Comprehensive investigation into the influence of oxygen vacancies on the ferroelectric properties of spin coated bismuth ferrite thin films

Bismuth ferrite (BFO) is a prime candidate for room-temperature magnetoelectric coupling and multiferroic applications. The rhombohedral R3c phase of BFO is the source of many properties, but the phase purity and oxygen vacancies are still the biggest obstacles to its real-world application. Considering these facts, the present work investigates the effects of oxygen vacancies on the functional properties through manipulation of drying temperatures of spin-cast films, especially at temperatures around 280 °C, where both the secondary phase and oxygen vacancies are prevalent. One of the biggest sources of oxygen vacancy is bismuth volatilisation, and our work deals with the situation head-on, uncovering the effect of bismuth volatilisation on functional properties. The structural properties were studied using x-ray diffraction (XRD), and deeper insights into the surface topography of the samples were obtained using AFM imaging. The electrical and dielectric characteristics help distinguish and analyse the samples in terms of the presence of resistive switching. PUND studies were performed to determine the ferroelectric properties of the samples. A fifty percent reduction in the oxygen vacancies in the presence of secondary phases was observed when compared with the phase-pure sample, as shown by the XPS analysis. Deeper insights were provided into the valence band spectra by first-principles studies. This work shows that phase purity may not be the singular condition for enhancing functional properties, and fine-tuning the presence of secondary phases and oxygen vacancies may be the way forward. The ferroelectric polarisation in one of the samples exhibits a notably higher value when using chemical solution deposition methods, making it a promising candidate for memory devices.


Introduction
Bismuth ferrite (BFO) has been the subject of extensive research for many decades because of its multiferroic properties at room temperature [1].The presence of G-type antiferromagnetism and its ferroelectric nature are highly dependent on its structure.It is well known that BFO exhibits multiple phases during its formation; hence, various methods have been employed to obtain phase-pure bismuth ferrite.Research on bismuth ferrite is ongoing because of its secondary phases, other than the parent rhombohedral phase, which exhibits functional properties.For example, orthorhombic Bi 2 Fe 4 O 9 exhibits photocatalytic activity [2], and the tetragonal phase exhibits enhanced ferroelectric properties [3].
Chemical solution deposition methods have been well explored because of their simplicity and relative ease with which various parameters can be analysed.Control over stoichiometry is essential, particularly with hightemperature heat treatment.One of the groundbreaking discoveries of bismuth ferrite is its large saturation polarisation [4].Research has been conducted in various directions, including both pristine and doped BFO.One of the important factors that have been of interest, even at present, is the thickness of the film, which controls its multiple functional properties.For example, thickness affects photocatalytic behaviour [5], photovoltaics [6], magnetic properties [7], and ferroelectric properties [8].Another critical factor is the strain on the films, which affects various properties and can be controlled by varying the thickness of the film.Reports have claimed a lack of correlation between the thickness and (and oxygen vacancy) ferroelectric properties [9]; however, this was later verified experimentally and found to be incorrect.Pristine BFO shows potential for various applications, but requires enhancement and fine-tuning for practical applications.Apart from the obvious methods of doping and complex formation, tailoring the preparation method is vital for optimising the properties of the pristine BFO.There are multiple reports on the effects of the annealing temperature [10,11], duration [12], and even the annealing environment [13].However, there is a clear void in understanding the role of oxygen vacancies and secondary phases on the film properties, especially in the vicinity of the temperatures where the volatility of bismuth is maximum (near 280 °C).Therefore, the present work aims to systematically analyse the variation in the chemical, structural, and electrical properties of the film in temperature ranges that can aid the growth of oxygen vacancies and secondary phases (namely, at drying temperatures of 200 °C, 250 °C, 300 °C, and 350 °C).The objective of this work is to address the impact of bismuth volatisation directly and to explore potential methods for fine-tuning the concentration of oxygen vacancies in a more precise manner.
First, we explored the effect of the thickness on the structure and morphology of the film.Using the optimised preparation conditions, we investigated the impact of the drying temperature.Structural and x-ray photoelectron spectroscopy (XPS) analyses were conducted to validate the presence of the secondary phases and oxygen vacancies.The presence of resistive switching was explored in the samples through I-V characterisation, and the dielectric behaviour was studied through capacitance versus frequency characterisation.Another important property is the roughness of the film, particularly for applications such as gas sensing, which is measured using atomic force microscopy.Our work also utilised first-principles studies by calculating the projected density of states to explain the valence-band spectra obtained through XPS studies.Raman modes were also determined.Ferroelectric behaviour was examined through PUND measurements.The significance of our work lies in the simple and efficient methodology used for exploring and controlling the properties of pristine bismuth ferrite through systematically understanding the effects of bismuth volatisation and hence this can be critical for applications in photocatalytic activity, photovoltaic activity, and water splitting [14][15][16], apart from having applications as phase shifters, microwave absorbers, and tunable filters beyond simple memory devices.

