Characterisation of self-organised memristive random nanowire networks

The growing demand for power and speed in digital computers is hampered by their architecture, Von Neumann architecture, in which the memory unit and the processing unit (CPU) are spatially separated and connected by buses. In order to overcome such limitations, commonly known as Von Neumann Bottleneck, a change in the computing paradigm and new technologies are needed. Taking inspiration from the way our brain is structured and works, a new kind of architecture has been proposed: memory and processing are accomplished in the same physical location, allowing spatio-temporal correlation of information. Thanks to its resistive switching and memory properties, the Memristor, a non-linear two-terminal passive component, is an optimal candidate to mimic the non-linear behaviour of brain synapses, and architectures composed of memristive elements find application in the implementation of Reservoir Computing, by which temporal input can be handled. In this work a self-organised memristive random network of Ag nanowires (NWs) is presented and characterised. The memristive function is carried out by both the junction between NWs and the single NW itself, making the whole network able to emulate brain properties like homo- and heterosynaptic plasticity, short-term plasticity and paired pulse facilitation. First, two-terminal measurements have been performed to investigate the pristine resistance state of the network and the contribution of the contact resistance. Then, different kinds of voltage stimuli have been applied to see how the network responds to excitation by short and long pulses, pulse trains and constant voltages and in how much time it relaxes to its initial resistance state. This study has been conducted on several samples that differ between each other in NW density. For the ones with a low density, an additional electroforming process had to be performed, since the poor amounts of initial connections made the network behave almost like an open circuit. The work has been carried on with the purpose of finding the optimal NW density that gives the best network dynamics.