Resistive switching properties for fluorine doped titania fabricated using atomic layer deposition

Flash memory has shown unprecedented lithographic development, including double patterning, three-dimensional (3D) vertical processing, data processing for multi-bit operation, and error correction code (ECC) optimization.^1–3 However, scaling limitations have also been demonstrated due to leakage current between word line (WL) and bit line (BL), coupling among adjacent cells, reduced saved charge as cell size decrease, etc. Since resistive random access memory (RRAM) exploits resistive switching (RS) behaviors relying on material inner ion movements, it can avoid flash memory problems that store data using charge traps.^4,5 Therefore, the RRAM scaling limit is relatively low, promoting RRAM as an ideal candidate for the next-generation non-volatile memory (NVM).^6–8 NVMs with a simple two-terminal structure are well-known for their high density using crossbar array architecture (CAA),^9–11 such as spin-torque transfer magneto-resistive RAM, phase change RAM, ferroelectric tunnel junction, and RRAM. The CAA structure addresses and operates memory cells by WL and BL for reading and writing, respectively, since each memory cell does not have a selector. Thus, crosstalk issue is induced during operation due to sneak current caused by parallel CAA circuitry;^6,12,13 hence, the sneak current increases with increasing array size. Previous studies have proposed various designs to manipulate sneak, including one selector–one resistor (1S–1R) structure^14–17 with various selectors, e.g., diodes,^6,18 Mott metal–insulator transition materials, potential barrier materials, and complementary resistive switches (CRSs).^19,20 However, reliable selector fabrication and their layout complexity undermines the advantages from the compact CAA structure.

ALD enables low-temperature deposition to produce highly uniform thin films.^21–23 This process aims to grow thin films by adopting self-limiting surface reactions, simplifying film thickness manipulation. Specifically, repetitive pulsing and purging composes chemical reactions with separately kept precursors while the deposition process continues. Thus, ALD produces thin films at low temperature with several advantages, such as high step coverage, uniformity, and precise film thickness control.²⁴ Separated precursor dosing also limits gas phase reactions, providing not only exceptionally reactive precursor adoption but also extending the time to complete each reaction step. These advantages help improve the scalability and reliability of complex 3D electronic devices. Thus, there is considerable research interest in producing 3D RRAM structures, focusing strongly on reliable resistance material fabrication using ALD.^25,26

However, reliable RRAM fabrication using the ALD process is not simple since RRAM RS behaviors require defect control. RRAM materials incorporate considerable oxygen vacancies to induce RS properties by oxygen migration. However, ALD exploits self-limiting surface reactions between reactants during film growth, which makes it fundamentally difficult to form oxygen defects within the film. Precursor dependency for ALD is also an obstacle to develop RS materials. On the other hand, oxygen defect control has been identified as a major mechanism underlying RRAM reliability.^27–29 Therefore, this study proposed and fabricated an F based RRAM device using fluorine as an alternative to oxygen defect for the main RS behavior agent.

We formed F:TiO₂ thin films by ALD using titanium tetra-isopropoxide (TTIP), deionized (DI) water, and an in-house fluorine source (HF dissolved in water) as precursors for Ti, O, and F reactants, respectively. Figure 1(a) shows the proposed ALD growth procedure for undoped TiO₂ films: TTIP (0.3 s) → N₂ purge (13 s) → H₂O pulse (0.1 s) → N₂ purge (13 s), where we pulsed water (H₂O) and F source (H₂O/HF) at a controlled y:x ratio to incorporate the F dopant into the films. Figure 1(b) shows one super-cycle for Ti–O:Ti–O/F (9:1) samples, i.e., one (x = 1) repetition of the ALD sequence [TTIP (0.3 s) → N₂ purge (13 s) → H₂O pulse (0.1 s) → N₂ purge (13 s)] and nine (y = 9) ALD sequences {TTIP (0.3 s) → N₂ purge (13 s) → H₂O/HF pulse (0.1 s) → N₂ purge (13 s) [Fig. 1(b)]}.

