Gaseous perfluorocarbons (PFCs), commonly used in dry etching processes, are highly climate damaging due to their outstanding global warming potential. This study investigates the feasibility of using optical emission spectrometry (OES) and varying plasma power to control the composition of a plasma containing oxygen (O2) and fluorine compounds etched in-process from a polytetrafluoroethylene (PTFE) precursor, as a step towards a climate friendly, closed system proposed by Kintzel et al. (2023) [3]. The resulting spectra were compared to plasma containing carbon tetrafluoride (CF4) as precursor to analyse the composition and behaviour of the process atmosphere over time and under different power conditions. We found that an increase of plasma power resulted in an increase of intensity in the spectrum associated with fluorine content relative to the intensity of oxygen. On the basis of these findings a closed loop control system using a PTFE precursor was designed as part of a two step process where the reactive fluorine species were generated in one and supplied to a second plasma chamber. While OES signals were used in the first chamber to control plasma power, the spectrum in the second chamber was measured to verify a steady output of reactive fluorine species. All findings support the feasibility of an effective closed loop controlled system with solid precursors, using OES signals and plasma power as in- and output variables, respectively.
Keywords: plasma, etching, CF4, PTFE, OES
Paper: Date of submission for publication: 26.01.2024 Date of review finishing: 26.03.2024 Publication acceptance: 28.05.2024
1 Introduction
Optical emission spectrometry (OES) is a widespread method of optical spectrometry allowing conclusions about particles from light emitted by those same particles [2]. These particles such as atoms, ions and molecules can, with an input of energy, be put into an excited state where electrons move to a higher energy level [3]. The energy to achieve this change in quantum states is individual for every element and discreetly quantifiable [4]. Upon return to the ground state photons are emitted with this exact same discreet energy which, according to the Planck-Einstein relation, is correlated inversely to the wavelength of the photon [4]. This relation between the energy necessary to excite the atoms and the wavelength emitted allows conclusions about the composition of a plasma to be drawn from the emitted spectrum [5].
Under constant input configurations time invariant values along the spectrum indicate a stable composition of the plasma. Changes of input parameters e. g. in plasma power will cause proportional change in the measured intensities of the plasma – assuming a constant plasma composition. To establish a steady state in a time variant system a closed loop control can be implemented to manipulate inputs dependent on a feedback signal from the output. The output in turn is causally dependent on the input and thereby closing the loop.
This approach was applied to a desmearing process used to remove excess epoxy from printed circuit boards presented by Kintzel et al. [1], in which the commonly used climate damaging gaseous perfluorinated hydrocarbon precursors such as carbon tetrafluoride (CF4) were substituted with solid state polytetrafluoroethylene (PTFE).
2 Process Indicators
2.1 Overview
The spectra of etching processes with both a solid and a gaseous precursor respectively were measured over time and at different plasma powers. With the step responses and the continuity of the spectra as process indicators the feasibility of a closed loop controlled process environment was investigated. The spectrum was measured with emphasis on the intensity at two wavelengths with prominent peaks: 288.8 nm and 777.0 nm. These wavelengths match ionized fluorine and oxygen respectively [6] allowing to draw conclusions about their relative concentration.
2.2 Experimental
Plasma was generated in a reaction chamber made from an ISO-K 6-way cross with a volume of about 10 L and a hollow cathode with a diameter of 76 mm and a length of 150 mm, which was connected to a plasma generator (PE 1000, Advanced Energy GmbH, Germany) with 50 kHz and a maximum power of 1000 W. The vacuum was created with a rotary vane pump (Trivac D65BCS, Leybold GmbH, Germany) in combination with a turbomolecular pump (Turbovac 50, Leybold GmbH, Germany). In the experiments with PTFE as precursor a porous PTFE rod (ElringKlinger AG, Germany) was placed inside of the hollow cathode, allowing the oxygen plasma to release fluorine compounds by etching the PTFE. Mass flow controllers (MFCs) were used for the supply of oxygen and carbon tetrafluoride (both Linde AG, Germany, quality 5.0). The flow rate was set to a total of 50 sccm, of which 10 % are carbon tetrafluoride when used as precursor. The emission spectra were measured using a UV-VIS spectrometer (FLAME-T-XR1-ES, Ocean Insights Inc., USA) with an integration time of 30 ms and plasma power ranging from 0 W to 500 W in steps of 20 W. Plasma was turned off between measurements.
