Abstract
Volatile threshold-switching (TS) devices have been used as selectors and to simulate neurons in neural networks. It is necessary to find new ways to improve their performance. The randomness of conductive filament (CF) growth and the endurance of the devices are urgent issues at present. Here, we explored embedded Ag nanoislands (NIs) in HfO2-based TS devices to limit the position of the CF and facilitate its growth at the same time. The Au/Ag(2 nm)/HfO2(4 nm)/Ag NIs/Au volatile TS devices exhibited forming-free characteristics with improved endurance compared with the devices without Ag NIs, which was ascribed to the enhanced localization of the electrical field and increased oxygen vacancies in HfO2 induced by the Ag NIs. A mechanism was proposed to explain the volatile TS behaviors of the devices. The Ag NIs and the thickness of the HfO2 layers played key roles in whether the devices required forming. This work shows that the use of metal NIs is an effective and convenient way to improve the performance of TS devices.
1. Introduction
Computer systems based on the von Neumann architecture, where storage and computation are separated, face bottlenecks in big data processing [1, 2]. Brain-inspired neural networks have the potential to process information with much lower power consumption than conventional processors and to move the control from data centers to edge devices. Memristors can be used to realize these neural networks as the basic elements to reduce the power consumption and the footprint of the circuits [3]. According to the resistive state retention characteristics of memristors, they can be divided into volatile threshold-switching (TS) devices [4, 5] and nonvolatile memory devices [6, 7]. The volatile TS devices can leverage their large ON/OFF ratios and low-leakage currents to be used as selectors or combine simple external circuits to simulate partial neuronal functions [8, 9]. When combined with external circuits to simulate neuronal function, the volatile TS devices need to be connected in parallel with an external capacitor and then in series with a resistor [10, 11]. The pulse signal generated by the whole circuit can be encoded and used for neuromorphic network calculations [12, 13]. Recently, it has been proven that volatile TS devices can simulate Hodgkin–Huxley and leaky integrate-and-fire neurons [14–16]. To simulate neurons, volatile TS devices with mechanisms of electrochemical metallization (ECM), metal–insulator transition (MIT), or the chalcogenide phase change materials (PCMs) are explored [17–21]. MIT- or PCMs-based volatile TS devices are based on the transformation between the high-resistance phase and the low-resistance phase of the dielectric layer of the devices [22]. Although they have an extremely fast switching speed, there are still many problems. First, the thermal cross talk of large-scale integration can make the whole circuit unstable [23]. Second, the transition temperature of the MIT-based material is too low, requiring a harsh working environment. Third, the devices have a high-leakage current in the off state (IOFF) [24–28]. The working principle of ECM-based volatile TS devices is that with an applied bias voltage, active atoms such as Cu or Ag are oxidized to metal ions at the positive electrode and diffuse toward the negative electrode, after being reduced by the electrons encountered, the metal atoms finally form a conductive filament (CF) [16, 29]. The resistance of the device changes from the HRS to the LRS, and when the bias voltage is removed or reduced to become insufficient to maintain the CF, the CF will spontaneously break, and the resistance of the device returns to the HRS. At present, there are still some challenges for ECM-based volatile TS devices, such as poor threshold voltage uniformity and endurance. To obtain high performance of the volatile TS devices, relying on patterned controllable Ag nanodots prepared by e-beam lithography, the randomness of CF growth was decreased and the threshold voltage uniformity was improved in HfO2-based TS [20], but the complicated process limited the application. Forming has a heavy impact on the performance of volatile TS devices. The initial forming process can usually lead to large parameter variations, poor endurance, and even worse [30], induce permanent damage to themselves due to the applications of large voltages and currents because excessive metal ions are injected into the dielectric layer so that a stable CF is formed during the forming process [30, 31]. Forming-free HfO2-based volatile TS devices were prepared by rapid thermal processing to eliminate the forming process and improve endurance [32], however, the diffusion of active metals was random and uncontrollable in the rapid thermal process, which is not the best choice for improving the integrated performance of volatile TS devices.
In this work, we explored Ag nanoislands (NIs) in Au/Ag(2 nm)/HfO2(4 nm)/Ag NIs/Au volatile TS devices to improve their electrical performance. It was found that the introduction of Ag NIs made the devices exhibit forming-free TS behavior. The endurance of the devices was improved as compared with that of the devices without Ag NIs. This work demonstrated an effective and convenient method by using metal NIs to eliminate forming, decrease the TS voltage and increase the endurance of ECM-based volatile TS devices.
