Embodied intelligence, which realizes adaptive behavior through dynamic physical interaction between an agent and its environment, relies critically on hardware capable of integrated perception, storage, and computation (PSC). Ferroelectric neuromorphic devices, which emulate synaptic functions, offer a promising path toward such PSC integration and toward overcoming the energy limitations of von Neumann architectures. However, incompatibility with mainstream semiconductor platforms has always hindered the practical application of traditional oxide ferroelectrics. Recently, wurtzite-structured nitride ferroelectrics have emerged as highly attractive candidates for neuromorphic devices, combining the merits of compatibility with mainstream semiconductor platforms, enhanced remanent polarization (Pr) and piezoelectric polarization, scalability to ultrathin thicknesses, high Curie temperature (Tc), and robust ferroelectric phase stability. While prior reviews have covered basic properties, growth methods, and memristive operation mechanisms of AlScN-based devices, achieving the deep integration of physical systems with artificial intelligence demands memristors with functionalities beyond mere storage and computation. A critical future direction involves embedding multisensory capabilities into neuromorphic devices to enable truly embodied intelligence. This review focuses on the application of wurtzite ferroelectrics in embodied intelligence neuromorphic devices. Given that neuromorphic computing is tightly linked to ferroelectric domain evolution and material properties, the domain dynamics of wurtzite ferroelectrics, including reverse domain nucleation and domain wall motion mechanisms during polarization switching, are systematically discussed. Additionally, we analyze the key factors influencing ferroelectric performance and their modulation strategies, which are critical for ensuring the functionality of neuromorphic devices. For device applications, we summarize the working principles and latest progress in neuromorphic devices, with particular emphasis on two-terminal memristors based on AlScN/n-GaN heterojunction and three-terminal memristors based on two-dimensional materials or two-dimensional electron gas channels, highlighting their potential to integrate sensing, memory, and computation within a single platform. Finally, we outline current challenges and future directions, aiming to provide insights for advancing wurtzite ferroelectrics in high-performance neuromorphic devices for embodied intelligence.
Wang et al. (Fri,) studied this question.