In this work, TiMoN thin films were deposited by plasma-enhanced atomic layer deposition with an equal number of Ti and Mo precursor exposures at a substrate temperature of 250 °C. Tetrakis(dimethylamido) titanium and bis(tert-butylimido)bis(dimethylamido) molybdenum were used as sources for Ti and Mo, respectively. N2 and N2/H2 plasma were used, respectively, for TiN and MoN cycles as a source for N. Negative RF substrate bias voltage of magnitude, |Vbias|, of 0, 31, 62, 125, and 188 V were applied during the plasma half cycle. Nanocrystalline rock salt crystal structures were found by x-ray diffraction for films deposited on single-crystal Si and Si-thermal oxide substrates. Applying |Vbias| generated voids by the bombardment of high-energy ions, lowering the density. Further increase of |Vbias| caused the annihilation of voids and a slight increase in density. Four-point probe measurement showed increased electrical resistivity due to a reduction in grain size caused by continuous renucleation during growth. High-energy ions at high |Vbias| sputtered away the films resulting in low growth rates. Stripe test revealed inferior wear rates and coefficients of friction at higher |Vbias| due to low-density porous films. Epitaxial films deposited on c-plane sapphire had (111) orientation and considerable mosaicity with twinned domains rotated at 60° to each other.
I. INTRODUCTION
Transition metal nitrides, often those based on TiN, have applications as wear-resistant coatings for cutting tools,1 electrode materials,2,3 and biomedical implants.4 One of the challenges for this class of materials is to make them even more resistant to high-temperature oxidation and abrasive wear.5 Ternary solid solutions of TiN have been studied to improve the tribological properties of TiN.6,7 For example, apart from solid solution strengthening in the TixAlyN ternary system,8 the addition of Al to TiN improves oxidation resistance by forming aluminum oxide on the coating surface to prevent oxygen penetration into the film.9 Furthermore, TixMoyN7,10,11 and TixVyN12,13 have been reported to reduce the coefficient of friction by forming lubricious oxides of Mo and V, respectively, during dry sliding conditions.
Other than lowering the friction coefficient, the addition of Mo to TiN substantially improves mechanical and tribological properties, such as increasing microhardness,7,14 reducing grain size,14 and reducing wear rates.7,14 Generally, TixMoyN solid solutions are reported to have a rock salt cubic B1 crystal structure.7,10,11,15 Although the solid solubility of Mo in α-TiN is reportedly very low under equilibrium condition,16 a number of solid solutions of TixMoyN films have been reported, grown mostly by physical vapor deposition.7,10,11,14,17 We previously reported the growth of TixMo1−xN by plasma-enhanced atomic layer deposition (PEALD).15
Beyond composition, several tunable parameters are available for modifying PEALD growth conditions that may also affect physical properties including gas composition, gas pressure, substrate temperature, and RF substrate bias.18–21 This work is focused on the application of applied RF substrate bias. In vacuum-based plasma systems, the plasma bulk is quasineutral with equal densities of negative (mostly) electrons and positive ions. Due to the much higher mobility of electrons, surfaces in contact with the plasma will develop a negative charge relative to the plasma bulk. A thin plasma sheath establishes a potential gradient between the plasma bulk and the surface such that positive and negative charge surface losses are equilibrated. Electrons are repelled from surfaces, while positive ions are accelerated into them. Under typical remote plasma conditions, plasma ions will strike an electrically floating or grounded substrate with a distribution of energies up to approximately the plasma potential (∼10–50 V for inductively coupled plasmas, depending on plasma conditions including RF power, pressure, and gas composition). Applying RF power to the substrate holder when it is in contact with a plasma (substrate bias) increases the average energy of the positive plasma ions striking the substrate surface. This additional ion bombardment energy is an effective way to tune the structure and properties of the deposited film.
