The authors have examined the effect of two molecules that form self-assembled monolayers (SAMs) on the subsequent growth of TaNx by atomic layer deposition (ALD) on two substrate surfaces, SiO2 and Cu. The SAMs that the authors have investigated include two vapor phase deposited, fluorinated alkyl silanes: Cl3Si(CH2)2(CF2)5CF3 (FOTS) and (C2H5O)3Si(CH2)2(CF2)7CF3 (HDFTEOS). Both the SAMs themselves and the TaNx thin films, grown using Ta[N(CH3)2]5 and NH3, were analyzed ex situ using contact angle, spectroscopic ellipsometry, x-ray photoelectron spectroscopy (XPS), and low energy ion-scattering spectroscopy (LEISS). First, the authors find that both SAMs on SiO2 are nominally stable at Ts ∼ 300 °C, the substrate temperature used for ALD, while on Cu, the authors find that HDFTEOS thermally desorbs, while FOTS is retained on the surface. The latter result reflects the difference in the head groups of these two molecules. The authors find that both SAMs strongly attenuate the ALD growth of TaNx on SiO2, by about a factor of 10, while on Cu, the SAMs have no effect on ALD growth. Results from LEISS and XPS are decisive in determining the nature of the mechanism of growth of TaNx on all surfaces. Growth on SiO2 is 2D and approximately layer-by-layer, while on the surfaces terminated by the SAMs, it nucleates at defect sites, is islanded, and is 3D. In the latter case, our results support growth of the TaNx thin film over the SAM, with a considerable delay in formation of a continuous thin film. Growth on Cu, with or without the SAMs, is also 3D and islanded, and there is also a delay in the formation of a continuous thin film as compared to growth on SiO2. These results highlight the power of coupling measurements from both LEISS and XPS in examinations of ultrathin films formed by ALD.

Atomic layer deposition (ALD) is well-suited to the formation of ultrathin films,1 particularly concerning growth on topologically complex three-dimensional device features such as those found in interconnect layers.2 However, conformal deposition where a thin film is formed on all exposed areas of a substrate is not always desired, especially when only select features need to be deposited. For example, in many device manufacturing applications, deposition may be desired on one exposed surface (e.g., metal) and not on another (e.g., dielectric). Selective deposition is a process where materials are only deposited where desired, effectively eliminating additional patterning steps using lithography and etching. Many current selective deposition technologies rely on lithography to pattern areas and block growth on areas where thin film growth is not desired. With device features becoming three dimensional and shrinking every generation, lithography is becoming less feasible, and the development of ALD processes that are selective is of increasing interest.

A common approach to achieve selective deposition is to provide a masking material over areas of the substrate where deposition is not desired. In previous work self-assembled monolayers (SAMs) have shown promise as molecular-scale masks, preventing deposition at early stages of ALD.3 With the appropriate choice of head groups, SAMs can be chosen to react selectively with one surface, while not reacting at all with another. For example, thiols (R-SH) react with metal surfaces such as Ag, Au, and Cu,4 while silanes (R-Si-X3) react with hydroxylated (–OH) surfaces. SAMs also contain tail groups, which can effectively control the surface energy of the substrate as well as other properties. In addition, combining SAMs with ALD is a logical choice, since many substrate temperatures used for ALD are relatively low, which is important concerning the stability of SAMs and their ability to block growth.

There has been significant previous work done in the area of using SAMs as molecular masks to prevent ALD.5,6 One recent study has examined the use of a SAM to selectively block growth ALD on one surface (Cu), while permitting it to occur on another (SiO2).7 A modification of this approach has also shown promise concerning selective area molecular layer deposition.8 However, much of the previous work has focused on patterning a SAM that is attached to a single substrate such as SiO2, blocking growth in areas where the SAM is present. Several techniques have been employed to deposit and/or pattern SAMs, including microcontact printing,3,9–12 solution-based assembly3,13–23 and vapor deposition.24–28 From this work, important relationships between the density of the SAM and its effectiveness as a molecular blocking agent have been identified.3,13

In this study, we have investigated two molecules as blocking layers for ALD: 1H, 1H, 2H, 2H-perfluorooctyltrichlorosilane [FOTS, (Cl)3Si(CH2)2(CF2)5CF3] and (heptadecafluoro-1,1,2,2-tetrahydrodecyl)triethoxysilane [HDFTEOS, (CH3CH2O)3Si (CH2)2(CF2)7CF3] (Gelest, Inc., Morrisville, PA). Both of the SAMs, shown in Fig. 1 in space-filling models, can be deposited onto the substrates via vapor deposition, prior to thin film growth via ALD. Vapor deposition of the SAMs holds obvious advantages for process integration as one does not require a vacuum break between the formation of the SAM layer and the ALD process. Concerning the SAMs themselves, both of these molecules have long fluorinated alkyl tail groups, which should be unreactive to many thin film precursors as the CFx chains are coordinatively saturated and consist of the strongest single bond (C–F) in organic compounds. Since both molecules are silanes, they can be expected to bind preferentially to a dielectric material such as SiO2, and not to a metal such as Cu, because of the prevalence of hydroxyl groups on SiO2. Thus, here we will investigate the ability of these two SAMs to block ALD growth on SiO2, while leaving growth on Cu relatively untouched. Finally, we recognize that, while both molecules are silanes, they do possess different head group chemistries [–SiCl3 versus –Si(OCH2CH3)3], which may turn out to be important in terms of their interaction with SiO2 and Cu surfaces.

Fig. 1.

(Color online) Space-filling models for the two reactants we consider here, Ta[N(CH3)2]5 and NH3, and the two SAMs: FOTS, (Cl)3Si(CH2)2(CF2)5CF3, and HDFTEOS, (CH3CH2O)3Si(CH2)2(CF2)7CF3.

Fig. 1.

(Color online) Space-filling models for the two reactants we consider here, Ta[N(CH3)2]5 and NH3, and the two SAMs: FOTS, (Cl)3Si(CH2)2(CF2)5CF3, and HDFTEOS, (CH3CH2O)3Si(CH2)2(CF2)7CF3.

Close modal

The deposition process investigated in this study is the thermal ALD of tantalum nitride (TaNx) in a conventional viscous flow reactor using pentakis dimethylamido tantalum {Ta[N(CH3)2]5} and NH3 as precursors (Ts ∼ 225–300 °C). TaNx is a material currently used as a Cu diffusion barrier and relevant to interconnect layers in microelectronic devices. We have examined the ALD growth of TaNx on SiO2 and Cu, both clean surfaces and ones that have been exposed to FOTS and HDFTEOS. We used ex situ contact angle, spectroscopic ellipsometry, x-ray photoelectron spectroscopy (XPS), and low-energy ion scattering spectroscopy (LEISS) to examine both the SAMs themselves and the TaNx thin films that were deposited. We will demonstrate that these latter two surface-sensitive probes are especially effective in determining not only the degree of selectivity, but also the mode of growth.

The experiments described here involved a sequence of steps beginning with preparation of the starting substrates (SiO2 and Cu), formation and characterization of the SAMs, and finally, ALD TaNx thin film growth and characterization. Complete details can be found in the supplementary material.29 Following substrate preparation, the next step in most cases was the growth of the SAMs via molecular vapor deposition on the two substrates: SiO2 (chemical oxide) and Cu [a 900 Å thick Cu thin film deposited on SiO2 via physical vapor deposition]. The Cu substrates were used as received and possess a native oxide as shown by XPS in the supplementary material.29 These substrate surfaces terminated with the SAMs were then characterized using a number of techniques. First, we used ex situ measurements of the contact angle (VCA optima, Billerica, MA) using water, where multiple (∼5) contact angle measurements were obtained and then averaged. We also used ex situ spectroscopy ellipsometry (SE; Woollam, Lincoln, NE) to estimate the thicknesses of the SAMs (only on the SiO2 substrates). These two techniques (contact angle and SE) allow for a quick noninvasive evaluation of the SAMs, albeit lacking chemical specificity.

All substrates were placed in the viscous flow ALD reactor (Oxford FlexAL, Oxford Instruments, Oxfordshire, UK), heated to the desired temperature for growth, and exposed to a set number of cycles (0–200) of the TaNx ALD process. Following processing in the ALD reactor, a number of samples were investigated using contact angle measurements and spectroscopic ellipsometry. All samples, including some that were “as-received” and not subjected to the ALD process temperature, were transferred and placed in a custom-designed ultrahigh vacuum chamber30,31 for analysis using XPS and LEISS. XPS is a powerful technique for analyzing thin films that are <10 nm thick and can give absolute densities/concentrations if appropriate calibration standards are used, and it holds the potential for depth resolution using angle-resolved XPS. LEISS has been shown to have very good surface sensitivity and is extremely useful in determining thin film continuity and interface abruptness of ultrathin films.32,33

As described in Sec. II and in the supplementary material,29 the SAMs were formed via vapor deposition by exposing the substrates to FOTS or HDFTEOS. Once formed, the SAMs were characterized ex situ using contact angle measurements, spectroscopic ellipsometry, XPS, and LEISS.

