Artificial intelligence is driving growing demand for computing hardware that can process information faster and more efficiently. However, most modern computers still rely on the traditional von Neumann architecture, where memory and processing units are physically separated. This separation requires frequent data movement between storage and computation, creating a major bottleneck that increases energy consumption and limits performance. To overcome this challenge, researchers have been actively exploring new hardware technologies capable of integrating memory and computation within a single device.
In a recent review published in Science Bulletin, researchers systematically summarize the latest advances in two-terminal ferroelectric memristors, an emerging class of devices that combine information storage and processing functions. Unlike conventional memory technologies, ferroelectric memristors utilize reversible ferroelectric polarization switching to continuously modulate resistance states while maintaining nonvolatile data storage. This unique characteristic allows data to be processed directly where they are stored, significantly reducing data-transfer overhead and improving computing efficiency.
The review highlights recent progress in a wide range of ferroelectric material systems, including hafnium oxide-based materials, perovskite ferroelectrics, polymer ferroelectrics, and low-dimensional ferroelectric materials. Among them, hafnium oxide-based ferroelectrics have attracted particular attention because of their compatibility with existing CMOS manufacturing technology, making them strong candidates for future large-scale integration. The authors also discuss the fundamental mechanisms governing device operation, including ferroelectric polarization switching, domain-wall motion, electric-field modulation, and interface engineering, all of which play crucial roles in determining device performance, reliability, and energy efficiency.
Beyond materials and device physics, the review emphasizes the growing potential of ferroelectric memristors for neuromorphic computing and artificial intelligence. Their polarization dynamics can mimic key characteristics of biological synapses, enabling learning, memory, and information-processing functions within hardware. Recent studies have demonstrated applications ranging from pattern recognition and sensory information processing to artificial neural networks. Although challenges such as device uniformity, long-term stability, and large-scale manufacturing remain, the authors believe that ferroelectric memristors are becoming an important foundation for next-generation intelligent computing systems and may help bridge the long-standing divide between memory and computation.
Science Bulletin
Systematic review