The application of RF bias during the plasma half cycle has been shown to impact crystallinity, density, refractive index, stress, resistivity, impurities, stoichiometry, and wet etch rate.19 AlN deposited by PEALD with N2/H2 plasma showed decreased surface roughness, improved crystallinity, higher growth per cycle (GPC), more Al–N bonds, and improvements in electrical properties like higher relative permittivity and drastic reduction of leakage current, within a narrow window of bias magnitude (−35 to −70 V).20 HfNx grown by PEALD with H2 plasma [CpHf(NMe2)3 precursor acted as the source for both Hf and N] had minimum optical and electrical resistivity and a reduction in oxygen impurity at −130 V substrate bias.22 At high bias, these films were porous and had columnar growth. Thin films of PEALD-grown TiNx and HfNx using Ar + H2 and H2 plasmas, respectively [N in TiNx and HfNx were supplied from Ti(NMe2)4 and CpHf(NMe2)3 precursors, respectively], exhibited desirable film properties such as minimum electrical resistivity, maximum density, and maximum residual stress at RF substrate bias of −130 V.18 Density increased for TiNx and HfNx films with increasingly negative substrate bias, and the residual stress transitioned from tensile to compressive. The oxygen impurity content was also reduced by the introduction of negative bias, which led to lower electrical resistivities. PEALD of HfNx using both H2 and Ar–H2 also showed a similar response to negative substrate bias.23 The lowest electrical resistivity, maximum Hf 3+ concentration in the cation mixture (Hf 3+ and Hf 4+), and lowest impurity content were found at −130 V for H2 plasma but at −225 V for Ar–H2 plasma. Therefore, the plasma ion species also plays a large role in determining the outcome of substrate biasing in addition to the magnitude of the bias voltage.
Negative substrate biasing has also been employed during physical vapor deposition, which results in increased hardness,6,8 a finer columnar grain microstructure,8,24–26 and a higher adatom mobility due to increased ion energy. The increased mobility causes the valleys of the columnar grains to fill in and results in a smoother film surface.8,24 Furthermore, the high-energy bombardment of ionized species on the reaction surface causes high compressive stress by embedding ions in the film8,26,27 and also densification by elimination of voids through faster adatom mobility.28,29 Ljungcrantz et al. demonstrated that the defect concentration is higher at low bias magnitudes, resulting in higher compressive residual stress compared to films grown at higher bias magnitudes where the compressive stress decreases by the annihilation of defects due to higher adatom mobility.25 In other cases, it has been shown that the defect concentration in the films increases with bias voltages.6,8 In terms of crystallographic structure, phase transformation,6 texure,26 and modification of lattice parameters due to ion implantation27 have been reported with the introduction of negative substrate bias.
The application of RF bias has also been extensively studied in the growth of rock salt TiN-based films, which are structurally and electrically similar to the TixMo1−xN films explored in this work. Sputtering of TiNx with negative substrate bias prevented oxidation during growth to form stoichiometric TiN, which resulted in higher hardness and elastic modulus in the absence of softer TiO2.30 It was also reported that higher adatom mobility due to high bias resulted in denser films by elimination of voids and the crystallographic orientation changed from (111) to (100). At high energy, the rock salt TiNx grows with (100) texture along which the surface free energy is minimum,31 whereas the most densely packed (111) plane is observed at lower bias.32 The film orientation is generally expected to be determined by the planes that present the lowest sum of the surface energy, the strain energy, and the stopping potential.33 There exists a critical thickness below which the surface energy term dominates resulting in (100) preferred orientation, and the strain energy term dominates above this critical thickness, which favors (111) oriented grains.34 Zhao et al. demonstrated this theory experimentally with filtered arc deposited TiN films.33 It was also reported that biasing the substrate for films below the critical thickness caused the preferred orientation to shift from (100) to (111). Conversely, films above the critical thickness are predominantly (111) oriented when unbiased and gradually change to (100) with increasing bias and eventually to (110) at very high bias (400–700 V),33 at which point the stopping potential starts to dominate.
The plasma species,25 ion energy, ion flux, and ion dosage govern the impact of substrate biasing on the film structure and properties.18 As a result, the impact of RF substrate bias is material/process specific, and straightforward predictions are difficult. These facts warrant further investigations into this field to better understand the behavior of substrate bias during plasma-assisted deposition. In our previous work, we have demonstrated that the stoichiometry of PEALD TixMo1−xN (0 ≤ x ≤ 1) can be tuned by varying the ratio of Ti and Mo precursor exposure.15 In this study, we varied the RF substrate bias voltage during PEALD growth of TixMo1−xN, with equal number of exposures of Ti and Mo precursors that results in a Ti:Mo ratio of slightly greater than unity, henceforth referred to as TiMoN for simplicity. These PEALD TiMoN films with a range of RF substrate bias voltages were deposited on different substrates and the film structure, chemistry, and tribological properties are reported.
II. EXPERIMENT
A. PEALD growth
A G2 Veeco-ALD Fiji system equipped with an optional substrate RF bias feature was used for the PEALD of TiMoN. Tetrakis(dimethylamido) titanium was used for the titanium precursor, and bis(tert-butylimido)bis(dimethylamido)molybdenum35 was used for the molybdenum precursor. The Fiji system allows for two reactor pumping techniques. For metal precursor pulsing, the reactor turbo is bypassed to increase the reactor residence time providing a longer precursor/substrate interaction. Switching to turbo pumping mode for the plasma steps provides higher radical fluxes due to higher plasma densities and reduced radical recombination.