Figure 1(c) shows the ALD-F:TiO₂ film sequence with a single reactant (H₂O/HF) [TTIP (0.3 s) → N₂ purge (13 s) → H₂O/HF pulse (0.1 s) → N₂ purge (13 s)] to maximize the F concentration. The proposed technique provides a new protocol to synthesize oxygen related defects, which passivate TiO₂ oxygen defects, with fluorine and induce self-rectifying capacity.

Fluorine doped titanium oxide thin films were deposited on Pt(111)/Ti/SiO₂/Si substrates using a Traveling Wave Lucida D100 system (NCD Technology, Inc. Korea) at 150 °C deposition temperature and ∼1 Torr operating pressure. Titanium tetra-isopropoxide [(Ti(OPri)₄ or TTIP] (EG Chemical Co., Ltd., Korea), DI water, and hydrogen fluoride were used as Ti, O, and F precursors, respectively. The mixture of 50 ml DI water and 0.5 ml diluted hydrogen fluoride (HF 48%–51% diluted in water) were used to produce the homemade F source. For uniform dispersion of HF, the solution underwent five-minute stirring before being charged to a canister. Hydrofluoric acid and/or liquid hydrogen fluoride comprised the chemical state of this solution.³⁰ We used a heater to maintain TTIP below 95 °C in high purity N₂ (99.999%) atmosphere.

The ALD growth cycle for TiO₂ film deposition progressed as TTIP (0.3 s) → N₂ purge (13 s) → H₂O pulse (0.1 s) → N₂ purge (13 s) and for F:TiO₂ film deposition as TTIP (0.3 s) → N₂ purge (13 s) → HF pulse (0.1 s) → N₂ purge (13 s).

After deposition, TiO₂ and F:TiO₂ films were post-annealed for 1 h at 500 °C. A 100 nm thick, 20 µm diameter Pt top electrode was formed by using an e-beam evaporator and shadow mask.

High-resolution x-ray diffraction (HR-XRD, SmartLab, Rigaku) was used to measure the crystal structure with Cu Kα1 radiation (λ = 1.5416 Å). F:TiO₂ thin films were investigated using field emission scanning electron microscopy (FE-SEM, 7610f-plus, Hitachi) at 15 kV operating voltage with atomic force microscopy (AFM, MultiMode 8, Bruker, USA) to examine surface morphologies. Hall effect measurements (Ecopia HMS-3000) allowed calculating carrier concentration (n), resistivity (r), and Hall mobility (μ_H) using the van der Pauw method at room temperature. We used an ultraviolet–visible–near infrared (UV–vis–NIR) spectrophotometer (V-570, JASCO, 200–900 nm) to obtain optical transmittance spectra and Fourier transform infrared spectrophotometer (FTIR; Perkin Elmer 1760X spectrophotometer, 600–1000 cm⁻¹) to measure Ti–F chemical bonding. X-ray photoelectron spectroscopy (XPS) from an Al Kα monochromator (1486.6 eV) with variable spot size (30–400 µm in 5 µm steps) was used to identify elemental presence on the F:TiO₂ thin films (XPS, K-alpha, Thermo UK). RS characteristics for F:TiO₂ based RRAM devices were measured at room temperature using a KEITHLEY 2635A Sourcemeter.

