Atomic layer deposition of vanadium oxides for thin-film lithium-ion battery applications

Amorphous VO2 thin films are deposited by atomic layer deposition (ALD) using tetrakis[ethylmethylamino] vanadium (TEMAV) as vanadium precursor and water or ozone as the oxygen source. The crystallisation and oxidation behaviour is investigated for different oxygen partial pressures between ambient air and 3.7 Pa, resulting in phase formation diagrams on SiO2, TiN and Pt substrates, demonstrating a series of stable vanadium oxide phases in the VO2–V2O5 series. Most of the obtained phases exhibit lithium intercalation behaviour in the 1.5–4.5 V vs. Li/Li potential range, and demonstrate high volumetric capacities in the order of V2O5 < VO2 (B) < V6O13 < V3O7 < V4O9, with the latter at more than twice the capacity of the best commercial cathode materials.


Introduction
The development of an 'internet of things' requires both wireless sensor networks and autonomous microsystems. 1,2While the electronics, sensors and wireless communication systems abide Moore's law, the energy storage devices lag behind. 2,3To advance from traditional secondary batteries to energy storage systems tuned for these micro-devices, all-solid state thin-lm batteries can provide the answer. 4While the removal of a liquid component in an all-solid state battery provides a huge advancement in battery safety and bio-compatibility, 5 the thin-lm approach provides integrability to on-chip devices as well as improved charge-discharge kinetics.
][10][11][12][13][14][15][16][17] Although V 2 O 5 and VO 2 (B) are the most investigated cathode materials in lithium ion battery research, [8][9][10][11][12][13][14][15][16][17] all vanadium oxides in the Wadsley series (V n O 2n+1 ) are related to one another 18 and show promising electrochemical properties. 19,20Even though vanadium oxides have a high theoretical energy density, they suffer from poor charge-discharge kinetics due to their moderate electronic conductivity and intrinsically low ionic diffusion. 21pon going from a typical thin-lm battery (1 to 10 mm thick electrodes) to truly thin-lm electrodes (10 to 100 nm), the resistance for electronic current drops by orders of magnitude, and interface intercalation dominates over the bulk diffusion. 22owever, the thin-lm nature of these batteries inherently leads to a low energy density.4][25][26][27] Many deposition techniques have been used to deposit a variety of vanadium oxides such as PVD, 28,29 sol-gel 29 and CVD, 30 however, only few techniques have the high step-coverage of atomic layer deposition (ALD) required to coat the high-aspect-ratio structures required in 3D thin-lm batteries. 31][34][35][36][37][38] In this work, we will investigate the formation of vanadium oxide thin lms in the VO 2 -V 2 O 5 series, based on the atomic layer deposition of amorphous VO 2 .We will show that we can obtain all vanadium oxide phases in the series by changing the post-deposition annealing conditions, and demonstrate electrochemical activity of different phases.

