Forgot your password?
typodupeerror
Data Storage Hardware

New Memristor Makes Low-Cost, High-Density Memory 86

Posted by timothy
from the call-doctor-if-effects-last-more-than-four-hours dept.
KentuckyFC writes "A group of electronics engineers have discovered that a thin layer of vanadium oxide acts as a memristor, the fourth basic component of circuits after resistors, capacitors, and inductors that was discovered last year. At a critical temperature, a current passing through the layer causes it to change from an insulating state to a metal-like state, thereby changing its resistance (abstract). The effect lasts many hours — which is what makes the layer a memristor (a resistor with memory). The team says this could be scaled up to make resistive random access memory, or RRAM, at very low cost, from little more than layers of vanadium oxide."
This discussion has been archived. No new comments can be posted.

New Memristor Makes Low-Cost, High-Density Memory

Comments Filter:
  • "The effect lasts many hours â" which is what makes the layer a memristor (a resistor with memory). The team says this could be scaled up to make resistive random access memory, or RRAM, at very low cost, from little more than layers of vanadium oxide.""

    If this was going into a SSD then it would have to last longer than that.

    • Re: (Score:3, Insightful)

      by pines225 (1413303)
      The article suggests use as resistive RAM rather than a solid state drive. As long as it doesn't need to be a permanent memory element it might be possible to refresh periodically on a schedule that's safely less than the lifetime of the state transition, i.e. boost the phenomenon every hour or two. Shouldn't cause much of a power hit.
      • The memristor is is just a way to model nonlinear circuit elements and is one of many components in a nonlinear expansion for circuit modeling. See this paper [ieee.org] by Leon Chua, the memristor's inventor. Note that in this paper the fourth element of the four element torus is negative resistance and not the memristor. All of the publicity over the memristor has been (sucessfull) marketing by some researchers at HP. .

        From the talk page for the memristor on wikipedia [wikipedia.org]

        "Resistance, Capacitance and Inductan
        • Re: (Score:2, Informative)

          by Anonymous Coward

          You sir are a troll.

          Nobody is talking about resistance capacitance and inductance when talking about fundamental passive circuit elements.
          There are four and only four passive circuit elements because they are defined in that paper as being the result of a relation between two of the four Fundamental Circuit Variables that are Voltage, Flow, Charge and Flux. So you can combine them in only 4 possible ways: The Resistor, The Capacitor, The Inductor and The Memristor(charge and flux).

          So stop pretending thats a

          • by sup2100 (996095)

            And you sir are a bigger troll. The paper I cited was not the original paper, but rather a follow up that puts the memristor in a better context. If you allow for various combinations of the derivatives and integrals of just current and voltage you can get a capacitor, resistor, inductor, and memristor. However you can also get the memductor and memacitor as well. This is all explained in that paper.

            Why are combinations of voltage, current, charge and flux fundamental?

            According to Websters fundamental i

          • by Teilo (91279)

            Actually, I would like to understand this point as well, so I'm not sure about all this "HP Conspiracy" stuff. Sounds like an honest question to me.

            If components were ever discovered that could couple Flow and Charge, or Flow and Flux, why would those not be "fundamental"?

            Why was there only one missing circuit element? Why not three?

    • by drinkypoo (153816) <martin.espinoza@gmail.com> on Sunday January 11, 2009 @10:02AM (#26407015) Homepage Journal

      Not really; if you only have to refresh it once every four hours, you can do that for a LONG time on a battery. What I really see as the potentially big win is if they can get the speeds up to SRAM levels. Technically it would be a kind of DRAM, but if you only have to refresh it once every four hours, then just using a LRU scheme when allocating memory might mean that in practical usage you would never need a refresh anyway.

      • by MathFox (686808)
        For a memory device that operates near in a 335-340K (65 C) temperature range you need a power budget for heating...

        On a second thought, if one taps into the excess CPUs and GPUs heat. Still a bad fit for a mobile device though

  • by Baron_Yam (643147) on Sunday January 11, 2009 @09:25AM (#26406885)

    I thought the more exciting announcement was that memristors could be tripled up to create transistors that were (despite being tripled up) still much smaller than a standard transistor.

    Then, there were bits about them supporting more than just binary states, which would increase complexity and density yet again.

    Denser memory may be the first pratical consumer product, but if the other possiblities work out, I'm pretty sure that memory will also be the least significant.

  • by trolltalk.com (1108067) on Sunday January 11, 2009 @09:25AM (#26406889) Homepage Journal

    " At a critical temperature,"

    "Gee, I had it stored on this memsistor chip - but I left it in my shirt pocket, and my data melted."

