Slashdot Mirror
Magnetic RAM from IBM
n8willis writes "ZDNet has an article about IBM's latest breakthrough, Magnetic RAM (MRAM) that actually works. The story is a bit fluffy, rolling out every possible buzzword (eg, wireless video will now be feasible due to faster RAM technology???), but the tech - in development since 1974 - is indeed going into production as we speak. Gotta hand it to IBM these days: copper interconnects, 200ppi LCD monitors and now this." I'll believe it when I can read/write from it.
Geez, from the questions here, such as "How will this affect my floppies/other media/hardware?" and "how will this be affected by my monitor/speakers?"
I must only presume that none of you has ever heard of shielding.
*GOOD* quality speakers and monitors (not the $20 speakers from compusa) have shielding that minimize the magnetic energy escaping the equipment. place your cheapo speakers against your monitor. if it affects the monitor image, then you know you bought crap.
If things are properly shielded, you won't have interference or suffer any negative effects.
Geez, you don't overclock without proper cooling, why would you think of using a magnetic storage media without shielding?
(course, now all the tin foil hat freaks will come out and claim they were right all along....)
A host is a host from coast to coast, but no one uses a host that's close
Magnetic RAM, MRAM, is a non volatile memory with unlimited read and write cycles. It also has the potential
...being refreshed over 15,000 times per second!
to be very fast, and very dense.
Virtually all products that involve digital electronics (which includes just about everything these days: computers,
cellular phones, but also household appliances like washers, dryers, microwave ovens, refrigerators, and entertainment
devices like televisions, CD players, VCR's...) need memory. The choice of memory has historically involved a tradeoff.
Non volatile memories (memories that hold the information whether or not power is supplied) like EEPROM's can be
expensive and may have long term reliability problems - you can read and write to them only so many times or cycles
before you degrade the ability of the memory to "remember".
Volatile memories (memories that hold the information only if power is supplied) can be static RAM's or dynamic RAM's.
SRAM's (static RAM's) consume a lot less power than DRAM'S (dynamic RAM's), but take up a lot more chip area for a
given memory size, so are a lot more expensive. Because cost is generally the driving force in electronics, DRAM's
tend to have the highest volume of sales...they are the lowest cost, but consume a lot more power. In fact, to get the
density in a DRAM, we use a simple capacitor as the memory device. A charge on the capacitor can represent a "1"
and no charge can represent a "0". However, over time, the charge leaks off....losing all the information. To avoid this,
the DRAM must be REFRESHED. Essentially, a REFRESH cycle involves reading all the information out, and then putting it
all back in. Looking at a data sheet for a DRAM, the example has the 64 Meg DRAM needing to be refreshed every 64
msec
That refresh cycle on the DRAM explains the much greater power consumption of the less expensive DRAM compared
to the more expensive SRAM. While it may not be typical, a quick browsing through data sheets found a 64 M DRAM
that consumed 1 Watt of power, and a 64 M SRAM that consumed less than 400 mW.
An MRAM can have the density of a DRAM...and since cost goes with density, it is possible that an MRAM can have a
cost that is competitive with a DRAM...but it will have much lower power consumption...effectively zero power
consumption in the standby mode, which is the mode the memory spends most of its time in.
The MRAM doesn't require the refresh of a DRAM, and has the nonvolatility of a EEPROM, but has reliability superior to
the EEPROM. So, an MRAM may be able to replace DRAM's, SRAM's. amd EEPROM's, without involving any major
compromise or tradeoff.
It is hard to judge the improvement in the battery life of a cellular phone. With digital cellular phones used just
sporadically for voice phone calls, the battery life now is very good, but might perhaps be doubled with an MRAM - this
is just my guess, and may be way off. If the digital cellular phone is heavily used for phone calls, the power
consumption involved in sending the signal out would tend to dominate over the impact of improved power
consumption in the memory, and the impact on battery life might be minimal.
However, future applications of wireless products like the cellular phone include adding more functionality beyond
voice calls. For communication products, MRAM may make it possible to have access to the internet with the ability to
get video in addition to data and voice.
Will this memory be susceptible to magnetic interference from ordinary magnetic sources?
(*Imagine*)
The SmartFridge(tm)... A billion dollar effort by GE brings you a Fridge that remembers what you put in and take out, generates shopping lists for your agent software, and doesn't open after midnight.