Alternating TiN and MoN deposition cycles in a 1:1 ratio were realized to obtain TiMoN composition. For the TiN cycle, the titanium precursor, heated at 75 °C, was pulsed in the vapor draw mode for 0.25 s. Following a brief purge step, TiN was completed with a 10 s, 300 W remote inductively coupled N2 plasma. The lower vapor pressure Mo precursor was delivered to the reactor via an optional precursor Boost™ technique that introduces a charge of Ar into the precursor cylinder before pulsing the precursor/Ar mix into the reactor. Each MoN deposition cycle included three repetitions of the Mo precursor boosting process followed by a purge and then a 300 W N2/H2 (1/16) plasma for 40 s. Depositions were performed with a substrate temperature of 250 °C.
During each plasma step of the TiN and MoN cycles, following a 1 s stabilization, substrate bias, |Vbias|, was achieved by applying 13.56 MHz power to the substrate holder. The bias was left on until the end of each plasma step resulting in 9 s of bias for the TiN deposition and 39 s for MoN. Voltages of |Vbias| = 0, 31, 62, 125, and 188 V were studied. See the supplementary material (Table SI)59 for the entire deposition process reported in a tabular form. Due to the requirements of the various analytical methods applied to the TiMoN films for this study, thinner and thicker films were deposited at each bias voltage except 188 V, which suffered from an extremely small growth rate per cycle, and only a thinner film was generated.
Substrates included silicon (100), silicon with thermal oxide (1000 nm), and c-plane sapphire. Substrates were placed on the substrate holder in the turbo pumped load lock and maintained at <10–6 Torr for 30 min before loading into the reactor where they spent 1 h before the start of the deposition process. When the deposition was finished, the samples were transferred from the reactor to the load lock for 1 h of cooling before removal.
B. Ex situ spectroscopic ellipsometry
Refractive index (n) and extinction coefficient (k) were measured on the films deposited on the Si (100) substrate, with a Horiba UVISEL, utilizing a wavelength range of 225–886 nm. The spectroscopic ellipsometry data were evaluated with a model combining a Drude term and two Lorentz oscillators. Only the thinner set of samples was analyzed with ellipsometry because the thicker samples were sufficiently reflective to prevent light from penetrating to the underlying substrate.
C. XRR and XRD
A PANalytical Empyrean diffractometer with a Cu x-ray tube at a wavelength of 1.541 Å was utilized for x-ray reflectivity (XRR) and x-ray diffraction (XRD) measurements. A Bragg–Brentano HD mirror with soller slit, 4 mm mask, 1/8° divergence slit, and a 1/16° antiscatter slit was used for the incident beam for XRR measurement. The diffracted beam path used a 1/16° antiscatter slit, a soller slit, and a PIXcel3D-Medipix3 1 × 1 area detector with 1 of 255 channels active with an active length of 0.055 mm. Gonio (symmetric) scans were performed with a step size of 0.005° (2θ) from 0° to 4° with a counting rate of 0.5 s/step. XRR data were fitted with x’pert reflectivity software (v 1.3a) with a layer model of a surface oxide on the ALD film above the substrate. Symmetric XRD measurements used a Bragg–Brentano HD mirror with appropriate slits and mask sizes to maximize irradiation of the incident beam without over irradiating the sample. A soller slit and 7.5 mm antiscatter slit were used on the diffracted beam path along with the PIXcel3D-Medipix3 1 × 1 area detector in scanning line 1D mode.
Bragg–Brentano HD mirror was utilized to condition the incident beam for grazing incidence XRD (GI-XRD) measurement with a soller slit, 10 mm mask, 1/8° divergence slit, and a 1/8° antiscatter slit. The diffracted beam path utilized a soller slit, parallel plate collimator of 0.18° opening, and a PIXcel3D-Medipix3 1 × 1 area detector. 2θ scans were taken keeping the incident angle ω fixed at an angle of 1° with respect to the film surface, with a step size of 0.04° and a counting rate of 0.55 s/step.
D. Four-point probe
Film sheet resistance values were measured using a Lucas Labs 302 Resistivity Stand with a Keithley 2400 SourceMeter. Resistance measurements were taken on the films deposited on the Si-thermal oxide substrate. Film resistivity values reported are the product of the four-point probe sheet resistance values and film thicknesses derived from the XRR fitting.
