Skip to main content

Retention Model of TaO/HfO x and TaO/AlO x RRAM with Self-Rectifying Switch Characteristics


A retention behavior model for self-rectifying TaO/HfO x - and TaO/AlO x -based resistive random-access memory (RRAM) is proposed. Trapping-type RRAM can have a high resistance state (HRS) and a low resistance state (LRS); the degradation in a LRS is usually more severe than that in a HRS, because the LRS during the SET process is limited by the internal resistor layer. However, if TaO/AlO x elements are stacked in layers, the LRS retention can be improved. The LRS retention time estimated by extrapolation method is more than 5 years at room temperature. Both TaO/HfO x - and TaO/AlO x -based RRAM structures have the same capping layer of TaO, and the activation energy levels of both types of structures are 0.38 eV. Moreover, the additional AlO x switching layer of a TaO/AlO x structure creates a higher O diffusion barrier that can substantially enhance retention, and the TaO/AlO x structure also shows a quite stable LRS under biased conditions.


Because NAND flash technology is facing a scaling limit, vertical resistive random-access memory (VRRAM) designs with low film stacks, high manufacturing yields, and no cross-coupling problems are promising candidates for high-density memory applications [1,2,3]. The 1TnR architecture with three-dimensional (3D) vertical structure helps realize ultralow bit cost for highly compact dense arrays [4,5,6]. Several researchers have proposed operating RRAM at low current levels by changing the resistance switching mechanism from a filamentary-type to a defect-trapping-, vacancy-modulating-, or interface-type conducting path model [7,8,9]. However, the questions central to retention failures and the migration of oxygen vacancies are still unsolved [3, 10]. In some filamentary-type retention studies, many different models have been proposed to explain retention losses [11,12,13]. The change of switching mechanism also indicates a different direction that might improve retention [11]. Our previous studies have shown that TaO/HfO x devices can show favorable nonlinearity values of approximately 40, endurance values exceeding 1000 cycles, and 85 °C data retention [6, 7]. Nevertheless, to obtain stable retention at such low operating current levels is still challenging. In this letter, a retention model is proposed to realize the retention loss in two different defect-trapping-type devices with the Arrhenius method. The extracted activation energy does not convincingly explain the retention improvement by the AlO x layer. Even though the original was ambiguous, the most likely interpretation is that dense bonding facilitates retention.


In the fabrication of TaO/HfO x and TaO/AlO x devices for the present study, the bottom electrode (BE) is composed of TiN metal deposited by physical vapor deposition (PVD) on 8-in. thermal oxide/Si substrates. Each BE was patterned and etched with a conventional lithography and etching process. After each TiN BE had been etched with chlorine-based gas, the remaining photoresist (PR) and etching residues were removed using a remote plasma system that applied O2 and H2O at 180 °C. During the PR removal process, a thin oxidation layer of TiON was formed on the surface of each TiN BE. Then, resistive switching layers of HfO x and AlO x were prepared through atomic layer deposition (ALD) with HfCl4-H2O and TMA-H2O precursors, respectively. The two resistive elements HfO x and AlO x were deposited at 300 and 250 °C. On the top of resistive switching layers, the TaO layer was then deposited by PVD through low-temperature plasma oxidation (LTPO) [14]. This fabrication deposits Ta metal at an ultralow rate (0.2 Å/s). Stable plasma oxidation was performed with a mixture of Ar and O2 gases. This TaO layer served as an internal self-compliance resistance, which was relatively leaky compared with prior resistive switching films [7]. The top electrode was also PVD-TiN. The cross-sectional views and thickness information of the TaO/HfO x and TaO/AlO x memory devices are illustrated in Fig. 1a, b respectively. The film thickness of TaO/HfO x was checked by transmission electron microscopy (not shown). After the cells had been patterned, the low-temperature oxide was deposited for passivation at 250 °C. Finally, a conventional back-end process was applied to finish the fabrication of contact and metal pad structures.

