Electroresistance mechanism of memristive Cu₂S thin films

Abstract

Cu₂S has been widely used as the resistive-layer material of memristors with the advantage of low-power consumption since the Cu-ions have high mobility and migrate quickly under the applied electric field or current. However, the resistance switching mechanisms remain unclear as direct experimental evidence is still lacking. Here, we prepare several prototypes of metal-Cu₂S-metal memristors, where the resistive layer is a thin film of polycrystalline Cu₂S fabricated by a simple magnetron sputtering process. By fitting the resistive curves, space charge limited conduction (SCLC) and electrochemical metallization (ECM) are identified in devices using different electrodes and compliance currents. The microstructural evolutions of the devices in their working conditions are then studied by in-situ transmission electron microscopy (TEM) to directly observe the formation and rupture of the electrically conductive pathways. We find the resistance switch at small current compliance is caused by the phase transformation between γ(L)-Cu₂S phase and the Cu-deficiency γ(D)-Cu_2-xS phase (corresponding to SCLC mechanism). Meanwhile, the origin of resistance change at a large current compliance is the formation of metal Cu filaments (corresponding to ECM mechanism). The study reveals the underlying resistive mechanisms of Cu₂S, providing insights for its potential applications as the resistive material in high-performance memristors.

Graphical abstract

1. Introduction

Cu₂S is a low-cost, nontoxic, chemically stable and highly conductive earth-abundant copper chalcogenide [1,2]. It has been widely used in many green energy devices, such as thin-film solar cells, photothermal conversion systems, lithium-ion batteries and thermoelectric devices [[3], [4], [5], [6], [7]]. At room temperature, Cu₂S has a monoclinic structure (a = 15.246 Å, b = 11.884 Å, c = 13.494 Å, α = 90°, β = 116.35°, γ = 90°, and space group P2₁/c), termed as γ(L)-Cu₂S [2,8,9]. At a slightly elevated temperature, Cu₂S is a fast ionic conductor and has a hexagonal β-phase structure (a = 3.89 Å, b = 3.89 Å, c = 6.88 Å, α = 90°, β = 90°, γ = 120°, and space group P63/mmc) [2,[8], [9], [10]], which consist of a rigid sublattice of S and “liquid-like” sublattice of transferable Cu-ions [2]. The high mobility of Cu-ions, attributed to their low migration barrier resulting from the relatively low bonding energy in copper chalcogenides, often leads to a low formation energy of defects, such as Cu-ion vacancies [11,12].

A memristor is a device with a metal-semiconductor-metal (MSM) sandwich structure, and its resistance state can be switched by applying an external bias [13]. It is supposed that the migration of ions and defects under the electric field is the fundamental mechanism behind the resistance switching [[14], [15], [16], [17]]. For memristors using Cu₂S as the resistive layer, the low migration energy of Cu-ions may lead to a relatively low energy consumption for the resistive switching and an enhanced response speed. It is worth mentioning that many memristor devices typically have switching voltages greater than 1 V [[18], [19], [20], [21]]. A lower switching voltage means that the resistance state can be switched with a small electric field or current, thereby reducing energy consumption. Additionally, a relatively low switching voltage can improve the device's response speed, enabling faster switching of the resistance state. However, the atomic mechanisms of ion activation and migration, as well as their effects on resistance changes, are not well understood since it was difficult to directly observe the microstructural evolution at the moment when there was a resistive change.

Transmission electron microscopy (TEM) can obtain both internal and surface information by investigating the cross-sectioned samples [11,22]. With the recent development of in-situ TEM, experiments can be conducted inside the electron microscope, allowing direct observation of the microstructural evolutions under a mechanical, thermal, electrical, or magnetic stimulus [23,24]. Compared to ex-situ TEM [23], in-situ TEM allows the observation of the intermediate states as well, which may provide more valuable insights on the structure-property mechanism [25,26].

