1 Introduction

Transition metal oxide-based resistive switching devices show their prominence in the recent times for nonvolatile memory (NVM) applications [1, 2]. Among them, high-\(\kappa\) dielectric materials have drawn the attention of the researchers for fabrication of NVM devices because the equivalent oxide thickness (EOT) of the gate dielectric layer can be scaled down to sub-nm dimensions. Among the high-\(\kappa\) elements, oxides of Hf [3], Zr [4], Ti [5], etc., have already settled their position in the NVM family. It is needless to mention that first resistive switching property of niobium oxide was observed by Hiatt et al. in 1965 [6]. Niobium (Nb) has now been considered because it offers various oxides with different physical and electrical properties [7]. Among them, NbO, NbO\(_2\), and \(\hbox {Nb}_{2}\hbox {O}_{5}\) are mostly studied that show metallic, semiconducting, and insulating electrical properties, respectively [8]. Having a relatively large value of dielectric constant, \(\hbox {Nb}_{2}\hbox {O}_{5}\) is a multi-purpose oxide and has been integrated into device structures which serve widespread applications [9]. Since niobium oxide (NbO\(_x\)) films possess several interesting properties, researchers have been exploring their properties for various applications where NbO\(_x\) films have been grown/deposited using sputtering [10], plasma oxidation [11], molecular beam epitaxy [12], pulsed laser deposition [13], etc. Researchers have so far concentrated on the electrochromic properties of niobium oxides [14]. Change in refractive index and deposition stress of sputter-deposited \(\hbox {Nb}_{2}\hbox {O}_{5}\) film has been reported recently by varying deposition pressure and oxygen content in the plasma [15]. In recent times, threshold switching properties of NbO-based metal–insulator–metal (MIM) devices have been presented [16, 17]. A multilayer NbO-based MIM structure with five-order on/off ratios is also reported [18]. There are several pending issues with NbO\(_x\) that require to be addressed.

In this communication, the niobium pentoxide films, deposited by dc reactive sputtering at different Ar/\(\hbox {O}_{2}\) ratios, were integrated to \(\hbox {Al}/\hbox {Nb}_{2}\hbox {O}_{5}/\hbox {Si}\) metal–oxide–semiconductor (MOS) structure to study their resistive switching properties. Grazing incidence X-ray reflectivity technique also is employed to estimate the porosity of \(\hbox {Nb}_{2}\hbox {O}_{5}\) thin films and to correlate its role with the resistive switching properties of the devices. Further, the XPS technique is employed for elemental mapping to explain various device properties. The results are presented and discussed here.

2 Experimental

n-type Si (100) substrates having a resistivity of 0.1–0.5 \(\Omega\)- cm and 1 cm \(\times\) 1 cm area were used for depositing the oxide films and subsequent MOS device fabrication. Niobium oxide films were deposited by dc reactive sputtering of a 99.9% metallic niobium target in \(\hbox {O}_{2}\)/Ar plasma at room temperature. A standard protocol for cleaning silicon wafers known as the Radio Corporation of America (RCA) cleaning technique was used for removing the metallic, organic, and inorganic impurities from the substrate surface [19]. The RCA cleansing was followed by a 1-min immersion in 1% hydrofluoric (HF) acid solution for stripping off the native \(\hbox {SiO}_{x}\) layer from the silicon surface prior to loading into the sputtering chamber. After achieving a base pressure of \(\sim\) 3\(\times\)10\(^{-6}\) mTorr, the deposition was carried out at a constant pressure of 5 mTorr for 40 min at 80 W dc power at different \(\hbox {O}_{2}\)/Ar ratios. The substrates were given a rotation of 30 rpm to ensure film uniformity. The sample nomenclature of the deposited films at various oxygen concentrations estimated as \({{\mathrm{O}}_{2}\over{\mathrm{O}}_{2}+\hbox{Ar}}\) is listed in Table 1. The thickness and surface roughness of the as-deposited films were extracted from XRR study using a Rigaku SmartLab 3 kW, X-ray diffractometer. The energies of the primary X-ray beam (\(\hbox {Cu} \hbox {K}_\alpha\)) used in XRR were 8 keV. To determine the stoichiometry of the \(\hbox {NbO}_{x}\) films and the oxidation states of niobium as well, XPS study was carried out at X-ray Photo-Electron Spectroscopy (PES) Beamline (BL-14) of Indus-2 at RRCAT, Indore, India. The energy of Indus-2 storage ring is 2.5  GeV. The achievable base pressure of the analytical chamber is \(\le\) \(5\times 10^{-9}\) mbar, while the electron beam energy was 4.4  keV. The XPS survey scan was taken by sweeping the binding energy from 604.2 to 0 eV with a pass energy of 150 eV [20]. A 300-nm-thick aluminum metal deposition on the niobium oxide films was performed by electron beam evaporation and patterned into 100-\(\mu\)m-diameter gate electrodes by UV photolithography technique. Figure 1 shows the schematic diagram of \(\hbox {Nb}_{2}\hbox {O}_{5}\)-based metal–oxide–semiconductor (MOS) structures. A voltage sweep was applied to the Al gate electrodes, and the corresponding gate current was measured using a keithley-make 4200 semiconductor parameter analyzer equipped with a 4200-PA amplifiers. The substrate chuck of the Signatone probe station was maintained at room temperature during the electrical characterizations. The probe station was isolated from light and shielded electrically. Ten devices for each set of samples have been measured.

