Pulse length dependence of a reactive high power impulse magnetron (HiPIMS) discharge

The pulse length dependence of a reactive high power impulse magnetron sputtering (HiPIMS) discharge with a tungsten cathode in an argon+oxygen gas mixture gas was investigated. The HiPIMS discharge is operated with a variable pulse length of 20–500 µs. Discharge current measurements, optical emission spectroscopy of neutral Ar, O, and W lines, and energy-resolved ion mass spectrometry are employed. A pronounced dependence of the discharge current on pulse length is noted while the initial discharge voltage is maintained constant. Energy-resolved mass spectrometry shows that the oxygen-to-tungsten (O+/W+) and the tungsten oxide-to-tungsten (WO+/W+) ion ratio decreases with pulse length due to target cleaning. Simulation results employing the SDTrimSP program show the formation of a non-stoichiometric sub-surface compound layer of oxygen which depends on the impinging ion composition and thus on the pulse length.


Introduction
Magnetron sputtering (MS) is a physical vapour deposition method [1]. Due to its simplicity, MS has become widely used for deposition of metallic and compound thin solid films [2]. A typical MS discharge consists of a negatively biased cathode which is operated in the glow discharge regime. A magnetic field supplied by permanent magnets underneath the cathode is employed to enhance the plasma (electron, ion) density in front of the cathode. The combination of magnetic and electric * Author to whom any correspondence should be addressed.
Original Content from this work may be used under the terms of the Creative Commons Attribution 4.0 licence. Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI. fields creates a magnetic trap confining electrons to a region above the cathode. This electron trap constitutes a region of high ionisation probability where electrons undergo ionising collisions with gas atoms. Positively charged ions are attracted by the negatively biased cathode. The resulting ion bombardment causes emission of secondary electrons which are repelled by the cathode and help to sustain the discharge. In addition, bombarding ions are implanted into the cathode's surface and cause the erosion of the cathode through sputtering of atoms.
MS discharges are frequently operated with argon as working gas for deposition of thin metallic films. Reactive MS has been invented for deposition of compound (e.g. oxide or nitride) thin films. The addition of a reactive gas, e.g. oxygen or nitrogen, to the discharge introduces several additional complications. Most notable are (i) the partial or total covering of the cathode's surface by the reactive gas followed by physical and/or chemical surface processes leading to the formation of an ultrathin, e.g. oxide or nitride layer and (ii) the formation of a sub-surface compound layer through implantation of plasma ions [3]. Surface modifications change the electronic structure of the surface which influences secondary electron emission and sputtering yield [4]. The sputtering yield of a compound is often reduced compared to a metallic surface and so is the deposition rate which shows a non-linear dependence on the reactive gas flow rate [2].
Magnetron discharges are operated in various regimes and with different, e.g. direct current, radiofrequency, and pulsed modes. In recent years, so-called high power impulse magnetron sputtering (HiPIMS) has received much attention due to its particular suitability for thin film deposition [2,[5][6][7][8]. HiPIMS utilises short pulses and low repetition rates and offers large plasma densities and larger ion energies compared to direct current MS (DCMS) [5][6][7][8][9][10][11]. Due to the pulsed operation, HiPIMS discharges are operated under non-stationary plasma conditions. This has certain consequences regarding the interaction with reactive gas species. The second process (ii) is essentially limited to the pulse on regime. On the other hand, process (i) is not limited to the pulse on time but continues during the pulse off phase. It implies that the surface coverage with the reactive gas will decrease during the pulse on time and increase again during the pulse off phase and eventually may reach the maximum possible surface coverage prior to the next pulse.
In this context, the surface coverage θ (see equation (1) below) is defined as the fraction of occupied docking sites by adsorbing atoms. The number of docking sites at the surface is determined by the area density of surface atoms which may depends on the crystal structure. The maximum (full) surface coverage θ 0 = 1 is reached once all docking sites are occupied. As an example, for O atoms on a W(100) surface, the maximum surface coverage θ 0 is reached when the oxygento-tungsten (O/W) ratio on the top surface equals 1 [12] which is much smaller than the stoichiometry ratio of a WO 3 crystal.
