Open Access

Excellent resistive memory characteristics and switching mechanism using a Ti nanolayer at the Cu/TaOx interface

  • Sheikh Ziaur Rahaman1,
  • Siddheswar Maikap1Email author,
  • Ta-Chang Tien2,
  • Heng-Yuan Lee3,
  • Wei-Su Chen3,
  • Frederick T Chen3,
  • Ming-Jer Kao3 and
  • Ming-Jinn Tsai3
Nanoscale Research Letters20127:345

DOI: 10.1186/1556-276X-7-345

Received: 19 March 2012

Accepted: 26 June 2012

Published: 26 June 2012


Excellent resistive switching memory characteristics were demonstrated for an Al/Cu/Ti/TaOx/W structure with a Ti nanolayer at the Cu/TaOx interface under low voltage operation of ± 1.5 V and a range of current compliances (CCs) from 0.1 to 500 μA. Oxygen accumulation at the Ti nanolayer and formation of a defective high-κ TaOx film were confirmed by high-resolution transmission electron microscopy, energy dispersive X-ray spectroscopy, and X-ray photo-electron spectroscopy. The resistive switching memory characteristics of the Al/Cu/Ti/TaOx/W structure, such as HRS/LRS (approximately 104), stable switching cycle stability (>106) and multi-level operation, were improved compared with those of Al/Cu/TaOx/W devices. These results were attributed to the control of Cu migration/dissolution by the insertion of a Ti nanolayer at the Cu/TaOx interface. In contrast, CuOx formation at the Cu/TaOx interface was observed in an Al/Cu/TaOx/W structure, which hindered dissolution of the Cu filament and resulted in a small resistance ratio of approximately 10 at a CC of 500 μA. A high charge-trapping density of 6.9 × 1016 /cm2 was observed in the Al/Cu/Ti/TaOx/W structure from capacitance-voltage hysteresis characteristics, indicating the migration of Cu ions through defect sites. The switching mechanism was successfully explained for structures with and without the Ti nanolayer. By using a new approach, the nanoscale diameter of Cu filament decreased from 10.4 to 0.17 nm as the CC decreased from 500 to 0.1 μA, resulting in a large memory size of 7.6 T to 28 Pbit/sq in. Extrapolated 10-year data retention of the Ti nanolayer device was also obtained. The findings of this study will not only improve resistive switching memory performance but also aid future design of nanoscale nonvolatile memory.


Ti nanolayer Nanoscale Resistive memory Nanofilament Charge-trapping.


Recently, many resistive switching random access memory (ReRAM) devices containing oxides such as SrTiO3[13], Al2O3[4], NiOx[57], Na0.5Bi 0.5TiO3[8], ZnO [9, 10], Ta2O5[11], ZrO2[1215], GdOx[16, 17], HfOx[18, 19], and TiOx[2123] have been reported for future nanoscale nonvolatile memory applications. However, the resistive switching mechanism of ReRAM devices is currently debated. On the other hand, conductive-bridge ReRAM devices with different solid-electrolytes, such as GeSex[2427], GeS [28, 29], Ta2O5[30, 31], ZrO2[32], SiO2[33], Ag2S [34, 35], HfO2[36, 37], SrTiO3[38], and Cu-Te/Al2O3[39] have also been reported by several groups. In these cases, silver (Ag) or copper (Cu) metal can be used as one of the electrodes to mobilize Ag+ or Cu2+ ions. Under an external bias on Ag or Cu electrode, metallic filament can be formed (or dissolved) into solid-electrolyte films. In general, Cu is a more preferable material than Ag because it is used as an interconnection metal in computer motherboards. High-κ Ta2O5 is considered the most promising as a resistive switching material. [30, 31] Therefore, a resistive switching memory device with a Cu/TaOx/W structure could be desirable but also may have the drawback of copper oxidation (CuOx) at the Cu/TaOx interface, which can hinder resistive switching characteristics. Moreover, controlling Cu ion transportation and recovery under external bias is difficult. Tada et al. [40] reported resistive switching memory based on a dual layered TiOx/TaSiOx solid-electrolyte under a high current compliance (CC) of 800 μA and a large operation voltage of 2.5 V. To prevent CuOx formation as well as promote easier Cu filament formation/dissolution through the high-κ TaOx solid-electrolyte, a Ti nanolayer at the Cu/TaOx interface is a promising approach to design an Al/Cu/Ti/TaOx/W resistive switching memory device, which has not been reported to date. Furthermore, Ti has good adhesion behavior and provides a good Cu diffusion barrier. The Gibbs free energies of TiO2, Ta2O5, CuO, and Cu2O films at 300 K are −889.5, –760.75, –129.7, and −149.0 kJ/mole, respectively [41], suggesting that the Ti nanolayer can be easily oxidized compared with the other possible materials. Therefore, a Ti nanolayer will consume more oxygen from the TaOx layer and will form TiOx/TaOx. Consequently, the Cu electrode will not form a CuO layer at the Cu/TaOx interface because of this greater oxygen consumption at the Ti nanolayer. This also has the benefit of easily controlling Cu migration and collection through the resulting higher defective high-κ TaOx solid-electrolyte under external bias. In this study, excellent resistive switching memory characteristics were observed in the proposed Al/Cu/Ti/TaOx/W structure with a Ti nanolayer at the Cu/TaOx interface annealed at 350°C in ambient N2 compared with a similar structure without a Ti nanolayer. This configuration will be useful for complementary metal-oxide-semiconductor (CMOS) processing after back end of line. The fabricated Al/Cu/Ti/TaOx/W structure memory device with a small area of 150 × 150 nm2 was observed by high-resolution transmission electron microscopy (HRTEM), X-ray photo-electron spectroscopy (XPS), and energy dispersive X-ray (EDX). In addition, the migration and collection (or formation/dissolution) of Cu ions through the defective TaOx solid-electrolyte under external bias, as well as its switching mechanism, were determined from capacitance-voltage (C-V) hysteresis. Improved resistive switching memory performance of the Al/Cu/Ti/TaOx/W structure compared with that of the Al/Cu/TaOx/W structure was also obtained, such as repeatable switching cycles with maintenance of a high resistance ratio (approximately 104), long extrapolated program and erase endurance of > 106 cycles, multi-level capability, and extrapolated 10 y data retention. Furthermore, Cu nanofilament diameters were calculated under current compliances (CCs) of 0.1 to 500 μA using a new approach. A large memory size of 28 Pbit/sq in. was achieved with a small nanofilament diameter of 0.17 nm under a CC of 0.1 μA.


