Hongjun Wang

The von Neumann memory wall severely restricts the performance of traditional computing systems, and memristors have emerged as premier candidates for neuromorphic computing by virtue of their ability to integrate storage and computation functions. However, conventional HfO _x memristors suffer from abrupt resistive switching and lack of intrinsic self‐protection, limiting their synaptic emulation performance. Herein, a novel nonvolatile HfO _x /MoS _x /HfO _x trilayer heterostructure memristor with intrinsic self‐compliance and analog conductance modulation was constructed. The optimized device exhibits stable cycling endurance, multilevel retention stability and fast response speed, and achieves linear and symmetric conductance potentiation and depression. Furthermore, the proposed memristor successfully emulates diverse biological synaptic plasticities, including EPSC, IPSC, PPF, and PPD. A constructed neural network based on this device achieves a high handwritten digit recognition accuracy of 93.84% on the MNIST dataset, with excellent noise immunity and device variation robustness. Furthermore, the memristor acts as an irreplaceable physical entropy source for zero‐trust XOR‐based image encryption, generating ciphertext images with near‐optimal information entropy reaching 7.978 bits. This work presents a feasible MoS _x interlayer modification strategy to optimize high‐performance artificial synapses on oxide memristors, enabling co‐integration of neuromorphic computing and hardware security.

Memristors are promising candidates for neuromorphic computing, but conventional oxide-based memristors suffer abrupt switching and limited synaptic plasticity. This paper reports a W/HfO2/Y2O3/Pt memristor that overcomes limitations through oxygen vacancy gradient and heterojunction interface engineering. The HfO2 layer (40.79% oxygen vacancies) forms primary conductive filaments, whereas the Y2O3 layer (33.68%) contains discontinuous residual filament segments that enable analog switching. Band discontinuity at the heterojunction localizes filament rupture, reducing filament randomness and improving switching uniformity. Combined with the oxygen vacancy gradient, this structure enables gradual synaptic modulation. The device exhibits gradual conductance modulation by adjusting the compliance current or reset voltage, with five stable resistance states retained for over 104 s. Under 50 consecutive negative pulses, the conductance decreases, demonstrating long-term depression behavior. A neural network simulation achieves 97.8% accuracy in MNIST handwritten digit recognition. This study promotes controllable conductance modulation and provides an effective strategy for artificial synapses and neuromorphic computing.

Gallium oxide (Ga₂O₃) has attracted extensive research for memristor applications due to its excellent properties, including a wide bandgap energy, oxygen-sensitive properties, and outstanding optoelectronic performance. However, existing devices typically suffer from high switching voltages and insufficient stability. Herein, high-stability memristors with resistive switching behavior are demonstrated, based on the a-TiO₂/a-Ga₂O₃ heterostructure. Compared to monolayer devices, the heterostructured memristors demonstrate remarkable robust performance, achieving a switching ratio of over 1 order of magnitude, excellent data retention (>10⁴ s), high endurance exceeding 500 switching cycles, and reliable long-term potentiation/depression (LTP/LTD) characteristics. This high-stability originates from its capacity to guide migration pathways and facilitates the ordered aggregation of oxygen vacancies, which facilitate enabling the formation of robust conductive filaments within the a-TiO₂/a-Ga₂O₃ heterostructures. The findings not only offer an effective optimization strategy for regulating the performance of Ga₂O₃-based memristors but also establish the development of high-efficiency artificial neural network computing systems.

By integrating rectification and resistive switching functionalities, self-rectifying memristors play a pivotal role for large-scale, high-density 3D integration, which can effectively suppress sneak-path currents. However, the limited reliability and poor overall performance of mainstream metal oxide-based self-rectifying memristors hinder their widespread practical application. Herein, robust self-rectifying memristors are fabricated using an amorphous WO₃/amorphous IGZO (a-WO₃/a-IGZO) heterostructure, and they exhibit excellent high-performance characteristics, including a high rectification ratio (>10⁴), low operating voltages, and outstanding operational stability. This enhanced self-rectifying switching performance is attributed to the formation of an interfacial space-charge layer, resulting from the mismatch in carrier concentration between a-WO₃ and a-IGZO. Furthermore, multisynaptic functions are subsequently emulated using a-WO₃/a-IGZO heterostructure memristors, whose conductance can be continuously modulated. This work presents a heterostructure strategy for constructing robust, high-performance self-rectifying memristors based on amorphous metal oxides, which effectively suppress crosstalk currents and serve as reliable artificial synapses for neuromorphic computing.

