Influence of transition metal doping on the structural, optical, and magnetic properties of TiO2 films deposited on Si substrates by a sol–gel process

Transition metal (TM)-doped TiO2 films (TM = Co, Ni, and Fe) were deposited on Si(100) substrates by a sol–gel method. With the same dopant content, Co dopants catalyze the anatase-to-rutile transformation (ART) more obviously than Ni and Fe doping. This is attributed to the different strain energy induced by the different dopants. The optical properties of TM-doped TiO2 films were studied with spectroscopic ellipsometry data. With increasing dopant content, the optical band gap (EOBG) shifts to lower energy. With the same dopant content, the EOBG of Co-doped TiO2 film is the smallest and that of Fe-doped TiO2 film is the largest. The results are related to electric disorder due to the ART. Ferromagnetic behaviors were clearly observed for TM-doped TiO2 films except the undoped TiO2 film which is weakly magnetic. Additionally, it is found that the magnetizations of the TM-doped TiO2 films decrease with increasing dopant content.


Background
Magnetic-ion-doped TiO 2 with room-temperature ferromagnetism is one kind of promising diluted magnetic semiconductors (DMS). It has been widely studied due to its potential applications in spintronics [1][2][3]. Many efforts have been made to understand the mechanism of ferromagnetism (FM) in magnetic-ion-doped TiO 2 . The most important point for industrial applications is if such room-temperature FM could originate from the doped matrices and not from the dopant clusters. Some theory models, such as the Ruderman-Kittel-Kasuya-Yosida exchange [4], super exchange [5], double exchange [6], magnetic polarons [7], and F-center exchange mechanism [8], have been used to explain ferromagnetism in transitionmetal-element-doped TiO 2 . However, many controversies still exist in the magnetic origin of DMS. Recently, roomtemperature FM [9] and reversible FM [10] in undoped TiO 2 films, and reversible FM in transition metal-doped TiO 2 nanocrystals [11], have been reported. These reports suggest that the structural defects can induce FM order, which brings new challenges in elucidating the magnetic mechanism in this kind of DMS.
In recent years, mixed crystal TiO 2 containing anatase and rutile phases has been more attractive because the anatase-rutile-phase junction (ARJ, related to phase composition) in the mixed crystal TiO 2 improves the spatial charge separation and enhances the photocatalytic activity [12][13][14][15][16]. Disorders in the mixed crystal TiO 2 affect the optical properties of TiO 2 [17,18]. The existence of the ARJs could enhance the disorders in the TiO 2 films, which will change the samples' physical properties. Our recent work indicates that both doping and phase composition affect the optical properties of TiO 2 films [19]. The ARJs could affect not only the optical but also the magnetic properties of the TiO 2 films [20]. However, to the best of our knowledge, the effects of phase composition on the magnetic properties of doped TiO 2 films have seldom been reported. Recently, Bahadur et al. found that the magnetic moment of the Ni-doped mixed crystalline TiO 2 powders increases and then decreases with increasing Ni content due to the change of spin ordering [21]. However, the influence of phase composition on the magnetic properties has not been taken into account in their studies.
In this paper, transition metal (TM)-doped TiO 2 films (TM = Co, Ni, and Fe) were deposited on Si(100) substrates by a sol-gel method. The influence of Co, Ni, and Fe doping on the crystalline structure of the TiO 2 films was compared. The magnetic and optical properties of the TM-doped TiO 2 films were investigated. The correlation between phase composition and magnetic and optical properties was studied, and the possible mechanism was discussed. These results will be useful for understanding the magnetic origin of oxide DMS.

