Abstract: The hafnium oxide ferroelectric tunnel junction (FTJ) has shown great potential in in-memory computing due to its unique material properties and device performance. This paper systematically examines the application of hafnium-based FTJ in artificial neural networks, providing both theoretical and experimental foundations for the hardware realization of non-von Neumann architectures. Material properties, device performance of HfO₂-based FTJs, and their potential applications in in-memory computing are thoroughly analyzed. The research background, global advancements, and emerging trends in hafnium-based materials are reviewed. The operational principles of FTJs are explored, with an emphasis on critical metrics such as switching ratio, endurance, and multi-state storage, alongside current strategies to enhance their ferroelectric characteristics. The integration of hafnium-based FTJs into neural networks is evaluated, and potential future development pathways are projected.