Hard disks store information magnetically, but typically have access times in the range of 2 to 8 milliseconds. That means that when a drive is asked to produce information from a certain sector on the disk, it takes somewhere between 2 and 8 milliseconds—depending on the rotation speed of the disk—for that portion of the disk to be spun around under the disk's magnetic head so that the data can be read or written. This latency leads some people to create "RAM disks," using a computer's RAM, which has much lower latency, to create virtual disk. Since the memory doesn't suffer from the comparatively long latency of magnetic disk, RAM disks are typically much faster than real disks. The drawback is that this memory is electrical, not magnetic, so when the power is shut off, everything in memory is lost.
Now Freescale Semiconductor is introducing MRAM, Magnetoresistive Random Access Memory, which they claim combines the speed of RAM with the non-volatility of disk. Information is stored magnetically, so a loss of power doesn't mean a loss of data. Without moving parts, though, they can eliminate the normal latency associated with hard disks.
Freescale says that the chips are now in volume production, and that they have customers already lined up, but they've declined to give any names. Right now the chips are considerably more expensive than either hard disks or conventional RAM, reportedly at $25 for a 512KB chip. Still, as MRAM is much faster and more durable than Flash memory, it will probably make inroads in embedded markets first before eventually spreading into other applications.
Their first product, part number MR2A16A, is designed as a replacement for battery-backed SRAM products.
Freescale was created from the semiconductor product sector of Motorola in July, 2004, and supplied Apple Computer with PowerPC chips until Apple's recent transition to Intel CPUs earlier this year. They have been building these chips for two months to build up stock, and they are shipping them beginning today.
I recently bought a 1GB Compact Flash card for my camera for $31 shipped. That's 2048 times the capacity of this chip, at a price not much higher. However, this chip is reportedly much, much faster for both reading and writing, and also will not degrade with time as Flash memory tends to do. Still, the price will need to come down considerably before it is ready for sale direct to end users, which is undoubtedly why Freescale is aiming at industrial consumers first.
Article comments
1 - Phillip Winn
I've seen a few people describe this as a major paradigm shift, an entirely new way of doing things, etc, etc, but I don't think I buy it. It's like a really fast hard disk more than it's like RAM at all. The fact that the data is stored persistently just means that now devices using this memory will have to manually erase it!
Or, yeah, they could just a leading bit and so on, but still. It's like disk, only smaller and more expensive for now. Come back in a few years when it's much, much cheaper.
2 - Amit shouldinel
Samsung is way ahead on this one.
3 - J Gruszynski
This very exciting for me personally because I first started working with MRAM back in 1985. I worked for Aerospace Corporation and was involved in new technology development for military space systems. It was in large part my research that caused Honeywell got funded to develop the first MRAMs for the military. Why did we care about MRAMs? In general semiconductor memories were inferior to magnetic memories for radiation hardness. There were plenty space systems still service use plated wire and core memory at the time. Though the semiconductor part of MRAM were still susceptible to upset, the memory material itself wasn't.
Will this news make a big difference, a revolution? From my perspective of being deeply involved in non-volatile memories (NVM) since the early 80s, I think not. Most people don't run into endurance limits of flash so there's no pain to be soothed. There are possible options for replacing harddisks in ruggedized applications and for fast boot applications. The problem is that MRAM doesn't exist in a vacuum. Hard disks (and even flash) are continuing to improve (e.g. Seagates perpendicular recording drives) and still offer a more attractive price-performance point for most applications.
A close relative of MRAM gives us an example of why MRAM might not take over hard disk: magnetic bubble technology. This was another one of our "eggs in the basket" back in 1985. Bubble held great promise to take over the applications occupied by hard disk but it simply couldn't keep up in terms of development and capital investment. One advantage over bubble that MRAM has is that it is relatively less exotic than bubble was: it's mostly a silicon process with added steps to handle permalloy or similar components. Bubble required a garnet substrate with permalloy patterning. On the plus side, permalloy isn't too different from other transition metals currently required for nanometer scales but it is still different. The one thing that could doom MRAMs is not technological; if Freescale remains the only source of the technology, I'd short it. Still I'll probably try to get a sample to play with.
This brings up another interesting point - the 20 year rule of innovation. The rule (akin to Moore's Law - it's more a rule of thumb than a law) states that it takes 20 years from first conception and prototypes to commerical viability and profit. It's been 20 years almost exactly for MRAM. The internet is another famous example. Another is the Seagate perpendicular recording drives mentioned above.
4 - Phillip Winn
Amit, Samsung is not shipping, and Freescale is. So Samsung is not in any way "way ahead" on this one.
5 - Phillip Winn
J, I wonder how operating system development might have been different had MRAM exists twenty years ago. As it is, computers take a long time to boot, loading things from disk into memory and so on. I use VMware, and I'm intrigued by the idea of pushing a button and flipping back to a known good state almost instantly. *Push* -- and I've got an almost-virgin install of Windows 2000, fully updated.
Of course, I have to boot up my *real* Windows 2000 in order to get to that push-button point, but with MRAM, and a BIOS intelligent enough to deal with it, I could turn on a computer and have my DRAM almost instantly populated from a known-good state from MRAM.
In any case, I agree with you overall. I think that the time when MRAM could have changed computing as we know it has likely passed, and we're left now with niche applications and hyperbole from marketing droids and journalists who don't know any better.
6 - David McCabe
If I'm not mistaken, it is IBM that has supplied Apple with PowerPC chips the last several years.
7 - Phillip Winn
David, you are mistaken, but it's an understandable mistake. Motorola and IBM jointly developed the PowerPC, and IBM continues to develop it for their own use, but Motorola (Freescale, after the division was split off) was the primary supplier for Apple. Freescale's inability to deliver processors Apple needed did force Apple to turn to IBM for some shipments, but the primary supplier was still Freescale.
Now that Apple gets their processors from Intel, both Freescale and IBM are out of luck.
Sources: One, two, three.
8 - Ralph
not really its dual core so although they use intel chips they also use the same chips as before too.
By the way in my opinion Apple is not very good
9 - Phillip Winn
If that comment made any sense, that would be better.
"Dual-core" doesn't mean that both a Freescale-produced PowerPC and an Intel chip live in the same processor. It means that two Intel cores live in the same Intel processor -- no PowerPC involved at all.