NEC Develops World's Fastest MRAM

Gary writes to tell us that NEC has developed a new SRAM compatible MRAM. The new memory module is capable of speeds up to 250MHz, the world's fastest to date. "MRAM are expected to generate new value and applications for future electronic devices thanks to their nonvolatility, unlimited write endurance, high speed operation, and ability to cut memory power dissipation in half. For example, these features could enable instant start up of PCs and prevent drive recorders from losing data after a sudden break in power in the future. As substitutes for system LSI-embedded SRAM, MRAM can provide even more value as they are expected to enable extremely low power dissipation of system LSIs because they can sleep when they are not in use and wake up instantly."

quality workmanship

[User 956]

MRAM are expected to generate new value and applications for future electronic devices thanks to their nonvolatility

That is, assuming they're not manufactured by Sony.

Nick of time for NEC

[goombah99]

NEC's stock has been in the toilet for 5 years. Actually Going down by 50% while the rest of the market rose. They need a potentially huge marketable break through. But I wonder if phase change ram will eat it's lunch. It too is supposed to be fast and non volatile.

Re:

[networkBoy]

One benefit for Ovonics based tech (PCM) is that it is inherently radiation and magnetic field resistant. While I realize that the former of those applications is nominal, the latter is not. It's only downfall is thermal stability, the temperatures experienced in reflow are sufficient to erase the memory. While this is beneficial from a security aspect (strip line heater on top of your memory bank, hit the panic button and poof the memory is blank) it may not be in other industrial applications.

Any indic

Re:Hooray!

[inviolet]

For example, these features could enable instant start up of PCs and prevent drive recorders from losing data after a sudden break in power in the future. As substitutes for system LSI-embedded SRAM, MRAM can provide even more value as they are expected to enable extremely low power dissipation of system LSIs because they can sleep when they are not in use and wake up instantly."

Now we can throw a rootkit into memory and have it chill there forever!

On a related note, non-volatile system memory will completely change the game for forensics experts. Right now, when they come and grab your computers, all memory contents are lost... and clever people also disable the swapfile. With MRAM, all that is out the window.

Watch for a new meme in the next years, categorizing the use of volatile RAM as a presumption of guilt.

Re:

[randyest]

At 250MHz I don't think MRAM will be replacing your system SDRAM anytime soon (since it's probably 2-3x that if you have a relatively recent system.) But your point is interesting -- when the ASICs and chipsets all over your mobo and peripherals have MRAM in them, the forensics guys will have a field day reading that data you thought wasn't being recorded at all :)

Re:

[afidel]

Most of the "forensic guys" use a COTS software package to look at your Windows, Outlook and IM logs, throw something as trivial as Linux at them and they are lost, asking them to try to look through the MRAM of ASIC's would be funny. Sure if you're a high profile suspect for the FBI/CIA there's a chance someone with the knowledge will look at those things, but very few of us have that much to worry about.

Re:Hooray!

[networkBoy]

System memory is currently running at 266-333MHz in the high end so this is a perfectly viable replacement in mid-range systems. Also for embedded systems where start-up time is more visible to the consumer than raw speed, again this is a viable replacement.

Remember DDR2 PC800 is 200MHz quad pumped not 800MHz.
-nB

Re:

[dogs4ar]

This is great news, and all. But I was wondering, how will they overcome the 640k limit?

I'll need to load all of my programs high, again, I suspect.

Re:

[randyest]

The high end is much, much higher than 333Mhz -- more like 1.8 GHz [newegg.com] for system memory. That link is for newegg's SDRAM DDR3 1866 (PC3 15000). But besides that, consider cache speeds, which this is more likely to replace.

Re:

[Anonymous Coward]

Yeah, you're not crazy.

Re:

[mOdQuArK!]

If privacy becomes a serious enough issue, then somebody will build an auto-encrypting memory interface (with a volatile key register).

Of course, the DRM folks will insist on a section of memory that only THEY have the key for, and the government will insist on key escrow, etc, etc...

MRAM is actually used

[Dwedit]

For small sizes (32KiB), MRAM already has a wide use in Game Boy Advance cartridges as a replacement for battery backed RAM.

Could the reverse be done?

[SanityInAnarchy]

I always wondered why more people didn't use battery-backed RAM with some slower, more persistent storage to dump it to when you lose power.

So really, the question is, which is cheaper: a gig of MRAM, or a gig of battery-backed RAM with a gig of flash or hard disk to dump to?

