Time-resolved optical emission spectroscopy of a unipolar and a bipolar pulsed magnetron sputtering discharge in an argon/oxygen gas mixture with a cobalt target

Reactive high power impulse magnetron sputtering (HiPIMS) of a cobalt cathode in pure argon gas and with different oxygen admixtures was investigated by time-resolved optical emission spectroscopy (OES) and time-integrated energy-resolved mass spectrometry. The HiPIMS discharge was operated with a bipolar pulsed power supply capable of providing a large negative voltage with a typical pulse width of 100 μs followed by a long positive pulse with a pulse width of about 350 μs. The HiPIMS plasma in pure argon is dominated by Co+ ions. With the addition of oxygen, O+ ions become the second most prominent positive ion species. OES reveals the presence of Ar I, Co I, O I, and Ar II emission lines. The transition from an Ar+ to a Co+ ion sputtering discharge is inferred from time-resolved OES. The enhanced intensity of excited Ar+* ions is explained by simultaneous excitation and ionisation induced by energetic secondary electrons from the cathode. The intensity of violet Ar I lines is drastically reduced during HiPIMS. Intensity of near-infrared Ar I lines resumes during the positive pulse indicating an additional heating mechanism.


Introduction
Plasma-based processing techniques are frequently used for applications in fundamental and applied research [1]. Magnetron sputtering is a common technique for deposition of thin solid films [2][3][4]. Magnetron sputtering can be engaged in different discharge modes, in particular, as direct current (DC), radiofrequency, or pulsed discharge. High power impulse magnetron sputtering (HiPIMS) has attracted much attention in previous years [5][6][7][8]. HiPIMS utilises short pulse durations and low repetition rates and, hence, allows for large plasma densities, high ionisation fractions, and large ion fluxes to a substrate [8][9][10][11][12]. A more recent implementation concerns the use of a positive pulse during the afterglow. Nakano et al investigated the influence of a positive cathode bias following the negative discharge pulse during the plasma off-time [13]. Bipolar pulsing with a negative discharge pulse followed by positive pulse was utilised by Wu et al [14]. Time resolved optical diagnostics and energy-resolved mass spectrometry have revealed details of the ion kinetics [15][16][17]. A positive pulse enhances the kinetic energy of plasma ions and provides further process control during film deposition. Bipolar pulsing can give rise to smooth surfaces, dense microstructures, reduces tensile stress, increases hardness and changes the crystallographic orientation of deposited films [15,17,18]. Both increased and reduced deposition rates have been reported thus far and a clear picture regarding this aspect has not emerged yet [14,18].
In the present paper we examine the properties of a HiPIMS plasma with a cobalt cathode for deposition of cobalt oxide films. The influence of a positive pulse following the main (negative) discharge pulse was investigated. Timeresolved optical emission spectroscopy (OES) and energyresolved mass spectrometry was employed for plasma characterisation. Cobalt is a ferromagnetic metal; it requires a strong magnetic field and a comparatively thin target material for a more efficient use during magnetron sputtering. Cobalt oxide is a material with many attractive optical, magnetic, spintronic, electrochemical, catalytic, and photocatalytic properties [19][20][21][22][23][24][25]. Cobalt oxide crystallises as rock salt-type CoO and cubic spinel-type Co 3 O 4 . Film properties and composition strongly depend on deposition conditions [26]. Models describing the reactive sputtering process and the dependence on plasma parameters have been developed by Nyberg and Berg and by Depla et al [27,28]. The hysteresis effect during HiPIMS was addressed by Strijckmans et al [29]. According to these investigations, the hysteresis effect is much smaller during HiPIMS compared to dc magnetron sputtering (DCMS).

