An international research team has created a synthetic neural network using nanotechnology, with potential to develop new systems in machine learning and artificial intelligence.
The team is led by the International Centre for Materials Nanoarchitectonics at the National Institute of Materials Science in Japan, in collaboration with the University of Sydney Nano Institute and School of Physics, and the California NanoSystems Institute at the University of California at Los Angeles.
The team found that when electrically stimulated, this “neuromorphic network” exhibited emergent brain-like behaviour resembling cognitive functions such as learning, memorisation and forgetting. A neuromorphic network is an artificial, large-scale system that mimics the biological functions of the nervous system, particularly its neurons.
Professor Zdenka Kuncic from Sydney Nano and the School of Physics was part of the team. She said: “This is exciting because it opens up the possibility of processing dynamically changing data that existing machine learning and AI methods can’t handle.”
The research was recently published in Springer Nature’s Scientific Reports.
The discovery has implications for the development of artificial intelligence (AI) networks. Although AI is brain-inspired, the underlying mechanism by which the brain processes information remains elusive. Therefore, creating novel materials and systems that mimic functions similar to the brain and understanding the mechanisms of those functions may open up new possibilities for neuromorphic information processing technologies.
The neuromorphic network was created by self-assembled nanowires that form numerous contacts between adjacent nanowires, with each contact exhibiting a synaptic-like response to electrical stimulation. A synapse is a junction between nerve cells.
Nanowires are measured in nanometres, which are a billionth of a metre in size. The nanowires used were made of a silver and polymer composite material. The average diameter of the nanowires used was 360 nanometres, or 0.00000036 metres wide. A human hair is about 100,000 nanometres wide.
The synthetic synapses used collectively regulated the overall functionality of the entire network. The research team investigated the process of electrical signal transmission across preferred paths in the complex network by time-resolved electrical resistance measurements.
This revealed continuous fluctuations that enable electrical signals to exploit multiple transport pathways across the network and spontaneously adapt to changing transmission routes. This process leads to emergent network properties related to learning, memorisation and forgetting of input signals.
Based on this discovery, the research team is now developing next-generation memory devices and neuromorphic information processing systems using nanowire networks. While current AI technology is based on an assumed model of brain-type information processing, this research provides a glimpse into what brain-type information processing actually is. The team hopes that the outcomes of this research will lead to new data processing capabilities beyond the reach of AI.
At Sydney Nano, Professor Kuncic is working with colleagues on a Grand Challenge project to unlock the neural interface. Along with Professor Gregg Suaning in the School of Biomedical Engineering, the multidisciplinary team is looking to unlock the neural code through the convergence of neural biology and electrical stimulation with nanotechnology.
The aim of the project is to harness the combined capacity of neural biology, electrical stimulation and nanotechnology to transform and restore neurons from a state of disease or dysfunction to a state of robust performance indistinguishable from normal function.
The research was funded by the International Centre for Materials Nanoarchitectonics, National Institute for Materials Science, Japan.