Scientists Create An Artificial Synapse That Can Learn Autonomously - Breaking News, Sports, Entertainment and more
The Brain

Scientists Create An Artificial Synapse That Can Learn Autonomously

Each person’s brain has billions of neurons and trillions of synapses – a neural assembly that’s shaped by time, environment, and experience in a way that is unique to everyone.

Now, researchers have taken inspiration from the inner workings of this convoluted organ to develop an artificial synapse that they say is capable of learning autonomously. They have even modeled the device, which is considered the next step in the creation of more complex circuits. The study is published in Nature Communications. 

The team created a nanoscale device called a memristor, whose resistance depends on the electrical signals it has previously received. The idea of the memristor is not new – it was first conceptualized in the 1970s and subsequently built in 2008. This study, however, takes it to the next level. 

The idea of the memristor is to create an electronic equivalent of the brain’s neurons and synapses – the biological “wiring” that is able to process and store information with incredible efficiency. Simply put, the synapse is the junction between two nerve cells that opens or shuts depending on the nerve impulses that reach it. Neurotransmitters cross that gap to pass the impulses on to the next neuron. Every time that crossing is made, the connection gets stronger and more efficient.

To achieve a biomimetic version of this, an ultrathin ferroelectric film was sandwiched between two electrodes, whose resistance can be tuned using voltage pulses. Thus, its plasticity (the ability to change and learn) is achieved via conductance – low resistance corresponds to a strong synaptic connection and high resistance to a weak connection.

content-1491493470-170403140249-1-900x60

The team then made a model of the device, and their “simulations show that arrays of ferroelectric nanosynapses can autonomously learn to recognize patterns in a predictable way, opening the path towards unsupervised learning in spiking neural networks.”

Essentially, the work takes us closer towards improving the speed at which artificial neural networks learn and adapt. Artificial intelligence (AI) systems have developed a lot in the last few years, with Google’s DeepMind and AlphaGo among the most popular examples. 

However, the brain is an incredibly intelligent machine and we are nowhere near replicating its sophistication. Even as you read this, neurons in your brain are firing a frenzy of electrical impulses and connecting to each other in ever-changing configurations. Such efficiency is a much sought-after goal in the creation of artificial brains. 

As the authors note, we are inching ever closer to an AI future: “These results pave the way toward low-power hardware implementations of billions of reliable and predictable artificial synapses (such as deep neural networks) in future brain-inspired computers.”

Image in text: Artist’s impression of the electronic synapse. The particles represent electrons, the flow of which depends on the ferroelectric domain structure. Credit: © Sören Boyn / CNRS/Thales physics joint research unit.

Leave a Comment