Experiment
The experimental set-up has been described in detail before [30][31][32]. A vacuum chamber is equipped with a planar magnetron (high strength magnetic field version) with a Co target (diameter 50.8 mm, nominal thickness 3 mm, purity 99.95%). Argon (purity 99.999%) and oxygen gas (purity 99.995%) is admitted to the vacuum chamber with the help of two gas flow controller. Typical argon and oxygen flow rates are 18-40sccm and 0-10sccm, respectively. The argon pressure is set to p=1.0 Pa with the help of a ultrahigh vacuum gate valve between pump and chamber; the gas pressure is measured with a capacitance vacuum gauge (Pfeiffer Vacuum, CMR 365, full range 11 Pa). The magnetron plasma is operated with the help a DC power supply (Advanced Energy MDX-1 K) operated in voltage regulation mode. Pulsed operation was achieved with the help of a home-built power switch delivering a negative pulse during the discharge [33][34][35][36]. A second home-built power switch was used for positive pulsing after the first (negative) pulse. A commercial two-channel arbitrary waveform generator (OWON AG 1022) was employed to set the repetition frequency f=100 Hz and the pulse width T n =100 μs and T p =350 μs of both negative and positive pulse, respectively, and to trigger the time-resolved measurements. Voltage and current waveforms were simultaneously recorded using voltage (Agilent 10076 A) and current probes (Textronix A622) connected to a two-channel digital oscilloscope (Agilent DSO 6012A).
OES was performed with a Shamrock SR500D spectrometer (focal length 500 mm) equipped with an iCCD detector (iStar DH334T-18-U-E3, Andor Technology, Belfast, Northern Ireland). According to the manufacturer's specifications, the quantum efficiency of the photocathode is ≈18% at 400 nm and ≈1.9% at 750 nm. The spectrometer is equipped with three gratings having 600, 1800, and 2400 lines per millimetre. An optical fibre was installed outside the vacuum chamber at an angle of 45°with respect to the target surface; it was connected on the other end to the entrance slit of the spectrometer. A typical slit width was 20 μm. Timeresolved OES measurements are performed with a gate width of 10 μs and with incremental gate steps of 10 μs in the range 0-500 μs. The spectrograph is equipped with a filter wheel in front of the entrance slit. The employed short-pass filter (FESH0800, Thorlabs Inc., Newton, New Jersey, USA) has a high transmission close to 100% in the range 490-800 nm; it is opaque elsewhere. Its purpose is to suppress the second order lines as shown in figure 1. The lower spectrum displays Ar I emission lines with wavelengths in the 750-772 nm range and O I lines at 777.19/777.42/777.54 nm. The upper spectrum additionally displays intense Co I lines in the wavelength range 372-393 nm which appear in second order and are, hence, suppressed in the lower spectrum.
Energy-resolved mass spectrometry is performed with a commercial Hiden EQP 1000 mass/energy analyser (Hiden Analytical Ltd., UK). Further details of the instrument and of the analyzer's settings can be found elsewhere [31,32]. The instrument is mounted opposite to the magnetron's race track at a distance of 5 cm from the cathode.

Current and voltage characteristics
Measured voltage and current waveforms of the magnetron discharge are displayed in figures 2 and 3. Figure 2 shows results for the HiPIMS discharge in pure argon. The voltage applied to the magnetron's cathode drops within 1 μs to −725V. The voltage remains at this level for about 7 μs prior to its increase to −590V followed by a slight decrease to about −630V. The discharge voltage is determined by the impedance of the power supply and of the plasma [37]. The increase marks the ignition of the plasma which is accompanied by an pronounced increase of the discharge current to about 36A followed by a gradual decrease to 25A towards the end of the pulse. The decrease of the discharge current after reaching the maximum is attributed to a rarefaction of the working gas. Also shown are the waveforms for the HiPIMS discharge with a positive pulse of +60 V for a duration of about 350 μs following the negative pulse. Due to experimental limitations, there is a small time gap of ≈5 μs between negative and positive pulse when the applied voltage is zero. The positive pulse induces a small current of opposite (negative) polarity compared to the regular (positive) discharge current; it rapidly decays with a time constant of  ≈30 μs. A reversed current with a magnitude of less than 5% of the main discharge current was observed by Britun et al during the positive pulse (up to 300 V) of a bipolar HiPIMS discharge [15]. The reversed current displayed two peaks separated in time by 150 μs. The current is presumably formed by negative charge carriers (electrons, negatively charged ions) since the cathode is charged positively. Figure 3 displays voltage and current waveform for the argon/oxygen gas mixture. Major differences compared to the pure argon case are a larger ignition delay of about 20 μs, a smaller ignition voltage of −570V which increases to −400V after ignition, and a larger discharge current of 52A which gradually decreases to 42A towards the end of the pulse.
