Design and characterization of a microreactor for spatially confined atomic layer deposition and in situ UHV surface analysis

Atomic layer deposition (ALD) is a technique for depositing thin films by exposing a substrate to alternating pulses of a thin film precursor and a coreactant, with purge steps in between.¹ Due to the self-limiting nature of the ALD process, one can precisely determine the thin film thickness at an atomic level by controlling the number of ALD cycles. ALD has been employed to deposit ultrathin films of many types of materials, including oxides, nitrides, and noble metals.² ALD also has the advantage of depositing conformal, ultrathin films on the surfaces present on 3D features, such as trenches and fins with high aspect ratios. Because of these and other features, ALD has been widely applied in semiconductor manufacturing for several different applications, such as high-k dielectrics,^3,4 diffusion barriers for interconnect,⁵ and spacers in double patterning.^6,7

ALD, similar to other technologically important processes such as chemical vapor deposition (CVD) and plasma etching, is typically performed under medium vacuum conditions, or pressures ranging from 10⁻³ to 10 Torr. In contrast, many of the most powerful surface characterization methods, such as electron and ion based spectroscopies and diffraction techniques, are performed under ultrahigh vacuum (UHV, p < 10⁻⁹Torr) conditions. Conducting surface analyses postdeposition and ex situ is always an option; however, this approach is susceptible to the occurrence of undesirable reactions such as surface oxidation, reactions with adsorbed species, and contamination. This is particularly true in the case of the ultrathin films one most often seeks to deposit using ALD, where postdeposition reactions occurring upon exposure to atmospheric air could totally obscure the composition and microstructure the deposited thin film. This situation is entirely analogous to the so-called “pressure gap” that has plagued the use of surface science approaches to examine reactions occurring in heterogeneous catalysis.⁸ Recent innovations in spectrometer design have resulted in techniques such as near ambient pressure x-ray photoelectron spectroscopy (XPS), where ALD has been examined in situ at partial pressures of ∼7.5 × 10⁻²Torr.⁹ Use of this technique at present, however, essentially requires the use of synchrotron x-ray radiation, thus significantly limiting its impact.

Historically, to overcome the pressure gap between the process and characterization conditions, two main approaches have been developed that involve employing separate chambers/zones for “high” pressure reaction and/or deposition and UHV surface analysis in a single apparatus.¹⁰ One approach is to “bring the substrate to the gas source,” where the substrate is transferred to and from the reaction chamber via a sample transfer mechanism to and from the analysis chamber for surface characterization.¹¹ An early example of this approach was provided by Goodman et al.¹² Another approach is to “bring the gas source to the substrate”: a translatable cuplike device, typically fitted with integrated lines for the introduction of reactive gases, makes contact with the substrate/substrate holder to form a small volume that acts as a small chemical microreactor, with the substrate surface providing one of the walls of the reactor. In this scheme, the microreactor is evacuated after the completion of the reactions, and surface characterization can be conducted once the chamber is brought back to UHV. Somorjai and coworkers were the first to utilize this design to study heterogeneous catalysis,^13,14 and more recently, George and coworkers applied this type of design to investigate ALD.^15–17 

In addition to the issue of the design of the reaction chamber/zone, a number of probes have been employed for in situ characterization of the thin films deposited by ALD, including techniques such as spectroscopic ellipsometry,¹⁸ XPS,^19–21 and x-ray fluorescence.²² Rubloff and coworkers have described a system that utilizes a microreactor enclosed in an UHV chamber, similar to those employed by Somorjai and coworkers, where the gases leaving the reactor are sampled using quadrupole mass spectroscopy.^23,24

We present here the design and characterization of a microreactor that draws from both of the two main strategies discussed earlier: it uses a cuplike device to define the reaction zone inside of a larger chamber, and this larger chamber is coupled by a gate valve to another chamber possessing tools for UHV surface analysis. In our case, an important criterion was that the design must be compatible with an existing sample holder and manipulator. In addition, another criterion we have invoked is that the design must not involve a physical seal (i.e., a gasket) to define the reaction zone, which essentially all previous designs have employed. The design we present here exhibits strong similarities to spatial ALD print head designs,^25,26 where the reaction zone is defined by the use of a curtain gas to limit the out-diffusion of reactive species into the surrounding larger chamber. In this paper, we will first provide a detailed description of our surface analysis system and how the probe is incorporated into the chamber and employed in experiments. We will describe a selected set of simulations at relevant reaction conditions using computational fluid dynamics (CFD) to evaluate the predicted performance of the microreactor probe. Finally, to demonstrate the functionality and performance of the microreactor system, we have conducted a gas–surface reaction between I₂(g) and Cu and ALD of thin films of ZrO₂ deposited on SiO₂ substrates, where the thin films have been analyzed in situ and without air break using XPS.