There is an ongoing challenge to further enhance the energy-influx during film deposition making use of so-called bipolar pulsing, a positively biased anode, and the use of an external anode [13][14][15]. Another approach is to vary the employed pulse length which influences ion composition and ion energy during the discharge and thus affects the deposition rate, the composition, and other properties of deposited thin films [16][17][18][19][20][21][22][23][24][25][26][27][28]. An alternative approach was employed by Magnus et al [29] varying the pulse frequency and, hence, the time between two subsequent pulses. A thorough understanding of the involved mechanisms during reactive HiPIMS and of the corresponding time dependence is yet lacking, however. In the present paper we examine the variation of discharge properties as function of pulse length of a reactive HiPIMS discharge with a tungsten target in an Ar+O 2 gas mixture. Discharge current, energy-resolved mass spectrometry, and time-resolved optical emission spectroscopy (OES) of selected Ar, O, and W lines are investigated as function of pulse length in the range 20-500 µs. Results are compared with theoretical estimates for the oxygen surface coverage and the formation of a sub-surface compound layer at the cathode.

Experiment
The experimental set-up used in this study has been described before [30][31][32]. A planar magnetron equipped with a W target (diameter 50.8 mm, nominal thickness 6 mm, purity 99.95%) is employed. The discharge is operated in an Ar+O 2 gas mixture using two gas flow controllers with equal gas flow rates of 10 sccm for Ar and 10 sccm for O 2 . The pressure is measured with a capacitance vacuum gauge. The nominal gas pressure is set to p = 1.0 Pa. The magnetron sputtering discharge is powered by a DC power supply (Advanced Energy MDX-1 K) which is operated in constant voltage mode. Pulsed operation is achieved in combination with a home-built power switch delivering a negative pulse during the discharge [33][34][35][36]. Repetition frequency and pulse length are both controlled with the help of an arbitrary waveform generator (OWON AG 1022). Measurements are carried out by varying the pulse length between 20 µs and 500 µs with the same repetition rate of 100 Hz. In this way, the time between subsequent pulses is kept constant which is important to achieve the same surface coverage prior to each pulse. Voltage and current waveforms are simultaneously recorded using voltage (Agilent 10076 A) and current probes (Textronix A622) connected to a two-channel digital oscilloscope (Agilent DSO 6012A). The floating potential (FP) is measured with an electrical probe hanging down from the top flange. The probe tip is placed 5 cm from the cathode in axial direction. The FP is recorded with the same voltage probe and oscilloscope as described above [15].
The discharge voltage measured at the cathode and the measured FP as a function of time for a 100 µs pulse length are displayed in figure 1. The discharge voltage initially drops to −878 V, temporarily increases to −863 V within the next 5 µs, and slowly decreases to −878 V towards the pulse's ending. The simultaneously measured FP reaches a minimum of −1.8 V after 8 µs; it becomes positive after 14 µs and increases further to +3.5 V after 60 µs. It approximately remains at this level until the end of the pulse. The FP shows a rapid decay after the pulse's ending followed by an approximately exponential decay during the afterglow with an estimated decay time of about 3.3 ms.
The mean discharge power as shown in figure 2 was obtained by numerical integration of the measured current and voltage characteristics; the calculated values agree well with the discharge power shown by the power supply. The discharge power for a constant initial discharge voltage of −810 V increases monotonously as a function of pulse length; the increase can be linked to an increasing duty cycle. Also shown in figure 2 is the gas pressure which decreases with increasing pulse length indicating a significant gas (oxygen) consumption with increasing discharge power.
The discharge power as function of oxygen gas flow rate is shown in figure 3. During the measurements, the initial discharge voltage was maintained at −810 V. The discharge power drops from 300 W without oxygen to about 27 W at an O 2 gas flow rate of 1 sccm; it increases to about 43 W when the O 2 gas flow rate is raised to 10 sccm. Similar dependencies, though less pronounced, have been observed before and  attributed to the formation of a (partial) surface coverage of the cathode with oxygen [37,38].
