Nondispersive ultraviolet monitoring of pulsed H₂S gas delivery

Hydrogen sulfide (H₂S) is used as a sulfur source in the vapor phase deposition of sulfur-containing thin films such as transition metal dichalcogenides^1–4 and chalcogenide glasses,^5,6 some of which are candidate materials for future transistor channels and nonvolatile phase change memory. Unlike the low-volatility metal precursors used in chalcogenide film deposition processes, the chalcogen source (e.g., H₂S and H₂Se) is a compressed gas requiring pressure reduction and flow control components downstream of the source. In a typical gas delivery system, H₂S is supplied from a high-pressure cylinder equipped with a pressure regulator, a downstream metering valve or critical orifice, and an injection valve. Operating conditions such as upstream pressure, metering valve conductance, and injection pulse width all contribute to the time-varying output of the H₂S delivery system. For continuous deposition processes where a steady-state flow of H₂S can be established, H₂S concentration and flow rate may be conveniently measured employing a mass flow meter using a sensing element based on heat loss or Coriolis forces. In processes that involve pulsed injections of H₂S, for instance, pulsed chemical vapor deposition and atomic layer deposition (ALD), measurement of H₂S delivery system output requires time resolution significantly shorter than the injection pulse width. Pulse widths on the order of 0.1–1.0 s are common for ALD processes, giving rise to pressure and flow transients during each injection. Commercial thermal mass flow meters achieve sampling rates of 10 Hz or lower, making it difficult to characterize the pulse profiles of H₂S under typical injection conditions. Further, the pressure drop introduced by the flow meter itself can alter the injection profile of a pulse. Therefore, a noninvasive, fast, and direct measurement of H₂S concentration and flow rates is needed for the characterization and monitoring of pulsed processes such as ALD. Optical absorption spectroscopy meets these requirements and is well-suited for in situ gas sensing in deposition processes. Our prior work has focused on the delivery monitoring of low-volatility organometallic precursors using nondispersive infrared (NDIR) measurements.^7,8 This work demonstrates the application of a nondispersive ultraviolet (NDUV) gas analyzer for the characterization of pulsed H₂S delivery from a high-pressure source.

Optical absorption measurements are commonly used in the monitoring of environmental pollutants, toxic gases, and indoor air quality.⁹ While NDIR is the more commonly used technique due to its high selectivity to gases of interest, NDUV instruments have also found use in gas detection.^10–13 In environmental applications, H₂S is challenging to measure using NDIR and NDUV due to interference from species such as SF₆ in the infrared and CS₂, SO₂, and mercaptans in the ultraviolet region.^14,15 Various methods have been developed to prevent interference in gas mixtures, such as wavelength modulation measurements or the use of multivariate optical elements.^15,16 Interference is less of a concern in semiconductor gas delivery applications where ultrahigh purity gases are delivered in dedicated gas lines. In this case, a direct absorption measurement using NDUV can provide low-cost and high-speed monitoring of gas pulsing. A typical NDUV measurement setup consists of a light source, an optical cell containing the sampled gas, and a detector. Selectivity to the target gas is provided by matching either the light source output or the detected wavelengths to characteristic absorption bands of the analyte. Quantitative measurements are obtained by calibrating sensor response against a known concentration or partial pressure of analyte. The use of nondispersive gas sensors in semiconductor processes has largely focused on steady-state flow measurements of metalorganics during continuous processes such as vapor phase epitaxy.¹⁷ Instruments making use of NDUV sensing have also been demonstrated for the steady-state monitoring of trimethyl aluminum delivery¹⁸ and NO₂ flow visualization.¹⁹ We have taken a similar approach and employed a high-speed NDUV absorption measurement to characterize H₂S concentration and flow rates under fast pulsing conditions used in ALD. The sensor is based on the H₂S absorption band centered at ≈195 nm and a bandpass filter that selects for this feature.²⁰ The NDUV gas sensor is installed on an H₂S delivery line consisting of pressure reduction and flow control elements representative of ALD reactors incorporating high-pressure gas sources.^4,21 We evaluate sensor response linearity and detection limit using an in-place calibration procedure. Using the calibrated sensor, we demonstrate the use of the sensor during pulsing to characterize H₂S partial pressure and mass flow rate as a function of delivery system parameters such as injection timing, upstream pressure, and the flow coefficient of a metering valve. The uses and limitations of the measurement approach are discussed with respect to dynamic measurements of gas delivery for pulsed processes.