2.3 Results
Figure 1a and Figure 1b show the intensities measured over of 20 s at wavelengths of 288.8 nm and 777.0 nm, respectively. Solid lines indicate PTFE and dashed lines carbon tetrafluoride as precursor while the colors indicate the power. The measurements include the step response when the plasma was turned on after about 2 seconds. The data was smoothed with a moving average of 30 measurements. The highest intensity at 288.8 nm was measured for both precursors with the most power applied, while at 777.0 nm the highest peak was measured with the highest power only when using carbon tetrafluoride as precursor. When PTFE was used as precursor the highest intensity at 777.0 nm was measured at 320 W. At 288.8 nm the gaseous carbon tetrafluoride led to very constant intensities over time, PTFE showed a slight increase, while the oxygen peak at 777.0 nm was constant for both precursors after the step response.
Fig. 1: Intensities of light emitted at different powers in an oxygen plasma with different precursors (moving average of 30 measurements, coefficient of variation after 8 s < 9 % – not shown for improved visibility); Intensities at a wavelength of 288.8 nm (left) and 777.0 nm (right)
Fig. 2: Intensities at different wavelengths over plasma power in an oxygen plasma; Intensities with carbon tetrafluoride as precursor (left) and with PTFE as precursor (right)
In Figure 2a and Figure 2b the intensities at different wavelengths are shown as a function of power for both carbon tetrafluoride and PTFE respectively. The values are the average intensities of the last 12 s of each measurement thereby filtering noise while omitting the step responses. The six wavelengths depicted were those with the highest intensity measured when using PTFE as precursor at 500 W. A direct correlation between plasma power and intensity was apparent for all wavelengths when carbon tetrafluoride was used as a precursor. When PTFE was used as a precursor, the intensity at 777.0 nm showed a maximum at 320 W.
Figure 3a and Figure 3b show the same intensities relative to the intensity of the oxygen peak at 777.0 nm. While Figure 3a shows almost no change in relative intensity for all wavelengths, independent of plasma power, Figure 3b shows an increase of all the intensities depicted, besides the oxygen peak used as basis.
Fig. 3: Intensities at different wavelengths relative to oxygen peak at 777.0 nm over plasma power in an oxygen plasma; Intensities with carbon tetrafluoride as precursor (left) and with PTFE as precursor (right)
2.4 Discussion
As shown in Figure 1a the step response of the intensity at 288.8 nm with PTFE as precursor varied from four to eight seconds and increases proportional to both the increase in plasma power and the final value. The step response when using carbon tetrafluoride as precursor showed very little delay, as the compounds were directly available to be ionized in the gaseous phase while with the solid precursor these compounds first had to be dissociated from the PTFE surface. In Figure 1b where the intensity of the oxygen peak at 777.0 nm is shown the step responses with both precursors are steep. The incline can mostly be attributed to the filter smoothing and dampening the response with the moving average over the course of 30 measurements equating just short of one second. The dampening of the step response can be reduced by using a different method to filter the signal than the moving average. Albeit slower than with a gaseous precursor the response when using PTFE is expected to be sufficiently dynamic in a system of high inertia such as an industrial dry etch processes. Furthermore, the slopes of responses at different power show that delay of the step response can be improved in a closed loop control system by overshooting the plasma power output allowing the peak intensity to settle faster.
The increase of intensity at 288.8 nm over time visible in Figure 1a and the decrease at 777.0 nm visible in Figure 1b suggest the release of fluorine compounds was facilitated with time when PTFE was used as precursor. This might be caused by thermal effects, induced by the energy of the plasma on the solid’s surface as increasing temperature can facilitate the reaction of PTFE compounds with oxygen or the desorption of reaction products. The change over time occurring due to an increasing amount of shorter and partially broken polymer chains on the deteriorating surface is considered unlikely due to the lack of saturation over the overall course of the measurements which used the same PTFE rod as precursor. When carbon tetrafluoride was used as a precursor these effects did not appear, indicating that the chemical reactions did not change over time and therefore no significant change in temperature or its effects occur, resulting in a constant composition of the plasma overall. With no abrupt changes in the process atmosphere the continuity of the process indicates to be adequately controllable. Even without compensation of the slight increase of intensity over time attributed to thermal effects, a stable atmosphere is to be expected, once the PTFE surface temperature reaches a steady state. This is still to be confirmed in measurements exceeding the investigated process time of 20 seconds. In a closed loop control system using the intensity as input these effects would innately be accounted for.