The microreactor we describe here has been incorporated into an existing custom-designed, multiple-stage stainless steel UHV chamber as illustrated schematically in Fig. 1(a), which we have described in detail previously.^27,28 For the experiments described here, the intermediate chamber plays an important role, as it is coupled to the main UHV surface analysis chamber via a gate valve, a load-lock chamber via a gate valve, and the microreactor, which can also be isolated in its own chamber by a gate valve. The main analysis chamber (volume ∼22 l) is pumped by both a 400 l s⁻¹ magnetically levitated high-throughput turbomolecular pump and a titanium sublimation pump. A base pressure of ∼9 × 10⁻¹⁰Torr is routinely achieved after a 24–36 h bakeout at ∼150 °C. The sample manipulator, mounted on the intermediate chamber, is designed to accept sample holders that can accommodate either full 100 mm wafers or smaller “coupon” size (e.g., 1.8 × 1.8 cm²) substrates. The substrates are introduced using a magnetically coupled transfer arm mounted to the load-lock chamber, which is separately pumped by a 60 l s⁻¹ turbomolecular pump. The substrates are mounted on a precision sample manipulator that allows for motion of the sample along three linear axes as well as polar and azimuthal rotation. Translation of the sample along the long axis of the manipulator takes the sample from the intermediate chamber to the main analysis chamber. The substrates are heated radiatively by a pyrolytic boron nitride heating element that is incorporated in the manipulator and possesses a maximum power output of 3 kW. This sample stage can heat Si substrates to a peak temperature of 1200 °C and continuous working temperatures of 1000 °C, although much lower substrate temperatures are employed for the experiments we describe here.

The main analysis chamber houses a concentric hemispherical electron energy analyzer, which serves as an analyzer for both XPS (photoelectrons) and low-energy ion scattering spectroscopy (scattered ions).^29,30 A twin anode Mg/Al x-ray source and an ion source (SPECS ion source IQE 12/38) are also located in the main chamber. The main analysis chamber is also coupled to a doubly differentially pumped supersonic molecular beam, and there is also a quadrupole mass spectrometer mounted on this chamber that can analyze both the molecular beam itself and residual gases in the chamber.²⁷ 

As shown in Fig. 1(b), the microreactor probe consists of a precision machined head, a fluidic extension piece, and a fluidic and electrical feedthrough. This assembly is mounted on a linear translator, which affixed to a gate valve, is attached to the intermediate chamber [for clarity the translator and gate valve are not shown in Fig. 1(a)]. A key goal is to minimize any possible contamination to the UHV chamber when the probe is not in use (e.g., due to outgassing), which is achieved in our design by retracting the probe into its own subchamber and closing the gate valve. For experiments involving thin film deposition, the microreactor probe is translated into the intermediate chamber where it combines with the sample holder to form the reaction zone. The exact geometry of this reaction zone (and the probe head) is critical concerning the performance of the microreactor, and we consider these issues in Sec. III B, where we describe results from CFD calculations designed to optimize these dimensions. In the actual construction, the optimized values were appropriately modified by considering limitations associated with machining and other practical constraints.

