Gas temperature measurements in a pulsed, low-pressure inductively coupled plasma in oxygen

Optical emission spectroscopy (OES) of the magnetic dipole allowed O2(b1Σg+) to O2(X3Σg−) transition was investigated as a non-intrusive gas temperature diagnostic for E-mode and H-mode inductively coupled plasmas (ICP) in oxygen. It was compared to tunable diode laser absorption spectroscopy using Ar admixtures, and OES of the nitrogen Second Positive System with nitrogen admixtures. O2 OES provided accurate results for the E-mode ICP, 400–600 K for powers of 100–300 W, but in H-mode the method was unsuitable probably because of excitation of O2(b1Σg+) by metastable atomic oxygen. Rotational temperatures were measured, using N2 OES with N2 admixtures, for pulsed operation of the ICP with a 30 ms pulse duration and 15% duty cycle. It took 1–3 ms before the steady-state rotational temperatures were achieved. In addition, a small variation of matching network settings affects the plasma ignition delay time by several ms.


Introduction
Low pressure oxygen plasmas find use in several industrial applications e.g. for etching or functionalisation of materials. 1) The gas temperature in such oxygen plasmas plays a key role in the plasma chemistry and therefore in optimization of these plasmas for specific applications. 2) Well-established techniques for measuring the gas temperature, from Doppler broadening or as rotational temperature, in low-temperature plasmas include tunable diode laser absorption spectroscopy (TDLAS) on argon metastable atoms 3) and optical emission spectroscopy (OES) on the nitrogen band N 2 (C 3 Π u , ν′ = 0) to N 2 (B 3 Π g , ν′ = 0). 4) However, for pure oxygen plasmas both these methods require a gas admixture for diagnostic purposes, which can be a limiting factor. OES on the magnetic dipole transition between O 2 (b 1 Σ g + , ν′ = 0) and O 2 (X 3 Σ g − , ν″ = 0) 5,6) could overcome these limitations. Zyryanov et al. used the O 2 OES diagnostic to reliably measure gas temperatures in a dc glow discharge at ∼660-6600 Pa (5-50 Torr), using a high-resolution spectrometer. 5) Here, we investigate whether this technique can also be applied in an industrially relevant plasma; an inductively coupled plasma (ICP) at pressures of 20-50 Pa. 6) Measurements of the new O 2 OES diagnostic as a function of ICP power and pressure are compared to measurements using conventional N 2 OES and TDLAS to establish its feasibility and accuracy.
In the second part of this paper, we use the most appropriate gas temperature diagnostic, N 2 OES in this case, to investigate the temporal evolution of rotational temperature in a pulsed O 2 ICP plasma. The use of powermodulation, or pulsing, of plasmas limits the time-averaged power deposition, and therefore gas temperature, but maintains high plasma densities during the plasma pulse. It therefore offers an independent control over the electron temperature and gas temperature of the plasma, both of which are important for the plasma chemistry of the discharge. 7) In addition, power-modulation allows treatment of heat-sensitive materials, for example plastics, with high-power, highdensity plasma processes. 8) In practice, plasma pulsing introduces transient effects when the plasma is switched on, or off, [9][10][11] which are not fully understood, making pulse optimization often an empirical process. Temporal evolution of electron properties is regularly reported, e.g. [9][10][11] however, evolution of gas temperature during a plasma pulse is not often studied or taken into account in modeling. 12,13) Duten et al. reports gas temperature measurements in a pulsed, medium pressure microwave discharge in hydrogen, 7) while Macko and Veis measured gas temperatures in a pulsed glow discharge in oxygen. 14) We measure the temporal evolution of gas temperature in a plasma pulse with the aim to inform decisions about pulse duration for a desired gas temperature as well as investigate plasma formation mechanisms in a pulsed ICP.

Experimental methods
First, we describe the plasma system under study, then we will describe the different diagnostics that were used to measure the (time-resolved) rotational and translational temperatures of the plasma.

