16.2.1 Resistive switching in oxide nanowire

Resistive switching in oxide nanowire was first reported by Kim et al. (2008). They employed NiO nanowire for anionic restive switching. The NiO nanowires were fabricated by the anodized aluminium oxide (AAO) template method as shown in Figure 16.4. The AAO template was fabricated via the two-step anodizing process. First, high-purity aluminium was anodized under a constant voltage in oxalic acid solution. Then the aluminium oxide layer was chemically removed. The anodizing process was repeated, and the AAO template with hexagonally aligned nanopores was obtained. Ni nanowires were then obtained by the electrochemical deposition of Ni into the nanopores of AAO template. Finally polycrystalline NiO nanowires with 70 nm diameter were obtained by thermal oxidization of Ni nanowires for 7 h at 450 °C. The grain size was larger than or comparable to the diameter of NiO nanowires. The thermal treatment was carried out by checking the resistance of the nanowire, and the nanowire resistance was controlled to be 100 MΩ.

The electrical transport property was performed using an individual NiO nanowire device. Prior to the construction of a NiO nanowire device, the AAO template was completely dissolved. The device was fabricated onto a silicon substrate covered by a thermally oxidized SiO₂ layer. The Au/Ti electrodes were patterned onto a single NiO nanowire by conventional lithographic technique. The gap size between electrodes was 1 μm. First, a forming process, which is an initiation process for the resistive switching with a relatively high voltage was performed at 2.5 V. After the forming process, the resistance of the device changed into a conductive state. Subsequently, the device showed the continuous unipolar resistive switching under the positive electric field as seen in Figure 16.5. The SET voltage, the RESET voltage and the RESET current of the NiO nanowire device were 1.2 V, 0.52 V and 2.3 × 10^− 4 A, respectively. Interestingly, a weak dependence of RESET current on the device size (RESET current ∝ A^0.38) was found, indicating that the transport property of NiO nanowire device was dominated by the localized conducting filaments. It should be noted here that the electric field intensity for forming in a 1 μm long NiO nanowire device (25 kV/cm) was almost two orders of magnitude lower than that of 20–300 nm thick NiO thin film device (> 0.1 MV/cm) (You et al., 2006). Because the forming voltage of the thin film device strongly depends on its thickness (Yanagida et al., 2013), the forming voltage over 100 V should be expected for a 1 μm long NiO nanowire device. The small forming voltage might be understood by the percolation nature of conducting filaments through the incompletely oxidized NiO in nanowire or at grain boundary rather than the localized resistive switching at the nanowire/electrode interface. This was experimentally confirmed by the length dependence on the forming voltage. The clear increase of forming voltage (20 V) was observed when a 25 μm long NiO nanowire was employed.

The large voltage for resistive switching was also found in the report by Oka et al. (2009a). Oka et al. first demonstrated the resistive switching in a single crystalline MgO–NiO core–shell nanowire. To fabricate the MgO–NiO core–shell nanowires, the in situ heterointerface fabrication technique was employed (Nagashima et al., 2008a; Marcu et al., 2008; Oka et al., 2009b). First single crystalline MgO nanowires were grown by a vapour–liquid–solid (VLS) mechanism through the pulsed laser deposition (PLD) technique (Nagashima et al., 2007a,b, 2008b; Marcu et al., 2007; Yanagida et al., 2007, 2008a,b). For the VLS growth, Au metal catalyst was utilized to conduct the MgO nanowire growth. The diameter of MgO nanowires can be definitely controlled by the size of metal catalyst. After the MgO nanowire growth, NiO was deposited onto the nanowire at 300 °C and 0.1 Pa of oxygen partial pressure. Because MgO (a = 4.21 Å) and NiO (a = 4.17 Å) are both cubic structure with a lattice mismatch of 0.7%, a single crystalline NiO shell layer was epitaxially grown onto MgO nanowires. The fabricated core–shell oxide nanowires were composed of 10 nm diameter of MgO nanowire and 10 nm thickness of NiO shell layer, resulting in 10^− 3 μm² of cross-sectional that are of NiO for resistive switching.

