6106
Langmuir 2007, 23, 6106-6112
In Situ Auger Electron Spectroscopy Study of Atomic Layer Deposition: Growth Initiation and Interface Formation Reactions during Ruthenium ALD on Si-H, SiO2, and HfO2 Surfaces Kie Jin Park, David B. Terry, S. Michael Stewart, and Gregory N. Parsons* Department of Chemical and Biomolecular Engineering, North Carolina State UniVersity, Raleigh, North Carolina 27695 ReceiVed June 30, 2006. In Final Form: February 28, 2007 Growth initiation and film nucleation in atomic layer deposition (ALD) is important for controlling interface composition and achieving atomic-scale films with well-defined composition. Ruthenium ALD is studied here using ruthenocene and oxygen as reactants, and growth initiation and nucleation are characterized on several different growth surfaces, including SiO2, HfO2, and hydrogen terminated silicon, using on-line Auger electron spectroscopy and ex-situ X-ray photoelectron spectroscopy. The time needed to reach the full growth rate (typically ∼1 Å per deposition cycle) is found to increase as the surface energy of the starting surface (determined from contact angle measurements) decreased. Growth starts more readily on HfO2 than on SiO2 or Si-H surfaces, and Auger analysis indicates distinct differences in surface reactions on the various surfaces during film nucleation. Specifically, surface oxygen is consumed during ruthenocene exposure, so the nucleation rate will depend on the availability of oxygen and the energetics of surface oxygen bonding on the starting substrate surface.
I. Introduction Ruthenium metal has a relatively high metal work function (4.7-5.2 eV), good thermal stability, and low resistivity in both reduced and oxidized states. These properties make it valuable for many possible applications in metal oxide semiconductor (MOS) device fabrication, including metal gate electrodes to replace common polycrystalline silicon gates,1-3 and nucleation seed layers for copper interconnect formation.4 Ruthenium is also considered for capacitor electrodes in dynamic and ferroelectric random access memories.5,6 Ruthenium is widely deposited by various deposition techniques, including physical vapor deposition,7 chemical vapor deposition,8 and atomic layer deposition (ALD).3,4,9,10 Atomic layer deposition is of particular interest because it enables very high conformality and, in principle, atomic level control over film thickness. In general, the use of highly conformal very thin films in ALD processing requires better understanding of interfaces and interface processing. While many different precursors and processes have been studied for Ru ALD, significant concerns remain regarding the mechanisms associated with metal nucleation and growth initiation. Nucleation is an important issue in ALD,11 and it is well-known that Ru (1) Wen, H.; Lysaght, P.; Alshareef, H.; Huffman, C.; Harris, H.; Choi, K.; Senzaki, Y.; Luan, H.; Majhi, P.; Lee, B.; Campin, M.; Foran, B.; Lian, G.; Kwong, D. J. Appl. Phys. 2005, 98, 043520. (2) Zhong, H.; Heuss, G.; Misra, V.; Luan, H.; Lee, C.; Kwong, D. Appl. Phys. Lett. 2001, 78, 1134. (3) Park, K. J.; Doub, J. M.; Gougousi, T.; Parsons, G. N. Appl. Phys. Lett. 2005, 86. (4) Kim, H. J. Vac. Sci. Technol., B 2003, 21, 2231-2261. (5) Aoyama, T.; Kiyotoshi, M.; Yamazaki, S.; Eguchi, K. Jpn. J. Appl. Phys., Part 1 1999, 38, 2194-2199. (6) Bandaru, J.; Sands, T.; Tsakalakos, L. J. Appl. Phys. 1998, 84, 11211125. (7) Krusinelbaum, L.; Wittmer, M. J. Electrochem. Soc. 1988, 135, 26102614. (8) Park, S. E.; Kim, H. M.; Kim, K. B.; Min, S. H. J. Electrochem. Soc. 2000, 147, 203-209. (9) Aaltonen, T.; Alen, P.; Ritala, M.; Leskela¨, M. Chem. Vap. Deposition 2003, 9, 45. (10) Kwon, O. K.; Kim, J. H.; Park, H. S.; Kang, S. W. J. Electrochem. Soc. 2004, 151, G109-G112.
Figure 1. The schematic of the cold-wall ALD reactor connected with the UHV analysis chamber for Auger electron spectroscopy.