*enter a small child*
"Oooh, this magnet stick to the fridge!"
(future engineer here)
SmartFridge orders 10lbs of Guacamole because its megnetic memory is scrambled like the eggs you can't have for breakfast because of the former.
1) It's integrated. Cores were a whole lot of tiny magnetic donuts, threaded into a two or D array by wires passing through the center holes. In most cases, three wires were used per core, X and Y selects and Read Out. Core memory arrays were enormous and they cost dollars per bit -- a small college I was attending in 1972 had ONE computer with just 16K-bytes of core memory. It sounds like IBM has worked out a way to grow the magnetic material and the read and write circuits on a chip--smaller and cheaper. (Whether it will ever be small and cheap enough to compete with other forms of non-volatile memory -- battery-backed static CMOS, EEPROM, and Flash -- is questionable. Let alone competing with DRAM, which would be necessary for many of those predictions in the article to come true.)
2) In the 1970's there wasn't a good way to read a constant magnetic field electronically. So what they did for readout was to send out the erase (write 0) signals on the X,Y wires. If a core was 1, this caused a pulse on the Read wire. After reading, it had to write the data back. Now, there are silicon sensors that can directly read out a magnetic field, so I assume that IBM is building these into the chips and you don't have to erase to read.
3. Power and interface requirements: It took a lot of power to write to those cores. Because of the erase-to-read, it took just as much to read them. They were low power only when you didn't use them. So the wire drivers had to use higher voltages than solid-state logic devices can handle, requiring level-changing interface circuits. The article doesn't say, but I would assume that the MRAM has much smaller magnetic thingies, requiring little power to switch, and can run at normal voltages.
I agree that it is a scenario unlikely to ever happen, but it has many problems like the one I've mentioned, should the appropriate "perfect" storage technology ever arise. These problems offer us insight into how other things work, and their solutions would perhaps help us predict and design things better in the long term.
IBM announced a colabrative effort with the US DOE to develop a "reactionless space drive" based on thier new MRAM technology. The idea came about when casing vibrations were observed during photoshop benchmarking of computers equiped with MRAM. Brokehaven researchers are modeling the effect in order to determine the optimum layout of memory slots for propelling laptops, spacecraft etc.
This is like MS claiming to have invented a new technology when they came out with their "optical mouse".
Well, they did. All previous optical mice required grids. Microsoft's mouse uses a camera and software capable of comparing images very quickly to determine speed and direction. A far cry from grid-based optical mice.
"And like that
The cool part is that it works by wiggling atoms around in the crystal structure:
... an idea, the fugitive fermentation of an individual brain ... -- T. Jefferson
I'll believe it when I can read/write from it.
:)
And he does not mean with a pen.
IBM recently announced their astonishing, new, unprecedented breakthrough in MAGNETIC memory that ACTUALLY WORKS, (some old industry wags are referring to it as core, for some unknown reason). Shortly afterward, they announced a startling new breakthrough method of storing data by punching holes in cards, a development no doubt inspired by the efficient vote tabulations in Florida.
include $sig;
1;
D%$#it! spintronics.com is already registered.
For all intensive purposes, "whom" is no longer a word. That begs the question, "who cares"?
Not no power consumption. No power consumption when not reading or writing data. And that isn't even 100% sure, it may need sense and positiong circuits powered up for any sort of acceptable memory access time. For all we know the actual power needed to read and write may be far more then DRAM (or far less).
We don't know what kind of memory this thing will actually replace until there is more info on operating power/speed/tempature, and storage tempature/duration, and density, and sheilding required (which is effectavly density).
This could be the next DRAM, or the next FLASH, or just the next Bubble Memory.
Well, if you're running a basic workstation (that is, nobody needs 24/7 remote access to it), You can save a fair chunk of energy by powering down your system without losing system state--a fair step up from current suspend methods, which basically need to write the contents of RAM to disk and then read and reinstate said contents to RAM (which can be a tricky/time consuming proposition.) Extend this tech to other components (processor, periphrials, etc) and you could eventually have a truly persistent-state computer, where the concept of "powering off" falls by the wayside, as power is simply siphoned when needed; since your computer wouldn't need a constant flow of electricity to maintain state, it could effectively draw zero power in an when idle, thus being (theoretically) as effecient as if it were off.
So basically, in the short term, snapping the power on and off when you finish using the computer becomes a quick, simple thing to do. If you get a little visionary with this, you may not even need a power switch on your computer eventually. Funky thought, huh?
