III. RESULTS AND DISCUSSION

For the preparation of air stable ALD precursors with suitable volatility and reactivity, a new ligand concept was developed based on tridentate enaminones: N,N-(4,4,4-trifluorobut-1-en-3-on)-dimethylethyldiamine (Htfb-dmeda, 3a) and N,N-(4,4,4-trifluorobut-1-en-3-on)-dimethylpropyldiamine (Htfb-dmpda, 3b), which differ in the length of the alkyl amino chain (Scheme S1).

These ligands feature a similar ligand backbone as our previously reported tetradentate enaminones, which were successfully used in the CVD of CeO₂.^{Reference Schlafer, Graf, Fornalczyk, Mettenborger and Mathur56} Replacement of the diaminoalkyl backbone by a single dimethylamino alkyl function (ethylene, 3a and propylene, 3b) was expected to enhance the reactivity of the corresponding cobalt and nickel precursors toward surface hydroxyl groups with the N-metal bond being the predetermined bond-breaking site. Based on our previous reports, the presence of fluorine-containing groups in the ligand periphery enhances the volatility of the precursor.

3a and 3b were synthesized in a two-step process. First (E)-4-ethoxy-1,1,1-trifluorobut-3-en-2-on (tfae, 1) was prepared according to the literature^{Reference Hojo, Masuda, Kokuryo, Shioda and Matsuo55} and subsequently reacted with either N,N-dimethylethyldiamine (2a) or N,N-dimethylpropyldiamine (2b). The resulting products, 3a and 3b, were purified via distillation (80 and 100 °C, 10⁻³ mbar) and obtained in yields of 87% and 74%, respectively. As shown by ¹H NMR spectroscopy, both ligands predominantly exist as Z-isomers in CDCl₃ stabilized by the formation of intramolecular hydrogen bonds between the acidic NH protons and the carbonyl oxygen atoms. As observed with structurally related enaminones, the E:Z ratio of these ligands increases in more polar solvents.^{Reference Schlafer, Graf, Fornalczyk, Mettenborger and Mathur56} The corresponding cobalt and nickel compounds [Co^II(tfb-dmeda)₂] (4a), [Co^II(tfb-dmpda)₂] (4b), [Ni^II(tfb-dmeda)₂] (5a) and [Ni^II(tfb-dmpda)₂] (5b) were synthesized in a straight-forward salt metathesis reaction using either anhydrous CoCl₂ or [Ni(NH₃)₆]Cl₂ and the corresponding in situ prepared potassium salts of the ligands. The products were purified via sublimation (140–180 °C, 10⁻³ mbar) and were obtained in yields ranging from 46 to 72%.

Single crystal diffraction analysis of all four compounds revealed both cobalt (4a and 4b) and nickel (5a and 5b) centers in octahedral configuration formed by tridentate chelation of substituted enaminone ligands around Co²⁺ and Ni²⁺ cations. The metal atoms (M = Co²⁺/Ni²⁺) exhibit a distorted octahedral coordination sphere, wherein the ligand (L = tfb-dmeda) coordinates in a meridional fashion (Fig. 1). The O1 and N1 atom of each ligand span a plane with the central metal atom and are tilted by 87.9° with respect to each other. In contrast to that, 4b and 5b (L = tfb-dmpda) reveal a facial coordination isomerism.

FIG. 1.

Molecular structures of 4a, 4b, 5a, and 5b. Thermal ellipsoids are shown at 50% probability level and hydrogen atoms are omitted for clarity. Selected bond lengths for: 4a [Å] Co–O1 2.069(3), Co–N1 2.044(4), Co–N2 2.456(5); 4b [Å] Co–O1 2.048(2), Co–N1 2.075(2), Co–N2 2.352(2); 5a [Å] Ni–O1 2.039(2), Ni–N1 2.002(3), Ni–N2 2.313(3), Ni–O2 2.049(2), Ni–N3 1.995(3), Ni–N4 2.305(3); 5b [Å] Ni–O1 2.038(1), Ni–N1 2.043(1), Ni–N2 2.298(2).

Comparison of metal–oxygen (M–O) and metal–nitrogen (M–N1) bond lengths (∼2.05 Å) of the chelating enaminone backbone in all cobalt and nickel complexes revealed no remarkable difference, whereas the metal–nitrogen bonds of the amine groups (M–N(Me)₂) were elongated by ∼0.3–0.4 Å (cobalt) and ∼0.25–0.3 Å (nickel). From the higher atomic distances, a weaker interaction between the metal center and the dialkyl amino nitrogen can be deduced, which accounts for a facile dissociation of the respective (N(Me)₂)-function.^{Reference Hausmann and Gordon57} This can be regarded as molecular predetermined breaking point and thus accounts for the reactivity of the complexes toward the substrate (Scheme S2).

