O 3. /Ni RRAM for synaptic applications

Transcription

1 Semiconductor Science and Technology PAPER Understanding the gradual reset in Pt/Al O /Ni RRAM for synaptic applications To cite this article: Biplab Sarkar et al Semicond. Sci. Technol. Manuscript version: Accepted Manuscript Accepted Manuscript is the version of the article accepted for publication including all changes made as a result of the peer review process, and which may also include the addition to the article by IOP Publishing of a header, an article ID, a cover sheet and/or an Accepted Manuscript watermark, but excluding any other editing, typesetting or other changes made by IOP Publishing and/or its licensors This Accepted Manuscript is IOP Publishing Ltd. During the embargo period (the month period from the publication of the Version of Record of this article), the Accepted Manuscript is fully protected by copyright and cannot be reused or reposted elsewhere. As the Version of Record of this article is going to be / has been published on a subscription basis, this Accepted Manuscript is available for reuse under a CC BY-NC-ND.0 licence after the month embargo period. After the embargo period, everyone is permitted to use copy and redistribute this article for non-commercial purposes only, provided that they adhere to all the terms of the licence Although reasonable endeavours have been taken to obtain all necessary permissions from third parties to include their copyrighted content within this article, their full citation and copyright line may not be present in this Accepted Manuscript version. Before using any content from this article, please refer to the Version of Record on IOPscience once published for full citation and copyright details, as permissions will likely be required. All third party content is fully copyright protected, unless specifically stated otherwise in the figure caption in the Version of Record. View the article online for updates and enhancements. This content was downloaded from IP address... on 0/0/ at :

2 Manuscript version: Accepted Manuscript The Accepted Manuscript is the author s original version of an article including any changes made following the peer review process but excluding any editing, typesetting or other changes made by IOP Publishing and/or its licensors. During the embargo period (the month period from publication of the Version of Record of this article), the Accepted Manuscript: is fully protected by copyright and can only be accessed by subscribers to the journal; cannot be reused or reposted elsewhere by anyone unless an exception to this policy has been agreed in writing with IOP Publishing As the Version of Record of this article is going to be/has been published on a subscription basis, this Accepted Manuscript will be available for reuse under a CC BY-NC-ND.0 licence after a month embargo period. After the embargo period, everyone is permitted to copy and redistribute this article for Non-Commercial purposes only, provided they*: give appropriate credit and provide the appropriate copyright notice; show that this article is published under a CC BY-NC-ND.0 licence; provide a link to the CC BY-NC-ND.0 licence; provide a link to the Version of Record; do not use this article for commercial advantage or monetary compensation; and only use this article in its entirety and do not make derivatives from it. *Please see CC BY-NC-ND.0 licence for full terms. View the Version of Record for this article online at iopscience.org This content was downloaded from IOPscience

3 Page of CONFIDENTIAL - AUTHOR SUBMITTED MANUSCRIPT SST-.R Understanding the Gradual Reset in Pt/Al O /Ni RRAM for Synaptic Applications Biplab Sarkar, Bongmook Lee, and Veena Misra Department of Electrical and Computer Engineering, North Carolina State University, Raleigh, North Carolina,, USA Abstract In this work, a study to understand the gradual reset in AlO RRAM has been performed. Concentration of vacancies created during the forming or set operation is found to be playing a major role for in the reset mechanism. Reset was observed to be gradual when a significantly higher number of vacancies are created in the dielectric during set event. A The vacancy modulation technique is concentration inside the dielectric was increased using a multistep forming method then used to realize which resulted in a diffusion dominated gradual filament dissolution during reset in AlO RRAM. This gradual dissolution of the filament allows to control the conductance of the dielectric during reset. The RRAM devices with gradual reset show excellent endurance and retention for multi-bit storage. Finally, the conductance modulation characteristics realizing a synaptic learning is are also confirmed in the RRAM. Keywords RRAM, ALD, AlO, Set, Reset, Forming, synapse, drift, diffusion. I. INTRODUCTION Neuromorphic computing is expected to emulate brain functions in near future enabling the massive parallel computing and less complex architecture []. Current technology to emulate brain function of animals like spider, cat etc. requires excessive number of processors, memory, and computation time []. These barriers can be overcome with bio-inspired computing [-]. Commented [BS]: Title changed accordingly to Answer question (a) of Reviewer- and question () of Reviewer- Commented [BS]: Abstract changed accordingly to Answer question (a) of Reviewer- and question () of the Reviewers: In this paper, we present our work on understanding the diffusion dominated reset and the possibility of creating such reset in RRAMs. Our focus was to demonstrate that the device which usually shows abrupt reset can be changed into a gradual reset device with engineering the vacancy concentration inside the dielectric.