E. X-ray photoelectron spectroscopy
X-ray photoelectron spectroscopy (XPS) measurements were collected using a Thermo K-alpha system using monochromatic aluminum K-α radiation with an excitation energy of 1486.7 eV and a hemispherical detector. High-resolution scans were collected using a step size of 0.1 eV and an appropriate number of scans to achieve suitable signal-to-noise ratios. Compositional information was determined by fitting XPS data using relative sensitivity factors obtained from Thermo using thermo avantage 5.973 software. Peak locations were calibrated to the C1s peak energy for adventitious carbon at 284.8 eV. Samples were sputtered for 30 s using a 200 eV Ar+ ion beam to remove surface contamination from air exposure.
F. Friction and wear testing
Friction and wear testing were performed to study the effects of varying RF bias voltage and substrate material on performance. Films were tested in a linear bidirectional reciprocating ball-on-flat configuration. A fixed 3 mm diameter ruby ball was mounted to a biaxial load cell to measure normal and friction forces. Wear and friction were monitored incrementally throughout the experiment by performing a stripe test, a method that preserves sections of the wear track at incremental sliding cycles to investigate the evolution of the wear behavior.36,37 All samples underwent a total of 10 000 sliding cycles with a starting stroke length of 10 mm (20 mm total cycle length). Stripe testing was performed with cycle intervals of 500, 1k, 1.5k, 3k, and 5k, with stroke lengths of 10, 8, 6, 4, and 2 mm respectively.
Before testing, samples and ruby counter-samples were rinsed with methanol and allowed to dry. Samples were tested in lab air with an RH of 30 ± 5%. A normal load of 0.1 N and a sliding speed of 1 mm/s were applied corresponding to a max Hertzian contact pressure of 820 MPa. The wear volume of each cycle interval was determined from the worn surface topographies measured with a scanning white-light optical profilometer (Bruker Contour GT, Billerica, MA). Five line scans (10 per stripe region) for each worn surface profile were individually analyzed to assess the cross-sectional area of the wear scar.38 Wear rate {K [mm3/(N m)]} shown in Eq. (1) is the volume lost [V (mm3)] per unit normal load [FN (N)] multiplied by sliding distance {d (m)]. The wear rate is determined by calculating the average measured worn cross-sectional area, A (mm2), divided by the product of the applied normal load FN and the number of total sliding cycles N (1 cycle = 2 sliding passes, s),
The friction coefficient was calculated by dividing the frictional force over the normal force. The average friction coefficient was calculated incrementally every 500 cycles of sliding.
G. Roughness measurements
Roughness measurements on TiMoN thick films were performed on a Bruker Dimension Icon atomic force microscope (AFM; Bruker, Billerica, MA, USA). Surface topography was measured using a sharp triangular silicon nitride cantilever, ScanAsyst-Air, with a radius of ∼2 nm. The data were collected at 1 Hz rate in the PeakForce mode. The RMS and Ra values were determined using nanoscope analysis software v 2.0 (Bruker, Billerica, MA, USA).
III. RESULTS AND DISCUSSION
A. PEALD growth
TiMoN films grown on Si (100) substrates received a varying negative RF substrate bias magnitude, |Vbias|, ranging from 0 to 188 V. The GPC is defined here as the average increase in thickness (derived from the XRR fitting) per metal (or N plasma) exposure (Fig. 1). Two sample sets, hereto referred to as thick and thin, were analyzed and both had identical growth parameters except for the total number of ALD cycles. The thin set was grown with 444 total metal precursor pulses, whereas the number of cycles for the thick samples were adjusted to keep the thickness ∼70 nm (0 V = 1332, 31 and 62 V = 1100 cycles and 125 V = 2200 cycles). Films grown on the unbiased substrates had growth rates (∼0.5 Å/cycle) similar to the previously reported case that were grown using the similar method.15,18 Introducing RF substrate bias increased the GPCs to a maximum at |Vbias| = 31–62 V, possibly due to nucleation of voids or gas bubbles by impingement of high energy ion species.22,29 Further increases in |Vbias| caused the growth rate to decrease, which can be explained by sputtering of deposited atoms by incoming ions with excessive energy brought about by the high bias voltage.39–42 Note that the film thickness at |Vbias| = 188 V was only 2.7 nm for 444 ALD cycles giving a very low GPC of 0.06 Å/cycle. This low sample thickness is not ideal for XRD or XPS analysis, so this sample is not discussed in the following analyses.