Fig. 1
figure 1

A schematic diagram of Al/\(\hbox {Nb}_{2}\hbox {O}_{5}\)/Si device

Full size image
Table 1 Nomenclature of the reactively sputter-deposited niobium oxide thin films
Full size table

3 Results and discussion

The XRR data of the above four films deposited on Si substrate under different conditions are demonstrated in Fig. 2a. The XRR profiles were fitted using the Parratt formalism to estimate the film thickness [21]. During the Parratt fitting, the films were sectioned into multilayers such that the surface roughness and interface roughness were extracted accurately [22]. The inset of Fig. 2a displays the electron density profile (EDP) for all the samples. The mass density of \(\hbox {Nb}_{2}\hbox {O}_{5}\) was considered during the fitting process.

Fig. 2
figure 2

a XRR profiles for all as-deposited samples: colored symbol indicates experimental data for different \(\hbox {O}_{2}\) to Ar ratios. Solid black line shows the fitted data. The inset shows electron density profile of all the samples. b Deposition rate of \(\hbox {Nb}_{2}\hbox {O}_{5}\) films at different \(\hbox {O}_{2}\) concentrations. Error bars in the figure denote standard deviation

Full size image
Table 2 Different thin-film parameters extracted from the XRR data
Full size table

The parameters extracted from the Parratt fitting of the XRR profiles are depicted in Table 2. Upon analysis, it is revealed that the thickness of N1 sample is maximum, and the film thickness follows a decreasing trend with increasing \(\hbox {O}_{2}\) content in the reactive plasma. The sputter-deposited films at various \(\hbox {O}_{2}/\hbox {Ar}\) ratios show a very smooth surface with roughness ranging from 5 to 2 Å. These extremely low values are reproducible over thin film because of the advantages of the XRR technique.

Fig. 3
figure 3

AFM image of N3 sample

Full size image

Since Nb sputtering yield by Ar is more than that by \(\hbox {O}_{2}\), the deposition rate decreases with increasing \(\hbox {O}_{2}\) ratio in the plasma. The surface roughness also increases with \(\hbox {O}_{2}\) plasma content. To study the top surface morphology of the films, atomic force microscopy technique has been employed using AFM(CSI, Nano-Observer). The measurements have been taken for all the samples and found that roughness data are more or less similar as obtained from GI-XRR technique. A representative AFM image of N3 sample is shown in Fig. 3. The average roughness obtained from AFM data using Gwyddion software is 6±2 Å, and the same obtained from GI-XRR data is 5±0.4 Å. The N3 film shows the highest surface roughness of \(\sim\) 5 Å. It appears from the profile that a very thin porous \(\hbox {Nb}_{2}\hbox {O}_{5}\) film is deposited over relatively denser film at higher \(\hbox {O}_{2}\) content in the plasma which is evident from N3 and N4 samples. Deposition rates of \(\hbox {Nb}_{2}\hbox {O}_{5}\) film at different \(\hbox {O}_{2}\) concentrations in the plasma are shown in Fig. 2b. A sample was deposited at 66% \(\hbox {O}_{2}\) content in the plasma only to see its role on the deposition rate at higher concentration. It is also concluded that the deposition rate becomes nearly constant after a critical \(\hbox {O}_{2}\) content in the reactive plasma.

Fig. 4
figure 4

XPS spectra of \(\hbox {Nb}_{2}\hbox {O}_{5}\) films deposited at room temperature: a Nb 3d and b O 1s

Full size image

The deconvoluted XPS spectra of Nb 3d and O 1s for all the samples are shown in Fig. 4a and b, respectively. A software PeakFit 4.11 is used to deconvolute the XPS spectra. The deconvoluted Nb 3d spectra show two peaks at 207 and 209.7 eV for N1 sample denoting the \(\hbox {Nb}^{5+}\) 3d\(_{5/2}\) and 3d\(_{3/2}\) states, respectively, with an energy difference of 2.7 eV indicating the presence of \(\hbox {Nb}_{2}\hbox {O}_{5}\) [23]. The peak positions for all other samples are nearly the same with an energy difference of 2.7 eV. The results indicate formation only of \(\hbox {Nb}_{2}\hbox {O}_{5}\) oxides for all the samples. It further confirms complete oxidation of Nb atom/molecules during reactive sputtering due to the absence of any peak corresponding to binding energy of Nb metal. The hump in the binding energy of O 1s has been deconvoluted into two peaks as shown in Fig. 4b. The peak at \(\sim\) 530 eV corresponds to Nb–O bond, whereas the one at 532 eV is due to the presence of non-lattice oxygen in the film [24]. Signature of non-lattice oxygen in all the films is confirmed from the XPS spectra. The non-lattice oxygen content in all the films is depicted in Table 3. The percentage of non-lattice oxygen represents the percentage of oxygen vacancies (\(\hbox {V}_{O}\)) [25] that increases with increasing oxygen concentration up to a certain value beyond which it decreases. In other words, the lattice oxygen percentage decreases with increasing oxygen content beyond a certain oxygen concentration in the plasma, i.e., up to 40% in this case. The deposited film at lower oxygen content removes the oxygen defects and produces more lattice oxygen (Nb–O lattice) [26]. At higher oxygen flow rates, the system prefers to form stoichiometric \(\hbox {Nb}_{2}\hbox {O}_{5}\). Above 40% oxygen content of the plasma, the film composition remains constant and the oxygen content does not have any significant influence on the stoichiometry of the film.