忆阻器作为新一代非易失性存储存储器，可应用于神经形态计算等新型计算范式。TiO₂材料因其高介电常数与热稳定性被广泛应用于忆阻器的功能层，然而TiO₂基忆阻器仍存在稳定性较差及模拟性能欠优等瓶颈。本文引入非晶态的二维材料MoS₂（（a-MoS₂）与非晶态TiO₂（（a-TiO₂）构建了异质结结构，获得了超200次循环、高数据保持时间（>10⁴ s）的模拟忆阻器。此外，W/a-MoS₂/a-TiO₂/Pt器件可通过控制扫描电压实现多级电导调制功能，通过导电机制拟合分析构建了一种基于局部纳米点诱导的导电细丝形成及断裂的物理开关模型，分析了器件实现多级电导调制的物理机制。最后，本文在W/a-MoS₂/a-TiO₂/Pt器件上实现了LTP、LTD功能。本文设计构建了W/a-MoS₂/a-TiO₂/Pt异质结结构，为过渡金属氧化物基忆阻器性能改进与应用提供了有效方案。

Memristors are promising candidates for next-generation data storage due to their exceptional resistive switching (RS) capabilities, characterized by ultra-fast response times, low power consumption, and outstanding stability. However, current memristor technology achieves a storage capacity of only one bit per cell, which severely restricts the storage density of individual devices. Herein, we demonstrate Y2O3/TiO2 heterostructure-based memristors capable of binary and ternary data storage, with tunable functionality achieved by modulating electrical stress conditions. Under low-electrical stress conditions (−1.50 V/1.50 V), the Y2O3/TiO2 heterostructure memristor exhibits reliable binary RS behavior, featuring low cycle-to-cycle variability, narrow switching voltage distributions, and robust retention (>104 s). In contrast, ternary RS operation is activated under high-electrical stress conditions (+2.0 V/−2.5 V), achieving an enhanced resistance ratio (>102) with distinct intermediate and high-resistance states. Furthermore, the mechanisms and physical models for switching performance under different electrical stresses are thoroughly elucidated. This work introduces a heterostructure strategy to enhance memristor storage capacity, offering a competitive pathway for developing high-density logic hardware systems with advanced memory capabilities.

Hafnium oxide (HfOx) films are highly valued as functional layers in nonvolatile resistive switching (RS) memristors due to their scalability, compatibility with CMOS technology, and high dielectric constant. However, the low reliability of HfOx-based memristors is the key factor hindering their widespread practical applications. Herein, amorphous HfOx (a-HfOx) films are used as the switching layers to construct memristors, and the nitrogen treatment strategy is employed to enhance the switching characteristics. All the fabricated Al/a-HfOx/ITO memristors demonstrate bipolar digital RS behaviors, and specifically, the 500 °C-treated a-HfOx device exhibits highly reliable RS performance, including low cycle-to-cycle variability, concentrated distributions and low operating voltages, long-term retention capacity (>104 s), and good cycle endurance (>200 cycles). The mechanisms and physical models for enhanced switching performance are thoroughly elucidated, revealing that the formation of stable oxygen vacancy–dinitrogen complexes confines the conductive filament path and significantly reduces filament randomness during formation and rupture. This work renders an effective material engineering strategy for widening a path toward designing highly reliable nonvolatile data storage devices with striking switching performances.

Implementation of tunable digital and analog resistive switching (RS) behaviors for memristors is crucial toward their development in data storage and energy-efficient computations. Herein, the CsPbBr3 film-based RS memristors with different switching film thickness and adaptable electrodes are constructed, which are implemented by the solution processed method. All the constructed memristors demonstrate stable bipolar switching behaviors, and in special, the gradual analog RS behavior with multi-level conductance is observed for the thin memristor, making it as the promising artificial synapse. As a comparison, the thick memristor demonstrates digital abrupt RS behavior, accompanying with a high memory window (>103), which can be applied in data storage field. Furthermore, the evolution of conductive filaments, stemming from synergetic bromine vacancies and aluminum atoms, is proposed to clarify the distinguished dynamic changes in the RS process. This work provides an optimization strategy to tune RS behavior for CsPbBr3-based RS memristors and paves the way for both data storage and artificial synapse applications.

The volatile and nonvolatile resistance switching (RS) characteristics can be, respectively, used for the selector and memristor, which have received much attention. Thus, it is essential to find a simple and effective method to control the specific RS behavior of NbOx materials due to the co-occurrence of memory RS and threshold RS behaviors. Here, the NbOx film with a thickness of 100 nm was prepared by magnetron sputtering at 80 °C. The tungsten steel tip/NbOx/Pt device exhibited the co-existence of memory RS and threshold RS behaviors. By properly regulating compliance current (Icc), the specified memory and threshold RS behaviors were observed: the memory RS behavior occurred with an Icc of 5 mA, the threshold RS behavior occurred with an Icc of 10 mA, and integrated one selector-one-RRAM (resistive random access memory) (1S1R) RS behavior occurred with an Icc of 50 mA. Moreover, individual RS behavior showed good performance, e.g., good stability of memory RS, good repeatability and concentrated voltage distribution of threshold RS. The memory RS behavior occurred mainly due to the formation and fracture of oxygen vacancy conductive filaments (CFs). Meanwhile, mediated by local Joule heating, thermally induced conductivity change was responsible for the threshold RS behavior. Under an Icc of 50 mA, the oxygen vacancy CFs and a thermally induced conductivity change triggered the 1S1R RS behavior, which significantly suppressed the leakage current in RRAM 3D integrated structures. This work provides an efficient and convenient method for modulating and obtaining the desired RS behavior and better understanding the conversion mechanism between them.