$ man reality
Obliteracy: Words with explosions
This is nothing new, they were called CORES, and they've been around since the '70s!!!
This is like MS claiming to have invented a new technology when they came out with their "optical mouse".
Most electronics up until now work within the charge domain. That is, devices all deal with moving, changing, transferring, etc, electric charge and the lack of electric charge. Amplifiers, for instance, can amplify current, which is the flow of charge, or voltage, which is the energy/charge ratio. Semiconductors exploit all kinds of funky physics to do these things.
However, there's a whole other degree of freedom of the electron that's virtually unused. Spin. In any elementary quantum mechanics course you'll learn rather soon on that electrons are Fermions with total spin 1/2, which means there are two spin states an electron can be in, usually called spin-up and spin-down.
So the new world of spintronics aims at manipulating the spin of the electrons, instead of the charge. Spin is a different beast than charge, in that it can be manipulated by magnetic fields and light, in vastly-different ways than does charge flow. Spin is a fundamental nature of angular momentum, so whereby the total charge is conserved within a small sample, so too is the angular momentum.
Some of these MRAM's were specifically mentioned yesterday, in that the parity of the spin can be used to store bits. One nice fact about this could be that information isn't lost if power is turned off, unlike DRAM's and many SRAM's.
It's a VERY new field, spintronics. I did a search on Google last night for only 'spintronics' and only 665 sites were listed. It's been around for a few years so far, but there have been problems with finding the right magnetic materials. You need the right combination of ferromagnetic and antiferromagnetic layers, and certain ways to test materials, before you can really start doing some good stuff. However, materials are starting to be found, so it's an exciting time for this potentially huge field.
Hopefully soon there'll be spin-like transistors, leading to spin amplifiers, and all sorts of other goodies. Sorry I don't have any specifics about this, but I just found out about it yesterday.
make world, not war
Instant on? Of course. Think about. Saving all of the startup files and bootloader in the ram, that will store information all the time. How could it losse power? It wont. It doesn't need an electric charge to store the bits in the ram, so will hold, in theroy, for ever. This is a huge breakthrough. And the amount of mem it could hold in one chip could easily be much higher then the 256MB that the early chips will run. Think about having 1GB of RAM on your home machine with just one chip, that will boot up your machine instantly and will never losse the information stored in the memory if your power get's cut off from not paying the electric company :) awwww, that will be sweat.
:)
And yes, faster surfing and faster downloads as well. The high-end servers will be able to store (or cache) the file's its serving up in the lighting fast memory. It will access the memory faster than a scsi raid getting thousands of hits per minute, or even per second.
And I'm SURE that the chips will be shielded, so you dont need to worry if it will wipe out your hard drive or floppy when its turned on. I would think that IBM wouldn't be THAT stupid
--we are the music makers, and we are the dreamers of the dreams.
--we are the music makers, and we are the dreamers of the dreams.
Billy
You can find an article on Stuart Parkin (also developer of the GMR head in your hard-drive) who is the lead scientist on this project at http://www.wired.com/wired/archive/8.04/mram.html
http://www.technews.com/news/00/159086.html
"The speed increase is thanks to subtle a twist on the solid-state memory technology that has driven computers for almost for decades. The technology is known as interlocked pipelined CMOS (complimentary metal oxide semiconductor) and will allow memory chips to reach speeds - in theory, at least - of between 3.3 gigahertz (GHz) and 4.5 GHz, using conventional silicon transistors. "
"The key to the new technology is a distributed "clock" function. In computer chips, the clock paces the speed of the circuits. Standard designs use a centralized clock to synchronize the operations of an entire chip, ensuring that all operations run at the same interval, or cycle. The clock waits for all the operations on a chip to finish before starting the next cycle, so the speed of the entire chip is limited to the pace of the slowest operation. To increase the speed, the IBM researchers decentralized the clock, using locally generated clocks to run smaller sections of circuits. Infineon says that it is working closely with IBM's New York Fishkill research operation on developing MRAM technology still further, with the intention of allowing the memory chips to function like bubble memory, which retails the computer data, even when power is removed from the chipset. "
We emerge from our mother's womb an unformatted diskette; our culture formats us.
