Neuromorphic behaviour in nanoparticle films (2019)
AuthorsShirai, Shotashow all
Conventional computer power has increased dramatically over the last 50 years due to reduction in the physical size and increased areal density of the fundamental electrical components, but this is now facing physical and economical limitations. Neuromorphic computing is a novel computing framework inspired by the biological brain. Advances in neuromorphic computing are expected to make it possible to overcome these limitations.
This study aims to explore electrical properties in percolating nanoparticle devices as candidates for neuromorphic hardware. The devices are fabricated in a cluster deposition system under optimised conditions to stabilise the devices. The electrical properties of the devices fabricated near the percolation threshold are characterised by measuring their conductance.
Sn-based devices show switching behaviour. This is observed as a change in conductance, induced by atomic-scale wire formation and destruction under application of voltage stimuli. Switching behaviour is investigated using voltage pulses, voltage amplitude and temperature. Also, the results of the analysis of the switching dynamics and temporal correlation indicate that the devices exhibit burst activity, scale-free dynamics as well as long-range temporal correlation (LRTC). In neuroscience, these features are associated with memory functions and information processing in cortex. The LRTC in the Sn-based devices originates from the underlying scale-free network structure.
As an alternative device, Ag/Ag2S-based devices are fabricated by the sulphidisation of Ag nanoparticles. The switching behaviour in the Ag/Ag2S devices originates from the electrochemical process of Ag filament formation and dissolution. Also, the device exhibits a stimulus frequency dependent plasticity which is similar to the synaptic activity which characterises the learning abilities of brains.
These brain-like characteristics can be exploited within the context of a neuromorphic computing framework. One example is reservoir computing: this allows practical signal processing such as voice recognition and time series prediction.