ALD growth begins differently on different surfaces,12 but the general mechanisms that relate substrate surface properties to film nucleation are not well understood. In this article, the substrate dependence of ALD Ru film nucleation is studied, including analysis of surface roughness, growth rate, and growth initiation time. In addition, we use a unique ALD reactor connected to a UHV surface analysis tool to examine and quantify growth of the first few monolayers of ruthenium on various substrates, including hydrogen-terminated silicon (Si-H), SiO2, and HfO2. The results obtained give important insight into the role of oxygen on the growth surface in enabling and promoting Ru film nucleation. Results also suggest that nucleation can consume oxygen at the dielectric/metal interface which may be important for controlling electronic defects and charge at the interface. (11) Green, M. L.; Ho, M.-Y.; Busch, B.; Wilk, G. D.; Sorsch, T.; Conard, T.; Brijs, B.; Vandervorst, W.; Raisanen, P. I.; Muller, D.; Bude, M.; Grazul, J. J. Appl. Phys. 2002, 92, 7168. (12) Haukka, S.; Tuominen, M. Proceeding of AVS-ALD2005, August 2005, San Jose, U.S.A. 2005.
10.1021/la061898u CCC: $37.00 © 2007 American Chemical Society Published on Web 04/27/2007
Ruthenium ALD on Si-H, SiO2, and HfO2 Surfaces
Figure 2. ALD cycle number versus Ru film thickness deposited on chemical SiO2, thermal SiO2, and hydrogen-terminated Si. Solid symbols are thicknesses measured using profilometry, and open symbols refer to thickness estimated from XPS data. The uncertainty in the thickness determined from profilometry is ∼(10%. A larger uncertainty is expected for the results from XPS as indicated by the typical error bar shown.
Figure 3. Ru3p XPS spectra with ALD cycle on (a) hydrogenterminated Si and (b) thermal SiO2
II. Experimental Procedures For this study, atomic layer deposition of ruthenium was carried out in two separate custom-made reactor systems. A hot-wall flow tube reactor was used for the growth rate and growth saturation analysis, and it has been described previously.3 For the nucleation studies, a cold-wall chamber system was used, where the reactor chamber was connected through a load-lock to an ultrahigh vacuum analysis chamber equipped with Auger electron spectroscopy. A schematic of the cold-wall reactor system is shown in Figure 1. In this system, the sample temperature is controlled using a resistive block heater, and gas flows into the top of the reactor and down over the sample stage. Samples are mounted on a 3 in. diameter sample puck and transferred from the deposition zone to the UHV chamber without air exposure using a fork-type sample transfer arm. In the UHV chamber, the sample is moved onto a stage that enables full sample rotation and translation to position the sample surface within ∼1 cm of the electron beam and Auger detector. In both reactor systems, deposition is done using bis-(cyclopetadienyl)ruthenium (ruthenocene, RuCp2) precursor and dry air or oxygen as the second reagent. Argon was used as the precursor
Langmuir, Vol. 23, No. 11, 2007 6107
Figure 4. The full-growth initiation cycle number, determined from the extrapolation of the data in the linear regime in Figure 2, is plotted vs the static water contact angle measured on the starting substrate surface. A smaller surface energy (i.e., larger contact angle) correlates with a longer time to reach the full growth rate. carrier gas. For the hot-wall flow system, growth took place in a heated quartz tube approximately 2 in. in diameter, and the Ar and air gas flow rates were set at 100 standard cm3 per min (sccm). The temperature of the ruthenocene bubbler was fixed at 80 °C, and the processing pressure and substrate temperature were fixed at 1.2 Torr and 325 °C, respectively. In the cold-wall reactor, the precursor and reactant gases were delivered to the ∼2 in. diameter heated growth zone through 1/4 inch stainless steel tubes inside the reactor. Nitrogen was used as the purge gas during the entire process, and all gas flows were set at 100 sccm. The process pressure and substrate temperature were fixed at 1 Torr and 275 °C, respectively. In this reactor, a lower set point temperature of 50 °C was used for the ruthenocene bubbler to reduce the precursor partial pressure. Because this system uses a smaller gas delivery area and the same gas flow rate as the hot wall reactor, a higher precursor partial pressure would lead to a much higher precursor flux on the surface as compared to the hot wall system. Under these conditions, the