In this work, resistive switching memory devices (memristors) were constructed by fabricating a series of approximately 100 nm thick Cu₂S thin films (as resistive layers) and metal electrodes on the [001]-oriented single-crystal silicon wafers using physical vapor deposition (PVD) methods. Electrical measurements revealed that the fabricated three-layer resistive random-access memory (RRAM) devices exhibited relatively low switching voltages (≈ 0.5 V), a narrow distribution of switching voltages, and a high on/off resistance ratio (≈ 10⁵). In-situ transmission electron microscopy experiments were conducted to explore the resistance switching mechanisms, providing clear microstructural evidence explaining the origins of the space-charge limited current (SCLC) and electrochemical metallization (ECM) mechanisms, respectively. The atomic mechanisms revealed in this work demonstrate that Cu₂S will have significant potential in the applications of non-volatile memory technology.

2. Experimental section

2.1. Memristor preparation

Cu₂S and other target materials used in the experiment were purchased from Qi Jin New Material Technology. The XRD pattern of the Cu₂S target is shown in Fig. S1. A 2.6 cm × 2.6 cm square monocrystalline silicon wafer was cut out from a wafer-level monocrystalline silicon wafer with a diamond pen. The cut silicon wafer was immersed in an ultrasonic bath for 10 min with propyl alcohol, anhydrous ethanol and deionized water successively. After ultrasonic cleaning, the silicon wafer was taken out and placed in boiling water of deionized water for 10 min. After cleaning, the wafer was removed and the surface was dried by a spray gun. The silicon wafer was covered with the designed mask plate, and the W bottom electrode of approximately 50 nm was sputtered by direct-current (DC) power supply (power was 30 W, argon (Ar) flow was 50 sccm, and the pressure in the cavity was 0.8 Pa). After the bottom electrode was sputtered, the mask plate was replaced, and the Cu₂S functional layer of about 100 nm was sputtered by radio-frequency (RF) power supply (power was 30 W, Ar flow was 50 sccm, and the pressure in the cavity was 0.6 Pa). After that, the sample was taken out and annealed at 150 °C in the Ar atmosphere for 1 h. Finally, the mask plate was replaced, and the W/Cu top electrode of about 50 nm was sputtered again with a DC power supply (the parameters were the same as the bottom electrode).

2.2. Materials and device characterization

The purchased Cu₂S target was analyzed by X-ray diffraction (XRD) using the D8 Discover X-ray diffractometer equipped with Cu Kα radiation source (λ = 1.54056 A). The I-V characteristics of the prepared memristor devices were measured by a semiconductor parameter analyzer (B1500A, Keysight).

The cross-section of the device was prepared by a focused ion beam (FIB) system (Helios NanoLab G3 UC). The FIB has a Schottky FEG source with a peak resolution of 0.6 nm. It has an adjustable electron beam acceleration voltage ranging from 20 V to 30 kV, and an ion beam acceleration voltage ranging from 500 V to 30 kV. For sample preparation, we initially use an electron beam acceleration voltage of 5 kV to pinpoint the suitable area. Then, the sample is cut and thinned down with an ion beam operating at 30 kV. Throughout the thinning stage, the ion beam current is gradually decreased from 0.79 nA to 80 pA, to ensure the integrity of the sample structure and less sample damage.

TEM/STEM images and electron diffraction patterns are collected from a Thermo Fisher Talos F200S and a Titan Themis G2 scanning/transmission electron microscope (S/TEM), both equipped with a Schottky electron source and the latter with double spherical aberration correctors. The Talos F200 operated at an acceleration voltage of 200 kV in both TEM and STEM mode, with an ultimate point resolution of 0.25 nm and 0.16 nm, respectively. With Talos F200 S/TEM, the images and patterns are collected by a Thermofisher Scientific Ceta 16 M camera, which is capable of capturing images at a rate of up to 25 frames per second. During data collection, the electron beam current was carefully controlled to remain approximately under 1 nA. The Titan S/TEM is operated at 300 kV, with an ultimate point resolution of about 0.07 nm. The TEM images were collected using a Thermofisher Scientific Ceta CMOS camera equipped with a 16-megapixel CMOS sensor. The camera provides high-resolution images, enabling detailed visualization of samples. The EDX spectra were collected using Bruker's four-probe Super-X detection system. During data collection by the Titan S/TEM, the electron beam current was carefully controlled to remain approximately under 0.05 nA. This setup allows for an EDS energy resolution of 132.3 eV under 10 kcps counting conditions.

