin... I know Nuthin...."

Bubble memory

[threaded]

I can't be the first one to read this and think, eh, isn't this just bubble memory?

Re:

[Raineer]

So fill me in on the details, did this work back then? If it's already in existence and supposedly better, why is no one making it? Or is it just not better? Honest questions, I have no idea and don't always trust wiki 'opinions' :) Thanks

Re:

[account_deleted]

Comment removed based on user account deletion

Re:

[Raineer]

Touche' Hit me with some logic and see how it goes :) lol

Re:

[SleptThroughClass]

Hit me with some logic

DL, RTL, DTL, or TTL?

Re:

[flyingfsck]

Original HP pocket calculators used bubble memory. Yes, I am that old...

Re:

[leighklotz]

Original HP pocket calculators used bubble memory.

Which model? MoHPC [hpmuseum.org] says there was a non-production prototype series called Roadrunner that was to use bubble memory but makes no mention of it in any production device.

Re:

[IAN]

Original HP pocket calculators used bubble memory.

Um, no. The HP 9100A used core memory (both parts of ROM and all of RAM), but that was a honking big desktop machine. All pocket calculators used PMOS RAM until the appearance of "continuous memory" (CMOS) models.

Re:

[kevmatic]

Well, it worked, alright. But it took forever for the information to get around the loop, leading to large seek times, and they couldn't push it over a mbit a chip. I'm thinking they used actual wire while IBM is probably lithographically defining it like a CPU transistor. And yes, they used it. If you look closely at any CNC machine shop that's been around for a while, you'll probably find one or two machines from the era with bubble memory, still whirring away. CNC machines, for many many years, had to k

Imagine an infinite-length wire "track"

[BadAnalogyGuy]

If you had an infinite-length track, you could theoretically encode data which could itself be interpreted as processor instructions. Then, given these instructions, you could move back and forth within this track and read data and further instructions. With a fairly minimum number of instructions, it would be possible to synthesize more complex instruction batches.

This sounds like such a great idea. I wish I had it already!

Re:

[Deltaspectre]

If I had an infinite length track, I'd sell it for infinite money.

Turing Machine!

[dazedNconfuzed]

Turing Machine! [wikipedia.org]

Re:Turing Machine!

[somersault]

Lego Turing Machine [umontreal.ca]

Lego Difference Engine [woz.org] :D

Re:

[Oktober Sunset]

Revoke my geek card if you must, but on being chalenged to make a lego computer, I'd just use an RCX. ;) Yea, I'm lazy and have no sense of wonder.

Re:

[hey!]

Well, as a practical matter you probably wouldn't make linear memory with a single read head that was billions of bits long. Nor would you be likely to treat it as such in your programs, although you might have clever adjustments to your algorithms that take its overall performance characteristics into account, the way that people take the performance characteristics of hard disks as a kind of unspoken assumption.

For that matter, modern random access memory is really more of an abstraction than a reality.

Re:

[Jurily]

I'd say the single thing most likely to really change over the next twenty years is this neat two way division of memory, especially as mobile and embedded devices become more common.

Oh, please no. I already lost an mp3 player because of that. Memory got corrupted, the whole thing froze, and it's persistent across battery changes. Even the firmware updates were software only.

Of course, I previously voided the warranty as I did a firmware update (the updated version had no arbitrary volume limit imposed on it, and also it enabled the built-in radio).

From that day on I only buy things I can reboot.

Re:

[Oktober Sunset]

I find it extra amusing that Turing machine is offtopic, but Lego Turing machine is informative.

Re:

[LarsG]

It is quite obvious what the gp was hinting at, so answering "Turing Machine!" is firmly in the whoosh, missed the point, Mr. Obvious, no really?, duh, fail category.

"Lego Turing Machine" is fun/informative.

Re:

[LarsG]

Explaining an obvious joke should really be modded redundant.

You mean DNA?

[EgoWumpus]

I assume that is what you were getting at; after all that is pretty much how DNA works, only without needing an infinite track.

Re:

[thewiz]

...you could move back and forth within this track and read data and further instructions.

If you play instructions to a CPU backwards, would the computer spout Satanic messages?

Re:

[Shinmizu]

I don't believe so. I've heard it decompiles Windows Vista into COBOL for your reading pleasure.

Re:

[thewiz]

decompiles Windows Vista into COBOL

Definitely Satanic on both ends.

Re:Imagine an infinite-length wire "track"

[itlurksbeneath]

Hmm... To use a car analogy (this is Slashdot, right?): if you had a car on an infinite-length track, would it be the ultimate Touring Machine?

Cue groans...

Re:

[A nonymous Coward]

To ring the patent office or not to ring, that is the question.

The old is new again

[hcdejong]

So, this is basically a solid-state version of mercury delay lines?