deposition rate in both reactor systems was approximately 1 Å/cycle which is higher than the rate of 0.5 Å/cycle reported by Aaltonen et al.13,14 We note that deposition rates in excess of 1 Å/sec can be observed in the ALD reaction15 and are consistent with the subsurface oxidation model proposed by Aaltonen.13,14 For the experiments described here, several substrate materials were prepared in our lab, including thermal SiO2, chemical SiO2, HfO2, and hydrogen-terminated silicon (Si-H). For all these substrates, silicon surfaces were oxidized by wet chemical treatment (BakerClean JTB-100), followed by buffered HF acid dip, deionized water rinse, and N2 flow dry. To form the Si-H surface, the silicon was treated with a 1% HF acid dip after the chemical oxidation step, followed by deionized water flush and N2 flow dry. All steps were done immediately prior to loading the sample into the vacuum reactor system. To form thermal SiO2, the wafers were oxidized in air at 900 °C for ∼20 min, resulting in ∼100 Å of SiO2. To form the chemical oxide,11 the substrates were further treated with SC1 solution for 30 min, resulting in ∼5-10 Å of oxide. HfO2 on silicon was formed by sputtering thin Hf metal films, followed by thermal oxidation in N2 (with ∼20 ppm O2) at 600 °C for 1 min.16 The HfSiOx films were deposited by chemical vapor deposition with Hf/Si ratio of approximately 0.6, and were used as received from industry. In the ALD process in either reactor systems, the substrates were transferred into the reactor then heated in vacuum (5 × 10-6 Torr) to the deposition temperature (275 or 325 °C), where they equilibrate for 30 min. The inert carrier gas flow is then initiated and the pressure stabilized, followed by binary cycles of reagent and precursor (13) Aaltonen, T.; Alen, P.; Ritala, M.; Leskela¨, M. Chem. Vap. Deposition 2003, 9, 45-49. (14) Aaltonen, T.; Rahtu, A.; Ritala, M.; Leskela¨, M. Electrochem. Solid-State Lett. 2003, 6, C130-C133. (15) Parsons, G. N.; Park, K. J.; Terry, D.; Stewart, S. M.; Hyde, G. K.; Hojo, D. In Proceeding of the 2006 International Atomic Layer Deposition Conference; AVS: Seoul, Korea, 2006. (16) Gougousi, T.; Parsons, G. N. J. Appl. Phys. 2004, 95, 1391-1696.
6108 Langmuir, Vol. 23, No. 11, 2007
Park et al.
Figure 5. SEM and AFM images for ALD Ru deposited on (a) chemical SiO2, (b) thermal SiO2, and (c) hydrogen-terminated Si. exposure. The pressure was maintained in the reaction zone by a variable orifice valve downstream from the reactor. Some variation in pressure was observed during the first ∼1-2 s of each cycle step because of the limited time response of the pressure control valve. The ALD growth was initiated by flowing the metal precursor, followed by inert gas purge and then the reagent (oxygen or air). A similar procedure is followed in both reactor systems, but the optimized cycle times were dependent on the reactor. In the hot wall system, typical cycle times were: RuCp2 3 s; Ar 20 s; O2 6 s; and Ar 20 s, resulting in ∼1 Å/cycle at 325 °C. In the cold wall system, typical cycle times were as follows: RuCp2 1 s; Ar 25 s; O2 3 s; and Ar 35 s, resulting in ∼1 Å/cycle at 325 °C. Differences observed in optimum growth conditions in the two reactor systems give insight into surface mechanisms in bulk metal ALD (after films are fully nucleated).15 Careful analysis is needed to quantify differences observed in the reactors used here. However, this article focuses on mechanisms involved in growth initiation which are expected to be qualitatively independent of reactor geometry. Static contact-angle analysis with deionized water was used to characterize the surface energy of the initial starting substrates. For films in excess of ∼150 Å, thickness was measured using a Tencor Alpha-Step 500 surface profilometer. For thinner films deposited in the hot wall system, X-ray photoelectron spectroscopy (XPS) was used to characterize composition and estimate film thickness. The photoelectrons were excited using non-monochromatic Mg Ka radiation (hν ) 1253.6 eV). Survey and detailed spectra were collected at a takeoff angle of 75° using 1 and 0.1 eV step sizes, respectively. The adventitious C 1s peak was set to a binding energy of 285.0 eV to compensate for any surface charge. The roughness and surface texture of the deposited Ru films were analyzed on the various substrates surfaces using atomic force microscopy (AFM) and scanning electron microscopy (SEM). Crystalline structure of Ru deposited on various surfaces was analyzed using X-ray diffraction (XRD).
