Many oxide materials such as NiO_x 1234, HfO_x 56, Cu₂O 7, Gd₂O₃ 8, TaO_x 910, TiO₂ 11, ZrO₂ 1213, AlO_x 141516, Na_0.5Bi_0.5TiO₃ 17, SrTiO₃ 18, and so on have been reported for nanoscale nonvolatile resistive switching random access memory (ReRAM) applications. Basically, a single layer of resistive switching material has been investigated by many groups. Due to stochastic nature of the conducting filament in a single binary-oxide resistive switching material 19, it results poor switching cycles. To improve the ReRAM device performance, many nano-dots (NDs) such as ruthenium (Ru) 19, gold (Au) 20, copper (Cu) 21, nickel 22, and so on have been reported. External electric field will control conducting filament formation/rupture through the NDs in the same pathway 22, which can improve the switching parameters. However, it is still an issue to improve the resistive switching parameter using NDs in the insulating materials. Even though many types of NDs have been reported to improve the resistive switching memory characteristics, the core-shell iridium-oxide (IrO_x) NDs embedded in high-κ Al₂O₃ film for IrO_x/Al₂O₃/IrO_x-NDs/Al₂O₃/WO_x/W ReRAM device have not been reported yet. The electrically conducting iridium (Ir) or iridum-oxide (IrO_x) metal is an attractive one due to its high work function (> 5 eV) and noble metal. Furthermore, formation polarity is also one of the important keys to improve the resistive switching performance because it can also control the conducting pathways through the NDs. It is known that the Gibbs free energies of the Al₂O₃, IrO₂, WO₃, and WO₂ films are -1,582.3 23, -183.75, -506.63, and -526.0 kJ/mole, respectively, at 300 K 24. It is suggested that the high-κ Al₂O₃ film will be easily oxidized than those of the WO₃, WO₂, and IrO₂ films. Under external bias, the oxygen ions (O^2-) will be generated from the Al₂O₃ layer, while oxygen vacancies (V_o) will be supplied from the WO_x layer. It is known that the Al₂O₃ film has negative-type defects. The defects density in the Al₂O₃ film will be increased by including core-shell IrO_x-NDs. It is expected that the core-shell IrO_x-NDs embedded in the Al₂O₃ films will not only increase the defect density but also will guide the conducting filament through a single nano-dot, which can be controlled also by external formation polarity. Due to noble metal of Ir, it will not be oxidized under external bias; however, the surface of the IrO_x-ND will control the oxygen-vacancy filament. Even the IrO_x metal, it will also not be oxidized under external bias. Due to IrAlO mixture on the IrO_x-ND surface, this will control the filament formation/rupture. In order to understand the defective IrO_x-NDs embedded in the Al₂O₃ films, the memory capacitors in an IrO_x/Al₂O₃/IrO_x-ND/Al₂O₃/SiO₂/n-Si structure have also been reported. Improved resistive switching parameters such as set/reset voltages, low resistance state (LRS)/high resistance state (HRS), switching cycles, read endurance of > 10⁵ times, and extrapolated 10 years data retention at 85°C of the IrO_x-ND ReRAM device under positive formation polarity (PF) have been reported. Formation polarity dependent resistive switching phenomena have been also explained by oxygen ions migration under external bias. The ReRAM device with layer-by-layer has been observed by high-resolution transmission electron microscopy (HRTEM), energy dispersive X-ray spectroscopy (EDX), and X-ray photoelectron spectroscopy (XPS) analyses.

