Modeling and Simulation of Ferroelectric-based Devices for Neuromorphic Computing Applications

In recent years, there has been a growing diffusion of computing systems based on artificial neural networks and artificial intelligence, which can be used in a very wide range of applications. However, these bio-inspired systems are often very energy-intensive due to continuous data transfer between the processing unit and memory; this continuous data transfer is also known as the Von Neumann bottleneck. A new computing paradigm, called neuromorphic computing, suggests to overcome this bottleneck by integrating the logic and memory functionalities within a single chip thus reducing the continuous data and emulating the working principle of biological neural networks. Memristive devices, which store information in a non-volatile manner as the value of their resistance, have gained appeal as possible hardware candidates to enable neuromorphic computing. Among the many candidates proposed as memristors, one of the most prominent is represented by ferroelectric-based devices. Ferroelectric materials are a sub-class of dielectric materials that possess a hysteretic polarization versus electric field characteristic. The energy required for their operation is significantly lower than it is in competing technologies for memristive applications, thus explaining their appeal in the scientific community. This thesis aims to study and model the behavior of ferroelectric devices as hardware for neuromorphic computing. This manuscript provides insights into the operation of ferroelectric materials in Ferroelectric Tunnel Junctions, as well as a novel approach to the modeling of antiferroelectric materials within the context of the Landau-Ginzburg-Devonshire framework. Overall, this work contributes to advancing the understanding and application of ferroelectric devices for energy-efficient computing applications.