    The article doesn't say what temperature, so there's probably an issue there. Until that issue is solved, it's about as useful as write-only memory.

    Also, looking at the required voltage (50 volts @ 0.6 amp), this is NOT going to be either high-density, or portable,or particularly energy-efficient.

    • Re: (Score:1, Interesting)

      by Anonymous Coward

      Yes, the combination of 240W/byte and "critical temperature" looks like a problem. I hope they find a way to scale it down a lot.

      • by Anonymous Coward on Sunday January 11, 2009 @09:50AM (#26406979)

        correction:
        it's actually 0.6 mA, so that would be 0.24W/byte, and only during the (very short) write pulse. Still some work to do, but it could possibly end up more efficient than flash memory.

        • The problem is that you're going to need a significant amount of insulator material at 50 volts, which means that the physical spacing will be much further apart than on today's devices. So we will see much higher latency, much more bulk, and because of the physical separation, much higher internal current consumption on the interconnects.

          You won't be seeing this particular implementation displacing flash.

        • Well I know that SD cards can draw a few amps for very small periods of time.
          So 0.6mA is a definate improvement.

    • Re: (Score:3, Interesting)

      well, in its current state, it probably won't be very good for much; but small additions of other elements will probably give you a compound which does what you want. After all, the properties of most alloys change quite remarkably with small composition changes.

    • Re: (Score:3, Interesting)

      by hairyfeet (841228)
      Not to mention according to Wikipedia its main component is HIGHLY [wikipedia.org] toxic to humans. Considering we've been phasing out solder for an inferior replacement(which I personally believe will lead to MORE E-waste due to whiskers) just to cut down the poisons in our electronics I can't seen the green folks jumping onto something that is highly toxic AND water soluble. So IMHO this will end up a non starter before it even gets off the ground.
  • fourth type? (Score:1, Insightful)

    by Anonymous Coward

    I've seen this statement repeated about memristors. Many devices cannot be duplicated with L's, C's and R's (diodes, fuses, etc.). Those parts all have one thing in common, they are nonlinear and/or time varying. Memristors are not LTI and therefore not a "fourth type" of circuit element.

    • That's a bit of a weird conclusion. It would make sense for it to be the 4th type; so that all components can be made up of the passive circuit elements.

    • by frieko (855745)
      There are four fundamental circuit variables: current, voltage, charge and flux.

      Resistance relates voltage and current.
      Capacitance relates charge and voltage.
      Inductance relates flux and current.
      Memristance relates flux and charge.

      Thus M is the 'missing relation'. I'm not so sure I buy it either, but that's where they're coming from. Perhaps LTI was just an incredibly useful coincidence.
      • I like thinking of a memristor as the REAL flux capacitor :)

      • You're forgeting there's two physical laws covering the relationships between charge and current (dq = i dt) and flux and voltage (dphi = v dt). The reasons memristors are not LTI is because (unlike the other three) they don't involve either of the time invariant quanties of current or voltage. Of course, one wouldn't expect them to be entirely like the other three. If they were, you could just make a memristor from resistors, capacitor, inductors and it wouldn't be a fundamental component. To put it an
      • And what about the other two relations (voltage and flux, and charge and current)? Are they for some reason not missing?

        • by frieko (855745)
          As stormydragon pointed out, I included the passives but forgot the two supplies:

          A current source relates charge and current.
          A voltage source relates flux and voltage.
    • Re:fourth type? (Score:5, Informative)

      by Stormy Dragon (800799) on Sunday January 11, 2009 @12:58PM (#26407823) Homepage

      There are four basic quantities that are of concern in an electrical circuit: charge(q), current(i), voltage(v), and magnetic flux(phi). Those four quantities can be matched into pairs 6 different ways. Two of those pairs are time constrained by basic physical law: dq = i dt and dphi = v dt. Three of the remaining are determined by the properties of resistance(R), capacitance(C), and inductance(I): dv = R di, dq = C dv, and dphi = L di. Resistors, capacitors, and inductors are ultimately just devices that have a lot of one of those three properites and nearly none of the others.

      It was speculated in the early 70's that there must be a fourth property, called memristance(M), that describes the 'missing' relationship: dphi = M dq. The memristor, then, is the corresponding device that has a lot of memristance but none of the other three properties. While memristance has been previously measured in complex systems, no one figured out how to build an feasible isolated memristor until just recently.