The electrical transport of the individual MgO–NiO nanowires was carried out with Pt/MgO–NiO nanowire/Pt-coated cantilever using conductive atomic force microscopy (C-AFM) (Oka et al., 2009a). The nanowires were first developed onto 300 nm SiO₂ coated silicon substrate, then Pt electrode was deposited by using a metal mask to create the electrical contact with nanowire. After checking the exact position of nanowire on the substrate by tapping mode AFM, the electrical measurement using C-AFM was carried out at a position around 1 μm away from the Pt electrode. Before the resistive switching, the resistance of NiO was around 10¹¹ Ω, which were three orders of magnitude higher than that of polycrystalline NiO nanowire reported by Kim et al. (2008). This is because the NiO shell layer was fully oxidized, and the defect concentration of single crystalline NiO shell layer might be less than that of polycrystalline NiO nanowire in Kim et al. (2008). Clearly, the MgO–NiO core–shell nanowire exhibited the bipolar resistive switching behaviours with the current range of 10^− 8 A (Figure 16.6). The SET and RESET voltage were, respectively, around 15 and 10 V, which were significantly higher than the report in Kim et al., 2008. The nonvolatility of the MgO–NiO nanowire resistive memory was confirmed via the data retention property, where both the ON state and the OFF state were stably maintained without degradation for 10⁴ s. Because the contact between the nanowire and the cantilever was deteriorated during the continuous switching processes, the endurance was measured as only six cycles.

The rigid electrode/nanowire contact might be crucial for the resistive switching of nanowire resistive memory. The performance of oxide nanowire-based memory was improved by constructing a rigid nanowire/electrode contact. Oka et al. reported the bipolar resistive switching in Pt/individual MgO–NiO core-shell nanowire/Pt structure (Oka et al., 2010). The nanowire was fabricated by the in situ heterointerface fabrication technique. The electrical contact with a 300 nm gap between electrodes was created by the electron beam lithography. Figure 16.7 shows the stable bipolar resistive switching in an MgO–NiO nanowire device at the current range of 10^− 10 A. The endurance was confirmed up to 10⁶ cycles. In addition, multistate resistive switching was demonstrated for the first time in a nanowire device (Oka et al., 2010). Better endurance and stable multistate resistive switching were achieved by choosing the precise fabrication condition. Nagashima et al reported 10⁸ cycles of endurance and stable multistate resistive switching by using Pt/individual MgO–CoO_x core–shell nanowire/Pt structure as shown in Figure 16.8 (Nagashima et al., 2010a). The single crystalline MgO–CoO_x nanowires were fabricated by the in situ heterointerface fabrication technique. The gap size between electrodes was about 250 nm. For the multistate switching, each resistance state was stably maintained without significant changes up to 10⁴ cycles of endurance. These results strongly encourage the scaling of ReRAM down to sublithographic scale to increase the density and the performance of memory. Although the current ranges of the MgO–NiO nanowire device and MgO–CoO_x nanowire device were extremely low (10^− 10–10^− 8 A), which is a strong advantage for the low power consumption of the memory device, the required voltages for both oxide nanowire devices (15–40 V) were too high for the acceptable voltage in practical applications. Despite the large operation voltage in such a nanowire device, the electric field of the nanowire device was 0.6–1.6 MV/cm and comparable to that of conventional thin film device (0.2–1 MV/cm) (Shima et al., 2008). Because the operation voltages for thin film devices are typically in the range of 1–5 V, the operation voltage can be reduced by designing the device structure.

In contrast to anionic restive switching in a nanowire device, the cationic resistive switching in a nanowire device interestingly exhibited the different memory characteristics. Yang et al. reported resistive switching in a Cu/ZnO nanowire/Pd device (Yang et al., 1917–1921). The single crystalline ZnO nanowires were prepared by Au assisted VLS growth using the chemical vapour deposition (CVD) method. The diameter of the ZnO nanowires was about 150 nm. The individual ZnO nanowire was electrically contacted with Cu and Pd electrodes by using a two-step photolithography process onto the 300 nm SiO₂ covered silicon substrate.

