Threshold Switching and Resistive Switching in SnO₂-HfO₂ Laminated Ultrathin Films

Abstract

Polycrystalline SnO₂-HfO₂ nanolaminated thin films were grown by atomic layer deposition (ALD) on SiO₂/Si(100) and TiN substrates at 300 °C. The samples, when evaluated electrically, exhibited bipolar resistive switching. The sample object with a stacked oxide layer structure of SnO₂ | HfO₂ | SnO₂ | HfO₂ additionally exhibited bidirectional threshold resistive switching properties. The sample with an oxide layer structure of HfO₂ | SnO₂ | HfO₂ displayed bipolar resistive switching with a ratio of high and low resistance states of three orders of magnitude. Endurance tests revealed distinguishable differences between low and high resistance states after 2500 switching cycles.

1. Introduction

Combinations of functional metal oxide films with highly different physical properties, such as conductivity, polarizability, hardness, etc., represented by, e.g., oxides of hafnium and tin, may offer opportunities to expand the usage of functional layers in devices, alongside the possibility of observing interesting properties which may emerge in material stacks. Thereby, HfO₂ thin films have a wide range of applications, including in optical coatings [1,2,3], gas sensors [4,5,6], waveguides [2,7], and gate dielectrics [8,9,10]. HfO₂ thin films have also been studied from the perspective of memory materials, specifically for use in resistive switching [11,12,13] or ferroelectric memory [14,15,16] devices. SnO₂ is a compound material which, in its thin film form, is of considerable interest owing to its well-defined structure and useful physical properties [17]. SnO₂ thin films also have many applications, some of which overlap with the applications of HfO₂, such as in gas sensors [18,19,20] and waveguides [21,22]. SnO₂ films have been studied as transparent conductors [23] and magnetic thin films [24]. In some cases, SnO₂ films have also been described as resistive switching materials [25,26,27].

SnO₂ and HfO₂ thin films have been prepared by sol–gel spin-coating [13,28], pulsed laser deposition [29,30], electron beam evaporation [31,32], reactive sputtering [33,34], spray pyrolysis [5,35], and chemical vapor deposition [36,37]. Atomic layer deposition (ALD) of SnO₂ has been studied extensively and realized by many chemical routes, listed in one comprehensive review [38]. The same can be said about HfO₂, which has been deposited using a number of different precursor combinations [39].

When it comes to discussing the fabrication methods of specifically resistive switching films, the behavior of SnO₂ as a switching media in memristive devices has been studied in few cases, while the deposition method chosen to date has, obviously, been sputtering [40,41,42]. Thereby, another compound, otherwise well known as a high-permittivity dielectric gate oxide, HfO₂, is also recognized as probably one of the most widely studied resistive switching materials, and it can be synthesized using a notable variety of techniques before mounting between different electrodes [43]. Specifically, the performance of HfO₂-based memristors has been explored via modeling as well [44].

The pairing of SnO₂ and HfO₂ has also been studied from a few different standpoints. For example, a layer of SnO₂ was inserted into HfO₂ to act as an oxygen stopping layer [45]. A SnO₂ anode was passivated by a HfO₂ surface layer [46], and a transistor was made with a SnO₂ channel which had a HfO₂ gate dielectric on top of it [47].

Methods for fabricating SnO₂/HfO₂ combinations in electronic devices have not yet included ALD, as far as the authors know. Spin-coating has recently been applied for the deposition of SnO₂/HfO₂ double layers as channel/gate junctions for field-effect transistors, providing appreciable charge mobility in relatively conductive SnO₂ film [48]. Spin-coating accompanying sol–gel synthesis is, however, a technique suited to the growth of rather thick functional films. The aforementioned SnO₂/HfO₂ junctions for transistor devices had also been fabricated by physical deposition methods [47], though one should account for the suitability of such methods, mainly to two-dimensionally devised structures.

ALD is the thin film growth method most properly suited to the conformal deposition of functional layers on three-dimensional device substrates, e.g., in the case of HfO₂ gate dielectrics supporting FinFET transistors [49]. Thereby, thin HfO₂ films are considered dielectric insulator material layers applicable together with channel layers made of a two-dimensional crystalline conductor, such as MoS₂, on planar [50] as well as on 3D [49] device substrates. In regard to contemporary and prospective concepts of three-dimensional resistive switching memory devices, ALD appears again as the method suited to the synthesis of HfO₂-based switching media [51].

The observations made to date imply that combinations of these two physically very different oxide materials may result in attractive properties, and the possible functionality in terms of the appearance of memory effects is worth examination. Prior to the fabrication of 3D device prototypes, the first investigations are to be conducted on solid layers grown onto planar substrates, in order to gain primary knowledge on their structural ordering as well as electrical performance.

The purpose of this work was to use ALD to create a nanolaminated thin film structure consisting of SnO₂ and HfO₂ layers, and to study it from the standpoint of resistive switching characteristics, after primary structural and compositional evaluation. Since HfO₂ and SnO₂ both have been reported to exhibit bipolar resistive switching characteristics, as referenced above, a hypothesis was made that layering these different materials together may induce interesting results.

