Yin Zhi-Xiang, Chen Hao, Yin Sheng-Feng, Zhang Dan, Tang Xin-Gui, Roy Vellaisamy A L, Sun Qi-Jun
School of Physics and Optoelectronic Engineering & Guangdong Provincial Key Laboratory of Sensing Physics and System Integration Applications, Guangdong University of Technology, Guangzhou, Guangdong, 510006, P. R. China.
School of Science and Technology, Hong Kong Metropolitan University, Hong Kong, 999077, P. R. China.
Small. 2025 Apr;21(17):e2412851. doi: 10.1002/smll.202412851. Epub 2025 Mar 19.
Memristors and artificial synapses have attracted tremendous attention due to their promising potential for application in the field of neural morphological computing, but at the same time, continuous optimization and improvement in energy consumption are also highly desirable. In recent years, it has been demonstrated that heterojunction is of great significance in improving the energy consumption of memristors and artificial synapses. By optimizing the material composition, interface characteristics, and device structure of heterojunctions, energy consumption can be reduced, and performance stability and durability can be improved, providing strong support for achieving low-power neural morphological computing systems. Herein, we review the recent progress on heterojunction-based memristors and artificial synapses by summarizing the working mechanisms and recent advances in heterojunction memristors, in terms of material selection, structure design, fabrication techniques, performance optimization strategies, etc. Then, the applications of heterojunction-based artificial synapses in neuromorphological computing and deep learning are introduced and discussed. After that, the remaining bottlenecks restricting the development of heterojunction-based memristors and artificial synapses are introduced and discussed in detail. Finally, corresponding strategies to overcome the remaining challenges are proposed. We believe this review may shed light on the development of high-performance memristors and artificial synapse devices.
忆阻器和人工突触因其在神经形态计算领域的应用潜力而备受关注，但与此同时，能耗的持续优化和改进也非常必要。近年来，已经证明异质结在改善忆阻器和人工突触的能耗方面具有重要意义。通过优化异质结的材料组成、界面特性和器件结构，可以降低能耗，提高性能稳定性和耐久性，为实现低功耗神经形态计算系统提供有力支持。在此，我们通过总结异质结忆阻器的工作机制和最新进展，从材料选择、结构设计、制造技术、性能优化策略等方面，综述了基于异质结的忆阻器和人工突触的最新进展。然后，介绍并讨论了基于异质结的人工突触在神经形态计算和深度学习中的应用。之后，详细介绍并讨论了限制基于异质结的忆阻器和人工突触发展的剩余瓶颈。最后，提出了克服剩余挑战的相应策略。我们相信这篇综述可能会为高性能忆阻器和人工突触器件的发展提供启示。