Experimental arrangement
The plasma under study here is an inductive version of the GEC reference cell, 15) with a 100 mm diameter grounded bottom electrode, and a powered top electrode consisting of a water-cooled three-turn double-spiral copper coil inside a 140 mm diameter quartz glass cylinder flange of about 7 mm wall thickness, which separates the coil from the plasma. The discharge gap is 40 mm. A turbomolecular pump achieved a base pressure of less than 10 −5 Pa. During plasma operation a scroll pump adjusted the pressure between 20 and 50 Pa of pure oxygen (with optional admixtures for diagnostic purposes as described later). Gas is supplied to the chamber using thermal mass flow controllers, operating in the range 5-20 sccm. Gas pressure is set by manually adjusting the admitted gas flow (5-20 sccm) and adjustable valve towards the pumping system. There is no real-time control of the gas pressure. The applied RF power is between 100 and 900 W. The driving frequency was 13.56 MHz and an L-type matching network was used for matching the effective load impedance to the RF generator impedance of 50 Ohm. The RF power is applied to the central connection of the coil, with the opposite end of the coil grounded. For all pressures, the E-H mode transition, i.e. a transition between a capacitively coupled electron heating mode and an inductively coupled heating mode, 16,17) occurs within this power range.
The TDLAS diagnostic consists of a narrowband (100 kHz) tunable diode laser (Toptica Photonics DL Pro) that operates around 810.4 nm. A beam splitter directs 10% of the beam to a confocal Fabry-Perot interferometer (Thorlabs SA200-8B) with a free spectral range of 1.5 GHz. The transmitted light is detected by a photodiode, connected to an oscilloscope, for laser wavelength calibration. The main laser beam is attenuated by a neutral density filter (OD 4.5) to avoid saturation broadening of the absorbing transition, before it enters the plasma vessel through an anti-reflection-coated optical window. The transmitted beam is detected on the opposite side of the vessel, using a fast (14 ns rise time), biased Si photodiode (Thorlabs DET36A/M), with an interference filter to isolate the laser wavelength from the plasma and background light.
For the OES diagnostics, an optical fibre is positioned at the quartz window of the vacuum system, approximately 300 mm from the plasma centre. Light from the entire plasma volume is collected by the fibre. This light then enters a Czerny-Turner spectrometer (Andor Shamrock 500i), coupled to an intensified charge couple device (ICCD) camera (Andor iStar 334T). A grating with 1200 lines mm −1 is used giving a spectral window of about 40 nm detected on the ICCD with a resolution of 300 pm (instrumental broadening). The full experimental setup, including diagnostics is schematically shown in Fig. 1.

OES of the
The rotational band of interest is from the low-lying, long-lived O 2 (b, v′ = 0) state to the ground state of oxygen, It is an allowed magnetic dipole transition, resulting in the 4 rotational branches, R R, R Q, P P, P Q. This specific transition was first described from its detection in stellar and atmospheric physics in 1961, 18) and since has been used for the determination of gas temperatures in DC glow discharges through OES, in some cases down to pressures of 70 Pa (0.5 Torr). 5,19,20) It has only recently been applied to a low-pressure RF ICP in the work of Wegner et al. 21) where measurements between 1 and 35 Pa were taken, however the technique used there was microwave interferometry, and not passive OES as utilised here.
The intensity I of an individual J′ to J″ line of the rotational band is given by:  22) Our simulation of line positions and intensities as a function of the gas temperature is based on the scheme described in 18) and data stated in the references within. An assumption is made that the population distribution in the upper state is that of the lower state, and described by a Boltzmann distribution, which requires a sufficient number of collisions between molecules for this to occur.
In order to derive a temperature from experimental spectra, they are fitted with modelled spectra, with the temperature T rot as a fitting parameter. It was found that there was a slight discrepancy between the positions of the experimental and modelled peaks. In particular, the gap between the RQ, RR and PQ, PP branches was not reproduced accurately in the model. This is most likely due to inaccuracies in the calculation of the level energies [Eq. (3)] and in particular the values for the rotational constants. A compression of the wavelength scale of the model by 2% around 762 nm was applied to match the positions of the modelled spectrum to the experimental ones. The result is a good fit between measurement and simulation and the associated value for T rot can be interpreted as reasonably accurate. Figure 2 shows an example of a measured and modelled spectrum.

Optical emissions spectroscopy of the
The molecular transition of interest here is the N 2 (C) to N 2 (B) transition, also known as the second positive system. In particular we focus on the band: This transition is widely used for determining rotational temperatures in plasmas. The main idea is the same as for the O 2 OES method; theoretical spectra with T rot as a fitting parameter are fitted to experimental spectra to derive the rotational temperature. A recent review by Bruggeman et al. 4) discusses the theoretical background in more detail. In addition, a detailed description of the model describing this transition can be found in. 23,24) For our case, the use of N 2 OES does mean that an admixture of N 2 needs to be added to our O 2 plasma. An admixture of 10% was used, which had negligible effect on the rotational temperatures of the plasma. i.e. rotational temperatures determined by O 2 OES gave the same value for pure O 2 and O 2 with 10% N 2 plasmas. Of course, a 10% N 2 admixture will certainly change other properties of the plasma significantly, e.g. plasma chemistry, making it unsuitable for in situ monitoring in industrial processing conditions. However, in this paper we only use N 2 OES to measure rotational gas temperature to compare with O 2 OES measurements and to study the fundamental plasma physics of gas heating during a plasma pulse. For both these studies, the N 2 admixture has negligible influence on the measured plasma parameters. Figure 3 shows an example of a measured and fitted simulated spectra for a 50 Pa, 300 W plasma in O 2 with 10% N 2 .