Re:

[addaon]

The battery-backed SRAM used in devices like game cartridges is medium-speed, very-low-power SRAM. It's pretty standard to see battery ratings of five years; twenty years is available pretty readily. With times like these, there's really no reason to have secondary persistent storage, especially since the energy for doing the dump (which you must reserve) is likely to be enough to power the SRAM for another couple years.

Re:

[mzs]

Yeah we use a cap and coin cell combo here at work. The capacitor provides about a week's worth of time, then the battery can do years. The cap is also useful for replacing the battery.

Re:

[photon317]

Your non-MRAM scenario (battery-backed fast DDR volatile ram caches with flash drives behind them) is exactly what Texas Memory Systems has been doing for a while. If you really want to throw hardware at certain performance problems, their solutions are quite useful.

http://www.superssd.com/products/ramsan-500 [superssd.com]

Oh, I know.

[SanityInAnarchy]

That's why I said "I wonder why more people don't use..."

Mostly, I'm just curious about the economics of this -- is tech like MRAM ultimately going to be any cheaper than battery+RAM+backup? If so, when, and for what applications?

Re:

[randyest]

It can be. Just as ECC is an optional, additional circuit plus some redundant bits in SRAM, DRAM, SDRAM, etc. ECC could be added to any conceivable type of storage. Given the much lower failure rate of MRAM (almost zero soft error rate) I don't think it'd be worth the overhead though.

Re:

[FredFredrickson]

The correct answer was +1 Funny, or -1 Lame Joke, but it was about the memory, so off topic doesn't quite apply here.

Hmmmmmm

[Hanging By A Thread]

"because they can sleep when they are not in use and wake up instantly."

Reminds me of my cat.

Re:

[ajs318]

You weren't searching very hard, were you? link [johnlittle.us]

Be sure to have the hardware modified to avoid spawning unwanted child processes. Core dumps are unavoidable, but you can mitigate the worst effects by running in a sandbox.

Re:

[mgblst]

I wonder what "uses" you put your cat to?

Awesome!

[IdeaMan]

Now all they need now is faster WOM [national.com]

Actually I did play with the serial MRAM's back when I was an embedded systems engineer, they were pretty cool. As I recall they didn't have the write cycle count issues that EEPROMs had and had way faster write cycle times.

Absolute values are nice ...

[maxwell demon]

... comparisons would be nicer.
How does the MRAM speed compare to typical SRAM speeds? And to typical DRAM speeds?

And what about the size, compared to SRAM and DRAM?

Re:

[Dan Ost]

As long as the speed is faster than flash (which it is, by many orders of magnitude), there will be a market for this stuff once it becomes available.

Re:

[Smidge204]

From the article:

"...the new design achieves an operation speed of 250MHz; double that of conventional MRAMs and almost equivalent to that of recent LSI-embedded SRAM."

Doesn't say anything about size though, assuming you mean physical size/bit density.
=Smidge=

Re:

[mzs]

Compared to the slower lower power SRAMs you would use in a battery backed scenario (measured in years), this new MRAM would be faster. MRAMs I have seen are bigger in size and way more expensive than SRAMs though.

Re:

[maxwell demon]

From the article:

Oops. I didn't read the whole article, but I scanned through it and didn't find that info. Obviously I have to work on my scanning skills ...

BTW, yes, I indeed meant physical size.

Re:

[WalksOnDirt]

And what about the size, compared to SRAM and DRAM?

From the article:

The unique MRAM was designed and fabricated by NEC and has a memory capacity of 1 megabit.

I was hoping for better.

Re:Absolute values are nice ...

[SnowZero]

... comparisons would be nicer.

This new MRAM can process 1.1 million operations in the time it takes an egg to fall the width of a human hair. In fact, it's so fast, it can output 91 words in the time it takes light to travel the length of a football field.

Hope this helps...

WTF is MRAM?

[bobdotorg]

I was unfamiliar with it, so I looked it up:

Magnetoresistive Random Access Memory - two magnetic plates separated by an insulator. One plate has is a permanent magnet, the other holds the temporary charge.

Cost?

[show me altoids]

Nowhere does the article mention anything about cost. Anyone have an idea of the relative cost? With the magnetic plates for each bit, sounds like it might be expensive.

Re:

[P3NIS_CLEAVER]

The other plate doesn't store a 'charge', it is magnetized or demagnetized, much like bits on a hard drive.

Re:

[Z34107]

The other plate doesn't store a 'charge', it is magnetized or demagnetized, much like bits on a hard drive.

Magnetic charge?

Re:

[default luser]

No, when you magnetize a material (e.g. screwdriver), you are not adding charge of any kind, you are simply aligning the atoms in a different arrangement that produces a magnetic field. When you demagnetize an object (i.e. smack the screwdriver against something), the aligned arrangement is lost, and so the magnetic field is lost.