Figure 1a shows proportional increase of all intensities when the power was increased, Figure 1b shows the same for carbon tetrafluoride, but not for PTFE, where the oxygen peak at 777.0 nm was negatively correlated with plasma power after peaking at 320 W. In the PTFE plasma the relative decline of the oxygen peak with power and the proportionality of the other intensities are indicating a shift in chemical reactions, such that the consumption of oxygen and the etching of PTFE both increase. This causes a surplus in the release of fluorine species leading to a change of the relative fluorine concentration in the process atmosphere.
This is also apparent with the absolute intensities at different wavelengths over plasma power shown in Figure 2a. The increase in power was reflected in the increase in excitation, increasing all wavelengths’ intensities when supplied with the mass flow controlled, gaseous carbon tetrafluoride as precursor. The decrease of the oxygen peak at powers above 320 W with the continued increase of the fluorine peak in Figure 2b, is supporting the proposed shift in chemical reactions of the oxygen with the PTFE, likely binding with carbon and forming stable bonds in the plasma. This suggests a surplus of released fluorine species, implying that the fluorine content can be controlled by adjusting plasma power and thereby further validating the feasibility of the proposed control system. The comparison of intensities relative to the oxygen peak at 777.0 nm at different plasma powers in Figure 3a and Figure 3b show that the effect of changes in plasma power on the composition of the plasma was negligible when carbon tetrafluoride was used as a precursor, reflecting the expected proportional change of the spectrum when plasma power was increased in a constant plasma composition. Contrary to this when PTFE was used as precursor the changes in power altered the relative intensities in the plasma to a point where the oxygen peak was no longer the highest in the spectrum, having been surpassed by the intensity at 288.0 nm, suggesting significant changes to the plasma’s composition which are deductible from these wavelengths.
In conclusion, the presented data suggests the feasibility of a closed loop control system. While plasma power as regulating output variable allows manipulation of the concentration of fluorine species, the investigated wavelengths at 288.8 nm and 777.0 nm show promise as feedback signal, reflecting the changes in relative concentration of fluorine species and oxygen.
3 Verification of a steady state
3.1 Overview
The system was modified to accommodate a second plasma system for generating the reactive fluorine species in an external system, simulating the replacement of gaseous precursors in an existing system as illustrated in Figure 4. The local emission spectrum in Chamber 2 and changes therein when igniting plasma in Chamber 1 were investigated by firstly igniting plasma in Chamber 2. With about 2 seconds delay plasma was ignited in Chamber 1 with the closed control loop implemented independently of Chamber 2. Thereby the transient response in Chamber 2 was recorded and can be investigated for a stable final value which is indicative of a steady state. The measurements continue for about one second after Plasma was turned off in Chamber 1 to include the second transient response. The occurrence of a steady state then can be verified by rejecting the null hypothesis of a unit root with a Dickey-Fuller test or an augmented Dickey-Fuller test in case of autocorrelated measurements [7, 8].
Fig. 4: Test set-up with two plasma chambers
3.2 Experimental
The modified set-up consisted of a plasma system PlasmaFlecto 10 (PlasmaTechnology GmbH, Germany) modified with a hollow cathode (Chamber 1), a generator with a varying frequency of 40 kHz to 100 kHz at a maximum power of 300 W and a UV-VIS spectrometer (BroLight BIM-6002A-01, Hangzhou Brolight Technology Co. Ltd., China). The chamber was connected to the 6-way cross reaction chamber (Chamber 2) as described in section 2.1. The chambers were evacuated until a starting pressure of 4 Pa was achieved in Chamber 1. Using a MFC, 50 sccm of oxygen (Linde AG, Germany, quality 5.0) was supplied to Chamber 1. After the pressure stabilized, plasma was ignited in Chamber 2 with a constant power of 300 W. Plasma was ignited in Chamber 1 about 2 seconds later, where a PTFE body (ElringKlinger Kunst-stofftechnik GmbH, Germany) was placed inside of the hollow cathode. Plasma power in Chamber 1 was controlled with the closed loop system introduced before, using a simulated PID controller and the emission spectrum in Chamber 1, maintaining a stable intensity at 288.8 nm. After about 12 minutes, plasma was turned off in Chamber 1, while the measurements in Chamber 2 continued before plasma was turned off in Chamber 2 after a total of 13 minutes.
3.3 Results
The results of the measurements carried out are shown in Figure 5.
Fig. 5: Intensity measured at 288.8 nm in Chamber 2 with transient responses and mean intensity in steady state
When plasma was turned on in Chamber 2, the mean intensity measured at 288.8 nm before igniting plasma in Chamber 1 at 1.8 min was 2,963 a.u. as ground level signal with a standard deviation of 25 a.u., resulting in a coefficient of variation of 0.86 %. When plasma was ignited in Chamber 1, the transient response in Chamber 2 peaked at an intensity of 5,055 a.u., overshooting due to PID settings before declining within 5 min to a steady state with a mean value of 3,433 a.u.. Standard deviation during the steady state was 28 a.u., leading to a coefficient of variation of 0.81 %.