The microreactor probe consists of several components, as indicated earlier. The mounting flange has five fluidic feedthroughs for gas inlets or exhausts, and four mini flanges for electrical feedthroughs (e.g., for heating elements and thermocouples). One of the feedthroughs, intended for use as a gas inlet and lying on the central axis of the probe, can double as a viewport, providing optical axis to the substrate surface. The five fluidic feedthroughs are connected to the probe head using an extension piece (∼36 cm long). The probe head itself is a cylindrical machined piece of stainless steel, ∼5.6 cm in dia., and 2.5 cm in height. At the end of the probe head, the central fluid line expands into a cylindrical space that is combined with two other fluid lines through slanted channels inside of the probe head, which is shown in the horizontal cross-section in Fig. 1(c). The cylindrical space, along with the substrate surface, defines the reaction zone that has a height above the substrate surface of ∼0.5 cm, and a diameter of ∼1.5 cm, equivalent to a volume of ∼1 cm³. These three gas lines provide the ability to control the introduction of three independent species to the reaction zone. The final two gas lines are connected to two peripheral semicircular grooves (tracing an arc of 150°) that are machined into the end of the probe, and these are ∼1.5 cm from the centerline, ∼0.3 cm wide. The two lines act together to provide the exhaust lines for the microreactor, which is shown in the vertical cross-section in Fig. 1(c). Four physical “stops” are located on the surface of the end of the probe head, and these act to ensure a precise gap (∼0.05–0.1 cm) between the probe head and the substrate surface. The microreactor system is constructed entirely of bakeable materials so that it can undergo a long bake (>24 h) at ∼150 °C. A closed-loop feedback control system is employed to control the pressure in the reactor during thin film deposition, which includes the main process mechanical pump, a capacitance monometer, and an exhaust throttle valve.

We have designed a gas delivery system that can deliver the reactive species to the microreactor and provide the needed performance for ALD, including fast switching between reactive species and inert purge gases, and this is displayed in Fig. 1(d). All gases (reactive and inert) are connected to the microreactor feedthrough using stainless steel tubes and VCR connections. In all cases, we make use of a “vent-run” configuration for both the reactive and purge streams, where each stream flows continuously, but the two streams alternate in sequence between being directed to a microreactor inlet and a bypass line to a pump. The volumetric flows of the reactive and purge streams are essentially the same, which minimizes any fluctuations of the reactor pressure. Each “toggling” network is comprised of four pneumatic valves driven by fast-switching solenoids (response time of 1.5–3.4 ms). An input/output board (National Instruments cDAQ-9174) with a power source sends 24 V DC signals to solenoids, as determined by a computer-programmed dosing sequence. The delivery lines of the reactants are heat-traced from the outlet of the bubbler (if present) to the microreactor probe feedthrough. In the case of vapors delivered by an inert carrier gas, these lines are heated to a temperature of ∼20 °C higher than the temperature used for the bubbler.

We have conducted simulations of the microreactor probe at relevant reaction conditions using CFD to model the flow, diffusion, and heat transfer. As indicated earlier, confinement of the reactive gases to the reaction zone is an important design goal, and it will depend on both geometric factors and the reaction conditions (flow and pressure). The geometry of the reactor-substrate system was reduced to a two-dimensional, axisymmetric coordinate system to reduce the computation time. The simplified model consists of a cylindrical planar substrate (thickness ∼0.076 cm) and a cylindrical sample heater (thickness ∼0.51 cm) that are separated by a gap of ∼0.25 cm. Both of these have a radius of ∼5.1 cm. The microreactor probe head (outside radius ∼3.25 cm and height of ∼2.1 cm) includes one common cylindrical central feed (inside radius ∼0.6 cm) for all input species (reactants, carrier, and purge) and an annular exhaust channel (inside radius of 1.2 cm and outside radius of 1.4 cm). There is a cylindrical chamber (∼14.4 cm in height, ∼9.5 cm in radius) that surrounds the substrate holder and microreactor probe head, which represents the intermediate chamber on our actual system. There are two inlets for an inert gas inlet on this surrounding chamber, which represent the injection points of the curtain gas. One of these is at the top of the surrounding chamber, on-axis, with an opening radius of ∼1.0 cm. The other is an annular region with an inner radius of ∼3.25 cm, and an outside radius of ∼4.25 cm. In this approach, the curtain gas flows radially inward between the small gap (∼0.05–0.1 cm) between the substrate surface, and flows out the annular region in the probe head to the exhaust, along with the reactant and purge flows.

We carried out CFD calculations based on the finite element method, which involved numerically solving all the governing equations that describe the physics, namely, multicomponent, nonisothermal, compressible flow. The domain was discretized into many triangular subdomains in which simple trial functions are used to approximate the true function, and the numerical solution was found across the whole domain at any given time with proper boundary conditions. Heat and mass transport and fluid flow are coupled due to the composition dependence of the viscosity and the thermal conductivity, and the temperature and pressure dependence of the diffusivity.

