A Magnetic Memory Alternative to Hard Disk

Dr Occult writes "Finally, a magnetic memory chip has been manufactured in volume and released by the U.S. company Freescale. Christened MRAM (magnetoresistive random-access memory),this chip will hold information even after power has been switched off. From the BBC news article: 'Unlike flash memory, which also can keep data without power, Mram has faster read and write speeds and does not degrade over time,' and 'MRAM chips could one day be used in PCs to store an operating system, allowing computers to start up faster when switched on.'"

Freescale's PR

[austinpoet]

Information about the product from the Freescale Press Release [freescale.com]

Price?

[bookemdano63]

It will be a while before they get their $25 / 4 megabit wholesale price to anywhere close to reasonable.
http://online.wsj.com/public/article/SB11524917130 4801944-v71_ITCad7JIwzqJZ_nfN_pacDg_20060809.html? mod=tff_main_tff_top [wsj.com]

Re:NOT a hard drive alternative

[TransEurope]

_Today_ they are larger. But tomorrow Freescale

plans to shrink their new chips (29nm) under the

scales of the future standard 6T-SRAMs (still 45nm).

http://www.heise.de/newsticker/meldung/75243 [heise.de]

Re:When?

[cyfer2000]

From what I am remembering, Freescale began small batch sampling of this kind of 4Mbit MRAM two years or 1 and half years ago, and now a available in large batch.

Re:more vaporware

[TransEurope]

First: freescale aims on the usage of MRAM in embeddet devices and microcontrollers. There will be no MRAM-Harddisk next month in the shops. 2nd: There is not only Freescale. Micromem will produce MRAM-Chips for the Aerospace industry. And IBM/Infineon already have an 16-MBit-MRAM-Chip since last year. There are also Renesas/Toshiba in the race. It's a completely new tech, you heard about years ago, when the first theories about mram came from the labs. But such a thing needs everytime many years to go to the serial production lines.

Old news

[dpaton.net]

Freescale's MRAM technology isn't all that new...it's an old Motorola technology that they kept running with when they were spun off. It's taken them a few years to get going again, but it's already been done [spacedaily.com] for a while.

That said, MRAM ain't a HD replacement yet. No one outside the aerospace industry is using it for storage right now that I'm aware of, and even if someone was, making a large enough FRAM based drive with 4Mb chips is HARD. 2 chips for every MB. 2048 chips for every GB. a 500GB FRAM disk would require 1,024,000 of these chips, requiring nearly 2,500 sqft of PCB space, and more power than a pile of overclocked P4s (~9mA * 3.3V * 1,024,000 chips = 30.4128kW at IDLE). Even if someone could build that, it'd be farking huge, run inconcievably hot, be incredibly power hungry, and sell for an obscenely expensive price, even for the most extr33m gadget hunters.

Wait for 32 and 64Mb chips. Then we'll talk.

Right now I'm too busy working with a serial FRAM from Ramtron [rantron.com] to write more.

Re:more vaporware

[Yusaku Godai]

Did you RTFA? This is the first commercial MRAM product which is being produced in volume. They have customers for it, and they've already built up a stock of the stuff. Can't call it vaporware if it supposedly actually exists somewhere and is ready to be shipped.

Re:more vaporware

[barawn]

We've been hearing vaporware mram chip stories for almost a decade now... When is it going to be on the market for people to purchase and use?

Now, apparently. That's what this story is about. Here's [freescale.com] a link to the actual chip's spec sheet. Here's [freescale.com] a link to the chip's page on Freescale, where you can order it for $25/chip in 1000 unit quantities.

It's not in any consumer products yet, no, but it is available to purchase, which means it isn't vaporware.

512kB chip: $25

[Demon-Xanth]

http://www.freescale.com/webapp/sps/site/prod_summ ary.jsp?code=MR2A16A&srch=1 [freescale.com]

"The MR2A16A is a 4,194,304-bit magnetoresistive random access memory (MRAM) device organized as 262,144 words of 16 bits"

Not ready for PC time yet.

