Web Concepts

Posts Tagged ‘memristors’

Like a delicious, biocompatible computer

The prototypes of the design haven’t yet been outfitted with much memory yet, but the researchers say the capacity is there. That aspect of the material works like a memristor, existing in one of two states at any given time: conductive or resistive. These two states can represent the 1s and 0s of binary computer code, and could someday be used to program the stuff to work inside or on the human body, perhaps in medical monitoring devices or biological sensors.

It’s a pretty cool development, at least for those who think we’re all going to be cyborgs with various machinery augmenting our bodies and optimizing our lives and keeping us healthy at some point in the future. Plus, it has the consistency of Jell-O. I can’t be the only one envisioning a new generation of scrumptious, intelligent desserts.

[NCSU]

Memristors are the fourth fundamental piece of electronic circuitry after capacitors, inductors, and resistors. They are basically resistors that remember what state of resistance they were in the last time a current passed through them. They were proposed more than three decades ago but were only first created in recent years by HP. And they are expected to drive revolutionary advances in future electronics for several reasons, not least of which is their ability to behave more like neurons than conventional electronic circuits.

Which brings us back to mazes. A maze can have varying degrees of complexity. It might have several solutions. A conventional processor would solve a maze the same way a person might--start at the starting point and weave itself through, noting dead ends and retrying until it gets all the way through to the end. Depending on the complexity of the maze, this could take some time.

To demonstrate how memristors could do better, Yuriy Pershin at the University of South Carolina and Massimiliano Di Ventra at UCSD constructed a sort of universal maze out of a grid of memristors which could be adopted to reflect any maze by disabling certain connections between memristors through which electricity cannot flow. Using this memristor array, once any maze design is imposed on their processor it can be solved for by simply applying voltage to the maze entrance and a ground to the finish.

That doesn’t sound so groundbreaking until you think it through. The memristor grid, unlike a conventional computer program, actually works in parallel, with all of the memristors working on solving the maze simultaneously. If there are multiple solutions, those are solved for simultaneously as well. Beyond that, the memristors will “remember” the solution(s) in their states for recall or use later.

That’s not such a big deal if you’re just solving a maze, but if you’re applying this power to robotics, graph theory, network optimization, or a slew of other computing models, it has the power to work much faster and more efficiently than computing complex problems in series.

Perhsin and Di Ventra’s processor amounts to the first application of memristor networks to massively parallel computing. Considering it such computing more closely mirrors the way the human brain works--or the way a brain-like computer might work--solving a maze with a memristor is about thrilling solving a simple puzzle can get.

[Technology Review]

MoNETA, software that runs a memristor brain, could make artificial intelligence a reality

There’s reason to be optimistic that this attempt might be different from all the previous AI let-downs that have come before it. Why? The memristor, a concept that HP first realized in 2008. The memristor, put simply, is an electronic component in which the resistance is dependent on the amount of charge that passed through it at a previous time. In other words, it remembers the state it was in the last time charge was applied, unlike a conventional RAM cell (which requires constant power to maintain the same state).

Their ability to both store and process information as it transfers charge (and to do so with far less power consumption) makes memristors more analogous to the neurons in the brain than any other previously developed electronic component, and they are small enough, cheap enough and efficient enough to someday be used to build computing platforms that function more like the brain: learning, making decisions, and even using a machine version of intuition to execute their roles.

The Boston U. team, by its own admission, doesn’t yet know exactly what these platforms will look like, but they seem very confident that they will soon be a reality. They also admit that, due to their benefactor (the DoD) they will likely first find a home in military tech; think autonomous vehicles that don’t just prowl the skies, but that actively engage in learning behaviors and problem solving to, say, search for IEDs or patrol territory for hostile intent. But the researchers envision a much broader role for MoNETA – and brain mimicking machines on the whole – in the near future.

Decide for yourself if MoNETA is the real deal by clicking through the source link below. It’s a somewhat lengthy but entirely interesting read.

[IEEE Spectrum]

The stated goal is a brain that works "at the cat level" and fits in a two-liter soda bottle

Memristors can take on the role of traditional transistors in such computers by opposing the flow of current. They can also remember the last voltage they experienced, not unlike the brain's synapses. Those signal junctions build stronger connections among neurons based on the strength and timing of signals, and form a basic part of how the brain's memory and learning processes work.

Similarly, researchers hope to build a computer network capable of learning. Yet a massive supercomputing cluster consisting of 140,000 processing units still performs 83 times slower than a cat's brain, said Wei Lu, a computer engineer at the University of Michigan. (Is he taking a catty swipe at Blue Gene?)

But Lu has now made progress, by connecting two electronic circuits with one memristor, and showing that they can remember the signal spikes which form the basis for learning and memory.

"We show that we can use voltage timing to gradually increase or decrease the electrical conductance in this memristor-based system," Lu said. "In our brains, similar changes in synapse conductance essentially give rise to long term memory."

Lu plans to develop a supercomputer packed into the size of a two-liter soft drink bottle several years down the line. DARPA's hope is for that supercomputer to have the electronic equivalent of cat-brain intelligence -- for instance, merely recognizing the shape of an object and being able to follow that object regardless of changing locations.

Of course, there's cat brains, and then there's cat brains. But it does seem like DARPA genuinely wants a supercomputer capable of thinking on some feline level, rather than just a clumping of simulated brain cells without any meaningful cognitive abilities.

[University of Michigan]

"In theory we can connect thousands of layers in a very straightforward fashion," Stan Williams, and scientist at HP, told the BBC. "It could provide a way of getting a ridiculous amount of memory on a chip."

Memristors improve on transistors in three key ways. First off, they allow the same device to serve as the processor and the memory. Right now, computers need separate devices for memory (such as solid state flash memory or regular magnetic hard drives) and processing (the computer chip itself). By eliminating the communication time and energy between those different parts of hardware, a memristor system would work far faster, and with far less energy, than a traditional computer.

Second, memristors can be much smaller than transistors. Quantum mechanics limits how tiny transistors can be, a limit that current technology is rapidly approaching. Memristors would allow computer chips to continue getting smaller past that point, all without resorting to exotic tricks like graphene chips or quantum computing.

Lastly, unlike transistors, which only work linearly, memristors can form three-dimensional networks. This added dimension exponentially expands the number of connections, and thus the power, of a memristor computer. In fact, the 3-D network capability of memristors is so profound that Leon Chua, the man who first theorized the existence of memristors in the 1971, believes that this technology could enable the creation of electronic brains. "We have the right stuff now to build real brains," he told the Times.

Hewlett Packard has already created a few simple devices that run on memristors as proof of concept, and they think that they can have the first working models capable of replacing some current computer parts within three years. However, with memristors enabling chip development for decades past where transistors would have hit their physical limit, the true value of this advance may not be realized for years to come.

[BBC News, and The New York Times]