ENGINEERING THE COMPLEX ELECTRICAL RESPONSE OF METALLIC CLUSTER-ASSEMBLED FILMS FOR NEUROMORPHIC COMPUTING APPLICATIONS

Mammalian brain outperforms modern computers in executing complex tasks with high energy efficiency. These capabilities rely on the processing of stimuli by a network consisting in thousand billion neurons intricately interconnected through synapses. Neurons integrate signals in a non-linear way. Synapses can modulate their effectiveness according to stimuli (synaptic plasticity), providing the basis for learning. Nontrivial responses to stimuli are due to the collective dynamics regulating the network operations along with network topology. Neuromorphic Computing (NC) draws inspiration from the structure and operation of the brain to offer an alternative to the present Von Neumann paradigm of computation. Metallic cluster-assembled films (MCAFs) constitute a potential physical substrate for this scope due to their topological complexity, non-linear conduction properties, and plasticity. A key role in determining MCAF conduction properties is played by their ability to undergo reversible structural modifications, thus producing changes in their electric conductivity. This phenomenon is referred as Resistive Switching (RS). Even if MCAF based devices have been already exploited in different computing tasks, we still have poor insight on the physics underlying the complex dynamics that govern RS activity. The main goal of my thesis is to exploit a novel approach relying on micro-thermography to characterize the evolution of conductivity in MCAF and gain a deeper understanding of this rich phenomenology. My analysis led to the identification of “switching sites” undergoing the structural modifications responsible of RS. This strategy also allowed to study the statistical properties of RS in space and time, unveiling the presence of non-trivial correlations in these systems. My analysis also highlighted the impact of sample geometry and dimension on the number and distribution of the switching regions. Furthermore, I probed the electrical response to different stimuli, and I observed stereotyped behaviours that emerge from the average response of the switching regions. Modelling the deterministic component of the electrical response, offers the possibility to enhance controllability of MCAF based devices, while the intrinsic stochasticity provides a natural mechanism for efficient state-space exploration, ultimately supporting efficient programmability of such systems.