Table 3 Percentage of non-lattice oxygen extracted from the XPS spectra
Full size table

The resistive switching properties of \(\hbox {Al}/\hbox {Nb}_{2}\hbox {O}_{5}\)/Si devices are depicted in Fig. 5. The voltage was swept across the devices following a sequence 0\(\rightarrow\) -20 \(\rightarrow\) +20 \(\rightarrow\) 0 V, and the gate current was measured. Current change of three to five orders is observed for all the devices in the negative cycle of the voltage sweep. The SET voltages for N1, N2, N3, and N4 devices are −17.5 V, −16.4 V, −11 V, and −9.4 V, respectively. The devices change from high-resistance state (HRS) to low-resistance state (LRS) at a certain threshold negative voltage bias. Thus, the SET process is attained. When the voltage sweeps from negative to positive direction, it shows hysteresis indicating slow removal of charges upon reverse voltage sweep to achieve the RESET state. Upon perusal of the film thickness and SET voltage data, it is seen that the electric field at which the SET process took place is \(\sim\) 3  MV/cm.

Fig. 5
figure 5

Room temperature \(I-V\) plots of the devices: a N1, b N2, c N3, and d N4

Full size image

The resistive switching behaviors of the MOS devices may be explained on the basis of porosity [27] and presence of non-lattice oxygen in \(\hbox {Nb}_{2}\hbox {O}_{5}\) film. It is revealed from EDP profile in Fig. 2a that N1 samples are denser, whereas the rest samples exhibit comparable porosity. As shown in Fig. 5a, N1 device offers the poorest performance, whereas N3 shows better device performance than the other devices. Analysis of XPS data confirms that N3 device contains relatively more oxygen vacancies. It is seen earlier that the oxygen vacancies in the insulating oxide layer directly affect the memory properties of the device [24]. Figure 6 shows the cycle-to-cycle variation of LRS and HRS current. Device with higher oxygen concentration in sputtering chamber gives better uniformity in current distribution. A five-order on/off ratio has also been observed using MOS structure, whereas the same order is also reported with MIM devices [18].

Fig. 6
figure 6

Variation of current at HR and LR states with number of cycles. Error bars in the figure denote deviation from the mean value

Full size image

The whole phenomena may be explained on the basis of oxygen filament theory. When a negative voltage sweep is applied to the top metallic Al electrode, oxygen ions present in the \(\hbox {Nb}_{2}\hbox {O}_{5}\) layer are pushed toward the bottom silicon. The accumulation of oxygen vacancies (\(\hbox {V}_O\)) is induced at the interface between top electrode and oxide layer. At a certain negative voltage, the forming process is finally completed creating a \(\hbox {V}_{O}\)-based conducting filament connecting the top and bottom electrodes. This CF forms a conduction path for electrons to travel from Al gate to the silicon substrate. The device then readily attains its LRS. In the reset process, the oxygen vacancies merge with the oxygen ions, thereby breaking the conductive filament. From the XRR data, it is evident that the porosity of N2, N3, and N4 samples shows nearly the same value. But the XPS data reveal more non-lattice oxygen content in the N3 film. The porosity in the film generates defects during sputtering, and the defects in the thin film contribute to trap sites. When a critical number of traps are filled, a sudden increase in the magnitude of current is observed. Therefore, a critical porosity along with non-lattice oxygen content is required to get large swing in the resistive switching behavior of the \(\hbox {Nb}_{2}\hbox {O}_{5}\)-based MOS devices. In other words, N3 exhibits larger HRS/LRS implying a combined role of porosity and non-lattice oxygen on the RS device performance.

4 Conclusion

\(\hbox {Nb}_{2}\hbox {O}_{5}\) films, deposited using reactive dc sputtering at different \(\hbox {O}_{2}/\hbox {Ar}\) ratios on Si substrate, have been characterized by X-ray photoelectron spectroscopy and X-ray reflectivity techniques. Metal–oxide–semiconductor devices were later fabricated with aluminum as top metal electrode with \(\hbox {Nb}_{2}\hbox {O}_{5}\) film as gate dielectric. A dependence of the switching properties on the porosity of the films is evident. Further, the non-lattice oxygen content in the film also plays a role in determining the on/off ratio of the devices. Therefore, it is concluded that both film porosity and presence of non-lattice oxygen determine the switching behavior of the \(\hbox {Nb}_{2}\hbox {O}_{5}\)-based devices.