The initial discharge voltage versus O 2 gas flow is displayed in figure 4. Without oxygen the required initial discharge voltage is about −870V. The (negative) voltage decreases to −970V when oxygen is added to the discharge at a gas flow rate of 2sccm. Further increasing the gas flow rate results in smaller discharge voltages; the discharge voltage reduces to −415V at an oxygen flow rate of 10sccm. For comparison, the initial discharge voltage for pulsed magnetron operation at 50 kHz (duty cycle 60%) is also shown. The observed behaviour can be explained by a variation of the ion-induced secondary electron emission due to the presence of oxygen on the cathode's surface [28,38]. As indicated in figure 4, the observed oxygen gas flow dependence seems to depend on the direction of the gas flow increase or decrease. The perspectives of a hysteresis during reactive HiPIMS was addressed by Strijckmans et al [29]. It has become clear from that investigation that the possibility of a hysteresis, if observed at all, is much smaller during HiPIMS compared to DCMS. A key point concerns the density of free oxygen atoms. Recent studies provide evidence that the O atom density is large during so-called poisoned mode while it is small in so-called metallic mode [39,40].

Energy-resolved mass spectrometry
3.2.1. Unipolar regime. Ion energy distributions (IEDs) of plasma ions was investigated by means of energy-resolved mass spectrometry. Figure 5(a) displays energy distributions of Ar + and Co + ions obtained with a HiPIMS discharge in pure argon. Significant differences in both peak position and energy-integrated intensity are observed between the two ions. The IED of Co + has a maximum at ≈12 eV which roughly coincides with the shoulder at ≈11 eV of Ar + . The IED of Ar + ions is dominated by a pronounced maximum at rather small kinetic energies. The energy-integrated intensity of Co + is about 40×larger compared to Ar + (figure 7). We want to point out that the measured Ar + ion intensities are rather small in comparison to previous measurements [32,41]. There are at least two reasons, in particular, the smaller ionisation probability of Ar compared to Co due to a larger ionisation energy (15.76 eV compared to 7.88 eV, respectively [42]). Secondly, Ar + ions are produced by ionisation of thermal gas atoms. Hence, the majority of Ar + ions has small kinetic energy, in contrast to Co + ions which have received a significantly larger kinetic energy during sputtering. We infer from the rather small Ar + ion intensity that a majority of Ar + ions is produced in a region with a negative plasma potential which accelerates positive ions towards the cathode. A responsible factor in comparison with previous measurements could be the different magnetic field configuration (high field strength and more balanced) confining electrons closer to the cathode thereby preventing slow ions to escape from that region. We want to emphasise that measured ion intensities and IEDs also show some dependence on the actual target thickness as it influences the magnetic field strength.
Adding oxygen to the HiPIMS discharge (figure 5(b)) leads to a decrease of the energy-integrated intensity of Co + ions by approximately one order of magnitude while the Ar + ion intensity shows a much weaker decrease. The pronounced decrease of the Co + intensity is somewhat compensated by a strong increase of O + and + O 2 ions. Aiempanakit et al for an Ar+O 2 plasma with a Ti target even observed that O + ions have the highest intensity of all ionised species which is in some agreement with our result [43]. The temporal variation of several ion species was investigated by Lundin et al using a reactive ionisation region model [40]. Three different (metallic, transition, and poisoned) cathode modes were considered. The model predicts a strong increase of the O + density for the poisoned mode which is caused by electron impact ionisation of metastable O( 3 P) atoms. The pronounced increase of the O + density observed in our experiment agrees well with their result. Energy distributions of O + and Co + are similar but differ significantly from Ar + ions. O + ions extend well beyond 100 eV which, in our opinion, suggests that a large fraction of these ions originates from oxygen atoms which are absorbed at the cathode's surface rather than from oxygen atoms and/or molecules in the gas phase. It could also explain the 8×larger intensity of O + compared to + O 2 ions. In addition, we observe a small component of CoO + ions which, after an initial increase, displays only a small variation with O 2 gas flow ( figure 7).
