Mechanisms of Atomic Layer Deposition (ALD)

The main objective of the work proposed here is to develop a molecular-level knowledge of the reactions associated with atomic layer deposition (ALD).  ALD relies on the alternate use of two or more self-limiting and complementary reactions to deposit materials on solid surfaces one monolayer at a time.  The figure below depicts schematically how this works for the case of the deposition of TiN films using TiCl4 and ammonia.  ALD techniques have been recently advanced as a powerful way to deposit thin films in a controlled manner for the microelectronic industry.  There are, however, a number of still unresolved problems involving the stoichiometry, structure, and chemical quality of the deposited films.  We expect to shed some light on those issues by advancing the understanding of the mechanism of the underlying surface reactions.

Schematic representation of the reaction mechanism for the atomic layer deposition (ALD) of TiN films from TiCl4 and ammonia.  This process relies on two self-limiting steps, the deposition of Ti from its chloride, and the reduction of that Ti and the deposition of N with ammonia.  The diagram points to the complexity of the overall process, and also to some poorly understood aspects of the mechanism such as the stoichiometry of the TiClx species formed on the surface, the extent of replacement of the chlorine by NHx ligands (which relates to the level of impurities left behind in the film), and the coverage and morphology of the TiN layer after each cycle.

Some years ago we carried out extensive studies on the thermal chemistry of metal carbonyls on nickel and platinum substrates in connection with metal chemical vapor deposition (CVD) processes [F. Zaera, J. Vac. Sci. Technol. A7 (1989) 640; F. Zaera, Langmuir 7 (1991) 1188; F. Zaera, Surf. Sci. 255 (1991) 280; F. Zaera, J. Phys. Chem. 96 (1992) 4609; M. Xu and F. Zaera, Surf. Sci. 315 (1994) 40; M. Xu and F. Zaera, J. Vac. Sci Technol. A14 (1996) 415].  In those studies it was found by using TPD and XPS that all the metal carbonyl precursors investigated (Fe(CO)5, Cr(CO)6, Mo(CO)6, and Co(CO)6) adsorb molecularly at low temperatures, but undergo complete decarbonylation upon heating of the substrate [M. Xu and F. Zaera, J. Vac. Sci Technol. A14 (1996) 415].  It was also observed that for all metal carbonyls except iron the removal of all the CO ligands occurs irreversibly in a narrow range of temperatures.  In the case of iron pentacarbonyl, on the other hand, this chemistry was found to take place in a stepwise and reversible manner.  Infrared spectroscopy was used to isolate and identify Fe(CO)4 and Fe(CO)3 adsorbed intermediates (see figure below) [F. Zaera, Surf. Sci. 255 (1991) 280], and coadsorption experiments with isotopically labelled 13CO were employed to prove that the decarbonylation could be easily reversed [M. Xu and F. Zaera, Surf. Sci. 315 (1994) 40].  Additional undesirable side reactions were also detected between the adsorbed carbon monoxide and the metal atoms left on the surface which lead to the incorporation of carbon into the growing films as well as to the oxidation of the deposited metal [F. Zaera, J. Phys. Chem. 96 (1992) 4609; M. Xu and F. Zaera, J. Vac. Sci Technol. A14 (1996) 415].

RAIRS data used to identify the intermediates that form during the thermal chemistry of iron pentacarbonyl on metal surfaces.  The traces shown here, which correspond to the C–O stretching region of the spectra, provide evidence for an initial isomerization of the Fe(CO)5 precursor upon adsorption into a more reactive C4v geometry, and also for subsequent stepwise decarbonylation steps to produce Fe(CO)4 and Fe(CO)3 surface intermediates at around 185 and 240 K, respectively.  Additional TPD experiments using isotopically labeled carbon monoxide were carried out to show that these steps are in fact reversible.

