As the scale of semiconductor processes continues to shrink, detecting even minute changes in the composition of process gases or by-products becomes imperative. Therefore, identifying gas species and measurement of their exact proportions within a process chamber has become crucial. A residual gas analyzer (RGA), which analyses the gases flowing from a process chamber to an exhaust line, can detect leaks in real time, respond rapidly to process anomalies, and detect the endpoints of etching and cleaning [1–7].

Currently, different analyzers, depending on the method of residual gas analysis, are available. A quadrupole mass spectrometer (QMS)-RGA is used to identify the composition of the gases entering the analyzer. The ionization source of a QMS-RGA converts the gas molecules into ions, and the mass filter separates the ionized gas molecules based on their mass-to-charge ratio. Finally, the QMS-RGA detector captures the ions filtered by the mass-to-charge ratio and determines the concentrations of different gas species. The operational principle of a time-of-flight (TOF)-RGA, which is another type of gas analyzer, is similar to that of a QMS-RGA. However, the mass filter is replaced by a flight tube, which separates the gas species based on their massto-charge ratios and different flight times in the tube before detection by the detector. The time resolutions of TOF-RGAs are better than those of QMS-RGAs. A self-plasma-optical emission spectrometer (SP-OES) is a different type of RGA that identifies gas compositions in a vacuum. SP-OESs consist of a plasma module to excite gas molecules and an optical spectrometer to collect the light photons emitted from the plasma module. The identified wavelength corresponding to a specific element or molecule reflects the gas composition.

Although QMS-RGAs and TOF-RGAs exhibit high detection sensitivities at the ppm level, they require high maintenance costs because of contamination induced during reactive gas monitoring [8–11]. By contrast, SP-OESs exhibit a high contamination resistance and can monitor diverse gas species in real time; therefore, it is widely used in process monitoring [12–15]. According to Jeon et al. [12], during plasma-enhanced chemical vapor deposition, an SP-OES can detect by-products such as SiH4, which primarily affects the reproducibility of the process, by comparing the optical emission spectra before and after processing and help maintain the chamber environment under optimal conditions. Kang et al. [13] suggest that SP-OESs can prevent chamber damage due to excessive cleaning by detecting the endpoint, and Kim et al. [14] used an SP-OES to detect the endpoint, even at a low open ratio of less than 1 %, during etching.

Nevertheless, with the gradually shrinking scale of semiconductor processes, which warrants the detection of minute changes in the compositions of process gases, the sensitivity of SP-OESs must be analyzed and optimized for advanced process monitoring. However, studies on the quantitative evaluation and optimization of the SP-OESs’ detection sensitivity remain scarce.

Therefore, this study was conducted to analyze and optimize various parameters that influence the detection sensitivity of an SP-OES (the SP-OES used in this study was the same as that used in a previous study [15]) using a gas-mixing system. The analysis was focused on the evaluation of the sensitivity’s dependence on a predetermined gas-mixing ratio. The limit of detection (LoD) was measured to quantify the sensitivity of the target SP-OES, and various parameters, including the integration time for photon collection, pressure under the self-plasma module, and selected wavelength were examined to improve the sensitivity of the target SP-OES.

In this study, we constructed a gas-mixing platform, with precisely controlled trace concentrations, to evaluate the detection sensitivity of an SP-OES (the SP-OES used in this study was the same as that used in a previous study; the detailed information is reported elsewhere [15]). As illustrated in Fig. 1, two mass flow controllers (MFC) with different capacities of 10 slm and 1 sccm were installed at the inlet of the mixing bombe for injecting a background gas, N₂, and a trace gas (Ar). The background and trace gases were injected independently into the mixing bombe, with the injection of the trace gas prioritized over that of the background gas to produce a homogeneous mixed gas of a uniform concentration in a short time. The concentration of the trace gas was varied from 0 to 1.8 % by adjusting the flow rate and time, using each MFC and an air valve, to produce the mixed gases. Table I lists the flow rates and times of the background and trace gases injected to realize different concentrations of trace gases. The mixed gas, transferred from the mixing bombe through the piping via a 300-sccm-capacity MFC, exited through the exhaust system, and a fraction of this gas was directed into the SP-OES installed in the piping, where this gas transitioned into a plasma state. The emitted optical signals were transmitted through an optical fiber cable (diameter: 400 μm) and then classified according to their wavelengths by a high-resolution spectrometer to analyze the types of residual gases. The measured wavelengths ranged from 300 to 1,100 nm. The pressure in the piping was adjusted to 0.2−3.5 Torr using a throttle valve. Following the completion of the measurements, the gases remaining inside the mixing bombe were exhausted using a connected diaphragm pump.