2.3. Electrical performance test

Memristors with MSM sandwich structure using Cu₂S as the resistive layer were prepared, and their electronic characteristics were measured. For all of the memristors, the I-V characteristics are measured at room temperature using a B-1500A semiconductor characterization system and a probe station. During the measurement, voltages were applied through two W-probes while the bottom electrode was grounded. With a positive bias, the current flows from the top to the bottom electrode.

2.4. Preparation and testing of in situ electrical samples

The in-situ experiments were conducted utilizing the Lightning in-situ TEM Biasing & Heating sample holder by DENSsolutions. Firstly, we utilized the Focused Ion Beam (FIB) technique to carve out a block from the device, measuring 20 µm × 2 µm × 10 µm in size. This block, containing the device with a sandwich-like structure and the underlying silicon substrate, was then welded onto a transfer mechanical arm (FIB-easylift). Subsequently, the in situ powered chip was placed into the FIB chamber, and the sample was attached to the electrode pins of the chip using metal Pt welding, as illustrated in Fig. S2a. To ensure that the current passes through the device along the vertical direction, the silicon substrate in the dashed region of the thin film was removed by ion beam bombardment (Fig. S2b). This step was to prevent the current from flowing laterally through the silicon substrate without passing the top device layer. Subsequently, the focused Ga ion beam was used to thin the central part of the sample down to a thickness of approximately 100 nm, allowing observation under TEM. The chips with welded samples were loaded into a DENS solutions chip holder (Fig. S2c) for transmission electron microscopy, and in situ electrical testing was performed using electrical testing software combined with a Talos F200 S/TEM or a Titan Themis S/TEM.

3. Results and discussions

We fabricated memristors with both Cu/Cu₂S/W and W/Cu₂S/W sandwich structures (Fig. 1a) and measured the I-V curves of the devices (Fig. 1b). The Cu₂S resistive layer was deposited using an RF magnetron sputtering process in an Ar gas atmosphere at a pressure of 0.6 Pa, while the metal electrodes were deposited using a DC magnetron sputtering process in an Ar gas atmosphere at a pressure of 0.8 Pa. By annealing at 150 °C for 1 h in Ar atmosphere after sputtering, we obtain well-crystallized polycrystalline and dense Cu₂S thin films (Fig. 1c). We tested the IV curve of the Cu/Cu₂S/W device at a compliance current of 5 mA for 100 cycles, respectively, as shown in Fig. S3. The device exhibited stable and reproducible bipolar resistive behavior with a concentrated distribution of SET/RESET voltages.

Fig. 1.

Structure of Cu/Cu₂S/W memristor. (a) Schematic of the Cu/Cu₂S/W memristor in crossbar array architecture. (b) Optical microscope schematic of memristor array unit. (c) Cross-sectional TEM image showing the Cu/Cu₂S/W memristor.

The morphology and crystalline structure of the prepared Cu₂S thin films were characterized by scanning transmission electron microscopy (STEM). We employed a focused ion beam (FIB) to prepare cross-sectional samples cut from the devices. The STEM image collected by the high-angle annular dark-field (HAADF) detector indicates that Cu/Cu₂S/W device (Fig. 2a) and W/Cu₂S/W device (Fig. S4) have uniform elemental distribution and smooth interfaces. A high-resolution TEM image (Fig. 2b) reveals that the Cu₂S resistive layer consists of polycrystalline grains, with vertical grain boundaries that traverse the entire layer (Fig. S5a). By fast Fourier transform (FFT) analysis (inset in Fig. 2b), the columnar nanoparticle is identified as monoclinic γ(L)-Cu₂S phase. The phase identification is also confirmed by the atomic HAADF-STEM images taken from the small grains, by imposing the atomic model upon the images as shown in Fig. 2c-d. While γ(L)-Cu₂S is identified as the major phase in the thin film, impurity phases, including the tetragonal phase (T) [27] and face-centered cubic phase (F) [2] are occasionally observed, by their STEM images along the [110] zone axis as shown in Fig. S5b-c.

