Researchers at AMBER, the Science Foundation Ireland funded National Materials Science Research Centre, hosted at Trinity College Dublin, in collaboration with Duke University have discovered the emergence of winner-take-all connectivity pathways in random networks, where memory is distributed across the network but encoded in specific connectivity pathways, similar to that found in biological systems. Their research was published today/this week in the prestigious journal, Nature Communications*.
Establishing the optimum pathway across complex networks is a ubiquitous problem: from information networks such as the internet to physical networks of roadways to highly interconnected biological networks within the brain. These findings may help in the development of hardware based neural network systems with brain-inspired architectures for cognitive signal-processing, decision-making systems and ultimately neuromorphic computing applications. Neuromorphic computers outperform conventional computers at tasks that are natural to our brain such as ultra-fast sensory processing, high-level pattern recognition, and motor control.
The research was a collaboration between Professors John Boland and Mauro Ferreira, AMBER researchers in Trinity’s Schools of Chemistry and Physics, Professor Justin Holmes, AMBER researcher at University College Cork as well as researchers from Duke University. Through experiment and simulation, the collaborative team elucidated the properties of nanowire networks that give rise to singular or multiple connectivity pathways.
Nanowires are similar to normal electrical wires but are extremely small, typically a few hundred atoms thick or thinner than one thousandth of the thickness of a human hair. Just like normal wires, nanowires can be made from a variety of different materials and typically have surface coatings either from their growth process or an engineered coating to stop them clumping together in solution. By changing the nanowire material, or the coating on the nanowire the team found that networks can develop different types of connectivity pathways, and importantly identified the conditions required for the emergence of a single lowest-energy most-efficient pathway.
To understand preferred pathways, think of walking through a University campus or business park with some grassy areas and paths connecting the different buildings. There will be foot-worn short cuts in the grass that people take to save time and energy. The combination of frequently used paved and unpaved pathways are the most practical or preferred pathways for moving efficiently. The human brain develops preferred communication pathways that link together different brain circuits or loops to quickly and efficiently complete specific tasks and this research shows evidence for the same behaviour in a nanowire network.
Prof John Boland, AMBER and Trinity’s School of Chemistry, said, “Nanowire networks offer promising architectures for neuromorphic applications due to their connectivity. Where one nanowire is in contact with another nanowire a junction is formed that behaves like a memory switch, and the behaviour of the network is dominated by the response of these junctions. In this work, we discovered a special symmetry that allows a network of junctions to respond as if it is a single junction. A particular class of junctions then naturally leads to the emergence of a “winner-takes-all” electrically conducting path that spans the entire network, and which we show corresponds to the lowest-energy connectivity path.”
“Even more surprising was that for silver nanowires, which prefers to self-select a single lowest energy pathway across the random network, once the pathway is established it forms a series of discrete memory levels. These results point to the possibility of developing and independently addressing memory levels in complex systems and which we expect to have important implications for computers that operate in a more brain-like fashion.”
The next goal of the research is to understand how to engineer this single or multipath behaviour, and to develop logic systems based on these nanowire network materials for cognitive signal-processing, decision-making systems and ultimately neuromorphic computing applications.
This publication has emanated from research supported in part by Prof John Boland’s Advanced Grant from the European Research Council.