4 CONFIDENTIAL - AUTHOR SUBMITTED MANUSCRIPT SST-.R Page of Several non-volatile memory candidates such as phase change RAM (PCRAM) [], conductive bridge RAM (CBRAM) [], Resistive RAM (RRAM) [-] etc. has been proposed as synapse element, but RRAM is a promising candidate because of its CMOS compatibility and low energy switching properties []. RRAM have been reported to have excellent retention and endurance, and can be arranged in the form of cross-bar array thereby enabling high-density memory architecture []. The RRAM memory operation is achieved with a filament formation in the dielectric with a forming or set operation (low resistance state or LRS), followed by the rupture of filament for a reset operation ((high resistance state or HRS) []. During forming or set operation, oxygen ions are moved out of the dielectric leading to formation of a filament comprising of oxygen vacancies in the dielectric and oxygen ions are stored into the electrode []. The dielectric is prevented from going into hard breakdown by setting a compliance current (CC) limit []. During reset, oxygen ions from the electrode comes back to the dielectric and fills up those vacancies thereby rupturing the filament. However, all those vacancies are not replenished during reset event, and hence set event starts at higher current level than forming event []. In bipolar RRAM reset process, two main mechanisms has been proposed as (i) drift of oxygen ions by electric field and (ii) diffusion of oxygen ions caused by concentration gradient of oxygen ions stored in the electrode []. It should be noted that the temperature of the dielectric caused by Joule heating also plays a major role in the reset process [,]. The reset in dielectric has found to be either instantaneous [], or gradual where the current drops gradually during the reset []. One prime requirement of RRAM to be applicable for synaptic applications is the feasibility of gradual reset characteristics []. A recent report on reset in HfO RRAM was observed to be dependent on the concentration of oxygen ions stored in the electrode []. The

5 Page of CONFIDENTIAL - AUTHOR SUBMITTED MANUSCRIPT SST-.R reset behavior in the RRAM having less oxygen storage in the electrode was observed to be abrupt whereas the reset behavior with electrode having excess of oxygen storage was observed to be gradual []. Similarly gradual reset was observed in bipolar RRAM [] and unipolar RRAM [] where the filament comprised of higher number of vacancies, which in turn stores higher number of oxygen ions in the electrode []. Thus gradual reset is possibly because of oxygen ion concentration gradient diffusion dominated reset process which ruptures the filament gradually. Even though some binary oxide RRAMs has been reported showing gradual reset [,], but in many case these oxide typically shows abrupt reset thereby eliminating the possibility of realizing synaptic behavior [,,]. In this work, we studied the resistive switching properties of AlO RRAM to understand the gradual reset. It was observed that a gradual set process comprising of significantly higher number of vacancies in the dielectric while reaching CC leads to a gradual reset behavior. A multi-step forming technique was then employed to confirm the hypothesis gradual filament dissolution thereby realizing a gradual reset in AlO RRAM with excellent multi-state retention and endurance behavior. This gradual reset behavior was further extended to realize synaptic learning in the RRAM. II. EXPERIMENTAL In order to obtain bipolar resistive switching behavior, AlO RRAMs were fabricated on Si/SiO wafer with Ni and Pt as metal electrodes. with Pt-Ni electrode. nm Ni followed by nm Pt metal bottom electrode was deposited on Si/SiO wafer using UHV-RF sputtering system.. nm AlO dielectric as resistive switching layer was deposited using atomic layer deposition (ALD) technique at 0 0 C substrate temperature. nm Ni followed by 0 nm W capping layer Commented [BS]: Answer to question (c) of Reviewer- Answer to question () of the Reviewers: A similar observation has been shown in unipolar RRAM where the filament with higher number of vacancies resembles a gradual reset. In our work, we proposed that gradual reset is controllable by controlling the corresponding forming or set operation where we intentionally generate a higher number of vacancies during filament formation.