Fig. 2.

Microstructure of the Cu₂S memristor. (a) Cross-sectional STEM image and EDS mapping of Cu₂S memristor. (b) Cross-sectional high-resolution TEM image of Cu₂S resistive layer. (The red boxed area is the γ(L)-Cu₂S along [010]) (c-d) Cross-sectional high-resolution STEM images of γ(L)-Cu₂S, overlapped with the atomic models (Cu is shown by blue dot, while S by yellow dot).

When memristors are fabricated using different types of electrodes (e.g. inert vs active electrodes) and tested under different current compliance, different resistance switching characters are observed. Firstly, by simply fitting the I-V curves, we find that the W/Cu₂S/W memristors with symmetric inert electrodes follow the space charge-limited current (SCLC) mechanism. The SCLC theory uses the ‘injected carrier’ to explain the change in device conductivity [28,29]. As shown in Fig. 3a, the W/Cu₂S/W memristor employing a symmetric inert W electrode exhibits typical bipolar voltage-current characteristics (Fig. S6a) with a narrow distribution of switching voltages and stable high/low-resistance states (HRS/LRS). As shown in Fig. 3b, where the current and voltage data are subjected to logarithmic processing, the curve can be divided into three segments based on the magnitude of the applied bias: V < V_ON, V_ON < V < V_TFL, V > V_TFL (where V_ON is the transition voltage from ohmic conduction to complete filling, and V_TFL is the voltage at which the trap is completely filled). When the applied voltage is low (V < V_ON), the injected carrier concentration is much lower than the intrinsic carrier concentration, and the device conductivity is mainly governed by the intrinsic carriers. At this stage, the slope of log(I)/log(V) is 1, indicating a linear relationship between current and voltage, in accordance with Ohm's Law. As the applied voltage increases (V_ON < V < V_TFL), the number of filled traps increases, and the injected carriers gradually dominate. At this stage, the log(I)/log(V) slope changes to 2, indicating a quadratic relationship between current and voltage, signifying a transition from Ohmic conduction to Child's law. With a further increase in applied voltage (V>V_TFL), it is supposed that all traps in the Cu₂S functional layer are filled by the injected carriers. Thus, at V_TFL the conductivity experiences a sudden jump and rapidly increases, and the log(I)/log(V) slope is greater than 2. After exceeding V_TFL, the device conductivity continues to follow Child's law, where the log(I)/log(V) slope is 2 [[28], [29], [30]]. At this point, the device has completed the transition from HRS to LRS, and the conductivity of the W-Cu₂S-W memristor has increased by approximately three orders of magnitude (which refers to the SET process). The conduction mode soon after the SET process in the LRS is Ohmic conduction. A reverse bias is needed to initialize the RESET process, to switch the device back to the HRS. As shown in Fig. 3c, it follows Ohmic or Child conduction before reaching the transition voltage (V_T). Once V_T is reached, the conductivity of the device gradually decreases, eventually switching to the HRS. Thus, the RESET process in the W/Cu₂S/W device is a gradual increase in resistance. It is shown that the SCLC mechanism dominates the resistance switching in the W-Cu₂S-W memristor, which is different from the rapid decrease in conductivity caused by the ECM mechanism.

Fig. 3.