The optical emission from the plasma is recorded with a Shamrock SR500D spectrometer (focal length 500 mm) equipped with an iCCD detector (iStar DH334T, Andor Technology, Belfast, Northern Ireland). The spectrometer is equipped with three gratings. The reported measurements are carried out employing a grating with 600 lines per millimetre which offers the largest spectral range of the three gratings

Simulation
Yield and target composition calculations are performed with the Monte Carlo program SDTrimSP [40]. The program is based on the binary-collision approximation (BCA) [41] and assumes an amorphous (randomised) target structure at room temperature and an infinite lateral size [42]. It can be run in static or dynamic mode on sequential or parallel systems. In the static mode, the composition of the target is predefined and kept fixed during the simulation, while in the dynamic mode the compositional changes of the target material are calculated self-consistently. Examples of modelings and comparison with measurements of a tungsten target are shown in [42]. Bombardment with noble gas projectiles are in good agreement with experimental data [43].
The implantation of atoms in the target changes the density and the composition inside the solid and has, therefore, an influence on the depth profile and on sputtering. Due to the low binding energy of gas atoms (nearly zero for noble gases) they get more easily sputtered. Another effect of the low binding energy is the possible out-gasing of gas atoms. The implanted and not bounded gas atoms in the target cannot move independently. Only the presence of defects during the bombardment causes a local diffusion [40].

Current waveforms
The temporal evolution of measured current waveforms for different initial discharge voltages (−600 V to −950 V) and for pulse lengths of 25 µs, 50 µs, 100 µs, 300 µs, and 500 µs are displayed in figure 6. The discharge current during, e.g. the 50 µs pulse shows an ignition delay varying between 21 µs and 6 µs for initial discharge voltages of −600 V and −950 V, respectively. After the ignition, the discharge current increases, reaching a maximum after 20-25 µs followed by a gradual decrease during longer pulses which is attributed to gas rarefaction [44,45]. The discharge current maximum (below 100 µs) strongly increases as function of the applied discharge voltage (figure 6(f)). We want to emphasise that the discharge current is significantly higher for short pulses compared to long pulses during the first 25 µs. It indicates that the evolution of a new pulse is influenced by the previous pulse. The discharge current rises again after 100 µs as is evident for the 300 µs and 500 µs pulses. A current increase for long pulses was observed by Aiempanakit et al for operation in so-called oxide mode [46]. In this time regime, the current waveform approaches the triangular shape typical for reactive HiPIMS [47,48]. It is believed that the main reason for this behaviour is due to the cathode's surface covered by adsorbed and implanted oxygen atoms and a changing surface composition during the pulse. The observation is in agreement with results of Kubart and Aijaz [49,50] who attribute the enhanced discharge current to a different surface stoichiometry of oxygen.

Optical emission spectroscopy
The time evolution of Ar I, W I, O I, and Ar II optical emission lines and of the oxygen-to-tungsten (O/W) ratio for different pulse lengths in the range 50-500 µs are displayed in figure 7. In order to maintain the same discharge conditions, the initial discharge voltage was kept at a constant value. The intensity of the Ar I emission line rapidly increases during the first 15 µs right after ignition followed by a fast decrease to a medium intensity level during the next 30 µs and, for the longer pulses, a slow decreases towards the pulse's ending. Where comparable, the temporal behaviour is rather similar and independent of the pulse length. The intensity of the W I line behaves differently. Again, there is a rapid although somewhat slower increase during the first 25 µs right after ignition but the maximum intensity reached during the first 50 µs of the pulse differs significantly as function of pulse length with the shortest pulses achieving the largest intensity. A renewed increase in intensity is noted for longer pulse lengths of 300 µs and 500 µs. The W I intensity, hence, is reminiscent of the behaviour of the discharge current, in particular, a larger intensity/current for short pulses, a falling intensity between 25 µs and 100 µs, and a renewed increase of current/intensity after 100 µs for longer pulses. Propagation of sputtered metal atoms as observed for Ti atoms may also contribute to the observed intensity increase after 100 µs [25,51].