6 CONFIDENTIAL - AUTHOR SUBMITTED MANUSCRIPT SST-.R Page of serving as top electrode was deposited using UHV-RF sputtering system, and patterned with conventional photolithography. All the devices under measurement have area of 0 µm x 0 µm, CC was kept at 0 µa during forming and set operation of the RRAM, and memory operations were performed by applying bias to the Pt electrode []. Fig. : (a) Forming; (b) reset after forming and (c) two kinds of set-reset behavior observed in the RRAM. III. RESULTS AND DISCUSSION Fig. shows the forming, reset after forming and set-reset behavior of Pt/AlO/Ni RRAM. Forming was observed to have an abrupt rise in current before reaching the CC, shown in fig. (a). The reset after the forming was observed to be abrupt, shown in fig. (b). Previously, it has Commented [BS]: Answer to question (C) of Reviewer- Answer to question () of the Reviewers: In many cases, applying bias at PT electrodes has been observed to show bipolar resistive switching. This reference uses Ni/Pt as electrode choice and bias applied to Pt electrode.

7 Page of CONFIDENTIAL - AUTHOR SUBMITTED MANUSCRIPT SST-.R been reported that successive set-reset cycles increases the number of vacancies in the dielectric thereby reducing the Ion/Ioff ratio [-]. This is primarily because of vacancies which are created during set are not recovered completely during reset, leaving behind additional vacancies in the dielectric. A similar trend was observed during successive set-reset operations in AlO RRAMs, two different set-reset behaviors were observed in the AlO RRAM, shown in fig. (c). These two behaviors can be categorized as case-i: a set operation where abrupt rise in current leads to reach CC, the corresponding reset was also observed to be abrupt. Whereas in case-ii: a set where gradual rise in current leads to reach CC, and the corresponding reset was also gradual. Forming was observed to be similar to case-i, and the reset after forming was also similar to case-i reset. It is known that a gradual rise in current during forming or set event leads to generation of vacancies inside the dielectric thereby creating a filament []. Considering point A and point B in the set curves before the CC is reached in case of abrupt set (or case-i) and gradual set (or case-ii) respectively, point A have lower current at higher voltage signifying lesser number of vacancies present in the dielectric. Whereas in the case of point B of gradual set, higher number of vacancies are present in the dielectric resulting in higher current at lower voltage. This indicates the gradual set leads to generation of significantly higher number of oxygen vacancies and hence higher number of oxygen ions are stored in the Pt electrode in case of gradual set as compared to abrupt set. Higher number of vacancies inside the dielectric is known to create higher temperature gradient in the dielectric near the filament [,]. And Thus oxygen ions can recombine with the vacancies at lower reset currents because of concentration gradient dominated diffusion of oxygen ions in case of gradual reset []. Hence the reset in case-ii was possibly because of a diffusion dominated mechanism filament dissolution where vacancies are Commented [BS]: Answer to question (b) of Reviewer- Answer to question () of the Reviewers: In a few cases, this kind of decrease in the Ion/Ioff ratio occurs due to successive endurance cycles where all the vacancies generated during set are not completely recovered during reset [-]. Similarly, it can also happen because of random variation in filament creation process []. Commented [BS]: Answer to question () of We have cited the references to state that higher concentration of vacancies leads to increased temperature inside the filament.

8 CONFIDENTIAL - AUTHOR SUBMITTED MANUSCRIPT SST-.R Page of ruptured gradually. A recent work also demonstrated a similar result where higher concentration of oxygen ions stored in electrode after the set process resulted in a gradual reset []. In other words, the higher concentration of oxygen ions stored in the electrode after the gradual set process caused thermal diffusion of oxygen ions during reset operation causing a gradual reset dissolution of the filament. Also note that gradual reset was observed to be occurring at lower current compared to the abrupt reset case. Fig. : (a) Multi-step forming method used to generate higher number of oxygen vacancies before reaching CC, (b) corresponding reset after multi-step forming, and (c) DC endurance of the device after multi-step forming (CC was kept at 0 µa during set operations). To reconfirm the diffusion dominated reset hypothesis, a vacancy modulation technique comprising of multiple forming steps were carried out a virgin device was analysed and forming process was broken into three parts in order to generate a higher number of vacancies in the dielectric before the device reach final the CC, as shown in fig. (a). This kind of multi-step forming is known to create a higher vacancy concentration in the dielectric []. Similarly, multi-step set has also been proposed which can control the filament size []. During forming, Commented [BS]: Answer to question () of After observing the degradation and understanding the corresponding physical insight, virgin device was selected and forming process was broken into three parts so that a gradual reset can be obtained (corresponds to fig. of the manuscript). The variation in different set-reset voltage corresponds to device-todevice variation, and this has been a general concern for RRAMs.