Experimental
The vanadium oxides were grown in an experimental highvacuum ALD setup with a base pressure of 10 À7 mbar with walls heated at 95 C. 31,38,39 The ALD process used here was characterised earlier by Rampelberg et al. 7,[34][35][36]40 and is based on tetrakis[ethylmethylamino]vanadium (V(NC 2 H 5 CH 3 ) 4 , or TEMAV) in combination with H 2 O and ozone. Th precursor was held in a stainless steel container at 70 C with argon as a carrier gas.TEMAV and argon were pulsed at a pressure of 6 Â 10 À3 mbar.As reactant gases H 2 O and O 3 were applied, pulsed at 7.5 Â 10 À3 mbar and 2 Â 10 À1 mbar, respectively.The ozone was produced with an OzoneLab™ generator from a pure O 2 gas ow, resulting in a minimal concentration of 150 mg ml À1 .In between pulses, reaction products and remaining precursor or reactant gas were evacuated using a turbomolecular pump to a pressure of <2.0 Â 10 À6 mbar.The lms were grown on silicon substrates covered with 100 nm thermal SiO 2 , 60 nm PVD TiN and 80 nm PVD Pt.Two thermal ALD processes were used here: the TEMAV-H 2 O process and the TEMAV-O 3 process.To ensure similar conditions, both processes were performed at a substrate temperature of 150 C with saturated process parameters, as determined by earlier experiments.7,36 Pulse times for the TEMAV precursor were 5 and 4 s in the water and ozone processes.Water and ozone pulse times were 5 s on the SiO 2 substrate.The Pt substrate inuenced the growth, as was also reported by Premkumar et al., 34 so higher ozone pulse times (8 s) were required to ensure uniform lms.The growth rate for the water process was 0.67 A per cycle.The growth rate for the ozone process was 1.05 A per cycle, a higher value than reported by Rampelberg et al., 7 but still lower than the reported value by Premkumar et al. 34 X-Ray techniques were used to determine lm thickness (Xray reectivity or XRR), crystallinity (X-ray diffraction or XRD) and lm composition and oxidation state (X-ray photoelectron spectroscopy or XPS).XRR and XRD were performed in a Bruker D8 Discover using a copper X-ray source (Cu-Ka radiation at 0.154 nm) and a point detector (for XRR) or a linear detector (for XRD).To determine the lm crystallinity during annealing in a controlled ambient, in situ XRD was performed using a linear detector and a controlled ambient/temperature system.XPS was performed using Al-Ka radiation (0.834 nm) under a take-off angle of 45 in a high-vacuum chamber (2 Â 10 À9 mbar).A resolution of 0.108 eV was obtained.
The lm morphology was probed using a Bruker Dimension Edge atomic force microscope (AFM) and lm rms roughnesses were determined from 1 mm Â 1 mm AFM images.
Electrochemical measurements were performed in an Ar-lled glove box (O 2 < 1 ppm, H 2 O < 1 ppm) with a Metrohm Autolab PGSTAT302 connected to a three-electrode setup.Electrical contact to the TiN or Pt current collector was made by contacting the side of the sample with silver paste to a Cu foil.Lithium ribbon (99.9%,Sigma Aldrich) was used as counter and reference electrodes, and 1 M LiClO 4 in propylene carbonate (99%, io-li-tec) was used as Li + electrolyte solution.Cyclic voltammetry (CV) measurements were performed at a 10 mV s À1 sweep rate between 1.5 and 4.5 V vs. Li + /Li, or relevant potential boundaries.Galvanostatic charge-discharge measurements were performed to examine the capacity and thin lm kinetics at varying current densities.

Results and discussion
3.1 VO x thin lm deposition by thermal ALD First, the properties of the as-deposited amorphous lms were investigated.The rms roughness of the as-deposited lms on a Pt substrate, as measured by AFM, was 1.95 and 1.77 nm for the H 2 O-and O 3 -based lms, respectively.The Pt substrate was also measured, and showed a roughness of 1.75 nm, so the deposition of a 10 nm lm on the substrate barely increases the roughness, indicating smooth lms.The surface morphology was investigated using SEM, and images are shown in Fig. 1 for both lms on the Pt substrate.A very similar granular morphology is observed for both processes.These results imply that the choice of process barely inuences the as-deposited lm morphology and topology.
Next, the lm composition was investigated using XPS.Since vanadium oxides are reported to be very sensitive to the preferential sputtering of oxygen during depth proling with XPS, only the surface spectra are evaluated to determine the composition and the oxidation state. 42The measured spectra were calibrated to the O 1s signals at 530.0 eV, since calibrating using the C 1s peak is not ideal for vanadium oxides. 41As the V 2p core level binding energy depends strongly on its oxidation state, 7,42 tting can be used to determine the oxidation state of the as-deposited lms.Furthermore, the separation of the V 2p 3/2 peak from the O 1s VO x -component at 530.0 eV was also used to compare the two as-deposited ALD lms. 42Fig. 2 shows the results of this analysis, alongside a V 2 O 5 reference, using the reported binding energies for the V2p components: 515.8 and 523.2 eV for the V2p 3/2 and V2p 1/2 , and 517.2 and 524.5 eV for the V2p 3/2 and V2p 1/2 for the V 4+ and V 5+ components, respectively.Both as-deposited lms clearly have vanadium in the 4+ oxidation state.
Even though the morphology and oxidation state of the asdeposited lms are shown to be quite similar, a difference in lm density was found by evaluating the V XRF counts, normalised to the Si-substrate XRF counts to account for surface and distance factors.When these values are compared to the lm thicknesses measured by XRR, it is clear from Fig. 3 that there is more vanadium per unit thickness in the H 2 O-deposited lms than in the O 3 -deposited lm, a direct proof that the lms grown by utilizing the water process are denser when compared to those grown from the ozone process, as is consistent with XRR and XRF measurements performed in earlier work.