Re:

[Nefarious Wheel]

Or plated wire memory.

FTFA

[Oxy the moron]

The first ever racetrack memory device is able to store and read three bits of data using the racetrack method.

Bit 1 - Did something?
Bit 2 - ??????
Bit 3 - Profited?

this won't take off soon

[OrochimaruVoldemort]

it will be a few years before this is practical, and even then, people won't buy it. new motherboards will have to be introduced, ultimately resulting in have to get a new pc (if you are a business). so slashdotters, i would hold off for, lets say, 7 years before getting this.

Re:

[CogDissident]

Why would it need new motherboards? Its just the internal method of data storage and retrieval. Its entirely possible that early versions of this drive could use SATA connections and just be cheaper and faster than current drives (rather than extraordinarily faster, if it needs a different kind of motherboard connection).

Re:this won't take off soon

[pipatron]

Oh! New motherboards would have to be introduced! That could take some time to switch to indeed, because it's quite rare that such a thing happens.

Except for the switch from DRAM to SDRAM. And the switch from SDRAM to DDR, and from DDR to DDR2, and from DDR2 to DDR3, and from AGP to PCI-e, and from IDE to SATA, and.. and.. ad infinitum.

Re:

[Anonymous Coward]

Your sarcasm is well targeted. I think techies forget how fast things evolve, because they fail to appreciate incremental increases.

Ten years ago we weren't using PCI Express, and AGP was extremely new.. we were mostly using PCI graphics cards! 3D support wasn't even common at the time. We were surfing the information superhighway at 33.6kbps, 20GB was considered a lot, and black and white laptops were still reasonably common!

Re:

[foniksonik]

Cause a new PC would be so expensive?

You should have stopped here:"It will be a few years before this is practical"

Re:

[LarsG]

Cause a new PC would be so expensive?

Yeah, I'm sure buying a new motherboard will be reeeeeal expensive compared to the cost of the first sticks of Racetrack RAM you buy.

Anyway, if people bothered to RTFA (hey, one can wish) they would see that they expect racetrack memory to first be used for storage - which means you will connect the stuff through sata/pci/pci-e or other existing buses. Using Racetrack as a replacement for RAM is in the 7year+ range.

Re:

[Waffle Iron]

I'm hoping it's a Formula 1 Circuit

Nope, more like NASCAR. Just a bunch of bits turning left for 500 ns.

Re:

[m.dillon]

I caught the tail-end of the bubble memory era, even bought a few chips and played with it, but it became extinct very quickly (probably due to dynamic ram going into volume production just as the TRS/Atari/Apple/Commodore era took off.

-Matt

Need an analogy

[Cryophallion]

Without a proper Light -Distance analogy [slashdot.org] I have no way of being impressed by the speed of device. Is it knuckle to knee? Nose to toe? People need to know these things!

Eh.. English ?

[bytesex]

'Those domain can be shunted along this 'racetrack' and past read heads.'

What does that sentence even mean ?

Re:

[hostyle]

Hang on. Somethings coming through on the teletype: dominatrix ... shunting ... racey ... red heads!

Older than Dirt!

[Number6.2]

http://en.wikipedia.org/wiki/Delay_line_memory

Correct me if I'm wrong,

[jockeys]

but hasn't this been done in the past with electrical pulses sent down a very long wire? In a loop? So long ago that registers were called accumulators?

I remember my OpSys prof showing us one of these things that was new and shiny when HE was in school. Basically just a long (couple km, I think) wire wrapped up in a small coil the size of a shoebox that acted as RAM by sending pulses around the loop, reading them and then sending them again... the delay of electrons traveling the loop acted as extra space, until you were sending pulses continuously. Sort of like a circular stack.

Anyone else see some similarities here?

Re:

[Mike McTernan]

Or even sound through Mercury Delay lines [wikipedia.org]...

That's nothin'

[Jupiter Jones]

Meh.

Wake me when they come up with "Hot Dog" or "Crashdown" memory.

JJ

Great for Microsoft Windows.

[splutty]

If this gets used a lot, at least Microsoft can blame all their problems on 'race track memory things crashing'.

This is brilliant for them!

Seriously, though. The idea of storing bits on top of each other instead of alongside each other does give a much smaller footprint, and from what I understand from what I've read some 3 years ago, also a much better speed vs thermal efficiency.

For primary storage

[foniksonik]

The interesting thing is that they feel it is capable of being primary storage...so we're talking Terabytes...

Could be interesting.

