Effect of dysprosium and lutetium metal buffer layers on the resistive switching characteristics of Cu–Sn alloy-based conductive-bridge random access memory

In recent years, conductive-bridge random access memory (CBRAM) devices have been intensively investigated as an exciting alternative for next-generation memory devices because of their high switching speed, low power consumption, high on/off ratio, and complementary metal oxide semiconductor compatibility [1–3]. The resistive switching (RS) of CBRAM is defined by the formation/dissolution of the conductive filament (CF), which acts as a bridge between the electrochemically active metal electrode (Cu, Ag) and the inert bottom electrode (Pt, W) [4–6]. However, uncontrolled growth of the CF leads to poor RS performance. Importantly, it is impossible to remove the CF entirely on applying a negative voltage; this affects the high resistance state (HRS) as well as the endurance of the device [7–9]. Moreover, the random growth of the CF in the RS layer leads to the growth of multiple CFs, considered to be constraints to device retention [10–12]. Therefore, several attempts have been made to overcome the above limitations in CBRAM. Among these, You et al embedded Ag metal nanocones into the RS layer by a solvent-assisted nanotransfer method [13]. Sun et al guided the growth of CFs by nanoindentation in RS media [14], and Shin et al used 2 μm pyramids of Ag/Al₂O₃/Pt [15]. Saremi et al used photodoping to control the Ag filament in chalcogenide glass [16–18]. Although this device shows better-quality RS performance with well-controlled filament growth, it requires specific material systems and fabrication processes.

Several reports have recommended that bilayer RS devices, such as those using AlO_x/Al₂O₃, SiO_x/TiO₂, TaO_x/Ta₂O_5–x, and a Ti/HfO₂ layer, show better performance than single-layer RS devices [19–22]. Moreover, bilayer devices with a metal buffer layer have attracted more attention because of the formation of an interfacial oxide layer between the metal (Ti, Hf, Ta, etc) and RS layers, which is vital to the formation of a single CF at the oxygen vacancies in the RS layer [23, 24]. In this study, we propose a method to control the random growth of CFs by reducing the concentration of copper in the active electrode using Cu_x–Sn_1–x alloys. Also, by inserting a lanthanide metal buffer layer dysprosium (Dy)/lutetium (Lu) between the newly modified active electrode (Cu_x–Sn_1–x) and RS layer (Al₂O₃), we achieve improved RS performance. The optimized devices show excellent RS with low SET/RESET voltage operation, large on/off resistance states (10⁶), high endurance, and stable data retention for 10 years at 85 °C. These improvements were clearly shown by the x-ray photoelectron spectroscopy (XPS) studies at the interface of the metal/Al₂O₃, and in addition, the conductive-atomic force microscopy (C-AFM) determines the filament formation in both devices.

2.1. CuSn- and CuSn/Lu or Dy-based CBRAM device fabrication

The schematic of the copper–tin (CuSn) alloy-based CBRAM is shown in figure 1(a).(inset). First, the bottom electrode of 10 nm Ti/20 nm Pt was deposited on a Si substrate with an e-beam evaporator (Daon-1000E; VTS). The deposition rate was approximately 0.5 Å s⁻¹ at a pressure of about 10⁻⁷ Torr, and the Pt/Ti/Si substrate was cleaned using an ultrasonic bath containing acetone and isopropyl alcohol. Next, a 3 nm Al₂O₃ thin film serving as the RS layer was deposited by laminar-flow atomic layer deposition at 300 °C. The metal precursor was trimethyl aluminum with a 1 s precursor time and the H₂O oxidant time was 1.5 s. A lanthanide metal thin film (Lu or Dy) was layered on to the Al₂O₃ film by an e-beam evaporation, at the rate of 0.5 Å s⁻¹. Subsequently, a Cu_x–Sn_1–x alloy-based active electrode with a different copper concentration (0.13 < x < 0.55) was deposited sequentially using an e-beam evaporator within a 30 × 30 μm² square pattern of the shadow mask. In addition, for comparison, devices without the lanthanide layer (Dy or Lu) were fabricated, as shown in figure 2(a).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

2.2. C-AFM measurements

2.3. Electrical & chemical characteristics

The CuSn alloy was used as the top electrode to control the amount of copper diffusion into the RS layer (Al₂O₃). Initially, we investigated the effect of different compositions of Cu_x–Sn_1–x alloy (0.13 < x < 0.55) on the RS performance, as shown in figure 1(a). The typical I–V relationships of the Cu_x–Sn_1–x/Al₂O₃/Pt/devices with an external current compliance of 100 μA were measured to confirm the RS performance of the Cu_x–Sn_1–x (0.13 < x < 0.55) alloy-based CBRAM devices. The optimal device was found to have a composition of Cu_0.27–Sn_0.73, which was sustained for more than 50 cycles of DC endurance, as shown in figure 1(b), to effectively control Cu diffusion and improve the RS performance of the device.