      The four are considered fundamental in that none of the four can be built from a combination of the other three (e.g. you can't make a resistor from some combination of capacitors, inductors, and memristors) but any device can be built from some combination of the four (e.g. you CAN make a diode from L's, C', R's, and M's).

  • by jridley (9305) on Sunday January 11, 2009 @09:55AM (#26406997)

    No more need to supercool RAM on seized computers in order to extract passwords - the RAM will just naturally hold state for hours.

    If they're going to use this, (some) people are going to want to have more secure operating systems that don't leak security data all over the place.

    • by JSBiff (87824)

      You know, this is the first thing that crossed my mind too. I don't *want* my memory to 'remember' what's stored in it for hours. How is that a good thing?

      "If they're going to use this, (some) people are going to want to have more secure operating systems that don't leak security data all over the place."

      How do you design an O/S that doesn't load security data into RAM? How do you design an O/S that doesn't load your encrypted files into memory when you decrypt them to view or work on them?

      AFAIK, with memor

      • Re: (Score:2, Insightful)

        by Anonymous Coward

        The situation is still somewhat rare that someone can go in and get all the data they like off your computer. They still need physical access to the computer before they can have a prayer at touching the data.

        And heck, since DRAM is also capable of storing data for a couple of minutes after power is cut, you have the exact same problem as someone who has the new memresistor based ram.

      • magnetic hard drives store data for years. so do you also abhor hard drives?

        if you're that paranoid, it's fairly easy to have the OS wipe all user & program data from the memory at shutdown. just because it's non-volatile doesn't mean it's not erasable or re-writable (or it wouldn't be very useful for computer memory). but most people who actually have a need for data security would simply encrypt the data just like people do on their hard drives.

        and this is a good thing because memristors can replace t

        • by TheLink (130905)
          "it's fairly easy to have the OS wipe all user & program data from the memory at shutdown"

          If you're paranoid that doesn't help. People can just yank the power cord, or even pull the chips from their sockets.
          • If you're *that* paranoid, you have your hardware locked in a good safe.

            • by TheLink (130905)
              If you are paranoid about keeping secrets, thermite is your friend.

              It'll be a while before they develop tech for recovering stuff erased with thermite.

              How to use it or rig it up is beyond the scope of this discussion :).
        • if you're that paranoid, it's fairly easy to have the OS wipe all user & program data from the memory at shutdown.

          IMO that is no solution. A system can be easily reset/halted before an OS has a chance to neatly shut down.

          Creating "Security RAM" modules would be more effective: Equipped with a capacitor for power, they could self-wipe at the hardware level when they detect a reset signal or power interruption. Given the precarious nature of info esp. on laptops, one would think this category of RAM would have already been developed.

        • I don't think this 'memristor' is suitable, at least in its current incarnation, to create Solid-State hard drives. Why not? It only retains your info for several hours. So, that essentially limits it to use as RAM. But RAM shouldn't store data for several hours.

          The distinction here is that, on the hard drive, the data is always encrypted, but in RAM, it must be decrypted. I don't mind the hard drive storing the *encrypted* data for years, but I don't want my memory storing decrypted data for hours after I

    • by Daimanta (1140543)

      Then we probably get software/OS that "wipes" the RAM when it shuts down.

  • Not vaporware... (Score:4, Informative)

    by Menkhaf (627996) on Sunday January 11, 2009 @09:57AM (#26406999)
    As the memristor was developed in HP Labs while working on fabrication techniques for "normal" memory, the fabrication technology is already here. It'll only be a short while before we'll see memristors in consumer products.

    "HP prototyped a crossbar latch memory using the devices that can fit 100 gigabits in a square centimeter.[10] HP has reported that its version of the memristor is about one-tenth the speed of DRAM.[27]"

    http://en.wikipedia.org/wiki/Memristor#Potential_applications [wikipedia.org]
    [27] http://www.nytimes.com/2008/05/01/technology/01chip.html [nytimes.com]

    • by trolltalk.com (1108067) on Sunday January 11, 2009 @10:11AM (#26407053) Homepage Journal

      As the memristor was developed in HP Labs while working on fabrication techniques for "normal" memory, the fabrication technology is already here. It'll only be a short while before we'll see memristors in consumer products.

      "HP prototyped a crossbar latch memory using the devices that can fit 100 gigabits in a square centimeter.[10] HP has reported that its version of the memristor is about one-tenth the speed of DRAM.[27]"

      So, knowing HP, we can expect memristors that need a new cartridge to "refill the memory" every few weeks.
      And your initial memsistor will have just a "starter cart" that only accesses 1/4 the data.
      And for best performance, you should only use genuine HP Brand electricity.
      And random blocks of memory in the memristors won't be accessible under linux. Especially when you try to send data via a wireless connection.