IEDs of Ar + and Co + ions during DCMS in pure argon are displayed in figure 6. Both distribution display a more or less pronounced maximum at a small kinetic energy of about 1 eV followed by a second maximum at ≈9 eV which is lower compared to the HiPIMS mode. Measured energy distributions are more narrow in DCMS compared to HiPIMS and hardly extend beyond 40-50 eV.
The measured Co + ion intensities without oxygen are one order of magnitude smaller during DCMS compared to HiPIMS which is expected since the ionisation degree during HiPIMS is generally much larger. The opposite holds for Ar + ions which are significantly more abundant during DCMS compared to HiPIMS. DCMS requires a smaller cathode potential (figure 4) and, in consequence, a weaker electric field compared to HiPIMS which could further the escape of Ar + ions from the electron trap region. With the addition of  oxygen the intensity of + O 2 ions strongly increases. Overall, the + O 2 intensity is 2-3×larger compared to O + in the DCMS discharge whereas the opposite was observed during HiPIMS (figure 7). + O 2 ions are produced in the gas phase and the energy distribution is similar compared to Ar + ions.

Bipolar regime.
IEDs obtained during bipolar HiPIMS with a positive pulse following the main negative pulse are displayed in figures 8 and 9. Figure 8 displays IEDs of Ar + and Co + ions in pure Ar with positive pulses of +0V, +60V, and +120V. The Co + IED displays an additional peak with a peak energy approximately corresponding to the applied voltage, i.e. at 60 eV and 119 eV for +60V and +120V, respectively. The intensity of the additional peaks amounts to about 2%-3% of the main peak at low energies. This is significantly less compared to the results of Keraudy et al [16].
The IED of Ar + ions differs significantly from that of Co + ions. For a positive pulse of +60V and +120V a small   peak which occurs at a somewhat smaller energy of 58 eV and 116 eV, respectively, is observed. The major difference is an orders-of-magnitude larger peak which occurs at a smaller energy of about 38 eV and 98 eV for a positive pulse of +60V and +120V, respectively. There are at least two explanations for the additional peak. Ar + unlike Co + ions are most likely produced in the plasma volume where the electric potential caused by the positive pulse is smaller and thus gives rise to a smaller kinetic energy. Keraudy et al based on their observations have proposed a plasma potential profile which extends over several cm from the cathode and essentially remains at the cathode potential over this range [16]; if so, the model will be a variance with our observation. A second interpretation rests on the assumption that the potential in some distance from the cathode is time-dependent which could affect Ar + and Co + ions in a different way. More experimental investigations will be required to solve this point.
The situation becomes more complicated if oxygen is added to the HiPIMS discharge. Figure 9 displays IEDs of O + , Ar + , and Co + ions in Ar+O 2 of a bipolar HiPIMS with a positive pulse of +120V. Observed peaks at about 85 eV, 95 eV, and 115 eV are indicated by arrows. Again, the intensity of the Ar + peak at 85 eV is rather strong while other peaks of O + and Co + ions are much weaker.     [42]. Other strong Co I lines are observed at wavelengths below 400 nm (not shown here). Co I lines originate from excitation of low-lying excited states including the ground state, e.g. Co(3d 7 4s4p) or Co(3d 8 4p). The required excitation energies from the ground state are about 4 eV and thus much smaller compared to Ar I and Ar II excitation energies (table 1). Ar I emission lines in this wavelength range originate from Ar(5p)  Ar(4s) transitions with excitation energies in the 14.46-14.74 eV range ( figure 12). Observed Ar II lines stem from excitation of Ar + (4p 4 D°) and Ar + (4p 4 P°) quartet or Ar + (4p′ 2 D°) and Ar + (4p′ 2 P°) doublet states. Addition of O 2 to the DCMS discharge reduces the overall line intensity by about 35% but does not change significantly the spectral distribution (table 1).