In Fig. 2, we display results for the contact angle of samples composed of (a) SiO2 and (b) Cu that have been exposed to HDFTEOS and FOTS. Results are shown for SiO2 and Cu substrates, with and without a SAM, at room temperature (as received), and after annealing the samples for 5 min to a temperature (∼300 °C) that was used for TaNx ALD. For these “0 cycles” data, the atmosphere in the ALD reactor consisted of Ar (99.9999%) at a pressure of 200 mTorr. Measurement of the water contact angles on the SiO2 surfaces (with and without SAMs) exhibited no significant time dependence. For example, for HDFTEOS|SiO2, the contact angle was measured to be within +2°/−4° over a time period of 7 min, a range within experimental uncertainties. On the Cu surfaces, however, we observed a change in the contact angle on the Cu surfaces as a function of time, most likely due to Cu oxidizing or reacting in the presence of water (as shown in the supplementary material,29 annealing the Cu samples to the substrate temperature for ALD reduces the surface oxides). Typically, for the first minute of exposure, the change observed was at most 4°, while after 7 min of exposure, changes of ∼50° were observed. To minimize subjectivity in the contact angle measurements, all results (on Cu and SiO2) were taken at the same time period after the water droplet made contact with the substrate surface (60 s).

Fig. 2.

(Color online) Contact angles measured for the six surfaces examined here for as-received samples, and those annealed to the substrate temperature for ALD (300 °C), 0 cycles: (a) clean SiO2 and FOTS and HDFTEOS on SiO2 and (b) bare Cu and FOTS and HDFTEOS on Cu.

Fig. 2.

(Color online) Contact angles measured for the six surfaces examined here for as-received samples, and those annealed to the substrate temperature for ALD (300 °C), 0 cycles: (a) clean SiO2 and FOTS and HDFTEOS on SiO2 and (b) bare Cu and FOTS and HDFTEOS on Cu.

Close modal

The most significant observations from these data first include the fact that the contact angles for both SAMs on the unannealed samples are quite similar, independent of substrate, ∼107° and ∼109° for FOTS on SiO2 and Cu, and ∼97° and ∼99° for HDFTEOS, respectively. Second, we observe that for both SAMs on SiO2, the change in contact angle is modest (increase of ∼11°–16°) after annealing to the process temperature for ALD. This suggests that these two SAMs may be relatively intact on SiO2 at the substrate temperature used for ALD growth here. It also suggests that both SAMs may become more ordered by annealing to the substrate temperature for ALD, possibly assisted by trace amounts of H2O (likely present in the reactor) assisting reconfiguration/movement of the head groups. On bare Cu substrates, however, we observed a significant decrease (∼27°) in the contact angle after annealing to ALD temperatures. In a previous study, we have found that similar Cu thin film substrates undergo significant surface morphology changes at ALD temperatures, resulting in smoothening.33 Results from XPS (cf. supplementary material)29 indicate that there are also chemical changes occurring on the Cu surface due to annealing to the ALD process temperature. In particular, features associated with oxides of Cu are greatly reduced after annealing. In addition, on Cu, we observe a difference in the effect of annealing on the contact angles for each SAM: for FOTS, there is a modest increase (∼7°) in the contact angle upon annealing, similar to that on SiO2, while for HDFTEOS, there is a significant decrease (∼22°) in the contact angle upon annealing, and the decrease in the value is similar to that observed for bare Cu upon annealing. The change in the contact angle for HDFTEOS|Cu upon annealing could mean a number of things, including thermal desorption of the molecule from Cu at the elevated temperatures (∼300 °C) used for ALD growth, and/or possibly chemical changes in the Cu substrate itself. We will now consider those possibilities directly.

We have also used XPS to characterize the SAMs on both the SiO2 and Cu surfaces, again for the unannealed samples, and those that have been annealed to the process temperature used in ALD. Here, we first focus on photoemission from the F(1s) feature, since it gives us a unique signal as fluorine is only present in the SAM molecules. In the supplementary material,29 we also present results from analysis of the C(1s) feature, as it exhibits a distinct feature associated with the CFx species in the SAM backbone. These results support those we report here for the F(1s). In Fig. 3, we present the integrated intensity of the F(1s) peak for all SAM-treated surfaces, both as-received pre- and post-0 cycles of ALD. After insertion into vacuum, these samples received no further treatment (e.g., Ar+ sputtering) before analysis via XPS. Halogens such as fluorine are notorious for artifacts associated with photoelectron induced desorption. To account for this, multiple F(1s) spectra were taken, and the results were extrapolated to represent the signal that would have been achieved at zero x-ray exposure time. As we can see, the F(1s) intensity remained essentially constant upon annealing for three cases: both SAMs on the SiO2 surface, and FOTS on Cu. However, for the HDFTEOS on Cu surface, a large decrease in the F(1s) intensity has been observed. This result supports the assertion, suggested by the contact angle data, that HDFTEOS or its adsorbed molecular fragment is not stable on the Cu surface and desorbs at the elevated temperature of ∼300 °C investigated here for the ALD of TaNx.

Fig. 3.

(Color online) Integrated intensities of the F(1s) peak from XPS for FOTS (squares) and HDFTEOS (diamonds) SAMs formed on SiO2 and Cu substrates for as-received samples, and those annealed to the substrate temperature for ALD (300 °C), 0 cycles.

Fig. 3.

(Color online) Integrated intensities of the F(1s) peak from XPS for FOTS (squares) and HDFTEOS (diamonds) SAMs formed on SiO2 and Cu substrates for as-received samples, and those annealed to the substrate temperature for ALD (300 °C), 0 cycles.

Close modal

Finally, the F(1s) [and C(1s) (from CFx)] feature observed in XPS can be used to estimate the absolute coverage of the SAMs on each of the surfaces. Details concerning this calculation, including possible effects due to the ubiquitous adventitious carbon are given in the supplementary material.29 In Table I, we summarize our calculations for the density of the two SAMs on the SiO2 and Cu substrates. The range given in the table represents the two limiting assumptions: no effect of the adventitious layer (lower limit); and factoring in attenuation by the adventitious layer (upper limit). First, focusing on the densities based on the F(1s) intensities, for both SAMs on SiO2, we see that the values for the as-received samples and those subjected to an anneal to the ALD process temperature give similar values, within ∼10%. Concerning the densities calculated from the C(1s)-CFx intensities, there is an apparent modest (∼30%) increase in the density upon annealing to the ALD process temperature for these SAMs on SiO2. These apparent increases are not physical and could be due to decreased effects of attenuation by contamination layers after the high temperature anneal, or other effects. If we take a simple average of the lower and upper limits obtained from the F(1s) data, we calculate densities of 2.51 × 1014 and 2.47 × 1014 molecules-cm−2 for as-received and 0 cycle ALD for FOTS|SiO2, and 2.20 × 1014 and 2.46 × 1014 molecules-cm−2 for as-received and 0 cycle ALD for HDFTEOS|SiO2. Concerning the former values, in previous work using essentially the same procedure to form the FOTS|SiO2 layers, we reported a density of (1.97 ± 0.12) × 1014 molecules-cm−2 based on results from synchrotron x-ray reflectivity.34 

Table I.

Properties of self-assembled monolayers.

SAM|substrateTreatmentDensity from F(1s) (molecules-cm−2)Density from C(1s) (molecules-cm−2)
HDFTEOS|SiO2 As-received 1.96–2.45 × 1014 2.09–2.61 × 1014 
 0 cycles of TaNx ALD 2.19–2.73 × 1014 2.65–3.31 × 1014 
FOTS|SiO2 As-received 2.23–2.79 × 1014 2.21–2.75 × 1014 
 0 cycles of TaNx ALD 2.20–2.74 × 1014 2.81–3.51 × 1014 
HDFTEOS|Cu As-received 1.24–1.55 × 1014 1.61–2.01 × 1014 
 0 cycles of TaNx ALD 0.23–0.29 × 1014 0.19–0.24 × 1014 
FOTS|Cu As-received 2.53–3.15 × 1014 3.22–4.02 × 1014 
 0 cycles of TaNx ALD 2.52–3.14 × 1014 2.31–2.88 × 1014 
SAM|substrateTreatmentDensity from F(1s) (molecules-cm−2)Density from C(1s) (molecules-cm−2)
HDFTEOS|SiO2 As-received 1.96–2.45 × 1014 2.09–2.61 × 1014 
 0 cycles of TaNx ALD 2.19–2.73 × 1014 2.65–3.31 × 1014 
FOTS|SiO2 As-received 2.23–2.79 × 1014 2.21–2.75 × 1014 
 0 cycles of TaNx ALD 2.20–2.74 × 1014 2.81–3.51 × 1014 
HDFTEOS|Cu As-received 1.24–1.55 × 1014 1.61–2.01 × 1014 
 0 cycles of TaNx ALD 0.23–0.29 × 1014 0.19–0.24 × 1014 
FOTS|Cu As-received 2.53–3.15 × 1014 3.22–4.02 × 1014 
 0 cycles of TaNx ALD 2.52–3.14 × 1014 2.31–2.88 × 1014 

Moving on to the results for the SAMs on Cu, we see that for FOTS|Cu, the densities found are quite similar to those found on FOTS|SiO2. Again, using the F(1s) data, we calculate densities of 2.84 and 2.83 × 1014 molecules-cm2, for the as-received and 0 cycle TaNx ALD samples. For HDFTEOS|Cu, however, both the F(1s) and the C(1s)-CFx data indicate a large (∼80%–90%) decrease in the density after the high temperature anneal, from ∼1.40 to 0.26 × 1014 molecules-cm2, reflecting the data we have already shown in Fig. 3. We note also that the absolute density for the as-received HDFTEOS|Cu is also less than (∼63%) what we observe for HDFTEOS|SiO2, suggesting further a relatively weak interaction between HDFTEOS and the Cu surface.