TDLAS on the Ar(1s 4 ) resonant state
TDLAS is another well-established technique for measuring gas temperatures in low-temperature plasmas. [25][26][27] Typically, visible or near-infrared tunable diode lasers are used to detect transitions in argon or helium atoms in the plasma, either as working gas or as admixture. In our case, we investigated absorption from a resonant state of argon instead of the more commonly used metastable state. Tian et al. 28) have shown in modelling that the resonant state has a density about an order of magnitude lower than the metastable state under lowpressure conditions. Our laser operates around 810.4 nm, corresponding to the transition Ar (1s 4 ) ⇒ Ar (2p 7 ). An admixture of 15% Ar in O 2 is used for all measurements in this paper. The value of 15% was chosen as the minimum value needed to perform TDLAS measurements under all plasma conditions. Again, it is likely that the plasma chemistry of the discharge will be affected by this admixture, however we only measure the gas temperature which is not affected much. i.e. rotational temperatures determined by O 2 OES gave the same value for pure O 2 and O 2 with 15% Ar plasmas. This is in line with measurements by Wegner et al. 29) of ion saturation current, line-integrated electron density and E-H mode transition voltages for a range of Ar/O 2 mixtures in a 10 Pa ICP. For pure O 2 and 25% Ar, the values are within roughly 10% of each other, giving some confidence that also the gas temperature is minimally affected by the admixture.
) band from a 50 Pa, 300 W plasma in O 2 . The resulting rotational temperature was 528 ± 15 K. The error estimate represents the standard deviation (of the fitting parameter) within the nonlinear least square fitting procedure used to match simulated spectrum to the measurement. Fig. 3. (Color online) Fitted simulated and experimental spectra for N 2 (C 3 Π u ) to N 2 (B 3 Π g ) band from a 50 Pa, 300 W plasma in O 2 with 10% N 2 admixture. The resulting rotational temperature was 561 ± 20 K. The error estimate represents the standard deviation (of the fitting parameter) within the nonlinear least square fitting procedure used to match simulated spectrum to the measurement.
The translational temperature of the plasma is determined from the Doppler broadening of the absorption transition. Other broadening mechanisms are negligible under these conditions. That is, pressure broadening can be estimated to be in the range 1.9-10 MHz for temperatures 300-1600 K, using the self-broadening data for this transition from Tachibana et al. 30) Broadening by other species in the plasma can be larger, nevertheless, it will be much less than the Doppler broadening which is in the order of 1 GHz. As discussed before, saturation broadening is avoided by including sufficient neutral density filtering. A Gaussian function is fit to the experimental data to determine the Doppler width, from which the gas temperature is derived. An example of an observed TDLAS signal and a fitted function is shown in Fig. 4. Figure 5 shows the results of translational and rotational temperature measurements using the three diagnostic techniques as a function of plasma power (100-900 W) for pressures of 20 and 50 Pa. The plasma was run in continuous mode for these measurements.