A dilute mixture of research grade H₂S in argon with an H₂S mole fraction $( x H 2 S)$ of (3.01 ± 0.06)% was used as the H₂S source. The H₂S component of the gas mixture had a specified purity of 99.9% with $x O C S≤0.06%$⁠, $x C S 2≤0.02%$⁠, $x C O 2≤0.01%$⁠, $x N 2≤0.01%$⁠, and $x hydrocarbon≤0.01%$⁠. A point-of-use purifier installed downstream of the H₂S cylinder was used to remove moisture, metals, and particles. Ultrahigh purity Ar was used as a carrier gas during all measurements and was passed through a purifier to remove moisture, H₂, CO₂, O₂, CO, organics, acids, bases, and metals.

A simplified schematic of the flow system used in this work is shown in Fig. 1.

The upstream portion of the gas delivery system consisted of 4.572 mm inner-diameter electropolished 316L stainless-steel tubing. The carrier gas flow rate was 100 cm³/min at standard temperature and pressure (STP), defined as 0 °C and 101.33 kPa. A pneumatic shutoff valve (PV1) was located downstream of the Ar mass flow controller (MFC). The H₂S delivery system including the gas cylinder was housed in a ventilated cabinet (dashed lines in Fig. 1). A pressure regulator (RV) was used to set the H₂S delivery pressure (P_R) upstream of the injection valving. Electronically controlled pneumatic diaphragm valves PV2 and PV3 were used to define injection, pressure equilibration, and purge timings. An all-metal metering needle valve (NV) installed between PV2 and PV3 was used as an adjustable flow restrictor. During injection, the H₂S mixture flowed into the delivery line by way of a tee junction. Past the tee, the combined gas flow was directed through a three-port pneumatic valve PV4 and the optical flow cell (FC). H₂S could be directed into the reference cell (RC) by opening PV4. A 2.6 kPa full-scale capacitance diaphragm gauge (CDG1) was directly connected to the RC to record cell pressure during sensor calibration. Gas lines downstream of the FC were transitioned to 25 mm inner-diameter stainless-steel tubing where a 13.3 kPa full-scale gauge (CDG2) was installed on a tee to measure downstream pressure. A throttle valve (TV) positioned downstream of the gauge was used to increase the delivery line pressure to ≈0.5 kPa, simulating ALD delivery line pressures under carrier gas flow. Gases exiting the flow system were released through an H₂S scrubber containing high surface area Fe₂O₃ granules and were pumped out using a dry pump.

Gas injection timings were controlled by opening and closing solenoid-controlled valves PV2 and PV3 driven by a digital delay generator. Valve actuation states corresponding to purging, H₂S injection, and delivery line pressure equilibration are summarized in Table I.

When idle, the system is in a purge state with PV1 open and carrier gas flowing. Injection valves PV2 and PV3 are closed during purging and the volume between the two valves contains the source H₂S mixture. After a time delay $( t d)$ of 1.0 s, H₂S is injected for a pulse time of $t p$ by simultaneously opening PV2 and PV3, allowing the H₂S mixture to flow through NV into the tee. At time $t d+ t p$⁠, the upstream injection valve PV2 is closed, allowing the volume between PV2 and PV3 to equilibrate with the delivery line pressure. After an equilibration time $( t e)$ of 1.0 s, downstream injection valve PV3 is closed at time $t d+ t p+ t e$ to return the system to a purge state. The delayed closing of PV3 is intended to reduce the buildup of pressure between the injection valves during purges, which would cause a large pressure transient in the delivery system upon subsequent injection. An alternative design incorporating a vent line between the flow restricting valve and the injection valve was described by Ahmido.²² For all experiments, $t d$ and $t e$ were set to 1.0 s, while $t p$ varied. The cycle time between injections was 10.0 s.