Figure 16.9 shows the I–V characteristics of a Cu/ZnO nanowire/Pd device. The gap size seems to be a few hundred nanometers. The forming process was completed by one to three times of unsuccessful voltage sweeps up to 3 V. After the forming process, the device showed bipolar resistive switching. Interestingly, the operation voltage was ≤ 3 V, which was much lower than the bipolar switching in single crystalline anionic nanowire memory (Oka et al., 2010; Nagashima et al., 2010a). In the Cu/ZnO nanowire/Pd device, the electromigration of Cu was experimentally confirmed the energy dispersive X-ray spectroscopy mapping taken in the ON state showed the existence of a Cu element along the ZnO nanowire. Furthermore, the control experiment using an Au/ZnO nanowire/Au device showed no resistive switching behaviour. These results highlighted the cationic resistive switching nature of the Cu/ZnO nanowire/Pd device. In the viewpoint of memory performance, the Cu/ZnO nanowire/Pd device had several advantages compared to anionic nanowire memory. The ON/OFF ratio of the Cu/ZnO nanowire/Pd device was as high as 10⁶, whereas that of the anionic nanowire memory was about 10² (Oka et al., 2010; Nagashima et al., 2010a). Also, the long retention time of 2 × 10⁶ s (23 days) was confirmed.

These excellent properties were commonly reported in cationic nanowire resistive memory. For example, Dong et al. reported the long retention for 5 months in cationic resistive switching of Cu/Zn₂SnO₄ nanowires/Pd (Figure 16.10) (Dong et al., 2012). The device required the forming process with a voltage of 5 V. After the forming process, the Cu/Zn₂SnO₄ nanowires/Pd device showed the bipolar resistive switching with an ON/OFF ratio of 10⁶. The operation voltage was ≤ 2 V. Also, the resistive switching could be completed within 20 ns. Thus the cationic resistive memory using oxide nanowire has significant advantages in the ON/OFF resistance ratio and operation voltage. However, the operation current in the cationic memory is commonly as high as the mA range. Also, the poor endurance caused by the large Joule heating during the memory operation is another issue to be solved for the nanowire-based cationic memory.

A summary of memory characteristics in both anionic and cationic resistive memory of oxide nanowire is provided in Table 16.1. As discussed earlier, cationic memory has advantages in the ON/OFF resistance ratio of retention and operation voltage, whereas anionic memory seems to be superior to cationic memory in endurance and operation current. Although there is much room for improvement in the performance of nanoscale memory, the literature review strongly supports the potential scalability of ReRAM at least down to 10 nm and the device application using oxide nanowire-based resistive memory.

16.2.2 Resistive switching in segmented oxide nanowire

As discussed in the previous section, the drawback of anionic nanowire memory is its large operation voltage. It is important to reduce the operation voltage for investigating the potential performance of nanoscale ReRAM and developing nanowire-based memory applications. To reduce the operation voltage, the active length of the resistive switching part should be decreased as thin as the conventional thin film device (e.g., several tens of nm range). In this section, one of the approaches using a multisegmented oxide nanowire, which possibly decreases the active length of the resistive switching part, is reviewed. Herderick et al. reported resistive switching in the Au–NiO–Au multisegmented nanowire (Herderick et al., 2009). The Au–NiO–Au nanowires were fabricated through sequential electrochemical deposition in the AAO template. Au with 6 μm length, Ni with 900 nm length and Au with 6 μm length were sequentially deposited into the AAO template. After the thermal annealing at 600 °C for 30 min, the Au–NiO–Au nanowires were obtained. The diameter and the length of segmented NiO nanowires were 250 and 900 nm, respectively. The typical grain size of NiO was about 8 nm. Au segment and NiO segment were seamlessly connected as shown in Figure 16.11 (Herderick et al., 2009). The electrical contact of the Au–NiO–Au nanowires was carried out by using the focused ion beam deposition.

Figure 16.12 shows the I–V characteristics of individual Au–NiO–Au nanowire devices. The device showed the bipolar resistive switching. The voltage range was ≤ 6 V, whereas the current range was as high as a few hundred μA. These characteristics were similar to the unipolar resistive switching observed in the polycrystalline NiO nanowire device (Kim et al., 2008); thereby, the direct comparison with the report in single crystalline anionic memory is rather difficult. Also, the NiO segment length of 900 nm was achievable by the conventional lithographic technique. However, the methodology introduced in this section is still promising to reduce the operation voltage because the length of resistive switching segment can be scaled down to tens in nm scale by controlling the thickness of Ni during the electrochemical deposition. In fact, it was demonstrated that the voltage was reduced below 2 V by decreasing the length of NiO segment to 20 nm (Heber et al., 2012).