Comparison of diagnostics
The first consideration to make is whether the rotational temperatures (measured by N 2 and O 2 OES) can be expected to be in equilibrium with the translational (gas) temperatures (TDLAS). Bruggeman et al. 4) conclude that as long as the main population mechanism for the excited state is direct electron impact from the ground state, the rotational temperature is a good estimate for the translational (gas) temperature. This includes N 2 OES for most plasmas, as long as they do not contain Ar (due to N 2 (C) production from Ar* collisions). For O 2 OES this will be the case for lowdensity plasmas, but for higher densities, significant population of the O 2 (b 1 Σ + g ) state through electronic excitation transfer from metastable O( 1 D) atoms will become important, which can prohibit the equilibrium between rotational and translational temperatures. 31) The sharp increase in temperature observed at 425 W for 50 Pa is associated with the E to H mode transition in the plasma. Visual inspection confirmed a mode change at these powers, as shown in Fig. 6.
For 20 Pa, there is a similar, but less pronounced step at about 250 W. This is likely to be a transition from E-mode to an E/H hybrid mode, as identified by Wegner et al. 21) for plasmas below 30 Pa. It is not clear from these measurements whether there is the transition into full H-mode for high powers.
Across the E to H mode transition, a sharp increase in electron density is expected, which leads to an increase in dissociation of O 2 , releasing thermal energy and increasing gas temperature, as observed in. 21,32) In addition, the gas temperature is relatively insensitive to the gas pressure, similar to the observations of Wegner et al. for electron density for varying gas pressures. 21) When comparing the different diagnostic techniques, the first thing to note is that for E-mode plasmas, all three diagnostics agree well with each other. The exception is Ar TDLAS for the 50 Pa case, where no measurements could be made due to a lack of absorption signal. This is believed to be caused by significant depopulation of the Ar (1s 4 ) state by dissociation and Penning ionisation reactions with O 2 , as discussed by Takechi and Lieberman 33) 21,33) Therefore, for both H-mode at 50 Pa and all measurements at 20 Pa, there is a lower density of O 2 in the plasma, compared to 50 Pa E-mode, and therefore a reduction in the reaction rates between Ar(1s 4 ) and O 2 , leading to a higher density of Ar(1s 4 ), allowing reliable TDLAS measurements for these conditions.
For H-mode operation Ar TDLAS and N 2 OES agree well, within 5%-10%, for all powers. The O 2 OES diagnostic only gave measurable results for relatively low powers (300-400 W at 20 Pa and 500-600 for 50 Pa). However, these were not in agreement with the other two diagnostics. The most likely reason for this is that in H-mode, there is a significant increase in O 2 dissociation by electrons, leading to a higher atomic oxygen density. 21,32) This in turn means that there will be a significantly higher density of the metastable O( 1 D) atoms, which can populate the O 2 (b) state: 31) The metastable O(1D) atoms are mainly produced by electron impact dissociation of O 2 which means their temperature is not in equilibrium with the gas temperature, resulting in a non-equilibrium rotational distribution of the O 2 (b) state, resulting in the O 2 OES rotational temperature fits deviating from the other two diagnostics.
In addition, the O 2 OES diagnostic did not provide any measurable results for high powers (>400 W for 20 Pa, >600 W for 50 Pa), due to a reduction of the observed emission signal. Reason for this is that the upper state, O 2 (b), is believed to be heavily quenched by ground state atomic oxygen: 31,34,35)

S +
Nevertheless, Fig. 6 shows that O 2 OES can provide accurate gas temperature measurements in low-density (Emode) O 2 plasmas, in agreement with well-established techniques such as N 2 OES and Ar TDLAS. Of course, many industrial processing applications work at high densities, i.e. H-mode, for which this diagnostic is not suitable, limiting its industrial applicability to specific low-density applications.