The violet spectrum taken in HiPIMS mode in pure argon displays strong Co I lines with about one order of magnitude larger intensity compared to DCMS while neutral Ar I lines are virtually absent. The intensity of Co I decreases by one order of magnitude when oxygen is added to the HiPIMS discharge (table 1). The pronounced decrease of the Co I line intensity is readily explained by the decreasing sputtering yield due to cathode poisoning with oxygen [27,28,49]; it is supported by the observed order-of-magnitude reduction of the deposition rate measured with a quartz micro balance [50]. The intensity of Ar II lines strongly increases with the addition of oxygen to the HiPIMS discharge in contrast to the DCMS discharge where the intensity slightly decreases. A similar rise of Ar + line emission was reported for a HiPIMS discharge with oxidised Nb and Fe cathodes [51,52].
The time dependence of several Ar I (763.51 nm), Co I (412.10 nm), and Ar II (434.80 nm) lines during the HiPIMS pulse in pure argon is shown in figure 13(a). The Ar I line intensity increases rapidly right after ignition; it is followed by a pronounced decrease towards the end of the pulse. Intensities of Co I and Ar II lines increase significantly slower and remain nearly constant during the entire pulse. It indicates that the initial phase is dominated by excitation (and ionisation) of Ar atoms whereas in the later phase excitation (and ionisation) of sputtered Co atoms becomes dominant. The decreasing Ar I intensity during the pulse is explained by gas rarefaction due to the strong sputtering wind from the cobalt cathode. In addition, a decrease of the electron temperature during the pulse could play a role [53]. The decreasing Ar + ion current to the cathode is (partly)  compensated by Co + ions which return to the cathode and give rise to self-sputtering. We would like to emphasise that the sputtering yield of Co by Ar + or Co + ions is comparatively large, about 1.04 or 1.16, respectively, at 500 eV, which is, e.g. more than twice compared to Ti sputtering [54]. We mention in passing that the Co I line intensity rather closely follows the time dependence of the discharge current ( figure 2). Similar observations and conclusions were also noted in other experiments with different target materials [38,55,56]. The behaviour of the Ar II ion line is intermediate between the Ar I and Co I lines: the Ar II intensity increases somewhat slower compared to Ar I and faster compared to Co I lines. Figure 13(b) displays the time dependence of Ar I, Co I, O I, and Ar II line intensities for the Ar+O 2 gas mixture. Some differences compared to the pure Ar case are noted, in particular, the larger ignition delay of ≈20 μs and the less pronounced decrease of the Ar I line intensity by approximately one third after reaching its maximum. The observation indicates that gas rarefaction has become less important due to the reduced sputtering yield caused by cathode poisoning. Co I and O I line intensities remain nearly constant until the end of the pulse while the Ar II line intensity shows a weak decrease.
Several processes can contribute to the formation of excited Ar +* states, in particular, direct excitation of Ar + ions by electron (e − ) impact, e Ar e Ar *, and simultaneous excitation and ionisation of argon atoms, . Direct excitation requires an energy transfer in the range of 19.22-19.64 eV or 21.35-21.50 eV for the Ar + (4p) or Ar + (4p′) states, respectively, of interest here. Significantly larger is the energy transfer for simultaneous excitation and ionisation which, for the same excited states, requires 34.98-35.40 eV or 37.11-37.26 eV ( figure 12). The relative importance of the two processes depends on the energy distribution of electrons, the respective cross sections, and the fraction of Ar + ions. The electron energy distribution during magnetron sputtering has at least two components, a (thermalized) low-energy component with typical electron temperatures in the range T e ≈2-3 eV and T e ≈3-6 eV for DCMS [57] and HiPIMS [36,53], respectively, and a high-energy component of energetic secondary electrons from the cathode with kinetic energies of several 100 eV. It has been shown that highenergy electrons constitute a significant fraction of the electron energy distribution function during HiPIMS, in particular, at the beginning of the pulse [53]. Evidence of a high-energy component in DCMS discharges is derived from the measured intensity ratio I 420 /I 419 of the 420.07 nm and 419.83 nm Ar I lines. The lines are not resolved in figure 11. However, measurements carried out with a better spectral resolution employing the 1800 lines mm −1 grating enables the separation of the two lines and a determination of their relative intensities. We measure an intensity ratio I 420 /I 419 ≈0.89 and I 420 /I 419 ≈0.90 for the DCMS and DCMS-O 2 discharge, respectively. According to Boffard et al [58], a ratio of I 420 /I 419 <1 is an indication of a significant high-energy electron component with kinetic energies of more than 25 eV.