2. Materials and Methods
2.1. Preparation of the Devices
Figure 1(a) shows the schematic diagram elucidating a device in the cross-point of the bottom electrode (BE) and top electrode (TE). Figure 1(b) shows the SEM vertical view of the cross-point structure. An ∼300 nm thick layer of SiO2 was grown on a cleaned 2-inch wafer using plasma-enhanced chemical vapor deposition. The pattern of the BE (Figure 1(b)) was formed by photolithography, and 100 nm thick Ti and 100 nm thick Au were successively deposited as the BE using magnetron sputtering. Ag NIs with a thickness of 2 nm were prepared using an ultrathin anodic aluminum oxide (AAO) template and electron beam evaporation deposition [33]. In this experiment, the AAO template was used as a mask with a hole diameter of 40 nm, hole–hole center distance of 100 nm, and thickness of 200 nm. After cleaning, the HfO2 layer was deposited using plasma-enhanced atomic layer deposition (PEALD) at 280°C using oxygen plasma and a tetrakis(ethylmethylamino)hafnium precursor. The pattern of the TE (Figure 1(b)) was formed by photolithography, and 2 nm thick Ag and 100 nm thick Au protective layers were deposited sequentially as the TE using magnetron sputtering.
(a)
(b)
(c)
(d)
(e)
2.2. Characterization Methods
Scanning electron microscopy (SEM, Gemini SEM 300, Germany and Hitachi S-4800, Japan) was used to characterize the morphology of Ag NIs and the fracture surface. Grazing incidence X-ray diffraction (GIXRD, Smartlab 9 kW, Rigaku, Japan) with the incident angle of 0.5° and the copper target was used to analyze the crystal structure of the HfO2 film. X-ray photoelectron spectroscopy (XPS, Thermo Fisher K alpha, Thermo Fisher Scientific, USA) was used to analyze the element valence and composition of the HfO2 film, and the diameter of the X-ray beam was ∼100 μm, which was focused on the HfO2 film surface (red circle in Figure 1(b)). For depth analysis, argon ion bombardment was used to etch the HfO2 film. The data were analyzed using the computer software Avantage, and the peaks were calibrated by the C 1 s peak at 284.8 eV. The electrical properties of the devices were measured by using a semiconductor parameter meter (Keithley 4200-SCS, KEITHLEY, Cleveland, USA) and a source meter (Keithley 2636B, KEITHLEY, Cleveland, USA). Two 5 μm diameter tungsten-plated steel probes were connected to the TE and BE.
3. Results
3.1. XRD and SEM Analyses
As shown in Figure 1(b), it can be measured that the linewidths of the BE and TE are ∼21.1 and 20.8 μm, respectively, so the area of the cross-point is ∼438.9 μm2, which is consistent with the designed size. Figure 1(c) shows the SEM morphology of the Ag NIs, which were from the same batch of devices but without deposition of the hafnium oxide layer and TE. The Ag NIs prepared with the AAO template and electron beam evaporation deposition of Ag have a diameter of ∼45 nm, and the center distance between two Ag NIs is ∼103 nm, so it can be estimated that the average number of Ag NIs per square micron is ∼109. The layered structure of HfO2(4 nm)/Ag NIs/Au (area outside the cross-point) observed by SEM is shown in Figure 1(d). Figure 1(e) shows the GIXRD spectrum of Au/Ag(2 nm)/HfO2(4 nm)/Ag NIs/Au device. The hafnium oxide (JCPDS#34-0104) prepared by PEALD has one feature diffraction peak belonging to the (21) crystal plane [34, 35], and strong feature diffraction peaks were not observed possibly due to the special crystal texture of the HfO2 deposited.
3.2. Volatile TS Behavior of the Devices
To analyze the TS behaviors of the devices, the I–V curves of the Au/Ag(2 nm)/HfO2 (4 nm)/Ag NIs/Au device were measured by a Keithley 4200-SCS (Figure 2(a)). The device abruptly switched from the high-resistance state (HRS) to the low-resistance state (LRS) at a Vth of ≈−0.04 V when the negative voltages were swept from 0 to −0.5 V (blue arrows in Figure 2(a)); the device abruptly switched from the HRS to the LRS at a Vth of ≈0.03 V, when the positive voltages were swept from 0 to 0.5 V (green arrows in Figure 2(a)). The electrical states of the device returned to the HRS when the voltages were reduced to ∼0 V, showing typical volatile TS behaviors. The volatile TS behaviors could be repeated for ∼80 cycles under a compliance current (ICC) of 10 μA, and then the device failed by remaining in the LRS. Figure 2(d) shows the cumulative probability distributions of cycle to cycle and device to device in the positive and negative threshold voltages of the Au/Ag(2 nm)/HfO2(4 nm)/Ag NIs/Au devices. The positive threshold voltage distribution ranged from 0.01 to 0.08 V, and the negative threshold voltage distribution ranged from −0.02 to −0.08 V. This was consistent with the fact that NIs could greatly decrease the threshold voltages [36].