Density of states (DOS) was calculated from first principles using density functional theory (DFT) and the CASTEP code.^31,32 We employed the Perdow–Burke–Ernzerhof exchange–correlation function with generalized gradient approximation (GGA) for geometrical optimization and GGA+U for DOS calculation.^33,34 Pseudopotentials were constructed using the ultrasoft method with cutoff energy = 380 eV and self-consistence field tolerance = 1.0 × 10⁻⁶ eV/atom. Brillouin zone integration was performed using 2 × 2 × 2 k-points following the Monkhorst–Pack approximation.³⁵ 

Hydrogen fluoride in one super-cycle can be calculated from the HF content in the diluted fluoride source solution, i.e., 0.5 ml 48%–51% diluted HF in 50 ml DI water ≈0.5% HF in the source canister, where HF(%) = [y/(x + y)] × 0.5%.³⁰ Both TiO₂ and F:TiO₂ (1.2) films were deposited at 150 °C and post-annealed at 500 °C, and then Ti 2p, O 1s, and F 1s were measured using XPS to examine fluorine presence and its relationship with TiO₂ to assure surface oxygen vacancy passivation on TiO₂.

Figures 2(a)–2(c) show Ti 2p_2/3, O 1s, and F 1s peaks at of 458.6, 529.9, and 684 eV binding energies, respectively, confirming F doping into TiO₂.^36,37 Figure 2(a) compares the Ti 2p_2/3 peaks for undoped TiO₂ and F:TiO₂ (1.2) films, exhibiting 0.2 eV shift attributed to the Fermi level receding from the conduction band due to oxygen vacancy passivation.³⁶ 

Figure S1 (supplementary material) shows XPS spectra of all F:TiO₂ (0, 0.5, 1.0, and 1.2) films, with corresponding DOS for TiO₂ (with oxygen vacancy) and F:TiO₂ (passivated oxygen vacancy) in Fig. S2 (supplementary material). Valence and conduction bands for TiO₂ and F:TiO₂ mainly came from the O p-orbital and Ti d-orbital, respectively. However, the TiO₂ Fermi level in the middle of the bandgap and the electron state located around the Fermi level can be attributed to oxygen defect states. The Fermi level moved to the top of the valence band for F:TiO₂ compared with TiO₂, the and oxygen vacancy defect energy level exhibited in TiO₂ vanished in F:TiO₂ due to oxygen vacancy passivation by F doping.

Figure 2(b) shows O 1s XPS and Gaussian fits to identify how oxygen and fluorine doping impacted oxygen vacancy passivation. Two constitutional O 1s peaks at ∼529.9 eV (O_I) correspond to the Ti–O binding energy, whereas the peak at ∼531.8 eV (O_II) corresponds to oxygen vacancies and non-lattice oxygen.³⁷ Reduced O_II peaks verify that fluorine doping passivated oxygen related defects. Figure S3 (supplementary material) shows that the peak area ratio rose from 5.9 to 15.17 eV after fluorine doping on the TiO₂ lattice_oxygen/defect_oxygen. Thus, fluorine successfully passivated TiO₂ oxygen related defects and non-lattice oxygen.

Figure 2(c) shows the F 1s peak for F:TiO₂, confirming fluorine presence in F:TiO₂ thin films. The peak at 684 eV corresponds to Ti–F binding energy. The F 1s binding energy intensity (peak at 684 eV) occurs above the 45 HF pulse portion. Table I shows that the F proportion in F:TiO₂ thin films increased with increasing HF pulse portion, reaching 1.2 % in the TiO₂ matrix. Fluorine proportion was calculated from XPS spectra for TiO₂ and F:TiO₂ films deposited on Pt substrates.

Figure S4 (supplementary material) shows FT-IR spectra for F:TiO₂ at F = 0, 0.5, 1.0, and 1.2 %. The main absorption area 750–850 cm⁻¹ corresponds to Ti–O bending, and the Ti–F peak at 890 cm⁻¹ confirms the Ti–F bond presence in the TiO₂ lattice.^38,39

Figure 3 shows HR-XRD patterns and AFM images for F:TiO₂ thin films for different fluorine concentrations. Anatase (JCPDS No. 21-1272) and rutile (JCPDS No. 21-1276) phases were found in all samples, with the main phase being anatase TiO₂. Anatase/rutile phase ratios were calculated as^40,41

and

respectively, where W_A and W_R represent anatase and rutile TiO₂ mass portions, respectively; and I_A and I_R represent diffraction peak integral intensity for anatase and rutile phase TiO₂, respectively. Anatase and rutile TiO₂ comprised 60.18% and 39.81% undoped TiO₂, respectively, and 59.22%, and 40.76% for F:TiO₂ (1.2), respectively.