Tungsten (W) metal was used as the bottom electrode (BE) and was deposited by sputtering onto SiO2/Si substrate. The thickness of the SiO2 layer was approximately 200 nm whereas that of the W layer was approximately 100 nm. A further SiO2 layer, which had a thickness of approximately 150 nm, was then deposited on the W/SiO2/Si substrates. A small device with an area of 150 × 150 nm2 was fabricated by lithography. Photoresist was coated on the patterned wafers and both the active and top electrode (TE) regions of the memory device were then exposed for a lift-off process. A high-κ Ta2O5 solid-electrolyte with a thickness of 18 nm was deposited from pure Ta2O5 granules using an E-gun evaporator. The resulting high-κ Ta2O5 solid-electrolyte film was mixed with Ta metal (i.e.,TaOx where x < 2.5), as characterized by XPS. Cu, which provided mobile ions, was deposited at a layer thickness of approximately 50 nm by a thermal evaporator. A 160-nm-thick layer of aluminum (Al) was then deposited in situ using the same thermal evaporator to protect the Cu surface layer from oxidation by the external environment. The total thickness of the top electrode (Cu + Al) was approximately 200 nm. Finally, the lift-off process was performed to obtain the resistive switching memory device. Cu mobile ions play a major role in the Al/Cu/TaOx/W structure resistive switching memory device (device S1). In this S1 device, Cu is oxidized at the Cu/TaOx interface and hinders the resistive switching memory performance. Therefore, Cu oxidation is expected to be avoided by inserting a Ti nanolayer at the Cu/TaOx interface. A thin Ti layer with a thickness of approximately 3 nm was deposited in situ using an E-gun evaporator and Ti granules during the described process to obtain a device with an Al/Cu/Ti/TaOx/W structure (device S2). The memory devices were annealed by rapid thermal annealing at 350 °C in ambient N2 for 1 min. Oxygen accumulated in the Ti nanolayer and formed a TiOxnanolayer, as characterized by XPS. More than twenty resistive switching memory devices were fabricated and measured randomly to determine and compare their memory performance. The microstructures and thicknesses of all layers were investigated by HRTEM at 200 keV. The memory devices were observed by TEM using an FEI Helios-400 s system (FEI Co., Hillsboro, OR, USA) with a Ga+ ion source at an operating voltage of 5 kV. All layers and materials were characterized by both XPS and EDX. TaOx and TiOx films were analyzed by XPS using an Al (Kα) monochrome X-ray source at 1,486.6 eV (Figure 1). The analysis area was 650 μm in diameter, and the base vacuum in the analytic chamber was 1 × 10−9 Torr. All spectra were calibrated using a reference C1s peak at 284.6 eV. Memory characteristics, such as current–voltage (I-V), C-V, read endurance, and retention were measured using an HP4156C semiconductor parameter analyzer and HP4284A LCR meter (Agilent Technologies Inc., Santa Clara, CA, USA). The frequency applied during the C-V measurements was 1 MHz. The capacitance was measured in parallel capacitance-conductance mode. The bias applied to the TE and the BE was grounded during I-V measurements, and CC was controlled by an HP4156C analyzer.
Figure 1

XPS spectra of (a) Ta 4f and (b) Ti 2p core-level electrons.

Results and discussion

Figure 1a shows the XPS spectra of Ta 4f core-level electrons. Peaks were fitted using Gaussian functions. The peak binding energies of Ta2O5 4f7/2 and Ta2O5 4f5/2 electrons in the sample without (w/o) Ti were centered at 26.7 and 28.6 eV, respectively. For the device with (w/) Ti, the peak binding energies of Ta2O5 4f7/2 and Ta2O5 4f5/2 electrons were centered at 26.48 and 28.39 eV, respectively. The w/ Ti sample showed slightly lower binding energies than the w/o Ti sample, attributed to oxygen accumulation in the Ti layer from the Ta2O5 layer. Chang et al. [42] previously reported a larger binding energy shift of approximately 4.5 eV after heat treatment of the Ti/Ta2O5 layers due to oxygen accumulation in the Ti nanolayer (i.e., formation of TiO2 or TiOx film). The binding energies of the Ta 4f7/2 and Ta 4f5/2 electrons for the w/o Ti sample were centered at 21.77 and 23.74 eV, respectively, while those of the w/ Ti sample were centered at 21.99 and 23.39 eV, respectively. These results suggest that the high-κ Ta2O5 film in both structures contained Ta metal (i.e.,TaOx where x < 2.5). The peak height ratios of the Ta 4f7/2, Ta 4f5/2, and Ta2O5 4f5/2 core levels for the w/o Ti samples with respect to the Ta2O5 4f7/2 peak height were 0.03, 0.03, and 0.77, respectively, while those for the w/ Ti samples were 0.27, 0.16, and 0.77, respectively. This also suggests that the Ta content was higher in the w/ Ti samples than in the w/o Ti samples. Furthermore, the TiO2 2p3/2 binding energy in the w/ Ti samples was 459.57 eV (Figure 1b), which is close to that reported in previous studies [43, 44]. These results indicate that a higher Ta metal content was present in the w/ Ti sample because of oxygen migration from the TaOx film to the Ti film, resulting in TiOx/TaOx bilayers. Because of this oxygen accumulation property of the Ti nanolayer, the Cu layer was expected to be protected from oxidation at the Cu/TaOx interface. In addition, the high-κ TaOx solid-electrolyte is expected to be more defective, which will lead to improve resistive switching memory characteristics in the Al/Cu/Ti/TaOx/W structure compared with the Al/Cu/TaOx/W structure. The resulting Cu oxidation and memory characteristics are explained below.

