Neural Network Chip Constructed With Memristors

 Neuromorphic Computing  

Researchers have created a transistor-free metal-oxide memristor network that can learn to recognize imperfect 3 × 3 pixel black-and-white patterns as one of three letters of the alphabet. The approach could be scaled so that larger neuromorphic networks capable of more challenging tasks may be possible.

Researchers at the University of California and Stony Brook University have, for the first time, created a neural network chip that was built using just memristors.

The study, published in the journal Nature, describes how they built their chip and what capabilities it has.

Memristors are electronic analog memory devices that are modeled on human neurons and synapses. Human consciousness, some believe, is in reality, nothing more than an advanced form of memory retention and processing, and it is analog, as opposed to computers, which of course are digital.

The concept for memristors was first worked out University of California professor Leon Chua back in 1971, but it was not until a team working at Hewlett-Packard in 2008, first built one. Since then, a lot of research has gone into studying the technology, but until now, no one had ever built a neural-network chip based exclusively on them.

Until recently, most neural networks have been theoretical or just software based. Google, Facebook and IBM, for example, are all working on computer systems running such learning networks, mostly meant to pick faces out of a crowd, or return an answer based on a human phrased question.

The gains in such technology have been obvious, the limiting factor is the hardware—as neural networks grow in size and complexity, they begin to tax the abilities of even the fastest computers.

Unlike other brain-inspired neuromorphic chips, which use the same silicon transistors and digital circuits that make up ordinary computer processors. The memristor-based chip is better suited for mimicking synapses, says Dmitri Strukov, an assistant professor at the University of California, Santa Barbara, who led work on the new memristor chip.

Many transistors and digital circuits are needed to represent a single synapse. By contrast, each of the 100 or so synapses on the UCSB chip is represented using only a single memristor.

Related articles Researchers Create Nanoscale Memory Devices That Mimic The Human Brain   Memristor Technology Offers The Promise of Learning Computers  Stanford Researchers and Google Create World's Largest Artificial Neural Network

The next step, according to experts, is to replace transistors with memristors. Each memristor is able to learn, in ways similar to the way neurons in the brain learn when presented with something new. Constructing the memristors on a chip would of course reduce the overhead needed to run the neural network.

"This demonstration is an important step towards the implementation of much larger and more complex memristive neuromorphic networks."

The new chip, the team reports, was created using transistor-free metal-oxide memristor crossbars and represents a basic neural network able to perform just one task—to learn and recognize patterns in very simple 3 × 3 pixel black and white images. In the diagram at the top, the memristors are the yellow components.

The experimental chip is an important step towards the creation of larger neural networks that tap the real power of memristors. It also makes possible the idea of building computers in lock-step with advances in research looking into discovering just how exactly our neurons work at their most basic level.

"We believe that this demonstration is an important step towards the implementation of much larger and more complex memristive neuromorphic networks," state the researchers.

Commenting on the work, Robert Legenstein, an associate professor at Graz University of Technology in Austria, wrote: “If this design can be scaled up to large network sizes, it will affect the future of computing … Laptops, mobile phones and robots could include ultra-low-power neuromorphic chips that process visual, auditory and other types of sensory information.”


SOURCES  PhysOrg and MIT Technology Review

By 33rd Square

Embed