Fig. 1
figure 1

Cell schematic plots with thickness information for a TaO/HfO x devices and b TaO/AlO x devices. Both plots describe devices for which PVD deposited the TaO layers with LTPO processes, and the bottom TiON interfacial layers were formed by plasma oxidation during photoresist removal

Results and Discussion

The electrical measurements were performed with a HP4156C semiconductor parameter analyzer. The set and reset current density (J) versus voltage (JV) curves of TaO/HfO x and TaO/AlO x devices are shown in Fig. 2a, b respectively. Both initial resistance states (R initial) of the TaO/HfO x and TaO/AlO x devices were HRS. The virgin memory devices were programmed to LRS with positive bias and were swept back. Then, each cell was switched from LRS to HRS by applied negative voltage. Both J–V plots contain three cell sizes, namely, 0.1, 0.56, and 25 μm2. In the J–V plots, all curves from devices with different areas resemble each other, which indicates both TaO/HfO x and TaO/AlO x devices had (i) the same current density in the initial state, (ii) similar set and reset voltages, and (iii) the same current density in LRS and HRS. Moreover, the constant current density property is clearly illustrated by the resistance versus area (RA) plots in Fig. 2c, d. The strong area dependence in both R initial and LRS can be observed by the control of current density. Regardless of the scale of cell area and compliance current, the same on/off resistance ratio is kept in both devices. This constant current density switch characteristic implies the memory cells are uniformly programmed or erased by the electrical field. These devices are considered to have trapping-type switching properties, which strongly relate to the modulation of vacancies [8]. In the case of trapping-type RRAM, no sharp current jump has been observed during the set process, but sharp current jumps have been commonly observed for filamentary-type RRAM. In the present research, different switching voltages were observed for the different switching layers with HfO x or AlO x . The set voltage range of a TaO/AlO x device is 4 to 4.5 V, which is larger than that of a TaO/HfO x device (3 to 4 V). The reset voltage range of a TaO/AlO x device is −1.5 to −2.5 V, which is larger than that of a TaO/HfO x device (−0.5 to −1.5 V). An AlO x system consumes more energy to complete the set and reset switches than a HfO x system consumes. During the setting of switches, the switching layers HfO x and AlO x achieve soft breakdowns at voltages of approximately 3 and 3.5 V, respectively. In both types of devices, before filaments form in the switching layer, the current is limited by the internal resistance of the TaO layer. During the self-compliance process of trapping-type RRAM, excessive oxygen vacancies are generated inside the switching layer [7]. Those oxygen vacancies are recombined during the negative biasing reset process. Unlike filamentary-type RRAM, the HRS is always lower than the initial resistance state (IRS) after a reset operation [15,16,17]. To summarize, defect-trapping is a process that modulates vacancies through oxygen ion–vacancy recombination to control the resistance variation in the switching layer. Compared with a HfO x switching layer, defect-trapping causes higher voltage and power in the AlO x layer during both the setting and the resetting of a switch.

Fig. 2
figure 2

Current density with voltage plot of a TaO/HfO x devices with different cell sizes. b TaO/AlO x devices with different cell sizes. The resistance versus the area plot of c a TaO/HfO x device and d a TaO/AlO x device. Both plots contain the IRS and LRS with reading voltage = 2 V. Each data point provides the average of 10 devices and the corresponding standard deviation

After the switching behavior had been investigated, the HRS and LRS retention behaviors of the trapping-type memory units were investigated. The plots of resistance variation versus time at 85 °C and 1 V for the TaO/HfO x and TaO/AlO x devices are shown in Fig. 3a, b. In both plots, the LRS variation is more pronounced than the HRS variation. The resistance stability of TaO/AlO x is higher than that of TaO/HfO x . The figures illustrate that the HRSs tended to drift toward the IRSs for both types of devices; the IRSs are marked by dashed lines in Fig. 3a, b. The trend of resistance coming back to device’s virgin state is depicted in Fig. 3c for TaO/AlO x and in Fig. 3d for TaO/HfO x . To realize this, both types of devices were initially programmed to LRS at room temperature, as shown in the I–V sweeps (black line). Then, the TaO/AlO x and TaO/HfO x devices were baked in ovens at 150 °C for 48 h and at 120 °C for 120 h, respectively. For both cases, the I–V sweep after having been baked was similar to the initial sweep. By this procedure, the LRSs of trapping-type devices were returned to the original states after time in a high-temperature environment. Unlike filament-type devices, which feature notable movement of oxygen atoms, trapping-type devices have pairs of oxygen ions and vacancies separated by short distances. The tendency of resistance drifting to the initial state is related to its original crystallinity, which is mainly controlled by the process temperature of ALD. As a result, the LRSs in both types of devices can be reset to HRSs (or IRSs) by negative bias or thermal energy. This property is different with filamentary RRAM.