16.2.3 Resistive switching in a nanowire crossbar array

Although the oxide nanowire-based resistive memory was reviewed earlier, and the oxide nanowire was found to be a powerful tool to investigate the feasibility of resistive switching at the sublithographic scale range, the integration of these nanowires for a high-density memory device is a big challenge. Because two electrodes are required for addressing the target memory device independently, the total number of electrodes is estimated to be 2N, where N is the number of nanowire memory, which limits the density of memory integration. Nevertheless the number of electrodes can be reduced to N + 1 by wiring the multielectrodes on the nanowire. The footprint of individual memory cell is determined by the gap size between the electrodes, and therefore the number density of the memory device is still limited by the lithographic technique. The ultimate architecture for high-density memory can be achieved by the crossbar array, where the switching area is located at the junction of crossbar structure. The crossbar array architecture allows reduction of the number of electrodes down to (2N)^0.5. To realize the crossbar array structure using oxide nanowires, two points must be addressed. First, the switching part should be only at the crossbar junction, whereas the other area should be electrically conductive. Second, the adjacent nanowires should be electrically isolated while keeping its conduction. In this regard, the metal-oxide core–shell nanowire structure may be promising to realize the crossbar array architecture (Figure 16.13).

Cagli et al. reported the proof of concept for the crossbar nanowire memory array (Cagli et al., 2011). They utilized the Ni–NiO core–shell nanowires fabricated by combining the AAO template method and the moderate thermal oxidization treatment. First, the Ni nanowires were fabricated by the electrochemical deposition of Ni in AAO template. Then the Ni nanowires with 200 nm diameter were aligned onto the silicon substrate with 150 nm thick SiO₂ layer. Because Ni has a ferromagnetic nature, the orientation of Ni nanowires can be arbitrarily controlled by the external magnetic field (Figure 16.14). After aligning the Ni nanowires, the thermal oxidization treatment was carried out. The bottom nanowires were annealed at 300 °C for 3 h in air while the top nanowires were spontaneously oxidized at room temperature in air to avoid overoxidization of the bottom nanowires. Figure 16.14 also shows the transmission electron microscope (TEM) image and Ni–NiO core–shell nanowire and electron energy-loss spectroscopy (EELS) line profile. The presence of an amorphous NiO shell layer with a thickness of 15 nm was confirmed.

Figure 16.15 shows the device structure and the I–V characteristics of Ni–NiO nanowire crossbar junction (Cagli et al., 2011). The device shows the unipolar resistive switching continuously. The resistive switching at the crossbar junction was justified by analysing the local resistance of the nanowire crossbar device including the contact between nanowire and metal electrode (Table 16.2). Thus the proof of concept for the nanowire crossbar array memory was demonstrated. These results highlighted that the nanowire crossbar junction might be a promising approach to integrate the nanowire memory for high-density memory device applications.

In summary, the resistive switching in oxide nanowire was reviewed, covering the cationic memory and the anionic memory. Resistive switching was ensured at about the 10 nm scale and up to 10⁸ cycles of endurance. Also, the low power consumption effect would be expected as a derivative of the scaling. Thus the review strongly supports the scaling of ReRAM down to the sublithographic scale to increase the density of memory device integration and the memory performance. Furthermore, the crossbar array using the metal-oxide core–shell nanowire was demonstrated as proof of concept for the high-density memory device. This result opens up opportunities for nanowire-based nonvolatile memory applications.