Summarizing our results concerning just the SAMs, we find that both molecules, HDFTEOS and FOTS, bind strongly to the SiO2 surface, and form layers with a density of ∼2–3 × 1014 molecules-cm2. These layers remain essentially intact after annealed to the ALD process temperature of 300 °C in an inert environment. On the Cu surface, the behavior is quite different. For FOTS on Cu, SAMs are formed that have a density similar to those on SiO2, and these layers are stable after annealing to the ALD process temperature in an inert environment. For HDFTEOS, however, the as-received layers that form are about 2/3 the density of those formed on SiO2, and the density of these layers decreases further to a value that is about 10% of those formed on SiO2.

1. Spectroscopic ellipsometry

We have employed SE ex situ to measure the thickness of thin films deposited by ALD as a first check of the effect of the self-assembled monolayers on the growth of TaNx. In Fig. 4, we plot the thin film thickness deduced from SE as a function of the number of ALD cycles for growth of TaNx on bare SiO2 (chemical oxide), and a SiO2 surface that has a SAM formed from HDFTEOS. Here, we made use of the Cauchy dispersion relationship for the optical constants for the deposited thin film in the analysis. We found very similar results (see the supplementary material)29 when we assumed that the thin film of TaNx could be described by the optical constants for Ta2O5. For example, as shown in the supplementary material,29 XPS results indicate that air exposure of the thin films resulted in significant oxidation of the TaNx thin film, particularly for thin films <30 Å in thickness. We can make a number of observations from the data shown in Fig. 4. First, there is an offset between the thin film thicknesses for growth on SiO2 versus HDFTEOS|SiO2 for all conditions examined, and the value on SiO2 is greater by ∼9 Å. Second, an incubation period is clearly indicated for growth on HDFTEOS|SiO2, but not on SiO2. Third, a linear fit to the thickness (D) deposited per cycle for n = 10–200 cycles gives growth rates (dD/dn) on the two surfaces that are nearly identical: 0.385 Å cycle−1 on SiO2 and 0.383 Å cycle−1 on HDFTEOS|SiO2.

Fig. 4.

(Color online) Thin film thickness as measured by spectroscopic ellipsometry as a function of the number of ALD cycles for the deposition of TaNx on clean SiO2 and HDFTEOS on SiO2 at Ts = 300 °C. The solid lines are simply to guide the eye, while the dashed lines are linear fits of the data for n = 10–200 cycles of ALD.

Fig. 4.

(Color online) Thin film thickness as measured by spectroscopic ellipsometry as a function of the number of ALD cycles for the deposition of TaNx on clean SiO2 and HDFTEOS on SiO2 at Ts = 300 °C. The solid lines are simply to guide the eye, while the dashed lines are linear fits of the data for n = 10–200 cycles of ALD.

Close modal

The results presented in Fig. 4 indicate clearly that the SAM formed by HDFTEOS has a significant effect on the early stages of growth, i.e., n < 20 cycles, or D < 10 Å. Unfortunately, analysis of TaNx thin film formation on Cu using SE is problematic and was not attempted here. Moreover, for many applications using ALD, the goal is to form films with ultrathin thicknesses, where D < 20 Å. Thus, there is a clear need for the use of techniques that are well-suited to the analysis of ultrathin films. In the next two sections, III B 2 and III B 3, we will apply two such techniques: XPS and LEISS, and we will find that the application of these two techniques is decisive in determining the effects of the two SAMs on ALD growth of TaNx.

2. X-ray photoelectron spectroscopy

In all cases to be presented below, analysis of the substrate and the thin films deposited on them were conducted ex situ. In these cases, we always find that the samples possess an ultrathin layer of adventitious carbon. As described elsewhere,33 a short time (∼1 min) exposure to a 3 keV Ar+ ion beam can remove ∼70%–80% of this contamination layer without adversely affecting the samples for subsequent analysis using XPS. Moreover, this procedure is required for LEISS (vide infra). We note that this previous work involved essentially exclusively analysis of 5 and 10 Å thick TaNx thin films. Consequently, unless otherwise noted, the results presented below involve samples that have been subjected to this short time Ar+ sputter etch. In selected cases, however, we will report analyses on as-received samples, similar to the results presented above in connection with the characterization of the SAMs themselves.

In Fig. 5(a), we present a plot of the integrated intensity of the Ta(4d5/2) peak as a function of the number of ALD cycles for growth of TaNx on bare SiO2, and SiO2 modified with SAMs formed from FOTS and HDFTEOS. These results are for Ts = 300 °C, and we also observed similar behavior at Ts = 225 °C (see the supplementary material29). These intensities have been normalized to the intensity from a ∼150 Å thick TaNx thin film, one that is sufficiently thick that photoemission from the underlying substrate is completely extinguished and attenuation effects are saturated. We get an expected result for film growth on bare SiO2: a decaying exponential approach to a constant value. Given the absolute growth rate from the results employing SE, and accounting for the takeoff angle of 38.5° employed for XPS in Fig. 5(a), the rate of the exponential decay corresponds to an effective photoelectron decay length of ∼18 ± 3 Å, which can be compared to our previous estimate for the attenuation length of the Ta(4d5/2) photoelectrons in TaNx of 19.6 Å.35 Most important, the rate of TaNx thin film growth on both surfaces terminated with the SAMs is clearly attenuated. The intensities also exhibit upward curvature with increasing numbers of ALD cycles. For these two cases, the smooth curves represent interpolation between the data points, simply to guide the eye. The dotted curves are a fit to a model we will consider below. We see that the effect of these two SAMs on the ALD growth of TaNx on SiO2 is essentially the same up to ∼20 cycles of ALD.

Fig. 5.

(Color online) Integrated intensities of the peak from XPS associated with the thin film, Ta(4d5/2), as a function of the number of ALD cycles for the deposition of TaNx at Ts = 300 °C on (a) clean SiO2 and FOTS and HDFTEOS on SiO2; and on (b) bare Cu and FOTS and HDFTEOS on Cu. The solid lines in (a) for bare SiO2 and in (b) for all Cu surfaces are fits to an exponential decay function. In (a), the dashed lines for all samples represent fits to a model that accounts for an initial suppressed rate of growth that decays exponentially toward the steady-state rate of growth (see text). In (a), the solid lines for the two SAMs on SiO2 are to guide the eye.

Fig. 5.

(Color online) Integrated intensities of the peak from XPS associated with the thin film, Ta(4d5/2), as a function of the number of ALD cycles for the deposition of TaNx at Ts = 300 °C on (a) clean SiO2 and FOTS and HDFTEOS on SiO2; and on (b) bare Cu and FOTS and HDFTEOS on Cu. The solid lines in (a) for bare SiO2 and in (b) for all Cu surfaces are fits to an exponential decay function. In (a), the dashed lines for all samples represent fits to a model that accounts for an initial suppressed rate of growth that decays exponentially toward the steady-state rate of growth (see text). In (a), the solid lines for the two SAMs on SiO2 are to guide the eye.

Close modal

In Fig. 5(b), we present a plot of the integrated intensity of the Ta(4d5/2) peak as a function of the number of ALD cycles for growth of TaNx on bare Cu, and Cu modified with SAMs formed from FOTS and HDFTEOS. Again, the data have been normalized to the intensity from a thick TaNx film. These results show clear differences when compared to those found on the SiO2 surfaces. First, while these data are also well described by a decaying exponential function, the implicated rate of growth on bare Cu is ∼0.237 ± 0.24 Å cycle−1, or about 60% of the rate of growth on bare SiO2. The reasons for this change are not clear at the moment. Perhaps most interesting concerning these results is the fact that both SAMs do not affect TaNx growth on the Cu surface. A simple fit to these data give values for the rate of growth of 0.236 ± 0.24 Å cycle−1 on HDFTEOS|Cu, and 0.276 ± 0.29 Å cycle−1 on FOTS|Cu, values that are within experimental uncertainties to that observed on bare Cu.