Resistive switching memory characteristics using IrO_x-NDs in an IrO_x/Al₂O₃/IrO_x-NDs/Al₂O₃/WO_x/W metal-insulator-metal structure were investigated. First, tungsten (W) bottom electrode (BE) with a thickness of approximately 100 nm was deposited by sputtering on SiO₂ (200 nm)/Si substrate. To design the ReRAM device, the SiO₂ layer with a thickness of approximately 150 nm was deposited. Different via sizes from 0.6 × 0.6-8 × 8 μm² were patterned using standard lithography. Then, photoresist was coated and opened the active and top electrode (TE) regions. Then, high-κ Al₂O₃ film with a thickness of approximately 3 nm was deposited by radio frequency (rf) sputtering system using Al₂O₃ target. Then, the IrO_x nano-layer with a nominal thickness of approximately 1.5 nm was also deposited by rf sputtering system. The Ir target was used during deposition of the IrO_x layer. The ratio of argon (Ar) to oxygen (O₂) gasses was 1:1 (i.e., 25 sccm:25 sccm) during deposition. Then, high-κ Al₂O₃ film with a thickness of approximately 3 nm was deposited using the same conditions above. Finally, the IrO_x metal as a TE with a thickness of approximately 350 nm was deposited. The resistivity of the IrO_x metal layer was approximately 600 μΩ cm, which is similar to reported result (approximately 500 μΩ cm) of IrO₂ 25. This suggests that the IrO_x is a metallic film. During deposition of the Al₂O₃ and IrO_x nanolayers, the bottom W electrode was partially oxidized, which results in a WO_x film on the W surface. To fabricate the ReRAM device, a lift-off process was used. Schematic view of our ReRAM device is shown in Figure 1. Microstructure of all layers in the ReRAM device was investigated by HRTEM. Material compositions were studied by XPS with Al K_α monochrom X-ray and energy of 1,486.6 eV. X-ray with a small-diameter of 500 μm was focused on IrO_x/Al₂O₃/WO_x/W structure. The surface spectra were taken by etching the layers one-by-one. All spectra were calibrated by C1s spectrum at an energy of 284.6 eV. Resistive switching memory characteristics were measured by HP 4156C semiconductor parameter analyzer {Agilent Technologies, 5301 Stevens Creek Blvd, Santa Clara, CA, 95051, USA} for a device size of 0.6 × 0.6 μm².

Furthermore, charge-trapping behaviors of the IrO_x-NDs in an IrO_x/Al₂O₃/IrO_x-NDs/Al₂O₃/SiO₂/n-Si metal-insulator-semiconductor (MIS) structure were also investigated. After cleaning the n-type Si (100) wafers, a SiO₂ tunneling oxide with a thickness of approximately 3.6 nm was grown. Then, high-κ Al₂O₃ tunneling oxide with a thickness of approximately 1.2 nm was deposited. So the thickness of stack tunneling oxide (SiO₂ + Al₂O₃) was approximately 4.8 nm. The IrO_x metal gate electrode with a gate area (Φ) of 3.14 × 10^-4 cm² was deposited by a shadow mask. Capacitance-voltage (C-V) hysteresis characteristics of the MIS structure were measured by HP 4284A LCR meter {Agilent Technologies, 5301 Stevens Creek Blvd, Santa Clara, CA, 95051, USA}.

Figure 2a shows the size and density of the IrO_x-NDs in MIS structure. The thickness of Al₂O₃ blocking oxide is 8 nm. As-deposited IrO_x nanolayer shows self-assembly NDs because as-deposited layer is very thin (1.5 nm). Isolated IrO_x-NDs are observed clearly (Figure 2b). The IrO_x-NDs have high density of approximately 1 × 10¹³/cm² (Figure 2c), which results a large memory size of > 60 Tbit/sq in. The nano-dots are core-shell structure (Figure 2d). Core region of the NDs is Ir-rich, and shell region is oxygen-rich or IrAlO mixture. It is indicated that V_o can be observed in the core region, and oxygen accumulations (O^2-) can be observed in the shell region. The IrO_x-NDs have a nanoscale average diameter of approximately 1.3 nm (Figure 2e). The hole or electron trapping can be measured from the C-V hysteresis characteristics of the MIS capacitors. Figure 3a shows high-frequency (1 MHz) clockwise C-V hysteresis characteristics of the MIS capacitors. A small equivalent oxide thickness of approximately 8 nm is obtained. A large memory window of approximately 1.2 V is observed under a sweeping gate voltage of ± 7 V for the IrO_x-ND device, while a small memory of 0.3 V is observed for the pure Al₂O₃ chrage-trapping layers (Figure 3b). The neutral flat-band voltage (V_FB) is found to be +0.15 V for the IrO_x-ND devices (not shown here). The hole and electron trapping densities can be calculated using equation in reference 26. High hole-trapping densities for the pure Al₂O₃ and IrO_x-NDs capacitors are found to be approximately 5.6 × 10¹¹/cm² and approximately 3.1 × 10¹²/cm², respectively (Figure 3b). Almost no electrons are trapped in the core-region IrO_x-ND under positive voltage, but holes are trapped on the shell region under negative bias on the IrO_x electrode. This suggests that the holes can be trapped on the nano-dot surface easily because the surface has negative-type defects. These trapping holes as well as oxygen vacancy on the surface of the NDs will lead the conducting filament as well as resistive switching memory parameters under formation polarity, which have been explained below.