The next point to consider is the fraction (ionisation degree) of Ar + ions. Typical electron densities in DCMS are a few 10 16 /m 3 which corresponds to an ionisation degree of about 10 −4 (at a gas pressure of 1 Pa) [57]. Corresponding numbers for the HiPIMS discharge are more than 10 18 /m 3 and a few 10 −2 [9,59]. Moreover, the ionisation degree of a HiPIMS discharge is time-dependent and starts from about zero at the beginning of a pulse. Thus, due to a small ionisation degree, the direct excitation process is rather unlikely and the simultaneous excitation and ionisation processes is expected to dominate.
Cross sections for simultaneous excitation and ionisation of selected Ar II lines (427.75 and 434.81 nm) and for direct excitation of Ar I lines (419.83/420.07 nm and 425.94 nm) are given in table 1 for an electron energy of 50 eV. All cross sections at this particular energy are approximately the same and if we generalise this argument to other electron energies the roughly equal line intensities during DCMS in pure Ar and in the Ar+O 2 gas mixture can be explained. The pronounced increase of the Ar + line intensity during HiPIMS with oxygen admixture points to an additional origin. We know from IED measurements that energetic O + ions dominate the ionic composition. Hence, interaction of energetic O + ions in, for example, charge exchange collisions with Ar atoms which proceed via formation of a transient quasi-molecule [60,61] could play an important role in this context. Reactions of this type are known for He + ions with kinetic energies as low as a few 10 eV which can interact with neutral Ar atoms to form excited Ar +* ions [62,63]. The disappearance of Ar I lines is explained by deexcitation to energetically close electronic states, for example, Ar(4d) with binding energies of 14.69-15.00 eV, respectively [42]. The violet Ar I lines originate from Ar(5p)  Ar(4s) transitions where the upper Ar(5p) states have binding energies in the range of 14.46-14.74 eV [42]. The density of Ar(5p) states is governed by excitation, de-excitation, and decay rates. The time evolution of the Ar(5p) density n 5p may be expressed as where k 1 and k 2 are the rate coefficients for excitation and deexcitation, respectively, of the Ar(5p) state, A≈10 7 /s [64] is the radiative decay rate, n Ar is the Ar atom density, and n e is the electron density. Excitation from the argon ground state by electron impact and de-excitation by electron and/or ion impact (ion density n + =n e ) is assumed. Under equilibrium conditions, we obtain m −3 and α are time-averaged electron density and duty cycle, respectively, which takes the time-structure and, hence, the temporally enhanced electron density of a pulsed discharge into account. α=0.01 and 1 is the duty cycle of the employed HiPIMS and DCMS discharge, respectively. For a qualitative understanding we include the two limiting cases k 2 n e =A and k 2 n e ?A which we expect for the DCMS and HiPIMS discharge, respectively, which holds if k 2 is in the range of 10 −5 -10 −3 cm 3 s −1 . The large value of k 2 compared to k 1 is readily justified by the orders-of-magnitude larger cross section for de-excitation compared to excitation from the ground state and the small fraction of electrons with kinetic energies larger than the excitation energy. Figure 14 shows the relative Ar(5p) density calculated with the help of equation (3) as function of duty cycle for different values of the de-excitation rate coefficient k 2 . It is noted that deexcitation can lead to a drastic reduction of the Ar(5p) density and thus can explain the disappearance of the violet Ar I lines.