In Fig. 6, we summarize the major results from Fig. 5, where we plot the initial rate of ALD growth of TaNx for all six surfaces examined here. Here, we make use of our estimate for λ = 19.6 Å to make these rates quantitative. For the two SAMs on SiO2, we have fit the early cycle data (0–20 cycles) shown in Fig. 5(a) to a straight line to assess the initial rate of growth. As may be seen, on SiO2, we find that the initial rate of growth of TaNx was significantly attenuated by the SAMs, by approximately a factor of 7–8. On Cu; however, we see that the SAMs have essentially no effect of the rate of growth of TaNx on the Cu surface. The results we presented in Sec. III B 1 can certainly shed some light on why we observe different effects of the SAMs on Cu and SiO2. Before doing so, however, we will consider additional results for the thin films.

Fig. 6.

(Color online) Initial rate of ALD of TaNx for the six surfaces examined here. For growth on bare SiO2 and the 3 Cu surfaces, the initial rates are derived from the fits of the Ta(4d5/2) integrated intensities to the exponential decay function displayed in Fig. 5. For the two SAMs on SiO2 the initial rates are based on a linear fit to the data for n = 0–20 cycles.

Fig. 6.

(Color online) Initial rate of ALD of TaNx for the six surfaces examined here. For growth on bare SiO2 and the 3 Cu surfaces, the initial rates are derived from the fits of the Ta(4d5/2) integrated intensities to the exponential decay function displayed in Fig. 5. For the two SAMs on SiO2 the initial rates are based on a linear fit to the data for n = 0–20 cycles.

Close modal

Analysis of the features due to emission from the substrate is also useful. For example, the attenuation of the substrate signal by an overlying thin film will depend on the nature of growth, e.g., 2D layer-by-layer (LbL) versus 3D islanded growth. In Fig. 7, we plot the integrated intensities of the two substrate peaks, in (a) the Si(2p) and in (b) the Cu(2p), as a function of the number of ALD cycles. All of these data have been normalized to a value of unity at 0 cycles of ALD. From Fig. 7(a), we see that the attenuation of the substrate signal on SiO2 with increasing numbers of ALD cycles is fastest on the bare surface. The data for the bare surface is fit very well by a simple exponential decay as shown by the solid line. Employing an attenuation length for the Si(2p) photoelectrons (λ = 20.8 Å), the rate of the decay of the data shown in Fig. 7(a) indicates a rate of TaNx growth equal to 0.44 ± 0.02 Å cycle−1, essentially identical to that found by an analysis of the data shown in Fig. 5(a) for Ta(4d5/2) (0.42 ± 0.07 Å cycle−1). The exponential decay observed for both the thin film and substrate signals for TaNx growth on bare SiO2, and the essentially identical growth rates that are implicated, are consistent with 2D LbL growth on this surface.

Fig. 7.

(Color online) Integrated intensities of the peaks from XPS associated with the substrates as a function of the number of ALD cycles for the deposition of TaNx at Ts = 300 °C on (a) clean SiO2 and FOTS and HDFTEOS on SiO2, Si(2p); and on (b) bare Cu and FOTS and HDFTEOS on Cu, Cu(2p). The solid lines in (a) for bare SiO2 and in (b) for all Cu surfaces are fits to an exponential decay function. In (a), the dashed lines for all samples represent fits to a model that accounts for an initial suppressed rate of growth that decays exponentially toward the steady-state rate of growth (see text). In (a), the solid lines for the two SAMs on SiO2 are to guide the eye.

Fig. 7.

(Color online) Integrated intensities of the peaks from XPS associated with the substrates as a function of the number of ALD cycles for the deposition of TaNx at Ts = 300 °C on (a) clean SiO2 and FOTS and HDFTEOS on SiO2, Si(2p); and on (b) bare Cu and FOTS and HDFTEOS on Cu, Cu(2p). The solid lines in (a) for bare SiO2 and in (b) for all Cu surfaces are fits to an exponential decay function. In (a), the dashed lines for all samples represent fits to a model that accounts for an initial suppressed rate of growth that decays exponentially toward the steady-state rate of growth (see text). In (a), the solid lines for the two SAMs on SiO2 are to guide the eye.

Close modal

In the cases of the SiO2 surface terminated by SAMs, we see that the rates of decrease/attenuation are slower, consistent with the behavior of the Ta(4d5/2) signal in Fig. 5(a). Even though the rate of the initial decrease is reduced, in the case of HDFTEOS|SiO2, we see that after 200 cycles, the substrate signal is essentially totally extinguished, just as it is on bare SiO2. The data for both SAMs are not well described by a simple exponential decay. The dotted lines are the prediction of the model fit to the Ta(4d5/2) data shown in Fig. 5(a). This model also does not capture the rate of decay well, although the agreement is better in the case of growth on HDFTEOS|SiO2.

We next consider the behavior of the Cu(2p) signals as a function of the number of ALD cycles. As may be seen in Fig. 7(b), the Cu(2p) data for bare Cu are well described by a decaying exponential. Employing an attenuation length for the Cu(2p) photoelectrons (λ = 10.9 Å), which is consistent with what we have employed for both Ta(4d5/2) and Si(2p), we find a growth rate for TaNx of 0.172 ± 0.007 Å cycle−1. This value is ∼23% smaller than the value implied by a fit to the Ta(4d5/2) data. The data for the two SAMs also show a decay that is approximately exponential, and the rate of decay is somewhat faster than that observed for bare Cu. Again employing an attenuation length for the Cu(2p) photoelectrons we find implicated growth rates of 0.208 ± 0.013 and 0.283 ± 0.021 Å cycle−1 for HDFTEOS |Cu and FOTS|Cu. These two values are much closer (within 12% and 2%, respectively) to the values implicated from the Ta(4d5/2) results.

For a final set of results concerning the use of XPS, we will consider the evolution of the F(1s) and C(1s)-CFx peaks as a function of the number of ALD cycles. These peaks are of course uniquely associated with the presence of the two SAMs, as there is no contribution from either the underlying substrates or the deposited thin film. If the SAMs remain intact, the ratio of these two peaks should be constant (see the supplementary material29). As described in the beginning of this section, we typically subject the sample to a short Ar+ sputter etch before analysis by XPS. Here, we report the intensities for the as-received samples as Ar+ ion bombardment had a significant effect on these results, similar to what we observe concerning the adventitious carbon layers. For example, as shown in the supplementary material,29 for data corresponding to 0–20 cycles of ALD, the short Ar+ sputter etch caused a decrease in the F(1s) [C(1s)-CFx] intensities by ∼80% (>90%) on SiO2, and ∼70% (>90%) on Cu.

First in Fig. 8(a), we consider the integrated (absolute) intensity of the F(1s) peak as a function of the number of ALD cycles for the two SAMs on SiO2. The results for the two SAMs are somewhat similar, with both showing a strong decrease in intensity with increasing number of ALD cycles. The dotted lines are the prediction of the model fit to the Ta(4d5/2) data shown in Fig. 5(a), where we have also assumed that the TaNx overlayer grows uniformly in thickness on the SAM covered substrate in a 2D layer-by-layer mode. The assumption of 2D layer-by-layer growth of TaNx on the SAM surfaces can be confirmed or disregarded with other analyses, which we will consider below. We see that the model is a reasonable description of the data for both SAMs. In Fig. 8(b), we consider similar results for the C(1s)-CFx intensities. These data show strong similarities to those we have just considered for the F(1s) peak. Again, the dotted lines are the prediction of the model fit to the Ta(4d5/2) data shown in Fig. 5(a), again assuming uniform 2D layer-by-layer growth of TaNx. The data for FOTS|SiO2 are in reasonable agreement with the model, whereas there is a sizable difference between the model and the results for HDFTEOS|SiO2. The decay observed experimentally is noticeably faster than what the model predicts. These data could indicate cleavage of C–F bonds due to the ALD process, where F may be transferred to other elements present in the thin film (e.g., Ta).

Fig. 8.

(Color online) Integrated intensities of the peaks from XPS associated with the SAMs as a function of the number of ALD cycles for the deposition of TaNx at Ts = 300 °C on FOTS and HDFTEOS on SiO2, (a) F(1s), (b) C(1s)-CFx; and on FOTS and HDFTEOS on Cu, (c) F(1s), (d) C(1s)-CFx. In all cases, the solid lines are to guide the eye. The dashed lines represent a prediction of the fits of the Ta(4d5/2) data shown in Fig. 5 where the SAM is assumed to be buried uniformly by the deposited TaNx thin film.

Fig. 8.

(Color online) Integrated intensities of the peaks from XPS associated with the SAMs as a function of the number of ALD cycles for the deposition of TaNx at Ts = 300 °C on FOTS and HDFTEOS on SiO2, (a) F(1s), (b) C(1s)-CFx; and on FOTS and HDFTEOS on Cu, (c) F(1s), (d) C(1s)-CFx. In all cases, the solid lines are to guide the eye. The dashed lines represent a prediction of the fits of the Ta(4d5/2) data shown in Fig. 5 where the SAM is assumed to be buried uniformly by the deposited TaNx thin film.