Figure 1. Schematic of the experimental setup.

Table I. Flow rates and time required by background and trace gases for preparing the mixed gas..

Concentration (%)	Flow rate of N2 (sccm)	Flow time of N2 (s)	Flow rate of Ar (sccm)	Flow time of Ar (s)
1.80	2,000	30	0.95	1,137
1.40	2,000	30	0.95	884
0.90	2,000	30	0.95	568
0.22	2,000	30	0.95	139
0.18	2,000	30	0.95	114
0.10	2,000	30	0.95	63

3.1. LoD test

In this study, the LoD (defined as the minimum concentration that can be reliably detected with reasonable confidence for a given analytical procedure) was determined to compare the detection sensitivity of the SP-OES under various operational conditions. For the LoD test, N₂ and Ar gases were used as background and trace gases, respectively. The signals of Ar at various concentrations were continuously measured with reference signals, originating from an Ar concentration of 0, for comparison. Throughout the test, the SP-OES remained continuously operational [Fig. 2(a)]. Each signal was measured for 200 s, and the signal intensities measured over a duration of 60 s were averaged to obtain normalized signals, which represented the ratio of the average signal intensity of Ar to that of the reference [Fig. 2(b)]. During the measurements, the SP-OES provided consistent gas signals from high to low concentrations; however, some unstable signals were occasionally observed as well [Fig. 2(a)]. To analyze the impact of these unstable signals on the slope of the calibration curve−a key factor in determining the LoD−we performed three experiments and evaluated the reproducibility of the slope across three calibration curves affected by random unstable signals. The results demonstrated that the slopes exhibited > 95.3 % reproducibility, indicating that the randomly observed unstable signals had a negligible effect on the calculation of the LoD [16]. The calibration curve represented the normalized signal intensities as functions of the trace gas concentration [Fig. 2(c)], and the slope and standard deviation were used to calculate the LoD as:

Figure 2. Signal processing method: (a) Time-dependent signal measurement in the SP-OES with various gas concentrations. (b) Average signal intensities of the mixed gas and reference average signal intensity. (c) Calibration curve for concentration and normalized intensity.

Here, σ_bn and m represent the standard deviation of the reference signal intensity and slope of the calibration curve, respectively. The LoD of the SP-OES reflects the lowest gas concentration that can be distinguished from background noise, including electronic and environmental factors. Using the standard deviation of the SP-OES’s noise level, we obtained a threshold that accurately represented the minimum signal intensity required to confidently distinguish a true SPOES signal from background fluctuations [17,18]. In addition, Eq. (1) indicates that the sensitivity of the SP-OES increases when the slope is steep, and the standard deviation of the instrument noise is small.