XPS was also employed to follow the kinetics of the film growth in situ during the CVD processes [F. Zaera, Langmuir 7 (1991) 1188; M. Xu and F. Zaera, J. Vac. Sci Technol. A14 (1996) 415].  It was found that the process is self-limiting at low temperatures, and that the sample needs to be kept at temperatures higher than those required for CO desorption to attain steady-state deposition.  Representative kinetic data is provided below for the case of W(CO)6.  Interestingly, the poisoning of the metal carbonyl surface activation by the CO byproduct suggests a procedure for growing metal films with these precursor in a ALD-type manner based on cycles of exposure of the substrate to the M(CO)x gas at low temperatures followed by pumping and annealing to higher temperatures to desorb the adsorbed CO.  The initial stages of this procedure were shown to work successfully for the case of W(CO)6 [M. Xu and F. Zaera, J. Vac. Sci Technol. A14 (1996) 415].

Growth kinetics for the deposition of tungsten films on a nickel surface using W(CO)6.  The main frame displays the W coverage uptake as a function of exposure at three different deposition temperatures, as measured by following the W 4f7/2 signal at 31.4 eV after appropriate calibration, while the inset shows the W 4f XPS signature for the resulting tungsten metallic films.  These data indicate that the film growth is self-limited at low temperatures, and can be carried out in a linear steady-state fashion only above 300 K.  Subsequent XPS, RAIRS and TPD experiments have indicated that the rate-limiting step in this case is the desorption of the carbon monoxide byproduct from the surface.

The composition and morphology of the deposited films were characterized as well.  The figure below reports an example of this work for the case of iron films grown using Fe(CO)5.  In that case, a combination of ISS and AES experiments were used to establish that the metal deposition occurs via the growth of three-dimensional islands, not in a more desirable layer-by-layer mode, and that it is followed by significant carbon incorporation.

ISS (LEIS) and AES spectra before (left) and after (right) the deposition of an iron film using Fe(CO)5.  Because of the different sampling depths of the ISS and AES techniques, the combined results reported here provide a fairly good indication of the morphology of the resulting films.  Specifically, it is seen that although the 100 L dose at 500 K used in this example leads to the deposition of an equivalent to seven Fe monolayers (according to the AES data), approximately 30% of the original platinum substrate remains uncovered still (based on the ISS information).  This means that the film has grown in a three-dimensional fashion.  Moreover, the AES spectrum shows the incorporation of significant amounts of carbon, whereas the ISS trace only identifies the terminal oxygen atom in the carbon monoxide chemisorbed on the topmost layer.

More recently we have carried out some preliminary experiments on the ALD of TiN films from TiCl4 and ammonia.  The figure below shows XPS data for each individual step involved in the deposition.  A number of interesting observations are readily derived from those data.  For one, it is seen there that the initial exposure of the surface to TiCl4 leads to the expected deposition of both titanium and chlorine atoms, only at a much richer Ti stoichiometry (TiCl) than that expected according to the surface chemistry previously proposed (TiCl2 or TiCl3).  Moreover, although it is clear that the subsequent treatment with ammonia allows for both the removal of the remaining surface chlorine and the deposition of nitrogen, the final film only displays a stoichiometry of approximately Ti1.8N.  In fact, the Ti 2p XPS signal recorded after many cycles shows a combination of oxidation states indicative of an incomplete reduction of the Ti4+ species, and an excess of nitrogen atoms.  It was determined that at least some of these observations are associated with the incorporation of oxygen impurities in the growing film (perhaps from water impurities in the gas supply), but some Ti3N4 seems to nevertheless be codeposited under the conditions of the experiment.  Finally, it was also seen that some superficial chlorine does remain in the film after repeated ALD cycles, most likely due to readsorption of the HCl byproduct.