Two separate optical cells were used for gas analyzer calibration and flow measurements. The cells were positioned such that their optical paths were collinear, allowing in-place calibration of the NDUV system. Both cells were made from stainless steel and had a central bore that matched the inner diameter of the upstream gas delivery lines. Pairs of recessed windows made of UV-grade fused silica were sealed to each cell using perfluoroelastomer O-rings. The optical path length between the windows within each cell was 9.2 mm, assuming 25% O-ring compression. For the FC, the optical axis was perpendicular to the gas flow direction. The RC was used only during sensor calibration and was evacuated prior to flow measurements. During measurement, light from a deuterium (D₂) lamp was focused onto a 450 μm core-diameter, 2 m-long, solarization-resistant silica fiber. The output of the fiber was coupled into the optical cells using a pair of f/2 plano convex singlet lenses made of fused silica. Light was focused to a spot between the two optical cells. A matching pair of lenses was used to direct the light exiting the flow cell through a bandpass filter, through another f/2 lens, which focused the light onto a UV-enhanced silicon avalanche photodiode (APD). The detector had an active area of 1.0 mm diameter and a transimpedance gain of 25 MV/A. The bandpass filter had a nominal center wavelength of 214 nm and a full width at half maximum of 25 nm. The measured transmittance of the filter and the molar absorption coefficient of H₂S at 22 °C are shown in Fig. 2. The transmittance peak near ≈185 nm in the filter spectrum may be an artifact of the charge-coupled device used to record the spectrum. Both detector and pressure signals were digitized at 1 kHz using a 24-bit analog-to-digital converter. All measurements were done at an ambient temperature of 22 ± 1 °C.

The carrier gas flow rate and the input mole fraction of H₂S are known and $x H 2 S , o u t$ can be estimated using the H₂S partial pressure measurement at the optical cell and the total pressure $(P)$ measurement at the downstream CDG. A steady-state mass flow rate for H₂S was calculated using Eq. (7) and converting the resulting volumetric flow (at STP) to mass flow $( m ˙ H 2 S)$ at the system temperature and pressure. Time-dependent mass flow rates were not calculated since the absorbance and pressure measurements were located at different points within the flow system. Fast pressure transients that could be observed optically in the flow cell were significantly distorted by the time the injection front reached the downstream CDG. This resulted in an underestimation of instantaneous system pressure at the start of each injection, which translated to unphysically large $x H 2 S , o u t$ values. However, at time points sufficiently far from the pressure transient (t > 2.5 s), steady-state values for $p H 2 S , o u t$ and $P o u t$ could be used to calculate a steady-state $x H 2 S , o u t$⁠. This was the approach taken in this paper. Furthermore, to obtain a steady-state pressure estimate closer to the flow cell, the pressure drop between the downstream CDG and the reference cell CDG was measured at representative pressures and a linear relationship was developed. A correction factor obtained from this analysis was applied to the downstream pressure measurement to approximate $P o u t$ near the flow cell.

The gas analyzer response in the reference cell during a single calibration run is plotted in Fig. 3. The results of linear regression (dashed line) give a calibration factor of (0.454 ± 0.003) m³/mol and an absorbance intercept of 8.09 × 10⁻⁶ ± 1.36 × 10⁻⁵. When the calibration sequence is repeated 16 times, the average value of S is (0.450 ± 0.003) m³/mol. Given in terms of H₂S partial pressure, the analyzer sensitivity is (1.83 × 10⁻⁴ ± 1.22 × 10⁻⁶) Pa⁻¹. Blank absorbance measurements were additionally performed in the absence of H₂S and referenced against an average $I 0$ taken during the first 1.0 s of measurement. The absorbance standard deviation (σ) for a 1.0 s long blank is ≈8.2 × 10⁻⁵. The detection limit of the NDUV analyzer, calculated as 3σ of the blank measurement, corresponds to a H₂S partial pressure of ≈1.4 Pa at a sampling rate of 1 kHz. Analyzer sensitivity and time resolution are both sufficient to characterize H₂S delivery under conditions relevant for pulsed processes such as ALD. For the remainder of this manuscript, we report on the results of a parametric study of delivery conditions and the corresponding H₂S output measured using our NDUV gas analyzer. Adjustable delivery parameters examined here are injection pulse width, H₂S cylinder outlet pressure, and the flow coefficient of the limiting constriction at the delivery system outlet.