I-V hysteresis curve and Logarithmic fitting results of I-V Curve of Cu₂S memristor. (a) I-V diagram of the W/Cu₂S/W device with Icc = 5 mA. (b) log(I)/log(V) curve fitting of the negative-bias region in Fig. a. (c) log(I)/log(V) curve fitting of the positive-bias region in Fig. a. (d) I-V diagram of the Cu/Cu₂S/W device with Icc = 5 mA. (e) log(I)/log(V) curve fitting of the negative-bias region in Fig. d. (f) log(I)/log(V) curve fitting of the positive-bias region in Fig. d. (g) I-V diagram of the Cu/Cu₂S/W device with Icc = 25 mA. (h) log(I)/log(V) curve fitting of the negative-bias region in Fig. g. (i) log(I)/log(V) curve fitting of the positive-bias region in Fig. g.

We then replaced the inert metal W with the active metal Cu as the top electrode and prepared Cu/Cu₂S/W memristors to further investigate the conduction mechanism of the Cu₂S resistive layer. At a current compliance of 5 mA, a positive voltage (i.e. the current flows from the Cu electrode to the W electrode) of up to 1.5 V did not trigger resistive switching. However, resistive switching was triggered when a negative voltage (about −0.6 V) was applied, and its I-V curve is shown in Fig. 3d. It exhibits a bipolar resistance switching behavior (Fig. S6b) similar to the W/Cu₂S/W device (Fig. 3a). The voltage and current were logarithmically processed and subjected to fitting analysis, as shown in Fig. 3e-f, indicating that the Cu/Cu₂S/W memristor also follows the SCLC mechanism. However, unlike the devices with W electrodes, when the positive voltage reaches 0.5 V, the current in the Cu electrode device decreases rapidly by nearly three orders of magnitude. This indicates a significantly faster RESET process in the Cu/Cu₂S/W memristor compared to the W/Cu₂S/W memristor. The resistance switching is clearly not the ECM mechanism, as the SET process can only be accomplished under a negative voltage. In the ECM mechanism, the formation of conductive filament pathways relies on the drift of active copper ions under an electric field and their deposition and reduction on inert electrodes (a process that should have occurred under the positive voltage). Thus, the Cu/Cu₂S/W memristor under a 5 mA current compliance also exhibits the SCLC mechanism. It could be due to the small current compliance (of 5 mA) with which the ECM mechanism could not be triggered.

We then increased the SET current compliance from 5 mA to 25 mA and found that the characteristic I-V curve of Cu/Cu₂S/W memristors conforms to the ECM resistive switching mechanism, which is distinct from the SCLC mechanism observed under low power consumption. The I-V curve under the current compliance of 25 mA is shown in Fig. 3g. The device still exhibits a typical bipolar resistive switching curve (Fig. S6c). Initially, the device is in the HRS with a resistance value of approximately 8 × 10⁴ Ω. When a voltage is applied to the copper electrode at 0.55 V, there is a sharp increase in device conductivity by about four orders of magnitude; concurrently the resistance reduces to 10 Ω. It is evident that the device undergoes a “SET” process under positive voltage, consistent with the ECM mechanism. Conversely, as the negative voltage gradually increases, the device's conductivity continues to decrease until reaching −0.45 V, at which point the device fully reverts to HRS. This implies that the internal conductive pathway gradually ruptures under the negative voltage. The logarithmic transformation and fitting analysis on the I-V curve (Fig. 3h-i) further demonstrate that the resistive switching mechanism follows the ECM mechanism.

Also, we tested the W/Cu₂S/W memristors under a SET current compliance of 25 mA and found it still follows the SCLC mechanism (as shown in Fig. S7). This demonstrates that the ECM mechanism is strongly dependent on the active electrode (which is the source of the metal-ion to form metallic filaments). Thus, for Cu₂S-based devices, there may be two competing resistive switching mechanisms within the device, depending on whether there is an active electrode and the energy consumption.