Similar time dependencies were reported by Shimizu et al [52] for a HiPIMS discharge with a tungsten target in pure argon. Accordingly, for a pulse length of 100 µs, the Ar I emission line reaches its intensity maximum within about 10 µs which is followed by a gradual decline. The intensity of other (Ar II, W I, and W II) emission lines show a less pronounced intensity increase reaching the maximum intensity after about 35 µs [52]. The intensity of the O I line also displays a large intensity difference during the first 50 µs for different pulse length. Unlike the W I line, a rising intensity for longer pulses is not observed, however. The largest intensity variation is observed for the Ar II emission line. The line is emitted by excited Ar + * ions which are produced either by direct excitation of Ar + ions or via simultaneous excitation and ionisation of neutral Ar atoms. The second process requires a rather large energy transfer and is believed to arise from collisions of energetic secondary electrons ejected from the cathode [38]. The existence of such energetic electrons has been reported before [53,54].
The O/W ratio deduced from the OES measurements displays a rapid decline within the first 25 µs which is followed by a more gradual decrease towards the end of the pulse. The different behaviour of, in particular, W I and O I emission lines indicates that the cathode's surface coverage with oxygen decreases during longer pulses yielding to an enhanced sputtering rate for tungsten atoms.

Energy-resolved mass spectrometry
In order to elucidate this point further, we investigate the ion energy distributions of plasma ions obtained by energyresolved mass spectrometry which are displayed in figure 8. Measurements are shown for W + , WO + , WO + 2 , O + , and Ar + ions. O + , O + 2 (not shown), and Ar + are the most abundant ions. Tungsten and tungsten oxide ions are produced via sputtering from the tungsten cathode and the subsequent ionisation in the plasma region close to cathode. The initial energy distribution near the cathode follows from sputtering theory and is characterised by a pronounced high-energy component [32,55]. We mention in passing that the energy distribution predicted by sputtering theory holds for sputtering of monoatomic atoms and ions (e.g. W + ) while for polyatomic atoms and ions (e.g. WO + and WO + 2 ) a steeper energy dependence is observed [56,57]. The present results are in qualitative agreement with previous observations. There are at least two different mechanisms for O + ion production, i.e. sputtering from the oxidised cathode and O + formation via ionisation of slow oxygen atoms or dissociative ionisation of O 2 molecules in the gas phase. The first process gives rise to the production of energetic ions with large kinetic energies while O + ions which form in gas phase should have small kinetic energies of typically less than 20 eV. The measured energy distribution of O + ions is indeed characterised by a pronounced highenergy tail above about 20 eV which we attribute to sputtering of oxygen atoms from the cathode (figure 8). For comparison, we also display our results for Ar + ions which are characterised by a pronounced low-energy contribution produced in gas phase collisions but do not show any high-energy tail.
The energy-integrated ion intensity of W + ions increases with pulse length (figure 9(a)). In part, the increase is due to an increase of the duty cycle. Intensities of WO + , WO + 2 , and O + ions at first also increase with pulse length. The increase is less pronounced, however, and intensities even decrease for the largest pulse length of 400 µs. Figure 9(b) shows the WO + /W + , WO + 2 /W + , and O + /W + intensity ratios. From 20 µs to 400 µs, the WO + /W + and WO + 2 /W + ratios decrease by factors of more than 6 and 9, respectively, (figure 9). The O + /W + ratio even decreases by a factor of 13. The result is in line with the OES measurements as the O I emission monotonously decreases with time while the W I intensity after an initial decline increases again ( figure 7). The results indicate that with increasing pulse length less oxygen is available at the cathode due to a lower surface coverage and, as a consequence, the intensity of oxygen and oxygen-containing ions decreases.

Discussion
It is known that the covering of a metal surface with oxygen influences sputtering as well as secondary electron and light emission processes processes [4,[58][59][60][61]. The here observed pronounced decrease of the discharge power (figure 3) is explained by a strong reduction of the secondary electron emission yield (SEEY) due to the cathode's oxygen coverage [4]. Three stages as function of oxygen coverage have been identified in a recent investigation [61]. Already at low oxygen (sub-monolayer) coverage, the formation of a surface potential barrier hinders secondary electron emission [61]. Further increasing the oxygen coverage, however, reduces the surface barrier and the SEEY increases again. In the third phase the surface is fully covered and an oxide layer forms. The workfunction of tungsten oxide is larger compared to clean tungsten, however, remains smaller compared to a partially oxidised (sub-stoichiometric) surface. The SEEY of a fully oxidised surface is, hence, larger compared to a partially oxidised surface but remains smaller compared to a clean tungsten surface [62]. The model, hence, can explain the power decrease for low and the power increase for larger oxygen gas flows. We would like to emphasise that the model presented by Li et al [61] is not fully compatible with our situation as the cathode's material (W vs. Al) and the employed ion energies are widely different.