Time-resolved H₂S partial pressure measurements and the corresponding total pressure are plotted in Fig. 4. Each trace is an average of eight consecutive injections. Injection pulse widths vary from 0.2 to 3.2 s with the H₂S cylinder regulator pressure (P_R) set to 68.9 kPa (10 psig) and the flow restricting needle valve at the 1.0 turn open position. Injection delay, pulse width, and postinjection equilibration durations are indicated by arrows for the longest injection. The total pressure measurement [Fig. 4(a)] at the downstream gauge shows a smooth increase in pressure ≈0.1 s after the start of the injection, reaching a steady-state value after ≈1.6 s of injection. Pressure starts to decay during the 1.0 s long equilibration as the volume between the injection valves equilibrates with the delivery line pressure. After both injection valves close, pressure continues to decrease as the unswept volume between the downstream gauge and the flow line is evacuated. The H₂S partial pressure measurement shown in Fig. 4(b) differs from the total pressure measurement in several ways. The rate of rise of $p H 2 S$ is sharper than the pressure gauge measurement and reaches a value of ≈65 Pa approximately 0.25 s after the start of the injection. After ≈0.5 s into the injection, $p H 2 S$ sharply drops to a new steady-state value of ≈55 Pa. The first plateau is due to the pressurized volume between the needle valve NV and the downstream injection valve PV3. Evidently, this volume takes ≈0.25 s to drain, after which point the pressure upstream of the needle valve drives flow, resulting in the second plateau. At this point, flow is fully established within the injection system, with the needle valve acting as a critical orifice under choked flow conditions $( P upstream ≫ 2 P downstream)$⁠. Upon closing the upstream injection valve, $p H 2 S$ decreases much more rapidly than the downstream pressure measurement would indicate. At the end of the injection, ≈1 Pa of H₂S is measured immediately before the downstream injection valve closes.

Next, we varied the flow coefficient (C_v) of the needle valve at the exit of the delivery system. This valve had an approximately linear C_v adjustment range over 0.1–2.0 turns of a vernier micrometer handle. Measurements from experiments spanning this C_v range are shown in Fig. 5.

Upstream regulator pressure and injection pulse width for these measurements were fixed at 68.9 kPa and 1.6 s, respectively. The downstream pressure traces in Fig. 5(a) show a slow rise followed by a plateau after ≈1.0 s of injection. The amplitude of the pressure signal increases by a factor of ≈9.7 as the needle valve is opened from 0.1 turns to 2.0 turns. Fast transients associated with the developing flow are not apparent in the total pressure data as the gauge is too far downstream of the injection point. However, the optical measurement of $p H 2 S$ in Fig. 5(b) reveals transients associated with flow development. As before, two regimes are apparent in the data. A first maximum in $p H 2 S$ is reached shortly after the start of the injection driven by the pressurized volume between the needle valve and the downstream injection valve. Once this volume equilibrates with the delivery line, $p H 2 S$ drops to its steady-state value where flow is established across the needle valve. As the needle valve is incrementally opened, the point at which steady-state flow is established shifts to the right. This trend is shown by the dashed arrow in the inset [Fig. 5(c)]. A larger flow coefficient (or conductance) permits higher pressures to develop at the point of injection. For a fixed equilibration time (t_e = 1.0 s), the trapped volume between PV2 and PV3 contains a higher density of gas at the end of each injection. Therefore, at the start of the subsequent injection, the time required to equilibrate this volume with the delivery line is longer. This could explain the positive shift in the initial pressure transient as a function of needle valve position.

Similar trends are evident in experiments where the H₂S pressure regulator output is varied. Figure 6 shows total pressure and partial pressure data for injections with an upstream cylinder outlet pressure $( P R)$ range of 17.2–96.5 kPa. For these injections, the pulse width and needle valve opening were fixed at 1.6 s and 0.4 turns, respectively.