Methods
Synthesis of TM-doped TiO 2 films, Ti 1 − x TM x O 2 (TM = Co, Ni, and Fe; x = 0, 0.01, and 0.03), was achieved on Si(100) substrates by sol-gel method. The precursor solutions of the TM-doped TiO 2 films were obtained from tetrabutyl titanate, cobaltous acetate, nickel acetate, and ferric nitrate with ethanol and acetylacetone as the solvent and the chemical modifier, respectively. The details of the preparation procedure are reported elsewhere [22]. For example, to prepare a Ni-doped TiO 2 solution, analytically pure nickel acetate (Ni[CH 3 COO] 2 ) and titanium butoxide (Ti[O(CH 2 ) 3 CH 3 ] 4 ) were used as the starting materials. Ni doping was achieved by dissolving nickel acetate in a solution with an appropriate volume ratio of ethanol (CH 3 CH 2 OH)/acetic acid (CH 3 COOH) at 60°C. Titanium butoxide and an equal amount of acetylacetone (CH 3 COCH 2 COCH 3 ) were dissolved in ethanol at 30°C. Then the two solutions were mixed slowly together at room temperature. In order to get a homogenous precursor, the mixture was stirred drastically in the magnetic stirrer for 2 h at 50°C. Finally, the 0.3 mol/L precursor solution was acquired and became transparent without precipitation even after 4 months.
The silicon substrates were cleaned in an ultrasonic bath for 20 min using acetone (CH 3 COCH 3 ), ethanol, and deionized water, respectively. The Ti 1 − x TM x O 2 thin films were deposited by spin coating the precursors on the silicon substrates at a speed of 3,500 rpm for 20 s. The films were thermally treated in a rapid thermal processor in air. Each layer of the films was initially dried at 200°C at a ramp rate of 15°C/s to evaporate the solvent and then rapidly heated to 380°C at a ramp rate of 20°C/s to remove the residual organics. Finally, the films were annealed at 700°C at a ramp rate of 20°C/s and naturally cooled down to room temperature. The each of the three steps of the rapid thermal treatment was held for 180 s. The spin coating and thermal treatments were repeated six times to prepare the samples.
The valences of the doping ions were determined by x-ray photoelectron spectroscopy (XPS, PHI 550 ESCA/ SAM; PerkinElmer Inc., Waltham, MA, USA) with a monochromatized AlKα radiation source (hυ = 1,486.6 eV) operated at 10 kV and 30 mA. The electron energy analyzer was operated at the constant pass energy of 50 eV. The structures of the samples were characterized by x-ray diffraction (XRD; D/max2200VPC, Rigaku Co., Shibuya-Ku, Tokyo, Japan) using CuK α radiation (λ = 0.15471 nm) with a resolution of 0.04°and the 2θ range from 10°to 65°. The ellipsometric measurements were carried out by a nearinfrared to ultraviolet (NIR-UV) spectroscopy ellipsometry (SE) in the wavelength range of 300 to 826 nm (1.5 to 4.1 eV) with a spectral resolution of 2 nm (SC630UVN; Shanghai Sanco Instrument, Co., Ltd., Xuhui, Shanghai, China). The incident angle for films was 70°corresponding to the experimental optimization near the Brewster angle of the Si(100) substrates. Magnetic measurements were performed at 300 K using a vibrating sample magnetometer (PPMS-9 Quantum Design, San Diego, CA, USA), and the measured sample size is about 2 mm × 10 mm. All measurements were performed at room temperature.