Replacing the ethylene spacing with a propylene moiety provides more freedom for relaxational alignment. The impact on the crystal packing is examined by comparing the distances of the fluorine atoms from the CF₃-group of the nearest neighboring molecules within a 2 × 2 × 2 supercell. Due to the different coordination isomerism (meridional versus facial), the F–F distances increase with the elongation of the ligand chain resulting in an increased volatility (Table S1).

TG profiles of all compounds exhibit a single weight-loss step with residues of 2–7 wt% for 4a, 4b, and 5b, pointing toward the intact evaporation of the precursor molecules [Fig. 2(a)] at the measurement conditions. The tail end of the TG-curve of 5a hints at a partial decomposition along with an evaporation of the precursor resulting in a residue of 6 wt%, which is lower than the calculated wt% for the expected elemental nickel (∼12 wt%). The evaporation of the precursor molecules during TG measurements does not allow any conclusion on the decomposition behavior of the precursors other than that their thermal stability and volatility allow an intact transport through the gas phase even at elevated temperatures. For verifying the suitability of the new precursors for ALD processes, the evaporation temperatures for all four compounds were examined [Fig. 2(b)].

FIG. 2.

(a) TG measurements of precursors 4a, 4b, 5a, and 5b. (b) Volatility data of the precursors to elucidate the gas phase saturation behavior.

Intact vaporization of 4a and 5a occurred at approximately 180 °C, whereas 4b and 5b already evaporated at approximately 140 °C. As presented in Fig. 2(b), the relative mass losses of 4b and 5b (∼40%) after shorter periods (4 min) denote a faster vapor phase saturation, compared to 4a and 5a (∼20%). These experiments indicated that 4b and 5b are more suitable as potential ALD precursors.

To investigate the mechanism of precursor adsorption and decomposition during the ALD process, detailed XPS studies were undertaken. In a first step, the chemisorption of the first precursor monolayer on the hydroxyl groups of the silicon substrate were elucidated. After the first precursor pulse and purge cycle, signals of the corresponding precursor atoms (N, F, Co/Ni) were visible in the survey scan XPS spectrum [Fig. 3(a)]. The O 1s narrow scan spectra were recorded to investigate the binding mode of the precursors (exemplarily of 4b) [Fig. 3(b)]. Whereas the spectrum of the bare silicon substrate [the inset of Fig. 3(b)] showed a single signal at 532.78 eV, the spectrum of the precursor exposed silicon substrate additionally displayed a small shoulder at 531.10 eV indicating a chemical bonding between 4b and the surface hydroxyl groups of the substrate. The shoulder at lower binding energy is due to the negative charge distribution around the metal atom, which decreased the bond polarity of the metal–substrate oxygen bond.^{Reference Keranen, Guimon, Liskola, Auroux and Niinisto58}

FIG. 3.

(a) Survey scan XPS spectra of bare and precursor exposed silicon substrate. (b) High resolution O 1s XPS spectra of the bare (inset) and precursor 4b exposed silicon substrate.

In a follow-up experiment, precursors 4b and 5b were used to deposit cobalt and nickel oxide films using oxygen plasma as the reactant to analyze the chemisorption of the precursors onto the growth layer. The high-resolution O 1s XPS spectra of the grown oxide surfaces before and after one precursor pulse were recorded (Fig. 4). The O 1s spectrum of bare nickel oxide displayed a broad signal at 530.2 eV, resulting from an overlap of the Ni–O and surface –OH signals [Fig. 4(b)]. The spectrum of bare cobalt oxide showed two almost resolved signals at 529.9 and 531.9 eV, which could be attributed to the surface OH- and cobalt-oxygen signals [Fig. 4(a)].^{Reference Biesinger, Payne, Grosvenor, Lau, Gerson and Smart59} After one precursor pulse, both spectra showed the disappearance of the surface OH-signals and a small shoulder at lower binding energies. This confirmed the chemisorption of 4b and 5b to the surface hydroxyl-groups of the corresponding metal oxides. To further analyze the binding of the precursor to the growth layer, high-resolution spectra of the N 1s and O 1s binding energy region (see Supplementary Material, Fig. S1) of the chemisorbed precursor were compared to spectra of the free ligand. This revealed no observable change in the binding energy or peak shape suggesting that both, the cobalt and the nickel precursor, exclusively form a chemical bond between the metal and the surface hydroxyl group after cleavage of the amine-moiety without any interaction of the ligands to the substrate surface.

FIG. 4.

High-resolution O 1s XPS spectra of (a) cobalt oxide and (b) nickel oxide surfaces before (black line) and after (red line) one precursor pulse using 4b and 5b, respectively.