In order to explain the observed pulse length dependencies (figures 6 and 7), we consider the time-dependent tungsten surface coverage with oxygen during the pulse on and off time. Starting with a single oxygen layer on the tungsten surface which builds up during the pulse off time, the surface coverage decreases during the pulse on time by the interaction with plasma ions and the subsequent sputtering of oxygen atoms. The time evolution of the surface coverage can be expressed by a rate equation [60] where θ = θ(t) is the time-dependent surface coverage, Φ + is the ion flux density, Y O is the sputtering (desorption) yield of O atoms, Φ O is the oxygen flux to the surface, c s is the sticking coefficient of oxygen, and N s is the saturation surface density of oxygen. The saturation surface density (beyond which no sticking occurs) depends on the crystal orientation of the tungsten surface. The sticking coefficient c s depends on the surface coverage and on the surface orientation. The initial sticking coefficient is 0.5 for a clean W(100) and 1.0 for a clean W(110) surface [12]. The sticking coefficient decreases monotonously with increasing surface coverage. For simplicity, a constant (averaged) sticking coefficient c s = 0.4 and a surface density N s = 1.4 × 10 19 m −2 are here employed. According to Wang and Gomer [12], the maximum oxygen coverage of a W(100) surface corresponds to an oxygen-to-tungsten (O/W) ratio of 1 which is much smaller than the O/W ratio expected for a fully oxidised WO 3 solid. The sputtering (desorption) yield Y O ≈ 1.4 of oxygen is evaluated from the work of Taglauer et al assuming Ar + ion bombardment with a kinetic energy of 1 keV [59]. Cross sections σ d = 3.9 × 10 −19 m 2 for desorption of oxygen atoms from a Ni surface by Ar + impact were measured at an incidence angle ψ = 30 o with respect to the surface. Corresponding desorption cross sections for oxygen on a W surface are ≈17% smaller. For normal incidence, we have to apply a correction cos −f ϕ, where ϕ = 1 − ψ is the incidence angle with respect to the surface normal and f ≈ 1.67 [63], in reasonable agreement with experimental results [64,65]. The surface coverage θ obtained with the help of equation (1) is shown in figure 10, where θ = 1 corresponds to a fully covered surface. The surface coverage decreases during the on time. For a typical discharge current of 10 A, the surface coverage reduces within 54 µs by half (θ = 0.5) and to less than θ = 0.12 within 200 µs. The surface coverage is a pronounced function of the discharge current (power) which is in line with calculations performed by Rezek et al [18] who concluded that the O/W flux ratio decreases with increasing discharge power. The surface coverage recovers during the off time. At the present conditions, it takes about 0.7 ms, 1.9 ms, and 2.7 ms for pulse length of 25 µs, 100 µs, and 400 µs, respectively, to reach a surface coverage θ = 90%. In all investigated cases, a full surface coverage with oxygen is reached again prior to the beginning of the following pulse. The inset of figure 10(a) shows the time-averaged surface coverage as function of pulse length. It amounts to 85%, 56%, and 23% for pulse length of 25 µs, 100 µs, and 400 µs, which is in quantitative agreement with results of Ganesan et al [22]. It confirms that the average surface oxygen coverage is larger during short pulses and smaller during long pulses which explains the decreasing fractions of O + , WO + , and WO + 2 ions with increasing pulse length ( figure 9). According to the model, complete surface coverage with oxygen is achieved prior to a new pulse, regardless of the here employed pulse length. Therefore, the surface coverage alone does not explain the different initial behaviour of the pulse current as function of pulse length (figure 6). We know that the ion composition of the plasma changes during the pulse, as can be seen from the inset in figure 9. Positively charged ions are attracted by the negatively biased cathode and implanted into the surface which leads to the formation of an oxygen-containing sub-surface layer (SSL). The formation of such a SSL is studied with the help of SDTrimSp simulations ( figure 11) which takes sputtering of target atoms and implantation by impinging ions into account. For simplicity, only impinging Ar + , O + 2 , and W + ions with a kinetic energy of 970 eV are considered here. Two different simulations corresponding to low (≈20%) and high (≈50%) oxygen fractions of the bombarding plasma ions are carried out. The total fluence of impinging ions is 2.66 × 10 21 m −2 to ensure steadystate conditions. According to the simulation, the SSL shows a broad spatial distribution of O atoms at a shallow mean depth of about 3 nm. The oxygen content of the SSL reaches up to 30% and more than 60% for the two cases with low and high, respectively, impinging oxygen fraction onto the surface. As expected, the oxygen composition of the SSL is strongly influenced by the oxygen fraction of the bombarding ions. The present results are in qualitative agreement with SSL formation results reported by Ganesan et al [22].