We inserted a high-density lanthanide metal (Dy: 8.55 g cm⁻³ or Lu: 9.84 g cm⁻³) as a buffer layer between the Cu–Sn active electrode and Al₂O₃ layer, as shown in figure 2(a). Unlike previously reported metal buffer layers [23, 25, 26], the lanthanide metal buffer layer has a high density to control the penetration of Cu. Also, owing to the low electronegative potential of these metals, a significant interfacial oxide layer can be formed between the active electrode and the RS layer, which could act as an oxygen scavenger. The effect of the Dy or Lu metal buffer layer was investigated using XPS to identify the reaction of the lanthanide metals at the interface with the Al₂O₃ layer. Figures 2(b), (c) depict the binding energy of the core element (Lu–O, Dy–O, Al–O, OH) in the O1s peak at the Dy/Al₂O₃ or Lu/Al₂O₃ interface.

The O 1-s spectra at the interfaces of the devices are deconvoluted into three peaks. The lower binding energies of 529.2 eV, 529.4 eV, and 531.1 eV belong to Dy–O, Lu–O and Al–O respectively, and 532.2 eV of binding energy is shown for oxygen vacancies in the Al₂O₃ layer. The percentage of the oxidation is slightly high in Dy–O compared to Lu–O, due to the low electronegativity of dysprosium [27]. The oxidation of the buffer layers at the Dy//Al₂O₃ or Lu/Al₂O₃ interface was determined by the absorption of oxygen from Al₂O₃: the stronger the interfacial reaction between Lu or Dy and Al₂O₃, the greater the number of oxygen vacancies created on the Al₂O₃ layer. In the oxygen vacancies, the CF grew vertically along the direction of oxygen deficiency and a relatively thin CF was formed on the Al₂O₃ layer due to the Lu buffer layer [28, 29]. Further, the C-AFM measurements confirmed the controlled growth of the CFs in the devices. C-AFM is one of the most effective approaches to investigate internal filament formation in a solid electrolyte [30–32]. In the C-AFM measurements, a current amplification module and cantilever for contact-mode operation were used to measure the amplified current between the substrate and the cantilever.

A schematic sketch of the process flow of the C-AFM sampling is shown in figure S1 and is available online at stacks.iop.org/NANO/29/385207/mmedia (supporting information). Figure 3 shows the current mapping image obtained from C-AFM (left part) of the Al₂O₃ layer of devices with the Dy and Lu buffer layers and the corresponding current (right part) profiles of the CFs. The bright region in the C-AFM image denotes the filament area in the Al₂O₃ layer. It is the result found in the local region of the Lu buffer layer/Al₂O₃ interface. The left part of figure 3 shows the current profile of the filament along the x-axis. The diameters of the filament in the Dy and Lu buffer layer-based devices were different, but the maximum currents flowing through the CF were the same (∼120 nA). We verified that a thin CF (from 10 nm to 40 nm) could be formed by inserting a Dy or Lu buffer layer. Figures 4(a), (b) shows the I–V curves of a 2.5 nm thick metal (Dy or Lu) buffer layer-based CBRAM device for different compositions of CuSn in the modified active electrode.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