We first carried out CFD calculations under the assumption that the flow is time-independent, seeking to understand the behavior of the flow at the steady state, using the dimensions given earlier. In these simulations, heater power was set to produce a substrate temperature of 150 °C, which is within the process temperature “window” for a number of ALD chemistries. The flow rate of the reactant gas (assumed to be pure O₂) through the common central tube was set to 10 sccm, while the flow at the two inlets for the curtain gas (assumed to be pure N₂) were set to be 50 sccm each. The reactor pressure was set to be at ∼1 Torr by fixing the exhaust pressure at the end of the effluent groove to be exactly 1 Torr.

In this calculation, we sought to determine the effectiveness of reactant delivery to the substrate surface, and the confinement of these reactants to the central reaction zone, with minimal out-diffusion of reactants to surrounding subchamber. We first examined the effect of the dimension of the gap between the probe head and the substrate surface. An examination of the pressure distribution across the simulation cell revealed a pressure drop of ∼0.5 Torr associated with the flow of the curtain gas inward from the edge of the microreactor probe head to the annular exhaust groove. A somewhat larger pressure drop (∼0.6 Torr) was associated with the flow of the reactant outward from the reaction zone to the annular exhaust groove. Concerning the temperature distribution, we found that it was uniform across the substrate surface (∼150 °C), apparently unaffected by the flow of gases, even though the temperatures of the input reactant and carrier gases were set to 25 °C. A calculation of the mole fractions of each species indicated excellent segregation of the two components. For example, we calculate a spatially averaged mole fraction for the reactant gas, O₂, of 0.84 at the substrate surface. In contrast, the spatially averaged mole fraction for the reactant gas at the walls of the surrounding chamber of 7.4 × 10⁻⁴, which corresponds to a 3 order-of-magnitude decrease in the amount of the reactant reaching those surfaces. These steady-state concentrations of O₂ represent a worst case scenario in terms of using this probe for ALD, as in ALD typically the reactant pulses are short, followed by much longer purge pulses. Thus, we expect that the concentration of reactant species reaching the walls of the surrounding chamber walls will be much lower in a transient mode, as compared to steady state.

We also examined the effect of the shape of the surface of the probe head, and the placement of the annular region. In particular, we have considered three cases concerning shape variations: (1) a flat planar end surface, (2) one where the surfaces are dished in a way such that the gap is greatest at the annular exhaust region, and (3) one where the surfaces are dished such that the gap is smallest at the annular exhaust region. Of the three cases, the design where the gap is the largest at the annular exhaust region, and smallest at the edge of the probe and the edge of the reaction zone produced the best results in terms of minimizing the out-diffusion of the gaseous reactant. We also examined the effect of the placement of the annular region, by varying the relative length of radial travel from the reaction zone to the annular region, and from the annular region to the edge to the probe, where we considered the ratio of these lengths to be 1:3, 1:1, and 3:1. We found that the first of these produced the best result in terms of minimizing the out-diffusion of the gaseous reactant.

In order to examine the effectiveness of the microreactor for operating conditions more representative of ALD, we have conducted CFD calculations under transient conditions. A two-step calculation was used to simulate the reactor flows for multiple cycles of ALD. The first step calculates the steady state that is reached for the flow of single-component species (e.g., the inert purge and curtain gases), which acts to establish the initial temperature distribution (including that of the substrate) and the reactor pressure. During this first step, no reactant is introduced, so it is equivalent to flows of N₂ for both the reactant inflow and the curtain gas. Following this initiation step, we next consider the calculation of the time-dependent flows from the reactant inlet: alternating pulses of the thin film precursor (or surrogate) and the coreactant (e.g., H₂O), separated by flow of the purge gas (N₂). Xe was chosen as a surrogate of organometallic precursors because the transport properties (or collision cross-sections and intermolecular potentials) of such molecules are generally not known. We consider the following sequence of inlet flow compositions and cycle times: H₂O (0.2 s), N₂ (0.6 s), Xe (0.2 s), and N₂ (0.6 s), for a total ALD cycle time of 1.6 s. We found that the calculations had difficulty converging if we assumed that the initial concentrations of the reactant species (Xe and H₂O) in the surrounding chamber were set identically to zero. Instead, we set the initial concentrations of these species to be 10 ppb (mole fractions of 10⁻⁸ for each reactant) in the surrounding chamber. Transport properties in this calculation were dependent on the composition of the ternary mixture, as well as temperature and pressure.