3.2. Comparison of the LoD at different wavelengths

Figure 3 illustrates the optical emission spectra measured by the SP-OES for gases consisting of 100 % N₂ and 100 % Ar, and a mixed gas containing 95 % N₂ and 5 % Ar. Several peaks within the wavelength range of 650−900 nm originate from N₂ and Ar. For the mixed gas, several signals corresponding to Ar (low concentration of 5 %) are overshadowed by those of N₂ (high concentration of 95 %). However, strong signals at various wavelengths, such as 750.57, 811.55, and 842.27 nm, are still visible in the spectra of the mixed gases. Figure 4(a) illustrates optical intensities of Ar present in the N₂ + Ar mixed gases, with varying concentrations of 0.1 to 1.8 %. Signals with and without Ar are alternately measured to assess the differences in the peak intensities. Initially (beginning of the measurements), the intensity sharply increases owing to the unstable discharge of the plasma within the SP-OES. As time progressed and the plasma stabilized, the intensity remained steady. The signal intensities in the presence of Ar decrease as the concentration decreases, whereas the reference signal intensities without Ar remain unaffected. Among the signals originating from Ar, the signal intensity at 750.57 nm is the highest. In addition, the reference signal intensities at 811.55 and 842.27 nm are similar, whereas that at 750.57 nm is high. This result implies that owing to the spectrometer’s resolution of 1 nm, the signals can not be distinguished, and the signal at 750.14 nm, which is attributed to N₂, overlaps with that at 750.57 nm.

Figure 3. Comparison of the emission wavelengths of the trace gas (Ar) in the single and (N₂ + Ar) mixed-gas spectra.

Figure 4. Comparison of the trace-gas (Ar) signal intensity shown by the SP-OES at different wavelengths. (a) Time-dependent signal measurement. (b) Normalized intensity with changes in trace-gas concentration. (c) LoD with respect to wavelength.

Figure 4(b) illustrates the calibration curve of the normalized signal intensity, which is the ratio of the Ar signal intensity at each concentration to the reference signal intensity. For all the emission wavelengths of Ar, the normalized signal intensity increases as the concentration of Ar increases and exhibits strong linearity with the concentration; the R2 values at 750.57, 811.55, and 842.55 nm are 99.8, 99.6, and 99.6 %, respectively. Unlike the signal intensities shown in Fig. 4(a), the normalized signal intensity is highest at 811.55 nm and the lowest at 750.57 nm. Figure 4(c) illustrates the LoD for each Ar wavelength, which is calculated from the slope of the calibration curve and standard deviation of the reference signal intensity. The LoD at 750.57 nm is the highest (951 ppm) and that at 811.55 nm is the lowest (234 ppm). However, despite the highest signal intensity at 750.57 nm in the optical emission spectrum of Ar, the detection sensitivity is the lowest owing to interference from the N₂ signals. In contrast, although the signal intensities at 811.55 and 842.27 nm are lower than those at 750.14 nm, their detection sensitivities are higher owing to the absence of nearby wavelengths associated with N₂. Furthermore, the LoD at 811.55 nm is lower than that at 842.27 nm, indicating that when both wavelengths experience minimal interference from background gases, a high detection sensitivity is detected at the high-signal-intensity wavelength.

3.3. Comparison of the LoD with pressure

Figure 5(a) illustrates the optical emission signal intensities of Ar measured using SP-OES when the pressure was varied from 0.2 to 3.5 Torr. For each pressure, the concentration of Ar was varied from 1.8 to 0.1 %, and the Ar wavelength of 811.55 nm, which had the highest LoD, was selected. At 0.2 Torr, the difference between the Ar signal intensity and the reference signal intensity was visually distinguishable; however, as the pressure increased to 3.5 Torr, the difference between the two signal intensities became indistinguishable. Furthermore, during gas exhaust at pressures above 1.5 Torr, the signal intensity momentarily increased significantly before decreasing because the SP-OES passed through the pressure range where a strong emission occurred as the pressure decreased.

Figure 5. Comparison of the trace-gas (Ar) signal intensity shown by the SP-OES at different pressures. (a) Time-dependent signal measurement. (b) Normalized intensity with change in trace-gas concentration. (c) LoD with respect to pressure.