Figure 4a shows typical HRTEM image of a ReRAM device with a size of 8 × 8 μm². The stack switching layers with self-assembly IrO_x-NDs are observed clearly in Figure 4b. The IrO_x-NDs with a thickness of approximately 1.5 nm are sandwiched in between high-κ Al₂O₃ layers with a thickness of approximately 3 nm. A small size of approximately 1.3 nm in a diameter is observed. The IrO_x-NDs shows crystalline, while the high-κ Al₂O₃ film shows amorphous. The WO_x layer with a thickness of approximately 14 nm is also observed, which is expected from our deposition. The WO_x film shows polycrystalline. All layers are also confirmed by XPS analysis below.

Figure 5a shows the XPS spectra of the Ir4f core levels with the Ir₂O₃ and IrO₂ compositions. The peak fitting was performed using Shirley background subtraction and Gaussian-Lorentzian functions. Two peaks positions of the X-ray photo-electron spectra are strong indication of the presence of Ir₂O₃ 4f_7/2 and Ir₂O₃ 4f_5/2 spin orbital components with peak energies centered at approximately 62.04 and approximately 65.12 eV, respectively. The peak binding energies of the IrO₂ 4f_7/2 and IrO₂ 4f_5/2 doublets are centered at approximately 63.70 and approximately 66.42 eV, respectively. The IrO₂ peak intensity is smaller than that of the Ir₂O₃, which is evidently prove the presence of IrO_x NDs. This IrO_x-ND is embedded in the high-κ Al₂O₃ films, which is also confirmed by Al2p and O1s signals (not shown here). Choi et al. 27 reported the peak energy positions for IrO₂4f_7/2 and IrO₂4f_5/2 which are located at 61.9 and 64.9 eV, respectively. The presence of metallic W and oxygen-rich WO₃ can be confirmed by XPS analysis of W4f spectra (Figure 5b) in 14 nm-thick WO_x layer. The peak binding energies of the W4f_7/2 and W4f_5/2 electrons are centered at 30.85 and 33.06 eV, respectively. The energy peak separation is 2.2 eV. The positions of binding energies for WO₃4f_7/2 and WO₃4f_5/2 electrons are located at 35.43 and 37.56 eV, respectively, and these values are similar to the reported results at 35.0 and 37.6 eV for the WO₃4f_7/2 and WO₃4f_5/2 electrons, respectively 28. The W sub-oxide (WO_x) located at 35.0 eV could be formed during the deposition process, and the oxygen-vacancy could be found in the WO_x films. Due to both the IrO_x-NDs in the high-κ Al₂O₃/WO_x bilayer structure and high charge-trapping density, improved resistive switching memory performance has been explained below.