Time-solved investigation of gas heating in a pulsed ICP
In addition, gas heating during pulsed operation of the ICP was studied using N 2 OES. This diagnostic was chosen because it works well across all pressures and powers of interest (see Fig. 5) and it can provide adequate time-resolution. The natural lifetime of the upper N 2 (C 3 Π u , v′ = 0) state is 36.6 ns, 36) and the measurement time is 200 μs. Achieving adequate timeresolution with TDLAS is more complex since a complete wavelength sweep takes longer than 200 μs, so partial acquisition over multiple pulses would be needed, making N 2 OES a more suitable choice. Figure 7 shows the temporal evolution of the rotational temperatures during a 30 ms plasma pulse in a 5 Hz pulsed ICP.
The measurements were performed using an ICCD gate time of 200 μs, while the delay between the start of the plasma pulse and the ICCD gate was varied. Light from multiple plasma pulses was accumulated. Depending on the intensity of the light emitted during the gate interval, between 75 and 500 plasma pulses were integrated to obtain good signal-to-noise. Four plasma cases were investigated, 20 Pa at both 500 and 700 W RF power and 50 Pa at 700 W. For the 50 Pa, 700 W case, the measurement was repeated on a different day, re-optimising the matchbox settings (labelled as "50 Pa 700 W repeat" in Fig. 7). It should be noted that for both cases, good matching was achieved with <0.3% reflected power. Figure 7 shows that for all cases there is a rapid increase in temperature, taking about 1-3 ms, after which it is relatively constant. The difference is that for three of the cases (20 Pa both powers and one of the 50 Pa, 700 W cases), this increase happens right at the start of the pulse, while for the "50 Pa, 700 W repeat" case, there is a delay of about 4 ms before the increase happens. The transition time of 1-3 ms is slightly longer than what is typically reported in literature after a change in power, or start of a plasma pulse. In most situations, the plasma reaches steady-state operation in about 0.5-1 ms. [9][10][11] However, this is likely due to the low duty cycle (15%) and repetition rate (5 Hz), compared to the experiments in literature. This means that in our case, during the off-time the electron density drops more significantly, requiring a longer "ignition phase" during which the electron density builds up again at the start of the next pulse.
Comparing the 20 Pa cases, we notice that the 700 W case has a more rapid increase in rotational temperature compared to the 500 W case. This can be explained by the fact that higher power will lead to a more rapidly increasing electron  density and therefore rotational temperature. Also, in steadystate, i.e. after ∼5-10 ms, the 700 W case will have a higher rotational temperature due to the higher electron density compared to the 500 W case.
Comparing the 700 W cases, i.e. 20 and 50 Pa (red dots and green upright triangle in Fig. 7), we can see that the rotational temperature for 20 Pa is higher than for 50 Pa during the temperature rise. Brockhaus et al. 9) established through modelling that during the ignition phase, the electron density decreases with increasing pressure. Here, the higher electron density for 20 Pa leads to higher rotational temperatures during the ignition phase.
Because electron density will reduce significantly during the off-period, ignition of the next pulse will not be in H mode, it has to start up in capacitive or E mode. The power supply and matching network will therefore also play a significant role in the plasma ignition since the matching network settings are fixed and optimised for inductive mode, which means that the capacitive power coupling is not optimal. Depending on the exact settings of the matching network, this could be a significant effect, see for example the "50 Pa, 700 W repeat" case in Fig. 7. Here, the matching settings turn out to be such that there is limited power deposition and a much slower plasma development. This effect can be amplified by the power supply output level regulation, i.e. if there is high reflected power, the forward power will be limited for safety reasons. This power regulation typically works on a timescale of ∼1-10 ms for our power supply. Eventually, after about 4 ms, there is sufficient plasma to trigger an E to H mode transition. Since with increasing inductive mode contributions, the power coupling efficiency increases (since this was optimized for H mode) and electron density rapidly increases and a H mode is established. It should be noted that there was no visual difference between the two 50 Pa, 700 W cases and matching was good with <0.3% reflected power for both cases. It is only when time-resolved plasma parameters, such as rotational temperature, are measured it becomes clear that there is a significant difference between the two cases. This means that for short pulses (at low duty cycles), typically less than about 10 ms, the power supply and matching network will need to be considered and monitored carefully to enable reliable operation of the pulsed plasma.
One thing that is not explicitly clear from these measurements is whether the rotational and gas temperatures follow changes in the electron density immediately or whether there is a delay between the two. However, times between collisions for electrons and neutrals under our conditions are typically less than 1 μs, which means that on the timescales of ∼100 μs investigated here, it is likely that the rotational and gas temperatures have adjusted themselves to changes in electron density. Time-resolved measurements of electron density would be needed to determine exactly how much delay there is between a change in electron density and a change in rotational and gas temperatures.

Conclusions
First, the feasibility of using a magnetic dipole transition in molecular oxygen as a non-intrusive gas temperature diagnostic for oxygen plasmas was investigated. Measurements were compared to more standard techniques requiring admixtures for diagnostic purposes. i.e. N 2 OES on the N 2 (C) to N 2 (B) transition and TDLAS of the Ar(1s 4 ) ⇒ Ar(2p 7 ) transition. It was found that O 2 OES can provide accurate gas temperature measurements in low-density (E-mode) O 2 plasmas, making it an appealing option for gas temperature monitoring for these types of discharges. For high-density (H-mode) operation at high powers, the O 2 OES did not provide accurate measurements, most likely due to nonequilibrium population of the O 2 (b) state by metastable atomic oxygen (O 1 D).
Second, the gas temperatures measured range from about 400 K in low-power E-mode to over 1800 K for high-power H-mode. A rapid increase in gas temperature is associated with the E to H mode transition that is observed for increasing power. An increasing electron density with power leads to an increase in dissociation of O 2 , releasing thermal energy and increasing gas temperature.
In pulsed operation at low duty cycles, it takes about 1-3 ms before the steady-state gas temperatures are achieved. This is slightly longer than the transition time of ∼0.5-1 ms that can be observed for pulsed plasmas with higher duty cycles. [9][10][11] The difference due to a longer afterglow period and therefore lower electron density at the start of the pulse. Rotational temperatures during the first 1-2 ms are found to be higher for higher powers (at the same pressure) and lower pressure (at the same power), both of which are in line with electron density trends during plasma ignition. 9) Finally, it was observed that the power supply output regulation and matching network settings can have a significant effect on the plasma ignition of a pulsed plasma. An ignition delay of up to 4 ms was observed for 50 Pa, 700 W with different matching network settings, but with the same 'match quality" (<0.3% reflected power). It can therefore be concluded that for short pulses, less than about 10 ms, the power supply and matching network will need to be carefully monitored and optimised to enable reliable operation of a pulsed ICP in oxygen.