We emerge from our mother's womb an unformatted diskette; our culture formats us. - Douglas Coupland
The point was speech recognition on cell phones, presumably because these MRAM chips are so tiny and energy efficient that you can cram 256 megs into a phone. I'm baffled by the "easier downloads" thing though... maybe MRAM will we very very fast, speeding up computers in general....
Or its just marketing crap. One Best Buy ad for an HP with a Pnetium IV claimed that MP3 downloads would go faster on the new chip....
---- I made the Kessel Run in under 11 parsecs.
Yes, it is a catchy buzzword, but at least it's an applicable one.
make world, not war
Does this mean that main memory and mass storage can now merge on the same medium? So essentially your filesystem IS your memory. The heap/stack/whatever is just located in some separate protected area. This would mean entirely redesigning the vast majority of operating systems which were designed to optimize the exact problem of memory/disk space usage, protection, etc. All of a sudden "disk cache", DMA, binary loaders, etc. are all history. Memory fragmentation and disk fragmentation are now the same thing. Anyway, sounds really cool. However, 256 MB mass storage is pretty damn small, and will be even orders of magnitude smaller 10 years from now. They better be able to ramp that number up.
It's 10 PM. Do you know if you're un-American?
The thought of not having to wait for some of the servers here to reboot is even more appealing.
Why should you not have to wait for your servers to reboot? If your servers reboot because of power failures (in which case I'd advice a UPS ;) then there is still the issue of fsck and mount that renders the RAM useless, if your server reboot because new hardware is added.. then you still need to reboot because you need to install a new kernel, if your machine crashes then a reboot won't help much, etc.
Monkey sense
Interestingly, the idea of uniform storage isn't used in humans. We have a short term and long term memory - which duals, at least in pedantic respects, that of RAM and magnetic media right now. It will be interesting to see how computers would evolve if all memory was addressable in the exact same direct (in terms of no buffer) mechanism.
What I don't understand is why this can't be achieved using EEPROMs or just regular old battery-powered NVRAM. I remember the old Macs used to have a great deal of the system on ROMSs ... why not take it to its logical conclusion and put all the libraries, kernel, system files, and for that matter OS applications on a ROM or EEPROM?
If the answer is that ROMs are slower than RAMs, can anyone explain why? I would think ROMs would be faster, not having to worry about writes...
Rambus claims to hold Magnetic Memory patent, will press IBM for licensing and royalties.
Microsoft invents Command Line Interpreter interface, says will replace Windows 2000
Shed 30 Lbs in time for the Holidays, with Linux powered dietary supplements!
Burroughs spins back out of Unisys, claims old taint associated with name forgotten.
HeathKit to introduce DNA Lab, l33t k1dd13s will be able to grow own dinos, munch San Diego, make bad movies.
OSX found to cause baldness in lab mice.
Black hole found forming in southern California desert, site said to contain Atari ET cartridges. Disney plans to use site for remake.
Warriors defeat Lakers, credit Buckminster Fuller for inspiration. "We kept thinking of them as Bucky balls and all the potential they encompass!"
Earth's moon actually stolen from Mars, water too!
--
A feeling of having made the same mistake before: Deja Foobar
>>Here's an article from Scientific American on the topic.
Which points out that the data density wasn't there, and that they were slow and expensive. Commercially available isn't what we're looking for: commercially viable is, and Honeywell's product clearly wasn't. Admittedly, it doesn't sound like IBM is going to be able to make them cheaply for quite some time either, but at least they have the access times down (or so it sounds--the article was a bit vague, as is ZDnet's want).
High-speed Road Trip (18.000KPH)
Sadly, with IBM's overly conservative estimates of "We won't have this in large scale production until after the next epoch" it's more torturous than vaporware.
--
A feeling of having made the same mistake before: Deja Foobar
...wouldnt the magnetic charge from say, turning on my monitor, cause problems with the MRAM, without some big shielding? Considering if I even think about magnets any floppy I have sitting around goes foom, whats going to happen when I'm relying on my 'instant on' MRAM to be ready?
Here's an article from Scientific American on the topic.
--
--
You are a fucking moron.
Seriously. What makes MRAM so special that it can beat out those other concepts?
On a slightly more serious note, some of the early palmtops had problems when they didn't have a way to clear memory and so had no equivalent of "reboot" if the system gets hosed. I think we'll see a BIOS ability to clear the "memory" on reboot and reload everything. Or maybe OS's will improve, but when something dies really badly I find it nice (windows or linux! I find this happening under both, though more for linux) to be able to start over with a freshly cleared memory.