Re:Old news

[Anonymous Coward]

I assume you meant Ramtron [ramtron.com], not the misspelled "rantron" in the URL.

Re:Price?

[barawn]

That is a reasonable price. Right now, MRAM replaces battery-backed SRAM: like the ZEROPOWER [st.com] series from ST Microelectronics, and a 4-Mbit version (M48Z512A) costs $45 in quantity, and the MRAM chip won't take up huge real estate with a gigantic DIP package.

At $25 in quantity for a 4-Mbit chip, it's about a factor of 5 higher than conventional SRAM. I'd guess that a factor of 5 in cost reduction isn't crazy to expect.

Too bad this chip didn't come out say, five to ten years ago - otherwise you likely would've been seeing it in video game cartridges for a while now.

Article misses the point almost completely...

[Gadzinka]

I am not really suprised, that no one bothered to google for MRAM, not even tried to look it up in WP [wikipedia.org]. What's missing in the article and most of the comments is that MRAM is one of those holy grails that most of the industry is chasing, because it promises great returns on investment. Basically MRAM (theoretically) can be:

• as fast as SRAM (i.e. cache in your processor)
• as small (i.e. as hight density) as DRAM; single MRAM memory cell is two magnets instead of two conductors of capacitor in DRAM, but the (theoretical) size is of the same order of magnitude
• non-volatile like Flash, but with random access and orders of magnitude faster, w/o "write penalty" and w/o erase/write cycles limit
• much less energy-hungry than SRAM, DRAM and Flash while working; when not working it can keep information at least as well as Flash

It's in development since the eighties and it will take time before we "get there" but it is possible, that one day MRAM could replace cache, main memory and memory cards in our computers.

When? I have no idea, but AFAIR transistors didn't get from prototype to 65nm in a decade. Hopefully engineergs and managers in some semiconductor companies have longer attention span than an avarage slashdot reader.

Robert

Still pretty small

[wonkavader]

Currently, it's 256K x 16-Bit

Here's the datasheet link: http://www.freescale.com/files/microcontrollers/do c/data_sheet/MR2A16A.pdf [freescale.com]

Re:Vaporware

[Kythe]

As was noted above, this is actually being produced, so by definition it isn't vaporware.

The problems MRAM could address are very real, and people have been working on using MRAM/GMR-based memory for a long time for that very reason.

Re:Vaporware

[1stpreacher]

I don't get the vaporware argument here... They're being SOLD... But aside from that, I see MANY uses for this - first in cell phones, according to this [wikipedia.org] "In February of 2006, Toshiba and NEC announced a 16 megabit MRAM chip with a new "power-forking" design. It achieves a transfer rate of 200 MB/s, with a 34 ns cycle time - the best performance of any MRAM chip. It also boasts the smallest physical size in its class -- 78.5 square millimeters..."

so we're looking at 'about' 3 inches for 16meg (in this case) ... I'd LOVE my phone to be able to use it's memory more quickly, and to be able to sync more quickly with my pc. And that's just the first thing I could think of...

Re:It doesn't quite seem to be (Re:) more vaporwar

[barawn]

On the other hand - sorry, I don't quite see how these will be in competition with hard-drives

Right now, they aren't. At that price point, they're competing with battery-backed SRAM (very nicely for the integrated stuff, and it depends on the product for the battery + battery monitor chip + SRAM solution).

Short primer on different memory technologies: SRAM is very fast, very low power, easy to interface, but it needs a battery for data integrity. DRAM is very cheap, but higher power, much harder to interface, and needs not only a battery but a controller for data integrity. EEPROM is everything SRAM is, but nonvolatile but is expensive, and writing requires awkward voltages, can't be done bytewise, and is slow anyway. Flash is similar to EEPROM, but beats it because it's cheaper, doesn't require weird voltages, but writing is still awkward.