Preparation of BFO thin films
The films were fabricated using the spin-coating method.Iron nitrate and bismuth nitrate acquired from Sigma-Aldrich were used along with 2-methoxyethanol (2-MOE) and glacial acetic acid.Suitable amounts of iron and bismuth nitrates were mixed with 2-MOE, and glacial acetic acid was used as the chelating agent.The preparation conditions used for different samples are listed in table 1.
All the samples were annealed at 550 °C for three hours to improve their crystallinity.The structural properties were investigated using x-ray diffraction (XRD).The morphology and topography of the films were examined using scanning electron microscopy and atomic force microscopy.The topography was analysed using WSxM 5.0 [17].The oxidation states of the samples were determined using x-ray photoelectron spectroscopy.The resistive switching properties of the samples were explored through I-V studies.Samples S5 and S6 were further characterised over 60 I-V sweeps at 5 V to check that the resistive switching behaviour decayed over a number of cycles (fatigue study).The dielectric properties were explored using capacitance versus frequency studies.The first-principles investigation of the projected density of states was conducted using the quantum espresso software suite [18,19] with Hubbard correction [20] on a supercell containing 80 atoms.Further information on the calculations can be found in our previous work [21] (see the supplementary information file for the calculated PDOS).

Device preparation
Films prepared on commercially obtained fluorine-doped tin oxide (FTO) coated on a glass substrate were used to prepare samples for electrical characterisation.FTO was chosen as the bottom electrode for electrical characterisation because of its stability, even after annealing.Thermal evaporation was used to coat Ag as the top electrode using a dot mask.A schematic of the entire process is shown in figure 1.
PUND measurements were performed at 1 V with a pulse delay of 1 ms each.2. This confirms the presence of the rhombohedral R3c phase of bismuth ferrite.The doublet peaks between 31°-32°are fingerprints of the rhombohedral phase.The bismuth ferrite peaks are indexed using a reference pattern (ICDD-01-074-2493).The impurity phase present in sample S3 can be indexed as selenite, which is a bismuth-rich phase (COD-96-901-1269), whereas the impurity in sample S1 can be indexed as an iron-rich mullite phase.The selenite impurity phase has no ferroelectric properties owing to its centrosymmetric cubic structure.All the samples, irrespective of the presence of a secondary phase, showed doublet peaks that remained intact, confirming that the rhombohedral R3c phase was the majority phase.With an increase in the thickness, the strain on the film was reduced owing to the decrease in the lattice mismatch between the substrate and the film.Another trend is that the (011) peak at approximately 22°increased with increasing film thickness.The preferential orientation of thicker films tends to shift the peaks [22], as shown in figure 2(b).All the samples were fabricated with excess Bi to compensate for the loss during annealing.The optimal compensation can be seen in sample S2, whereas sample S1, which has a relatively lower thickness, lacks the appropriate amount of Bi to compensate for its loss during volatilisation, resulting in an iron-rich mullite phase.Sample S3, which was the thickest sample, showed the presence of excess Bi, resulting in a bismuth-rich selenite phase.