VO
x crystallisation and oxidation state control for 2 # x # 2.5 The as-deposited lms using the TEMAV ALD processes are found to be amorphous by XRD with vanadium in a 4+ oxidation state.By post-ALD annealing in ambients with carefully controlled oxygen partial pressures, vanadium oxides in the V n O 2n+1 series could be reached.The crystallisation was monitored using in situ XRD and the oxidation states were in accordance to the crystal state, as was conrmed by XPS on selected quenches (not shown here).Ex situ XRD on selected quenches conrmed the crystal states at intermediate temperatures.The crystallisation of the thin lms was investigated on three substrates: SiO 2 , Pt and TiN.
First, we will discuss the results on the SiO 2 substrate.The systematic study of the crystallisation behaviour of the ALD VO x on the SiO 2 substrate is presented in Fig. 4(a) and (b) for the H 2 Oand O 3 -ALD deposited lms, respectively.As an example, the phase evolution of the ozone-grown lm during annealing in a 3.7 Pa oxygen partial pressure is discussed here (top section of Fig. 4(b)).One can observe that the lm is initially amorphous by the lack of diffraction peaks.When heating in a controlled ambient, the lm crystallises into VO 2 (B) just above 400 C, and nally oxidizes between 500 and 600 C to V 6 O 13 .
The results of the in situ XRD measurements are summarized in 'phase formation diagrams', as shown in Fig. 5 for the crystallisation and oxidation on a SiO 2 substrate.These should not be interpreted as thermodynamic phase diagrams, since no constant-temperature steps were made to allow lm stabilisation to the equilibrium state, but as phase formation diagrams displaying the kinetic path the lms go through while being heated at 0.25 C s À1 in the ambients under study (He with oxygen partial pressure of 3.7, 7.4, 14.8, 29.6 and 48.1 Pa, and ambient air).The in situ XRD data provides a wealth of information on the oen quite complex phase formation sequence.On SiO 2 substrates, phase-pure regions were only observed for the VO 2 , V 6 O 13 and V 2 O 5 phases.A clear trend emerges when examining the inuence of the oxygen partial pressure.When going through both formation diagrams in Fig. 5 from low to high oxygen partial pressure, we see that phases emerge and disappear in the order of increasing oxidation state: VO 2 -V 6 O 13 -V 4 O 9 -V 3 O 7 -V 2 O 5 with respective average oxidation states for the V of 4-4.33-4.5-4.67-5.The inuence of the temperature is more complex since the phases have no time to settle to an equilibrium state at each temperature, so thin-lm kinetics will complicate this behaviour.However, we can see that in general with increasing temperature the average oxidation state rises to higher values, showing high-temperature stability for the higher oxides.
Furthermore, we can see the remarkable inuence of the asdeposited lm on the initial crystallisation of the amorphous layers.Since all parameters are equal except for the ALD process gas during the ALD process (ozone or water), this translates to an inuence of the latter on the initial crystallisation.The ozone-based ALD crystallizes into the VO 2 (B)-phase, while that via the water-based process crystallizes into the hightemperature stable (>68 C) VO 2 (R)-state (which settles to the VO 2 (M1) state when quenched in that phase).Both these phases are based on a bcc lattice with vanadium on the octahedral sites.The difference lies in the mutual orientation of the fourfold axis of the oxygen octahedra, being aligned or perpendicular for the VO 2 (B) and VO 2 (R), respectively. 43This translates into a difference in structure as well as in density: VO 2 (R) has a higher density (4.67 g cm À3 ) than VO 2 (B) (4.03 g cm À3 ). 44Earlier, we showed that the water-based process leads to growth of a higher density amorphous VO x layer than the ozone process.The high density amorphous layer deposited using the H 2 O-based ALD process crystallizes into the high density VO 2 (R) phase, while the lower density amorphous layer deposited using the ozone process crystallizes initially into the lower density VO 2 (B) phase.So, by choice of ALD process, we can choose to crystallise the lm in either the VO 2 (B) or VO 2 (M1) phase for electrochemical testing.
We hypothesize that the use of ozone in the ALD process also has an inuence on the oxidation behaviour.Even though XPS shows that these lms have the same oxidation state for the vanadium (4+), we observe a lower temperature for the introduction of higher oxidation state vanadium oxides for the ozone grown lms.For the lowest oxygen partial pressure we obtain phase-pure VO 2 for both lms.For higher oxygen partial pressures, we see the presence of V 4 O 9 upon crystallisation of the ozone grown lms, having a higher oxidation state.For the water-grown lms this was not the case, and we obtained phasepure VO 2 for all oxygen partial pressures examined (except for air), before further oxidation occurred.Even though the oxidation state is the same, a stronger oxidant is used during the ozone-based ALD process, which could incorporate more oxygen into the lms.Another hypothesis is based on the fact that the lms are less dense when using the ozone process.This could allow faster oxygen diffusion into these lms, enhancing the oxidation kinetics during post-ALD annealing.
The crystallisation of the lms on the Pt and TiN substrates was examined next, since a current collector is required to test the vanadium oxides as LIB electrode materials.The results are shown in Fig. 6.Similar trends were observed, with again formation of the VO 2 (B) or VO 2 (M1) as rst phases for the ozone and water-based ALD, respectively, and the V 2 O 5 phase occurring at high-temperature and/or high oxygen ambients.However, one can note two main differences when comparing these phase formation diagrams to those on SiO 2 .Firstly, the SiO 2 formation diagrams are much richer in phase zones than their counterparts on Pt and TiN.Furthermore, we observed that on SiO 2 the transformation to higher oxidation state vanadium oxides occurs at lower temperatures than on the other investigated substrates.In the case of a TiN substrate, the vanadium oxide phases with the highest oxidation states appear at the highest temperatures, or not at all, when compared to SiO 2 and Pt.The Pt case lies in between both other cases.Here, we see the inuence of the TiN in the substrate.We hypothesise that the TiN-substrate acts as an oxygen drain, extracting oxygen from the lm, and thus preventing further oxidation of the vanadium oxides, as was also observed by the presence of TiO 2 rutile diffraction peaks in the XRD scans taken aer the oxidation experiments.The SiO 2 is completely oxidized already, so it does not act as an oxygen drain, lowering the oxygen pressure and temperatures required to oxidise the vanadium oxides.The Pt substrate has a TiN layer underneath.Since oxygen can diffuse through the Pt layer, the underlying TiN can still get oxidized, as was conrmed by XRD.However, this oxidation is delayed by the presence of the Pt layer.This causes both observed differences: the phase-richness in the formation diagrams likely also occurs for TiN or Pt substrates, but at higher oxygen partial pressures than examined here, when the supply of oxygen is higher than the harvesting effect of the TiN.