The same mechanisms will also work for Ar(4p) excitation. However, we expect in this case a significantly smaller deexcitation rate which, in combination with a larger excitation rate, should have a minor influence on the observed nearinfrared Ar I lines.  figure 15(a). The first 100 μs correspond to the standard HiPIMS (without positive pulse) and no significant difference between the unipolar and bipolar case is noted in this time interval. Large differences are noted for the time domain with the positive pulse (time t>100 μs), in particular, for the Ar I line intensity. After a short decline which resembles the time gap of ≈5 μs between negative and positive pulse, the Ar I line intensity resumes to a slightly larger value from which it gradually declines until the signal disappears at about 250 μs after the commencement of the positive pulse. It is well known that the electron density in the off phase (afterglow) shows a multi-exponential decay with time constants of ≈40 and ≈210 μs [36,66]. The electron temperature in the afterglow regime reduces significantly and, as is evident from our results with the unipolar pulse, becomes too small for excitation of atoms or ions [30]. Hence, we propose that the behaviour of the Ar I line during the positive pulse is caused by returning electrons of the expanding plasma produced during the main (negative) pulse which gain kinetic energy by the applied positive voltage. We have also varied the applied voltage during the positive pulse. As shown in ( figure 16)(a), a threshold voltage of ≈25 V is required which is about 10V larger compared to the excitation energy. Co I and Ar II line intensities diminish at the end of the negative pulse and do not resume during the positive pulse. Excitation of Ar II lines by simultaneous  (table 1). However, Co atoms are produced during the negative pulse only and with typical energies in the 10 eV range. It takes less than 10 μs for a 10 eV Co atom to travel a distance of 5 cm and only a few Co atoms should be present near the cathode region during the positive pulse. Figure 15(b) displays the time dependence of Ar I, O I, Co I, and Ar II line intensities during bipolar HiPIMS with an Ar+O 2 gas mixture. Noted differences to the pure Ar case are a significantly reduced Ar I line intensity during the positive pulse. Even weaker is the intensity of the O I line intensity during the positive pulse while CoI and ArII line intensities have diminished. The observation supports our hypothesis of a returning accelerated electron cloud as the electron density in Ar/O 2 is smaller compared to the pure Ar case [36]. The required minimum voltage is about +30V and thus 5V larger compared to the pure Ar case ( figure 16(b)).

Conclusions
Reactive HiPIMS of a cobalt cathode in pure argon gas and with different oxygen admixtures was investigated by OES and energy-resolved mass spectrometry. The HiPIMS discharge was operated in unipolar mode with a negative voltage and in bipolar mode with an additional positive pulse following the negative pulse. The unipolar HiPIMS plasma in pure argon is dominated by Co + ions; the intensity of Ar + ions is comparatively small. O + intensity strongly increases with increasing oxygen gas flow. It is attributed to an oxidised cathode. IED measurements with bipolar pulsing reveal the appearance of a new peak at an energy corresponding to the applied voltage. The intensity of this new peak amounts to a few percent of the main peak. For Ar + ions a second and  much stronger peak appears at a kinetic energy which is about 23-25 eV smaller than the first peak. The exact origin of this peak is not clear yet and will require further investigations. OES shows strong Ar I, Co I, O I, and Ar II line emissions. The near-infrared part of the OES spectrum is dominated by Ar I lines resulting from excitation of Ar(4p) states. The disappearance of Ar I lines in the violet spectral range is explained by quenching of Ar(5p) states by de-excitation processes which increase with plasma density. The enhanced intensity of excited Ar +* lines is explained by simultaneous excitation and ionisation of ground state Ar atoms induced by energetic secondary electrons from the cathode. Timeresolved OES reveals a change from an Ar-dominated to a Co-dominated discharge within the first 20 μs. Light intensity diminishes at the end of the unipolar (negative) pulse. The intensity of Ar I lines resumes during bipolar HiPIMS when a positive pulse is applied. Co I and Ar I lines do not show an enhancement in this time domain. It is explained by returning electrons which gain energy by the applied positive voltage. It does not affect the Co I lines as sputtering has ceased and Co atoms are no longer produced.