After turning plasma off in Chamber 1, the transient response in Chamber 2 took place within 9 s. The mean intensity in Chamber 2, after plasma was turned off in Chamber 1, was at 2,988 a.u., an increase of 0.8% compared to before plasma was first ignited, with a standard deviation of 26 a.u., resulting in a coefficient of variation of 0.86 %.
The associated autocorrelation plot of the intensities after 7 minutes until the second transient response is shown in Figure 6.
Fig. 6: Intensity measured at 288.8 nm during steady state with the associated autocorrelation
3.4 Discussion
Before the plasma in Chamber 1 was turned on, the measurements show the ground level signal in Chamber 2 including measurement noise inherent to OES. The noise is described in absolute and relative terms with the standard deviation and the coefficient of deviation respectively. While the absolute noise increased about 10 % when plasma in Chamber 1 was turned on compared to off, the measured mean increased by 15 %, leading to a reduction of the relative deviation expressed by the coefficient of deviation changing from 0.86 % to 0.81 %. The overshoot originating from PID settings with a high proportional term is expected to have no significant impact due to the high inertia of the etching process while reducing the duration of the transient response.
The least squares linear regression through the steady state shown in Figure 5 (top) shows a slight decline with a slope of -0.05, which equals a decrease of 0.5 % from 3,441 a.u. to 3,425 a.u. in intensity before plasma in Chamber 1 was turned off, which is still within the standard deviation.
The associated autocorrelation plot in Figure 6 (bottom) shows significant spikes only at 10 and 11 lags, suggesting little to no autocorrelation. Therefore a standard Dickey-Fuller test can be carried out to test the null hypothesis of a unit root within the steady state against the alternative hypothesis of the steady state being stationary. The null hypothesis is rejected at a significance level of 0.001.
4 Conclusions
In conclusion this study shows the feasibility of a closed loop controlled system for plasma etching with a solid precursor using OES. The results showed that while the least squares linear regression analysis of the steady state data in Chamber 2 included a slight slope, the measurements were well within standard deviation. With the autocorrelation analysis revealing minimal spikes at 10 and 11 lags, little to no autocorrelation is expected in the set-up. With this data the steady state was proven to be stationary in a Dickey-Fuller test at 99.9 % confidence.
References
[1] Wolfram Kintzel et al.: Novel method for plasma etching of printed circuit boards as alternative for fluorocarbon gases; in: International Journal of Plasma Environmental Science and Technology 17.2 (2023), e02002; doi: 10.34343/ijpest.2023.17.e02002; url: http://ijpest.com/Contents/17/2/e02002.html
[2] Ronald K. Hanson, R. Mitchell Spearrin, Christopher S. Goldenstein: Spectroscopy and Optical Diagnostics for Gases; 2016; Springer International Publishing, Cham; ISBN: 978-3-319-23251-5. doi: 10.1007/978-3-319-23252-2
[3] Regine Grafe: Umweltwissenschaften für Umweltinformatiker, Umweltingenieure und Stadtplaner; Lehrbuch, 1. Auflage, 2018; Springer Vieweg, Berlin, Germany; ISBN: 978-3-662-57746-2
[4] Douglas A. Skoog, F. James Holler, Stanley R. Crouch: Principles of instrumental analysis; Seventh edition, 2018; Cengage Learning, Australia; ISBN: 978-13-3746-803-9
[5] A. Kolpaková, P. Kudrna, M. Tichý: Study of Plasma System by OES; in: 20th Annual Conference of Doctoral Students, Charles University; Matfyzpress, Praha 2011, pp. 180–185; ISBN: 978-80-7378-185-9
[6] Alexander Kramida, Yuri Ralchenko: NIST Atomic Spectra Database, NIST Standard Reference Database 78, 2023; doi: 10.18434/T4W30F
[7] Tetiana Stadnytska: Deterministic or Stochastic Trend; in: Methodology 6.2 (2010), pp. 83–92; ISSN: 1614-1881; doi: 10.1027/1614-2241/a000009
[8] Rizwan Mushtaq: Augmented Dickey Fuller Test; in: SSRN Electronic Journal (2011). doi:10.2139/ssrn.1911068.
DOI: 10.7395/2024/Schmid1
* Corresponding author: Marvin Schmid,
email: marvin.schmid@hfu.eu