Electrochemical characterisation of the possible VO
All of the obtained crystalline vanadium oxides (VO 2 (B), VO 2 (M1), V 6 O 13 , V 4 O 9 , V 3 O 7 and V 2 O 5 ) were tested as LIB electrodes on Pt substrates.Film crystallisation was monitored by in situ XRD to ensure corresponding phase formations and Table 1 Conversion paths from the as-deposited films to their crystallised and oxidized forms on the Pt-substrate oxidations, † and results are summarised in Table 1.Only V 3 O 7 was not obtainable in a phase-pure form on the Pt substrate, but the oxidation path indicated in Table 1 shows only a small fraction of the V 3 O 7 lm crystallised to V 2 O 5 , as was conrmed by XRD and XPS.This makes the V 3 O 7 almost phase-pure, and will be denoted further simply as V 3 O 7 .The path to V 6 O 13 does not match the phase formation diagram, since phase-pure V 6 O 13 was obtained by performing a ramp slower than the one used for the formation diagrams.Cyclic voltammetry at 10 mV s À1 was performed down to 1.5 V vs. Li + /Li to examine the electrochemical intercalation behaviour of the different vanadium oxides.As can be seen in Fig. 7, all phases show a clear lithiation and delithiation behaviour.However, even though the VO 2 (M1) appeared to show activity, initial galvanostatic chargedischarge tests showed only very low electrochemical storage capacity.Furthermore, some peaks of the cyclic voltammogram appear closely related to the electrochemical activity of the Pt substrate, as shown in ESI.† Due to the absence of storage of lithium into these lms, VO 2 (M1) was not examined in-depth further.All other lms showed a clear lithium insertion/ extraction behaviour, indicating lithium can be stored in these lms reversibly.Galvanostatic charging and discharging was performed on the lms to determine the available storage capacity and compare it to V 2 O 5 , which is the most investigated form of vanadium oxide and is generally considered to be the most promising phase.The lms were cycled at varying current densities between 0.1C and 100C, and the capacity was extrapolated to the 1C capacity to allow direct comparison between the different VO x phases, as shown in Fig. 8. Extrapolation was performed by tting a linear relation to the logarithm of the capacity plotted to the logarithm of the C-rate, which provided ts with R 2 > 0.98, demonstrating the validity of this approach.Table 2 summarizes the extrapolated capacities at 1C.Both Table 2 and Fig. 8 clearly demonstrate that, although V 2 O 5 is the most investigated vanadium oxide for LIB cathodes, the capacity of the stable region (which corresponds to insertion of one lithium into V 2 O 5 ), has a limited volumetric capacity.This is related to the low density of this V 2 O 5 (3.36 g cm À3 ) compared to that of commercial lithium-ion cathodes such as LiCoO 2 (4.9 g cm À3 ). 45All of the other investigated vanadium oxide phases have capacities well above the commercial cathodes, which is spearheaded by LiNi 0.33 Mn 0.33 Co 0.33 O 2 (NMC) with a capacity of 600 mA h cm À3 .The higher capacity for other vanadium oxides has an origin in increased material density and insertion range.First, the densities of the lower oxidation state vanadium oxides are higher than that of V 2 O 5 , leading to a higher volumetric capacity density; V 3 O 7 , V 4 O 9 , V 6 O 13 and the VO 2 (B) lms have densities of 3.61, 3.78, 3.91 and 4.0 g cm À3 , respectively.Second, while V 2 O 5 can only be used in the narrow range that leads to LiV 2 O 5 , i.e. reduction of vanadium from V 5+ to an average oxidation state of V 4.5+ , the vanadium in the series of vanadium oxides tested here changes its oxidation state much more, ranging from a change in average oxidation state of 0.62 for VO 2 (B) to almost 1.2 for V 4 O 9 .So, while V 2 O 5 and VO 2 (B) are generally considered the most promising cathode materials, vanadium oxides in between them exhibit higher volumetric capacities.