Figure 2a shows a typical TEM image of an Al/Cu/TaOx/W resistive switching memory device (S1) of approximately 1 × 1 μm in size. The thicknesses of the W, Cu, and Al layers were approximately 100, 50, and 160 nm, respectively. A rough interface between the Cu and TaOx layer was observed by HRTEM (Figure 2b,c), attributed to the formation of copperoxide in an approximately 3-nm-thick CuOx:TaOx mixture layer at the Cu/TaOx interface. Considering the Gibbs free energies [41], the formation of CuOx at the Cu/TaOx interface is not possible. However, a CuOx:TaOx mixture was expected to form at the Cu/TaOx interface during the device fabrication process. We have attempted to overcome this problem but in our experience to date it is difficult to avoid because the Cu first diffuses into the TaOx layer and then intermixes, forming a CuOx:TaOx layer. Therefore, Ti barrier layer for Cu diffusion is generally used for interconnection of CMOS. Because of the observed thin (<3 nm) intermixing layer and the existence of the same materials (Cu and TaOx) on both sides of this interfacial layer, the exact composition at the Cu/TaOx interface was not easy to determine in this report. Further study is needed to investigate the formation and composition of the Cu/TaOx interface. The high-κ TaOx film had a thickness of approximately 18 nm and appeared amorphous (Figure 2d). Figure 3a shows a TEM image of the device with a Ti nanolayer at the Cu/TaOx interface (S2). The device size was approximately 150 × 150 nm. All layers were covered well at the active and outer regions of the device (Figure 3b,c). Oxygen accumulated in the Ti layer as expected, resulting in an approximately 3-nm-thick TiOxnanolayer on the high-κ TaOx solid-electrolyte. This accumulated oxygen originated from the TaOx layer and formed a more defective TaOx solid-electrolyte. The composition of all layers was also confirmed by EDX (Figure 4). The spectra in Figure 4 are labeled with corresponding positions in Figures 2 and 3. The observed Cu, W/Ta, W, Ta, and Ti signals had energies of approximately 0.94, 1.72, 8.38, 9.36, and 4.54 keV, respectively. The Cu count in the TaOx layer w/ Ti nanolayer was lower than that in the layer w/o Ti nanolayer at an energy of 0.94 keV (108 vs. 228, respectively). The oxygen counts at an approximate energy of 0.52 keV were 17, 238, 129, and 378 for W, TaOx w/o Ti, TaOx w/ Ti, and pure Ti layers, respectively. The oxygen count in the Ti nanolayer was the highest of any layer in the resistive switching memory device. These findings suggest that the Ti nanolayer is easily oxidized, which is in agreement with the Gibbs free energy, resulting in the formation of TiOx on the TaOx solid-electrolyte. TiOxnanolayer formation at the Cu/TaOx interface improved resistive switching memory characteristics as described below.
Figure 2

TEM and HRTEM images of Al/Cu/TaO x /W structure. (a) TEM image of an Al/Cu/TaOx/W structure. Enlarged images with (b) scale bars of 50 nm and (c) 10 nm. (d) HRTEM image from (a) with a scale bar of 10 nm.
Figure 3

TEM image of an Al/Cu/Ti/TaO x /W structure. (a). HRTEM images on (b) outside and (c) inside of the structure shown in (a).
Figure 4

EDX spectra from Figures2and3. Numbers indicate corresponding positions in Figures 2 and 3.

Figure 5 shows a typical I-V hysteresis loop for the Al/Cu/Ti/TaOx/W resistive switching memory device (S2). Ten consecutive switching cycles at a CC of 100 μA are indicated by arrows 1 to 4. The pristine device had a high resistance state (HRS) with a leakage current of 5.26 nA at a read voltage (Vread) of +0.1 V. When the bias voltage applied to the top electrode was larger than the SET voltage of + 0.6 V, Cu filament formed in the TaOx solid-electrolyte and induced a low resistance state (LRS) through a chemical reduction process (Cuz+ + ze → Cu, where z is 1 or 2). By applying a negative voltage of approximately −0.6 V, the Cu filament was dissolved or reverted to the TE through a chemical oxidation process (Cu → Cuz+ + ze), and the device was reverted to a HRS. The SET voltage varied between 0.2 and 0.6 from cycle-to-cycle and device-to-device (data not shown). It has previously been shown that a higher formation voltage (e.g. >5 V) is not needed to observe resistive switching characteristics of the pristine device [45]. It is interesting to note that the value of the RESET current (IRESET) increased as the number of switching cycles increased. No RESET current was observed for the first switching cycle of the pristine device. These results suggested that the filament diameter is very thin (or unstable) during the first switching cycle. When the number of switching cycles was increased to 10, the RESET current reached approximately −3.9 μA, and a stable filament was formed, suggesting that Cu migration was controlled by the TiOx layer. In this case, the initial filament formation of a pristine device depends on a few switching cycles rather than the formation voltage, which is commonly needed for a ReRAM device. Therefore, this provides an advantage for future real applications of Cu metallic filament resistive switching memory devices. The present device also only requires very low operation voltage of ±1 V with a current of only a few nanoamperes. Although it has been reported that Cu nanofilament has been successfully formed in solid-electrolyte, Cu migration in solid-electrolyte has not been described in the literature to date. To address this point, C-V hysteresis characteristics were measured to obtain an understanding of the solid-electrolyte under external bias.
Figure 5

Typical I-V hysteresis loop under a CC of 100 μA for the Al/Cu/Ti/TaO x /W structure. The RESET current increases with the number of cycles.