Figure 6a shows electrochemical formation process of our IrO_x/Al₂O₃/IrO_x-NDs/Al₂O₃/WO_x/W ReRAM device. The pristine devices show low leakage current. The currents increase with increasing voltage, and it is limited by current compliance (I_CC) of 500 μA. The average leakage current under negative formation (NF) is 229.7 pA at a read voltage (V_read) of -2.0 V, while it is 673.3 pA under PF at a V_read of +2.0 V. Asymmetry values of the leakage currents under NF and PF polarities are quite reasonable due to the different work function (Φ_m) values of IrO_x [(Φ_m)_IrOx ≈ 5.2 eV 14, and of W [(Φ_m)_W ≈ 4.3-4.91 eV 29] electrodes. Considering the electron affinities (χ) of 3.33 eV 30 for WO₃ and 1.25 eV for Al₂O₃ layers, the values of the effective barrier heights (Φ_b) are found to be [Φ_b = (Φ_m)_W-(χ)_WO3] 1.3 eV and [(Φ_b) = (Φ_m)_IrOx-(χ)_Al2O3] 3.95 eV, respectively. So the electron injection at the IrO_x/Al₂O₃ interface will be lower than that of the W/WO_x interface. In consequence, the currents under PF increase rapidly which results a lower and tight distribution of formation voltages as compared to NF (6-7 V vs. -7 to -9 V). This suggests that lower formation voltage can protect device degradation 22. During the formation process, the high electric field breaks the Al₂O₃ film and forms the Al-O bonds with oxygen vacancy as well as oxygen vacancy filament. In consequence, oxygen ions (O^2-) will be released from the conduction channels. These oxygen ions will move from the Al₂O₃/IrO_x-NDs/Al₂O₃ stacks toward WO_x layer under NF because the TE has negative polarity, and the BE is grounded during measurement. Under PF, the oxygen ions will move toward the TE because the TE has positive polarity. In this case, the oxygen ions will be accumulated at the Al₂O₃/TE interface, i.e., oxygen-rich AlO_x film. So the oxygen vacancy filament can be formed in the Al₂O₃/IrO_x-NDs/Al₂O₃ stack. The WO_x layer will behave as a conductiong layer under PF. After the first reset operation, the ReRAM devices under both NF and PF show bipolar resistive switching characteristics. Figure 6b shows typical I-V characteristics with 50 DC cycles at room temperature (RT), as shown by arrows: 1→4 after NF. The LRS currents are fitted with ohmic, and the HRS currents are fitted with Schottky emission (not shown here). Large variations of the set and reset voltages at a V_read of -0.2 V are -1.7 to -3.6 V and +2.0 to +3.2 V, respectively (Figure 6d) and large variations in LRS and HRS are also observed (Figure 6e). Typical resistance ratio under NF is approximately 10³. Under PF, excellent DC I-V switching cycles are observed (Figure 6c). A tight distribution of the set voltages from +2.0 to +2.9 V as well as reset voltages is observed (Figure 6d). Average HRS and LRS values are approximately 2 MΩ and 20 kΩ, respectively (Figure 6e). So an acceptable resistance ratio of approximately 10² is obtained under PF. Even at high temperature of 85°C, excellent consecutive 100 DC switching cycles is observed under PF (Figure 6f). To investigate the current conduction mechanism under PF, I-V curves of LRS and HRS are fitted linearly in log-log plot, which are in agreement with the trap-charge controlled space charge limited current model for both LRS and HRS (not shown here).