You can only magnetize certain materials that have crystal arrangements susceptible to magnetic fields (ferrous materials, and a handful of others). If it were simply a matter of

Re:

[Myrcutio]

i wonder if this means its magnetically sensitive, as in leave it too close to your speakers and your data gets corrupted.

MRAM Solid State Hard Drives

[Nonillion]

Cool, this sounds like the ticket to fast Solid State hard drives. I know there are some flash drives being produced, but the limited read/write cycle is what has kept me from trying one. I would most certainly like to have a drive where slew rate and rotational latency are non-existent.

Re:

[goofy183]

I'm nearly positive that the read/write cycle issues have long been moot. http://www.storagesearch.com/ssdmyths-endurance.html [storagesearch.com] I realize this is an industry sponsored site but even with taking very pessimistic views of their numbers a flash drive will last far longer than most disk based drives on the market will.

Re:

[TheRaven64]

For those too lazy to read the linked article, the author indicates that a flash drive will wear out from the finite rewrites problem after 51 years of being constantly written to at 80MB/s. I did a similar calculation in a different /. post a few months back based on the average write rate from my laptop hard disk over the preceding month. In my calculation, a flash drive would last a shade under 1,000 years if used in the same way as my laptop disk. In short, your flash drive is almost certain to fail

Re:MRAM Solid State Hard Drives

[Alsee]

Cool, this sounds like the ticket to fast Solid State hard drives.

The automobile is the ticket to moving horses around faster.

Or you can just have a car and skip the horse altogether.
And have 200 Gig of RAM and skip the Solid State drive altogether.

Whenever you buy new software you just put the software in the drive, load the software into your 200 Gig RAM, then you can just hit the power switch on the computer. Then whenever you want you just tap the power switch for an instant power-on and ALL of your software and ALL of your photos and ALL of your music and everything else, it's all already live in your 200 Gigs of RAM.

Yeah you'd want to change some aspects of the operating system to adapt to this new paradigm, in some ways you want to add new "hard drive style" management features in how you handle RAM, but you could throw the entire buggy-whip notion of a hard drive right out the window.

The only issue here is whether this is too expensive to have 100Gig+ bulk memory... but if that's the case then it would be too expensive for a "Solid State Hard Drive" anyway.

-

Re:

[Courageous]

The only issue here is whether this is too expensive to have 100Gig+ bulk memory... but if that's the case then it would be too expensive for a "Solid State Hard Drive" anyway.

Your conclusion paints a false dichotomy, with the unstated assumption being that any form of SSD technology has to at least be as expensive as RAM. This assumption appears unmerited, from direct observation of buyable SSD's today...

C//

Re:

[Seraphim1982]

Your conclusion paints a false dichotomy, with the unstated assumption being that any form of SSD technology has to at least be as expensive as RAM.

How does he do that?

As far as I can tell he states:
1) If this technology is expensive might be suitable for RAM, but not for SSDs
2) If this technology is cheap, it might be suitable for SSDs but would also be suitable for RAM, so remove the disk/ram distinction and have one large bank that acts as both

Re:

[Courageous]

Oh. Rereading, I see I misunderstood. But according to Slashdot, I'm 'Insightful'.

*chortle*

C//

Re:

[Alsee]

I'll chuckle too. I didn't see Seraphim1982 already answered you, and actually I think I like his short clean answer better than my own :)

-

Re:

[Alsee]

Your conclusion paints a false dichotomy, with the unstated assumption being that any form of SSD technology

No, I did state "this", meaning *this* technology.
If you have nonvolatile memory that is as fast as system RAM, then you may as well use it as system RAM.

A paradigm shift. When RAM is nonvolatile, it possesses all the capabilities of a drive. When a drive has the speed of RAM, it possesses all the capabilities of RAM. With this technology the capabilities RAM and drive are unified. The very concept an

Re:

[k8to]

Well, when you have nonvolatile memory that is as fast as DRAM and as *cheap* as DRAM, you might as well use it as system ram. We've had the former for a long time. In fact, DRAM has been slower than some forms of persistent ram for a long time. It's just much cheaper.

Re:

[Anonymous Coward]

Be honest, I think the limited write cycles of your wallet are the real reason that's kept you from trying one.

Re:

[John Whitley]

Why in the heck would you want to put fast nonvolatile memory behind a klunky, slow disk interface? Are you nuts? Just map it directly into the system address space and away we go! Use a ramdisk if you really need to work with that storage as a filesystem.