Fig. 3
figure 3

Plots of resistance variation versus time for a TaO/HfO x and b TaO/AlO x devices. Both plots contain HRS and LRS variation at reading voltage = 1 V in 85 °C. After the IV sweeps of each virgin device had been set, the device was baked and then programmed to LRS again: c TaO/AlO x (150 °C for 48 h); d TaO/HfO x (120 °C for 120 h)

In standard retention testing for nonvolatile memory, data retention is tested at both room temperature and at high temperature; devices must be able to retain data at both room temperature and at high temperature to be useful in real applications. Activation energy (E a) extraction by the Arrhenius method in the retention plot is a common method to evaluate data retention [18, 19]. As can be seen in Fig. 3a, the LRS variation is more pronounced than the HRS variation. Therefore, the resistance ratio (R ratio) of LRS versus baking time at temperatures ranging from 30 to 150 °C was analyzed. One example of retention time extraction from a TaO/AlO x device is shown in Fig. 4a. The resistance degradation rate can be calculated by the slope of linear fitting in log(R ratio)-log(time) scale. By considering the maximum on/off resistance ratio of approximately 103 for a TaO/AlO x device, as shown in Fig. 3c, a retention time with 103 times the LRS variation can be calculated. The estimated LRS data retention at measurement temperatures ranging from 30 to 150 °C is shown in Fig. 4b. Each data point represents information from more than 18 devices for both device types. In a TaO/AlO x device, data retention is as high as 106 s at 150 °C and 2 × 108 s (approximately 5 years) at room temperature; those times are almost 101.5 times longer than those of a TaO/HfO x device. The most interesting point is that both TaO/HfO x and TaO/AlO x devices show the same E a = 0.38 eV, as calculated from the extracted slope. The same E a implies that both types of devices undergo similar chemical reactions in the LRS degradation process. This E a is involved in all thermally activated kinetic processes, including the release of oxygen ions near TaO interfaces and the oxygen diffusion processes in AlO x and HfO x layers. However, the oxygen self-diffusion coefficients of HfO x and AlO x layers are different at high temperatures (>1000 °C); exact measurements can be found in the literature [20, 21]. The oxygen diffusion coefficient at low temperature (<200 °C) also depends on the thickness of HfO x dielectrics [22]. If the diffusion processes in switching layers dominate the chemical reaction, then the E a values should be different due to the different diffusion coefficients in HfO x and AlO x layers. Both types of devices in this work exhibited the same E a = 0.38 eV; this was related to the fact that both types of devices had the same capping layer of TaO on the top of the switching layers. LRS degradation is a process of recombination of vacancies and ions, which means the TaO layer controls this chemical reaction and most of the vacancies are crowded near the interface between the TaO and the switching layer. Those vacancies prefer to stay on the TaO/switching layer interface; this phenomenon could be supported by the thermodynamic stability point of view, as reported by Zhong et al. [23]. In their simulation of TiN/Ta/HfO x /TiN stacks, the oxygen ions preferred to stay on the Ta/HfO x interface because a low energy difference existed between Ta and HfO x [23]. In their simulation, as in the present experiments, the TaO resistive layer trapped most of the oxygen ions and dominated this vacancy recombination process. LRS degradation is schematized in Fig. 4c. The oxygen ions return to the previous thermal equilibrium state during the baking process, which results in retention loss. Differences can be noted between the Ta/HfO x device as proposed by Zhong et al. and the TaO/HfO x device in this study, but in both studies, the TaO layer was formed by several cycles of metal Ta deposition and LTPO processes [14]. Because of the LTPO process, the metal-rich TaO/HfO x interface can be considered as an oxygen ion reservoir. During the recombination process of oxygen ions and vacancies, the atom packing density plays an essential role. The superior LRS retention properties obtained in the AlO x switching layer could be explained by the high atomic density of the AlO x layer. It is well known that the bond length of Al–O is shorter than that of Hf–O [24, 25]. The short bond in the AlO x reduces the oxygen ion mobility due to high coulomb interaction, which results in a high oxygen vacancy diffusion barrier. This barrier causes retention time to be longer in a TaO/AlO x device than in a TaO/HfO x one.

