Cross-Layer Optimizations for Ferroelectric Neuromorphic Computing

Hardware paradigms for neuromorphic computing mimic the functionalities of brain primitives in order to replicate similar area and energy advantages of biological nervous systems. Discovery of ferroelectricity in doped Hafnia has thrust Ferroelectronics, specifically Ferroelectric Field-effect transistors (FeFETs), at the forefront of realizing such device platforms for future data-centric applications. Despite the utility afforded by its intrinsic properties, e.g. CMOS-compatibility and scalability, harnessing their full potential requires a cross-layer design approach combining devices, circuits, and algorithms. In this paper, we review the recent developments looking at FeFETs and their application to enable low-power on-chip learning. We outline the unique opportunities emerging from device characterization and modelling results, that ultimately translate in novel algorithms and system-level benefits.