Thriving In-Memory Computing and Neuromorphic Applications of Ferroelectric-Based Devices
The rapid expansion of artificial intelligence (AI) model sizes to trillions of parameters has intensified the demand for computational paradigms that overcome the von Neumann bottleneck. Emerging memory technologies, while advancing, fall short of meeting the massive requirements of large-scale models. Ferroelectric materials, with their continuous tunability of domain patterns, offer a promising route to emulate synaptic weights in biological learning. This review systematically examines four fundamental ferroelectric-based device architectures: ferroelectric capacitors, ferroelectric field-effect transistors (FeFETs), ferroelectric tunnel junctions (FTJs), and ferroelectric domain wall memories. We analyze their latest progress, application domains, and inherent advantages, while critically assessing the challenges impeding their commercialization. Key issues include scalability, endurance, retention, and integration with CMOS technology. We also highlight optimization strategies for material and device performance, array-level design, and neuromorphic computing architectures. Future research directions are proposed, emphasizing the expansion of novel applications and the realization of energy-efficient, high-density in-memory computing systems. This review provides a comprehensive framework for researchers and engineers aiming to harness ferroelectric devices for next-generation computing.