Figure 7 shows typical AC endurance characteristics for both NF and PF devices up to 10³ cycles. Both the devices are programmed at a voltage of ± 4 V, a low I_CC of 500 μA, and a V_read of ± 0.1 V. Inferior AC cycles are observed for the NF devices, where as the PF devices show excellent switching stability. Improved resistive switching parameters of the IrO_x/Al₂O₃/IrO_x-NDs/Al₂O₃/WO_x/W structure under PF can be explianed as follows. For the NF devices, the oxygen ions will move from the Al₂O₃/IrO_x-NDs/Al₂O₃ stack toward the WO_x layer under a negative voltage (-V < V_set) on the TE or the oxygen vacancies will move from the WO_x layer toward the Al₂O₃/IrO_x-NDs/Al₂O₃ stack (Figure 8a). It seems that the oxygen ions will be hidden in to the WO_x layer. It results the oxygen vacancy filament through the IrO_x-NDs as well as LRS. Under reset operation (+V > V_reset), the oxygen ions will move from WO_x layer toward Al₂O₃/IrO_x-NDs/Al₂O₃ stack and oxidize the conducting filament as well as HRS (Figure 8c). Due to the lower barrier height at the W/WO₃ interface, the injected electrons will be higher. This result indicates a higher O^2-ions migration and randomized rupture of the conducting filaments. According to our previous report, improved resistive switching characteristics of the IrO_x-ND ReRAM devices are observed as compared to the pure Al₂O₃ ReRAM device under NF due to oxygen-vacancy filament confinement by the IrO_x-NDs 31. Even though IrO_x-NDs are embedded in the Al₂O₃ switching material, but higher O^2-ions migration will reset HRS dispersion resulting a variation of switching cycles. This suggests that the controllable reset operation is also a crucial issue to improve the repeatable switching cycles. So oxygen ions migration leads to the filament formation/rupture in the Al₂O₃/IrO_x-NDs/Al₂O₃ stack under NF, which results an inferior switching cycles. In previous literature, similar phenomena using oxygen vacancies generation and recombination to form/rupture a conducting filament to switch the LRS and HRS are also reported in some other ReRAM devices 3233. For the PF devices, the oxygen ions will move from the Al₂O₃/IrO_x-NDs/Al₂O₃ stack toward the Al₂O₃/TE interface under a positive voltage (+V > V_set) on the TE or the oxygen vacancies will move from the Al₂O₃/TE interface (oxygen-rich AlO_x) toward Al₂O₃/IrO_x-NDs/Al₂O₃ stack and sets LRS (Figure 8b). This suggests that an insulating layer as a series resistor with LRS will be created at the Al₂O₃/TE interface under set operation, which will limit electron injection at the W/WO_x interface. This result indicates a thinner and stable conducting filament formation, as shown in Figure 8b. So the LRS of the PF devices is higher than that of the NF devices (Figure 6e) due to series insulator (oxygen-rich AlO_x layer). Under reset operation (-V < V_reset), the filament can be ruptured by oxygen ions migration from the Al₂O₃/TE (i.e., oxygen-rich AlO_x layer) and oxidize the partial filament, which results also a stable HRS because the electron injection (O^2- ions migration from oxygen-rich AlO_x layer) from the TE is very low, and this will be limited by the IrO_x-ND/Al₂O₃/TE region, as shown in Figure 8d. The conduction filaments in the WO_x/Al₂O₃/IrO_x-NDs region will remain after reset operation. Due to this remaining filament, the next set operation will be also easier. Due to this stable reset, repeatable switching cycles are expected for the PF devices. So the HRS of the PF devices is lower as compared to the NF devices (Figure 6e), due to shorter length filament oxidized. Basically, the limitation of electron injections (O^2- ions migration) can control the charge trapping (filament oxidation)/de-trapping (filament formation) through the IrO_x-ND boundary. A similar study has been reported by other group 34. For the PF devices, the filament can be formed or ruptured repeatedly than that of the NF devices. This reveals that the filament formation/rupture is more controllable (i.e., localized filament) when it is formed at the IrO_x-NDs/oxygen-rich AlO_x interface region because it has short length, and an oxygen-rich AlO_x layer is in series which can protect the higher electron injection. So the improved resistive switching characteristics are expected for the PF devices. It is noted that a small size (1.3 nm) NDs are observed, and the filament diameter can be adjusted through a single ND in future. It indicates that the device size can be down scale to 1.3 nm in future for advanced high-density nonvolatile memory applications.

Figure 9a shows stable resistance ratio with an excellent non-destructive read up to 10⁵ times for the PF devices. No fluctuation can be detected in LRS (25 kΩ) and HRS (2.2 MΩ) values at a V_read of +0.2 V. Figure 9b shows the retention characteristics of the PF devices at 500 μA with a program/erase voltage of ± 4 V. The data retention ability is quite remarkable for both at RT and 85°C. Extrapolated 10 years of data retention with a resistance ratio of > 10² at RT and > 10 at 85°C are obtained. It is expected that a single IrO_x-ND in the Al₂O₃/IrO_x-NDs/Al₂O₃/WO_x stack can control the resistive switching characteristics in future.

Formation polarity dependent ReRAM devices using IrO_x/Al₂O₃/IrO_x-NDs/Al₂O₃/WO_x/W structure have been investigated. The devices are confirmed by HRTEM, EDX, and XPS analyses. The improved resistive switching parameters such as stable set and reset voltages, HRS/LRS with a acceptable ratio of > 10, long read endurance of 10⁵ times, and extrapolated 10 years data retention at 85°C under PF are obtained owing to the filament controlled at the IrO_x-NDs/oxygen-rich AlO_x interface. Due to the high-charge trapping density, controllable O^2- ions migration and nano filament formation through the core-shell IrO_x-ND under PF, the improvement of the resistive switching memory characteristics is evident and the switching mechanism is explained successfully. A large memory size of > 60 Tbit/sq in. can be obtained. It is expected that the scalability potential of this resistive switching memory device could be the diameter (approximately 1.3 nm) of a single IrO_x-ND in future.

The authors declare that they have no competing interests.

WB carried out the device fabrication, measurement, and data analysis under the instruction of SM. CSL helped to WB for experiments. YYC performed the TEM measurements of MIS capacitors under the instruction of JRY and SM. TCT performed the XPS measurements. HYL, WSC, FTC, MJK, and MJT performed via structure design and fabrication. All the authors contributed to the preparation and revision of the manuscript, and approved its final version.