We know that the composition of the impinging ion flux depends on the pulse length, see figure 9. For example, the fraction of O + , WO + , and WO + 2 ions is large during short pulses and a larger fraction of oxygen ions and atoms is implanted in comparison to long pulses. As a consequence, the fraction of implanted oxygen ions and atoms decreases with pulse duration and the oxygen composition of the SSL becomes smaller for long compared to short pulses. Calculations by Rezek et al also show a decrease of the target oxide fraction as function of pulse length [18]. However, the calculated decline is rather weak and quantitatively deviates from our observations. The surface composition of two target surfaces, aluminium and tantalum, after reactive magnetron sputtering in argon with a 30% O 2 admixture was examined by Schelfhout et al [66]. Accordingly, the Al surface consists of Al 2 O 3 and is therefore completely oxidised. In contrast, under otherwise similar conditions, the Ta target contains a significant amount of different suboxide states and even metallic tantalum is observed. The results also indicate that there are significant differences in the oxidation state for different target surfaces and that even with large oxygen admixtures of 50% or more, the target surface may not reach the fully oxidised Ta 2 O 5 state. Similar conditions may prevail in the present case of a W target, which is chemically similar to Ta.

Conclusions
The pulse length dependence of a HiPIMS discharge with a W cathode was investigated. Discharge current, time-resolved optical emission spectra, and energy-resolved mass spectrometry results show pronounced dependencies on pulse length. In particular, the peak discharge current is larger for short (25 µs) and smaller for long (up to 500 µs) pulses. Energy-resolved mass spectrometry reveals that the fraction of oxygen and oxygen-containing ions like WO x + (x = 1, 2) decreases during long pulses. Time-resolved OES indicates a decreasing density of excited Ar and O atoms while the W density increases during long pulses. There are at least three mechanisms to explain this behaviour. Firstly, gas depletion during the pulse reduces the Ar and O 2 density in front of the cathode. Secondly, the cathode is covered by oxygen atoms which desorb during the pulse on phase. Our simulations show that the resulting mean surface coverage with oxygen is larger for short compared to long pulses. Lastly, energetic oxygen and oxygen-containing ions impinging onto the cathode are implanted into the surface forming a sub-surface compound layer of implanted oxygen atoms which was investigated by SDTrimSp simulations. Our simulation results show that the cathode' surface coverage with oxygen changes during the pulse and that less oxygen is bound to the cathode for longer pulse length. The same holds for the oxygen content of the forming sub-surface layer which, due to the smaller oxygen fraction of impinging ions, also contains less oxygen for longer pulses. The changing oxygen surface coverage as well as the changing oxygen content of the sub-surface layer modifies the electronic properties of the surface and are thus responsible for different secondary electron emission yields as a function of time or pulse length. It explains the enhanced discharge current and the larger oxygen and oxygen-containing (WO + x ) ion fraction during short pulses. The changing ion ratio should have a significant impact on film properties deposited during reactive HiPIMS. More detailed investigations are yet necessary to understand the pulse length dependence in more detail.

Data availability statement
The data that support the findings of this study are included within the article. Numerical data are available upon reasonable request from the corresponding author.