The average grain size of F:TiO₂ (0, 0.5, 1.0 and 1.2) films were calculated from the Scherrer equation using HR-XRD results,

where β is FWHM of the peak, K = 0.9 (i.e., the shape factor), and λ = 0.154 06 nm (Cu Kα wavelength).

Thus, TiO₂ and F:TiO₂ (1.2) film grain sizes increased from 15.19 to 19.62 nm with increasing F concentration.^38,42 Figures 3(b)–3(e) show AFM images for F:TiO₂ on Pt substrates. F:TiO₂ root-mean-square (rms) roughness over 2 × 2 μm² increased with increasing fluorine concentration, achieving 1.37, 1.50, 1.92, and 2.28 nm rms roughness for F at. % = 0, 0.5, 1.0, and 1.2, respectively. Increased rms values were attributed to the crystalline growth and different growth rates of two TiO₂ phases of anatase and rutile. The proper surface roughness promoted nucleation and grain growth. Increasing rms value implies the TiO₂ crystallinity and grain size enhancement with increasing fluorine doping ratio.⁴² Thus, the fact that crystallinity in F:TiO₂ thin films improved as the F concentration increased can be confirmed based on the calculation of accurate grain size through XRD data and an increase in the rms value from the AFM images.

Figure S5 (supplementary material) shows that XPS peaks for F:TiO₂ film C 1s reduced as F doping increased, confirming fluorine improved TTIP ligand reactivity, thereby contributing to increased TiO₂ crystallinity. F:TiO₂ films residual carbon concentration changes provide further evidence for improved F:TiO₂ crystallinity. Figure 4(a) shows optical bandgaps with and without F:TiO₂, exhibiting interband absorption edge shift in the short wavelength region with increasing fluorine doping. Thus, fluorine doping increased the TiO₂ optical bandgap.⁴³ The optical bandgap was evaluated based on the absorption edge using Tauc’s equation,⁴⁴ 

where hν is the photon energy, n = 2 for the direct bandgap in the materials, and C is a constant.

Figure 4(a) shows optical bandgaps, estimated by extrapolating the linear portion of the curve to α = 0. The optical bandgap for undoped TiO₂ increased slightly (3.29, 3.30, 3.33, and 3.34 eV for 0, 0.5, 1.0, and 1.2 F doping, respectively), attributed to improved TiO₂ crystallinity due to F doping.⁴³ A band tail of TiO₂ refers to the localized electronic states located at the valence and edges of the conduction band. They were produced by disordered structures and defects (such as impurities or vacancies) of semiconductors. As F doping increases, the localized states of F:TiO₂ decrease because of the crystallinity enhancement and oxygen defect passivation.^45,46 Therefore, the valence band maximum decreased and the optical band of F:TiO₂ slightly enhanced more than the actual bandgap of TiO₂.