9 Page of CONFIDENTIAL - AUTHOR SUBMITTED MANUSCRIPT SST-.R CC was kept at 0 na, µa and 0 µa respectively for successive steps, shown in fig. (a). Forming operation was observed to start at higher current levels with successive forming steps signifying increase in the concentration of vacancies in the dielectric with each step. Possible vacancy profile after successive sweep are also illustrated in fig. (a). The forming current at step- was observed to be similar to set at case-ii of fig. (c), which indicates higher number of vacancies are present in the dielectric during step- of the multi-step forming compared to abrupt forming of fig (a). The corresponding reset after the multi-step forming is shown in fig. (b), and the reset was also observed to be gradual which ascertains the hypothesis of concentration gradient dominant diffusion reset mechanism. Successive set-reset endurance cycles after obtaining gradual reset are shown in fig. (c). Excellent endurance was observed without any degradation for 0 DC endurance cycles. Unlike the abrupt reset case where the Ion/Ioff ratio varies randomly with successive endurance cycles, the gradual reset after vacancy diffusion showed almost no degradation even after 0 DC cycles. The set-reset uniformity was also observed to be greatly improved. This kind of controllable gradual reset is advantageous over abrupt reset, and are of particular interest to analog memory applications where RRAMs can resemble multi-bit operation with excellent retention and endurance []. Commented [BS]: Answer to question () of Previously, it has been mentioned that the endurance degrades within 0 set-reset cycles. We also observed the similar behavior, but after adopting the gradual set-reset behavior with a virgin device, we observed that there is almost no degradation even after 0 DC cycles. A similar result was also observed in HfO RRAM. Commented [BS]: Answer to question () of Tthere is device to device variation in set and reset currents, and this has been a concern in the RRAM community. This paper addresses the idea that gradual reset is caused primarily due to excess of oxygen vacancies in the dielectric. These measurements are completely repeatable, and has been observed over other AlO RRAMs and in HfO RRAM.

10 CONFIDENTIAL - AUTHOR SUBMITTED MANUSCRIPT SST-.R Page of Fig. : (a) Conductance change in AlO RRAM during reset with respect to reset stop voltage and reset pulse time; (b) retention characteristics of multi-state of the AlO RRAM. Next, the gradual reset characteristics was further used to study the conductance modulation of the AlO RRAM for realizing a synaptic element. Set operation was performed initially in the AlO RRAM keeping the CC at 0 µa, followed by the reset operation performed with either DC sweep or AC pulse. Read was performed at 0.V. Change in the conductance of the dielectric during reset with respect to reset stop voltage during DC reset and pulse time during AC reset are shown in Fig. (a). Both DC and AC reset were observed to result in dielectric conductance modulation, and the change in conductance of the dielectric was observed to higher in case of DC reset using a reset stop voltage as compared to AC reset. The retention characteristics of the multiple memory states obtained by choosing different DC reset stop voltage is shown in fig. (b). Excellent retention was observed in the AlO RRAM for four different memory states achieved with set and DC gradual reset operation. Commented [BS]: Answer to question () The retention measurement was carried out with AC pulse measurement at a Read voltage of 0.V. At this read voltage, the window is lower than X.

11 Page of CONFIDENTIAL - AUTHOR SUBMITTED MANUSCRIPT SST-.R Fig. : Synaptic learning in AlO RRAM with steps of set and 0 steps of reset, read was performed at 0.V. Finally, the synaptic operation was carried out in the RRAM where the conductance of the dielectric was changed gradually using AC pulses. The primary requirement of a synaptic circuit used in neuromorphic system is to detect very small change in resistance of the memory, and it has been reported that a smart synaptic system should detect % change in resistance per synaptic activity []. Synaptic RRAM element needs very fine control of conductance, and this The controllable resistance or conductance change in RRAM can be achieved by gradually varying the CC during the set and gradually increasing the pulse voltage during the reset []. Fig. shows the conductance change in AlO RRAM with steps of set and 0 steps of reset. During set of the device, CC was varied from 0 µa to 0 µa keeping the pulse voltage fixed at +. V, ms, whereas during reset, the pulse voltage was varied gradually from -0.V to - 0.V, read was performed at 0.V. A controllable conduction change was observed in the RRAM which confirms the applicability of using AlO RRAM as a synaptic memory after obtaining the gradual reset. Commented [BS]: Answer to question (d) of A small window is very common in reported RRAM synaptic devices and RRAM analog memories. The main requirement of synaptic circuit is to detect very small change in resistance of the memory, and it has been reported that a smart synaptic system should detect % change in resistance[].