Close modal

In Fig. 8(c), we consider the integrated (absolute) intensity of the F(1s) peak as a function of the number of ALD cycles for the two SAMs on Cu. Unlike the results for SiO2, here we observe a large difference between the two SAMs, consistent with the results presented in Fig. 3. After ∼40 cycles of TaNx ALD, we see that there is no evidence of the presence of the HDFTEOS SAM on Cu. For FOTS, however, the SAM is clearly present after 50 cycles. Again, the dotted line is the prediction of the model fit to the Ta(4d5/2) data shown in Fig. 5(b), where a simple exponential decay could explain both the data for Ta(4d5/2) and Cu(2p) [Fig. 7(b), albeit a different rate of decay]. We see that the model seems to slightly over predict the rate of decay of the F(1 s) intensity. Moving on to the results for the C(1s)-CFx intensities displayed in Fig. 8(d), we see that these results mostly mirror those for the F(1s) peak. The biggest difference may be associated with the larger difference observed between the results for FOTS and the prediction of the model based on the fit to the Ta(4d5/2) data shown in Fig. 5(b). Some of this behavior could be explained by the larger than expected F-to-CFx ratio (∼2.4) observed for 0 cycles and FOTS|Cu, which could again be reconciled by C–F bond cleavage and transfer of F to the Cu substrate. Thus, the data shown in Fig. 8(d) for the C(1s)-CFx intensities may better represent the presence of the FOTS SAM. If this is the case, the reduced rate of decay could be explained by FOTS being at the surface, as opposed to being buried by the TaNx thin film. Additional information concerning the mechanism of growth may be found by considering results from other analyses, which we now consider.

3. Low-energy ion scattering spectroscopy

We now consider results from LEISS, which is extremely surface sensitive, detecting only those elements that exist in the topmost monolayer. In all cases, these samples were subjected to a short time (∼1 min) exposure to a 3 keV Ar+ ion beam to remove adventitious carbon, which has a significant detrimental effect on all features observed in LEISS. Example spectra, using He+ ions (∼1 keV), are provided in the supplementary material.29 In Fig. 9, we present the integrated intensities of the Ta and Si peaks from LEISS as a function of the number of ALD cycles for growth of TaNx on (a) bare SiO2 and (b) SiO2 terminated by the two SAMs. First, we see that on clean SiO2 the Ta and Si signals are essentially mirror images of each other: there is a sharp linear increase (decrease) in the Ta (Si) signal up to 20 cycles, followed by a region of nearly constant intensity. This behavior indicates that the TaNx thin film formed on SiO2 is continuous after ∼20 cycles of ALD, or ∼0.8 nm in thickness. On the SiO2 surfaces terminated by the SAMs, the behavior of the LEISS signals is qualitatively similar in the sense that decreases in the substrate Si signal are matched by increases in the thin film Ta signal. The rate of the increase (decrease) in the Ta (Si) signal is much slower than that observed on bare SiO2. For growth on HDFTEOS|SiO2, for example, the substrate feature is only extinguished after ∼200 cycles of ALD growth. This result is consistent with that suggested by the results from XPS, which we considered in Figs. 5 and 7. There are important differences observed from the LEISS results; however, after 80 cycles of growth, the substrate is clearly not covered in the case of HDFTEOS|SiO2, whereas the results from XPS for 80 cycles of growth were quite similar for growth on SiO2 and HDFTEOS|SiO2.

Fig. 9.

(Color online) Integrated intensities of the peaks from LEISS associated with the thin film, Ta (left ordinate), and substrate, Si and Cu (right ordinate) as a function of the number of ALD cycles for the deposition of TaNx at Ts = 300 °C on (a) clean SiO2; on (b) FOTS and HDFTEOS on SiO2, on (c) bare Cu; and on (d) FOTS and HDFTEOS on Cu. In (c) and (d), the behavior of the Ta LEISS signal for growth on clean SiO2 is reproduced.

Fig. 9.

(Color online) Integrated intensities of the peaks from LEISS associated with the thin film, Ta (left ordinate), and substrate, Si and Cu (right ordinate) as a function of the number of ALD cycles for the deposition of TaNx at Ts = 300 °C on (a) clean SiO2; on (b) FOTS and HDFTEOS on SiO2, on (c) bare Cu; and on (d) FOTS and HDFTEOS on Cu. In (c) and (d), the behavior of the Ta LEISS signal for growth on clean SiO2 is reproduced.

Close modal

In Fig. 9, we present the integrated intensities of the Ta and Cu peaks from LEISS as a function of the number of ALD cycles for growth of TaNx on (c) bare Cu and (d) Cu terminated by the two SAMs. For bare Cu, the intensity of the Ta and Cu peaks mirror each other, exhibiting a steady increase (decrease) in the Ta (Cu) signal. Saturation is apparent at 200 cycles, where the substrate Cu signal is finally extinguished. For comparison, the intensity for the Ta peak observed for TaNx growth on clean SiO2 is also plotted in Fig. 9(c) and we see that the behavior on Cu is quite different. In fact, the dependence of the thin film (Ta) and substrate (Cu) peaks is much more similar to that observed on the SiO2 surfaces covered with the two SAMs. In Fig. 9(d), we display the results for TaNx growth on Cu terminated by the two SAMs. Unlike the results for SiO2 and bare Cu, here the intensities of the thin film (Ta) and the substrate (Cu) do not mirror each other so well. For both surfaces terminated by the SAMs, there is an initial period (up to ∼30–40 cycles) where there is little change in the substrate (Cu) signal, while there is a strong increase in the thin film (Ta) signal. These results suggest that the deposited TaNx thin film may be replacing the SAM on the surface, either by desorption or displacement. The other feature that is obvious in Fig. 9(d) is the substantial difference in the intensity of the substrate (Cu) signal depending on the SAM. The Cu intensity for HDFTEOS|Cu for the first 40 cycles is essentially that for bare Cu. The much lower intensity for the Cu peak for FOTS|Cu is entirely consistent with the presence of FOTS (or molecular fragments) at the surface.

We have presented results for the growth of TaNx by ALD on two different substrates, SiO2 and Cu, examining both the clean/bare surfaces and those that have been terminated with two SAMs. To this point, we have employed four different analytical techniques, and these have been applied to characterization of the substrates, the modifying SAMs, and the thin films. In this section, we will attempt to reconcile all of these results in terms of a mechanistic model that is consistent with all of the data.

Growth of TaNx by ALD on SiO2, and, in particular, chemical oxide, is a well-characterized system. Indeed, in previous work,33 we examined the ALD growth of ∼0.5 (10 cycles) and 1.0 nm (20 cycles) thick TaNx on clean SiO2. In this previous work, we found results from XPS and LEISS that were consistent with 2D layer-by-layer growth. The results we present here are consistent with this picture. Namely, from both ellipsometry and XPS, there is no indication of an incubation period, and the thickness (D) deposited per cycle (n), dD/dn, is approximately constant. The dependencies of the integrated intensity from XPS for the thin film [Ta(4d5/2)] and the substrate [Si(2p)] with the number of ALD cycles are very consistent with each other. Finally, from LEISS, the signal associated with the thin film saturates at ∼20 cycles (D ∼ 0.8–0.9 nm), and the signal from the Si substrate decreases approximately linearly over this same range.

Growth of TaNx by ALD on SiO2 surfaces modified by the SAMs is clearly different from clean SiO2. First, we have confirmed using contact angle measurements and XPS that both SAMs on SiO2 examined here appear to remain largely intact after annealing to the ALD process temperature of ∼300 °C. In many previous studies, this issue—the survival of the SAMs at the ALD process temperature—goes mostly unexamined. Their presence is further verified by their effect on subsequent ALD growth of TaNx. The amount of TaNx deposited for the first 20 cycles of ALD is greatly reduced by the SAMs, but after about 40–50 cycles, the thickness deposited per cycle begins to approach that for clean SiO2.

As indicated above, we used a model employed in previous work31,35–38 to describe the initial attenuation of the thickness deposited per cycle (dD/dn = D′). As described in the supplementary material,29 this model assumes that growth is initially αD, which decays exponentially [exp(−n/m)] to the steady state rate of D. To fit the data shown in Fig. 5(a), we also have accounted for the attenuation of the photoelectrons by the depositing thin film. As may be seen, the fits to the data using this model shown in Fig. 5(a) describe the data quite well. The values determined for the attenuation of the initial thickness deposited per cycle (α) are 1 as expected, but the exact values are not particularly significant as the uncertainty in these values are larger than the values themselves. The values for the constant that characterizes the length of the attenuated growth period (m) are more meaningful, and we find m ∼ 27 and 75 cycles for HDFTEOS|SiO2 and FOTS|SiO2. This model predicts that the growth rate will be within ∼5% and essentially indistinguishable of the steady-state growth rate after n ∼ 3m cycles, which for HDFTEOS|SiO2 is ∼80 cycles.