XPS data after each step involved in the ALD of TiN.  Some of the expected behavior is corroborated by these data, namely: (1) the incorporation of Ti and N on the surface after TiCl4 and NH3 treatments, respectively; (2) the removal of most of the chlorine upon ammonia treatment; and (3) the attenuation of the XPS signals from the substrate by the TiN deposited film.  On the other hand, some results were unexpected: (1) the final film from this cycle is not stoichiometric but rich in titanium (Ti/N~1.8), perhaps because of the incorporation of some oxygen impurities (thicker films appear to be nitrogen rich instead); (2) the average stoichiometry of the surface species after the first half of the cycle amounts to only about TiCl, indicating a significant elimination of chlorine from the surface before the reducing step; (3) the titanium XPS peaks display a complex shape, especially after the ammonia treatment, suggesting the presence of several Ti oxidation states on the surface; and (4) the XPS peaks from the substrate (Ni) grow back in intensity upon annealing in vacuum in between both half-cycles, indicating an initial deposition of three-dimensional films which spread on the surface at higher temperatures.

The buildup of thicker films was tested by performing multiple cycles with alternating exposures to TiCl4 and NH3.  An example of the XPS obtained after one of such studies is shown in the figure below, for a case where the film reached a close-to-stoichiometric composition and displayed reasonably low (but non-negligible) levels of O and Cl impurities.  A total deposition of about three TiN monolayers was reached after 27 cycles.  Interestingly, though, both the extent of the deposition and the stoichiometry of the film were found to depend in a complex non-linear fashion on the pressures of the precursors used in each half-cycle (Figure inset).

Main Frame: Survey XPS trace from a TiN film deposited by 27 ALD cycles using TiCl4 and NH3 as precursors.  Clear signals are seen here for both Ti and N (as well as for the nickel substrate), and also for small amounts of chlorine and oxygen impurities.  The final film thickness at the end of the 27 cycles was estimated from the attenuation of the nickel signals at approximately three monolayers.  Inset:  Film uptake and Ti/N stoichiometry as a function of the number of ALD cycles used for the deposition showing that both the rate of deposition and the nature of the resulting film may depend not only on the numbers of cycles but also on the pressure of the precursors used in each cycle.  This dependence is complex, and does not simply scale with exposure: higher pressures appear to increase the rate of deposition in a non-linear fashion, and also lead to films richer in titanium.

Finally, a few probing studies have been carried out to test the surface chemistry of the bis[(N,N'-di-sec-butyl acetamidinate)Cu] precursor provided to us by Prof. Roy Gordon.  TPD and XPS experiments were performed on both a Ni(110) single crystal and a cobalt polished polycrystalline foil.  No significant differences were observed between the two surfaces, and a small deposition of copper could be detected by XPS on both.  Also, clear shifts seen in the C 1s and N 1s XPS peaks as a function of temperature, indicative of a significant surface conversion (figure below).  Interestingly, at least two steps could be identified by the different onsets in the chemistry followed by the carbon and nitrogen signals in the XPS data.  It appears that the amidine ligands may detach from the copper center at temperatures as low as 250 K, and that the resulting surface species may then decompose starting about 300 K, to produce butene among other products.  Predosing of the surface with hydrogen, as required by this ALD processes, seems to inhibit the decomposition of both the copper complex and the adsorbed ligands.

XPS data for 10 L of bis[(N,N'-di-sec-butyl acetamidinate)Cu] adsorbed on nickel at 90 K..  The C 1s XPS data displayed in the main frame, obtained as a function of annealing temperature after a 15 L dose, show both the loss of signal intensity slightly above 200 K due to molecular desorption and the shift of the remaining signal from 285.8 to 284.2 eV indicative of further surface conversion.  The inset reports the estimated coverages of the new surface species identified by the XPS.  Three things become apparent from those data (1) the amidine ligands react below 250 K to yield a new surface species characterized by a N 1s XPS peak at 397.6 eV; (2) a second reaction starts around 300 K leading to a surface moiety characterized by the C 1s XPS peak at 284.2 eV; and (3) predosing of hydrogen on the surface leads to the inhibition of the decomposition of the surface species, which in that case starts at higher temperatures and occurs to a lesser extent.  A small amount of copper deposition is also detectable by XPS at the end of this thermal cycle.