Figure 6 shows the typical C-V hysteresis characteristics for both resistive switching memory devices under sweeping voltages from −1 V → +1.6 V → –2.5 V. The typical size of the device (A) was 8 × 8 μm. The switching cycle is indicated by arrows 1 → 3. The capacitances of pristine S1 and S2 devices at HRS were approximately 14.5 and 13.5 pF, respectively, at a Vread of + 0.1 V. The capacitances at LRS were very high at approximately 1.87 and 2.03 nF for the S1 and S2 devices, respectively, under SET operation at a Vread of + 0.1 V. This result was likely due to the movement of Cu ions through defect sites, which were trapped in the defective high-κ TaOx solid-electrolyte (i.e., Cu filament formation). The charge-trapping density was calculated from C-V hysteresis characteristics using the Equation 1 below [46].
N charge = Δ V . C LRS q . A ,
where ‘q’ (= 1.602 × 10−19 C) is the electronic charge and ‘A’ is the area of the resistive switching memory device. The hysteresis memory windows (ΔV) were approximately 3.4 and 3.5 V for the S1 and S2 devices, respectively. The charge-trapping densities of the S1 and S2 resistive switching memory devices were approximately 6.2 × 1016 and 6.9 × 1016/cm2, respectively. C-V was measured using a 4284A LCR meter (Agilent Technologies Inc., Santa Clara, CA, USA). This system does not have capacitance compliance function. To confirm the capacitance values, more than fifty devices on each wafer were measured randomly. The Ti nanolayer of the Cu/TaOx device (S2) had a higher charge-trapping density due to trapping of more Cu ions at the defect sites of the TaOx solid-electrolyte, which improved the resistive switching performance as described below.
Figure 6

C-V hysteresis characteristics for w/ and w/o Ti nanolayer capacitors.

Figure 7 shows the typical I-V hysteresis loops for both resistive switching memory devices. One hundred consecutive switching cycles are indicated by arrows 1 to 5 under a CC of 500 μA. The S1 memory devices showed a leakage current of approximately 1.5 pA at a Vread of + 0.1 V for a pristine device, lower than that of the S2 devices (23 pA) (Figure 7a,b). Average values (± standard deviations (SD)) of the leakage currents for the S1 and S2 memories were 1.5 (4.7) and 180 (10.7) pA, respectively (Figure 7c). Because of a higher charge-trapping density (Figure 6), the average leakage current of the S2 memory devices was higher than that of the S1 devices. The HRS currents at a Vread of +0.1 V increased after the second cycle to approximately 32 and 0.57 μA for S1 and S2 devices, respectively. The lower HRS current of the S2 devices compared with that of the S1 devices was attributed to a longer filament length remaining in the S1 devices. This suggests that the recovery of Cu ions in the S2 devices was caused by a higher amount of defects in the TaOx layer. The SET voltages of the S1 devices showed larger variation (+0.3 to 0.9 V) compared with those of the S2 devices (0.23 to 0.5 V). In addition, the IRESET of the S2 devices was lower than that of the S1 devices (100 vs. 1,000 μA, respectively); this was because of the control of Cu migration by the Ti nanolayer at the Cu/TaOx interface. The average device-to-device LRS values (± SD) of the S1 and S2 devices were 5.5 × 102 (3.2 × 102) and 8.4 × 102 (3.4 × 102) Ω, respectively, and the average HRS (± SD) values were 5.9 × 103 (4.0 × 103) and 3.2 × 107 (10.4 × 107) Ω, respectively (Figure 8a,b). For cycle-to-cycle, the average LRS values (± SD) of the S1 and S2 devices were 1.6 × 102 (1.3 × 102) and 8.2 × 102 (0.88 × 102) Ω, respectively, whereas the average HRS (± SD) values were 7.1 × 103 (1.34 × 104) and 7.4 × 106 (1.5 × 107) Ω, respectively. The resulting resistance ratios (HRS/LRS) of the S1 and S2 devices were approximately 10 and 3.8 × 104 for device-to-device and 44 and 9 × 103 for cycle-to-cycle, respectively. The resistance ratio of the S1 device was smaller than that of the S2 (10 vs. 104, respectively), which was attributed to the smaller gap (i.e., dissolved filament length) between the Cu electrode and remaining filament. Because of Cu ion migration under external bias, the filament formation/dissolution can be explained with the following hypothesis. Considering the charge-trapping phenomenon shown in Figure 6, the Cu ions (as positive charges) are transported through the defect sites into the high-κ TaOx solid-electrolyte under external bias. These ions are then trapped in the defect sites (starting from the W BE) and consequently neutralized by electrons from the BE. This causes metallic Cu filament to grow from the BE and form a conical-type metallic filament between the BE and TE (Figure 9a), a different mechanism to that recently reported for ZnO [10]. Under RESET operation, the Cu filament then starts to dissolve from the Cu/TaOx interface (Figure 9b) because the electric field at the Cu/TaOx interface will be higher as a result of its conical-shape. In this case, under negative bias to the TE, the filament Cu ions are transported through the similar defect sites into the high-κ TaOx solid-electrolyte. The Cu ions then either capture electrons from the TE, resulting in neutralization or return to the Cu electrode. In this case, the Cu ions will partially return to the Cu electrode because of the copper oxide present at the Cu/TaOx interface. On the other hand, the Cu filament forms, and ions migrate through the TaOx solid-electrolyte easily because of the higher charge-trapping density (i.e., defect sites), as shown in Figure 9c. It is clear that the observed lower diameter of the Cu filament is due to the Ti nanolayer at the Cu/TaOx interface controlling the migration of Cu ions under SET operation. Therefore, one of the major reasons that the control of the Cu ion mobilization is easier in the device with Ti is because of the higher defective TaOx solid-electrolyte. Under RESET operation, either the filament is almost dissolved (Figure 9d) or the gap length (i.e., the length between the Cu electrode and the remaining filament, or length of filament dissolution) in the S2 devices is longer than that of the S1 devices. The longer gap length of the S2 devices produced a higher resistance ratio (approximately 104 vs. 10, respectively) than that exhibited by the S1 devices because of the easier dissolution of the Cu filament through the defective TaOx solid-electrolyte. Through observation of several switching cycles (Figure 8), we found that the standard deviation of HRS for the S2 devices was larger than that for the S1 devices (1.5 × 107 vs. 1.34 × 104 Ω, respectively), which was attributed to the longer gap length. However, the SD of LRS for the S2 devices was smaller than that for the S1 devices (88 vs. 130 Ω, respectively) because of the Ti nanolayer controlling Cu ion migration. Therefore, LRS dispersion was better in S2 devices compared with S1 devices. It is also interesting to note that under a CC of 500 μA, the LRS of the S2 devices was higher than that of the S1 devices (840 vs. 550 Ω, respectively) because of the series resistance effect of Ti acting as a Cu barrier layer. Notably, the observed longer gap length in the S2 devices was attributed to either easier dissolution of the Cu filament or higher amount of electron injection from the Cu electrode through the Ti nanolayer. It has been reported that the electron affinities of Ta2O5 and TiO2 films with respect to the Si conduction band are 3.77 [47] and 3.15 eV, respectively [32, 47]. Therefore, our results indicate that the barrier height at the Cu/Ta2O5 interface was reduced because of the Ti nanolayer, increasing electron injection from the Cu electrode. Similar barrier height lowering at the TiOx/ZrO2 interface and improved resistive switching memory performance were reported by Li et al. [48]. These findings suggest that Cu recovery from the filament through the TaOx solid-electrolyte is easier in the device with Ti. The defective high-κ TaOx solid-electrolyte, Cu ion migration control, and lower barrier height of the Ti nanolayer at the Cu/TiOx interface in the Al/Cu/Ti/TaOx/W structure were therefore expected to maintain repeatable and improved resistive switching memory characteristics, as described below.
Figure 7