The most logical interpretation of the initial attenuation of the thickness deposited per cycle involves removal of active sites for nucleation by the chemisorption of the molecules that constitute the SAM. Based on the steady-state growth rate and assuming deposition could be described by growth of cubic TaN along the (111) direction, there is a density of Ta of 2.0 × 1014 sites-cm2 at steady-state. This density is close to and somewhat smaller than the density of the SAMs themselves as may be seen from the results presented in Table I. The SAMs, if uniform in density, could likely block ALD growth. Thus, it is likely that growth nucleates at defect sites that are present in the SAM, e.g., areas of low local concentration, and these are at densities of much less than 1014 sites-cm−2. In this case, growth will at first be limited to defect sites, and it could be complex, depending on the fate of the SAM. Provided the SAMs are stable during ALD, thin film growth should first be mostly vertical, perpendicular to the surface. Once the ALD thin film has grown to a thickness comparable to the surrounding SAM (∼1 nm), growth can also proceed in a horizontal direction, parallel to the surface. In this scenario, forming a conformal thin film will depend on the distance between nucleation sites—thus, the point at which the thin film “closes out” (i.e., becomes continuous) could be used to estimate the nucleation density. Close out is indicated directly by the disappearance of the substrate signal from LEISS, and indirectly by the deposited thickness per cycle reaching the steady-state value.

To facilitate the discussion, we consider the schematic in Fig. 10. Here, we present snapshots of layer occupancies for thin film growth that proceeds in a 2D layer-by-layer mode, and one that proceeds in a 3D island growth mode. For the latter example, the layer occupancies have been chosen to be also consistent with random deposition (RD). Also displayed in this figure is a plot of the surface coverage (θ) versus the total coverage (Θ, all layers) for these two modes of growth. As may be seen, for these two growth modes, the predicted behavior, θ = f(Θ), is quite different, and based on these models, the deviation becomes apparent for total coverages ∼0.5 < Θ < 3 ML, and is largest at Θ = 1 ML.

Fig. 10.

(Color online) Simple models for thin film growth involving perfect LbL growth (left panel) and random deposition (right panel), where in the latter case, the depositing species are allowed to coalesce into islands. The central panel is a plot of the surface coverage as a function of the total coverage, which highlights the differences between the two models.

Fig. 10.

(Color online) Simple models for thin film growth involving perfect LbL growth (left panel) and random deposition (right panel), where in the latter case, the depositing species are allowed to coalesce into islands. The central panel is a plot of the surface coverage as a function of the total coverage, which highlights the differences between the two models.

Close modal

In Fig. 11, we consider a plot of the intensity from LEISS for the Ta peak versus the TaNx thin film thickness based on an analysis of the data from XPS shown in Fig. 5(a). In this case, the data as plotted account for attenuation effects in XPS and conform to the analysis presented in Fig. 10. From these data, we see that the results for growth on SiO2 are very indicative of 2D layer-by-layer growth, where close-out of the TaNx thin film occurs at a thickness of ∼7.5 Å. This thickness corresponds to the diagonal of the cubic TaN unit cell, i.e., 3 MLs along the TaN(111) direction. Thus, for this system, the depiction of 2D growth shown in Fig. 10 is accurate in the case where a unit cell thickness, and not a single monolayer, is required for thin film close-out. This is not unexpected as a truly 2D monolayer of TaN may not represent a stable structure (unlike, e.g., graphene), whereas a somewhat thicker thin film may provide the stability provided by more bulk like 3D bonding.

Fig. 11.

(Color online) Integrated intensities of the peaks from LEISS associated with the thin film, Ta, as a function of the thin film thickness for the deposition of TaNx at Ts = 300 °C on clean SiO2 and FOTS and HDFTEOS on SiO2. Here, the TaNx thin film thicknesses are that predicted by the fits to the Ta(4d5/2) data shown in Fig. 5. This representation is the equivalent to the theoretical expectation given in the central panel shown in Fig. 10.

Fig. 11.

(Color online) Integrated intensities of the peaks from LEISS associated with the thin film, Ta, as a function of the thin film thickness for the deposition of TaNx at Ts = 300 °C on clean SiO2 and FOTS and HDFTEOS on SiO2. Here, the TaNx thin film thicknesses are that predicted by the fits to the Ta(4d5/2) data shown in Fig. 5. This representation is the equivalent to the theoretical expectation given in the central panel shown in Fig. 10.

Close modal

For TaNx growth on HDFTEOS|SiO2, we see that the shape of the curve in Fig. 11 is distinct from that for SiO2; downward curvature is clearly observed, which is very indicative of 3D islanded/RD growth. Thus, combined with the attenuation of the initial rate of growth, these results clearly support islanded growth for this system, where thin film closure occurs between 80 and 200 cycles of ALD in this case. How do our other results support or possibly contradict this viewpoint? First, the decay of the LEISS signal for the Si substrate shown in Fig. 9(b) mirrors that of the increase in the Ta signal, supporting close-out between 80 and 200 cycles. Data from XPS for both the Si substrate [Fig. 7(a)] and the HDFTEOS SAM [Figs. 8(a) and 8(b)] are described reasonably well by the model used to fit the Ta thin film signal [Fig. 5(a)]. Indeed, concerning the fate of the HDFTEOS SAM on SiO2, the behavior of the F(1s) and C(1s)-CFx signals are in good agreement with the model that assumes the SAM layer is eventually buried by the TaNx thin film. If degradation of the SAM were significant, we would expect a faster decay in these signals with the number of ALD cycles. We have seen similar behavior (i.e., the SAM is buried by ALD) concerning Al2O3 growth on top of a surface containing an ultrathin (∼0.5 nm) poly(ethyleneimine) layer.31 

For TaNx growth on FOTS|SiO2, these results are quite similar to those for HDFTEOS|SiO2. For example, downward curvature is also observed for the LEISS Ta signal versus TaNx thickness, as shown in Fig. 11. These data do, however, lay above those for HDFTEOS, indicating that for a nominal thickness of ∼5 Å more of the surface is covered by TaNx in the case of FOTS|SiO2. The drop in the Si LEISS signal after 50 cycles for growth on FOTS|SiO2 is also consistent with a more significant coverage of the substrate by the TaNx thin film. These observations indicate that the growth on FOTS|SiO2 is more 2D than that on HDFTEOS|SiO2, suggesting that growth in-plane on FOTS|SiO2 might be less attenuated for these thicknesses.

Finally, as indicated above, we can use the data from LEISS to estimate a defect density in the case of TaNx growth on HDFTEOS|SiO2. Using the disappearance of the Si LEISS signal at 200 cycles (a minimum) to represent close-out, and assuming growth proceeds at equal rates (∼0.4 Å cycle−1) perpendicular to the surface and parallel to it, we estimate a spacing of defect sites of ∼8 nm, and a defect density of ∼2 × 1012 cm−2. This value is ∼1% of the steady-state density of Ta deposited per cycle and corresponds to a value of α ∼ 0.01, which is within experimental uncertainties for the value of this parameter found by a fit to the data in Fig. 5(a).

Growth of TaNx on Cu, bare or modified, is different from that on SiO2. First, for bare Cu, we found for these conditions that the data from XPS, both from the thin film [Fig. 5(b)] and substrate signals [Fig. 7(b)], indicate a steady-state rate of growth of TaNx that is nevertheless ∼60% of that we find on clean SiO2. These data are described by a model that assumes that the rate of growth is steady, and that no incubation period exists. Second, growth on the surface terminated by the SAMs is essentially the same as that on unmodified Cu. We will attempt to reconcile these observations in the following, starting first with growth on bare Cu.

Based on our prior work,33 the growth of TaNx on Cu is indeed different from that on SiO2. For thin films of ∼5 and 10 Å nominal thickness, we found that the TaNx/Cu interface was not abrupt, that neither thin film was continuous at these thicknesses and that the near surface layer was likely a mixture of TaNx and Cu. The results we present here are consistent with that viewpoint. Unlike clean SiO2, on bare Cu, complete coverage of the substrate has definitely not occurred after 80 cycles of TaNx growth and may not even have occurred after 200 cycles of growth based on the size of the LEISS Ta signal at this point [cf. Fig. 9(c)]. For example, after 10–20 cycles of ALD, there is little change in the Cu LEISS signal, and after 40 cycles, this signal is still about ½ of that for the bare Cu substrate. Thus, these results from LEISS indicate that the surface consists of a mixed TaNx/Cu layer for at least the first 80 cycles. It is possible that clean, low index [e.g., (111)] faces of Cu do not provide active sites for TaNx nucleation, whereas defect sites, and certainly regions where TaNx is present could provide such sites. Thus, it is entirely feasible that the initial rate of ALD on bare Cu may be less than that on clean SiO2, and its rate may continue to be suppressed until the entire Cu substrate is covered. For our conditions, closure of the TaNx thin film on bare Cu may be almost complete at ∼200 cycles where, unfortunately, we are not particularly sensitive to changes in the growth rate due to attenuation effects on the signal from XPS. Thus, based on our calculations of the growth rate from XPS (Ta and Si signals), we estimate that a nominal thickness of ∼35–47 Å after 200 cycles of ALD is required for closure of TaNx on bare Cu.