Re:

[MtViewGuy]

If they can get this to work (including withstanding billions of re-writes), we're talking a completely solid-state storage device that could store at minimum couple of terabytes in the same form factor as today's ATA-100/133 and Serial ATA drives. Imagine loading Linux--including the full graphical environment!--or even Windows in under 10 seconds! :-)

Re:

[Kehvarl]

An AC has already pointed this out, but "Primary Storage" in the typical computer model is what we would consider RAM. Hard Drives and other "storage" media are considered to be "Secondary Storage". Primary Storage also consists of cache, and I believe processor registers. Removable media, is often called "Offline Storage" or "Tertiary Storage"

The relevant Wiki article is: Computer Storage [wikipedia.org]; specifically the sections: Primary Storage [wikipedia.org], and Secondary Storage [wikipedia.org].

The overall point I'm trying to make here is, we

Re:

[foniksonik]

From TFA:

His calculations suggest it could provide faster, cheaper and higher capacity storage than RAM or hard disk storage.

I'll stand by my statement.

Compared to PMC?

[babymac]

Does anyone have any idea how this compares to programmable metallization cell [wikipedia.org] technology which made the news recently? How close to production is PMC vs racetrack memory?

For some reason...

[Anonymous Coward]

...'RaceTrack' reminds me of 'TokenRing'

Re:

[Chaoticmass]

...'RaceTrack' reminds me of 'TokenRing'

I came here to say that.

business machines

[Anonymous Coward]

first they "Ship Fastest CPU on Earth" now this. all after loosing 300K jobs and being fucked by EPA. what the hell is going on at IBM? has the The Large Hadron Collider been turned on already? should we be expecting lots of zombies on pink unicorns anytime soon? sheeeeeet.

Sequential Access

[GottliebPins]

So they're saying sequential access is better than dynamic access? What's the advantage in that? Oh wait, I need to access data at location $ff00, let me spin the bits around the track a few thousand times til I get to that memory location. Duh?

Re:

[badboy_tw2002]

Oh man you hit the nail on the head buddy! What a bunch of idiots overthere! Don't they know anything? I always say, for any resarch project everyone should just post their idea to slashdot and see what comes up with the comments before they even spend a dime on it. Often you'll always find that:

a) What you're doing doesn't make any sense whatsoever based on that article I read a few years ago.
b) What you're doing was done 50 years ago and was better because that's what I did in my day.
c) What you're do

ah...

[apodyopsis]

I'm too young to remember bubble memory, but don't worry Google is my friend and the (ever so) helpful /.r's will correct any errors in their normal polite, restrained and helpful fashion.

From what I understand you created a circular magnetic field and then loaded a pattern into in, moving it around the field and past the read area to determine the data value. The pattern density was very high and the loops were printed onto the surface of the device, a bit like the old 80s game donkey kong they were joi

Sounds kinda like..

[ChronoFish]

A miniature token ring.... That went over well for IBM...

-CF

first computer memory was a "racetrack" device

[peter303]

Before they perfected magnetic storage, acoustic delay queues [wikipedia.org] in mercury were used as secondary memory.

Place your bets ..

[OzPeter]

Come on down to Racetrack Memory.

Set your data running the loop and then sit back and watch the fun.

Have drunken parties as you watch you bits speed around the course.

Place bets on your bits coming in first, a place, or if you are feeling really lucky plonk your money down for a trifecta.

Its so much fun for the whole family that even if you lose you still win!!

Timeline

[audubon]

The first ever racetrack memory device is able to store and read three bits of data using the racetrack method.

Assuming memory capacity doubles [wikipedia.org] every two years, IBM expects to have a 64 kilobyte version ready by mid 2025.

No betting

[UnknowingFool]

on the racetrack memory results. "Come on, NAND gate#7. Lucky #7! Daddy needs a new iPod"

Re:

[martin_henry]

I've got some inside info on #3's mean free path...

Why?

[toporok]

Is it me or old is new again? I'm getting tired of seeing things we've already done and moved on being broght back as the next best thing. And why would we want to move from solid state to a something that has moving parts. What if you drop it? Will it jump the "racetrack"?!

Re:

[toporok]

My bad, that's what I get for trying to multitask. Did not read the whole article and "ass-u-me-d" that wire was moving.

there are still some issues to be resolved...

[Khopesh]

Ars Technica picked this up yesterday [arstechnica.com] and has a pretty good run-down of how it works (complete with a pretty illustration).

They also provide Links to the Science articles themselves:

• Magnetic Domain-Wall Racetrack Memory [sciencemag.org]
• Current-Controlled Magnetic Domain-Wall Nanowire Shift Register [sciencemag.org]

It's promising, but there are still some lingering issues:

There is still work to do before an entire three-dimensional memory chip will replace your current memory solutions. The biggest problem may be heat; moving DW

Prior Art

[Jaazaniah]

There are magnetized wires out there with analog sound recordings on them from the days before vinyl. This is nothing new, and without a better control and recall system, is highly inefficient. Now, if the patent deals with the control system, kudos to IBM. If it deals with just the idea of selectively magnetizing a wire set, this has already been done. Google wire recording if you want more info.