(a)
(b)
(c)
(d)
To perform comparative studies on the effect of Ag NIs on volatile TS behaviors, we prepared Au/Ag(2 nm)/HfO2(4 nm)/Ag(2 nm)/Au device with a Ag layer instead of Ag NIs and Au/Ag(2 nm)/HfO2(6 nm)/Ag NIs/Au device (with a different thickness of HfO2). Figures 2(b) and 2(c) show the volatile I–V curves of Au/Ag(2 nm)/HfO2(4 nm)/Ag(2 nm)/Au and Au/Ag(2 nm)/HfO2 (6 nm)/Ag NIs/Au devices, respectively. The devices needed forming voltages of −2.85 and −5.48 V under a ICC of 100 µA to become operational, respectively, then the changing from the HRS to the LRS of the electrical state of the volatile devices occurred at a Vth of 2.11 and 2.41 V when the voltages were swept from 0 to 3 V (the blue curve in Figures 2(b) and 2(c)), respectively. When the voltages were reduced to ∼0 V, the electrical states of the volatile devices changed from the LRS to the HRS (volatile TS behaviors). However, these volatile TS behaviors could be repeated only several times, after which the volatile TS devices failed by remaining in the LRS, and the TS behaviors of the devices also could be repeated only several times under the ICC of 1 and 10 mA, after the forming process (Figure S1(a)–S1(c)).
We further investigated the endurances of the devices under voltage pulses. All devices were measured by a Keithley 2636B instrument. Before the pulse measurements, the Au/Ag(2 nm)/HfO2(4 nm)/Ag(2 nm) /Au devices and Au/Ag(2 nm)/HfO2 (6 nm)/Ag NIs/Au devices were electroformed by voltage sweeping from 0 to −3 V and 0 to −6 V, respectively. The LRS of the devices were read at a pulse (5 V, 200 μs), the HRS of the devices were read at a pulse (0.01 V, 200 μs) after the 0 V lasted for 600 μs (delay time for relaxation) (Figure 3(a)–3(c)). The LRS and HRS of Au/Ag(2 nm)/HfO2(4 nm)/Ag NIs/Au device could still be read out after 5,000 cycles (Figure 3(a)). For the Au/Ag(2 nm)/HfO2(4 nm)/Ag(2 nm)/Au device and Au/Ag(2 nm)/HfO2(6 nm)/Ag NIs/Au devices, the HRS could not be effectively read out after about 55 cycles (in Figure 3(b)) and 500 cycles (in Figure 3(c)), respectively.
(a)
(b)
(c)
These comparison results show that the Au/Ag(2 nm)/HfO2(4 nm)/Ag NIs/Au devices did not need an initial forming process, and the devices had relatively high endurance. The devices prepared with the Ag nanodots on top of the hafnium oxide layer have been reported, but the devices needed a rapid thermal processing at 500°C to eliminate the forming process [32]. We differently used embedded Ag NIs, which were embedded below the hafnium oxide layer in our devices that did not require the forming process in their pristine state. The Au/Ag(2 nm)/HfO2(4 nm)/Ag(2 nm)/Au and Au/Ag(2 nm)/HfO2(6 nm)/Ag NIs/Au devices needed forming, and they had much poor endurances. This indicated that the Ag NIs and the thickness of the hafnium oxide layer played a key role in the volatile-switching behaviors of the TS devices.
3.3. X-Ray Photoelectron Analysis
The computer software Avantage was used to analyze the XPS peaks with fitting by Gaussian and Fourier methods. Figure 4(a) shows the full XPS spectra of HfO2 on SiO2/Si (the XPS measurement was located on HfO2 that was directly deposited on SiO2/Si) and HfO2 on Ag NIs/SiO2/Si (the XPS measurement was located on the red circle area as shown in Figure 1(b), and there were Ag NIs underneath). Figure 4(b) shows the O 1 s feature peaks of HfO2 on SiO2/Si and on Ag NIs/SiO2/Si at the surfaces.
