New research led by the University of Southampton, UK, has demonstrated that a nanoscale device called a memristor could be used to power artificial systems that can mimic the human brain.

Artificial neural networks (ANNs) exhibit learning abilities and can perform tasks that are difficult for conventional computing systems, such as pattern recognition, on-line learning and classification. But practical ANN implementations are currently hampered by the lack of efficient hardware synapses—a key component that every ANN requires in large numbers.

In a study published in Nature Communications, the Southampton research team experimentally demonstrated an ANN that used memristor synapses supporting sophisticated learning rules in order to carry out reversible learning of noisy input data. Memristors are electrical components that limit or regulate the flow of electrical current in a circuit and can remember the amount of charge that was flowing through it and retain the data, even when the power is turned off.

A memristor chip. Image credit: University of Southampton.Acting like synapses in the brain, a metal-oxide memristor array was shown to be capable of learning and re-learning input patterns in an unsupervised manner within a probabilistic winner-take-all network. This is potentially useful for enabling low-power embedded processors (required for the Internet of Things) that can process big data in real time without any prior knowledge of the data.

“If we want to build artificial systems that can mimic the brain in function and power, we need to use hundreds of billions, perhaps even trillions, of artificial synapses, many of which must be able to implement learning rules of varying degrees of complexity," says Alex Serb, a postdoctoral research fellow in electronics and computer science. "[While] currently available electronic components can certainly be pieced together to create such synapses, the required power and area efficiency benchmarks will be extremely difficult to meet—if even possible at all—without designing new and bespoke 'synapse components.'"

According to Serb, memristors offer a possible route to that end by supporting many fundamental features of learning synapses—memory storage, on-line learning, computationally powerful learning-rule implementation and two-terminal structure—in extremely compact volumes and at low energy costs.