III. Results and Discussion The ruthenium film thickness deposited in the hot wall flow reactor is plotted versus number of ALD cycles for deposition on chemical SiO2, thermal SiO2, and Si-H surface in Figure 2. Data are shown for film thicknesses directly measured by step profilometry, as well as thicknesses estimated from XPS analysis. After approximately 200 cycles of growth, the film thickness is observed to increase linearly with the number of cycles, and the
Figure 6. Root-mean-square surface roughness measured by AFM plotted vs the number of Ru ALD cycles on Si-H and thermal SiO2. The rate at which the roughness increases depends on the starting surface, but the saturation roughness is the same on both surfaces.
slope of the linear increase is found to be ∼1 Å/cycle for growth on Si-H, SiO2 and HfO2 surfaces under the conditions used. For less than ∼200 cycles of growth, the growth rate depends significantly on the nature of the substrate. Film thicknesses in this thin-film regime were estimated from XPS data shown in Figure 3 for Ru signal versus cycle time for growth on Si-H and SiO2 surfaces. For film thickness less than the photoelectron escape depth (∼45 Å at 462 eV), the XPS counts per second is expected to be related to the film thickness. Therefore, to estimate thickness from the XPS data, the intensity of the Ru 3p3/2 peak is plotted as a function of cycle number, and the trend is fit to the results measured by the step profilometry for the thicker samples in Figure 2. The film thickness was also obtained from the XPS results using attenuation length analysis.17 Thickness (t) was estimated from: t ) -λ ln (Io/I∞), where λ is the mean free path of Si 2p photoelectrons (25 Å), I∞ is the Si 2p intensity for a clean silicon surface, and Io is the Si 2p intensity of the silicon substrate peak when an overlayer is present. This approach resulted in thickness values within reasonable experimental variations from data fitting in Figure 2. As a convenient estimate of the time required to reach the bulk growth rate, the linear data for thickness versus cycle number (17) Cumpson, P. J.; Seah, M. P. Surf. Interface Anal. 1997, 25, 430-446.
Ruthenium ALD on Si-H, SiO2, and HfO2 Surfaces
Langmuir, Vol. 23, No. 11, 2007 6109
Figure 7. XRD analysis for ALD Ru films deposited on HfO2, HfSiOx, and thermal SiO2
Figure 8. AES spectra on thermal SiO2 after the O2 exposure step and after the RuCp2 exposure step.
in Figure 2 can be extrapolated to the abscissa to define a fullgrowth initiation time. This value is somewhat arbitrary, but it may be related, for example, to the number of cycles needed to achieve complete nuclei coalescence on the various growth surfaces. Even without a precise physical definition, the fullgrowth initiation time is a simple and convenient parameter to compare growth on the different surfaces. For example, we find that the full-growth initiation time can be phenomenologically correlated to the surface energy of the starting surface, estimated from the static water contact angle, as shown in Figure 4. The contact angle on the chemical SiO2 surface was found to be smaller than that on thermal SiO2. A larger contact angle is observed on the Si-H surface, consistent with a lower surface energy than on the oxidized surfaces. The data show that substrates with higher surface energy (i.e., smaller contact angles) reach full-growth initiation relatively quickly, whereas lower energy surfaces require more ALD cycles to achieve full growth. In previous work,3 we have shown that when low-energy methylterminated self-assembled monolayers with water contact angle of 100° to 110° are exposed to 300 cycles of Ru ALD, no Ru nucleation is observed, suggesting a nearly infinite full-growth initiation time for this surface, consistent with the trend shown in Figure 4. To further characterize differences in film nucleation and growth on different sample surfaces, ruthenium film morphology was compared by AFM and SEM analysis as shown in Figure 5. Figure 5a shows that growth on chemical SiO2 results in a smaller grain size and smoother surface texture as compared to growth on Si-H, shown in Figure 5c, consistent with a more two-dimensional growth morphology on the higher energy SiO2 surface. The root-mean-square (rms) surface roughness was measured for a range of film thicknesses on the substrates of
Figure 9. AES spectra after (a) 2 Ru ALD cycles; (b) 10 cycles; (c) 20 cycles on hydrogen-terminated Si, thermal SiO2, and HfO2. All spectra are shown on the same intensity scale and were collected after the oxygen exposure half-cycle.