(a)
(b)
(c)
(d)
(e)
(f)
We performed argon ions etching on the HfO2 layer and used XPS to analyze the distribution of oxygen at different depths in HfO2 on SiO2/Si and in HfO2 on Ag NIs/SiO2/Si. Figure 4(c) shows the O 1 s feature peaks of HfO2 on SiO2/Si after the HfO2 layer was etched by argon ions for 7 and 14 s. Figure 4(d) shows the O 1 s feature peaks of HfO2 on Ag NIs/SiO2/Si after the HfO2 layer was etched by argon ions for 7 and 14 s. From the full XPS spectra obtained after etching time of 7 (Figure 4(e)) and 14 s (Figure 4(f)), the peaks of Hf 4 f in the full XPS spectrum indicate that the etching was performed on hafnium oxide layers, and the characteristic absorption peak of Ag 3d orbit could be detected which was from the Ag under the HfO2. The peaks located at ∼530 and 532 eV correspond to lattice oxygen and nonlattice oxygen, respectively [37]. The percentages of nonlattice oxygen in HfO2 on SiO2/Si and in HfO2 on Ag NIs/SiO2/Si at different etching time are listed in Table 1. The high proportion of nonlattice oxygen at the surface of HfO2 might be due to the contribution of adsorbed oxygen on the surface [38]. The amount of nonlattice oxygen decreased at first in HfO2 of the two devices after etching for 7 s. However, the amount of nonlattice oxygen significantly increased in HfO2 on Ag NIs/SiO2/Si after etching for 14 s. It can be inferred that the presence of Ag NIs led to a high concentration of oxygen vacancies that contribute to nonlattice oxygen in HfO2 near the Ag NIs [39], which might be attributed to the reaction between the active metal NIs and HfO2 [33, 36].
4. Discussion
4.1. Mechanism for the Volatile TS Behaviors of the Au/Ag(2 nm)/HfO2(4 nm)/Ag NIs/Au Device
The Ag ions (Ag ⟶ Ag+ + e−) in the top Ag layer first diffused toward the HfO2 layer when a positive voltage bias was applied to the TE [40]. Because the diffusion barrier of Ag ions in HfO2 is higher than that in HfOx, the diffusion rate of Ag ions in HfO2 is slower than that in HfOx [41–43], thus, in HfO2-based switching devices, it is supposed that the Ag ions were reduced to Ag atoms (Ag+ + e− ⟶ Ag) before arriving at the BE, so the Ag CF most likely grew from near the top Ag layer [43, 44]. In the present work, because the Ag NIs had a tip charge effect, i.e., an enhanced localization of the electric field (ΔE1) induced by the accumulated electrons (Figure 5(a)) could be formed at the tip of the NIs [33, 36], the Ag ions diffused faster above the tip of the Ag NIs along the electric field, the reduced Ag formed a CF faster at this position. A preferential CF could be formed under a low bias of ∼0.03 V (Figure 2(a)). The Au/Ag(2 nm)/HfO2(4 nm)/Ag NIs/Au device did not require an initial forming process. After the Ag CF formed, the electrical state of the volatile device switched from the HRS to the LRS (Figures 2(a) and 5(a)). According to the Gibbs–Thomson effect, the instability of the atoms on the fiber surface (Ag CF) would be increased [45], and the Ag CF formed under a ICC of 10 μA was unstable after the voltage bias decreasing because of the Rayleigh instability, the cylindrical-shaped structure of the Ag CF tended to break up easily to minimize surface energy [41, 46]. Therefore, when the voltage bias was reduced to ∼0 V, the CF instantly dissolved, and the electrical state of the volatile device returned to the HRS (Figures 2(a) and 5(b)).
(a)
(b)
(c)
(d)
When a negative voltage was applied to the TE, the Ag ions diffused from the Ag NIs to the TE, which could be accelerated by the oxygen vacancies [41] formed near the Ag NIs (Table 1). Similarly, the Ag ions were rapidly reduced to Ag atoms near the tip of the Ag NIs under the enhanced localization of the electric field (ΔE2) induced by the accumulated Ag ions at the tip of the Ag NIs (Figure 5(c)). The Ag cluster grew and extended upward from the tip of the Ag NIs, and a preferential Ag CF was formed with a low bias of ∼−0.04 V (Figure 5(c)). The electrical state of the volatile device switched from the HRS to the LRS (Figure 2(a)). The CF channel formed under an ICC of 10 μA was unstable after the voltage bias decreasing, when the voltage bias was reduced to ∼0 V, the CF instantly dissolved, and the electrical state of the volatile device returned to the HRS (Figures 2(a) and 5(d)).
The XPS analyses (Figure 4(d) and Table 1) show that the Ag NIs led to higher oxygen vacancy concentrations in the HfO2 layer near the Ag NIs (Figure 5). The oxygen-vacancy defects might promote Ag ion diffusion because of the low-diffusion barrier, which further promoted the formation of conductive channels [41]. The oxygen vacancies in the HfO2 layer also might participate in the formation of the CF channel, so the CF might consist of oxygen vacancies and Ag atoms [47].