Figure 4(b) shows carrier concentration, mobility, and resistivity for TiO₂ and F:TiO₂ on glass substrates. Hall effects confirmed that F passivated vacancies and hence decreased carrier concentration. n-type conductivity was found in all films, and increasing fluorine doping concentration reduced the carrier concentration to 1.77 × 10¹⁴ (TiO₂) and 1.81 × 10¹³ cm⁻³ [F:TiO₂ (1.2)] due to the passivation of oxygen vacancies with fluorine ions. Unlike carrier concentration, mobility seems to increase in the F:TiO₂ (0.5) film but tends to decrease according to the increase of F concentration. It is noteworthy that in the case of the F:TiO₂ (0.5) film, the passivation effect reduced both the carrier concentration and grain boundary defect acting as the scattering center.^47,48 The reduction contributed to the mobility enhancement of the F:TiO₂ (0.5) film. However, for F:TiO₂ (1.0) and F:TiO₂ (1.2) films, the change of band tail state played a main role compared to the effect of scattering center mentioned above. The optical bandgap was slightly enhanced by F doping on TiO₂. This enhancement attributed to the decrease of TiO₂ band tail state due to F doping. The reduced band tail made it difficult for the electrons in F:TiO₂ films to move from the valence band to conduction band, leading to the decrease in mobility.⁴⁵ Thus, the mobility of F:TiO₂ (1.0) and F:TiO₂ (1.2) films decreased as the F concentration increased. As a result, the reduced charge carrier concentration and mobility contributed to the electrical resistivity increase to 774 (TiO₂) and 3433 Ω cm [F:TiO₂ (1.2)], respectively,

where ρ is the resistivity, σ is the conductivity, μ is the carrier mobility, and n is the carrier concentration.

Figure 5 shows that F:TiO₂ thin films exhibit different rectifying characteristics in I–V curves. Devices exhibit forward characteristics under positive voltage because TiO_2−x was formed locally around the bottom interface (in Pt/TiO₂/Pt/Ti/SiO₂) by Ti adhesion layer diffusion during post-annealing at 500 °C.⁴⁹ Yang et al.⁵⁰ previously showed Ti adhesion layer diffusion in the annealing process, where diffusion reduces TiO_2−x at the bottom interface and induces diode characteristics. Figure S6 (supplementary material) shows the XPS depth profiles of TiO₂/Pt/Ti/SiO₂ samples before and after post-annealing at 500 °C. Before post-annealing, the Ti 2p signal was not detected, as shown in Fig. S6(a), while 0.7. % of the Ti 2p signal in the Pt substrate region was detected after post-annealing [Fig. S6(b)]. Figure 6(c) shows the calculated O/Ti atomic ratio before and after annealing based on the depth profile data (etching time from 25 to 400 s). The atomic ratio before annealing was 1.95–2 with 400 s etching time. After post-annealing, the ratio was not changed with 200 s etching time, while it ranged from 1.8–1.85 under 200–400 s etching time. These results suggest that post-annealing induced the diffusion of the Ti adhesion layer under the Pt bottom electrode, which led to the formation of the TiO_2−x layer at the bottom interface and self-rectifying characteristic.

Although RS behaviors were exhibited by F:TiO₂ films, TiO₂ did not exhibit RS behaviors in Fig. 5. The blue arrows and numbers indicate sweep sequence (0 → +2.5 → 0 → −2.5 → 0 V, step size = 0.01). Pristine F:TiO₂ was in the high resistance state (HRS) and exhibited forming-less resistive switching and set-first behavior. All F:TiO₂ samples exhibited typical counterclockwise RS and asymmetric I–V curves.

The rectification ratio of F:TiO₂ films is shown in Fig. 6(a). The TiO₂ device exhibited a similar current level in both low resistance state (LRS) and HRS regardless of voltage polarity, whereas the F:TiO₂ (1.2) device showed a higher current level in the positive bias compared to that in the negative bias. In the case of the TiO₂ device, the rectification ratios of HRS and LRS were 1.4 and 1.6, respectively. On the other hand, for the F:TiO₂ (1.2) device, the rectification ratios of HRS and LRS increased to 4.0 and 10, respectively. The self-rectifying properties were indicated by the difference in height of the Schottky barrier at the top interface. While the amount of F doping increased in F:TiO₂, the Schottky barrier height at the top interface increased due to oxygen vacancy passivation. Therefore, the F:TiO₂ (1.2) films showed the largest Schottky barrier height difference at both electrodes interfaces. Based on this difference, it can be expected that the rectification ratio of F:TiO₂ (1.2) could be improved compared to TiO₂. Figure 6(b) shows the endurance characteristic of the Pt/F:TiO₂ (1.2)/Pt device measured at room temperature, which shows stable behavior of 250 cycles. Figure 6(c) shows the retention properties of LRS and HRS for V_read = 1 V at 85 °C over 24 h (time interval = 300 s) and the blue dash line indicates 10 years. Figure S7 (supplementary material) shows simulated CAA readout margin using the non-linear I–V curves in Fig. 5.