interest, and the results are shown in Figure 6. For these measurements, the roughness is measured over relatively large surface areas of 1 µm2. Results in Figure 6 show that roughness increases relatively quickly on the SiO2 surface, whereas it increases more slowly on the Si-H surface. After the full-growth initiation time (from data in Figure 2), the surface roughness increases at the same rate and saturates at the same value on both surfaces, indicating that after film nucleation, the growth mechanism are likely the same on these different surfaces. To compare the bulk properties of Ru deposited on the different surfaces, the crystalline structure of the films was examined by X-ray diffraction, and results are shown in Figure 7. The primary crystalline structure of Ru on SiO2 and HfO2 is found to be (002) with some contribution from (101) structure. The ratio of (002) to (101) peak intensity is very similar for Ru on these surfaces. On HfSiOx, (002) and (101) are also both observed, but there is a much stronger contribution from the (002) facets, suggesting
6110 Langmuir, Vol. 23, No. 11, 2007
Figure 10. AES Ru peak height as a function of ALD cycle number on hydrogen-terminated Si, thermal SiO2, and HfO2. Note a larger rate of increase on HfO2 as compared to SiO2, indicating different nucleation rates on those surfaces.
some difference in bulk structure. It is interesting to note that the surfaces that show the largest difference in nucleation (SiO2 and SiH) show very similar bulk crystallographic orientation. More work is needed on a broader range of samples to better understand the relation between bulk crystal orientation and film growth. The effect of high temperature annealing on the structure of the deposited films was also studied. Films ∼150 Å thick on HfSiOx were heated in argon using rapid thermal annealing at 1000 °C for 10 s, and SEM was compared before and after anneal. The SEM results show no evidence for ruthenium film agglomeration or delamination under these conditions, indicating thermal stability on these surfaces.1 Using the Auger electron spectroscopy system connected to the ALD reactor, it is possible to gain more detailed insight into the substrate dependence of film nucleation and growth. Auger spectroscopy has some advantages over other spectroscopy tools in that it is very surface sensitive, allowing small changes in surface composition to be observed and quantitatively analyzed.18 The capability for direct transfer of the sample from the ALD reaction chamber without exposure to the atmosphere is important to analyze the surface composition without interference from ambient surface contamination. With this reactor set up, for example, clear differences in surface composition are observed when film growth is stopped and the sample is transferred to the AES chamber after the RuCp2 exposure step as compared to samples transferred after the O2 exposure step. Example results comparing AES spectra collected after RuCp2 surface exposure and O2 exposure are shown in Figure 8. For the spectra labeled “After O2”, collected after oxygen exposure, clear evidence of surface O is observed, whereas the spectra labeled “After RuCp2”, the oxygen KLL signal is significantly smaller, and the Ru can C signals become more predominant. The result indicates that surface oxygen is removed during Ru precursor exposure, likely due to oxidation of the Ru precursor ligands, resulting in reduction of the Ru metal and film deposition. These results are consistent with the mechanisms proposed by the Helsinki group for ALD of Ru and other noble metals when oxygen is used as a reagent for ligand oxidation.14 The Auger electron spectroscopy system was used to characterize the Ru film growth as a function of number of growth cycles on each of the surfaces of interest. Figure 9 shows the (18) Somorjai, G. A. Introduction to Surface Chemistry and Catalysis; Wiley: New York, 1994. (19) Childs, K. D.; Carlson, B. A.; LaVanier, L. A.; Moulder, J. F.; F. Paul, D.; Stickle, W. F.; Watson, D. G. Handbook of Auger Electron Spectroscopy; Physical Electronics, Inc: Edenprairie, MN, 1995.
Park et al.