For the Au/Ag(2 nm)/HfO2(4 nm)/Ag(2 nm)/Au device (without Ag NIs), the Ag CF might form anywhere through a competing CF formation process. The device needed a forming process to form the CF at −2.85 V, and the positive and negative threshold voltages of the devices had a large variation (Figure 2(b)) from 0.33 to 2.12 V and from −0.8 to −2.2 V, respectively.
4.2. Endurance of HfO2-Based Volatile TS Devices
From the electrical performances evaluated by DC voltage cycle sweeping, it can be found that the volatile TS behaviors of the Au/Ag(2 nm)/HfO2(4 nm)/Ag NIs/Au device, which did not need a forming process, could be repeated 80 times (Figure 2(a)). The Au/Ag(2 nm)/HfO2(4 nm)/Ag (2 nm)/Au device (using a Ag layer instead of Ag NIs) and Au/Ag(2 nm)/HfO2(6 nm)/Ag NIs/Au devices (the thickness of the HfO2 layer increased to 6 nm) needed forming processes, and the volatile TS behaviors for these devices could be only repeated few times (Figures 2(b) and 2(c)). Under the voltage pulse measurement, the LRS and HRS of the Au/Ag(2 nm)/HfO2(4 nm)/Ag NIs/Au device were read out over 5,000 cycles (Figure 3(a)); however, the HRS could not be effectively read out after about 55 cycles (Figure 3(b)) and 500 cycles (Figure 3(c)) for the Au/Ag(2 nm)/HfO2(4 nm)/Ag(2 nm)/Au device and Au/Ag(2 nm)/HfO2(6 nm)/Ag NIs/Au device after the forming process, respectively. This implies that the forming process affects the endurance of the devices greatly, this is because the forming process might result in the excessive injection of Ag in the dielectric layer, leading to a stable Ag CF so that the electrical state of the devices no longer exhibited volatile TS behavior because the devices failed by remaining in the LRS, and therefore leading to a poor endurance [38, 48, 49].
The Ag NIs and the thickness of the HfO2 layers played key roles in whether the devices needed forming, and a forming-free process could avoid excessive Ag injection into the HfO2 layers, thus improving the endurance of HfO2-based volatile TS devices. It can be concluded that limiting the excessive injection of Ag into the switching layer would be an effective method to increase the devices endurance.
5. Conclusion
By a comparison study, it has been demonstrated that the Au/Ag(2 nm)/HfO2(4 nm)/Ag NIs/Au volatile TS devices exhibited forming-free characteristics with increased endurance and low-threshold voltages. These effects can be ascribed to the enhanced localization of the electric field at the tip of the Ag NIs, which were conveniently embedded into the devices with the help of an ultrathin AAO template. The Ag NIs and the thickness of the HfO2 layer played key roles in whether the devices needed forming. The forming-free Au/Ag(2 nm)/HfO2(4 nm)/Ag NIs/Au device could avoid excessive Ag injection into the HfO2 layers and attain high endurance. It can be expected that the effective and convenient NI strategy can be used to further improve the performance of other TS devices.
Data Availability
In our manuscript, we have provided supporting information as a word document, applied for Section 3.2.
Conflicts of Interest
The authors declare that they have no conflicts of interest.
Authors’ Contributions
Fanlin Long was responsible for investigation, writing the original draft, and review and editing. Yichuang Zhang was responsible for formal analysis, writing the original draft, and review and editing. Zhaozhu Qu was responsible for methodology and formal analysis. Peiwen Lv was responsible for methodology, data curation, and formal analysis. Baolin Zhang was responsible for conceptualization, supervision, writing the original draft, and review and editing.
Acknowledgments
This work was supported by the National Natural Science Foundation of China (52162022), the Guangxi Natural Science Foundation (2021JJA160015), Open Fund of Key Laboratory of New Processing Technology for Nonferrous Metal & Materials, Guilin University of Technology, Ministry of Education (19AA-1, 20KF-29).
Supplementary Materials
Figure S1: (a) The I–V curves of the Au/Ag(2 nm)/HfO2(4 nm)/Ag (2 nm)/Au device under the compliance currents of 1 mA (red and blue curves) and 10 mA (green curves). (b) The I–V curves of the Au/Ag(2 nm)/HfO2(6 nm)/Ag NIs/Au device under the compliance currents of 1 mA. (c) The I–V curves of the Au/Ag(2 nm)/HfO2(6 nm)/Ag NIs/Au device under the compliance currents of 10 mA. Figure S2: The SEM image of the Ag film prepared by electron beam evaporation deposition. (Supplementary Materials)