Fig. 4
figure 4

a Resistance variation ratio versus baking time for different temperatures in TaO/AlO x devices. The average initial resistance was 179 MOhm with a reading voltage of 2 V, and the LRS resistance degradation rate was calculated by the linear fitting method in log(R ratio)–log(T) scale. b Estimated retention time (1000×) versus 1/kT plot. Each point contains data from 18 devices taken at a reading voltage of 2 V. The extracted activation energies were 0.38 eV in both the TaO/AlO x and TaO/HfO x devices. c Retention schematic diagram of different oxygen diffusion barriers in HfO x or AlO x with a TaO capping layer

In addition, the retention loss model of a filamentary-type device is different from that of a defect-trapping-type device. The retention behavior for filamentary-type RRAM is related to filament rupture, and the vacancy diffusion direction is lateral [11, 19, 24]. In defect-trapping RRAM, the defect diffusion direction is longitudinal, which is parallel to the external electric field. Therefore, the retention behavior can be affected by the biasing direction and magnitude. Figure 5a, b shows the on-bias retention through the resistance ratio for the two devices. The resistance ratio is defined as the resistance of the stress device to the resistance of the LRS. A positive bias can help to maintain the LRS, but a negative bias accelerates the degradation process. Those on-bias properties could be explained by the interaction between the localized field of pairs of oxygen ions and vacancies and the external electrical field. If the direction of external field is the same as the set direction (positive), it extends retention time; if the external field is in the reset direction (negative), it causes degradation. In a low electric field with ±100 mV, the on-bias degradation is the same as the no-bias degradation in both types of devices. This ±100 mV bias might be covered by the band offsets of TiON-HfO x , TiON-AlO x , and TiN-TaO junctions. A TaO/AlO x device under a high positive bias of 500 mV shows no obvious degradation.

Fig. 5
figure 5

On-bias resistance ratio versus stress time for a TaO/HfO x and b TaO/AlO x devices at room temperature


In summary, we compared two types of self-rectified RRAM devices through their switch characteristics and analyzed their retention behaviors. The TaO/AlO x device showed a higher switching voltage and a more robust LRS thermal stability than the TaO/HfO x device did. The benefit of robust retention from the AlO x switching layer is due to the high oxygen diffusion barrier rather than activation energy. The activation energy of retention loss is related to the ion de-trap process in the TaO resistive layer. The high atomic density of AlO x film may improve LRS retention. A retention loss schematic model has been proposed and the on-bias retention results supported this model. This model could be beneficial for the development of low-current, long-retention, self-rectifying RRAM devices for future high-density memory applications.


  1. Baek IG, Park CJ, Ju H, Seong DJ, Ahn HS, Kim JH, Yang MK, Song SH, Kim EM, Park SO, Park CH, Song CW, Jeong GT, Choi S, Kang HK, Chung C (2011) Realization of vertical resistive memory (VRRAM) using cost effective 3D process. Int Electron Devices Meet 31.8.1. doi:10.1109/IEDM.2011.6131654.

  2.  Park SG, Yang MK, Ju H, Seong DJ, Lee JM, Kim E, Jung S, Zhang L, Shin YC, Baek IG, Choi J, Kang HK, Chung C (2012) A non-linear ReRAM cell with sub-1 μA ultralow operating current for high density vertical resistive memory (VRRAM). Int Electron Devices Meet 20.8.1. doi:10.1109/IEDM.2012.6479084.