B. Composition
XPS was used to measure the composition of samples grown on silicon substrates. Compositional data are shown in Fig. 2, where a number of trends emerge from increasing substrate bias. Overall, N content increases with increasing substrate bias, while both Ti and Mo content decreases, resulting in an increasing N:metal ratio from 0.9 to 1.01 and decreasing Ti:Mo ratio from 1.9 to 1.5, shown in the right axis of Fig. 2(a). These data indicate an increased incorporation of MoN at higher substrate biases, although the films are still Ti rich relative to the (1:1) pulse ratio. Data from 188 V substrate bias were omitted, as those films were too thin for reliable XPS quantification of the bulk material. Overall, TiMoN films deposited at all substrate biases exhibit oxygen contamination below 2.5% and carbon contamination below 2% [Fig. 2(b)]. Our ALD reactor is equipped with a fused silica (“quartz”)-based ICP source, which could act as a source for the oxygen contamination.
High-resolution XPS spectra (Fig. 3) indicate negligible changes to the chemical binding environment of the constituent elements for up to |Vbias| = 125 V. The Ti2p region exhibits a doublet at 455.5 eV with 2p3/2 and 2p1/2 separated by 6 eV, along with the associated shakeup peaks, both indicative of Ti3+. The Mo3d region exhibits two doublets, one at 228.8 eV with 3d5/2 and 3d3/2 separated by 3.2 eV associated with Mo3+ and a second at 230.0 eV with 3d5/2 and 3d3/2 separated by 3.4 eV associated with shakeup peaks. These features are consistent with a bonding environment of mixed ternary Mo-Ti oxides and nitrides.43 Specific peak fits are not shown for clarity to better compare the constituent XPS spectra; for a more comprehensive treatment of the analysis and fitting, see our previous publication.15 The O1s peak location is consistent with a mixed TiO2-MoO3 surface oxide, while the C1s peak exhibits carbide bonding at 282.7 eV. At |Vbias| = 188 V, significant chemical changes are seen in the XPS data; however, due to the low growth rate (as determined by XRR) at 188 V, we assign these chemical species to those at the TiMoN-SiO2 film-substrate interface and not representative of the bulk of the films. The formation of TiON and additional TiO2 species are evidenced by the Ti2p and O1s high regions, while the formation of pyrrolic N species is indicated in the N1s high-resolution spectrum, consistent with the higher nitrogen content for films deposited at 188 V. As Ti is more electropositive than Mo, it would be expected to preferentially bond with O species at the SiO2 substrate interface, forming this TiON bonding, while the pyrrolic N is likely formed from energetic radicals impinging on the sample surface.44
C. Crystal structure
Both the thick and thin films grown on Si (100) substrates showed evidence of crystallinity at all |Vbias| in GI-XRD measurements (Fig. 4). Well-defined Bragg peaks were observed near 36°, 42°, and 62° 2θ corresponding to (111), (200), and (220) of the FCC rock salt structure. Weak reflections from (311), (222), (400), (331), and (420) planes of the FCC structure were also observed near 73°, 77°, 92°, 104.5°, and 108.5°, respectively. All the peak positions were found to be at an intermediate location between the cubic TiN (00-038-1420) and cubic MoN (01-079-5757), indicating solid solution between rock salt TiN and MoN. The larger Mo atom substituted the Ti atom in the FCC rock salt structure of TiN decreases the 2θ values of the Bragg reflections.10,15
Preferred crystallographic orientation plays an important role in the wear performance of TiN-based coatings.45 TiN films with a (111) texture are known to have excellent wear and abrasion properties.46 We define the Bragg peak intensity fraction, Nhkl, as the ratio of the integrated intensity, Ihkl, of the corresponding Bragg peak over the sum of the integrated intensity of (111), (200), and (220) reflections of symmetric θ-2θ XRD scan taken on the thick TiMoN samples deposited on the Si substrate. See the supplementary material for the θ-2θ XRD scans (Fig. S1)59 and Ihkl and Nhkl (Table SII)59 at various |Vbias|. A plot of Nhkl against |Vbias| shows that the unbiased film was almost entirely (111) oriented [Fig. 5(a)]. An increase of N200 was seen at |Vbias| of 31 V, where equal proportion of (111) and (200) oriented grains were found. Similar changes in orientation with increasing RF bias [from (111) to (200)] were observed for sputtered TiN30 and for ∼75 nm HfNx films grown by PEALD.22 Increasing the |Vbias| to 62 V caused N111 to decrease to a minimum and N200 decreased to zero. N220 increased from zero for the 62 V sample. A similar appearance of (220) preferred orientation at very high negative substrate bias has also been reported for filtered arc deposited TiN.33 Further increasing the |Vbias| to 125 V did not change the preferred orientation significantly compared to what was found at 62 V.