Results and discussion
XPS of the TM-doped TiO 2 films Figure 1 shows the XPS survey spectra of the TM-doped TiO 2 thin films. The carbon peak comes from surface contamination because of exposure to air [23]. All the peaks are calibrated with the carbon 1 s peak at 284.6 eV. The survey indicates that titanium, oxygen, iron, cobalt, and nickel are the major components on the surface of these films. Figure 2 shows a high-resolution XPS spectrum of the Ti 2p region for Ni-doped TiO 2 thin films, respectively. The core level binding energy of Ti 2p 3/2 is 458.4 eV and that of Ti 2p 1/2 is 464.16 eV. The difference of 5.7 eV in the two peaks indicates a valence state of +4 for Ti in the TiO 2and Ni-doped TiO 2 samples [24,25]. The same analysis also shows a valence state of +4 for Ti in the Fe-and Co-doped TiO 2 samples (not shown). Figure 3 depicts the TM 2p core level XPS spectra for TM-doped TiO 2 thin films. A Gaussian (80%) + Lorentzian (20%) fit was carried out and showed that the binding energy of Ni 2p 1/2 is 873.1 eV; the binding energy of Ni 2p 3/2 is 855.4 eV. This is different from those of metal Ni 0 (852.6 eV) and Ni 3+ (856.1 eV) [25,26] and very near to that of Ni 2+ (855 eV) [21,25,27]. This indicates that the chemical valence of Ni in the films is +2. Furthermore, the difference of 17.7 eV between Ni 2p 3/2 and Ni 2p 1/2 peaks also indicates a valence state of +2 for Ni in the Ni-doped TiO 2 films [25]. The same analysis also shows a valence state of +2 for Co in Co-doped TiO 2 and a valence state of +3 for Fe in Fe-doped TiO 2 (in Figure 3).  Further, TM doping may also result in oxygen vacancy due to the replacement of Ti 4+ by TM ions to maintain crystal charge neutrality, and the vacancy content may increase with increasing dopant content. As an example, the O 1 s peaks for TiO 2 , Ti 0.90 Co 0.01 O 2 , and Ti 0.97 Co 0.03 O 2 thin films are shown in Figure 4a. Both the O 1 s core levels display an asymmetric shape and are located at about 530.4 eV. The O 1 s peak was fitted by the two-peak Gaussian curves. The two fitting peaks are defined as OI and OII, respectively (Figure 4b,c,d). The OI peak is due to the oxygen atoms of TiO 2 [24,28], and the OII peak is attributed to the oxygen vacancies [24,26,29]. The OII peak appears as a function of oxygen vacancies. The increase in the area ratio of OII peak to OI peak indicates the enhancement of oxygen vacancy content [24,29,30]. The area ratio is 0.18, 0.28, and 0.32 for TiO 2 , Ti 0.90 Co 0.01 O 2 , and Ti 0.97 Co 0.03 O 2 films, respectively. These results indicate that the oxygen vacancies increase with increasing Co content. The same analysis also suggests that oxygen vacancies increase with increasing dopant content for Fe-and Ni-doped TiO 2 samples (not shown).

XRD of the TM-doped TiO 2 films
The XRD patterns of the TM-doped TiO 2 films on silicon substrates are shown in Figure 5. All the films are mixed crystal with diffraction peaks of A(101) and R (110), respectively [20,21]. Except the diffraction peaks of the anatase and rutile phase, no impurity phase is observed, which indicates that the TM atoms have been successfully incorporated into the TiO 2 matrix. The change in the rutile and anatase lattice constant was shown to follow Vegard's law (Figure 6a,b respectively), in which a linear relation exists between the crystal lattice constant of a material and the concentrations of the constituent elements at constant temperature [31]. Of course, TM ions may also be at the interstitial site, but the matrix compound, TiO 2 , has a relatively closepacked structure, and this is not generally favorable for interstitial defects [32]. Moreover, the interstitial defect in this case is highly charged, which is another detrimental factor [32].
With increasing dopant content, rutile-related peaks gradually increased. For the Co-and Ni-doped TiO 2 films, when dopant content reaches 0.03, the diffraction patterns of the rutile phase become predominant. On the contrary, for the Fe-doped TiO 2 films, the diffraction patterns of the anatase phase are still dominant. These results indicate that the addition of dopant catalyzes the anatase-to-rutile transformation (ART), which are similar to those of the Co-doped [23,33], Ni-doped [34,35], and Fe-doped [36][37][38][39] TiO 2 powders.