Growth of TaNx thin films on the Cu substrate modified by the two SAMs are very similar to that found for bare Cu. First, for HDFTEOS|Cu, the reason for this behavior is clear. Based on results from XPS [Figs. 3, 8(c), and 8(d)], it is obvious that the majority of this SAM thermally desorbs at the substrate temperature used here for ALD. An interesting question to ask is why does HDFTEOS desorb and FOTS does not? The two molecules have similar fluorinated alkyl tail groups, –(CH2)2(CF2)nCF3, where n = 5 (FOTS) and 7 (HDFTEOS). The difference in binding between these tail groups based on van der Waals interactions is expected to be small. The answer likely lies in the head groups. If, for example, the Cu surface can activate the Si–Cl bonds in the FOTS head group, but not the Si–OR bonds in the HDFTEOS head group, this could explain the observed difference.

Growth of TaNx on FOTS|Cu occurs in the presence of a significant coverage of the FOTS SAM, or possibly partially dissociated molecular fragments of this molecule. Our results from XPS [Figs. 3, 8(c), and 8(d)] clearly indicate the presence of these species. Why do these fragments apparently affect growth so minimally? First, as discussed above, the bare Cu surface does not possess the intrinsic activity of the SiO2 surface, even after 80 cycles of ALD and a nominal TaNx thin film thickness of ∼14–23 Å. As the TaNx thin film grows, and a mixed TaNx/Cu surface is formed, what is a likely fate for the FOTS SAM? One possibility is that the SAM may segregate to those areas that are Cu rich, ones that may not be effective in nucleating further growth of TaNx. If we look at the XPS results for the FOTS SAM [Figs. 8(c) and 8(d)], we see that the rate at which these signals are attenuated do not match the prediction of the model that assumes TaNx overgrowth, namely, the intensity from XPS is higher than what the model predicts. Also, we see from the LEISS results [Fig. 9(d)] that the substrate Cu signal is strongly attenuated, even for 0–30 cycles of ALD on FOTS|Cu—this signal is about 30% of that observed for bare Cu and HDFTEOS|Cu for this region of growth. These results strongly suggest that the FOTS SAM is located mostly in the Cu-rich areas of the surface, and these act to attenuate the Cu LEISS signal, while those signals associated with XPS and the SAM are not strongly attenuated.

In Fig. 12, we consider a construct similar to that shown in Fig. 11 for growth on clean and SAM-modified SiO2, where we plot the intensity from LEISS for the Ta peak versus the TaNx thin film thickness based on an analysis of the data from XPS shown in Fig. 5(b). Again these data for growth on Cu are plotted to account for attenuation effects in XPS, conforming to the analysis as presented in Fig. 10. We see that these data for all three cases of growth on Cu exhibit downward curvature, which is very indicative of 3D islanded/RD growth. Closer examination of these data indicates a strong similarity to the results for growth on SAMs|SiO2. Unlike the growth of TaNx on SAMs|SiO2, however, the amount of material deposited per cycle on Cu, clean or modified, does not change markedly for the number of ALD cycles examined here, remaining at a value of ∼60% of that observed on clean SiO2. A constant, relatively high growth rate indicates that TaNx once nucleated on Cu does not provide the only sites for further growth; otherwise, the rate of growth should accelerate as it does on SAMs|SiO2. In particular, areas of Cu surrounding TaNx islands must remain active for adsorption/growth. For this mechanism to be in play, the material that is added in each cycle in these surrounding areas must eventually join the existing TaNx islands via transport and incorporation. For growth to remain self-limiting, this transport most likely occurs after the half cycle involving Ta[N(CH3)2]5, possibly during the half cycle involving the NH3 exposure.

Fig. 12.

(Color online) Integrated intensities of the peaks from LEISS associated with the thin film, Ta, as a function of the thin film thickness for the deposition of TaNx at Ts = 300 °C on bare Cu and FOTS and HDFTEOS on Cu. Here, the TaNx thin film thicknesses are that predicted by the fits to the Ta(4d5/2) data shown in Fig. 5. This representation is the equivalent to the theoretical expectation given in the central panel shown in Fig. 10. The behavior of the Ta LEISS signal for growth on clean SiO2 is reproduced.

Fig. 12.

(Color online) Integrated intensities of the peaks from LEISS associated with the thin film, Ta, as a function of the thin film thickness for the deposition of TaNx at Ts = 300 °C on bare Cu and FOTS and HDFTEOS on Cu. Here, the TaNx thin film thicknesses are that predicted by the fits to the Ta(4d5/2) data shown in Fig. 5. This representation is the equivalent to the theoretical expectation given in the central panel shown in Fig. 10. The behavior of the Ta LEISS signal for growth on clean SiO2 is reproduced.

Close modal

Is there evidence for transport in the case of growth on Cu? Unlike SiO2, where strong covalent linkages, e.g., Ta–O–Si, will form on chemisorption of the Ta complex, no such bonds will form on the Cu surface. In previous work,33 we have shown that growth of TaNx on Cu results in significant smoothening of the surface, and it is more than is observed to occur based on simple annealing to the ALD process temperature in either an Ar or NH3 atmosphere. Moreover, the amount of smoothening can only be reconciled if transport of Cu is also occurring. The picture that we are left with concerning TaNx growth on Cu, both on bare and SAM modified surfaces, is a dynamic one, where species are free to diffuse during the ALD process, perhaps most effectively during the NH3 half-cycle. Proof of such a mechanism would likely require in situ surface sensitive measurements for each half-cycle, experiments that are beyond the scope of the work presented here. Finally, concerning an ultimate goal of selective growth on a patterned Cu/SiO2 surface, for either SAM examined here, selective area growth would likely be achieved but would be limited to TaNx thin films on the order of 1 nm.

To conclude, we present in schematic form in Fig. 13 our understanding of the effects of the SAMs examined here on the ALD growth of TaNx on the SiO2 and Cu surfaces. For growth on clean SiO2, as may be seen in Fig. 13(a), the thin film roughness indicated by the cartoon is consistent with our previously reported values (∼0.7–0.8 ML RMS) for this system for similar thicknesses.33 On clean SiO2, the TaNx thin film grows mostly in a 2D layer-by-layer mode, but not strictly in the way that epitaxial growth is observed in many thin film systems where each (n) layer essentially completes before the next (n + 1) layer starts to grow. In epitaxial systems LbL growth is facilitated by interlayer transport, coupled with preferred incorporation at step edges. These features likely play a limited role in ALD, which leads to more random deposition and the observation of a continuous thin film only after the deposition of ∼3 MLs of TaNx.

Fig. 13.

(Color online) Schematic representation of the early stages (0–40 cycles) of ALD growth of TaNx for the systems examined here, where these representations are designed to replicate our results from both XPS and LEISS. The systems fall into three classes: (a) quasi-2D LbL growth on SiO2; (b) islanded 3D growth on SAMs|SiO2 where the TaNx thin film eventually overgrows the SAM; and (c) growth on Cu where on all surfaces growth is 3D and islanded, and the TaNx|Cu interface is diffuse.

Fig. 13.

(Color online) Schematic representation of the early stages (0–40 cycles) of ALD growth of TaNx for the systems examined here, where these representations are designed to replicate our results from both XPS and LEISS. The systems fall into three classes: (a) quasi-2D LbL growth on SiO2; (b) islanded 3D growth on SAMs|SiO2 where the TaNx thin film eventually overgrows the SAM; and (c) growth on Cu where on all surfaces growth is 3D and islanded, and the TaNx|Cu interface is diffuse.

Close modal

For the ALD of TaNx on the SiO2 surface terminated by the SAMs, growth begins at defect sites, where free –OH(a) is present, as indicated schematically in Fig. 13(b). For the early stages of growth, 0–20 cycles of ALD, the amount deposited per cycle is small, and a small fraction of the surface is covered by the TaNx thin film after 20 cycles of growth. On bare SiO2, in contrast, the surface is nearly entirely covered by the TaNx thin film after 20 cycles of ALD. After about 20 cycles of ALD, growth accelerates, and the amount deposited per cycle begins to approach that observed on clean SiO2, after about 40 cycles of ALD. Although the kinetics of growth indicated by both XPS and spectroscopic ellipsometry indicate that steady-state conditions have been achieved after 40 cycles, and certainly after 80 cycles, the results from LEISS indicate that the growth is very 3D and islanded for growth on these SiO2 surfaces terminated by SAMs. For example, although approximately the same amount of TaNx has been deposited after 40 cycles on SAM|SiO2 and 20 cycles on clean SiO2, only ∼1/3 of the surface has been covered for the former case. Based on our results, both SAMs, FOTS and HDFTEOS, are essentially stable at the substrate temperature used here (300 °C) for ALD growth. Moreover, the rates of decay of the signals from XPS uniquely associated with fluorinated SAMs are consistent with TaNx overgrowth of these organic monolayers, and only perhaps modest degradation during ALD.