Results and discussion
From table 2, we can see a change in the crystallite size and microstrain with increasing thickness.The marginal reduction in the crystallite size of sample S3 could be due to restrained growth with an increase in thickness (however, it should be noted that this reduction falls within the error range).In sample S2, we observed a slight relaxation of the microstrain, while with a further increase in thickness, we observed that the microstrain increased.The lattice constant a has been found using where d is the interplanar spacing; h, k, and l are the Miller indices; and α is the rhombohedral angle.
The crystallite size t and the microstrain e has been found using where β is FWHM (rad).
Thicker films tend to produce more uniform and well-defined particles, and sample S2 shows better crystallinity with sharper peaks than S1.The crystallinity between samples S2 and S3 could not be compared, as the FWHM values fell within the error range of each other, but as a whole was more crystalline than that of sample S1.

Effect of heat treatment
Based on the results from the structural (and the following morphological section) section, the preparation conditions of sample S2 were considered for further investigation of the effect of drying temperature on the film properties.Drying temperatures of 250, 300, and 350 °C were considered, along with the previously prepared films at 200 °C.During the typical spin-coating method, when crystallisation occurs before the densification of  the films, the porosity of the film dramatically increases.By rapid heating, we can reduce crystal nucleation and hence achieve complete densification and reduce porosity [23][24][25].At temperatures near 280 °C, Bi-O readily breaks into Bi and O, owing to the volatility of Bi.This loss in Bi ion concentration can lead to a loss of stoichiometry; hence, an excess amount of Bi has been considered in the precursor, which is not uncommon [26].This variation in the Bi ion concentration can lead to the formation of impurity phases, which can also affect the oxidation state of iron, leading to impurity phases that are now caused by iron.The Effect of drying temperature on the structural properties of the films is reported in table 3. Phase-pure bismuth ferrite could be obtained at an annealing temperature of 550 °C; however, impurity phases were observed at drying temperatures close to the bismuth volatilisation temperature of approximately 280 °C (figure 3).Sample S4 showed the signature of iron oxide, which was dried just below the volatilisation temperature.Sample S5 exhibited both iron oxide and bismuth-rich Bi 2 Fe 4 O 9 phases, and its drying temperature was slightly higher than that of Bi.Even at higher drying temperatures, phase-pure bismuth ferrite with relatively better preferential orientation towards the (012) plane was obtained, as shown in figure 3. The fingerprint doublet peak of the rhombohedral R3c phase was significantly reduced.

Surface morphological and topography studies
Figure 4 shows FESEM images of spin coated bismuth ferrite films.The morphology appeared uniform and homogenous upon increasing the thickness from S1 to S2.No cracks were observed in any of the samples, whereas pores were observed in sample S1.The porous nature of bismuth ferrite films has been previously reported in the literature and is one of the causes of poor electrical and ferroelectric performances [3,27].Despite the phase purity, porosity can give rise to higher DC leakage currents; hence, measures such as a twostage spin process (slower spin rotation followed by higher spin rotation) need to be considered to reduce the porosity [25].
Cross-sectional images of the samples were obtained, and a cross-sectional image of sample S2 is shown in figure 5.The thicknesses of samples S2, S4, S5, and S6 were similar because the number of layers remained identical.
From the topography studies using AFM (figure 6), we can conclude that the surface roughness of the films decreased with increasing drying temperature (up to 300 °C), after which the roughness increased.The average roughness values of all samples are tabulated in table 4. We also deduced by comparing samples S1, S2, and S3  that after crossing a threshold thickness where the substrate can no longer affect the roughness, the roughness increases with an increase in the number of layers [28].
Skewness and kurtosis are critical parameters describing the surface and topography of a film, respectively.A skewness of 0 refers to a symmetrical height distribution, whereas a kurtosis of 3 refers to a moderate height distribution using a Gaussian distribution.A positive skewness refers to a surface with tall peaks and filled valleys, whereas a negative skewness indicates deeper grooves and a lack of peaks.A kurtosis larger than 3 indicates a surface that has high peaks and/or low valleys, and a kurtosis smaller than 3 signifies low peaks and/ or valleys [29].
The skewness of the samples varies from −0.28-0.5.With respect to skewness, we see that for samples S1, S2, and S3, S2 has tall peaks and filled valleys.With increasing drying temperature, we see that skewness decreases, and sample S5, which has a relatively high concentration of impurity phase, shows negative skewness, referring to deeper valleys.Kurtosis also decreased with the drying temperature, with sample S5 being an exception.From the results, we deduce that sample S5 may have tall peaks with deep valleys, as shown in figure 6, which may affect the conduction of the film.The variation in kurtosis, skewness and RMS roughness can be viewed in figure 7.
Another probable cause for smoother films may be the reduction in Marangoni forces with increasing drying temperatures (up to 300 °C).The main requirement is that a less volatile solution (2-MOE) has greater surface tension than a more volatile solvent (in this case, glacial acetic acid), which has a lower surface tension.As the drying temperature increased, this was because of the relatively longer presence of 2-MOE, which tended to become smoother during the drying process [30].In sample S5, the presence of an impurity phase might have  contributed to the greater smoothness than in samples S4 and S2.In sample S6, there was a shift in the predominant peak in the XRD pattern, which might have contributed to the greater roughness of the film.
The general behaviour of bismuth ferrite thin films includes a reduction in surface roughness with increased thickness, which shows better electrical properties (lower leakage current, higher dielectric constant, and lower tangent loss, considering ideal conditions with no impurities) [31].