Lots of interesting reading
"It is a greater offense to steal men's labor, than their clothes"
With conventional memory this thing eat battery even if it's turned off and when the batteries run out - whooosh! All data stored in the RAM is gone.
Microsoft? Is that some kind of a toilet paper?
In production by 2004
There will be some test versions out in 2003, but IBM, which will make the chips at its own plants, doesn't expect volume production until sometime in 2004, Davari said.
If this is correct, you will not want any of the initial run. Who things we'll still want a 256Mb ram chip in 2004? More likely we'll be at 512 as a minimum, and as such, these will be worthless. Not to mention that they will be expensive, in all likelyhood.
I'll wait until it's less expensive, personally, and more usefull - like say a 1Gb chip. THAT would be cool. I'd think it has OTHER applications as well - imagine a storage CUBE of these, say 100Gb? I'd find that VERY useful. Since it's magnetic, it could be treated like a big drive. Imagine the access time!
Fawking Trolls!
"Going to war without France is like going deer hunting without your accordion." - Jed Babbin
Even though I'm well aware of the fact that MRAM is decidedly far more advanced than the magnetic RAM of old, I can't help but chuckle at the article's starry-eyed vision of tomorrow's "instant-on" computers. I can close my eyes and imagine...
$ man reality
Obliteracy: Words with explosions
Well its great and all that the inital chips will be available in 256mb varities...and a smaller physical chip is great for MP3s player and such... but the article says IBM doesn't think it'll be in volume production for another 10 years. So um...what difference does it make on the products we have today? Do you seriously think that 10 years from now we are still going to have portable mp3 players?
I don't like how the article makes it seem like MRAM is the answer to today's memory design challenges, when most of the products its going to benefit probably won't be around in 10 years. Or of they are around, they'll be completely different.
---
That I got from the IBM site here
I demand a million helicopters and a DOLLAR!
What I am waiting for is a 10Gig (or more) bank of non-volitile memory.. It wouldn't even have to be fast.. Hell, if it were 100 times slower than SDRAM it would still be worth while.. The reason is that you could treat main-memory as a cache into a very very special and protected read-write region of the nvram. With this you could do away with all swapping and most of the complexities of disk-writes.
In fact, you could finally treat files as character devices instead of block devices (though you could maintain the latter for compatibility). When you do this, memory mapping of disk-space is nothing more than literally allocating a chunk of nvram into your address space.
You'd still have to perform disk-syncing, though it would be more like cache flushing than anything else. The disk sector would shrink from 512Bytes to 128Bytes (a typical cache line)
Through the use of memory mapping and independant read/write/execute permissions on memory pages, you could achieve a whole new software design structure which completely eliminates the need for OS-calls for disk access for anything other than setup and shut down. Databases could achieve phenominal performance gains.
I know that nvram disks have been around for a while, but they've been incredibly expensive, and only support very small disk sizes (couple gig max).
Assuming the power requirements on MRAM are low, and that the cell size can be comparable to that of DRAM, then we should be able to achieve 40Gig MRAM disk drives at some point, which are a good size for serious servers. So long as you can daisy-chain them as with SCSI, then you're good to go.
On 64bit machines, all you need are the appropriate drivers and you're golden.. Sadly, 32bit machines would require some sort of banked paging. Thankfully, this would enhance the general desire for 64bit machines and then things like AMDs x86-64 and Intels Italium would receive a brand new source of demand. SUN and Alpha would be the immediate benificiaries, of course.
-Michael
I don't wait for my laptop to boot - I suspend and resume it like most laptop users. The people who write articles for ZDnet are clearly so inexperienced with computers they've never even used a laptop (or maybe they have but they were too scared to select 'suspend' from a menu). No wonder they can't figure out what faster RAM is all about.
--
-- SIGFPE
I've always wanted to be able to stick computer memory to my fridge door!
--
The gift of death metal does not smile on the good looking.
I'm truly impressed, though, since I saw that tiny little 340Meg hard drive that could hide behind a Kennedy half-dollar. They've figured out how to make all those tiny little wires and ferrite beads on a micron scale, eh? At last we can upgrade the 360/40! Kewl! ;-)
--
A feeling of having made the same mistake before: Deja Foobar
--
Trolling using another account since 2005.