For the ALD of TaNx on the Cu surface, both on bare and Cu surfaces terminated by the SAMs, the situation is quite different as displayed in Fig. 13(c). Consistent with our previously reported work on clean Cu,33 growth differs from clean SiO2 in that the growth is not 2D LbL, and the interface that is formed is not abrupt. First, the polycrystalline Cu thin films are not nearly as smooth as the SiO2 surface examined here (for 0 cycles, annealed samples, RMS roughness is ∼3 nm for Cu and ∼0.2 nm for SiO2), which is indicated schematically in Fig. 13(c). The initial rate of growth is about 60% of that observed on SiO2, and it remains approximately constant up to the point where we are still sensitive to the growth rate measured using XPS (<80 cycles or <2 nm). Results from LEISS indicate 3D islanded growth, where the surface is not completely covered by the TaNx thin film until 200 cycles of ALD. Concerning the effect of the SAMs on ALD growth on Cu, we observe with both SAMs that there is essentially no effect. In the case of HDFTEOS, results from XPS indicate that essentially all of this SAM thermally desorbs from the Cu surface at the substrate temperature examined here for ALD (300 °C), thus no effect. For FOTS, however, this SAM is largely retained on the Cu surface at the ALD process temperature examined here. Indeed, from XPS, signals uniquely associated with this fluorinated SAMs are larger than one would expect if a TaNx thin film were formed uniformly across the surface and buried the SAM. One possible explanation for this observation is that the growing TaNx thin film mostly displaces the SAM to the Cu rich parts of the surface. The difference observed between the two SAMs on Cu likely reflects the head group chemistry, where the Cu surface can activate the Si–Cl bonds in FOTS, while it does not activate Si–OR bonds nearly as effectively. Finally, the bound FOTS molecules or molecular fragments are likely mobile on the Cu surface, where they apparently have little effect on the growth kinetics. This observation could reflect the differences between the very directional covalent bonds between the SAMs and the SiO2 surface versus the metallic bonding that occurs on Cu.

This research was supported in part by the Semiconductor Research Corporation (Task 2149.001). This work was performed in part at the Cornell NanoScale Facility, a member of the National Nanotechnology Infrastructure Network, which was supported by the National Science Foundation (Grant No. ECCS-0335765). The authors would also like to thank Clay T. Long for technical contributions.

2.
O.
Van der Straten
,
S. M.
Rossnagel
,
J. P.
Doyle
, and
K. P.
Rodbell
,
ECS Trans.
1
,
51
(
2006
).
3.
X.
Jiang
and
S. F.
Bent
,
J. Phys. Chem. C
113
,
17613
(
2009
).
4.
P.
Laibinis
and
G.
Whitesides
,
J. Am. Chem. Soc.
113
,
7152
(
1991
).
5.
H.
Kim
,
H.-B.-R.
Lee
, and
W.-J.
Maeng
,
Thin Solid Films
517
,
2563
(
2009
).
6.
A. J. M.
Mackus
,
A. A.
Bol
, and
W. M. M.
Kessels
,
Nanoscale
6
,
10941
(
2014
).
7.
F. S. M.
Hashemi
,
C.
Prasittichai
, and
S. F.
Bent
,
J. Phys. Chem. C
118
,
10957
(
2014
).
8.
C.
Prasittichai
,
K. L.
Pickrahn
,
F. S. M.
Hashemi
,
D. S.
Bergsman
, and
S. F.
Bent
,
ACS Appl. Mater. Interfaces
6
,
17831
(
2014
).
9.
M. H.
Park
,
Y. J.
Jang
,
H. M.
Sung-Suh
, and
M. M.
Sung
,
Langmuir
20
,
2257
(
2004
).
10.
E. K.
Seo
,
J. W.
Lee
,
H. M.
Sung-Suh
, and
M. M.
Sung
,
Chem. Mater.
16
,
1878
(
2004
).
11.
X.
Jiang
and
S. F.
Bent
,
J. Electrochem. Soc.
154
,
D648
(
2007
).
12.
K. J.
Park
,
J. M.
Doub
,
T.
Gougousi
, and
G. N.
Parsons
,
Appl. Phys. Lett.
86
,
051903
(
2005
).
13.
R.
Chen
,
H.
Kim
,
P. C.
McIntyre
, and
S. F.
Bent
,
Appl. Phys. Lett.
84
,
4017
(
2004
).
14.
H.-B.-R.
Lee
,
W.-H.
Kim
,
J. W.
Lee
,
J.-M.
Kim
,
K.
Heo
,
I. C.
Hwang
,
Y.
Park
,
S.
Hong
, and
H.
Kim
,
J. Electrochem. Soc.
157
,
D10
(
2010
).
15.
M.
Li
,
M.
Dai
, and
Y. J.
Chabal
,
Langmuir
25
,
1911
(
2009
).
16.
K.
Park
,
Y.
Lee
,
K. T.
Im
,
J. Y.
Lee
, and
S.
Lim
,
Thin Solid Films
518
,
4126
(
2010
).
17.
H.-B.-R.
Lee
and
H.
Kim
,
ECS Trans.
16
,
219
(
2008
).
18.
A.
Amassian
,
V. A.
Pozdin
,
T. V.
Desai
,
S.
Hong
,
A. R.
Woll
,
J. D.
Ferguson
,
J. D.
Brock
,
G. G.
Malliaras
, and
J. R.
Engstrom
,
J. Mater. Chem.
19
,
5580
(
2009
).
19.
J.
Huang
,
M.
Lee
, and
J.
Kim
,
J. Vac. Sci. Technol. A
30
,
01A128
(
2012
).
20.
A. S.
Killampalli
,
P. F.
Ma
, and
J. R.
Engstrom
,
J. Am. Chem. Soc.
127
,
6300
(
2005
).
21.
M. J.
Preiner
and
N. A.
Melosh
,
Langmuir
25
,
2585
(
2009
).
22.
R.
Chen
,
H.
Kim
,
P. C.
McIntyre
,
D. W.
Porter
, and
S. F.
Bent
,
Appl. Phys. Lett.
86
,
191910
(
2005
).
23.
R.
Chen
,
H.
Kim
,
P. C.
McIntyre
, and
S. F.
Bent
,
Chem. Mater.
17
,
536
(
2005
).
24.
J.
Hong
,
D. W.
Porter
,
R.
Sreenivasan
,
P. C.
McIntyre
, and
S. F.
Bent
,
Langmuir
23
,
1160
(
2007
).
25.
R.
Chen
and
S. F.
Bent
,
Chem. Mater.
18
,
3733
(
2006
).
26.
E.
Färm
,
M.
Vehkamäki
,
M.
Ritala
, and
M.
Leskelä
,
Semicond. Sci. Technol.
27
,
074004
(
2012
).
27.
O.
Seitz
,
M.
Dai
,
F. S.
Aguirre-Tostado
,
R. M.
Wallace
, and
Y. J.
Chabal
,
J. Am. Chem. Soc.
131
,
18159
(
2009
).
28.
E.
Färm
,
M.
Kemell
,
M.
Ritala
, and
M.
Leskelä
,
Chem. Vap. Deposition
12
,
415
(
2006
).
29.
See supplementary material at http://dx.doi.org/10.1116/1.4931722 for further details concerning substrate preparation, SAM formation, process and equipment details concerning ALD, and experimental procedures involving spectroscopy ellipsometry, XPS, and LEISS. Selected results concerning calculations of the absolute densities of the SAMs, XPS of the C(1s) region, the effects of sputtering on XPS, determination of the C/F elemental ratio in the SAMs, XPS of the Cu substrate, the substrate temperature dependence of TaNx ALD, determination of the N/Ta ratio of the TaNx thin films, “raw” LEISS spectra, and the model used to fit the thickness deposited per cycle.
30.
S. E.
Roadman
,
N.
Maity
,
J. N.
Carter
, and
J. R.
Engstrom
,
J. Vac. Sci. Technol. A
16
,
3423
(
1998
).
31.
K. J.
Hughes
and
J. R.
Engstrom
,
J. Vac. Sci. Technol. A
30
,
01A102
(
2012
).
32.
A.
Satta
,
J.
Schuhmacher
,
C. M.
Whelan
,
W.
Vandervorst
,
S. H.
Brongersma
,
A.
Vantomme
,
M. M.
Viitanen
,
H. H.
Brongersma
, and
W. F. A.
Besling
,
J. Appl. Phys.
92
,
7641
(
2002
).
33.
W.
Zhang
,
R. K.
Nahm
,
P. F.
Ma
, and
J. R.
Engstrom
,
J. Vac. Sci. Technol. A
31
,
061101
(
2013
).
34.
T. V.
Desai
,
S.
Hong
,
A. R.
Woll
,
K. J.
Hughes
,
A. P.
Kaushik
,
P.
Clancy
, and
J. R.
Engstrom
,
J. Chem. Phys.
134
,
224702
(
2011
).
35.
K. J.
Hughes
,
A.
Dube
,
M.
Sharma
, and
J. R.
Engstrom
,
J. Phys. Chem. C
116
,
21948
(
2012
).
36.
A.
Dube
,
M.
Sharma
,
P. F.
Ma
, and
J. R.
Engstrom
,
Appl. Phys. Lett.
89
,
164108
(
2006
).
37.
A.
Dube
,
M.
Sharma
,
P. F.
Ma
,
P. A.
Ercius
,
D. A.
Muller
, and
J. R.
Engstrom
,
J. Phys. Chem. C
111
,
11045
(
2007
).
38.
K. J.
Hughes
and
J. R.
Engstrom
,
J. Vac. Sci. Technol. A
28
,
1033
(
2010
).

Supplementary Material