Forgot your password?
typodupeerror
HP Hardware Technology

HP Creates First Hybrid Memristor Chip 155

Posted by timothy
from the stitch-in-time dept.
An anonymous reader writes "HP researchers have built the first functioning hybrid memristor-transistor chip. Lead researcher Stanley Williams and his team built the very first memristor — the '4th fundamental element' of integrated circuits after resistors, capacitors and inductors — back in April. Memristors can remember their resistance, leading to novel electronic capabilities. The new FPGA circuit uses memristors to perform tasks normally carried out by (many more) transistors and is therefore smaller, more power efficient and cheaper to make, HP says. Memristors could also turn out to be a more compact, faster alternative to flash memory."
This discussion has been archived. No new comments can be posted.

HP Creates First Hybrid Memristor Chip

Comments Filter:
  • Re:size? (Score:5, Informative)

    by The-Pheon (65392) on Tuesday November 25, 2008 @01:09PM (#25887859) Homepage

    Does anyone know what size features the chip was etched at? um? nm? That might give a clue how close it is to being used in other products.

    The memristors made in April were 50 nm wide

  • In general there are two ways to store data (at least in a RAM type memory module). You can either use a capacitor (like DRAM) or a flip flop (SRAM). The problem with a capacitor is that it looses it's charge after a certain amount of time, and has to be refreshed (which is why DRAM is relatively slow). A flip-flop can be used to store one bit, but in order to do that two separate latches are needed. One latch has 2-4 gates each having 2-8 transistors. All of a sudden there are over a dozen transistors used to store one but. A memristior does not actually help any calculations, it just hopefully make storing data much more efficent
  • Re:Spice model (Score:5, Informative)

    by TheNecromancer (179644) on Tuesday November 25, 2008 @01:40PM (#25888355)

    Who is the idiot modding this as Funny? A SPICE model is an engineering tool used to diagram circuits. It is NOT related to any of the Spice Girls!!! :p

    Where's the link to MetaModerate?? grrr...

  • by Chirs (87576) on Tuesday November 25, 2008 @01:42PM (#25888409)

    They can replace SRAM cells, which take multiple transistors. They also retain state without needing power (similar to FeRAM, which can also replace SRAM). However, unlike FeRAM the memristor can also store analog values.

    In the digital realm they're likely useful mostly as memory. However, given the analog properties, they could be useful in creating "fuzzy" neural nets.

  • Re:Spice model (Score:5, Informative)

    by evanbd (210358) on Tuesday November 25, 2008 @02:04PM (#25888767)

    I think you could build one from components if you had data about the memristance function. Start with a voltage controlled resistance element (Gxxx) connected to the two exposed terminals. Add a hidden element of a current controlled current source, sensing on the VCR element, injecting current into a hidden capacitor. The voltage on that capacitor is proportional to the total charge that has passed through the memristor device. The exposed VCR element senses voltage on the hidden capacitor, and uses an interpolated table of resistance vs voltage rather than a linear relationship.

    That doesn't capture the hysteretic behavior of the current devices, which stop integrating at the some limiting points, but it's a starting point. You could add such behavior with a few more hidden components (back-to-back ideal zeners across the hidden cap, for example), though getting the right behavior might be a little tricky.

  • by rolfwind (528248) on Tuesday November 25, 2008 @02:04PM (#25888771)

    With memristors (once they are perfected) can have multi-state such as trinary (base 3) or decinary (base 10) eliminating all of the conversion that is neccessary in the present binary system that require cpu cycles. 123 in the decinary system represents 123 where in binary it would be 1111011 and need to be converted in order to be meaningful.

    Um, for the most part, the computer only has to convert from binary to decimal when it displays base 10 numbers on the screen (ie using the calculator). It's hardly computer intensive. All the operations (add, subtract, multiply, divide) are going to be in it's native binary, no conversion needed.

    Computers convert data all the time - this text you are reading now is really just a series of binary numbers converted to ascii or unicode or whatever with lots of other conversions needed to throw it on the screen.

    Native base 10 has been done before (basically ignoring bits representing 10-15) and all that was found was that it wasted space as conversion in those scenarios are beyond trivial. Here's a book for you:
    http://www.amazon.com/Code-Language-Computer-Hardware-Software/dp/0735611319 [amazon.com]

  • FPGA FUD from TFA (Score:3, Informative)

    by fpgaprogrammer (1086859) on Tuesday November 25, 2008 @02:05PM (#25888779) Homepage
    From TFA: '''In order to be so flexible, however, FPGAs are large and expensive. And once the design is done, engineers generally abandon FPGAs for leaner "application-specific integrated circuits."'''

    This isn't really true. The rising fixed costs of an ASIC is prohibitive for low volume embedded projects where a $1 FPGA will do just fine. High performance FPGA chips are about the same cost as a CPU and they are commonly used as reconfigurable co-processors for supercomputing applications or embedded DSP. And I get way more GigaOps per dollar with FPGAs than with a CPU and for much less power.
  • by evanbd (210358) on Tuesday November 25, 2008 @02:09PM (#25888843)
    Modern SRAMs usually use fewer transistors than that. 6T SRAMs are common, for example. They use a pair of inverters to store the state (4 transistors) and a pair of transistors to connect the inverters to the data lines (6 total). The write operation then involves a drive signal with more power than the inverters, thus forcing the state change.
  • by TheRaven64 (641858) on Tuesday November 25, 2008 @02:19PM (#25888999) Journal
    The grandparent isn't as silly as it sounds. Multi-level flash cells don't store binary data, they work, typically, in base 4. Each cell stores a value from 0-3, meaning you only need 4 cells instead of 8 for a byte. Memristor-based systems can accurately store analogue values, and so can potentially store more than one bit of data in a single element. If you could store an entire byte in a single element then you'd get much higher density, although you'd probably want to translate it back into binary when you loaded it into a register, since binary arithmetic circuits are very easy to build (try doing bitwise or in decimal...).
  • by bitrex (859228) on Tuesday November 25, 2008 @02:48PM (#25889407)
    I believe what they mean is that the memristor is "fundamental" in that it provides a symmetrical link between electric charge and magnetic flux. A resistor is the link between current and voltage, a capacitor - voltage and charge, an inductor - current and magnetic flux, and a memristor - magnetic flux and charge. I think. I'm sure someone with a better grounding in electromagnetic physics can clarify this a bit.
  • Re:size? (Score:3, Informative)

    by DoubleReed (565061) on Tuesday November 25, 2008 @03:10PM (#25889679)

    1Hz? Next!

    The application they talk about is the programmable transistors in FPGAs.

    You can find FPGAs in all kinds of consumer electronics. Typically, the chip has some onboard flash from which it loads its configuration during power on. These transistors which load from flash during power on are the ones the researchers are talking about replacing with memresistors (memristors?). For this kind of application, a 1Hz write speed is completely reasonable. The write will be done once in the factory.

  • by Falstius (963333) on Tuesday November 25, 2008 @05:15PM (#25891573)

    It is the fourth fundamental linear circuit element. Transistors (I = k*V^2) and diodes (I = e^V) are not linear. Resistors V = k*I, capacitors I = k*dV/dt (the derivative is a linear operation), inductors V = k*dI/dt, memristors V=k(t)*I are linear.

    The reason there are four linear circuit elements is clear if you write these equations in terms and flux and charge.

How often I found where I should be going only by setting out for somewhere else. -- R. Buckminster Fuller

Working...