X-ray photoelectron and Raman spectroscopy
X-ray photoelectron spectroscopy is an effective tool for determining the oxidation states of the elements in a sample.It is essential to quantify the presence of iron in BFO, which can be in the +2 or +3 state, thereby confirming the presence of oxygen vacancies.The role of oxygen vacancies in resistive switching is discussed in a  later section.It also provides vital data for bismuth, which can be present in its zero oxidation state (Bi-metal state), as bismuth is added in excess to compensate for its loss during annealing.The area ratios calculated from XPS studies are listed in table 5.The spin-orbit coupling energy of bismuth and iron was approximately 5.31 eV and 13.6 eV, which matches well with standard values [32].Doublet Bi 4f 7/2 and Bi 4f 5/2 peaks appear at 158.6 and 163.9 eV, respectively.This result was in good agreement with those reported in the literature [33,34].The splitting of the Bi 4 f peak into a doublet is due to spin-orbit coupling between the electron spin and angular momentum spin, which can be parallel or antiparallel.The anti-parallel spin alignment is more favourable for stability, and hence, it is justified that Bi 4f 5/2 (l = 3, s = −1/2) appears at a higher binding energy.The area ratio between the doublets is a function of its degeneracy g, which is given by g = (2j+1), where j is the total angular momentum (figure 8).
The O 1 s peak (figure 10) appears at approximately 531 eV; the s orbital is not degenerate and is therefore not expected to split.The subpeak of the 1 s peak (at approximately 531 eV) belongs to the oxygen vacancy and is, hence, likely to be minimal or even absent, if possible.The presence of oxygen vacancies indicates the existence of dual oxidation states of iron.We observed that the presence of oxygen vacancies is indeed high, to about 70%, but this is a common feature among bismuth ferrite films, as oxygen vacancies as high as 83% have been reported in [35].
The Fe 2p peaks split into doublets with the Fe 2p 3/2 peak appearing at a lower binding energy.The Fe 3+ state is the state of iron in bismuth ferrite.The loss of oxygen during film formation results in oxygen vacancies, where Fe 3+ -O-Fe 2+ hybridised states are formed to negate charge imbalance, and magnetic double-exchange interactions appear [36,37] (figure 9).
The core spectra of Fe in the samples S2, S4, S5, and S6 are shown in figure 9.The presence of satellite peaks confirms the presence of Fe 3+ [38][39][40], which is typically observed when a 2p photoelectron excites an electron at the 3d-4 s level [41].Fe 2+ is observed in the Fe 2p 3/2 and Fe 2p 1/2 peaks.The total area ratio of Fe 3+ to Fe 2+ is greater than 1.4.This relatively high ratio can aid in Fe-O-Fe electron hopping [42], which may explain the higher magnitude of the current in the I-V measurement.From the area ratios, we realise that sample S2 shows relatively higher oxygen vacancy (OV) values than when other impurity phases are present.Sample S5 contained iron oxide, which could have led to the relatively lower OV values.
The significance of the qualitative measurement of oxygen vacancies comes from the fact that bismuth ferrite is capable of showing approximately four phases experimentally (at least seven, as seen in computational studies [43]).In the present study, all phases other than the rhombohedral phase are considered as secondary phases (usually, bismuth ferrite crystallises in the rhombohedral phase at room temperature).Oxygen vacancies lead to the formation of Fe 2+ ions, which is an obstacle to the practical implementation of applications.Generally, in  secondary phases, the FeO 6 octahedra are tilted, which changes the Fe-O-Fe bond angle, thereby controlling the charge transfer mechanism between O and Fe and reducing the formation of Fe 2+ (and consequently the oxygen vacancies) [44].For instance, in Pr-doped BFO [45] (orthorhombic distortion), the ferroelectric properties have been reported to improve for the same reason mentioned above.Similarly, La doping into BFO [46] transforms the structure into tetragonal (and orthorhombic at higher concentrations), which greatly helps reduce the leakage current and improve the ferroelectric properties.
Besides doping, a strain-driven morphotropic boundary between the rhombohedral and tetragonal phases, is also known to improve ferroelectric and piezoelectric properties [47].Likewise, the orthorhombicrhombohedral morphotropic phase boundary also improves ferroelectric properties [48].
The authors in their previous work, using first principles studies, have shown that lesser hybridization of Fe and O might be one of the reasons for better electrical properties [21] in secondary phases.
Raman spectroscopy is an essential investigation that sheds light on the type of bonding present and helps detect phases that go undetected in XRD characterisation.The rhombohedral R3c phase has 13 Raman modes (4 A and 9 E modes) figure 11.It is well known that lower-frequency A-mode vibrations up to 167 cm −1 correspond to Bi-O covalent bonds.These bonds are also responsible for the dielectric and ferroelectric behaviours [3].
The A 1 −1 (LO) mode is responsible for electric and magnetic ordering, which is easily affected by induced stress in the film and other changes in the Bi-O octahedron [49].The high-frequency E modes are related to the Fe-O bond and are significantly affected by oxygen vacancies, especially the modes near 470 cm −1 .This E 7 Raman mode corresponds to the motion of oxygen planes with a large in-plane amplitude and is also a mixed (LO and TO) mode [50].This mode also corresponds to relative disorder in the sample.In sample S2, which is the phase-pure sample, we measured and located the A 1 modes at (140, 169.9, and 217.7) cm −1 , while the E modes were found at (127.6, 261.8, 284.2, 348.9, 379.2, 469.5, 433.2, 552.1, and 607.7) cm −1 .These modes are in excellent agreement with [51,52,53].From the other samples, we observe that low-intensity modes such as E 4 and E 6 disappear relatively quickly.The A 1 −1 and E7 modes are effective indicators of the ferroelectric property because they signify the 6s 2 lone pair and disorder in the system, respectively.Samples S2 and S6 showed the most intense A 1 −1 mode, whereas S2 also showed an intense E 7 mode.