Even when using hydrogen plasma instead of oxygen plasma as the activant, the initial formation of metal hydroxides was observed (Scheme S3). The high oxophilic nature of nickel and cobalt facilitates preferred scission of the C–O bond of the enol functionality in the ligand instead of the M–O bond in the precursor molecules resulting in the formation of homogenous and pinhole-free hydroxide films at the growth stage. Thus, the role of hydrogen plasma as etchant in a subsequent step was essential to induce a recrystallization process, transforming as-deposited hydroxide films into pure metallic coatings.^{Reference Sasinska, Ritschel, Czympiel and Mathur60,Reference Choi, Yun and Rhee61}

SEM (Fig. S2) and AFM analysis (Fig. S3) of the hydrogen plasma treated films after the etchant step showed a smooth surface with a RMS roughness value of 0.2 ± 0.06 nm for the cobalt film. Surface analysis of the metallic nickel film displayed a similar surface decorated with larger crystallites in the micron range (Fig. S3). The cross-sectional SEM analysis (Fig. S2) showed that growth rates for nickel films are lower than those observed for cobalt, which can be explained by a faster chemisorption of the cobalt precursor 4b. This observation is in line with previous investigations reporting lower growth rates for nickel.^{Reference Lim, Rahtu and Gordon32} ALD processes onto trenched silicon wafers with ranging aspect ratios from 4:1 up to 30:1 revealed that homogeneous deposition can be achieved even on structured substrates (Fig. 5); however, due to the high recombination rate of hydrogen atoms at the surface, the conformity is limited.^{Reference Kariniemi, Niinisto, Vehkamaki, Kemell, Ritala, Leskela and Putkonen62–Reference Schindler, Logar, Provine and Prinz64}

FIG. 5.

SEM analysis: (a) Metallic Ni film deposited onto a trenched silicon wafer. (b) Magnified view of a conformal coating across the whole trench width.

X-ray diffraction (XRD) analysis (Fig. 6) of hydrogen plasma treated cobalt films after the etchant step showed a small peak at 2θ = 44.4° corresponding to the (111) plane of the cubic crystal structure of the pure metal (JCPDS-No. 15-0806). The nickel films exhibited a small peak at 2θ = 44.3° and 51.8°, which can also be assigned to the (111) and (200) plane of the cubic crystal structure (JCPDS-No. 04-0850) of the pure metal. As-deposited cobalt and nickel films were found to be amorphous (XRD) and post-growth hydrogen plasma treatment was found to be essential for inducing crystallization through the reduction of metal oxide and hydroxide domains present in the ALD films.^{Reference Miikkulainen, Leskela, Ritala and Puurunen28,Reference Choi, Yun and Rhee61}

FIG. 6.

XRD patterns of (a) cobalt and (b) nickel thin films after hydrogen plasma post-treatment.

The XPS measurements performed on the post-treated films (Fig. 7) showed metallic cobalt and nickel deposits with remaining impurities from ligand atoms. The relatively high oxygen and carbon contents are due to atmospheric oxidation under ambient conditions that was inevitable when transferring the samples from the ALD reactor to the XPS chamber. Besides the characteristic 2p transitions, both samples exhibited intensive metal 2s-, 3s-, 3p-, and Auger-transitions still underlining a good sample quality (Fig. 7).

FIG. 7.

Survey scan XPS spectra of (a) cobalt and (b) nickel thin films before and after surface etching with argon plasma.

High-resolution scans of the Co 2p- and Ni 2p-signals revealed the presence of metal oxide passivation layers, which were removed via argon plasma etching for 120 s (see Supplementary Material, Fig. S4) in the XPS chamber.^{Reference Biesinger, Payne, Grosvenor, Lau, Gerson and Smart59,Reference Mcintyre and Cook65} To exclude argon plasma induced reduction,^{Reference Choudhury, Saied, Sullivan and Abbot66} control experiments were performed subjecting as-deposited metal hydroxide films (prepared without subsequent hydrogen plasma treatment) to argon sputtering. Narrow scan spectra of these films in the Co 2p- and Ni 2p-binding energy region only showed the presence of cobalt(II) and nickel(II) oxide (see Supplementary Material, Fig. S5). In general, metallic nickel films have a lower tendency to undergo oxidation reactions due to their passivation behavior as confirmed by additional annealing experiments under argon atmosphere (see Supplementary Material, Fig. S6).

Resistance measurements revealed an increase of metallic conductivity up to 500 cycles of post-deposition hydrogen plasma treatment. Raising up to 1000 cycles of post-deposition, hydrogen plasma treatment resulted in a diminution of the metallic conductivity (Fig. S7). For the determination of the hydrogen content within the films, nuclear reaction analysis has been performed. An excessive hydrogen plasma treatment was found to be detrimental as it reacted with the metallic films, probably transforming them into metal hydrides (Fig. S8).

This competitive interplay of deposition and in situ etching enables tailoring the chemical composition of the deposited thin films and also provides fresh insight into the reactivity of etchants that can alter the composition of the film material (Fig. 8).

FIG. 8.

Schematic illustration of the role of hydrogen plasma as the activant and etchant during the ALD process.