Gou Kaiyun, Li Yanran, Song Honglin, Lu Rong, Jiang Jie
Hunan Key Laboratory of Nanophotonics and Devices, School of Physics, Central South University, Changsha, Hunan 410083, China.
State Key Laboratory of Precision Manufacturing for Extreme Service Performance, College of Mechanical and Electrical Engineering, Central South University, Changsha, Hunan 410083, China.
iScience. 2024 Nov 6;27(12):111327. doi: 10.1016/j.isci.2024.111327. eCollection 2024 Dec 20.
With the advent of the post-Moore era and the era of big data, advanced data storage and processing technology are in urgent demand to break the von Neumann bottleneck. Neuromorphic computing, which mimics the computational paradigms of the human brain, offers an efficient and energy-saving way to process large datasets in parallel. Memristor is an ideal architectural unit for constructing neuromorphic computing. It offers several advantages, including a simple structure, low power consumption, non-volatility, and easy large-scale integration. The hardware-based neural network using a large-scale cross array of memristors is considered to be a potential scheme for realizing the next-generation neuromorphic computing. The performance of these devices is a key to constructing the expansive memristor arrays. Herein, this paper provides a comprehensive review of current strategies for enhancing the performance of memristors, focusing on the electronic materials and device structures. Firstly, it examines current device fabrication techniques. Subsequently, it deeply analyzes methods to improve both the performance of individual memristor and the overall performance of device array from a material and structural perspectives. Finally, it summarizes the applications and prospects of memristors in neuromorphic computing and multimodal sensing. It aims at providing an insightful guide for developing the brain-like high computer chip.
随着后摩尔时代和大数据时代的到来，迫切需要先进的数据存储和处理技术来突破冯·诺依曼瓶颈。模仿人类大脑计算范式的神经形态计算提供了一种高效节能的方式来并行处理大型数据集。忆阻器是构建神经形态计算的理想架构单元。它具有多种优点，包括结构简单、低功耗、非易失性以及易于大规模集成。使用大规模忆阻器交叉阵列的基于硬件的神经网络被认为是实现下一代神经形态计算的潜在方案。这些器件的性能是构建大规模忆阻器阵列的关键。在此，本文全面综述了当前提高忆阻器性能的策略，重点关注电子材料和器件结构。首先，研究了当前的器件制造技术。随后，从材料和结构角度深入分析了提高单个忆阻器性能和器件阵列整体性能的方法。最后，总结了忆阻器在神经形态计算和多模态传感中的应用及前景。旨在为开发类脑高性能计算机芯片提供有见地的指导。