  3. Ielmini D (2016) Resistive switching memories based on metal oxides: mechanisms, reliability and scaling. Semicond Sci Technol 31:063002

    Article  Google Scholar 

  4. Zhang L, Cosemans S, Wouters DJ, Govoreanu B, Groeseneken G, Jurczak M (2013) Analysis of vertical cross-point resistive memory (VRRAM) for 3D RRAM design. Int Memory Workshop :1–4. doi:10.1109/IMW.2013.6582122.

  5. Chen HY, Yu S, Gao B, Huang P, Kang J, Wong HSP (2012) HfOx based vertical resistive random access memory for cost-effective 3D cross-point architecture without cell selector. Int Electron Devices Meet ᅟ:20.7.1–20.7.4. doi:10.1109/IEDM.2012.6479083.

  6. Lin YD, Chen YS, Tsai KH, Chen PS, Huang YC, Lin SH, Gu PY, Chen WS, Chen PS, Lee HY, Rahaman SZ, Hsu CH, Chen FT, Ku TK (2015) Highly robust self-compliant and nonlinear TaOx/HfOx RRAM for 3D vertical structure in 1TnR architecture. Intl Symp VLSI Tech Sys and App. doi:10.1109/VLSI-TSA.2015.7117559

    Google Scholar 

  7. Chen YS, Lee HY, Chen PS, Chen WS, Tsai KH, Gu PY, Wu TY, Tsai CH, Rahaman SZ, Lin YD, Chen F, Tsai MJ, Ku TK (2014) Novel defects-trapping TaOx/HfOx RRAM with reliable self-compliance, high nonlinearity and ultra-low current. IEEE Electron Dev Lett 35:202–204

    Article  Google Scholar 

  8. Govoreanu B, Crotti D, Subhechha S, Zhang L, Chen YY, Clima S, Paraschiv V, Hody H, Adelmann C, Popovici M, Richard O, Jurczak M (2015) a-VMCO: a novel forming-free, self-rectifying, analog memory cell with low-current operation, nonfilamentary switching and excellent variability. VLSI Technology Symposium 59:T132–T133

    Google Scholar 

  9. Prakash A, Jana D, Maikap S (2013) TaOx-based resistive switching memories: prospective and challenges. Nanoscale Res Lett 8:418

    Article  Google Scholar 

  10. Subhechha S, Govoreanu B, Chen Y, Clima S, Meyer KD, Houdt JV, Jurczak M (2016) Extensive reliability investigation of a-VMCO nonfilamentary RRAM: relaxation, retention and key differences to filamentary switching. IEEE Int Reliab Phys Symp :6C.2.1–C6.2.5. doi:10.1109/IRPS.2016.7574568.

  11. Choi S, Lee J, Kim S, Lu WD (2014) Retention failure analysis of metal-oxide based resistive memory. Appl Phys Lett 105:113510

    Article  Google Scholar 

  12. Yu S, Chen YY, Guan X, Wong HSP, Kittl JA (2012) A Monte Carlo study of the low resistance state retention of HfOx based resistive switching memory. Appl Phys Lett 100:043507

    Article  Google Scholar 

  13. Chen CY, Fantini A, Goux L, Degraeve R, Clima S, Redolfi A, Groeseneken G, Jurczak M (2015) Programming-conditions solutions towards suppression of retention tails of scaled oxide-based RRAM. Int Electron Devices Meet :10.6.1–10.6.4. doi:10.1109/IEDM.2015.7409671.

  14. Chen YS, Lee HY, Chen PS, Lin YD, Tsai KH, Hsu CH, Chen WS, Tsai MJ, Ku TK, Wang PH (2016) Low power/self-compliance of resistive switching elements modified with a conduction Ta-oxide layer through low temperature plasma oxidization of Ta thin film. Intl Symp VLSI Tech Sys and App. doi:10.1109/VLSI-TSA.2016.7480496

    Google Scholar 

  15. Nardi F, Larentis S, Balatti S, Gilmer DC, Ielmini D (2012) Resistive switching by voltage-driven ion migration in bipolar RRAM—part I: experimental study. IEEE Electron Dev Lett 59:2461–2467