The microstructural evolutions of the memristors with inert electrode, or active electrode but under a low power consumption, was investigated by in-situ TEM to explore the structural basis of the SCLC and ECM mechanism. Fig. 4a is a low-magnification cross-sectional TEM image of the memristor with Cu/Cu₂S/W structure, which was cut and transferred onto an electric biasing chip (Fig. S2). In order to study the microstructure evolution under the SCLC mechanism, the applied voltage direction was first applied from the W electrode to the Cu electrode (to eliminate the influence of ECM mechanism). Fig. 4b-c show the selected-area electron diffraction (SAED) patterns of the Cu₂S functional layer before and after the bias was applied, respectively. By comparing the two diffraction patterns, it can be observed that the (042) and (14–4) diffraction rings of the γ(L)-Cu₂S phase (indicated by the yellow circles in Fig. 4b) disappear after the SET process. Meanwhile, the (91–1) diffraction ring of γ (D)-Cu₂S (indicated by the white circle in Fig. 4c) appears. Although the lattice spacing of the γ(D)-(91–1) plane is close to that of the γ(L)-(042) lattice plane, they can be distinguished by overlapping the two patterns (Fig. 4d). It is obvious that the (91–1) diffraction ring of the γ(D) phase is slightly smaller than the (042) diffraction ring of the γ(L) phase, which is consistent with the fact that the lattice spacing of the γ(D)-(91–1) is slightly larger than that of the γ(L)-(042). Transformation from γ(L) phase to the Cu-deficiency γ(D) phase is thus identified. Under the applied bias, some of the copper ions in the γ(L) phase migrate and are lost, thus forming γ(D) phase containing a large number of copper vacancies [31]. Because the electrical conductivity of the γ(D) phase is significantly higher than the γ(L) phase [2], the device will be set to the LRS state. We successfully observed the same phenomenon in the W/Cu₂S/W device, as shown in Fig. S8. Thus, a Cu-ion concentration modulated phase transition is found as the origin of the resistive switching when the inert electrode is used.

Fig. 4.

Phase transformation corresponds to the SCLC mechanism. (a) cross-sectional TEM image of Cu/Cu₂S/W memristor (SAED obtained from the black circle). (b) electron diffraction pattern of Cu₂S before powering. (c) electron diffraction pattern of Cu₂S after powering. (d) comparison of diffraction patterns before and after electrification.

When using an active electrode and under high-power consumption, the formation of metal-Cu filaments is directly observed by in-situ TEM due to the migration and reduction of Cu-ions under the applied voltage. Fig. 5a is a cross-sectional TEM image of the sample when the applied voltage is at 0 V. In the experiment, a voltage was gradually applied to the thin film starting from 0 V. At the initial stages, slight strain-induced contrast changes in the Cu₂S grains can be seen (Video S1).

Fig. 5.

Process of forming Cu filament. (a-d) Cross-sectional TEM images of device at 0 V, 1.5 V, 1.6 V and 1.66 V bias. (e-f) Enlarged image of the black shadow in fig. d and EDS element scanning.

When the applied voltage reached 1.5 V, an area with dark contrast appeared on the right side of the functional layer. As the voltage continues to increase, the dark contrast continues to deepen and gradually permeates the entire functional layer. The dark contrast region is rich in Cu as identified by EDS mapping (Fig. 5f). At the same time, part of the electrode on the top of Cu becomes very thin (Fig. 5, Video S2), which indicates that migration of the Cu in the top electrode occurs. This migration occurred very rapidly, which is observed within a single frame of the TEM video recording (i.e. 500 ms in the current experimental setting). Meanwhile, it was observed that many bright nanoparticles were formed along the crystalline boundaries, forming discrete, yet filament-like chains along the direction of the applied voltage (Fig. 6b). The nanoparticle is rich in Cu as identified by EDS mapping (Fig. 6c), and have metal Cu face-centered cubic structure as revealed by the atomic HAADF-STEM image shown in Fig. 6d. Due to the high copper ion mobility in fast ion conductors, copper ions receive electrons and are reduced back to metallic copper when they reach the inert electrode. It is worth noting that metallic copper tends to accumulate at grain boundaries within the functional layer, forming conductive filaments. This suggests that the presence of a substantial number of grain boundaries in the polycrystalline thin film could be one of the benefits contributing to the excellent resistive switching performance of the device [32].