The total pressure profiles in Fig. 6(a) show similar time-varying features as the series of experiments where the needle valve opening was varied. The amplitude of the pressure traces scale with upstream pressure, but the gain associated with upstream pressure is weaker than that observed for flow coefficient adjustment. Here, the maximum total pressure increases by ≈1.6× when the upstream pressure is raised by a factor of ≈5.6. This scaling is readily discerned in the total pressure measurement because H₂S is delivered from a dilute source, with Ar constituting most of the injected gas. An equivalent H₂S injection from a 100% source would result in a small deflection in steady-state P in the delivery line, which would be challenging to detect using a pressure gauge alone. The optical measurement in Fig. 6(b) shows $p H 2 S$ scale as a function of $P R$⁠, similar to the total pressure measurement. Sensitivity to $P R$ is weaker when compared with the scaling observed with C_v. The positive shift in the time it takes to reach a steady-state value for $p H 2 S$ is shown in the inset [Fig. 6(c)]. As was the case when the needle valve opening was adjusted, increasing $P R$ results in a longer duration transient that occurs at the start of the injection. This could be explained by a higher initial pressure (number of moles) between the injection valves at larger $P R$ values. A variation in initial pressure occurs because the postinjection equilibration time was constant for all experiments.

Using total pressure and partial pressure data presented thus far, the impact of adjustable delivery parameters on H₂S partial pressure and mass flow rate can be evaluated. Figure 7 shows the average values for $p H 2 S$ and $m ˙ H 2 S$ calculated at a steady state between t = 2.50 and 2.65 s.

The first adjustable parameter is the upstream regulator pressure. Panels (a) and (b) show that both $p H 2 S$ and $m ˙ H 2 S$ vary linearly with $P R$⁠. This is consistent with sonic injection (choked flow) where the flow rate is expected to vary linearly as a function of pressure upstream of the flow-limiting element (needle valve). The degree of adjustment afforded by $P R$ is notably small for both partial pressure and flow rate. For a sixfold increase in upstream pressure, we observe an increase of ≈1.9× in $p H 2 S$ and of ≈1.6× in $m ˙ H 2 S$⁠. The second parameter of interest is the flow coefficient of the needle valve. The numerical value of C_v was provided by the valve manufacturer and given in volumetric flow rate of water, as defined in NFPA/T3.21.3 R1-2008 (R2018).²⁵ However, relative changes in C_v are expected to map to changes in vacuum conductance. Therefore, we show changes in $p H 2 S$ and $m ˙ H 2 S$ as a function of a normalized C_v value and the corresponding valve position (top axis, number of turns) in panels (c) and (d). At a fixed upstream pressure of 68.9 kPa, increasing normalized C_v by a factor of ≈6.5 results in an increase in $p H 2 S$ of ≈9.7×. Partial pressure increases nonlinearly at first but becomes linear as the valve is opened past 0.5 turns. In contrast, $m ˙ H 2 S$ shows a similar response up to 0.5 turns but stops increasing thereafter (the shaded region in the figure). Therefore, after 0.5 turns, the H₂S mole fraction stops increasing in the delivery line, while $p H 2 S$ continues to rise. With the needle valve position between 0.1 and 0.5 turns open, measured flow rates of 0.05–0.21 mg/s are within a factor of ≈2.5 of the estimated flow rates (dashed lines) based on choked flow calculations and manufacturer-provided C_v values.²⁶ This is reasonable since the flow does not account for any downstream flow-reducing components and reported C_v values have a typical error of >10%.²⁵ A lack of change in mass flow in response to increasing C_v suggests that beyond a certain needle valve position, a downstream flow element becomes the limiting constriction in the delivery system. This could be the injection valve (PV3) or the throttle valve downstream of the pressure gauge.