The outputs of the calculations that we are most interested are the concentrations/partial pressures/fluxes reaching the surface of the substrate, and those reaching the surfaces of the surrounding chamber. First, we consider the spatial variation of the partial pressures of the two species across the surface of the substrate. In these calculations, the mesh size at the substrate surface was 145 ± 34 μm. In Fig. 2, we plot the partial pressures predicted by the CFD analysis for two points in time: after 0.18 s time has elapsed for the 0.2 s pulse of H₂O, and after 0.18 s time has elapsed for the 0.2 s pulse of Xe. In both cases, these represent the approximate point in time of the pulse where the partial pressure of both species reached their maxima. As may be seen, the partial pressures of both species are quite uniform across the reaction zone, where the fraction of that achieved at the edge of the reaction zone, with respect to that in the center is 87% for H₂O and 90% for Xe. This amount of nonuniformity could be an issue for continuous processes such as CVD, but should not be important for ALD, unless it is conducted in a subsaturation mode.

In Fig. 3(a), we plot the partial pressure of the two reactants as a function of time, averaged over the substrate surface. We see that the peak partial pressure is ∼7.6–7.7 Torr for H₂O, while that for Xe is 6.9–7.8 Torr, compared to the total pressure of ∼10 Torr (with N₂ making up the balance). We see that there is minimal overlap between the two species. This suggests that CVD reactions (coexposure to both reactants) are unlikely for this dosing sequence, although increasing the purge time would certainly be an option. If we integrate the data presented in Fig. 3(a), we can produce the results we present in Fig. 3(b), where we plot the cumulative exposure of the substrate to the two reactants. As may be seen we observe a staircaselike increase in the cumulative exposure for both H₂O and Xe, but that for H₂O approaches the absolute limit (20 Torr s) closer given the number of cycles simulated (10) and the inlet partial pressure (10 Torr) and cycle time (0.2 s). The cumulative precursor exposure reaches only 70% of the theoretical limit. This may reflect a difference in diffusivities of the two species (the binary diffusivity of Xe in N₂ is about 60% of that for H₂O in N₂ for these reaction conditions).

The behavior at the surfaces of the surrounding chamber is quite different, which is of course a major objective of the design of the microreactor. In Fig. 3(c), we plot the partial pressure of the two reactants as a function of time, averaged over the surfaces of the surrounding chamber. As indicated earlier, we set the mole fractions of the reactants at zero time to be 10⁻⁸ in the surrounding chamber, or a partial pressure of 10⁻⁷Torr. First, concerning the partial pressure of Xe, it exhibits a continuous decrease with time, exhibiting no periodicity, and in fact is fit extremely well by a simple exponential decay. This indicates that no Xe that enters the microreactor probe via the reactant inlet reaches the surrounding chamber—it is perfectly confined in the volume defined by the reaction zone and the gap leading to the annular effluent channel. The partial pressure of H₂O is quite different, as may be seen—it shows clear oscillations, but the peak partial pressure seems to be approaching a constant value. Indeed these peak values are very well described by an exponential decay to a constant value of ∼1.8 × 10⁻⁶Torr. This value is still an approximately 8 orders-of-magnitude reduction in the amount of H₂O reaching those surfaces as compared to that on substrate surface. It would seem that the inward radial flow of the curtain gas is sufficient to confine Xe, but not H₂O, due to a difference in their diffusivities. In Fig. 3(d), we plot the cumulative exposure of the surfaces of the surrounding chamber to H₂O as a function of time. Consistent with results shown in Fig. 3(c), after an initial transient period of ∼10 s, the increase in the exposure becomes approximately constant, reaching a value of ∼1.3 × 10⁻⁵Torr s after 10 ALD cycles. This would represent a 13 Langmuir (L) exposure of the surfaces of the surrounding chamber to H₂O, which should not sufficiently degrade the ability to reach UHV after pump-out.

In summary, the CFD calculations indicate that using this design one can produce short pulses of reactant species, separated sufficiently by purge streams. The use of a curtain gas predicts that large molecules such as typical thin film precursors should be effectively confined to the reaction zone. On the other hand, smaller molecules, including typical coreactants such as H₂O, may be present at trace levels in the surrounding chamber, but at levels that may be compatible with achieving UHV conditions after pump-out.