As illustrated in Fig. 5(b), the normalized signal intensities of Ar increase with increasing concentration, regardless of the pressure. The corresponding R2 values of 99.6, 99.8, 99.1, 94.1, and 92.7 % at 0.2, 0.5, 1.5, 2.5, and 3.5 Torr, respectively, indicate strong linearity between the measured parameters. The normalized intensity was the highest at 0.2 Torr, decreasing as the pressure increased. Figure 5(c) shows the LoD, derived from the slope of the calibration curve in Fig. 5(b), and standard deviation of the reference signal intensity, as a function of pressure. Similar to the trend observed for the optical emission signal intensities, the LoD of Ar increases as the pressure increases. When the pressure changes, the electron density, electron temperature, and electron energy distribution function of the plasma also change, indicating that the number of electrons satisfying the minimum energy required for ionization and excitation varies accordingly [19,20]. These results indicate the presence of an optimum pressure to enhance the trace-gas detection sensitivity of SP-OESs. Therefore, depending on the target vacuum pressure under the gas sample, the discharge pressure and electrode structure of the SP-OES plasma module must be optimized to discharge plasma efficiently for exciting the gas species.

3.4. Comparison of the LoD with integration time

Figure 6(a) illustrates the optical emission signal intensities of Ar measured by the SP-OES for different integration times of 40 to 640 ms. At each integration time, the concentration of Ar is varied from 1.8 to 0.1 %, and a wavelength of 811.55 nm, which is associated with the highest LoD, is selected. The signal intensities obtained with and without Ar increase with the integration time.

Figure 6. Comparison of the trace-gas (Ar) signal intensity shown by the SP-OES for different integration times. (a) Time-dependent signal measurement. (b) Normalized intensity with changes in trace-gas concentration. (c) LoD with respect to integration time.

As shown in Fig. 6(b), the normalized signal intensities of Ar increase with the increasing concentration, irrespective of the integration time. The corresponding R2 values (99.9 % at 40, 160, and 320 ms; 99.8 % at 640 ms) indicate strong linearity. The normalized intensity was the highest at 640 ms, decreasing as the integration time decreased. Figure 6(c) shows the LoD, determined from the slope of the calibration curve in Fig. 6(b), and standard deviation of the reference signal intensity, as a function of the integration time. Similar to the trend observed for the optical emission signal intensities, the LoD of Ar decreases as the integration time increases. The electric circuit of the spectrometer regulates the light intensity by adjusting the integration time. We assume that the detection sensitivity is enhanced because of the large amount of information on the trace gases collected within the same sampling duration. However, a long integration time can restrict real-time monitoring, and therefore, setting an appropriate integration time is crucial for utilizing SP-OESs in real-time monitoring.

In this study, the sensitivity of an SP-OES was measured under various conditions for the detection of trace gases. To quantify the gas detection sensitivity, a platform capable of preparing mixed gases with precise concentrations was developed, and the LoD was determined. Even for the same type of trace gas, the gas detection sensitivity varied with wavelength. The optical signal intensity and selecting a wavelength that did not interfere with the background gas were deemed crucial for detecting trace gases and performing efficient data processing to increase the sensitivity of the SP-OES. In addition, the detection sensitivity varied with pressure, with an optimum pressure determined as a necessary condition for enhancing the sensitivity of the SP-OES. The findings of this study indicate that depending on the target vacuum pressure under the gas sample, the discharge pressure and electrode structures of the SP-OES’s plasma module must optimized to discharge an efficient plasma for exciting the gas species. In addition, increasing the integration time leads to an increase in the optical signal intensity, which consequently enhances the detection sensitivity of the SP-OES. Thus, we can conclude that the sensitivity of an SP-OES depends on controllable parameters and that wavelength selection techniques are essential for utilizing SP-OESs in semiconductor process monitoring in advanced applications.

This work was supported by the Korea Institute of Machinery & Materials (KIMM/NST) Institutional Program (NK248E) and the Industrial Strategic Technology Development Program (RS-2023-00264860, RS-2023-00221595), funded by the Ministry of Trade, Industry, and Energy (MOTIE, Korea) and Ministry of SMEs and Startups (MSS, Korea).

The authors declare no conflicts of interest.