Typical I-V hysteresis loop of S1 and S2 devices and cumulative probability of leakage currents. Typical I-V hysteresis loop with 100 consecutive cycles under a CC of 500 μA for (a) S1 and (b) S2 devices. (c) Cumulative probability of leakage currents for S1 and S2 devices.
Figure 8

The cumulative probability plots of LRS and HRS. Cumulative probability plots of LRS and HRS under a CC of 500 μA for (a) Al/Cu/TaOx/W and (b) Al/Cu/Ti/TaOx/W devices with cycle-to-cycle and device-to-device measurements.
Figure 9

The schematic view of filament formation/dissolution. Schematic view of filament formation/dissolution under SET and RESET operations for w/ and w/o Ti nanolayer resistive switching memories. Filament formation (a) and dissolution (b) for w/o Ti nanolayer devices. Controllable filament formation (c) and dissolution (d) for w/ Ti nanolayer resistive switching memory devices.

Figure 10 shows the dependence of average LRS with CC ranging from 0.1 to 500 μA for the S2 devices. The LRS decreased linearly with increasing CC. The observed control of LRS by the device operation current was independent of the device size, which confirms the filamentary conduction mechanism proposed for the present resistive switching memory device. The average (median) value of LRS under a CC of 1 μA was approximately 147 (145) kΩ for a typical device of 150 × 150 nm. All LRS data were captured from several devices within the same cycle number. As a comparison, the LRS value of Ag-GeSe solid-electrolyte is approximately 140 kΩ at a CC of 1 μA [28]. The resistance at LRS can be expressed in relation to CC by Equation 2
R LRS = 0.175 C C .
Figure 10

The average LRS and filament diameter. (a)Average LRS under current compliances from 0.1 to 500 μA and (b) filament diameter versus current compliances for the S2 structure.

The average LRS determined for the present S2 devices was slightly higher than that previously reported (0.14 /CC), which was due to both the series resistance of the Ti nanolayer and use of a different solid-electrolyte. Considering the cylindrical shape of the Cu filament, its diameter can be calculated from Equation 3
R LRS = ρ filament . L Φ ,
Φ = π . D 2 4 ,
D = 4. L . ρ filament π . R LRS ,

where L is the length (thickness of high-κ TaOx solid-electrolyte, approximately 18 nm), ρfilament is the resistivity (approximately 200 μΩ.cm [49]), Ф is the cross-sectional area, and D is the diameter of the Cu filament. Using RLRS and CC as obtained from equations 2 and 5, respectively, the filament diameter was found to decrease linearly with CC (Figure 10). The nanoscale filament diameter was 10.4 to 0.17 nm as the CC decreased from 500 to 0.1 μA. These calculated filament diameters were generally consistent with those in some reports [26, 32, 50] but were slightly different than others in the literature [33, 51], which may be due to different solid-electrolytes and structures used. The observed small diameter of 1.7 Å at a small CC of 0.1 μA indicates that this device will be scalable beyond the atomic scale in the future. Under a CC of 500 μA, a memory size of 7.6 Tbit/sq in will be obtained, higher than that previously reported (60 Gbit/sq in. [35]). If a low current operation of 0.1 μA is achieved, then an even larger memory size of 28 Pbit/sq in. will be obtained, a significant benefit to future high-density nonvolatile memory applications.