16.3.1 Chemical analysis of conducting filaments using oxide nanowire device

The chemical properties of conducting filaments can be evaluated by varying the surrounding atmosphere of the oxide nanowire device because the conducting filaments in the oxide nanowire are exposed to the environment and should sensitively interact with the ambient atmosphere. Nagashima et al. utilized the MgO–CoO_x core–shell nanowire device to investigate the chemical properties of conducting filaments for the anionic resistive memory (Nagashima et al., 2011). The individual MgO–CoO_x nanowire was electrically contacted with Pt electrodes. The reason why Pt was employed was to avoid the unexpected oxidization of electrodes, which makes the analysis complicated. The stable bipolar resistive switching in MgO–CoO_x nanowire was confirmed by the endurance properties up to 10⁸ cycles. To investigate the chemical properties of a conducting filament, first the resistance state of the device changed into the conductive ON state in air condition. Then the chamber was evacuated to 10^− 2 Pa within 2 min, followed by the introduction of reactive gas. Prior to the measurement, the pressure of reactive gas was controlled to be 10^− 1 Pa and kept for 5 min. Because the electrical transport in oxides is known to be varied by the concentration of oxygen, a redox might affect the mobile carries of oxides, and the experiment using redox gases enables identification of the conducting properties of resistive switching. On the basis of the earlier hypothesis, the effect of reactive gas such as O₃ (7% in O₂), pure O₂, dry air (20% O₂ in N₂) and H₂ (4% in Ar) for the conduction properties was examined. Figure 16.17 shows the retention measurement in the reactive gases. The reading of the nanowire resistance was performed only at the data points shown in the figure to avoid the unintentional resistive switching during measurement. The readout voltage was 2 V. Interestingly, the conduction of MgO–CoO_x nanowire tended to be enhanced when O₃ contained oxidization gas was introduced, whereas H₂ contained reduction gas suppressed the conduction. The conduction properties of MgO–CoO_x nanowire were found to vary systematically corresponding to the ability of the redox reaction of gases. Because cobalt oxide is known to show the p-type carrier transport where the conduction is caused by the cation deficiency or the excess oxygen, the trend observed in the retention measurement was consistent with the p-type transport nature of cobalt oxide (Wdowik and Parlinski, 2008; Hed, 1969; Shinde et al., 2006).

Similar experiments were reported in MgO–NiO core–shell nanowire and MgO–TiO₂ core–shell nanowire (Oka et al., 2010; Nagashima et al., 2012a). The results showed that the trend in MgO–NiO nanowire was almost same as MgO–CoO_x nanowire. In contrast, MgO–TiO₂ showed the opposite trend, that is, the H₂ contained reduction gas enhanced the conduction, but the O₃ contained oxidization gas strongly suppressed the conduction. Because NiO and TiO₂ are p-type and n-type oxides, these results were rational. Although the results from the atmosphere control measurement seems to be natural and not so surprising from the aspect of oxide electronics, these results demonstrated the significance of the redox event on the resistive switching and gave one answer for the discussion on the controversial conduction mechanism of anionic resistive switching. More practically, the knowledge provides the logical pathway to design the nanoscale ReRAM.

16.3.2 Electrical analysis of conducting filaments using oxide nanowire device

The planar structure of an oxide nanowire memory device offers the unique analysis tool to evaluate the electric properties of conducting filaments. Figure 16.16 shows a schematic of the oxide nanowire FET device to analyse the carrier type in the conducting filaments. The electrical transport of the oxide nanowire memory is in principle modulated by the electric field from the gate electrode. This methodology is difficult to be taken in a conventional thin film device due to the difficulty in introducing the gate electrode near the conducting filaments. The carrier property of resistive switching was directly evaluated by Nagashima et al. using MgO–CoO_x core–shell nanowire FET device (Nagashima et al., 2011). Figure 16.18 shows a scanning electron microscope (SEM) image of the MgO–CoO_x nanowire FET on the silicon substrate covered with 300 nm SiO₂. Pt was utilized as contact electrodes. The gap size between source (S) and drain (D) was about 400 nm.

Prior to the FET measurement, the device was set into the conductive ON state with the compliance current of 10^− 9 A. After confirming the small distribution of conduction within 0.3 × 10^− 10 A at 15 V, the measurement was carried out. When the positive + 8 V (V_G) was applied to the gate electrode, the source–drain current (I_SD) decreased, whereas I_SD increased for the negative − 8 V. The detailed trend could be seen in the transfer curve (I_SD–V_G) as shown in the inset of Figure 16.18b, clearly showing that I_SD increased as the gate voltage decreased. These results indicated that the electrical conduction in bipolar resistive switching of CoO_x is governed by the p-type carrier transport, which was consistent with the result of chemical properties analysis discussed earlier (Nagashima et al., 2011). Thus the oxide nanowire was found to be able to utilize as an analysis tool to evaluate the carrier transport of the resistive switching.