12 CONFIDENTIAL - AUTHOR SUBMITTED MANUSCRIPT SST-.R Page of IV. CONCLUSIONS Engineering the Higher number of vacanciesy createdion during forming or set event results in either a diffusion dominated gradual or abrupt reset filament dissolution process during the reset. In order to generate a higher number of vacanciesy in the dielectric, A multi-step forming technique enables a forming step was broken into three parts which resulted in a reliable gradual reset in AlO RRAMs by creating higher concentration of vacancies in the thereby allowing a diffusion dominated reset mechanism. Such a gradual reset resulted in a multi-bit analog memory characteristics in AlO RRAM. The devices also showed no degradation in gradual reset characteristics for after 0 DC endurance cycles. The property of controllable conductance change was observed in AlO RRAM using DC and AC reset techniques. The conductance change via controlled reset which was further used to demonstrate synaptic learning behavior in AlO RRAM. Hence, by using a technique to generatinge a higher number of vacancies inside the dielectric during forming or set process, AlO RRAM was demonstrated to have gradual reset characteristics suitable for multi-bit analog memory storage and neuromorphic computation applications. ACKNOWLEDGEMENTS This work was supported by the National Science Foundation (NSF) under Grant number CNS. REFERENCES. D. Kuzum, S. Yu and H. S. P. Wong, Synaptic Electronics: materials, devices and applications, Nanotechnology,, (). Commented [BS]: Conclusion changed accordingly to Answer question (a) of Reviewer- and question () of Reviewer-

13 Page of CONFIDENTIAL - AUTHOR SUBMITTED MANUSCRIPT SST-.R. M. Suri, O. Bichler, D. Querlioz, O. Cueto, L. Perniola, V. Sousa, D. Vuillaume, C. Gamrat and B. DeSalvo, "Phase change memory as synapse for ultra-dense neuromorphic systems: Application to complex visual pattern extraction," IEEE InternationalElectron Devices Meeting (IEDM), pp ().. M. Suri, O. Bichler, D. Querlioz, G. Palma, E. Vianello, D. Vuillaume, C. Gamrat and B. DeSalvo, "CBRAM devices as binary synapses for low-power stochastic neuromorphic systems: Auditory (Cochlea) and visual (Retina) cognitive processing applications," IEEE International Electron Devices Meeting (IEDM), pp ().. Y. Wu, S. Yu, H. S.P. Wong, Y. S. Chen, H. Y. Lee, S. M. Wang, P. Y. Gu, F. Chen and M. J. Tsai; "AlOx-Based Resistive Switching Device with Gradual Resistance Modulation for Neuromorphic Device Application," IEEE International Memory Workshop (IMW), pp. - ().. S. Ambrogio, S. Balatti, F. Nardi, S. Facchinetti and D. Ielmini; Spike-timing dependent plasticity in a transistor-selected resistive switching memory, Nanotechnology,, 0 ().. S. Yu, B. Gao, Z. Fang, H. Yu, J. Kang and H. S. P. Wong; A Low Energy Oxide-Based Electronic Synaptic Device for Neuromorphic Visual Systems with Tolerance to Device Variation, Advanced Materials,, pp. ().. Y. Wu, S. Yu, B. Lee and H. S. P. Wong; Low-power TiN/AlO/Pt resistive switching device with sub- µa switching current and gradual resistance modulation, Journal of Applied Physics,, 0 ().. C. Cagli, J. Buckley, V. Jousseaume, T. Cabout, A. Salaun, H. Grampeix, J. F. Nodin,H. Feldis, A. Persico, J. Cluzel, P. Lorenzi,L. Massari,R. Rao, F. Irrera, F. Aussenac, C.