The fraction of rutile phase in these films can be estimated from the XRD peak intensities by the following equation: X R = 1/[1 + 0.884(I A /I R )], where X R is the weight fraction of rutile phase in the samples, and I A and I R are the x-ray-integrated intensities of the A(101) and R(110) peaks, respectively [20]. The rutile fraction against dopant content of the TM-doped TiO 2 films is presented in Figure 6c. It can be seen that the contents of the rutile phase enhance with increasing dopant content. The influence of the Co and Ni dopants on the ART of the TiO 2 films is conspicuous, but minimal for the Fe dopant. At the same dopant content, the rutile content of the Co-doped TiO 2 films is the highest, and that of Fe-doped TiO 2 films is the lowest.  The ART is a nucleation and growth process at the expense of consuming the surrounding anatase in undoped TiO 2 [23,33]. The nuclei were formed at the anatase {112} twin boundaries. Half of the titanium cations in the twin slab displace and the rutile phase nucleates [40,41]. The transformation of bulk anatase ruptures 7 out of the 24 Ti-O bonds per unit cell and leads to the cooperative displacement of both Ti and O. After Ti 4+ is replaced by Co 2+ , Ni 2+ , and Fe 3+ ions, oxygen vacancies are introduced to keep the crystal charge neutrality. During the course of the ART, the presence of oxygen vacancies makes the number of Ti-O bond rupture less than 7/24 per anatase unit cell. In other words, oxygen vacancies make the ART [24].
The replacement of Ti 4+ by Fe 3+ leads to the following equation [32]: Similarly, the replacement of Ti 4+ by Co 2+ (Ni 2+ ) leads to the following equation: CoO Therefore, at the same dopant content, the oxygen vacancies due to Co 2+ or Ni 2+ doping are theoretically more than those of Fe 3+ doping. Thus, the rutile content of Co-or Ni-doped TiO 2 films is more than that of the Fe-doped TiO 2 films. In addition, the ionic radius of Co 2+ , Ni 2+ , Fe 3+ , and Ti 4+ are 0.72, 0.69, 0.64, and 0.605 Å, respectively. When the Ti 4+ ions are substituted by TM n+ (Co 2+ , Ni 2+ , and Fe 3+ ) ions, the difference in ionic radii between Ti 4+ and TM n+ results in the lattice deformation of anatase TiO 2 , and the strain energy due to the lattice deformation facilitates the ART [33]. Furthermore, the strain energy supplied by Co 2+ doping is bigger than that of Ni 2+ doping because the ionic radii of Co 2+ is larger than that of Ni 2+ . Thus, the rutile content of Co-doped TiO 2 films is more than that of Ni-doped TiO 2 films.

Ellipsometric spectra of the TM-doped TiO 2 films
With increasing dopant content, the optical properties of the doped TiO 2 films will change due to the increasing rutile content. SE is an appropriate tool to calculate optical constants/dielectric functions and the thickness of films because of its sensitivity and nondestructivity. The SE parameters Ψ(E) and Δ(E) are the functions of the incident angle, optical constants, and the film thickness. In our previous studies, the optical constants of some materials have been successfully obtained using the SE technique [42,43]. To estimate the optical constants/dielectric functions of TM-doped TiO 2 films, a four-phase layered system (air/surface rough layer/film/ substrate, all assumed to be optically isotropic) [43] was utilized to study the SE spectra. A Bruggeman effective medium approximation is used to calculate the effective dielectric function of the rough layer that is assumed to consist of 50% TiO 2 and 50% voids of refractive index unity [43]. Considering the contribution of the M 0 -type critical point with the lowest three dimensions, its dielectric function can be calculated by Adachi's model: where, E is the incident photon energy, ε ∞ is the high-frequency dielectric constant, χ 0 = (E + iΓ), E OBG is the optical gap energy, and A 0 and Γ are the strength and broadening parameters of the E OBG transition, respectively [42,44]. Figure 7 shows the measured SE parameters Ψ(E) and Δ(E) spectra at the incident angle of 70°for the TMdoped TiO 2 films on Si substrates. The Fabry-Pérot interference oscillations due to multiple reflections within the film have been found in from 1.5 to 3.5 eV (354 to 826 nm) [42,43]. Note that the interference oscillation period is similar across the film samples, except for the undoped TiO 2 that has the maximum thickness. The revised Levenberg-Marquardt algorithm in the nonlinear least squares curve fitting can extract the best-fit parameter values in the Adachi's model for all samples. The simulated data are also shown in Figure 7. The good agreement between experimental and fitting curves suggests that the optical constants attained by the simulation are reliable.