16.3.3 Spatial analysis of conducting filaments using oxide nanowire device

Identifying the exact location of resistive switching is one of the important clues to understanding the mechanism of ReRAM because it is related to the generation and the evolution of conducting filaments. Previously, the switching location was evaluated by investigating the effect of anode/cathode electrode material on the resistive switching event (Seo et al., 2005; Lee et al., 2007, 2008). However, such an evaluation method contains a serious problem, that is, the switching location would be changed when the electrode such as Ti, Ag and Al is oxidized, and the electrode/oxide interface governs the resistive switching. To avoid this problem, the direct identification of switching location using a planar-shaped nanowire device might be promising. Nagashima et al. applied a multiprobe measurement to identify the switching location of bipolar resistive switching in p-type cobalt oxide (Nagashima et al., 2011). Several Pt electrodes were contacted on the MgO–CoO_x core–shell nanowire as shown in Figure 16.19.

The location for resistive switching was examined by using three electrodes, namely electrode A, electrode B and electrode C in the figure. First, the forming process was performed between electrodes A–C, where the voltage was applied to electrode A while electrode C was grounded. After the forming process, the device showed the bipolar resistive switching as shown in Figure 16.19b and Table 16.3. The local resistance change was monitored before and after the resistive switching via the intermediated electrode B. Clearly, the resistance change between electrodes B and C (cathode side) was larger (ON/OFF ratio ~ 17.1) than that between electrodes A and B (anode side; ON/OFF ratio ~ 2.2), indicating that the resistive switching of cobalt oxide occurred at the cathode side. Although the switching location in n-type oxide, of which the conduction mechanism is based on the oxygen vacancies, had been known to be the anode side (Yang et al., 2008; Kwon et al., 2010), the switching location in p-type oxide was demonstrated for the first time using oxide nanowire. Taking into account the carrier generation mechanism of oxides, the opposite switching location between n-type and p-type oxides can be rationally understood.

The switching location of p-type oxide was also reported by Oka et al. using NiO nanowire (Oka et al., 2011). They utilized the asymmetric SiO₂ passivation layers onto NiO nanowire devices. The four types of the devices were prepared such as NiO nanowire device (Type I) passivated at anode side, (Type II) passivated at cathode side, (Type III) without passivation and (Type IV) fully passivated as shown in Figure 16.20. The I–V characteristics of the NiO nanowire device without passivation showed a symmetric bipolar resistive switching, whereas the fully passivated NiO nanowire device showed an asymmetric bipolar resistive switching. The prominent passivation effect was confirmed only when the cathode side was passivated, obviously indicating that the resistive switching occurred at cathode side. Although the role of passivation effect on the resistive switching was not well understood, these results highlighted that the switching location of bipolar resistive switching in p-type oxide commonly occurred at cathode side.

Thus, the methodology using oxide nanowire offered the powerful tool to identify the location of resistive switching. The extracted information on the switching location is very important not only to understand the underlying mechanism of resistive switching but also to design and improve the performance of ReRAM.

16.3.4 Stability of resistive switching in nanoscale resistive memory device

As the cell area scales down to nanometer, the size of conducting filaments becomes comparable to the device size, and the interaction with the surrounding atmosphere must be increasingly crucial for the performance of ReRAM. The electrically programmed resistance state should somehow survive by competing with the surrounding atmosphere for the nonvolatility. Such knowledge is rather important to design the resistive switching at nanoscale and therefore toward the high-density ReRAM. However, such interaction with surroundings had not been accessible due to the limitation in conventional lithography.