Electrical characterisation 3.4.1. I-V studies
Phase-pure bismuth ferrite exhibited resistive switching during I-V characterisation, which is characteristic of ferroelectric materials (figure 12).The high and low resistances can also be explained by changes in the barrier height and depletion region, as shown in figure 13.When the Au/BFO interface is subjected to a negative voltage, polarisation is generated in the downward direction, leading to repulsion of the electrons near the FTO interface regime.This also attracts holes from the BFO switching layer to the FTO layer.This leads to an increase in the depletion layer (at the BFO/FTO interface), and the major component of the current originates from the holes; hence, this state is supposedly a high-resistance state.
When positive polarisation was applied to the Au/BFO interface, charges accumulated near the BFO/FTO interface, and electrons were dragged from the FTO layers.The holes are repelled and the depletion region decreases, leading to a low-resistance state [54][55][56].Samples S2, S4, and S6 showed resistive switching, as shown in figure 14, whereas sample S5, which contained many impurity phases, exhibited an almost ohmic behaviour.The asymmetry in the behaviour under different biases is due to the variation in the work functions of Ag and FTO, which are the top and bottom electrodes, respectively.

Resistive switching
Resistive switching (RS) behaviour has applications in the field of resistive random access memory (RRAM) with the advantages of a simple device structure, possibility of miniaturisation, low power requirement, and relatively high endurance.There are two modes of resistive switching: unipolar and bipolar.Unipolar switching occurs when the magnitude of the applied field defines the switching between the high-resistance state (HRS) and the low-resistance state (LRS).In contrast, bipolar switching occurs when the polarity of the field determines the switching between the LRS and the HRS.Unipolar switching encompasses the mechanism of formation of conductive filaments (pathways), which leads to the SET phase (HRS to LRS), and the subsequent rupture of the filament, which leads to the RESET phase (LRS to HRS).In contrast, bipolar switching involves migration and/ or redox reactions of the electrode at the active-layer interface.Another critical parameter is the type of electrode used.They can be active electrodes (electrodes that take part in migration and/or redox reactions of electrode ions near the electrode-active layer interface, for example, Ag and Cu) or inert electrodes (which do not take part in any RS mechanism, for example, Au and Pt).
Unipolar switching has the advantage of utilising different top electrodes in the same device for SET and RESET, confirming that the SET process occurs locally but indicates that the conductivity inside the memory cell is inhomogeneous (Au/Co-doped BTO/Pt).In bismuth ferrite, the migration of oxygen vacancies to and away from the interface determines the RS mechanism.Oxygen vacancies distributed randomly in the films can tend to act like 'clusters' possessing a strong correlation among them.The energy required for the migration/ movement of these clusters was lower than the sum of the energies required to move the individual oxygen vacancies.The activation energies of the oxygen vacancies were reduced by cluster formation [57,58].