The fitted parameters of the TM-doped TiO 2 films determined by the SE spectra are listed in Table 1. From the table, the film thickness of undoped TiO 2 film is the largest and that of Co-doped TiO 2 films is the smallest. Compared with the undoped TiO 2 film, the addition of dopant decreases A 0 and increases Γ, which suggests that the Urbach tail absorption characteristics were formed. Note that it is common to observe the development of an Urbach tail on doping transition metal oxides [45,46]. Figure 8 depicts the variation in dielectric function of the TM-doped TiO 2 films with photon energy. In general, in all samples, we found that the real part ε r of the dielectric function increases and gradually nears the maximum, and then decreases due to the Van Hove singularities. This is the typical optical response of dielectric or semiconductor materials [44]. The imaginary part ε i of the dielectric function nears zero in the transparent region (E OBG > E) and sharply increases further with increasing photon energy in the absorption region (E OBG < E).
The dopant content dependence of the E OBG of the TM-doped TiO 2 films is presented in Figure 6c. It is can be seen that the E OBG of the TM-doped TiO 2 films decreases with increasing dopant content. Note that at the same dopant content, the E OBG value of the Co-doped TiO 2 films is the smallest and that of the Fe-doped TiO 2 films is the largest. By comparing the E OBG value with the rutile content of the TM-doped TiO 2 films (in Figure 6c), it is found that the change of E OBG value is related to the fraction of rutile phase, except the doping. This suggests that E OBG can be tuned not only by dopant type but also by dopant content. The undoped TiO 2 film has little rutile phase detected by XRD, and the E OBG value is about 3.58 ± 0.01 eV. For the x = 0.01 TM-doped TiO 2 films, the rutile phase is minimal, and the E OBG value is about 3.56 ± 0.02, 3.53 ± 0.01, and 3.48 ± 0.02 eV for Fe, Ni, and Co-doped TiO 2 films, respectively. However, when dopant content reaches 0.03, the rutile phase is prominent for Co-and Ni-doped TiO 2 films, and the E OBG value is about 3.43 ± 0.01 and 3.50 ± 0.01 eV, respectively. For Fe-doped TiO 2 film, the anatase phase is still prominent, and the E OBG value is 3.54 ± 0.02 eV. These values of E OBG for all samples are larger than those in the literatures [17,18,47] but near the reported values of rutile TiO 2 films [44].
As shown in Figures 5 and 6c, the results indicate that the undoped TiO 2 film is mainly composed of anatase phase and a minor rutile phase. Thus, the ARJs between the anatase and rutile phases are embedded within the anatase phase [15]. The electronic mobility from anatase-to-rutile phases is affected by the junctions. To some extent, the ARJ structure is electronically disordered. In addition, oxygen vacancies increase with increasing dopant content, which also results in the electronic disorder in the samples. Therefore, the increase of the disorder leads E OBG to shift to lower energy [17,18,47]. With the same dopant content, the disorder in the Co-doped TiO 2 films is the strongest and the E OBG value is the smallest.

Magnetic properties of the TM-doped TiO 2 films
Magnetization (M) versus magnetic field (H) curves of TM-doped TiO 2 films are displayed in Figure 9. The ferromagnetic hysteresis curves are clearly found for all samples, which indicate that the undoped and doped TiO 2 films exhibit ferromagnetic behavior. The results are similar to those of the literature [21,[48][49][50][51]. In addition, the M values of x = 0.01 Fe-, Ni-, and Codoped TiO 2 films at 10 4 Oe were the largest and about 419.7, 386.5, and 445.6 emu/cm 3 , respectively. The M values of doped samples decrease with increasing metal element contents, which is similar to the Ni-doped TiO 2 powders [21] and Fe-doped TiO 2 films [52]. Generally, the magnetization of samples should increase with increasing magnetic ions, but the magnetic data of these samples do not support it. These magnetic phenomena are extraordinary and different from the magnetic results of the literature [7][8][9][10][11]21], which suggest that there are complex magnetisms in these samples.