2. Experimental Details

The samples which were studied in this work were deposited in a flow-type ALD reactor [52]. The pressure inside the reactor was 1.8 mbar at room temperature and 2.2 mbar at the working temperature of 300 °C. The precursor for tin was Tin(IV) iodide, SnI₄, (99.999%, Sigma Aldrich, St. Louis, MI, USA), which was evaporated at 83 °C inside the reactor, from a glass rod that had an opening at one end. Hafnium precursor was HfCl₄, (99.9%, Alfa Aesar, Black Freyer, MA, USA), which was evaporated at 162 °C. Nitrogen, N₂ (99.999%, AS Linde Gas, Tallinn, Estonia), was used as the carrier gas and for purging in between precursor pulses. Ozone with a concentration of 220–250 g/m³ was used as an oxidizer, which was produced from O₂ (99.999%, AS Linde Gas). All samples were deposited at 300 °C.

The deposition cycle times for SnO₂ were 5-2-5-5 s for the following sequence of pulses: metal precursor − N₂ purge − O₃ − N₂ purge. The pulse periods for the HfO₂ deposition cycle were 4-3-2-7 s. All films were grown on two different substrates. Firstly, Si(100) substrates were used, having been cleansed and etched prior to the growth. Secondly, for electrical measurements, films were grown on a TiN-covered electrode substrate that had been previously cleansed. The bottom electrode substrates were pieces of Si(100) wafers that had been coated with a crystalline and a conductive 10 nm thick TiN layer, which was deposited via pulsed chemical vapor deposition, using a batch TiCl₄/NH₃ process [53,54] at temperatures of 450–500 °C in an ASM A412 Large Batch 300 mm reactor at Fraunhofer IPMS-CNT.

The films grown on TiN-covered substrates for electrical measurements were also supplied with Ti/Au electrodes (area 0.204 mm²) that were electron-beam evaporated on top of the films through a shadow mask.

X-ray fluorescence (XRF) spectrometer Rigaku ZSX 400 was used to measure the elemental composition of films, using the in-built software of the ZSX Version 5.55 equipment. A spectroscopic ellipsometer (SE), model GES5-E, was used to measure the thickness of the films. Ellipsometric data were modelled using the Tauc–Lorentz dispersion model. The crystal structure of the samples was evaluated by grazing incidence X-ray diffractometry (GIXRD), using an X-ray diffractometer SmartLab Rigaku. The equipment employs CuKα radiation, which corresponds to an X-ray wavelength of 0.15406 nm.

Cross sections of the laminated structures were studied using scanning transmission electron microscopy (STEM). This was carried out with the FEI Titan Themis 200 instrument with an electron gun operated at 200 kV. The device was equipped with a Bruker SuperX SDD Energy-dispersive X-ray spectroscopy (EDX) system. The EDX system was used for the elemental composition profiling of the SnO₂-HfO₂ nanolaminates. Electrical measurements were performed by using a Cascade Microtech EPS150TRIAX station, which was controlled by a voltage source and measuring device, namely Keithley 2636A. All electrical measurements were performed at room temperature.

3. Results and Discussion

3.1. Film Growth and Composition

The analysis of the resulting solid films was carried out mainly on samples grown to comparable thicknesses. The number of ALD cycles was kept high enough to ensure a total thickness of the double-layered films of about 20 nm and individual thicknesses of 10 nm for constituent HfO₂ and SnO₂ layers. This required about 100 deposition cycles for HfO₂ and about 50 cycles for SnO₂, since the growth rates of reference HfO₂ and SnO₂ films, measured separately, were 0.11–0.13 and 0.18–0.20 nm/cycle, respectively. The exact growth rate varied somewhat from sample to sample, because the exact growth rate depends on the order of laminates—hafnia grows differently on tin oxide compared to on the silicon substrate and vice versa—and on the number of deposition cycles. The growth rate of HfO₂, 0.1 nm/cycle, corresponded to that which was previously reported in the literature at the same temperature and using the same precursor chemistry [55]. The growth rate of SnO₂ was harder to compare with the values reported in the literature, since the ALD process based on SnI₄ and O₃ precursors has previously been reported only by our group [56]. In the present study, all resulting films possessed thicknesses in the range of 22–25 nm.

Since the precursors used contained I and Cl, the amounts of these elements in the films were assessed as well. Since the optimization of growth for both HfO₂ and SnO₂ had already been published elsewhere [55,56], the amounts of these residues could also be held at low values in the present study, as the content of I was <1.0 at% and the content of Cl was <0.1 at%.

3.2. Film Structure

When the constituent oxide layer thickness was less than 5 nm, no distinct diffraction maxima were observed on the diffractogram (top pattern in Figure 1) of a four-layer laminate. Constituent layer thicknesses exceeding 10 nm enabled the appearance of a well-defined set of maxima, which allowed one to recognize the formation and presence of the monoclinic HfO₂ and a mixture of cubic/tetragonal SnO₂ (bottom two patterns in Figure 1).