With a positive bias of a suitable magnitude and duration applied to the top electrode, electrons are extracted from the interface.This net positive charge (V o+ ) can lead to band bending, which reduces the barrier width (or lowers the barrier height), resulting in LRS.When a negative charge is applied, a large number of electrons accumulate near the interface, modifying the barrier height (and/or width) and resulting in an HRS [59,60].Figure 15 shows that the first cycle exhibited the best resistive switching properties, whereas the subsequent cycles exhibited relatively deteriorated properties.This could be mainly attributed to the incomplete migration of oxygen vacancies over different cycles, which depends on the magnitude and duration of the applied field.From cycle 1 (C-1) to cycle 26, we observed a degradation of resistive switching and hence, an increase in the conductance and loss of distinction between the LRS and HRS states.There is an improvement in the behaviour of C-40 and later a relative degradation at C-60.A similar behaviour was observed for sample S6.
Oxygen vacancies are a necessary evil, and a controlled and specific concentration of oxygen vacancies supports resistive switching.In Co-doped barium titanate, the voltage and resistance of the SET and RESET phases were observed after the samples were exposed to an oxygen environment at 800 °C [61].However, an effective solution to reduce the contribution of oxygen vacancies is to characterise the films at relatively lower temperatures, as seen for (Au/BiFeO 3 /SrRuO 3 ) devices [62].Oxygen vacancies also affect ferroelectric behaviour, where the films fail to show saturation polarisation in the applied field.The effect of migration of oxygen vacancies (or charge trapping or de-trapping) decreases with lowered temperature, after which polarisation switching dominates the resistive switching behaviour [63,64].At lower characterisation temperatures, better resistive switching behaviour was observed at higher sweep rates.However, with increasing    temperature, the variation due to the different sweep rates became almost non-existent.This indicates that slower mechanisms such as oxygen vacancies can be safely neglected in the presence of other resistive switching (RS) mechanisms.
In bismuth ferrite, polarisation reversal can aid the formation and rupture of conductive filaments at the charged domain walls [65,66].The changes in the charge carrier density with the applied voltage help to sustain the resistive switching properties.The polarisation of the domain walls tends to redistribute the carriers (with an applied field), leading to conductive pathways responsible for the resistive switching behaviour.Apart from polarisation reversal, the Valence Change Memory (VCM) mechanism can also be responsible for the resistive switching behaviour [67][68][69].The VCM depends on the motion of oxygen vacancies within the oxide layers and their exchange with the adjacent electrode (oxide layers).Similar to polarisation reversal, there can be VCMinterfaced RS and VCM filament RS.The VCM interface RS involves interaction between BFO and the electrode material.The changes in the charge distribution caused by the applied voltage resulted in the formation of conductive pathways through the migration of oxygen vacancies.The reversal of the applied voltage can lead to the redistribution of oxygen vacancies and, hence, the disappearance of the conductive pathways.In VCM filament RS, the conductive filament pathways are established through migration of oxygen vacancies serving as electron donors in BFO.Reversal of the applied voltage can trigger the recombination of oxygen vacancies, which can rupture and hence lead to the disappearance of conductive filaments.
The resistive switching mechanism may involve a complex mixture of multiple mechanisms and hence requires further extensive investigation into each mechanism.The presence of polarisation can be confirmed by the ferroelectric characterisation seen in later section.