AES signal in the Ru MNN spectral region for growth on Si-H, SiO2, and HfO2 measured after (a) 2, (b) 10, and (c) 20 cycles. All spectra are collected after the oxygen exposure step in the cycle. Note that after only 2 cycles, clear evidence for Ru is observed on the SiO2 and HfO2 surfaces, including the main peak near 270 eV, and satellite peaks at 195 and 225 eV. However, on the Si-H surface, no Ru is observed, consistent with the ex situ XPS and growth rate results presented above. A small feature near 280 is observed, but this is assigned to the C KLL peak at 275 eV, related to adventitious carbon present on the surface before being loaded into the growth chamber. After 10 and 20 cycles, the Ru peak is larger on all surfaces, and the relative difference on the various surfaces becomes less pronounced, consistent with more full coverage of Ru as the number of growth cycles proceeds. Figure 10 shows the Ru MNN signal intensity as a function of number of growth cycles collected from the data in Figure 9 and other similar data sets. A relatively rapid increase in Auger signal strength is observed on HfO2, and a slower increase is seen on Si-H, again consistent with the slower nucleation on the lower surface energy substrate. Further insight into initial film growth can be obtained by analyzing the AES signal of the other chemical species observed during film nucleation and coalescence, and a series of results obtained for Ru ALD is shown in Figure 11. These data show results from Si LVV, Hf LMM, O KLL, C KLL, and Ru MNN, after 2 and 20 cycles on Si-H and HfO2 surfaces, where all spectra are collected after the oxygen exposure step in the cycle. Regarding the data in Figure 11, it is important to note that: 1) the Si or Hf signal from the substrate is not visible after 20 ALD cycles; 2) oxygen is visible on all surfaces after the oxygen exposure step; and 3) the Ru signal increases on all surfaces with the number of growth cycles. Data similar to that show in Figure 11 were collected for different numbers of cycles and for deposition on SiO2, and the results are summarized in Figure 12. The figure shows atomic percentages of each component calculated from the AES peak heights normalized by the tabulated sensitivity factors.19 By this method, the relative magnitude of the AES signal is a reliable measure of changes in surface atom density. In the AES spectra, the primary carbon KLL and ruthenium MNN peaks occur at nearly the same electron kinetic energy, but the second Ru MNN line at 235 eV is readily observed. The amount of carbon in the film, therefore, can be estimated by comparing the relative intensities of the peaks at 273 and 231 with a standard elemental ruthenium spectrum.19 Atomic ratios were determined from the signals due to silicon at 96 eV, oxygen at 510 eV, and hafnium at 1625 eV. This approach may not be optimum for evaluating the precise elemental composition in the near surface region, but consistent analysis procedures applied to the different surfaces will give reliable relative results. The results in Figure 12 show the intensity of the AES signal, measured after the oxygen exposure step, from each element analyzed as a function of ALD cycle for growth on the three surfaces of interest. As mentioned, the Ru peak increases with the number of ALD cycles on all three surfaces, but distinct differences are observed on these surfaces. Specifically, on SiO2 and HfO2, the oxygen signal decreases from its initial starting level, reaching an approximate steady-state value consistent with significant oxygen content on the surface after 20 ALD cycles, consistent with subsurface oxidation during the oxygen exposure step.14 The relative atomic fraction of oxygen on film deposited on HfO2 is seen to decrease from the starting surface somewhat more rapidly with ALD cycle number as compared to the film on SiO2. The data in Figure 12 also show that the carbon level on the SiO2 and HfO2 surfaces is relatively small after the first
Ruthenium ALD on Si-H, SiO2, and HfO2 Surfaces
Langmuir, Vol. 23, No. 11, 2007 6111
Figure 11. AES spectra after 2 and 20 ALD cycles on (a) hydrogen-terminated Si and (b) HfO2. All spectra were collected after the oxygen exposure half-cycle.
Figure 12. Atomic fractions of various surface component species estimated from the AES data in Figure 11 (and other similar data sets) plotted vs the number of ALD cycles on hydrogen-terminated Si, thermal SiO2, and HfO2.