Comparisons of STEM images and the EDX composition profiling of the 25 × SnO₂ + 50 × HfO₂ + 25 × SnO₂ + 50 × HfO₂ sample demonstrate that a multilayer, i.e., nanolaminate structure has been formed (Figure 2). The appearance of distinguishable lattice planes is indicative of a nanocrystalline order, which is not clearly observed in the GIXRD diffractogram.

3.3. Electrical Properties

Resistive switching as a phenomenon manifests itself via the alternate stabilization of high and low resistivity states in the solid material layer. The persistence of such memory states is not necessarily flawless. One of the drawbacks is current leakage through low resistive state (LRS) cells in the memory crossbar device, i.e., the sneak path problem [57]. Usually, a selector device like a transistor can be used to restrict sneak path currents; however, simpler two-terminal selector devices are preferable to maintain the overall minimal footprint of the memory cell [58]. A selector device is a highly non-linear device like a diode or a threshold switch. In the case of a threshold switch, the device provides a high resistance state (HRS) until the applied voltage is increased to a certain value, and then it switches to an LRS. The LRS is retained while the voltage is reduced down to a threshold voltage value. In this way, an LRS forms a window of voltage that can be used to switch a resistive switching device that is connected in series. In an attempt to reduce the risk of contamination by common diffusive contaminants like Cu and Ag in silicon electronics, Sonde et al. [59] studied HfO₂-based conductive-bridge random-access memory (CBRAM), replacing the commonly used anode material, e.g., Cu or Ag, with Sb. CBRAM is based on the electrochemical metallization mechanism of resistive switching. The aforementioned study demonstrated the bipolar resistive switching and threshold switching capabilities of Pt/HfO₂/Sn structures. Figure 3 shows the current–voltage characteristic of a bidirectional threshold switching cell composed of a resistive switching medium sandwiched between Ti/Au and TiN electrodes. The resistive switching medium was deposited using a cycle sequence of 25 × SnO₂ + 50 × HfO₂ + 25 × SnO₂ + 50 × HfO₂, resulting in a multilayered structure consisting of 5 nm thick SnO₂, 7 nm thick HfO₂, 5 nm thick SnO₂, and 7 nm thick HfO₂ layers. Similar electrical behavior was observed by Sonde et al. [59] in a 4 nm thick HfO₂ device with Pt and Sn electrodes. Interestingly, in our study, the switching voltages at negative polarity coincide with the aforementioned study, whereas the asymmetrical I-V characteristics in our case may be related to the asymmetrical structure of the film [59]. In an LRS, the operating voltage window for a series device ranged approximately from −1 V down to −3 V in the case of negative voltage polarity, and from 1 V to nearly 2 V at positive voltage polarity. The ratio between current values in the HRS and LRS increased by over three orders of magnitude, whereas Sonde et al. [59] achieved approximately one order of magnitude.

Figure 4 shows the I-V curve of a bipolar resistive switching device containing a solid medium deposited using a cycle sequence of 70 × HfO₂ + 30 × SnO₂ + 70 × HfO₂. The ratio between the HRS and LRS was, again, measured to be over around three orders of magnitude (Figure 4a). Nevertheless, one can notice that the I-V curve is rather unstable; thus, optimizing switching voltage pulses for endurance measurement would need further improvement. This device withstood about 2500 switching cycles, noticeably demonstrating LRS currents below 10 nA. Thereafter, the distinguishable difference between the HRS and LRS collapsed due to the resistivity loss in the HRS (Figure 4b). It is plausible that at such low operating currents, the source-meter current compliance was not sufficiently fast, causing the degradation of the HRS and the noticeably smaller HRS to LRS ratio. As seen in the discussion above, two specific samples of laminated structures exhibited resistive switching characteristics. Other systems were examined as well, varying the number of layers and their order, but they did not exhibit publication-worthy resistive switching characteristics. It is plausible that a certain balance between the more conductive SnO₂ and the more insulating HfO₂ is necessary for successful switching properties, and the reported films had this balance, but an exact assessment of what this balance entails is beyond the scope of this work and should be left for further studies.

4. Summary

Nanolaminates consisting of SnO₂ and HfO₂ layers were deposited by ALD, using SnI₄, HfCl₄, and O₃ as precursors. SnO₂ was found to grow in its cubic or tetragonal phase, whereas HfO₂ was formed mostly in its monoclinic phase. Films containing relatively thin layers, grown to thicknesses below 7–8 nm, remained weakly crystallized. A film deposited using a cycle sequence of 25 × SnO₂ + 50 × HfO₂ + 25 × SnO₂ + 50 × HfO₂ and mounted between Au and TiN electrodes exhibited threshold switching characteristics, with the ratio between the HRS and LRS increasing by nearly three orders of magnitude. The film deposited using a cycle sequence of 70 × HfO₂ + 30 × SnO₂ + 70 × HfO₂ acted as a bipolar resistive switching device, with a ratio between the HRS and LRS of around three orders of magnitude. The latter device withstood about 2500 cycles before failure.