The NDUV analyzer and the optical measurements described here provide insights into delivery system design and operation for pulsed processes. A designer may choose to optimize a delivery system for $p H 2 S$⁠, $m ˙ H 2 S$⁠, or both. Control over the partial pressure reaching the wafer surface is important for reaction kinetics, which could be a source of variability in process performance. On the other hand, mass delivery may be a consideration in reactor scaling or changes in wafer load. Adjustable delivery parameters examined here show varying degrees of controllability over $p H 2 S$ and $m ˙ H 2 S$⁠. First, we examine injection pulse width. For a dilute H₂S source, as is the case here, injections lasting several seconds are needed to deliver sufficient mass to the reactor. As shown in Fig. 4, changes in $p H 2 S$ and P stabilize after 1.0 s of injection, after which point total mass delivery $( m H 2 S)$ should scale linearly with injection time. If supplied from a pure source, rather than a dilute mixture, the same mass of H₂S could be delivered with pulse widths that are ≈33× shorter. This would be preferable for high-volume process equipment, where shorter cycle times are desired for tool productivity. However, as the NDUV data show, stable flow in the delivery system is not fully established at t_p < 1.0 s as evidenced by the transients. Therefore, $p H 2 S$⁠, $m ˙ H 2 S$⁠, and $m H 2 S$ are likely to vary nonlinearly with $t p$ for short pulse widths. This, in turn, would make injection system outputs sensitive to uncertainties in injection timing.

Next, we consider the upstream pressure and injection system flow coefficient in terms of delivery system performance. Both parameters provide control over $p H 2 S$ and $m ˙ H 2 S$⁠, as shown in Fig. 7. Adjusting the delivery system conductance (needle valve opening), when compared with upstream pressure, provides a greater degree of control over $p H 2 S$⁠. However, $m ˙ H 2 S$ and $m H 2 S$ do not scale past a certain C_v value in our flow system. Note that a gas delivery system design employing higher conductance downstream flow elements may not exhibit this limitation. Upstream pressure regulation, on the other hand, enables linear control over both $p H 2 S$ and $m ˙ H 2 S$ but provides comparatively weak scaling. The set of delivery parameter choices that a process designer makes will depend upon the desired system output. High-speed and high-sensitivity measurements discussed here provide one means of evaluating parameter response for pulsed delivery applications. Although dilute H₂S is used as a test case for the optical measurements in this work, more common reactants such as NH₃, H₂, and O₂ are also delivered from high-pressure sources in pulsed processes. These gases all require similar pressure reduction and control elements to achieve controlled pulsed injections. Insights gained here from optical measurements are expected to be applicable to pulsed processes employing other high-pressure gas sources.

We described an NDUV gas analyzer for H₂S detection and demonstrated its use in a gas delivery line characteristic of a pulsed process such as ALD. The minimum detectable partial pressure was ≈1.4 Pa (at 22 °C) with a time resolution of 1 kHz. The analyzer was used to monitor hydrogen sulfide partial pressure and flow rate as dilute H₂S was injected into a flowing carrier gas. Flow transients that could not be resolved with a downstream pressure gauge were readily detected by the optical measurement. Steady-state flow development was observed to occur ≈1.0 s after the start of the H₂S injection. Adjustment ranges for H₂S partial pressure and steady-state flow rates were characterized as a function of upstream pressure and the flow coefficient of a downstream metering valve. The partial pressure of H₂S in the delivery line could be tuned by changing upstream pressure or the flow restricting valve opening. The mass flow rate could be controlled using the same parameters, but only a limited range of flow rates was accessible due to flow restrictions downstream of the gas delivery system. For long injections, total mass delivery could be further adjusted by changing the injection pulse width. The optical measurements demonstrated here could aid in the design of process equipment or in process parameter selection for pulsed gas delivery applications.

Certain commercial equipment, instruments, or materials are identified in this paper in order to specify the experimental procedure adequately. Such identification is not intended to imply recommendation or endorsement by NIST, nor is it intended to imply that the materials or equipment identified are necessarily the best available for the purpose.

The authors have no conflicts to disclose.

Berc Kalanyan: Conceptualization (lead); Data curation (lead); Formal analysis (lead); Investigation (lead); Writing – original draft (lead); Writing – review & editing (lead). Evan P. Jahrman: Conceptualization (supporting); Data curation (supporting); Formal analysis (supporting); Investigation (supporting); Writing – review & editing (supporting). James E. Maslar: Conceptualization (supporting); Data curation (supporting); Formal analysis (supporting); Investigation (supporting); Writing – review & editing (supporting).

The data that support the findings of this study are available from the corresponding author upon reasonable request.