The texture of a deposited film can be influenced by the growth process,47 substrate structure,46 and the film thickness.34 Studies showed that the evaporation ion-plated TiN grows with (111) texture, whereas the magnetron-sputtered ones have moderate (100) and (110) texture.47 Thermodynamically speaking, minimization of the total energy of crystal planes dictates the crystallographic orientation. Although the surface energy is minimum for the (100) plane in the rock salt structure,33,34 the strain energy is minimum for the (111) plane.34,48 The strain energy increases linearly with the film thickness, while the surface energy remains independent. As a result, at lower film thickness, surface energy term dominates making the film (100) oriented, while the orientation along (111) and (110) is expected at higher film thickness.46 Therefore, a critical thickness exists above which the strain energy dominates over the surface energy. In this study, we did not investigate the critical thickness explicitly, but the dominance of strain energy for thick samples and surface energy for thin samples suggest that the critical thickness for this system lies somewhere between these two thicknesses (20 and 70 nm). In terms of experimentally specified parameters, the preferential orientation of the films is a complex function of film thickness and RF bias.
The average lattice parameter was calculated considering a cubic unit cell from all the observable Bragg peaks in the GI-XRD data [Fig. 5(b)]. The boxplots represent the spread of lattice parameters obtained from different crystal planes. The theoretical unit cell dimensions shown in the figure were calculated considering a solid solution of cubic TiN and MoN using the XPS-derived composition. The lattice parameters were similar to, but slightly less than, the expected values based on the chemical compositions. The changes in the lattice parameter found were not significant with the increase in bias voltages or the film thickness. The apparent crystallite sizes were calculated from Scherrer broadening of the (111) Bragg peak as we have done previously.12,15,49 The films were nanocrystalline (∼12.5 nm) on the unbiased substrates [Fig. 5(c)]. With the introduction of bias voltage, the crystallite size decreased to about half and did not change significantly with the increasing bias or film thickness. The growth of crystallites is likely disrupted by the bombardment of high-energy ions causing continuous renucleation,26,28,29 resulting in finer grains with the introduction of bias voltages. The crystallite sizes estimated from Scherrer broadening here is also consistent with the finding from cross-sectional transmission electron microscope (TEM) imaging. See the supplementary material for cross-sectional TEM image (Fig. S2)59 showing the crystallite size of TiMoN deposited at 31 V.
D. Physical properties: Electrical resistivity and density
Low electrical resistivities (672 μΩ cm) were found for films deposited on unbiased Si-thermal oxide substrates as determined by four-point probe measurements [Fig. 6(a)]. Similar results (712 μΩ cm) were observed previously under identical growth conditions.15 The resistivities almost doubled with the introduction of |Vbias| and did not change significantly with further increases in bias magnitude. For thin films with such low impurity concentration, the electrical resistivity is governed by the limitation of mean free path of the charge carriers due to grain boundary scattering.50 Reduced grain size with |Vbias| [Fig. 5(c)] provides a large volume of grain boundaries, which contributes to the grain boundary scattering and thus may be responsible for the higher resistivity. PEALD TiN showed a similar increase in resistivity due to reduced grain size.50–52
The highest density was observed for the unbiased films as determined by the XRR-derived density [Fig. 6(b)]. Similar densities were observed for PEALD TixMo1−xN15 and TiN52 deposited with Ar added N2 plasma. Introducing RF substrate bias voltages caused a drop in density possibly due to the void generation in addition to the reduction in grain sizes. Grain size decreased almost 50%, increasing the volume of grain boundaries, which may also contribute to density decreases. Densification at higher |Vbias| is observed, which can be explained by elimination of voids by higher adatom mobility.28,29 Note that the initial decrease in density is larger than the subsequent rise at higher |Vbias|. This observation may be explained in that the initial decrease is due to the combination of void generation and reduction in grain sizes, but the grain sizes did not change at higher |Vbias| and the increase in density may be due to the elimination of voids. At all bias conditions, the film densities were significantly lower than the theoretical bulk density, estimated from the cubic unit cell of TiN (00-038-1420) and MoN (01-079-5757) considering a solid solution based on the XPS-derived compositions. This deviation from the theoretical density could result from the large number of grain boundaries from the nanocrystalline grains, impurities, and the possible existence of amorphous regions.