Note that high-speed, high-capacity non-volatile memory completely screws with many built-in assumptions in modern operating systems and the use of their APIs. What happens when a disk orders-of-magnitude slower than RAM no longer slow and isn't even t

mmhm

[kalvyn]

They've been promising us "instant-on" PCs forever. The technology is there now, but as Bruce Schneier indicates "... the current crop of major operating systems just don't" (from Freakonomics Q&A [nytimes.com]. I'll believe it when I see it. I'm from Missouri, so you'll have to Show-Me!

Re:

[MBCook]

I have an instant on computer now. It's called a MacBook Pro and I just put it to sleep. I never have a need to actually cold boot.

This kind of thing would allow you load an OS wicked quick, but there are still some problems. There is some hardware (do X, wait 200ms, do Y, wait 200ms... until the hardware is initted) that will slow things down. Then there is the problem of as computers get faster, they are asked to do more stuff so it takes longer (in absolute cycle counts) to boot them up.

Re:

[TheRaven64]

As you say, it's not particularly interesting for computers. The difference in power use between suspend to volatile RAM and suspend to non-volatile RAM is very small and not really worth bothering with. For devices which spend a lot of their time in their 'off' state, however, it would be very attractive. Things like set top boxes, for example, would be an ideal example.

Re:

[Overzeetop]

It will never come to pass, unless you equate instant on with standby, which we already have.

I have 2GB of RAM in my PC. It takes close to 4 minutes to go from cold to functional. Now, I have a slow hard drive - maybe 300Mb/s - but that still means that I should be able to load the memory completely full in less than 60 seconds. I can when I come out of hibernation, but somehow the disc will run continuously for 240+ seconds on boot? It has nothing to do with the memory, and more to do with the fact that a

Re:

[vranash]

You missed a decimal there: 300Mb/s should read 30.0Mb/s :) That's one of the reasons they take so long to boot up, the rest prehaps someone else can elucidate on.

Re:

[afidel]

It has little to do with resource usage and more to do with disk seeks. Using a SAN analyzer to watch a system boot is quite informative, even on a good SAN with 15K disks you only get a fraction of the theoretical throughput because the system does so many random seeks at boot time. MS has tried to address this with the boot time optimizers in XP and Vista with mixed success.

Re:

[MLS100]

Something is wrong, I think. 4 minutes is ludicrous.. I can go from cold to functional in about a minute on this ancient computer (1 GHz Pentium 3, 512MB RAM, XP SP2).

ASUS CUSL2-C, still kicking.

Re: Operating system startup time

[Douglas Goodall]

Contemporary PC operating systems (with plug and pray) are faced with identification and operation of a multitude of devices that can be inserted into the bus or attached to one port or another. Part of what takes a lot of boot time is probing for devices and identifying drivers appropriate to operate them. This is one area where Apple can excel as they have a smaller range of hardware to support (not counting usb...). I recently looked at the linux driver source code for handling serial ports and was astou

Re:

[Overzeetop]

Again, interesting in that the OS is essentially locked up at this time, with no ability to load other user programs or even accept user input. I realize that 2 billion operations a second seems trivial these days, but you'd think there's be a few million cycles left to serve the user.

Re:

[pimpimpim]

4 minutes boot time, I guess you count until all programs have started, if you'd look at the time to get the desktop visible (but without being able to actually do anything) it is shorter.

Still, the EEE pc takes 5 minutes [blogeee.net] to reinstall the complete OS from DVD. Full boot seems to be about 20 seconds. I think this is going somewhere finally!

I hope...

[Anonymous Coward]

I hope MRAM catches on and sets the world on fire, becoming cheaper than flash.

It's just better all around, especially because there isn't limited write endurance like flash has, and because of the speed, and because it's easier to drive (than, say, flash).

"But AC! There is wear levelling! Flash write limits are no longer relevant!"

If you have gobs of flash, sure. Not so in embedded devices however. And in those cases MRAM can easily replace battery backed ram. YEIGH!

Re:

[Tuoqui]

Wont happen, the big companies need to keep things with limited write endurance otherwise they'll go out of business. As long as they can keep you on the consumer treadmill they stay in business.

Re:

[afidel]

Write endurance is already good enough for quality flash that at interface speed it would take 20+ years to wear out the chip with wear leveling. Not only that but you have to remember that data is a gas, it expands to fill its container.

Re:

[mrv20]

Not true for storage. One only has to wait a few months before consumers are clamouring for a new version of whatever product it is with a larger data capacity. The obsolescence comes naturally from the rapid improvements in the field.

