Researchers improve neuromorphic computing by creating artificial neurons composed of complex nanomaterials. Previously, ‘spiking neuromorphic computers’ only simulated the connections between each neuron called ‘synapses’ – little organic cables that create a network of electricity in the brain
Now Japanese scientists from the Kyushu Institute of Technology and Osaka University have come up with a new method for improving neuromorphic computing so as to more accurately resemble the human brain – one of the main goals in this new and emerging field of computation.
Their computer chip is enabled by a large arrangement of customized molecular composite structures that store electricity just like a real neuron. The nano-material in question is composed of polyoxometalate (POM) molecules enclosed by single-wall carbon nanotubes (SWNTs). Akin to how neurons store action potential in the form of electric charge, POM contains a mix of metal and oxygen atoms that conduct electricity at a smaller scale.
This kind of microscopic storage is the norm in another field of computational science called quantum computing, where electrons and photons are stored inside of pre-arranged subatomic formations. However, neuromorphic computing is concerned with larger nano-scale resolution which is why scientists believe it is closer to being implemented.
However with that being said quantum computation is enabled by molecular quantum transistors, where electrons are stored inside of the empty space between interlocking atomic cages. In principle that approach is not so different from storing electricity inside of nano-material composite structures scientists have come up with for neuromorphic computing. Perhaps we are witnessing a significant cross over that could eventually lead to merging quantum and neuromorphic computing.
Of course both of them lie in stark contrast to contemporary computer science whereby information is stored inside pulsing chunks of silicon crystal and transmitted along insulated copper wires. Following the trend of miniaturization, neuromorphic computation continues to further the boundaries by simultaneously shrinking and replicating the human brain at nano-scale resolution.
Their new nano-material communicates through the emission of “spikes” i.e. electrical discharges along artificial synaptic corridors – the kind that neuromorphic computing has become well known for. Now, they’re also storing electricity in a way that resembles the action potential of organic neurons. Of course, that will add another layer of informational complexity that is likely to manifest as increased processing power.
Furthermore researchers demonstrate their molecular model can be utilized for next-generation artificial neural networks called “reservoir computing devices”
The study was published in Nature Communications