Figure 7(a) shows the Schottky barrier heights of both LRS and HRS in F:TiO₂ (1.2) at ±0.5 V. The Schottky barrier heights were extracted from temperature-dependent (300–380 K) experimental I–V data shown in Fig. S8 (supplementary material).^51,52 In HRS, the barrier heights were 0.24 eV (at +0.5 V) and 0.34 eV (at −0.5 V), and the Schottky barrier heights of LRS were 0.2 eV (+0.5 V) and 0.29 eV (−0.5 V), respectively. We found that Schottky barrier height changed in resistive switching behavior from the result.

Figure 7(b) shows the RS mechanism for F:TiO₂, inducing RS behaviors in F:TiO₂ films. We could expect the conduction mechanism of F:TiO₂ following an interface-limited conduction regime since the asymmetric I–V curve occurred. As shown in Figs. 2 and 5, despite the decrease in the amount of oxygen vacancies, the on/off ratio increased as the amount of F doping increased. Considering the change of on/off ratio with F doping concentration, it is suggested that the migration of the fluorine anion mainly contributed to the RS characteristics of F:TiO₂ because fluorine having a small ionic radius could migrate more easily. Under positive voltage (set process), the passivated fluorine anion in bulk migrated to the oxygen site and/or interstitial site of the top interface by the electric field or electron scattering. The migrated F ions at the top interface act as the n-type dopant, which decreases the Schottky barrier height and depletion width and also could provide defect level assisting charge injection. This change switches the resistance state of F:TiO₂ from HRS to LRS, and the applied negative voltage induces the reset as the opposite process.

This paper proposed and validated a simple ALD process for a new RS material F:TiO₂ for RRAM devices. Fluorine concentration in F:TiO₂ thin films can be controlled by engineering ALD precursor pulse portions for Ti–O/F and Ti–O. Increased fluorine doping concentration reduced oxygen defect in the fabricated TiO₂ films. XPS, FT-IR, and Hall measurements confirmed that F anion presence passivated oxygen vacancies in the film. TiO₂ crystallinity also increased because fluorine doping improved TTIP ligand reactivity, and the optical energy bandgap also slightly increased from 3.29 (TiO₂) to 3.34 [F:TiO₂ (1.2)] eV.

Thus, the proposed F:TiO₂ exhibited rectifying properties, forming-less operation, set-first process, and anti-clockwise RS. These phenomena were based on the interface limited conduction regime. The migration of fluorine mainly contributed to the RS characteristics by controlling the Schottky barrier height of the top interface of F:TiO₂. The on/off ratio and self-rectifying characteristics of F:TiO₂ films can be improved by optimizing the F doping concentration with inserting a tunnel barrier.^53,54 Enhanced asymmetric and RS properties due to fluorine presence in TiO₂ make the proposed system attractive to achieve sneak current-free CAA memory devices.

The supplementary material includes the XPS spectra, simulated DOS, oxygen/defect oxygen area ratio, FT-IR spectrum, C 1s spectra of XPS spectrum, and readout margins.

This work was supported by the South Korean Ministry of Trade, Industry, and Energy (Grant No. 10068075), the National Research Foundation of Korea funded by the Korean Government (Grant No. 2019R1A2C2087604), and the Creative Materials Discovery Program through the National Research Foundation of Korea funded by the Ministry of Science and ICT (Grant No. 2018M3D1A1058536).

The authors have no conflicts to disclose.

The data that support the findings of this study are available within the article and its supplementary material and from the corresponding authors upon reasonable request.