The only downside to MRAM currently is its cost - fast and easy reads, fast and easy writes, nothing required to maintain it, and low power to boot. If this becomes available at distributors within a factor of 2 of that $25 price point, there's literally no point to integrated battery-backed SRAM chips anymore. If the price drops by a factor of 2 or so, there's no point to battery-backed SRAM at all.

As the price drops, though, MRAM has the potention to challenge all of those technologies above, as well as hard drives, much like flash is starting to do now.

Re:I bet it's hard to make it very small

[Kythe]

Since magnetic fields from the electric current in write lines decrease as the square of the distance from the lines, you don't have to worry all that much about crosstalk at write time, either.

Oops -- slaughtered that one, didn't I? It's been too long since I looked at the equations. Straightfoward Ampere's Law: the decrease is linear, not with the square of the distance.

Still, the decrease is significant enough, and the resistance to switching state high enough, that you don't generally have to worry about write lines inadvertently flipping more than one bit.

Re:NOT a hard drive alternative

[harrkev]

It would be extremely expensive to create an "MRAM hard drive". This is just more pump and dump for Freescale daytraders.

Bzzzzt. Wrong. Thank you for playing.

OK. You are half-right. It would be expensive to crate an "MRAM hard drive." So, getting 20 gigs of MRAM would cost a small fortune. But this is NOT a "pump-n-dump." This is really cool stuff. I can easily imagine some embedded systems that could really use this stuff. This is non-volatile system memory. The problem with FLASH and EEPROM memory is that the cells wear out after a lot of writing (somewhere in the neighborhood of 10,000 to 1,000,000 write cycles will give you trouble). For some applications, this is not enough, so you have to resort to battery-backed SRAM. Now there is at least another option.

• Do you need to store data without having a constant battery backup?
  
• Do you need to store a relatively small amount?
  
• Will the data be changing rapidly enough to kill a FLASH in short order?
  
• Would a hard drive be too big or too power hungry?

If you answered "yes" to all those questions, MRAM might be for you:

Although, really, this seems to solve the exact same problem as Phase-Change RAM [wikipedia.org].

Re:Is bootup time really that big of an issue?

[fyngyrz]

Does anyone have any non-fluff stuff about wha power consumption, max transfer and the like is?

35 ns cycle time for read or write (about 28.57 MHz), read modes 50 ma to 80 ma max, write modes 105 ma to 155 ma max, 9 ma to 12 ma max for stanndby (no pins changing state) and 18 ma to 28 ma with pins flying but no selection enabled for the chip. This is with a 4 mbit chip organized as either 8- or 16-bit. Couldn't find a spec for "the like", you'll have to be more specific. :-)

Those specs were abstracted from the PDF data sheet easily found at on this page. [freescale.com]

Re:Old news

[dpaton.net]

That would be right on, but I quoted the CMOS sustained idle current. Write current jumps to 155mA, and read current is 80ish as I recall. Keeping the chips alive but idle is the easiest way to design the system. Switching the power to the chips required to store a chunk of data would require knowing the length and width of the memory required, and then knowing what blocks are free, and then powering on the required chips (with a huge current spike and associated noise), and then making the write. Designing the power switching infrastructure on that many sqft of PCB (damn TSSOP packages) would be impossibly problematic. The design of the chips isn't supportive of anything large scale...these are indeed for cellphone scratch pad use, or for NV storage in other small devices.

And remember, my numbers were only for a 500GB x 1 byte array. That's horribly inefficient. If they can bump the width and depth of the array up, then we can talk. Let's hope they scale it fast and well.

Re:Back to the past....

[WebCowboy]

MRAM is in some ways a modern take on 1960's era "Core Memory" technology. There are similarities between both, however core memory was not semiconductor-based--it was a plane of copper wires woven together with little ferrite rings strung on where wires intersected. As such is is pretty low density: 16 Kbit of core memory took up 250 cm^2 of area. With MRAM the method of operation is the same and it also involves reversing polarity of magnetic fields. However there are no ferrite cores; MRAM consists of a sandwich of conductor grids around memory cells. Like with core memory an entire row of a grid can be written to in one operation--you charge one "row" line on the write grid and all the columns you want to flip and they all change at once.

