Ferroelectric Domain Patterns for Brain-Inspired Information Processing

Ferroelectric thin fims can be engineered to combine several key features of relevance for efficient information processing: they provide non-volatile memory addresasble by electric fields and they self-organize in domain patterns, whose periodicity can be made of the order of nanometers for ultra-thin layers. In addition, in ferroelectric oxides, the domain walls have been often found to be conducting and memristive due to ion migration driven by strain gradients and internal bias field, therefore, showing multistate resistance and synaptic functionality. In addition, these materials can be tuned to exist the verge of a phase transition, by means of chemical composition or epitaxial strain. As shown by Kinouchi and Copelli (Nature Physics 2006), the dynamical range is maximized at non-equilibrium phase transitions and, it is well known that the dynamics of a system slows down by many orders of magnitude near critical points, where the correlation lengths reach the size of the entire system. Therefore, materials that undergo phase transitions are of great interest to achieve the largest sensitivity to input signals, the greatest non-linearities in their responses and, at the same time, the largest variations in the time scales of their output signals, all of these very useful ingreients for efficient learning. Here I will show two examples of ferroelectric systems with domain patterns that show some of these aspects: BaTiO₃ films at a boundary between two differently modulated domain phases that show signatures of edge-of-chaos dynamics and BiFeO₃ films that provide a dense network of interconnected (conducting) domain walls intersecting memory bits.