The ferromagnetic properties of the anatase or rutile phase TM-doped TiO 2 system have been widely investigated; however, there are few reports of magnetic change due to the phase fraction change. Recently, Bahadur et al. found that the magnetic moment of Ni-doped mixed crystalline TiO 2 powders increases and then decreases with increasing Ni content [21]. They suggested that the observed ferromagnetic states may originate from the spin ordering through exchange interactions between the holes trapped in the oxygen 2p orbital adjacent the Ni site, which substitutes Ti sites. However, in their reports, rutile content decreases with increasing Ni content, indicating that their theory may not fit for our samples because the rutile content of the present doped TiO 2 films increases. Additionally, Jiang et al. suggested that the decrease in the saturation magnetization may be related to the antiferromagnetic contribution with increasing dopant content in the Fe-doped TiO 2 films [52]. Although their samples are mixed crystalline, the authors had not taken the ARJs into account. The film thickness (df), the thicknesses of the surface rough layer (ds), and the parameter value of Adachi's model (A 0 ) for TM-doped TiO 2 films with dopant content extracted from the simulation of SE in Figure 7. The 90% reliability of the fitted parameters is shown with ± sign. The TM atom composition C TM derived by the XPS spectra is also listed. It is known that TiO 2 shows a strong polaronic effect in which the carrier effective mass becomes bigger due to strong electron-phonon interactions [53,54]. A polaronic electron will spend most of its time near an oxygen vacancy when it is trapped in the vacancy. Then the trapped electron can form an F-center. In the center, the trapped electron occupying an orbital effectively overlaps the d shells of the surrounding magnetic ions. Therefore, a possible origin of ferromagnetism is an F-center-bound magnetic polaron, which is formed by an electron trapped in an oxygen vacancy and its neighboring magnetic impurity ions [8,51]. In other words, the room-temperature ferromagnetism of TM-doped TiO 2 films was induced mainly by the magnetic polarons formed by the localized electrons surrounded by magnetic impurities.
There are oxygen vacancies in our samples and the vacancies promote the ART. Thus, the magnetic properties of the samples may be related to the influence of the ART on the magnetic polarons. According to XRD analysis, the ART easily occurs in anatase TiO 2 lattice with oxygen vacancies. The ARJs emerging during the course of ART will reduce the number of the trapped electrons. That is to say, these ARJs may destroy the magnetic polarons in anatase/rutile TiO 2 , which results in the decrease in magnetization. Of course, the magnetic mechanism of mixed crystal TM-doped TiO 2 is an open issue and needs further study in depth.

Conclusions
The TM-doped TiO 2 films (TM = Co, Ni, and Fe) have been deposited on Si substrates by a sol-gel route. The additives promote the ART of the TiO 2 films. The influence of Co, Ni, and Fe on the ART was compared. With the same dopant content, Co doping catalyzing the ART is more obvious than those of Ni doping and Fe doping, which is attributed to the different strain energy induced by oxygen vacancies and the difference in valence and ionic radii of Co 2+ , Ni 2+ , and Fe 3+ . The decreases of the E OBG are related to the enhancement of disorders induced by the ARJs in the samples. The undoped TiO 2 film exhibited weak magnetic properties, while ferromagnetic behaviors were clearly observed for TM-doped TiO 2 films. The magnetizations of the TM-doped TiO 2 films decrease with increasing dopant content, which may be related to magnetic polarons in the samples. The final explanation on their magnetic properties still remains a puzzle, and the true mechanism deserves further study.
of the study and participated in its design and coordination. All authors read and approved the final manuscript.