several cycles. However, on the Si-H surface the oxygen signal is observed to increase, then decrease to a relatively low level,
and the carbon level is observed to be significant. This suggests that on the Si-H surface, a larger oxygen exposure is required to achieve full precursor ligand oxidation and metal reduction. Other significant differences can be observed for Ru nucleation on Si-H, SiO2, and HfO2 surfaces. In particular, we note that on Si-H and SiO2, the Si signal monotonically decreases with the number of ALD Ru cycles, consistent with an increasing surface coverage of Ru. On both HfO2 and SiO2 the carbon signal is observed to decrease to below the detection limit within the first two metal deposition cycles. However, after two cycles of metal ALD on HfO2, the Hf signal intensity does not decrease, and may even increase somewhat relative to the signal from the starting surface, before decreasing below the detection limit after ∼6 cycles. This is in contrast to the case of Ru ALD on SiO2, where in the first two cycles the Si signal decreases by a factor of 2 from its starting value. Several possible mechanisms could account for this observation. For example, if mixing were to occur at the dielectric/metal interface during metal deposition, it could be possible for one component of the mixture to phase segregate at the surface, leading to a relatively larger AES signal. However, the thermodynamics of Hf-O, Si-O, Ru-Hf, RuSi, Ru-O, and Ru-Ru bond energies suggests that significant bond dissociation and interfacial mixing between HfO2 and Ru (or SiO2 and Ru) would not be favorable at the temperatures studied. The initial adsorption of the ruthenocene precursor on the starting surface is likely controlled by the availability of oxygen species, typically in the form of isolated or hydrogenbonded hydroxyl groups. Careful analysis of TiO2 nucleation on silica and alumina show that the precursor adsorption is sensitive to the density and type of hydroxyl groups present on the surface.20,21 Moreover, within the first few cycles of ALD, differences in initial surface groups are also expected to affect the initial island formation and structure.21 Likewise for the growth processes studied here, a different density of available oxygen groups on the SiO2 and HfO2 surfaces would result in different (20) Haukka, S.; Lakomaa, E. L.; Suntola, T. Appl. Surf. Sci. 1994, 82/83, 548-552. (21) Lindblad, M.; Haukka, S.; Kyto¨kivi, A.; Lakomaa, E. L.; Rautiainen, A.; Suntola, T. Appl. Surf. Sci. 1997, 121/122, 286-291.
6112 Langmuir, Vol. 23, No. 11, 2007
Ru surface coverage and a different density and structure of nuclei on the surface. Because of the higher oxygen affinity of hafnium versus silicon, it is likely that there is a larger number of available oxygen species on the SiO2 surface. This is consistent with the larger relative oxygen signal observed on SiO2 within the first 2-10 ALD cycles in Figure 12b as compared to that on HfO2 in Figure 12c. Another important consideration regarding the nucleation on silica versus transition metal oxides is the potential catalytic effect of the transition metal in oxygen decomposition. Distinctly different temperature dependence has been observed for Ti isopropoxide interaction with alumina as compared to silica, suggesting some affect of the metal center in the reaction21 and such effects may also extend to other transition metals. The energy associated with oxygen vacancy generation may also play a role. For example, if oxygen vacancy generation in HfO2 is energetically more favored over that in SiO2, this could promote the rate metal precursor reaction with the HfO2 versus SiO2. However, oxygen vacancy affects are likely to be quantitatively much smaller than other mechanisms, such as those associated with differences in hydroxyl group concentrations. Therefore, it is likely that the differences observed in the first few ALD cycles for ruthenium deposition on SiO2 and HfO2 result from differences in the density of oxygen, including hydroxyl groups, available on the starting growth surface. The rate and detailed mechanisms associated with metal growth initiation and nucleation during ALD is therefore expected to be strongly dependent on the reactivity and oxygen affinity of the metal center in the metal oxide growth surface.
IV. Summary and Conclusions Nucleation and growth initiation during Ru ALD are affected by the energy and composition of the growth surface. A surface
Park et al.
with relatively large surface energy can promote Ru film nucleation during ALD, whereas a smaller surface energy makes nucleation proceed more slowly. Moreover, the substrate surface composition and bonding can effect the initial growth reactions during film nucleation. For example, Auger electron spectroscopy analysis of the initial 1-20 cycles of ALD of ruthenium on SiO2 and HfO2 indicate qualitatively different mechanisms and rates for metal nucleation and growth on the different oxide surfaces. The thermodynamics of Hf-O-Ru and Si-O-Ru systems both suggest that the HfO2 and SiO2 will be stable oxide growth surfaces during Ru deposition. However, observed differences in the change in substrate signal intensity upon metal growth are ascribed to differences in the density of oxygen, likely in the form of hydroxyl groups, available on the different starting growth surfaces. Also, transition metal oxides may enable a higher rate of catalytic oxygen dissociation than silica, or a higher rate of oxygen vacancy generation which could also lead to differences in the rate of metal nucleation in oxygen-based noble metal atomic layer deposition processes including Ru deposition from ruthenocene. Differences in surface oxidation and reduction mechanisms on various oxide surfaces will also likely promote different ligand oxidation pathways. These different reactions can then lead to different interface composition which may influence electrical behavior or other performance criteria at the metal/oxide interface. Acknowledgment. Support is acknowledged from Sematech and the Semiconductor Research Corporation under the Center for Front End Processing. Support is also acknowledged from NSF CTS Grants #0072784 and #0626256 LA061898U