    Article  Google Scholar 

  16. Maikap S, Jana D, Dutta M, Prakash A (2014) Self-compliance RRAM characteristics using a novel W/TaOx/TiN structure. Nanoscale Res Lett 9:292

    Article  Google Scholar 

  17. Jana D, Samanta S, Roy S, Lin YF, Maikap S (2015) Observation of resistive switching memory by reducing device size in a new Cr/CrOx/TiOx/TiN structure. Nanoscale Res Lett 7(4):392–399

    Article  Google Scholar 

  18. Ielmini D, Nardi F, Cagli C, Lacaita AL (2010) Size-dependent retention time in NiO-based resistive-switching memories. IEEE Electron Dev Lett 31:353–355

    Article  Google Scholar 

  19. Chen YY, Komura M, Degraeve R, Govoreanu B, Goux L, Fantini A, Raghavan N, Clima S, Zhang L, Belmonte A, Redolfi A, Kar GS, Groeseneken G, Wouters DJ, Jurczak M (2013) Improvement of data retention in HfO2/Hf 1T1R RRAM cell under low operating current. Int Electron Devices Meet 10.1.1–10.1.4. doi:10.1109/IEDM.2013.6724598.

  20. Clima S, Chen YY, Degraeve R, Mees M, Sankaran K, Govoreanu B, Jurczak M, Gendt SD, Pourtois G (2012) First-principles simulation of oxygen diffusion in HfOx: role in the resistive switching mechanism. Appl Phys Lett 100:133102

    Article  Google Scholar 

  21. Oishi Y, Kingery WD (1960) Self-diffusion of oxygen in single crystal and polycrystalline aluminum oxide. Journal of Chemical Physics 33:480

    Article  Google Scholar 

  22. Zafar S, Jagannathan H, Edge LF, Gupta D (2011) Measurement of oxygen diffusion in nanometer scale HfO2 gate dielectric films. Appl Phys Lett 98:152903

    Article  Google Scholar 

  23. Zhong X, Rungger I, Zapol P, Nakamura H, Heinonen YAO (2016) The effect of a Ta oxygen scavenger layer on HfO2-based resistive switching behavior: thermodynamic stability, electronic structure, and low-bias transport. Phys Chem Chem Phys 18:7502–7510

    Article  Google Scholar 

  24. Traoré B, Blaise P, Vianello E, Grampeix H, Jeannot S, Perniola L, Salvo BD, Nishi Y (2015) On the origin of low-resistance state retention failure in HfO2-based RRAM and impact of doping/alloying. IEEE Tran Electron Devices 62:4029–4036

    Article  Google Scholar 

  25. Traoré B, Blaise P, Vianello E, Grampeix H, Bonnevialle A, Jalaguier E, Molas G, Jeannot S, Perniola L, DeSalvo B, Nishi Y (2014) Microscopic understanding of the low resistance state retention in HfO2 and HfAlO based RRAM. Int Electron Devices Meet 21.5.1–21.5.4. doi:10.1109/IEDM.2014.7047097.

Download references


This manuscript was edited by the Wallace Academic Editing.

Competing Interests

The authors declare that they have no competing interests.

Authors’ Contributions

YDL, KHT, WSC, and CHH fabricated the RRAM devices on 8-in. wafer under the instruction of PHW. YDL measured the devices under the instruction of SZR, HYL, and CJL. YSC also measured the devices. PSC, CJL, and YCK contributed to the understanding of the retention characteristics. YDL wrote the first draft, and the final draft was modified by PHW, SZR, and CJL. All the authors contributed to the preparation and revision of the manuscript, and they approved the final draft for publication.

Publisher’s Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Author information

Authors and Affiliations


Corresponding authors

Correspondence to Yu-De Lin or Chrong Jung Lin.

Rights and permissions

Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (, which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Lin, YD., Chen, PS., Lee, HY. et al. Retention Model of TaO/HfO x and TaO/AlO x RRAM with Self-Rectifying Switch Characteristics. Nanoscale Res Lett 12, 407 (2017).

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI:


  • Retention
  • TaO/HfO x
  • TaO/AlO x
  • Self-rectifying
  • Resistive memory
  • Trapping-type