New MRAM tech promises higher densities, lower power

Researchers at UCLA’s Henry Samueli School of Engineering and Applied Science have made a discovery that can dramatically lower the power consumption and increase the bit density of magnetoresistive random access memory, otherwise known as MRAM. In existing MRAM implementations, a process called spin-transfer torque is used to write to the memory. This technique relies on an electrical current to move the electrons responsible for representing data. The current generates heat, and it limits how densly bits can be packed.

Instead of using an electrical current, the UCLA researchers have been able to flip MRAM bits using voltage. They’ve dubbed their creation MeRAM, or magnetoelectric random access memory. The press release explains how it works:

MeRAM uses nanoscale structures called voltage-controlled magnet-insulator junctions, which have several layers stacked on top of each other, including two composed of magnetic materials. However, while one layer’s magnetic direction is fixed, the other can be manipulated via an electric field. The devices are specially designed to be sensitive to electric fields. When the electric field is applied, it results in voltage — a difference in electric potential between the two magnetic layers. This voltage accumulates or depletes the electrons at the surface of these layers, writing bits of information into the memory.

MeRAM is reportedly 10-1000 times more energy efficient than traditonal MRAM, and it’s supposed to have five times the bit density. Because MeRAM is non-volatile, it could potentially replace the flash memory used in solid-state drives, tablets, and smartphones. Best of all, there are no endurance limitations attached.

Magnetic-based memory isn’t just confined to research labs. Everspin is current sampling (PDF) a 64Mb MRAM device that promises bandwith of 3.2GB/s in a 16-chip configuration—that’s DRAM-like performance. Much higher densities will be needed before MRAM becomes a viable replacement for flash in consumer-grade devices, but MeRAM might get us there sooner.

Comments closed
    • Anonymous Hamster
    • 7 years ago

    At one level, this seems vaguely like hard-drive technology, except you have a read/write head per bit.

    At another level, this seems like magic (like nearly all advances in microelectronics do these days).

    • Tristan
    • 7 years ago

    We will see MRAM in Excavator produced by Samsung, after buying AMD.

    • HibyPrime
    • 7 years ago

    According to [url<]http://semiaccurate.com/2012/11/16/everspin-makes-st-mram-a-reality/[/url<] these MRAM chips are built on a 90nm process. Just a straight die shrink should get capacities up a fair bit higher. Semi-accurate says that you should get ~16x the density at the 'bleeding edge' process (22nm?) so we're looking at 1024Mb chips or 2GB per standard DRAM stick.

      • Airmantharp
      • 7 years ago

      That’s pretty dense- and memory can usually be spun to run on a smaller process than logic. That will take someone who really wants to push the nodes, but the promise that memory like this holds might just be the motivation needed.

      That said, at modern node sizes, 2GB modules wouldn’t be that bad at all; and who says they need to be on DIMMs (not that I’m not thankful for the comparison), but with the speeds listed above, it sounds like this technology might be more suited for PCIe cards. Which would be awesome.

    • Grigory
    • 7 years ago

    At this point I would like to see ANY technology actually take on Flash in the consumer market and not only for niche applications.

    • alloyD
    • 7 years ago

    Still holding out for [url=http://en.wikipedia.org/wiki/Memristor<]memristors[/url<].

      • MadManOriginal
      • 7 years ago

      lol, HP :*(

        • NeelyCam
        • 7 years ago

        That’s my biggest worry – the board will cancel the program and refuse to sell/license the technology

          • NoahC
          • 7 years ago

          They won’t cancel anything without first cutting off the program’s resources and letting it flounder for a few quarters to justify its cancellation. 😉

    • ludi
    • 7 years ago

    Interesting in theory…but let’s see it happen. At the present rate of press releases, I expect to see MRAM in the control module for my long-range electric car with an onboard fusion engine.

      • bcronce
      • 7 years ago

      edit: I think I mixed up MRAM with ReRAM, but I’m leaving my original post. I hate it when people change posts on me, so I don’t do it to them.

      Several memory and storage manufactures already announced MRAM being slated for around 2014/2015.

      They’ve already poured lots of money into their fabs to ramp up for for mass-production; I would hope this isn’t vaporware for their stock-holder’s sake.

      • Helmore
      • 7 years ago

      What about it’s use for my FTL space ship?

    • just brew it!
    • 7 years ago

    This was already posted in the Shortbread!

      • chuckula
      • 7 years ago

      The TR editors are known to make a more in-depth posting using a story selected from the Shortbread from time to time.

      • Arclight
      • 7 years ago

      Like many other news articles before…….i presume this happens only for the most important news

      • DPete27
      • 7 years ago

      Couple days behind [url=http://www.fudzilla.com/home/item/29839-ucla-researchers-come-up-with-new-energy-efficient-memory<]fudzilla[/url<] which had the same article posted on Monday.

    • Symmetry
    • 7 years ago

    What’s the write endurance look like? Traditional MRAM doesn’t have write endurance issues at all, but they’re expecting STT-MRAM to have a write endurances in the millions or billions range, the same as RRAM. While this is certainly in the Not An Issue range for using it as storage, it’s something that bears worrying about if it’s going to be used as main memory.

    I expect that using less power would tend to mean more endurance, but I’d still like to hear something from the researchers.

    RRAM or STT-MRAM will clearly be coming first and if we’re going to have a unified memory/storage revolution they’ll be in the vanguard. But I guess it’s possible this stuff will be the second wave.

      • radializer
      • 7 years ago

      You’re correct … most published STT-MRAM reports are talking about 1e12 type numbers which would be needed (as you surmised correctly) if this were to be used as some form of main memory. If you have the access, you should look for papers on the IEEE site – there are many companies publishing such results (Samsung, Toshiba, Fujitsu, Spintec, Grandis, etc).

      I’m not sure how this would translate to MeRAM – it would depend on the stack physics and what they plan to do with the spin filter layer.

      • NoahC
      • 7 years ago

      Funny to think we’ll soon be talking about the best way to secure erase main memory 😉

      But seriously, bring on the unified memory/storage revolution!

    • MadManOriginal
    • 7 years ago

    Coming to a product near you, in 5-7 years!

      • LostCat
      • 7 years ago

      No kidding. ReRAM will beat it to market since actual companies are working on it right now.

      • bcronce
      • 7 years ago

      [url<]http://forums.xkcd.com/viewtopic.php?f=9&t=98859[/url<] This is from a month ago. You can already purchase... 64MB.... sticks of DDR3 MRAM, but it was announced for MRAM to go mainstream in 2014, so give it another 1-2 years.

        • MadManOriginal
        • 7 years ago

        Well well, that is pretty huge. BUT it’s been in development since the ’90s, so the 5-7 years joke still holds up, hurrah!

Pin It on Pinterest

Share This