Nagashima et al. investigated the role of the surrounding atmosphere on the stability of nanoscale resistive switching (Nagashima et al., 2012b). They prepared the planar-type resistive memory devices utilizing several oxide nanowires, including TiO_2 − x nanowire (n-type), MgO–NiO nanowire (p-type) (Oka et al., 2010) and MgO–CoO_x nanowire (p-type) (Nagashima et al., 2010a, 2011). The TiO_2 − x nanowires were fabricated onto SiO₂/Si substrate by combining PLD and electron beam lithography (Nagashima et al., 2012b). The deposition of TiO_2 − x was performed at room temperature in 10 Pa of oxygen pressure. The thickness and width of TiO_2 − x nanowire were 80 and 100 nm, respectively. The fabrication of MgO–NiO nanowire and MgO–CoO_x nanowire was similar to that in Section 16.2. The oxide nanowires were electrically contacted with Pt electrodes, of which the gap size was 200 nm.

Figure 16.21 shows the I–V characteristics and the retention data of planar structured oxide nanowire devices measured at room temperature in air. The TiO_2 − x nanowire didn’t exhibit a resistive switching, whereas the MgO–NiO nanowire and the MgO–CoO_x nanowire clearly showed the bipolar resistive switching with 10⁴ s of retention. Because the bipolar resistive switching in TiO₂ thin film devices was reported (Nagashima et al., 2012a), no switching behaviour in TiO_2 − x nanowire might be caused by the interaction with surrounding atmosphere. Therefore, the surrounding condition was varied by changing the pressure of the measurement chamber. Figure 16.22 shows the I–V characteristics of TiO_2 − x nanowire devices measured under various ambient pressures. Interestingly, the hysteresis of I–V curves tended to increase as the pressure was reduced to 1 Pa and the clear bipolar resistive switching was seen. Further reducing the ambient pressure resulted in the decrease of hysteresis. The ON/OFF current ratio and the resistances of ON state and OFF state as functions of ambient pressure were shown in Figure 16.22b and c, respectively. The ON/OFF current ratio was found to have a maximum value at 5 Pa. Because it is well known that the oxygen vacancies act as a donor for the electrical conduction in n-type oxides, the results could be interpreted in terms of the amount of oxygen vacancies. These results demonstrated that the resistive switching can be stably performed in nonstoichiometric TiO_2 − x between stoichiometric insulative TiO₂ and highly conductive TiO_2 − x with plenty of oxygen vacancies. In the thermodynamic diagram of oxides, the equilibrium line of TiO₂ is located much below the atmospheric oxygen pressure. Therefore, the interaction with surrounding air condition strongly drives TiO_2 − x back to stoichiometric TiO₂. On the other hand, the equilibrium lines of NiO and CoO are close to the atmospheric oxygen pressure, enabling the stable resistive switching in air condition because the driving forces to drive back to the stoichiometric state for NiO and CoO should be much smaller than that for TiO₂. Although the dynamic reaction at the grain boundaries and the surface of oxides cannot be captured quantitatively, the discrepancy between TiO_2 − x, NiO_x and CoO_x on the resistive switching at atmospheric pressure should be qualitatively significant. Furthermore, the results in TiO_2 − x implied that it is important to design the surrounding condition, for example, via the passivation layer, toward the highly stable nanoscale ReRAM.

The impact of the passivation layer on the stability of resistive switching was examined in the TiO_2 − x nanowire device. Amorphous SiO₂ with 300 nm thickness was selected as the passivation layer because the diffusion of oxygen in SiO₂ is much slower than that in air. Figure 16.23 shows the device structure and the electrical characteristics of the SiO₂ passivated TiO_2 − x nanowire device. The measurement was performed under atmospheric conditions. In contrast to the I–V curve of the bare TiO_2 − x nanowire device as shown in Figure 16.21, the passivated device clearly showed the bipolar resistive switching. The endurance was stably repeated until 10⁶ cycles. Although further investigations need to be undertaken to realize a much higher stability of resistive switching, these results strongly encourage tailoring a resistive switching at nanoscale.

16.3.5 In situ analysis of resistive switching using transmission electron microscopy

As discussed earlier, the oxide nanowire structure enables identification of the mechanism of resistive switching more directly than conventional thin film form. However, these investigations are still indirect, and the dynamic behaviours of resistive switching cannot be captured. In situ analysis combining the electrical measurement and the TEM observation is elegant solution. Huang et al. reported the in situ analysis of resistive switching using ZnO nanowire device (Huang et al., 2013). In this report, ZnO nanowires were fabricated by a VLS mechanism in the furnace. The diameter and the length of ZnO nanowires were 100 nm and 5 μm, respectively.

