MRAM Contends for Green Memory Title
Magnetoresistive random access memory, aka MRAM, has been around for nearly 20 years, but it's just now beginning to be seriously considered for wider use. MRAM is non-volatile, but its performance and density are on par with volatile memory technologies. As a result, it's one of the more promising candidates for the holy grail of storage: universal memory.
|Everspin 64Mb MRAM chip|
MRAM is also among the greenest of memory technologies. Its non-volatility comes from the way it stores bits -- magnetically, rather than as electrical charges. Since no constant electrical current needs to be applied, MRAM devices can be designed to draw very little power in general use, and can be turned of completely when its data doesn't need to be accessed. Its also requires much less power to operate than NAND flash, the traditional technology used for non-volatile memory in thumb drives, handheld device storage, and solid state disks (SSDs).
The advantages of non-volatility aside, MRAM reading and writing is not particularly power-efficient compared to volatile memory. Reads, for example seem to require about the same amount of power as DRAM, the common technology used for main memory in most computers. Writing requires still more power, at least for the earlier iteration of MRAM using the so-called "toggle switch" technology. The newer STT (spin transfer torque) technology, sometimes just referred to as ST (spin torque), is more efficient for writes, using about the same power as reads, and has the additional advantage of allowing for denser, smaller circuitry. It does however rely on electric current to switch the magnetic state, which ultimately limits energy savings.
Other memory technologies like NAND flash can offer the same energy savings associated with its non-volatile nature, but MRAM has a huge advantage: its performance doesn't degrade with use. That is, it has virtually unlimited endurance, regardless of the volume of reads and writes it must contend with. That makes it a potential replacement for main memory and cache, as well as external storage. In fact, some of the earliest uses of MRAM are in cache replacement schemes, where it is used to substitute for battery-backed SRAM. Although MRAM is not quite as speedy as SRAM, its ability to keep its data intact without power can be a huge advantage, especially for the on-board computers that don't have the luxury of a wall socket connection.
At this point, transistor density, which is a proxy for cost, is probably MRAM's weak spot. DRAM is still considered the highest density memory around, making it the cheapest, and thus the most widely used. And since just about every computer, from smartphones to servers need gobs of it to keep up with the ever-increase influx of data, manufacturers are not likely to kick the DRAM habit anytime soon.
Also, there's usually a trade-off between performance and power consumption, and MRAM is no exception. Faster switching speeds require more power, even for magnetic devices. But Toshiba recently announced a variant of STT that improves memory speeds, while "reducing power consumption by 90 percent." As a result the company thinks it could even be more energy-efficient than SRAM cache. The technology is still in the lab, but Toshiba is already eyeing the smartphone and tablet market for a commercial play.
Physikalisch-Technische Bundesanstalt, a German research institute funded by the government, is also looking into boosting MRAM performance. Some recent work there involving an invention, known as "ballistic bit triggering, is able to boost data rates five-fold -- from about 400 megabits/second to 2 gigabits per second. The researchers say that the invention enables faster MRAM clock rates, which "can compete with those of the fastest volatile storage components." Power consumption is likewise reduced.
UCLA is also working on an energy-efficient version of MRAM known as MeRAM. According to researchers there, the device is 1,000 times more efficient than commercial versions. It appears to ditch the STT model entirely, instead relying on voltage to write the data.
According to the UCLA press release, by eliminating the reliance on electrons, as with STT, the technology "generates much less heat, making it 10 to 1,000 times more energy-efficient." Conveniently, the design can be five times denser, which suggests cost per bit will also improve. DARPA is helping to fund the work, so it's still very much still a research project. But given that Hitachi Global Storage Technologies and Singulus Technologies are also involved in the effort, the technology could eventually make its ways into commercial production.
The good news is that MRAM technology is already a commercial concern. Toshiba, IBM, NEC, Micron, Renesas Technology, Samsung, Hynix, Spingate, Cypress Semiconductor, Hitachi and Crocus Technology have all dipped their toes into the MRAM waters over the last decade or so, although not all have taken the full plunge. One company that seems especially serious about magnetoresistive memory is Everspin Technologies, which dubs itself "The MRAM Company."
Everspin, which spun off from Freescale in 2007, taking with it all the MRAM patents and expertise, is focused entirely on the technology and offers an array of commercial products today. The company sells devices for both serial and parallel interfaces and in densities ranging from 256Kb up to 64Mb. As of last year, the company claimed 300 customers and more than 100 products. In 2011, the first year they began shipments, they delivered over 2 million units.
And unlike many of the early versions of the MRAM, which targeted mobile and embedded computing, Everspin is also looking at the enterprise market. The company's latest product, its 64Mb DDR3 ST-MRAM, is aimed at complementing SSD setups in datacenters. In general, they they're looking to move the product into cloud storage, in-memory computing, and RAID storage systems/appliances. Everspin has been shipping the devices since November 2012 and is expected to announce broader availability sometime later this year.
All around, it looks like MRAM performance and power are starting to reach a happy equilibrium in commercial products. If some of the latest research pans out and density could be improved to the point where it could compete effectively with DRAM, MRAM could reach critical mass quickly across a broad range of applications.