Figure 11 shows typical program/erase (P/E) endurance characteristics of the resistive switching memory devices. The S2 device had improved switching cycles, with P/E voltage of +1.4/–1.2 V, pulse width of 500 μs, and P/E current of +300/–500 μA (Figure 11b). An extrapolated P/E endurance of > 106 cycles with a resistance ratio of > 20 was observed for the S2 devices, whereas the S1 device collapsed at 3.2 × 106 cycles. This result was attributed to the Ti nanolayer at the Cu/TaOx interface in the S2 devices. Excellent read endurance characteristics of the Al/Cu/Ti/TaOx/W memory device were obtained even after >105 cycles under a read voltage of +0.1 V and CCs of 0.1 and 200 μA (Figure 12a). Under a CC of 200 μA, a high resistance ratio of > 104 was obtained after 105 cycles, whereas under a low CC of 0.1 μA, a smaller ratio of 2 to 3 was obtained and there was no collapse even after 105 cycles. Figure 12b shows typical data retention characteristics as well as 10 y extrapolated data retention under a CC of 200 μA, performed at 25 °C. A high resistance ratio of > 104 was obtained after the extrapolated data retention, which will be of great advantage in future nanoscale low power nonvolatile resistive switching memory applications.
Figure 11

P/E endurance characteristics for (a) S1 and (b) S2 resistive switching memory devices.
Figure 12

The typical read endurance and data retention characteristic for the device with Ti nanolayer. (a)Typical read endurance and (b) data retention characteristics for the device with Ti nanolayer at the Cu/TaOxinterface (S2). The device is capable of operation at a low current compliance of 0.1 μA.


In summary, superior resistive switching memory performance of an Al/Cu/Ti/TaOx/W structure in a CC range of 0.1 to 500 μA was demonstrated compared with that of an Al/Cu/TaOx/W structure. The resistive switching mechanism was elucidated and explained by a schematic model, correlated with charge-trapping phenomena observed from C-V hysteresis characteristics. These findings suggest that the formation/dissolution of Cu nanofilament can be controlled by insertion of a Ti nanolayer at the Cu/TaOx interface or using a defective high-κ TaOx solid-electrolyte. The high-κ TiOx and TaOx films were amorphous as observed by HRTEM. The oxygen accumulation properties of the Ti nanolayer at the Cu/TaOx interface were confirmed by EDX and XPS analyses. Good uniformity with large resistance ratio of approximately 104, long P/E endurance of > 106 cycles, MLC operation, and 10 y data retention were observed for the devices with Ti nanolayer at the Cu/TaOx interface. The LRS increased linearly with decreasing CC in the range 500 to 0.1 μA, which resulted in a decrease of the Cu nanofilament diameter from 10.4 to 0.17 nm. These results suggest that a large memory size of 28 Pbit/sq in. at a small CC of 0.1 μA will be possible for the present devices for future nanoscale (0.17 nm) nonvolatile memory applications.



This work was supported by the National Science Council (NSC), Taiwan, under contract number: NSC-98-2221-E-182-052-MY3. We are grateful to NSC, Taiwan for their support. The authors are also grateful to MSSCORPS Co., Ltd., and MA-tek, Hsinchu for their HRTEM support and discussion of our resistive switching memory device. The authors are grateful to Prof. M. N. Kozicki, Arizona State University, USA for his insightful suggestions in this study.

Authors’ Affiliations

Department of Electronic Engineering, Chang Gung University
Material and Chemical Research Laboratories, Industrial Technology Research Institute
Electronic and Opto-Electronic Research Laboratories, Industrial Technology Research Institute