Conclusions
We demonstrated the deposition of two different forms of amorphous VO 2 using two different ALD chemistries based on the same TEMAV precursor.Films were shown to be similar in morphology and as-deposited oxidation state of vanadium, but Fig. 8 Measured delithiation capacity of the examined films on a logarithmic-linear scale (solid symbols).The data was extrapolated (dashed lines) resulting in a good match to the data (R 2 > 0.98).Micron-sized LMO is shown as a model bulk cathode, (red circles) 46 and more general bulk cathodes (shaded left corner) 45 from literature are shown for comparison.
Table 2 Delithiation capacities for the examined thin films, calculated by charging and discharging the films at varying current density over several orders of magnitude and extrapolating the measured capacity to 1C (Fig. 8 differ in density.By annealing these lms in controlled ambients with varying oxygen partial pressures, we demonstrated that all crystalline phases between VO 2 and V 2 O 5 could be obtained.The substrate was found to inuence the oxidation and crystallisation behaviour, with TiN acting as an oxygen drain delaying the oxidation.All obtained crystalline phases were characterised electrochemically, and showed activity as lithiumion electrodes.We demonstrated that, while V 2 O 5 is the most widely investigated vanadium-based cathode material, also VO 2 (B), V 6 O 13 , V 4 O 9 and V 3 O 7 can be used as cathodes, with capacities up to 1380 mA h cm À3 (V 4 O 9 ), more than twice the volumetric capacity for commercial cathode materials such as NMC.

