Fast-Switching Micromagnets

apirkle writes "Why can't we use magnetic media as RAM? Flipping the north and south poles of the tiny magnets we call bits simply takes too long. A collaboration of researchers has recently demonstrated a much faster method for reversing magnetization, described in Physical Review Focus."

Now that...

[CptCook]

is a cool (possible) application of Magnetic Theory!

Re:Now that...

[jmccay]

It may be cool, but the problem is reading and writing to magnetic media is slower than reading and writing to memory. I once drastrically cut the run time of a function from 20 minutes to under two minutes by reading more data into memory for processing than the function origannally did when I got my hands on it.

Re:Now that...

[Simon Field]

That may have been a problem you worked on, but it has nothing to do with the discussion at hand.

This thread is about replacing fast RAM with fast non-volatile magnetic memory in handheld devices.

The article says these new memories have a fundamental switching speed limit of 165 picoseconds. This is what I find interesting, as 0.165 nanoseconds is a little over 6 gigahertz. Maybe the speeds will increase if the magnetic particles can be made smaller, but it looks like by the time they get a commercial device on the market, 6 gigahertz may be too slow for general purpose computers.

On the other hand, if they can make my terrabit flash memory chip in my digital camcorder record at 6 gigahertz, I might not need a terrabit of battery draining RAM to record my grandkids at play.

Dumping core...

[orangesquid]

Actually, tiny ferrite cores used to be used as magnetic memory. It wasn't called "RAM" then, but "core"... it was pretty cool stuff, (mostly) nonvolatile if a system powered off, but when you read a bit you had to rewrite it. It was invented in the 50's I think, and the inventor licensed it to many computer companies and made a healthy profit.

It seems those old sci-fi stories predicting computers of the future to have microscopic core memory (but still core memory) may very well come true ;)

Re:Dumping core...

[Koos Baster]

Yep, they did in the 50s, 60s and early seventies. I got one of these antique core memory modules at home. It features 2000 tiny magnetic cores (2-3mm), connected by a grid of copper wires, to provide a massive 400 words (4 bits + parity). It's definitely hand-crafted, probably by some poor kid in an Asian sweat shop.

The module is 20 x 30 cm, weighs a lot and produces a lot of heat, and apparently is very slow. Not exactly what you would consider the future of fast memories. But I wonder what contemporary production techniques and future nanotech (and probably an expired patent) can do to revive this technology. I'm sure that as with any "novel" memory technique it will take a few decades before it will pay off, else why would we still use magnetic hard disks for fast mass storage and not optical (or opto-magnetic) stuff?

--
The human brain is a wonderful thing: It starts working the moment you are born, and never stops until you stand up to speak in public -- Sir George Jessel

Re:Dumping core...

[Anonymous Coward]

Yeah, fun stuff. But in the 50s and 60s, people still had some sense of decency and morality. The people assembling those memories were Asian, but they were highly skilled women, namely jewellers and artists, and were paid accordingly. When a computer cost millions in 50s money, what's a decent wage for the employees?

Of course, the proliferation of dirt cheap electronics and a 'throw-away' mentality is responsible for the modern electronics sweat shop.

Also, the situation that university is a cult and makes a Master's necessary for even the lowest job doesn't help the fact that most jobs in electronics these days are of sweat-shop caliber.

Re:Dumping core...

[GigsVT]

http://www.howstuffworks.com/mram.htm

Several companies are currently marketing FRAM as a rugged replacement for Flash.
http://www.e-insite.net/electronicnews/ind ex.asp?l ayout=article&articleid=CA257740

Re:Dumping core...

[whiskers]

IBM 1620 (1960's small scientific machine) memory which you may have- it was 4 bits + parity anyway. That got you a decimal digit, everything was decimal, even the instructions and addresses. 20 microseconds for an access (read-writeback) per digit. 240 microseconds to read an instruction (2 digit op plus two 5 digit addresses). Each core had 4 wires (later 3) through it, an x, a y, a sense and if I remember an inhibit. to read you put a current just less than enough to change a "1" state to a "0" state on the appropriate x and y wires if the core at the intersection was a "1" the combined x and y current would drive it to "0", the change of magnetic field put a pulse on the sense line. No pulse was a 0. Then if it was a 1 you drove the x and y opposite to before to reset the core to 1. The fastest core I recollect was of the order .6 microsecond.

Re:Dumping core...

[dunedan]

i've heard that even with modern dram the capacitors are so small they have to rewrite them after reading so thats really not that different.

Reasoning

[schmink182]

Flipping the north and south poles of the tiny magnets we call bits simply takes too long.

Well we tried spinning the magnets to flip the east and west poles but all the other magnets spun too. It was a real disaster.

IBM MRAM

[eluusive]

The simpsons did it ....
No really, IBM has been working on this stuff for a couple years now, the projected release date is 2004.
Here [ibm.com] and Here [ibm.com]

Magnetic Bubble memory

[goombah99]

Any one recall magnetic bubble memory? Iwas the last person at IBM working on this in the 80's when they dropped it because although it was cool, disk drives were where the money was. Fujitsu worked on it a little longer, but eventually flash memory ate its lunch (application space)

it was cool. it worked like this a slab of magnetic material could have small (circular) islands of magnetism. from above this looks like a sea of dots on a surface.

Now onto the surface one writes a set of T shapped (or chevron shapped) pieces of permaloy. thousands of these in rows. The manetic bubbles are attracted to the edges of these T's and stick to them. When a transverse magnetic field is applied the T's become bar magnets and the bubbles move to one end to the bar margnet. rotate the magnetic filed the the bubbles move from one side of the T to the other. If the magnetic field is strong enough they will Jump from one T to the text. seen from above the magnetic bubbles march in lock step from one T to the next each rotation of the magnetic field.

ones and zeros are encoded either as presensce or absence of bubbles in the series of marching bubbles, or better yet by using higher order magnetization patterns within each bubble (N-bits per bubble).

since the materials are transparent and the bubbles are optically active, you can read out the bubbles with light. These things were built as large bucket brigades, just as CCD chips often are. You dont try to address a location on the chip directly. instead to determine the value of a bit, you shffle the whole marching line along until that bit position reaches your detector and you read it out. Thus its a serial access device (with many read heads). It's all solid state. the media is fixed (does not spin), instead the bits in the media move!!!

the problem was that these things were slower than Ram, and faster than hard drives. they were denser then ram but not as fast as harddrives. thus they had a niche rolle as non-volatile hard drive caches. their role as non-volatile memeory got squezed out by flash memory on the low end and harddrives on the high end.

Re:Magnetic Bubble memory

[goombah99]

Here's a photo of one:

http://starfish.rcsri.org/rcs/bubble/info.html [rcsri.org]

Re:Magnetic Bubble memory

[kcelery]

I always thought the magnet bubbles were pulled out of the market by the dept of defence. These device is ideal for cruise missile.