The authors demonstrate that the addition of an ammonia coflow during the chemical vapor deposition of MoCxNy, Fe, or Ru thin films at ≤200 °C from the metal carbonyl precursors Mo(CO)6, Fe(CO)5, or Ru3(CO)12 affords area-selective growth: film grows readily on titanium metal or vanadium nitride substrate surfaces, but no nucleation occurs on air-exposed SiO2, TiO2, Al2O3, or MgO within the investigated times of 1–2 h. By contrast, in the absence of ammonia, nucleation and deposition on these oxide surfaces can either be slow or rapid, depending strongly on the oxide surface preparation. NH3 is also the source of N in MoCxNy, which has a resistivity of 200 μΩ cm and becomes superconducting at a critical temperature of 4 K. The authors hypothesize that the passivating effect of NH3 on oxide surfaces involves site blocking to prevent precursor adsorption, or an acid–base interaction to stabilize surface-bound metal subcarbonyl intermediates, or a combination of these mechanisms. A key finding is that surface selective growth is often crucially dependent on the sample history of the substrate, which must be specified in detail if reproducible results are to be obtained.

Superconducting thin films of vanadium nitride have been grown by low temperature (250–300 °C) chemical vapor deposition from tetrakis(dimethylamido)vanadium (TDMAV) and ammonia. For example, films grown from TDMAV (1 sccm Ar as carrier gas) and 7 mTorr ammonia at 300 °C are nanocrystalline (cubic δ-phase) with an average crystal size of 20 nm, have relatively low room temperature resistivities of 250 μΩ cm, and are superconducting with critical temperatures as high as 7.6 K (versus a bulk value of 9 K). The films have a V:N ratio of 1:1, with a carbon content of <5 at. % and an oxygen content of <3 at. % (as determined by high resolution XPS). The V 2p3/2 and N 1 s XPS binding energies of 513.5 and 397.3 eV, respectively, are consistent with the presence of a nitride phase. In contrast, films grown at lower temperatures <200 °C show carbon incorporation, have a much higher resistivity of ∼3000 μΩ cm, and are not superconducting. The results suggest that, at low temperatures, the thermally activated transamination reaction with ammonia becomes too slow to remove dimethylamido groups from the surface, resulting in carbon-rich films (10–15 at. % carbon). The conformal step coverage of the VN films depends on the growth conditions. For thermal growth of nonsuperconducting films at 150 °C, the step coverage is >95% in trenches of an aspect ratio of 4:1; for superconducting films grown at 250 °C, the step coverage is 65% for an aspect ratio of 3:1. At 150 °C, near-stoichiometric films with <2 at. % carbon and <3 at. % oxygen can be deposited if the gaseous ammonia is precracked by a remote plasma source; the resulting films have low resistivities of 320 μΩ cm but are not superconducting down to 4 K.

Three-dimensional nanodevice architectures require the coating and filling of deep vias and trenches, leading to an ongoing demand for dry processes with step coverages equal to or greater than one. We describe a new superconformal chemical vapor deposition process based on the use of two precursors: The first precursor readily deposits to afford film growth, but it cannot fill trenches when used alone because the coating is subconformal. The second precursor inhibits the deposition rate of the first precursor, and it grows film relatively slowly so that the overall film growth rate decreases when both precursors are present. In a trench, the inhibitor significantly suppresses the growth rate at the trench opening, but its pressure declines with depth due to consumption (film growth on the sidewalls) and the suppression effect weakens. Near the opening of the trench, where the inhibitor pressure is high, the consumption rate of the first precursor is small; it, therefore, diffuses deep into the trench to afford a growth rate that increases toward the bottom. If the flux of the inhibitor is not too high and the uninhibited growth rate of the first precursor is larger than that of the inhibitor, then the resulting film will be superconformal. We demonstrate this superconformal process for the growth of a metallic ceramic alloy, Hf1−xVxBy, in which the vanadium-bearing precursor serves as the consumable inhibitor. A continuous, single-step process is used to fill trenches with aspect ratios up to 10 with no void or seam along the centerline. We develop a model that captures the trench filling kinetics using Langmuirian growth kinetics, in which the two precursors compete for available adsorption sites and have different reaction rates. Calculations using physically plausible model parameters agree well with measured results and can be used to predict filling as a function of the aspect ratio. The model also indicates why filling fails at very high aspect ratios. In principle, a superconformal film of constant composition could be obtained using two precursors that each afford the same material.

