Distinguishing Oxygen Vacancy Electromigration and Conductive

Jun 12, 2017 - By replacing the top electrode layer in a metal–insulator–metal memory structure with easily removable liquid electrolytes, either ...
3 downloads 7 Views 7MB Size
Letter pubs.acs.org/NanoLett

Distinguishing Oxygen Vacancy Electromigration and Conductive Filament Formation in TiO2 Resistance Switching Using Liquid Electrolyte Contacts Kechao Tang,† Andrew C. Meng,† Fei Hui,‡ Yuanyuan Shi,‡ Trevor Petach,§ Charles Hitzman,∥ Ai Leen Koh,∥ David Goldhaber-Gordon,§ Mario Lanza,‡ and Paul C. McIntyre*,† †

Department of Materials Science and Engineering, Stanford University, Stanford, California 94305, United States Institute of Functional Nano and Soft Materials, Collaborative Innovation Center of Suzhou Nano Science and Technology, Soochow University, Suzhou 215123, China § Department of Physics and ∥Stanford Nano Shared Facilities, Stanford University, Stanford, California 94305, United States ‡

S Supporting Information *

ABSTRACT: Resistance switching in TiO2 and many other transition metal oxide resistive random access memory materials is believed to involve the assembly and breaking of interacting oxygen vacancy filaments via the combined effects of field-driven ion migration and local electronic conduction leading to Joule heating. These complex processes are very difficult to study directly in part because the filaments form between metallic electrode layers that block their observation by most characterization techniques. By replacing the top electrode layer in a metal−insulator−metal memory structure with easily removable liquid electrolytes, either an ionic liquid (IL) with high resistance contact or a conductive aqueous electrolyte, we probe field-driven oxygen vacancy redistribution in TiO2 thin films under conditions that either suppress or promote Joule heating. Oxygen isotope exchange experiments indicate that exchange of oxygen ions between TiO2 and the IL is facile at room temperature. Oxygen loss significantly increases the conductivity of the TiO2 films; however, filament formation is not observed after IL gating alone. Replacing the IL with a more conductive aqueous electrolyte contact and biasing does produce electroformed conductive filaments, consistent with a requirement for Joule heating to enhance the vacancy concentration and mobility at specific locations in the film. KEYWORDS: Conductive filament, TiO2, electromigration, Joule heating, ionic liquid

R

defects in the TMO or of metal atoms that migrate from the electrodes into the oxide. Apart from determining the structure and properties of the filaments in a given TMO, one of the key issues for better understanding the mechanism of filament-based RRAM operation is the role of Joule heating in resistance switching of the device.3,20−22 Some reports23 have suggested that resistance switching can occur in TMO films as a result of field-driven ion motion alone, based on observations of local resistance degradation and breakdown of bulk samples.24−29 It is difficult to separate the effects of thermal versus field-driven phenomena in resistance switching of thin films because large electric fields generate both large and locally varying electronic and ionic currents. Furthermore, direct physical characterization of the filaments is complicated by the fact that they

andom access memories based on bias-induced resistance switching (RRAM) in metal−insulator−metal structures have attracted great attention1−6 due to advantageous device performance,7−9 fabrication compatibility with silicon circuitry,10 and the potential for three-dimensional stacking for highdensity memory.11 RRAM devices are sometimes classified according to whether switching between low resistance (on) and high resistance (off) states is governed by primarily thermal, ionic, or electronic processes,8 although in practice all of these phenomena are often interrelated. When the insulator layer is a transition metal oxide (TMO) stacked between metal electrodes, a variety of switching mechanisms are possible depending upon the specific TMO insulator and metallic electrodes used. These include oxygen vacancy electromigration,5,12,13 electrochemical metallization,14,15 charge-trapping,16 and voltage-programmable ferroelectric tunnel barriers.17 Among these, formation and breaking of conductive filaments in the TMO layer that span the gap between the electrodes is the most widely studied mechanism for resistance switching.18,19 These filaments may be composed of interacting point © 2017 American Chemical Society

Received: April 7, 2017 Revised: June 8, 2017 Published: June 12, 2017 4390

DOI: 10.1021/acs.nanolett.7b01460 Nano Lett. 2017, 17, 4390−4399

Letter

Nano Letters

Figure 1. Schematic and electrical measurement for ionic liquid gating on TiO2. (a) Schematic of the device structure for liquid electrolyte (including ionic liquid) gating. The outer diameter of the top electrode metal ring is 500 μm. The thickness of the photoresist layer is ∼1 μm. (b) Schematic procedure for electrical characterization of the devices before, during, and after IL gating. (c) Comparison of measured I−V data before and after IL gating under different bias conditions (showing potential on the IL relative to the substrate). An increase in the device conductivity is detected under positive bias, and this effect can be reversed by a subsequent negative biasing. (d) The measured current during IL gating as a function of time for different bias conditions. The diameter of the TiO2 window for the devices in (c,d) is 40 μm. (e) I−V data measured for different TiO2 window sizes. The change of I−V curves after IL gating is indicated with colored vertical arrows. The current increase is dependent on window diameter. Detailed numerical comparison can be found in Table S1.