The RMS roughness of the films deposited on the Si substrate was determined by atomic force microscopy, shown on the right axis of Fig. 6(b). See the supplementary material for AFM images used for RMS calculation (Fig. S3)59 and the Ra and RMS roughness as a function of |Vbias| (Fig. S4).59 The initial increase in roughness with |Vbias| can be attributed to the voids generated by high-energy ion impingement that appears as roughness on the surface. The smoother film found at |Vbias| = 125 V may be attributed to the filling up the valleys of the columnar grains on the surface by higher adatom mobility8,24,28 and possibly due to preferential sputtering of the peaks of the columnar grains by high-energy ion bombardment. The roughness of the films deposited on Si-thermal oxide (data not shown) is very similar to the ones shown for the films on Si (100).
E. Optical properties: Refractive index (n) and extinction coefficient (k)
The refractive index, n, and extinction coefficient, k (reported at 633 nm), for the thin deposition samples are shown in Fig. 6(c). Both n and k decrease slightly with increasing bias 0–62 V. While k continues to decrease as the bias is further increased to 125 V, n showed a slight increase. A root cause for the variations in optical properties is not obvious with stoichiometry, impurities, density, and crystallinity variations all potentially contributing. It is to be noted that the changes in n with |Vbias| follows the trend for density very closely as is expected. The n and k values are reasonable intermediate values of the previously reported pure TiN (n: 1.2, k: 2.3)53 and MoN (n: 3.6, k: 3.3)35 values.
F. Film stress
The lattice strain for the thick polycrystalline TiMoN films was estimated from GI-XRD data and was used to calculate the film stress using the sin2 ψ method.54 The elastic modulus (E) and Poisson’s ratio (ν) were approximated to the values for TiN (E = 447 GPa and ν = 0.2),55 for the biaxial stress model. The film stresses at all bias voltages were found to be compressive between 3.5 and 4.4 GPa (Fig. 7). The compressive stress in the unbiased film (3.5 GPa) was consistent with previously reported stress of PEALD TixMo1−xN.15 With increasing bias magnitude, the film stress did not change considerably.
G. Friction and wear
The lowest total wear rates and friction coefficients were observed for films with the lowest |Vbias| (Fig. 8). TiMoN films deposited on both Si (100) and Si-thermal oxide with a |Vbias| of 0 V achieved the lowest total wear rates (1 × 10−6 mm3/Nm). All films with applied bias voltage delaminated or wore through to the substrate before the completion of 10k sliding cycles. The film failure point becomes clearer in (Fig. 9), where the average friction coefficient is plotted over the total number of cycles; average friction was calculated incrementally every 500 cycles of sliding and error bars represent the standard deviation of the measured friction coefficient during the 500 increment cycles. The transition to high friction coefficients (μ > 0.5) are indicative of large-scale film delamination, subsequently resulting in sliding on the substrate. Very large error bars represent significant fluctuations in friction coefficient that can be correlated with film failure or interrupted sections of intact film and delaminated regions. Film delamination was confirmed using scanning electron microscope (SEM) imaging and energy dispersive spectroscopy (EDS) of the wear track, bulk film, and delaminated regions. See the supplementary material for SEM images (Figs. S5–S7),59 EDS spectra (Figs. S8–S11),59 and EDS compositions (Table SIII)59 inside and outside the wear scar.
The friction coefficients of films on Si with any applied |Vbias| become larger than 0.4 after 1500 sliding cycles, which correspond to an increase in wear rate [Fig. 8(a)]. The sample with |Vbias| = 125 V reached extremely high wear after few sliding cycles and wear rates were not recorded after 5k cycles because the film had completely worn away. The 31 and 62 V films had comparable friction coefficients to the 125 V sample after 3k sliding cycles but had surprisingly low friction coefficients and wear rates below 2k cycles. The 31 V sample had the lowest wear after 1k cycles, while the wear behavior of the 62 and 0 V samples performed similarly until 3k cycles wherein the 62 V sample experienced failure from delamination. This could indicate that the 31 and 62 V samples are more wear resistant but have worse adhesion to the Si substrate. Similar trends in the coatings deposited on Si were observed for TiMoN films on Si-thermal oxide. The total wear of the 0 V sample continues to decrease over the total sliding cycles and achieves a similar wear rate to that of the comparable film on Si. The 31 and 62 V samples approached similar wear rates to the 0 V sample up until 3k cycles and then began to increase with continued sliding. On both substrates, the 31 V sample was initially more wear resistant than the 0 V sample but suffered delamination with prolonged sliding.