The authors describe the enhancement of the area-selective chemical vapor deposition of cobalt films on one oxide surface over another from the precursor Co2(CO)8 by addition of the nucleation inhibitor ammonia (NH3). In the absence of an NH3 coflow, the Co2(CO)8 precursor exhibits a weak intrinsic selectivity: at 70 °C, Co nucleates quickly on Al2O3 but more slowly on SiO2. The addition of an NH3 coflow, however, greatly amplifies the selectivity between different oxide surfaces. Thus, NH3 significantly inhibits nucleation on acidic oxides such as SiO2 and WO3 but has little effect on more basic oxides such as Al2O3, HfO2, and MgO. Comparison of growth on fully hydroxylated and dehydroxylated SiO2 suggests that hydroxyl groups are the nucleation sites that are affected by the addition of NH3. The mechanism of nucleation appears to be disproportionation of Co2(CO)8 to Co2+ (the intermediate that leads to nucleation) and Co(CO)4−: this disproportionation occurs readily on basic oxides but not on acidic oxides. The addition of NH3 has little effect on Co nucleation on basic oxides, probably because ammonia binds poorly to such surfaces, but NH3 greatly retards nucleation on acidic oxides such as SiO2; the authors propose that the latter result is either a site blocking effect or the result of conversion of Co2+ to inactive Co(NH3)x2+ species. Nucleation of cobalt is facile on gold (a very unreactive metal) even in the presence of NH3. The authors have found, however, that the deposition of Co on tungsten can be inhibited by exposing the surface briefly to ozone; no deposition occurs on the resulting thin tungsten oxide overlayer from Co2(CO)8 in the presence of NH3. In other words, this thin oxide overlayer affords the same selective inhibition behavior as seen on bulk WO3. In this way, both metal-on-metal and metal-on-oxide selectivity can be achieved. Cobalt films grown in the absence and presence of ammonia have resistivities of 11–20 and 15–25 μΩ cm, respectively.

In this paper, we develop multilayer optical theory to model the real-time (in operando) optical response of a growing dielectric film being deposited by chemical vapor deposition (CVD), with a particular emphasis on understanding the deposition mechanism through direct detection of the adsorbates responsible for film growth by infrared reflection-absorption spectroscopy (IRRAS). The model involves a four-layer stack consisting of a vacuum over a monolayer or submonolayer of molecules adsorbed on the surface of a dielectric thin film, which, in turn, is growing on a metal substrate. It is well known that, in IRRAS, the sensitivity of p-polarized light to absorption by the molecular adsorbates is a function of the incident angle of the IR beam: at high angles, the sensitivity is highest. We show that, for incident beam angles above 70° (which are typically used in IRRAS experiments), the sensitivity also depends on the thickness and refractive index of the insulating thin film; as a result, the sensitivity changes dynamically during the growth of the dielectric layer. Our analysis shows that, at incident beam angles of ∼60°–70°, the sensitivity to molecular adsorbates is somewhat lower, but is almost independent of the oxide thickness from 0 to 100 nm and also independent of the oxide refractive index from 1.0 to 2.5. Despite the loss of sensitivity relative to that achievable at higher incident beam angles, 1000 scans at an incident angle of 60° are sufficient to obtain IR spectra of the adsorbed molecules with reasonable signal-to-noise ratios even at submonolayer coverages. Because the sensitivity at this incident beam angle is not thickness dependent, it is not an issue (as it is at higher beam angles) that additional oxide grows during the time required to acquire 1000 scans. Experiments can be performed using a conventional vacuum deposition system, in which the internal beam path is tens of cm. We demonstrate the use of these smaller incident beam angles to study the mechanism of a CVD process in real time by polarization-modulation IRRAS, obtained by subtracting the s-polarized from the p-polarized infrared spectra in order to eliminate the unpolarized component due to molecules in the beam path and on windows. We explore the surface coverage of various adsorbed intermediates during CVD of HfO2 from tetrakis-(dimethylamido)-hafnium (TDMAH) and water in the presence of the consumable inhibitor magnesium N,N-dimethylamino-diboranate [Mg(DMADB)2]. We find that the addition of the Mg(DMADB)2 inhibitor causes a decrease in the IR absorption from the adsorbed TDMAH precursor that correlates with the observed decrease in the HfO2 growth rate; this result indicates that the mechanism of inhibition involves Mg(DMADB)2 acting as a dynamic site-blocker that lowers the surface coverage of TDMAH.