Reading MRAM is simpler than core memory becasue core memory had no read operation--it had "flip to zero" and "flip to one" and a "sense" line--the sense line would emit a pulse if a core element changed state. To read core memory, you had to do a "flip to zero" and watch the sense line--if it pulsed then a one was in the cell and you had to do a "flip to one" to restore it. If there was no pulse then it was already zero. With MRAM reading simply involves measuring the resistance of the insulating layer of a memory cell (the insulating material has the property where resistance increases as the magnetic field passing through it increases). IIRC there is nothing preventing parallel reads either. MRAMs are also much denser--megabits can fit in 0.25 cm^2

The "MRAM hard drive" thing may be hyperbole right now, but it looks like development of MRAM rechnology is significantly outpacing Moores Law. MRAM is also potentially as fast as SRAM and as dense as SDRAM--without the need for refresh circuitry so designs can be greatly simplified. Further downsizing could make it a good flash replacement. The biggest hurdle could be reduction...

Comment removed

[account_deleted]

Comment removed based on user account deletion

Re:NOT a hard drive alternative

[drinkypoo]

MRAM doesn't need to be a hard drive alternative; that wasn't even suggested in the summary. It only needs to make *boot-up* near instantaneous, holding enough (10 gigs?) to get the luser to login prompt and then desktop.

The fact that you would claim this gig crap proves that you don't have any idea of the issues involved here. Are you a troll, or just posting as AC so that no one calls you on your shit?

If you just replace your main memory with MRAM, then you can hibernate without writing out the contents of memory to disk. Think of it as a suspend-to-ram mode on steroids. Because the memory is not cleared, you can just restore processor state, pick up from where you were, reinitialize drivers, and move on with life. Just like coming back from hibernation mode.

This technology also opens up the possibility to use hard power switches on more systems, or low-power electronically switched relays. You don't need to maintain power to the device and run the power supply just so you can keep one memory chip going to retain settings.

Unfortunately

[Flying pig]

It lost its data the moment you read it if the read/modify/write circuitry failed. Anyone remember the PDP-8, whose accumulator cleared when you read it, presumably so if it was implemented in core, there would not be a wasted rewrite cycle if you didn't need the accumulator data again? Ah, the fun of early machines...and you could even use them in IBMs, which is more than you can do with a P4.

Re:I was alive in the 1980s

[goombah99]

I worked on magnetic bubble development before IBM canceled their program. it was a matter of finding a commercial niche.

Magnetic bubbles did exist and were sold and used in computers. But at the time their was no niche for them like their is flashram. bubbles were faster than disks but more expensive and slower than ram but cheaper. Thus they got caught in a squeeze play. Although they consumed no current when off they were not particularly low power devices so they were not suited for battery powered devices. It's the latter that allows flashram to get a commerical foothold around which it has matured.

Mram is supposedly going to be faster then ram and consume less power. So it too may have a niche that eluded bubbles. it's main competitor is not flash or disk but ram I think.

Re:NOT a hard drive alternative

[Anonymous Coward]

"somewhere in the neighborhood of 10,000 to 1,000,000 write cycles will give you trouble"

Industrial Compact Flash is >2,000,000 read/write cycles. And some advertise 3,000,000 cycles. http://www.psism.com/industrialcf.htm#industrialCF [psism.com]

Re:Flash FS. Hm. How 'bout something actually used

[julesh]

How much of a myth is it when you use least common denominator, portable file systems like FAT?

A decent flash disk will have write-spreading as a layer on top of the filesystem, so it will remap sectors on the fly to avoid wearout.

Re:Found the datasheet...

[stefanb]

Two clicks away from freescale.com homepage [freescale.com], instead of your nice ad-loaded page, and with quite a bit more info.