The TEM specimen for in situ analysis was constructed via the photolithography and electron beam lithography. First, 30 nm SiO₂ and 60 nm Si₃N₄ were deposited on silicon substrate by low-pressure CVD. A window for TEM observation was then created via the reactive ion etching by removing the SiO₂ and the Si₃N₄ layers, followed by the silicon etching through KOH chemical etching. After creating the window, the conventional lithography was applied to make an electrical contact with Ti/Au electrodes as shown in Figure 16.24.

The electrical measurement of ZnO nanowire was conducted in the TEM chamber. Figure 16.25 shows the series of TEM images taken from video when a positive voltage of 3 V was applied to the electrodes. Interestingly, the portion near the anode tended to expand after the SET process (resistive switching into conductive state). This trend was reproducibly observed in the Au/ZnO nanowire/Au device structure. Figure 16.26 shows another sample of the ZnO nanowire memory device composed of two ZnO nanowires between electrodes. The device continuously showed the unipolar resistive switching after a forming process. Similar to the image in Figure 16.25, the portion near the anode expanded after the SET process. The composition analysis was then performed to identify what changed during the resistive switching process. Energy dispersive spectroscopy (EDS) analysis in Figure 16.26c and d exhibits the compositional change at the middle area of ZnO nanowire during the resistive switching. In the virgin state, the Zn/O ratio was almost 1 (Zn:O = 46.0:54.0 at.%) as shown in Figure 16.26c. When the SET process was performed, the Zn/O ratio increased (Zn:O = 75.5:24.5 at.%), indicating that the oxygen vacancies were induced via the SET process. Contrary, the Zn/O ratio returned to be almost 1 (Zn:O = 42.6:57.4 at.%) after the RESET process (resistive switching into insulative state). These results supposed that the formation and rupture of the oxygen vacancies-based conducting path were crucial for the resistive switching phenomena in ZnO. EDS mapping near the anode further evidenced the dynamic migration of oxygen during the resistive switching (Figure 16.27). After the SET process, the Zn/O ratio near the anode was much higher than the virgin state (Zn:O = 9.4:90.6 at.%). These results give us an important clue to understand the dynamic mechanism of resistive switching. First, the oxygen was decomposed to oxygen vacancies and oxygen ions through the electrical breakdown process. Because the oxygen ions are negatively charged, the oxygen ions migrated toward the anode side under the electric field, leaving the oxygen vacancies based conducting path. This caused the low resistance, conductive ON state. The electrode should have an ability to restore the oxygen ions to exhibit nonvolatility. In the RESET process, the oxygen ions solely cannot be driven back from the anode under the electric field because the polarity of bias for SET and RESET processes was not changed in unipolar switching. The complementation of oxygen vacancies during the RESET process might be achieved by the removal of oxygen ions from the anode with the assist of Joule heating. Consequently, the conducting path was ruptured and the insulative state obtained with the stoichiometric composition of ZnO. Thus, the in situ analysis combining the electric measurement and TEM/EDS observation on oxide nanowire device was demonstrated as a powerful tool to capture and extract the intrinsic nature of conducting path for the full understanding of the resistive switching mechanism.

In summary, this section reviewed the mechanisms of ReRAM, which is especially for the anionic resistive switching, extracted by the oxide nanowire device. The oxide nanowire offered unique analysis tools, which had never been accessible by the conventional thin film device form. The carrier type of the conducting filaments was identified electrically and chemically by using the nanowire FET structure and the atmosphere controlled measurement, respectively. Also the location of resistive switching could be analysed via the multiprobe measurement and the asymmetric passivation technique. The interaction with surrounding atmospheres gave a fundamental understanding for the emergence of nonvolatility in oxide-based resistive switching. The in situ analysis combining the electric measurement and the TEM observation enabled evaluation of the dynamics of resistive switching phenomena. The results obtained from the oxide nanowire device revealed the new aspects of the resistive switching and allows one to tailor the performance of nanoscale ReRAM.