      • Re: (Score:3, Funny)

        by daveime (1253762)

        You forgot to mention the HP Memristor (tm) driver software, which despite being about 335 bytes in size, will come bundled in an installed package that is 37MB, just so HP software can show pointless splash screens and randomly create services and daemons that appear to serve no purpose whatsoever, while STILL not being able to cancel the printing of a document without cycling the power.

        • Re: (Score:3, Interesting)

          by ScrewMaster (602015) *

          You forgot to mention the HP Memristor (tm) driver software, which despite being about 335 bytes in size, will come bundled in an installed package that is 37MB, just so HP software can show pointless splash screens and randomly create services and daemons that appear to serve no purpose whatsoever, while STILL not being able to cancel the printing of a document without cycling the power.

          And which will periodically send copies of your memory to a remote HP server in order to "improve the customer experience."

          HP is not what it once was. Thanks for that, Carla.

    • by Glonoinha (587375)

      Although the speed is 1/10th that of DRAM, it appears that they can crank the density up and reduce the refresh rate. It still has effectively zero latency, which puts it squarely in the space between core memory and hard drive storage.

      Personally I see this as a future replacement for the iRAM and other such 'ramdrives' - with the bandwidth limits of the SATA interface, a device of these interlaced across four or eight banks would make a very sweet solid state working storage area. It would need to be tra

      • with the bandwidth limits of the SATA interface"

        Right, cos who would need transfer speeds faster than 300MB/s...

        --- "Determine never to be idle...It is wonderful how much may be done if we are always doing." -Thomas Jefferson

  • by Bender_ (179208) on Sunday January 11, 2009 @10:45AM (#26407205) Journal

    Sorry to be so harsh, but the specific experiment reported here is of little to none value outside of science. Why?

    Hysteretic resistive switching in metal oxide systems is a well known phenomenon (RRAM) and occurs in all transition metal oxides with noble eletrodes. This is what has been recristened as "Memrestor" by HP. It is widely agreed upon that this switching mechanism is due to a redox reaction where oxygen is added or removed from the insulator. The specifics (filament, interfacial barrier lowering etc.) are still subject of current research though.

    The experiment in the paper takes a slightly different approach: vanadium oxide has a very interesting property where its resistance switches apruptly by orders of magnitude at a certain temperature due to a reorganisation of its electronic structure. This phenomenon is known as metal to insulator (MTI) transition and has been research for at least 50 years.

    The MTI has a hysteretic behavior which means that it retains its state if you vary the temperature only a little above or below the critical MTI temperature Tc. The researchers have now shown that if you keep the temperature of the system close to Tc, you can use an additional electric current to switch the resistivity of the system. A possible explanation could be self heating.

    Why is it useless for practical application?

    1) The phenomenon instrinsically only works at a certain temperature. Deviations by fractions of degrees K will destroy all information.

    2) As far as I can see they only demonstrated electrical switching into one direction. To erase the memory both would be required.

    All in all a nice experiment, but again with typical university style hype, piggybacking on the Memristor craze.

    I am also relatively certain that current driven MTI switching has been reported before. I am aware of a couple of experiments where a field switched MTI transition was proposed for transistors. Those devices should exhibit exactly the same hysteresis and "memory" properties.

  • Grammar (Score:1, Offtopic)

    by matt_martin (159394)

    > resistors, capacitors and inductors that were discovered last year

    Wow, I don't know how we made ANYTHING until last year !

  • Patentability (Score:3, Insightful)

    by kimvette (919543) on Sunday January 11, 2009 @02:00PM (#26408191) Homepage Journal

    . . . just posting this in the very unlikely chance that a competent USPTO employee (I know you're out there) is not only reading this thread but also is assigned related patents.

    IMHO this kind of development is worthy of a patent; it includes a brand-new type of component, with no prior art in a single component appearing to exist, and a method by which it is manufactured.

    Now, I expect patent trolls will start the patenting insanity with "it's a PDA, but with memristors" and "it's a phone, but with memristors" and "it's an instant-on PC, but with memristors" and in all of those cases I would say that the patent should not be allowed, because those are "innovations" which are obvious to those skilled in the art.

    Also, the software to store to memristors should not be patentable. "method by which data is semi-permanently stored in a memristor-based storage device" should not be patentable, because that skill (putting data in memory or storage) is obvious to every literate computer user, let alone software engineers.

"Marriage is low down, but you spend the rest of your life paying for it." -- Baskins

Working...