14 CONFIDENTIAL - AUTHOR SUBMITTED MANUSCRIPT SST-.R Page of Carabasse, M. Coue, P. Calka, E. Martinez, L. Perniola, P. Blaise, Z.Fang, Y. H. Yu, G. Ghibaudo, D. Deleruyelle, M. Bocquet, C. Müller, A. Padovani, O. Pirrotta, L. Vandelli, L. Larcher, G. Reimbold, B. de Salvo, Experimental and Theoretical Study of Electrode Effects in HfO based RRAM, IEEE International Electron Devices Meeting (IEDM), pp ().. L. Vandelli, A. Padovani, L. Larcher, G. Broglia, G. Ori, M. Montorsi, G. Bersuker and P. Pavan; Comprehensive physical modeling of forming and switching operations in HfO RRAM devices, IEEE International Electron Devices Meeting (IEDM), pp ().. S. Yu, and H. S. P.Wong; A Phenomenological Model for the Reset Mechanism of Metal Oxide RRAM, IEEE Electron Device Letters, vol., no., pp. - ().. X. Yang, S. Long, K. Zhang, X. Liu, G. Wang, X. Lian, Q. Liu, H. Lv, M. Wang, H. Xie, H. Sun, P. Sun, J.Sune and M. Liu, Investigation on the RESET switching mechanism of bipolar Cu/HfO/Pt RRAM devices with a statistical methodology, Journal of Physics D: Applied Physics,, ().. L. Vandelli,A. Padovani, L. Larcher andg. Bersuker, Microscopic Modeling of Electrical Stress-Induced Breakdown in Poly-Crystalline Hafnium Oxide Dielectrics, IEEE Transactions on Electron Devices, vol., no., pp. - ().. C. W. Zhong, W. H. Tzeng, K. C. Liu, H. C. Lin, K. M. Chang, Y. C. Chan, C. C. Kuo, P. S. Chen, H. Y. Lee, F. Chen and M. J. Tsai; Effect of ITO electrode with different oxygen contents on the electrical characteristics of HfOx RRAM devices, Surface and Coatings Technology, vol., pp. - ().

15 Page of CONFIDENTIAL - AUTHOR SUBMITTED MANUSCRIPT SST-.R. N. Raghavan, R. Degraeve, A. Fantini, L. Goux, D. J. Wouters, G. Groeseneken, M. Jurczak; "Stochastic variability of vacancy filament configuration in ultra-thin dielectric RRAM and its impact on OFF-state reliability," IEEE International Electron Devices Meeting (IEDM), pp ().. S. Long, L. Perniola, C. Cagli, J. Buckley, X. Lian, E. Miranda, F. Pan, M. Liu and J. Sune, Voltage and Power-Controlled Regimes in the Progressive Unipolar RESET Transition of HfO-Based RRAM, Scientific Reports,, ().. H. Lv, X. Xu, H. Liu, R. Liu, Q. Liu, W. Banerjee, H. Sun, S. Long, L. Li and M. Liu, Evolution of conductive filament and its impact on reliability issues in oxide-electrolyte based resistive random access memory, Scientific Reports,, ().. P. Huang, B. Chen, Y.J. Wang, F. F. Zhang, L. Shen, R. Liu, L. Zeng, G. Du, X. Zhang, B. Gao, J.F. Kang, X.Y. Liu, X.P. Wang, B.B. Weng, Y.Z. Tang, G. Q. Lo and D. L. Kwong, Analytical model of endurance degradation and its practical applications for operation scheme optimization in metal oxide based RRAM, IEEE Electron Device Meeting (IEDM), pp ().. Y. Z. Tang, Z. Fang, X. P. Wang, B. B. Weng, Z. X. Chen and G. Q. Lo, "A Novel RRAM Stack With TaOX/HfOy Double-Switching-Layer Configuration Showing Low Operation Current Through Complimentary Switching of Back-to-Back Connected Subcells," IEEE Electron Device Letters, vol., no., pp. - ().. B. Long, Y. Li and R. Jha, "Switching Characteristics of Ru/HfO/TiO-x/Ru RRAM Devices for Digital and Analog Nonvolatile Memory Applications," IEEE Electron Device Letters, vol., no., pp. 0-0 ().

16 CONFIDENTIAL - AUTHOR SUBMITTED MANUSCRIPT SST-.R Page of. J. Sun, Q. Liu, H. Xie, X. Wu, F. Xu, T. Xu, S. Long, H. Lv, Y. Li, L. Sun, and M. Liu, In situ observation of nickel as an oxidizable electrode material for the solid-electrolyte-based resistive random access memory, Applied Physics Letters,, 0 ().. D. Ielmini, F. Nardi, and C. Cagli, "Universal Reset Characteristics of Unipolar and Bipolar Metal-Oxide RRAM," IEEE Transactions on Electron Devices, vol., no., pp. - ().