All the devices showed an on/off ratio improvement and reduced deterioration, as evidenced by the black (first sweep) and the red (10th sweep) lines. The electrical characteristics such as the on/off ratios and operating voltages were different for different compositions of CuSn; we confirmed that the device with the composition Cu_0.27–Sn_0.73 had a relatively uniform resistance distribution and showed stable operation. Figures 4(c), (d) show the DC sweeping of the optimized Cu_0.27–Sn_0.73/Lu CBRAM device for 50 successive cycles. The on/off ratio of the Lu buffer layer device (10⁶) was one order higher than that of the Dy buffer layer device (10⁵), which could be ascribed to the relatively fewer oxygen vacancies in the Lu buffer device, forming a thinner CF due to the low penetration of Cu ions [33, 34]. In addition, we compared the performance of the Cu_0.27–Sn_0.73 devices with and without the buffer layers in terms of variation (cumulative probability) in SET/RESET voltages, resistance states, AC endurance, and long-term retention at 85 °C. Figure 5(a) shows the statistical distributions of the SET/RESET voltage obtained from cyclic curves for the Cu_0.27–Sn_0.73 device with and without the Dy or Lu metal layer. The statistical approach was used to calculate the coefficient of variation (C_v) [35–37] for the average operating voltage of the devices with and without the buffer layer. Devices with Dy and Lu showed a smaller Cv value (SET/RESET: 30.8%/15.6% without buffer, 11.6%/9.5% for Dy buffer, 9.8%/9.2% for Lu buffer) and low operating voltages. The lanthanide buffer layer-based devices showed more uniform operating voltages due to the formation of relatively thin Cu filaments. Figure 5(b) represents the cumulative probability of the low resistance state (LRS)/HRS distribution of the Cu_0.27–Sn_0.73-based CBRAM devices with and without the Dy or Lu buffer layer. The C_v of the HRS for the devices decreased from 51.5% to 30.8% with the Dy buffer layer, and to 11.9% with the Lu metal layer. These data demonstrated the improvement in resistance uniformity and on/off ratio (10⁵–10⁶). From the cumulative probability results, it could be verified that the Lu buffer layer interferes with the Cu-ion injection from the CuSn active electrode and that a thin CF contributed to the improvement in switching performance by controlling the oxygen defects at the interface. The Dy and Lu buffer layers enhanced not only the RS characteristics but also the reliability of the CuSn-based CBRAM devices. AC endurance and retention were also analyzed to assess the operational reliability characteristics of the devices.

Zoom In Zoom Out Reset image size

Figure 5(c) shows the AC endurance of all three different devices, obtained by applying AC pulses using a WGFMU. The applied pulse condition involves a SET pulse (1.7 V/10 μs), read pulse (0.05 V/10 μs), and RESET pulse (−1.7 V/10 μs). The AC endurance performance of the Dy or Lu buffer layer-based devices was very stable up to 10⁴ cycles. We believe that stable endurance in the Dy and Lu buffer layers devices is due to the formation/dissolution of tiny CFs under a uniform electric field, where the local electric field strength is high compared to the randomly distributed and thick CFs [34]. Moreover, we evaluated the retention properties of the CuSn-based devices with lanthanide metal buffer layers at a temperature of 85 °C, as shown in figure 5(d); the results showed that the properties were retained without significant corrosion for 10⁴ s. Also, LRS retention was evaluated to verify the robustness of the Cu–Sn/Lu or Dy buffer at high temperatures.

The ON state retention failure time at three different high temperatures (160 °C, 200 °C, and 240 °C) is shown in figures 6(a)–(c). The measurement revealed that the devices with a Lu/Dy metal buffer layer were more stable than those without a buffer layer. This enhancement in retention could be explained as follows: when metal electrodes with two different standard potentials (Cu: +0.34, Dy: −2.29, and Lu: −2.28) were contacted, the electrode with the lower potential could undergo reduction and release electrons [38–40]. Therefore, the released electrons from the buffer layer were injected into the copper filament, reducing copper oxidation even at high temperatures. Also, the mean time to failure (MTTF) could be estimated using the Arrhenius expression ${\rm{MTTF}}\propto {e}^{\tfrac{{E}_{a}}{{k}_{B}T}},$ where E_a is the activation energy, k_B is the Boltzmann constant, and T is the temperature on the Kelvin scale. The MTTF curves of the devices are shown in figure 6(d). These curves indicated that the devices with the buffer layers were fail-safe over 10 years at 85 °C. These results indicated that the combination of the Lu or Dy buffer layer with CuSn as the modified active electrode could more effectively control the RS performance of CBRAM devices.

Zoom In Zoom Out Reset image size

In this work, we demonstrated a way to control the overgrowth of CFs in the RS layer of CBRAM devices by combining a modified active electrode with a Lu or Dy metal buffer layer. At the optimal combination of Cu_0.27Sn_0.73/Lu or Dy (2.5 nm), the device showed highly stable RS characteristics, such as a high on/off ratio (10⁶ for Lu-based devices and 10⁵ for Dy-based devices) with a long AC pulse endurance of about 10 000 cycles, low average operating SET (0.7 V) and RESET (−0.3 V) voltages, a good balance of HRS/LRS, and good retention up to 10 years. In addition, the Cu-filament formation is understood by the C-AFM. The improvements in the CuSn alloy along with the Lu/Dy buffer layer based RS memory suggests tha the new approach to improve the CBRAM devices for future non-volatile memories.

This research work was supported by the Basic Science Research Program through the National Research Foundation of Korea (NRF) grant funded by the Korean Government (MEST) (NRF-2017R1A2A2A05022574, 2015M3A7B7045496).