Fig. 1
Fig. 1 Morphology of the as-deposited VO x films on the Pt substrate as measured by SEM, showing the H 2 O-based (left) and O 3 -based (right) ALD films.

Fig. 2
Fig.2XPS spectra for the VO x film surface: (a) a crystalline V 2 O 5 reference film, (b) as-deposited film using the water-based ALD process and (c) as-deposited film using the ozone-based ALD process.The binding energies are referenced to O 1s calibrated at 530.0 eV.41

Fig. 3
Fig. 3 Relation between the XRF vanadium counts (normalised to the Si signal) and the measured film thickness (XRR) for different film thicknesses for both processes.The higher values for the H 2 O-based ALD films indicate a higher VO x density for these films.The dashed lines are a guide to the eye to illustrate this.

Fig. 4
Fig. 4 Crystallisation of the as-deposited VO x thin films on the SiO 2 substrate, as studied by in situ XRD in varying oxygen partial pressure at a heating rate of 0.25 C s À1 .Symbols indicate VO x crystal states (2 # x # 2.5), as shown in the legend (top left).

Fig. 5
Fig. 5 Phase formation diagrams for the formation of crystalline VO x phases (2 # x # 2.5), based on the in situ XRD measurements (Fig. 4) at 0.25 C s À1 for He ambients with oxygen partial pressures of 3.7, 7.4, 14.8, 29.6 and 48.1 Pa, and in ambient air.The same set of symbols is used as in Fig. 4 except for the extrema, namely VO 2 and V 2 O 5 states, since these dominate the phase formation diagrams to a large extent.

Fig. 6
Fig. 6 Phase formation diagrams for the formation of the VO x crystal states (2 # x # 2.5) on TiN substrates (left) and on Pt substrates (right), at 0.25 C s À1 for He ambients with oxygen partial pressure of 3.7, 7.4, 14.8, 29.6 and 48.1 Pa, and in ambient air.The absence of the melt-line on the left figure is related to the presence of rutile TiO 2 from the substrate, complicating the analysis of the VO 2 (R) phase on the TiN substrate at high temperatures.

Fig. 7
Fig.7Cyclic voltammetry on the selected vanadium oxide thin films, performed at 10 mV s À1 in a three-electrode setup with lithium as counter and reference electrodes, and 1 M LiClO 4 in PC as electrolyte.Potential ranges were chosen according to activity.
), which corresponds to the current necessary to reach the theoretical capacity in 1 hour.Volumetric capacities of commercial cathode materials are included for comparison45 Potential (V vs. Li + /Li)