Dielectric studies
The room-temperature dielectric constant and dielectric loss as functions of frequency were measured for S2.The response was measured from 100 Hz to 2 MHz, and a strong frequency dependence of the capacitance was observed (figure 16).A similar dielectric behaviour has also been observed in the literature.It has been observed in the literature that thin and inhomogeneous grain boundaries result in a high dielectric coefficient.Hence, there is an inverse relationship between the dielectric constant and thickness of the grain boundaries at lower frequencies.At higher frequencies, the material reaches an extremum where all the electric dipoles neither rotate nor align to a rapidly accelerating field; hence, saturation polarisation is lost, leading to a decrease in the dielectric constant.With an increase in the drying temperature, the films tended to become denser with a reduced thickness.Thus, increased density facilitated the formation of smoother films.The capacitance, and hence the dielectric constant, decreases when the porosity in the film decreases, which could be one of the reasons for the reduced capacitance over the drying temperature [28].
The behaviour of the dielectric constant (capacitance) was mainly due to the inability of the dipoles to follow variations in the electric field.The samples prepared through spin-coating depend on the phase purity, stoichiometry, drying temperature, annealing temperature, and device structure.The oxygen vacancies in the sample, which appear because of the dual valencies of iron can or even because of the volatisation of Bi, can significantly affect the grain boundary mobility, affecting grain growth [70].
Figure 17 shows the Nyquist plot, where the intercept of the semicircle on the real plot indicates the bulk resistance of the material.The dielectric behaviour of the film can be modelled in terms of the combination of resistance and capacitance.Currently, only the bulk contribution is observed, and the effect of grain boundaries is not observed.This can occur when there is a clear difference in the order of the contributions.The fitted data for the resistance and capacitor of Sample S2 are tabulated in table 6.
The minimum and maximum bulk resistances shown by sample S2 were approximately 29 and 283 Ω, respectively, which might explain the I-V characteristics of sample S2, where the distinction between the HRS and LRS was minimal.It is important to realise that the on/off ratio in our case (ratio between LRS and HRS) is close to 10, but is still considered as minimal as whenever bismuth ferrite is considered for memresistive applications, the on/off ratios previously reported are 61, 2500, and 1000 [71][72][73].