  1. Waser R, Aono M: Nanoionics-based resistive switching memories. Nat Mater 2007, 6: 833–840. 10.1038/nmat2023View Article
  2. Sawa A: Resistive switching in transition metal oxides. Mater Today 2008, 11: 28–36.View Article
  3. Sun X, Li G, Chen L, Shi Z, Zhang W: Bipolar resistance switching characteristics with opposite polarity of Au/SrTiO3/Ti memory cells. Nanoscale Res Lett 2011, 6: 599. 10.1186/1556-276X-6-599View Article
  4. Wu Y, Lee B, Wong HSP: Al2O3-based RRAM using atomic layer deposition (ALD) with 1-μA reset current. IEEE Electron Dev Lett 2010, 31: 1449–1451.View Article
  5. Ielmini D, Spiga S, Nardi F, Cagli C, Lamperti A, Cianci E, Fanciulli M: Scaling analysis of submicrometer nickel-oxide-based resistive switching memory devices. J Appl Phys 2011, 109: 034506. 10.1063/1.3544499View Article
  6. Panda D, Dhar A, Ray SK: Nonvolatile and unipolar resistive switching characteristics of pulsed laser ablated NiO films. J Appl Phys 2010, 108: 104513. 10.1063/1.3514036View Article
  7. Son JY, Shin YH, Kim H, Jang HM: NiO resistive random access memory nanocapacitor array on graphene . ACS Nano 2010, 4: 2655–2658. 10.1021/nn100234xView Article
  8. Zhang T, Zhang X, Ding L, Zhang W: Study on resistance switching properties of Na0.5Bi0.5TiO3thin films using impedance spectroscopy. Nanoscale Res Lett 2009, 4: 309–1314.
  9. Chiu FC, Li PW, Chang WY: Reliability characteristics and conduction mechanisms in resistive switching memory devices using ZnO thin films. Nanoscale Res Lett 2012, 7: 178. 10.1186/1556-276X-7-178View Article
  10. Peng S, Zhuge F, Chen X, Zhu X, Hu B: Mechanism for resistive switching in an oxide-based electrochemical metallization memory. Appl Phys Lett 2012, 100: 072101. 10.1063/1.3683523View Article
  11. Strachan JP, Ribeiro GM, Yang JJ, Zhang MX, Miao F, Goldfarb I, Holt M, Rose V, Williams RS: Spectromicroscopy of tantalum oxide memristors. Appl Phys Lett 2011, 98: 242114. 10.1063/1.3599589View Article
  12. Lin CY, Wang SY, Lee DY, Tseng TY: Electrical properties and fatigue behaviors of ZrO2resistive switching thin films. J Electrochem Soc 2008, 155: H615-H619. 10.1149/1.2946430View Article
  13. Lin CY, Wu CY, Wu CY, Lee TC, Yang FL, Hu C, Tseng TY: Effect of top electrode material on resistive switching properties of ZrO2film memory devices. IEEE Electron Dev Lett 2007, 28: 366–368.View Article
  14. Zuo Q, Long S, Liu Q, Zhang S, Wang Q, Li Y, Wang Y, Liu M: Self-rectifying effect in gold nanocrystal-embedded zirconium oxide resistive memory. J Appl Phys 2009, 106: 073724. 10.1063/1.3236632View Article
  15. Lin CC, Chang YP, Lin HB, Lin CH: Effect of non-lattice oxygen on ZrO2-based resistive switching memory. Nanoscale Res Lett 2012, 7: 187. 10.1186/1556-276X-7-187View Article
  16. Cao X, Li X, Gao X, Yu W, Liu X, Zhang Y, Chen L, Cheng X: Forming-free colossal resistive switching effect in rare-earth-oxide Gd2O3films for memristor applications. J Appl Phys 2009, 106: 073723. 10.1063/1.3236573View Article
  17. Jana D, Maikap S, Tien TC, Lee HY, Chen WS, Chen F, Kao MJ, Tsai MJ: Formation polarity dependent improved resistive switching memory performance using IrOx/GdOx/WOx/W structure. Jpn J Appl Phys 2012., 51: in press in press
  18. Lee HY, Chen PS, Wang CC, Maikap S, Tzeng PJ, Lin CH, Lee LS, Tsai MJ: Low-power switching of nonvolatile resistive memory using hafnium oxide. Jpn J Appl Phys 2007, 46: 2175–2179. 10.1143/JJAP.46.2175View Article
  19. Son JY, Kim DY, Kim H, Maeng WJ, Shin YS, Shin YH: A HfO2thin film resistive switch based on conducting atomic force microscopy. Electrochem Solid State Lett 2011, 14: H311-H313. 10.1149/1.3574526View Article
  20. Son JY, Shin YH, Kim H, Jang HM: NiO resistive random access memory nanocapacitor array on graphene. ACS Nano 2010, 4: 2655–2658. 10.1021/nn100234xView Article
  21. Kim KM, Choi BJ, Hwang CS: Localized switching mechanism in resistive switching of atomic-layer-deposited TiO2thin films. Appl Phys Lett 2007, 90: 242906. 10.1063/1.2748312View Article
  22. Yang JJ, Pickett MD, Li X, Ohlberg DAA, Stewart DR, Willams RS: Memristive switching mechanism for metal/oxide/metal nanodevices. Nat Nanotechnol 2008, 3: 429–433. 10.1038/nnano.2008.160View Article
  23. Kim GH, Kim KM, Seok JY, Lee MH, Song SJ, Hwang CS: Influence of the interconnection line resistance and performance of a resistive cross bar array memory. J Electrochem Soc 2010, 157: G211-G215. 10.1149/1.3478143View Article
  24. Kozicki MN, Park M, Mitkova M: Nanoscale memory elements based on solid-state electrolytes. IEEE Trans Nanotech 2005, 4: 331–338. 10.1109/TNANO.2005.846936View Article
  25. Linn E, Rosezin R, Kugeler C, Waser R: Complementary resistive switches for passive nanocrossbar memories. Nat Mater 2010, 9: 403–406. 10.1038/nmat2748View Article
  26. Rahaman SZ, Maikap S, Chiu HC, Lin CH, Wu TY, Chen YS, Tzeng PJ, Chen F, Kao MJ, Tsai MJ: Bipolar resistive switching memory using Cu metallic filament in Ge0.4Se0.6solid electrolyte. Electrochem Solid-State Lett 2010, 13: H159-H162. 10.1149/1.3339449View Article
  27. Yu S, Wong HSP: Compact modeling of conducting-bridge random-access memory (CBRAM). IEEE Trans Electron Dev 2011, 58: 1352–1360.View Article
  28. Kozicki MN, Gopalan C, Balakrishnan M, Park M, Mitkova M: Non-volatile memory based on solid electrolytes. Proceedings 2004 Non-Volatile Memory Technology Symposium 2004, 10–14.