Instant on PC

[Jeff1946]

Could you put a flash memory in a PC with a wide data bus and use it to store the hiberate sys file instead of the disk, with say a 128 bit wide bus should be able transfer memory to flash in a few seconds.

Not a replacement for SRAM... yet

[Kazymyr]

I have some MRAM samples waiting to be tested in my drawer (4Mb chips from Freescale). They look good as replacement for flash chips rather that SRAM, because of better reliability and lower power consumption, however the technology is quite young and hasn't reached yet the packing density of flash, or the speed of SRAM. Lots of potential though.

Re:Not a replacement for SRAM... yet

[Kazymyr]

Replying to my own post - here's a brief paper explaining the technology (PDF warning):
http://www.freescale.com/files/microcontrollers/doc/brochure/BRMRAMTECHGUIDE.pdf [freescale.com]

Re:

[afidel]

however the technology is quite young

Huh? MRAM has been the next big thing as long as I've been in the IT industry, coming up on a decade and a half. Noone has figured out how to make them cheaply and with enough storage space to rival flash or SRAM.

Re:

[Kazymyr]

...coming up on a decade and a half...

Precisely my point. Young.

Re:

[k8to]

My response to this is that flash is of approximately the same age, but has been a large success. I'm curious what should cause us to expect this situtation to change.

Re:

[default luser]

My response to this is that flash is of approximately the same age, but has been a large success. I'm curious what should cause us to expect this situtation to change.

Flash is a lot easier technology to implement, since it uses the very same basic transistor gate technology as SRAM. The only difference is, Flash uses a Floating Gate Transistor instead of feedback to store the charge long-term, at the expense of write speed. Thus, once the idea was hatched, it was pretty obvious how to go about implementin

Get me off these spinning platters.

[camperdave]

OK, so they're fast. They're non-volatile. They're low power. They're immune to "wear". So where's the MRAM solid state drives?

Re:

[llamalicious]

I have one powering the MP9 player in my flying car.

Oh shit, I think I've altered the timeline, again.

Re:

[Chirs]

You forgot, "they're expensive and low capacity".

Give it time....

Alternate link

[flatulus]

The linked article seems to be Slashdotted. http://www.nec.co.jp/press/en/0711/3001.html/ [nec.co.jp] is NEC's own press release.

expected to Generate New Value!

[Alsee]

And to syngerize stuff!

-

It's about the power, stupid.

[randyest]

Too bad the summary had to mention "instant-on PC" because most of the responses so far are about "No way" or "my kewl macbook does that already" and such. The biggest advantage of MRAM over SRAM/eDRAM is not that it retains data without power (though that's nice too), it's that MRAM about as fast as current eDRAM and half the power. And even zero power when not used (while retaining most recent data) is a bonus.

This is a huge plus for ASICs and other chips (ASSPs, COTS, etc.) that have a lot of memory on them (which is most of them.) It allows more memory on a chip without expensive packages/die sizes for thermal management or complex, time-consuming power management systems. LSI (large-scale integrated) circuits use a lot of memory, and power consumption is a huge problem, so cutting that in half will enable a lot of products to be made that wouldn't have been possible/affordable before, and a lot of other products will get to market faster.

MRAM has been around for a while, but the relatively slow speed made it unsuitable for most applications. Now it will be great enabling technology that will ripple through many products that use semiconductor devices.

Re:

[petermgreen]

How compatible is it with high speed processes? Most higher speed FPGAS use volatile configuration (usually loaded from either a computer system or from a serial flash chip on startup) because flash technology puts a lot of nasty constraints on your semiconductor process,

Re:

[randyest]

My understanding is that this MRAM can be created in the standard ASIC process. Not sure about FPGA, but I believe they're similar.

Re:

[mrv20]

Exactly. Assuming that the read/write/idle power levels of these cells are comparable to the SRAM/eDRAM they replace then the technology will be a great help in low power IC design.

Currently if you power-gate part of a core you have to worry about how to avoid losing any state information (often involving writing it out to off-chip memory or to shadowed, non-volatile storage. If all your embedded RAM is non-volatile MRAM then you can turn the power-gate the module as easily as clock-gating it, and automat

Re:

[repvik]

Step two is superflous

Write speed

[owlstead]

These things have really high write speeds and are non-volatile. So they will be used at places where write access should be really fast, and the need for reliability and persistence will be high. Because of size considerations (we are talking about multiple Mb per chip here, not Gb) I cannot see them replace flash soon. What I can see is the use in devices that don't need too much memory, but do need speed.

Are there any plans to use this memory as a cache for (solid state) disks? It seems to me that it mig