In the modern era marked by rapid technological advancements, ferroelectric materials have gradually emerged as highly promising candidates for a wide range of applications, including ferroelectric memories, sensors, and optoelectronic devices, due to their distinctive polarization characteristics. A common strategy to address the low ferroelectric polarization caused by lattice mismatch between the substrate and ferroelectric film is the insertion of a buffer layer. However, a thicker buffer layer tends to promote dislocation formation, which relaxes epitaxial strain and thereby deteriorates ferroelectric polarization, this mechanism has yet to be systematically explored. In this study, a method is presented that alleviates strain relaxation by modulating interfacial stress through precise control of the buffer layer thickness, thereby enhancing the ferroelectric polarization performance. Here, to reduce the strain between the PbZr_1−xTi_xO₃ (PZT) and substrates, which could induce pronounced lattice mismatch, increased defect density, and consequently reduced ferroelectric performance, a SrRuO₃ (SRO) buffer layer of optimized thickness was inserted between SrTiO₃ (STO) and PZT to mitigate the lattice mismatch. This approach increased the maximum polarization from 126.3 to 142.6 μC/cm², the remanent polarization from 86.52 to 116.03 μC/cm², and enhanced the photocurrent by 2.2 μA. On this basis, the material stack provided robust support for an intelligent traffic-intersection recognition system, achieving a recognition accuracy of 93.23% under diverse weather conditions. The methodology elucidated the fundamental interplay between strain and ferroelectric/photoelectric properties, offering new insights and strategies for the performance optimization of ferroelectric materials.

Mott insulator material, as a kind of strongly correlated electronic system with the characteristic of a drastic change in electrical conductivity, shows excellent application prospects in neuromorphological calculations and has attracted significant attention in the scientific community. Especially, computing systems based on Mott insulators can overcome the bottleneck of separated data storage and calculation in traditional artificial intelligence systems based on the von Neumann architecture, with the potential to save energy, increase operation speed, improve integration, scalability, and three-dimensionally stacked, and more suitable to neuromorphic computing than a complementary metal-oxide-semiconductor. In this review, we have reviewed Mott insulator materials, methods for driving Mott insulator transformation (pressure-, voltage-, and temperature-driven approaches), and recent relevant applications in neuromorphic calculations. The results in this review provide a path for further study of the applications in neuromorphic calculations based on Mott insulator materials and the related devices.

Ferroelectric memristors, as one of the most potential non-volatile memory to meet the rapid development of the artificial intelligence era, have the comprehensive function of simulating brain storage and calculation. However, due to the high dielectric loss of traditional ferroelectric materials, the durability of ferroelectric memristors and Si based integration have a great challenge. Here, we report a silicon-based epitaxial ferroelectric memristor based on self-assembled vertically aligned nano-composites BaTiO₃(BTO)-CeO₂ films. The BTO-CeO₂ memristors exhibit a stable resistance switching behavior at a high temperature of 100 °C due to higher Curie temperatures of BTO-CeO₂ films with in-plane compressive strain. And the endurance of the device can reach the order of magnitude of 1 × 10⁶ times. More importantly, the device has excellent functions for simulating artificial synaptic behavior, including excitatory post-synaptic current, paired-pulse facilitation, paired-pulse depression, spike-time-dependent plasticity, and short and long-term plasticity. Digits recognition ability of the memristor devices is evaluated though a single-layer perceptron model, in which recognition accuracy of digital can reach 86.78% after 20 training iterations. These results provide new way for epitaxial composite ferroelectric films as memristor medium with high temperature intolerance and better durability integrated on silicon.