Ferroelectric studies
Bismuth ferrite, in general, exhibits leakage behaviour during ferroelectric characterisation.The experimental ferroelectric loop (P-E loop) contains the contribution of three components, namely, the ideal ferroelectric component, leakge component (usually the resistive leakage component), and the dielectric (linear capacitor) component, as shown in figure 18.
In experimental analysis, the degree of contribution from each component determines the quality of the sample.The present set of samples shows a greater contribution of the resistive leakage component which confirms the presence of the leakage current.Usually, it appears due to leakage along grain boundaries, although it can occur from dopants or defects (oxygen vacancies) in the grains themselves.Figure 19 shows a representative image of resistive leakage in the hysteresis curve (from sample S4, characterised at 1 kHz and 1 V applied voltage).This typical nature of the resistive leakage in PE loops is greatly explained in the supporting library of Radiant (https://www.ferrodevices.com/support-library/,document name-'Ferroelectric components-a tutorial') Technologies website.We observe the typical trend seen in ferroelectric materials, where with the increase in frequency, the polarisation greatly reduces as the contributions to the polarisation reduce [74,75].
The measured polarisation values are reported in table 7.
It must be stressed that even though the polarisation values were obtained through PE loop studies, it would be inappropriate to conclude the behaviour of the sample; hence, a more comprehensive measurement of positive-up-negative-down (PUND) was made to confirm the actual ferroelectric behaviour.
Positive-up-negative-down (PUND) is a measurement task which performs standard ferroelectric memory characterisation using a series of five pulses.A preset pulse was used to set the polarisation state of the sample opposite to that of the first pulse with a voltage V max .The second pulse switched the polarisation and measured the amount that was switched.The non-remnant polarisation is made to dissipate by switching the voltage to  zero.After the sample was settled at zero volts, a second measurement was performed (P * r).The third pulse, similar to the second pulse, is given and the measurement of P ^is made.Here, the sample polarisation is not switched with respect to the second pulse; hence, it measures the non-remnant polarisation of the sample.The fourth and fifth pulses are similar to the second and third pulses, with the only difference being that they are given in the -V max direction.Similar polarisation measurements were made as before, as shown in figure 20.The PUND measurement provides eight parameters that are distinct from those found using P-E loop studies.The principle behind the PUND measurement is that it measures both the total polarisation (remnent + non remnent-P * parameters) (during switching pulse) and only the non-remnent polarisation (during non switching pulse-P^parameters) and hence the difference between these parameters provides the actual polarisation of the sample.The delta parameters dP (= P * −P^), dP r (= P * r −P^r), −dP (= −P * − (−P^)), −dp r (= −P * r − (−)P^r).Table 8 reports the actual ferroelectric polarisation of the samples which confirms the analysis in previous studies.Sample S4 exhibited better ferroelectric behaviour than the other samples.Samples S2 and S6 which are relatively phase-pure when compared to sample S5, show reasonably better polarisation than sample S5.

Conclusion
The growth and assessment of bismuth ferrite thin films and the role of oxygen vacancies and secondary phases in their functional properties were investigated in this study.The effectiveness of this study lies in the simplicity of the design, where the variation in the film properties is thoroughly explored by changing the drying temperature of the simple spin-coating method.XRD studies showed the presence of secondary phases near the volatilisation temperature of Bi.The XPS studies confirmed the presence of oxygen vacancies.It was also observed that the pure samples had more oxygen vacancies than those with secondary phases.Hence, secondary phases are vital for regulating the oxygen vacancy concentration.I-V characterisation also helps to recognise the presence of resistive switching in the sample.We predicted that the presence of a secondary phase may affect the surface roughness along with the drying temperature, which tends to smoothen films up to 300 °C.Raman spectroscopy was utilised to determine various vibrational modes which also helped identify a possible secondary phase present in the sample which could not be detected through XRD studies.Hence, we infer that oxygen vacancies are one of the biggest threats to device applications of bismuth ferrite.Phase purity in the presence of oxygen vacancies has no advantage; hence, methods to regulate oxygen vacancies must be considered.The PUND measurements for the samples confirmed that sample S4 exhibited properties which were better suited for potential ferroelectric applications.From our work, we realise that the limited presence of secondary phases can control the oxygen vacancies, which greatly aids the functional properties.The present work aims to boost the need for further investigation of mixed-phase bismuth ferrite, which might be a way forward in capitalising on the potential of bismuth ferrite.

Figure 1 .
Figure 1.Schematic representation of the spin coating method and prepared device.

Figure 3 .
Figure 3. XRD plots of samples prepared at different drying temperatures.

Figure 4 .
Figure 4. FESEM images of the spin coated samples.

Figure 5 .
Figure 5. Cross-sectional image of sample S2 with an average thickness of 570 nm.

Figure 7 .
Figure 7. Variation in surface roughness, skewness, and kurtosis of the samples.

Figure 15 .
Figure 15.IV characteristics of sample S4 at different cycles (C).

Figure 20 .
Figure 20.PUND waveform.Schematic showing measurement of polarisation at different intervals.

Table 1 .
Preparation conditions used for the present study.

Table 2 .
Structural parameters of spin coated samples.

Table 3 .
Structural parameters of spin coated samples.

Table 4 .
Roughness of samples determined by AFM.

Table 5 .
Area ratios of Bi, Fe, and O from XPS studies.

Table 6 .
The fitted parameters from the Nyquist plot of S2.

Table 8 .
Delta parameters found through PUND measurement.