To evaluate the performance of the microreactor, we have conducted two sets of experiments: (1) a gas-surface reaction between a single reactant [I₂(g)] and a substrate surface (Cu); and (2) ALD of a thin film (ZrO₂) on a substrate surface (SiO₂). In the first experiment, we have examined the reaction of I₂(g) vapor with a thin film of Cu. The Cu substrate in this case was a thin film of Cu (900 Å thick, sputter deposited onto a SiO₂ substrate possessing also an intermediate thin film of Ta, several Å thick), mounted on a metal (Mo) platen capable of holding a wafer of 10 cm in diameter. Before inserting a sample, the sample manipulator is positioned in the intermediate chamber and the gate valve separating the main surface analysis chamber and the intermediate chamber is closed. The sample is first placed in the load-lock chamber and is then transferred onto the sample manipulator in the intermediate chamber when the pressure in the load-lock chamber is below 1 × 10⁻⁶Torr. Once mounted, the sample platen stays on the manipulator during thin-film deposition as well as surface analysis.

To begin an experiment, the gate valve isolating the microreactor probe is opened and the probe is translated into the intermediate chamber with a probe-substrate gap of ∼1.5 cm. Next, the flow of the curtain gas (N₂ at 50–75 sccm) is initiated, and the sample is heated to the appropriate temperature for reaction or deposition. Once the sample temperature has stabilized, using a precision linear translator the probe is positioned in front of the sample, which produces the small gap (∼0.05–0.1 cm) between the probe and the sample surface, and flow of the reactant (and purge for ALD) gas stream (N₂) is started. Spring-loaded electrical probes aid in the precise positioning of the probe, as contact with the conducting sample holder produces a change in resistance.

For the reaction of I₂(g) with a Cu surface, we have delivered the I₂(g) as a vapor using a carrier gas (N₂ at 20 sccm). The vessel containing solid I₂ was cooled to a temperature of 0 °C [vapor pressure ∼11 mTorr (Ref. 31)] while the lines downstream leading to the microreactor were kept at room temperature. The Cu substrate was exposed to the vapor for 30 s. Following reaction, the intermediate chamber was evacuated, the microreactor was retracted and isolated by closing a gate valve, the gate valve separating the intermediate chamber and the main surface analysis chamber was opened, and the sample was lowered into the position for analysis using XPS. In Fig. 4, we display results from XPS where we have translated the substrate such as to produce a line scan, i.e., the integrated intensity of the I(3d) peak as a function of position. For this experiment, the area sampled by XPS was 0.12 cm dia., and the line scan step size was 0.254 cm, so there is no overlap between neighboring data points. As may be seen, we observe a flat-top profile, and over the reaction zone, the mean of the measured integrated intensity is 1.920 × 10⁵ eV counts s⁻¹, with a standard deviation of 5800 eV counts s⁻¹ (∼3% of the mean). Outside of the reaction zone, we observe a smooth drop in intensity, reaching a value of zero at the centerline of the annular exhaust region. This is in good agreement with the predicted variation of the partial pressure by the CFD calculations shown in Fig. 2. These experimental results indicate that we have achieved excellent confinement of the I₂(g), with uniform substrate modification over the reaction zone, and essentially no evidence of reaction beyond the annular exhaust region.

We have examined the ALD of ZrO₂ using as reactants tetrakis(ethylmethylamino) zirconium, Zr[N(CH₂CH₃)(CH₃)]₄ (purity 99.9999%, Air Liquide) and water vapor on SiO₂. The SiO₂ substrates were prepared starting from single-side polished, Si(100) wafers (B doped, resistivity 38–63 Ω cm). The native SiO₂ layer was removed from the substrates by dipping in buffered oxide etch (BOE) for 2 min. The substrates were then reoxidized by dipping in Nanostrip for 15 min at a temperature of ∼75 °C. The BOE/Nanostrip treatment was then repeated. This method is known to produce a 15–20 Å layer of chemical oxide with surface Si-OH density of ∼5 × 10¹⁴ cm⁻².^28,32