  29. Bruchhaus R, Honal M, Symanczyk R, Kund M: Selection of optimized materials for CBRAM based on HT-XRD and electrical test results. J Electrochem Soc 2009, 156: H729-H733. 10.1149/1.3160570View Article
  30. Sakamoto T, Lister K, Banno N, Hasegawa T, Terabe K, Aono M: Electronic transport in Ta2O5resistive switch. Appl Phys Lett 2007, 91: 092110. 10.1063/1.2777170View Article
  31. Tsuruoka T, Terabe K, Hasegawa T, Valov I, Waser R, Aono M: Effects of moisture on the switching characteristics of oxide-based, gapless-type atomic switches. Adv Funct Mater 2012, 22: 70–77. 10.1002/adfm.201101846View Article
  32. Liu Q, Long S, Lv H, Wang W, Niu J, Huo Z, Chen J, Liu M: Controllable growth of nanoscale conductive filaments in solid-electrolyte-based ReRAM by using a metal nanocrystal covered bottom electrode. ACS Nano 2010, 4: 6162–6168. 10.1021/nn1017582View Article
  33. Schindler C, Weides M, Kozicki MN, Waser R: Low current resistive switching in Cu-SiO2cells. Appl Phys Lett 2008, 92: 122910. 10.1063/1.2903707View Article
  34. Hasegawa T, Terabe K, Tsuruoka T, Aono M: Atomic switch: atom/ion movement controlled devices for beyond Von-Neumann computers. Adv Mater 2012, 24: 252–267. 10.1002/adma.201102597View Article
  35. Wang D, Liu L, Kim Y, Huang Z, Pantel D, Hesse D, Alexe M: Fabrication and characterization of extended arrays of Ag2S/Ag nanodot resistive switches. Appl Phys Lett 2011, 98: 243109. 10.1063/1.3595944View Article
  36. Wang Y, Liu Q, Long S, Wang W, Wang Q, Zhang M, Zhang S, Li Y, Zuo Q, Yang J, Liu M: Investigation of resistive switching in Cu-doped HfO2thin film for multilevel non-volatile memory applications. Nanotechnology 2010, 21: 045202. 10.1088/0957-4484/21/4/045202View Article
  37. Nagata T, Haemori M, Yamashita Y, Yoshikawa H, Iwashita Y, Kobayashi K, Chikyow T: Bias application hard x-ray photoelectron spectroscopy study of forming process of Cu/HfO2/Pt resistive random access memory structure. Appl Phys Lett 2011, 99: 223517. 10.1063/1.3664781View Article
  38. Yan XB, Li K, Yin J, Xia YD, Guo HX, Chen L, Liu ZG: The resistive switching mechanism of Ag-SrTiO3-Pt memory cells. Electrochem Solid-State Lett 2010, 13: H87-H89. 10.1149/1.3279689View Article
  39. Goux L, Opsomer K, Degraeve R, Muller R, Detavernier C, Wouters DJ, Jurczak M, Altimime L, Kittl JA: Influence of the Cu-Te composition and microstructure on the resistive switching of Cu-Te/Al2O3/Si cells. Appl Phys Lett 2011, 99: 053502. 10.1063/1.3621835View Article
  40. Tada M, Sakamoto T, Banno N, Aono M, Hada H, Kasai N: Nonvolatile crossbar switch using TiOx-TaSiOySolid electrolyte. IEEE Trans Electron Dev 2010, 57: 1987–1995.View Article
  41. Birks N, Meier GH, Pettit FS: Introduction To The High Temperature Oxidation Of Metals. Cambridge: Cambridge University Press; 2006. http://​www.​doitpoms.​ac.​uk/​tlplib/​ellingham_​diagrams/​interactive.​phpView Article
  42. Chang JP, Steigerwald ML, Fleming RM, Opila RL, Alers GB: Thermal stability of Ta2O5in metal–oxide–metal capacitor structures. Appl Phys Lett 1999, 74: 3705–3707. 10.1063/1.123227View Article
  43. Maikap S, Wang TY, Tzeng PJ, Lin CH, Tien TC, Lee LS, Yang JR, Tsai MJ: Band offsets and charge storage characteristics of atomic layer deposited high-k HfO2/TiO2multilayers. Appl Phys Lett 2007, 90: 262901. 10.1063/1.2751579View Article
  44. Moulder JF, Stickle WF, Sobol PE, Bomben KD: Handbook of X-ray Photoelectron Spectroscopy. Eden Prairie, MN: Perkin-Elmer Corp; 1992.
  45. Banerjee W, Maikap S, Rahaman SZ, Prakash A, Tien TC, Li WC, Yang JR: Improved resistive switching memory characteristics using core-shell IrOxnano-dots in Al2O3/WOxbilayer structure. J Electrochem Soc 2012, 159: H177-H182. 10.1149/2.067202jesView Article
  46. Maikap S, Das A, Wang TY, Tien TC, Chang LB: High-κ HfO2nanocrystal memory capacitors prepared by phase separation of atomic-layer-deposited HfO2/Al2O3nanomixtures. J Electrochem Soc 2009, 156: K28-K32. 10.1149/1.3070660View Article
  47. Luo ZJ, Guo X, Ma TP, Tamagawa T: Temperature dependence of gate currents in thin Ta2O5and TiO2films. Appl Phys Lett 2001, 79: 2803–2804. 10.1063/1.1412823View Article
  48. Li Y, Long S, Lv H, Liu Q, Wang Y, Zhang S, Lian W, Wang M, Zhang K, Xie H, Liu S, Liu M: Improvement of resistive switching characteristics in ZrO2film by embedding a thin TiOxlayer. Nanotechnology 2011, 22: 254028. 10.1088/0957-4484/22/25/254028View Article
  49. Rahaman SZ, Maikap S, Chen WS, Lee HY, Chen FT, Tien TC, Tsai MJ: Impact of TaOxnanolayer at the GeSex/W interface on resistive switching memory performance and investigation of Cu nanofilament. J Appl Phys 2012., 111: in press in press
  50. Rosezin R, Meier M, Breuer U, Kugeler C, Waser R: Electroforming and resistance switching characteristics of silver-doped MSQ with inert electrodes. IEEE Trans Nanotech 2011, 10: 338–343.View Article
  51. Hsiung CP, Liao HW, Gan JY, Wu TB, Hwang JC, Chen F, Tsai MJ: Formation and instability of silver nanofilament in Ag-based programmable metallization cells. ACS Nano 2010, 4: 5414–5420. 10.1021/nn1010667View Article


© Rahaman et al.; licensee BioMed Central Ltd. 2012

This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://​creativecommons.​org/​licenses/​by/​2.​0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.