Fig. 6.

Formation of conductive metallic Cu nanofilament corresponds to the ECM mechanism. (a) cross-sectional STEM image of Cu/Cu₂S/W memristor before the application of bias. (b) cross-sectional STEM image of Cu/Cu₂S/W memristor after the application of bias. (c) STEM image of conductive filaments in the functional layer and EDS line scan of the conductive filaments (d) STEM image of copper particles in the functional layer. (e) schematic diagram of the conductive filaments.

Meanwhile, reversible migration of Cu-ions and repeated formation and rupture of the chains formed by metal-Cu nanoparticles are also observed (Video S3). As the applied voltage reached 1.5 V, the copper electrode at the top rapidly migrated into the functional layer. With the voltage increased to 1.6 V, the copper in the TE completely entered the functional layer, reflected in a downward movement of dark contrast regions. When the voltage reached −0.4 V and the electric field reversed, copper within the functional layer began to precipitate due to the electric field. At −1.9 V, the copper within the functional layer completely filled the voids created by the migration of the TE into the functional layer (Video S4). This further confirms that copper ions can reversibly migrate within the Cu₂S functional layer under the influence of an external electric field.

Based on the observations, Fig. 6e illustrates the formation of a conductive pathway, formed by copper ions migrating into the Cu₂S functional layer and then reduced to form Cu metal nanoparticles along the grain boundary, driven by a positive bias. It is thus confirmed that under the active electrode and enough power consumption, the repeated migration and reduction/oxidation are the basis of the electrochemical metallization (ECM) mechanism.

By in-situ transmission electron microscopy characterization, we find that the SCLC mechanism of resistive switching is an intrinsic property of Cu₂S under the influence of an external electric field, which normally requires inert electrodes. However, the resistive switching caused by the electrochemical metallization mechanism depends not only on the intrinsic property of the semiconductor but also on the application of active electrodes (e.g. Ag or Cu).

4. Conclusions

In the study, we correlate the resistive switching mechanism determined from I/V curve processing to that studied by direct observation of microstructural and phase evolution in polycrystalline Cu₂S thin film. Through curve fitting, both the space charge limited current (SCLC) and electrochemical metallization resistance (ECM) change mechanisms are found in Cu₂S resistive switching devices. Space charge limited current mechanism plays a dominant role when the inert electrodes are used or when the power consumption is small even when the active electrode is employed. Under a high-power consumption and an active electrode, the resistive switch is dominated by the formation of conductive filaments through the ECM mechanism. By using in situ TEM to directly observe the dynamic structural changes under applied bias, it is found that the different resistive mechanisms originate from various activation states of the Cu-ions. Under low-power consumption or in the absence of an active electrode, the limited hopping of Cu-ions leads to the phase transition between a Cu-rich and a Cu-deficient phase, causing the change in resistance (which corresponds to the SCLC mechanism). However, under high-power consumption and with a sufficient supply of active ions from the active electrode, massive migration and reduction of Cu-ions lead to the formation of metallic Cu nano-filaments along the grain boundaries (which corresponds to the ECM mechanism). The study shows the potential of Cu₂S as a promising resistive material with the advantages of low power consumption and response speed, once the devices and working conditions are appropriately applied.

Declaration of competing interest

The authors declare that they have no conflicts of interest in this work.

Biography

Jinsong Wu, professor of the State Key Laboratory of Advanced Technology for Materials Synthesis and Processing, and executive director of Nanostructures Research Center at Wuhan University of Technology, is a recipient of the Senior Foreign Scholar Research Fund Program of the National Foundation of China (2022–2024). He had worked at world-renowned electron microscopy centers in France, Germany, and the United States, in electron microscopy research for over 20 years. He is a council member of the Chinese Society of Electron Microscopy. He has primarily focused on advancing in-situ high-resolution electron microscopy, applying it to study the lithiation mechanisms of battery electrode materials and the resistive switching mechanisms of semiconductor memristors. He has published more than 300 scientific papers.