The change in film characteristics with increasing voltage bias, such as increasing (111) orientation, decreasing grain size, and increasing compressive film stresses, has been shown to result in improved wear characteristics of mixed-metal nitride films when they are also accompanied by increased density.56 In the present study, the long-term wear durability of ALD mixed-metal nitrides decreases with increasing bias voltage, which is the opposite of what has been reported in more traditional sputtering deposition methods with bias voltage. Experiments at higher bias voltages tended to delaminate. There may be a competition between film delamination and film wear, as indicated by the 31 and 62 V samples that tend to delaminate yet have improved wear performance in the early stages. In several studies, there is a certain point at higher bias voltages that result in reduced adhesion due to low densification and/or high compressive stress.56 Such behavior is consistent with the 125 V sample in our study. At this point, we can only speculate the root of the decrease in long-term wear performance (which includes wear and delamination) with increasing bias voltage. Low-density porous films could lead to decreased wear performance at higher |Vbias|.
H. Epitaxial growth: Structure and symmetry
Films were also deposited on c-plane sapphire substrates to examine the influence of crystalline template upon film structure. These films grew epitaxially in the [111] direction on the (0001) plane of the sapphire as evident from the θ-2θ XRD scan [Fig. 10(a)]. At |Vbias| = 125 V, some grains oriented toward the [220] direction were found as a weak reflection at 62° 2θ. Such changes in orientation at very high bias voltages due to the dominance of stopping potential were previously demonstrated for filtered arc deposited TiN films.33 An increase in the FWHM of the (111) Bragg peak with the introduction of bias suggests deterioration of structural coherence [inset of Fig. 10(a)]. Further increase in bias resulted in narrower Bragg peak and longer length of structural coherence consistent with the nonmonotonic changes in other properties such as density, conductivity, and the FWHM in the nonepitaxial samples.
The rotational (azimuthal, ϕ) scans of asymmetric (020) planes of the epitaxial films showed peaks at every 60° [Fig. 10(b)]. Based on the rock salt structure, azimuthal scans of the (020) planes are expected to yield peaks with 120° separation.57 The observation here of 60 separation indicates that the film contains two domains rotated at 60° relative to each other, as demonstrated previously by similar ϕ scans and electron backscatter diffraction for (111) epitaxial rock salt MgO on GaN.57 (111) oriented epitaxial rock salt cubic ScN on hexagonal GaN substrate has also shown similar sixfold rotational symmetry due to the cubic-twined nature.58 The weak intensity for |Vbias| = 125 V samples relative to the unbiased films was possibly due to smaller domain sizes. The finer domain size can be explained by continuous renucleation by bombardment of high-energy ions at higher bias.26,28 The GPC and density derived from XRR for the epitaxial films were comparable to the polycrystalline films grown on the Si (100) substrate (data not shown).
IV. CONCLUSION
TiMoN films deposited by PEALD at 250 °C resulted in nanocrystalline films that appeared to incorporate voids/porosity and result in lower film density with the introduction of negative RF substrate bias voltage. At high |Vbias|, the deposited films were resputtered by the bombardment of incoming energetic ions causing lower GPCs. The introduction of |Vbias| also caused continuous renucleation resulting in smaller grains that may have caused higher electrical resistivity of the films due to the increased scattering of carriers from the large number of grain boundaries. The FCC rock salt polycrystalline films grown on Si and Si-thermal oxide substrates had (111) preferred orientation at lower |Vbias|, which changed to (110) at higher |Vbias|. The residual stress in the film was compressive and did not change significantly with |Vbias|. Delamination of films and high wear rates were observed at high |Vbias| possibly due to void-filled, low-density films. Epitaxial films deposited on c-plane sapphire substrate showed (111) orientation with twinned domains rotated at 60° to each other. With increasing |Vbias|, the crystal quality of the epitaxial films deteriorated possibly due to the reduced size of the twinned domains by continuous renucleation by high-energy ions.
ACKNOWLEDGMENTS
All the authors acknowledge the National Science Foundation for funding this research through Project NSF No. CMMI #1826251. This material is based on work supported by the National Science Foundation Graduate Research Fellowship Program under Grant Nos. 1449440 (K.E.V.M.) and 1842163 (T.F.B.). TEM work was carried out by Dr. Yan Xin at the National High Magnetic Field Laboratory, which is supported by National Science Foundation Cooperative Agreement No. DMR-1644779, and the State of Florida.
DATA AVAILABILITY
The data that support the findings of this study are available from the corresponding author upon reasonable request.