Once the Si substrate has been loaded, and the microreactor has been placed in position, the flows of the curtain and carrier gases are initiated and the substrate is heated to the temperature for ALD. First, we conducted a series of ½ cycle experiments to verify that for the conditions examined that they were in the so-called ALD window. For these experiments, the combined gas flow rate was 50 sccm [35 sccm N₂ curtain gas, 5 sccm (He) carrier/(N₂) purge gas for the (Zr) reactant, and 5 sccm N₂ each in the other two inlets], and the total pressure in the microreactor was 5 Torr. The substrate temperature was set to T_s = 180 °C. The temperature of the bubbler containing the Zr[N(CH₂CH₃)(CH₃)]₄ precursor was held at 40 °C, producing a vapor pressure of ∼30 mTorr.³³ We also note here that the temperature of the feed lines (both those outside the UHV chamber, and those making up the fluidic extension) were heated to a temperature intermediate (typically ∼60–80 °C) between that of the bubbler and that of the substrate, to minimize condensation of the Zr precursor. In Fig. 5, we display the integrated intensity of the Zr(3d) peak as a function of the exposure time of the SiO₂ surface to the Zr[N(CH₂CH₃)(CH₃)]₄ precursor. As may be seen, saturation is reached after an exposure time of ∼1 min. This time is not totally the effect of the kinetics of adsorption, rather it is mostly associated with the length of the feed lines between the bubbler and the microreactor, and may also include effects due to passivation of these feed lines. For example, particularly concerning this Zr precursor, we have found a correlation between the length of the feedlines and the minimum exposure required for saturation. We have also observed a dependence on the chemistry examined, with the Al(CH₃)₃ precursor exhibiting saturation in an exposure time of <10 s.

We next conducted ALD of ZrO₂ using as reactants Zr[N(CH₂CH₃)(CH₃)]₄ as the precursor (bubbler temperature of 40 °C) and H₂O as the coreactant (held at 0 °C in a stainless steel bubbler). For these experiments, we used a total flow of 95 sccm (curtain, carrier, and purge gases), which resulted in an operating pressure of 18 Torr. The dosing sequence was as follows: Zr[N(CH₂CH₃)(CH₃)]₄ for 2 min (delivered using 10 sccm of a He carrier gas), purge for 2 min with N₂, H₂O for 2 min (delivered using 10 sccm of a N₂ carrier gas), and finally purge for 2 min with N₂. The combined flow of purge (N₂) and carrier (He or N₂) gases through the two reactant inlet lines was always 20 sccm (the balance being 75 sccm of curtain gas N₂ flow). Once we had conducted the desired number of ALD cycles, the intermediate chamber was pumped down, first to a pressure of <100 mTorr by a roughing pump, and then to <10⁻⁶Torr by a turbomolecular pump. During this time period, the microreactor probe is also retracted and isolated from the intermediate chamber by closing a gate valve. The sample was then transferred to main chamber using the precision sample manipulator for surface analysis using XPS.

In Fig. 6, we consider a series of XP spectra taken on thin films of ZrO₂ deposited by ALD, where the samples were examined directly after growth, and no air-exposure of the samples had occurred. XPS is of course a surface-sensitive technique that probes the elemental composition and chemical states of the elements in a material. In Fig. 6(a), we display the XP spectrum of the Zr(3d) region for a thin film of ZrO₂ deposited on SiO₂ at 150 °C after 10 cycles of ALD. As may be seen, we clearly observe the Zr(3d_5/2) and Zr(3d_3/2) peaks with a spin-orbit doublet splitting of 2.4 eV.^34,35 In addition, we find a binding energy for the Zr(3d_5/2) peak of 183.3 ± 0.1 eV, which is in good agreement with previous reports for ZrO₂.³⁶ In Fig. 6(b), we display the XP spectrum of the O(1s) region for this same thin film of ZrO₂ deposited on SiO₂ at 150 °C after 10 cycles of ALD. As may be seen, the O(1s) feature is fit with two peaks, one at higher binding energy (532.9 ± 0.1 eV) and representing the underlying SiO₂ substrate and the other at lower binding energy (531.2 ± 0.1 eV) representing the ZrO₂ thin film. These values are in good agreement with previously reported values for the O(1s) peak for SiO₂ (Ref. 37) and ZrO₂.³⁴ We also have examined the N(1s) feature and these data are displayed in Fig. 6(c), and we observe a single peak at a binding energy of 399.5 ± 0.3 eV. Although detectable, this amount of N represents a submonolayer coverage of about ∼0.07 ML if we compare this to the density of O found in crystalline ZrO₂.