microscopic processes in RRAM switching. Compared to conventional metal electrodes, IL contacts greatly limit current flow during biasing due to their typically large electrochemical stability window.35 When an IL is placed on the surface of a TMO layer and a bias is applied between them, ions in the IL drift toward the TMO and accumulate at the interface. As a result, large electrical fields can be achieved in an initially insulating TMO layer without significant current flow, making it simpler to distinguish the effects of local field-induced drift of ions from those of local Joule heating. It is worth noting that the electrochemical stability window of the IL can be reduced by atmospheric contaminants, especially water vapor,36,37 so procedures to remove such contaminants, including IL gating in a vacuum environment, are desirable (see Supporting Information). Detailed discussions of the trends in stability of ILs depending on their molecular electronic structure and contact interface structure can be found in prior literature reports.38−43 In addition, the IL can be readily washed away and replaced by other solutions, such as aqueous salt electrolytes, that support higher currents under bias, thus providing an approach to investigate the effect of Joule heating on resistance switching. Finally, because such liquid top contacts can be readily removed after testing, without altering

usually occupy a very small volume fraction of the oxide and may have nanoscale radial dimensions. The filaments also typically form between metal electrode layers that contact the TMO film, preventing direct imaging. Attempts to remove the electrodes by etching or other means after RRAM fabrication and testing may alter the point defect population of the TMO film, producing results that are not representative of an actual device. In this paper, we describe the use of ionic liquid (IL) gating to study local conductivity switching in titanium oxide-based RRAM structures. Ionic liquids have been employed routinely as the top gate contact of electric double layer transistors (EDLTs). Their wide electrochemical stability window allows large electric fields to be achieved in the EDLT channel layer without chemical alteration of the IL.30 IL gating is reported to induce strong modulation of the in-plane conductivity of EDLT surface channels in a variety of single crystal materials, including SrTiO3,31 VO2,32,33 and rutile TiO2.34 The ability to induce insulator−metal transitions in planar devices by IL gating motivates us to explore the possibility of expanding this technique to the study of filament formation in RRAM devices. Ionic liquid gating has a number of advantages compared to conventional metal−electrode gating for investigating the 4391

DOI: 10.1021/acs.nanolett.7b01460 Nano Lett. 2017, 17, 4390−4399

Letter

Nano Letters

An increase in conductivity is observed in all cases and the current after IL biasing shows a clear dependence on the TiO2 window size. A detailed numerical comparison is provided in Table S1. This strong area scaling of the on-state current indicates that resistance changes observed after IL gating are unlikely to arise from formation and breaking of single conductive filaments. Instead, the data suggest a more uniform change of TiO2 conductivity. Local conductivity changes of the TiO2 layer were studied by C-AFM imaging of the TiO2 surface. Compared to the fresh TiO2 surface (Figure 2a), after setting by positive IL gating,

the programmed TMO surface, direct characterization of the TMO structure, composition, and local conductivity is greatly simplified.32 We report the first application of vertical-field IL gating to study local programming of conductivity and filament formation in atomic layer deposited (ALD) TiO2, a common TMO in RRAM device studies. Electrical testing using both a probe station and conductive atomic force microscopy (CAFM) mapping are employed to characterize resistance changes in the TiO2. Mechanisms underlying the conductivity changes are examined using oxygen isotope exchange and depth profiling, electron energy loss spectroscopy (EELS) in scanning transmission electron microscopy (STEM), and storage of the ALD-TiO2 layers in atmospheres of varying oxygen activity after IL gating and liquid electrolyte biasing. The details of device fabrication and measurement procedures can be found in the Supporting Information. A primary electrical measurement result is shown in Figure 1. Following the process flow in Figure 1b, I−V curves before and after steady IL biasing were measured (Figure 1c). A significant current increase is detected during low-bias I−V sweeps after applying +5 V IL bias for 5 min (factor ∼10× at negative bias and factor ∼5× at positive bias), indicating an increase of the TiO2 conductivity. This conductivity increase is highly reproducible. Increasing the initial biasing time to 15 min at +5 V, however, causes very little further increase in TiO2 conductivity compared to the 5 min data (Figure 1c). In addition, postbias conductivity of the TiO2 is unaffected by higher positive bias values up to +14 V (Figure S1). For biasing conditions less intense than +5 V for 5 min, the degree of conductivity increase diminishes and becomes negligible when the biasing voltage applied to the IL drops below +4 V or the biasing time drops below 10 s. The time scale for IL gating to induce a notable change in the TiO2 conductivity is on the order of minutes, comparable to the results in previous studies of IL gating of metal oxide materials.34,36 Representative current values measured during IL biasing are displayed in Figure 1d. The current stays low at ∼10 nA for the majority of the biasing time. This current is significantly lower than the typical current for a conventional metal−insulator−metal filamentary device, which is usually larger than 10 μA.6,18,22 Consequently, assuming comparable operating voltage, the power decreases by more than 1000 times compared to conventional filamentary RRAM devices, and the effect of Joule heating is expected to be minimal. The observed low current during gating is consistent with the reported large energy gap between the highest occupied molecular orbital (HOMO) and the lowest unoccupied molecular orbital (LUMO) of ILs.39,41,43 IL gating can also recover the TiO2 insulating state by subsequently applying a steady negative bias, as shown in Figure 1c. This insulating state recovery is also highly repeatable. If higher reverse bias is applied to a fresh TiO2 surface, it further decreases the conductivity below the level of the initial state (Figure S2). Therefore, for positive and negative biasing of the IL top contact, the conductivity of the TiO2 can be increased (set) and decreased back to original level (reset), respectively, while passing very low currents (