In Fig. 7, we plot the integrated intensity of the Zr(3d) feature as a function of positon on the sample, similar to the procedure we used in connection with the data shown in Fig. 4. Here, the step size varied between 0.254 and 0.508 cm. In Fig. 7, we have also indicated by the dashed line the results shown in Fig. 4. As may be seen, we also observe here a flat top profile and a drop in the intensity moving from the perimeter of the reaction zone to the effluent groove, indicating excellent confinement of the reaction to the central zone.

We have also examined thin films of ZrO₂ deposited on SiO₂ at 150 °C after 20 and 40 cycles of ALD, and we consider these results in Fig. 8. Here, we plot the integrated intensities of the peaks representing the deposited thin film [Zr(3d)] and the substrate [Si(2p)] as a function of the number of ALD cycles. As may be seen there is a continuous increase in the intensity of the Zr(3d) feature, and a continuous decrease in the intensity of the Si(2p) feature, consistent with expectations. We have fit these data to a simple model that assumes the growth rate per cycle is constant, and that a uniform 2D thin film of ZrO₂ is deposited on the SiO₂ substrate. In this case, the integrated intensities will exhibit opposing exponential decays, and the decay constant for the two peaks will differ only by a factor that depends on the inelastic mean free path of the photoelectrons (which scales with the square root of the kinetic energy). The fit to this model is shown by the smooth curves in Fig. 8, and as may be seen this model describes the data quite well.

In addition to analysis by XPS similar to those shown in Figs. 6 and 8, we have also determined the thicknesses of the ZrO₂ thin films by employing ex situ spectroscopic ellipsometry (SE, Woollam VASE ellipsometer) after the samples were unloaded from the UHV chamber. The thickness of ZrO₂ film on SiO₂ was determined by the difference in the ellipsometric thickness between a bare chemical oxide and the ZrO₂|SiO₂ film. This analysis made use of the n and k values for ZrO₂ and SiO₂. We plot these results in Fig. 9 where we display the thickness of ZrO₂ films deposited on SiO₂ as a function of ALD cycles, measured by ex situ spectroscopic ellipsometry. We also plot the thicknesses predicted by the integrated intensities of both the Zr(3d) and Si(2p) peaks, using the same assumptions that went into the fit we displayed in Fig. 8. As may be seen, for all sets of data the thickness of ZrO₂ deposited grows linearly with the number of ALD cycles, and there is no indication of a nucleation delay,²⁸ as the results are well described by a linear function passing through the origin. From SE, we find a rate of ∼0.80 Å cycle⁻¹, where from XPS we find rates of ∼0.64–0.70 Å cycle⁻¹. These values are similar to those reported previously for this same ALD process conducted under similar conditions, i.e., rates of ∼0.9–1.1 Å cycle⁻¹.^38–41 

We have described the design, construction, and performance evaluation of a microreactor probe that enables the study of gas-surface reactions at pressures on the order of 10 Torr, and subsequent in situ analysis of the surface under UHV conditions with no air-break. Our design makes use of the flow of a curtain gas to spatially confine the reactant species to a well-defined area on the substrate surface. Calculations using computational fluid dynamics determined the critical dimensions of the probe under typical reaction conditions necessary to achieve confinement and minimize out-diffusion of the reactant species to the surrounding UHV chamber. We evaluated the performance of the probe by examining two gas-surface reactions: (1) the adsorption of I₂ on a Cu surface, and (2) the atomic layer deposition of ZrO₂ using a Zr amido compound and H₂O as the coreactant. Confinement of the reactant species was verified by the use of in situ XPS, and the thickness of ZrO₂ deposited per cycle was consistent with results reported using conventional reactors designed for ALD. With suitable optimization of operating conditions of the microreactor, we anticipate that this design could be replicated and retrofitted onto a variety of UHV systems that possess the tools for in situ surface analysis, but the lack the ability to study gas-surface reactions at low-to-medium vacuum conditions.

This work was supported by the U.S. Department of Energy, Office of Basic Energy Sciences, Division of Materials Sciences and Engineering, under Award No. DESC0006647. Additional support was provided by the Semiconductor Research Corporation (Tasks 2524.001 and 2530.001). This work was performed in part at the Cornell NanoScale Facility, a member of the National Nanotechnology Coordinated Infrastructure (NNCI), which is supported by the National Science Foundation (Grant No. ECCS-1542081). The authors gratefully acknowledge the technical assistance of Harold Fu and Paul A. Pelletier.