Atomic Layer Deposition of Cobalt Using H2-, N2-, and NH3-Based Plasmas: On the Role of the Co-reactant.

This work investigates the role of the co-reactant for the atomic layer deposition of cobalt (Co) films using cobaltocene (CoCp2) as the precursor. Three different processes were compared: an AB process using NH3 plasma, an AB process using H2/N2 plasma, and an ABC process using subsequent N2 and H2 plasmas. A connection was made between the plasma composition and film properties, thereby gaining an understanding of the role of the various plasma species. For NH3 plasma, H2 and N2 were identified as the main species apart from the expected NH3, whereas for the H2/N2 plasma, NH3 was detected. Moreover, HCp was observed as a reaction product in the precursor and co-reactant subcycles. Both AB processes showed self-limiting half-reactions and yielded similar material properties, that is, high purity and low resistivity. For the AB process with H2/N2, the resistivity and impurity content depended on the H2/N2 mixing ratio, which was linked to the production of NH3 molecules and related radicals. The ABC process resulted in high-resistivity and low-purity films, attributed to the lack of NH x,x≤3 species during the co-reactant exposures. The obtained insights are summarized in a reaction scheme where CoCp2 chemisorbs in the precursor subcycle and NH x species eliminate the remaining Cp in the consecutive subcycle.

Introduction

Atomic layer deposin class="Chemical">tion (ALD) is a thin-film deposition technique, which relies on the cyclewise alternation of precursor and co-reactant doses. The self-limiting nature of the surface reactions during ALD generally allows for good uniformity on large-area substrates and excellent conformality on three-dimensional structures.[1] Although the precursor that is used for an ALD process generally receives considerable attention, the choice of the co-reactant is equally important because it can greatly affect the properties of the deposited material as well as the technological and industrial feasibility of the process. For the ALD of metals, a wide range of co-reactants have been explored, with gases or plasmas of O2, H2, and NH3 being the most common choices.[1−4] In addition, less common chemicals such as hydrazine (N2H4), silane (SiH4), disilane (Si2H6), formic acid (CH2O2), and tertiary butyl hydrazine (C4H12N2) have been used.[2,5−8] Moreover, certain ALD processes make use of what can be referred to as an advanced ALD cycle, in which either two or more co-reactants are dosed simultaneously or after one another in an ABC-type manner. For instance, mixed H2/N2 plasmas have been used for the ALD of a variety of materials.[9−12] Furthermore, Hämäläinen et al. deposited Ir, Pd, Rh, and Pt at low temperatures (120–200 °C) using consecutive O3 and H2 exposures, and similar ABC-type cycles were later reported for the ALD of Ru (at 150 °C) using subsequent O2 and H2 doses and for the ALD of Pt (at room temperature) using subsequent O2 and H2 plasmas.[13−16]

H2-, N2-, and n class="Chemical">NH3-based plasmas (e.g., plasmas using NH3, H2, N2, or H2/N2 mixtures as source gases) have previously been used as co-reactants for the ALD of a wide range of metals and metal nitrides. See Table S1 in the Supporting Information for an overview of selected metals and metal nitrides, which have been deposited using a NH3 plasma or a mixed H2/N2 plasma as the co-reactant. For instance, Kim et al. found that for the ALD of Ir using ((ethylcyclopentadienyl)(1,5-cyclooctadiene)iridium), NH3 plasma yielded a lower surface roughness in comparison to when using O2 gas as the co-reactant.[17] Furthermore, Ten Eyck et al. employed a H2/N2 plasma for the ALD of Pd on a polymer substrate and claimed that a H2/N2 plasma leads to the formation of reactive NH2 groups on the polymer, needed for chemisorption of the palladium(II) hexafluoroacetylacetonate precursor.[10] Moreover, the use of NH3 plasmas instead of H2 plasmas for the ALD of Ru, Ag, and Ni resulted in higher growth per cycle (GPC) and lower resistivity values.[18−20] The choice for the co-reactant is generally less straightforward for the ALD of Co, Ni, and Cu, as compared to noble metals, because their reduction potential is lower, which makes impurity incorporation more probable.[4] For this reason, H2-, N2-, or NH3-based plasmas can be preferred over the O-containing co-reactants commonly used for noble metal ALD. However, for certain elements, the use of N2, NH3, or H2/N2 plasmas and mixtures thereof can also result in metal nitride films (e.g., AlN, TiN, and TaN).[9,21−24] In general, deposition of a metal nitride becomes more likely for metals with a low reduction potential, as illustrated in Table S1.[4] Al, Ta, and Ti have reduction potentials between −0.6 and −1.7 V, whereas for Co, Ni, and noble metals, it is −0.26 V or higher.[25] Consequently, the use of, for instance, a mixed H2/N2 plasma leads to the deposition of AlN on one hand and metallic Co on the other hand.[26]

Co is a ferromagnen class="Chemical">tic transition metal used in, for instance, magnetoresistive random-access memory and CoSi2 contacts.[27−31] Currently, Co mostly receives much attention for applications in interconnect technology, in order to reduce the resistance–capacitance delay in state-of-the-art devices.[32] First, Co has been suggested as a viable candidate as the liner for Cu interconnects because Co can be thinner than the conventional Ta liner, which leaves more space for Cu.[33,34] Moreover, Co is also being investigated for the replacement of Cu or W in small-dimension interconnects in the front-end of line.[35,36] It is therefore valuable to select Co ALD as a model system for studying the influence of the co-reactant. A wide range of precursors and co-reactants have been investigated for the ALD of Co, as shown in Table . As compared to other precursors, the bis(cyclopentadienyl)cobalt(II) (cobaltocene, CoCp2) precursor has previously given good results, that is, a low resistivity and high purity, while also being a readily available and low-cost precursor. Interestingly, the ALD of Co using CoCp2 as the precursor can be achieved using different co-reactants. Specifically, the studies of Lee et al. and Yoon et al. reported the growth of high-quality Co films using NH3 plasma and H2/N2 plasma, respectively.[26,37] A direct comparison between the two different processes as well as a connection between the plasma composition and the obtained material properties has not been made so far.

Table 1

ALD Processes Reported in the Literature for the Deposition of Co, Listing Deposition Temperature T, GPC, and Resistivity ρ

precursor	co-reactant	T (°C)	GPC (Å)	ρ (μΩ cm)	refs
CoCp2	NH3 plasma	300	0.48	10	(37)
CoCp2	H2/N2 plasma	150–450	0.26–0.65	18	(26)
CoCp2	NH3a	100–300	0.37–0.97		(38)
Co(MeCp)2	NH3 plasma	100–350	0.4–1.9	31	(39)
Co(CpAMD)b	NH3 plasma	200–250	0.5	140	(40)
Co2(CO)8	H2 plasma	75–110	1.2		(41)
CpCo(CO)2	H2 plasma	125–175	1.1		(42)
Co(AMD)2c	H2	340	0.50	285	(43)
Co(AMD)2	NH3	350	0.26	50	(44)
tBu-allylCo(CO)3	dimethylhydrazine	140	0.5		(45)
CCTBAd	H2	125–200	0.8	90	(46)
Co(DBDB)e	formic acid	170–180	0.95	13f	(7), (47)
Co(DBDB)e	tert-butylamine	170–200	0.98	15f	(48)

Hot-wire ALD.

Cyclopentadienyl isopropyl acetamidinato-cobalt.

Bis(N,N′-diisopropylacetamidinato)cobalt(II).

Dicobalt hexacarbonyl tert-butylacetylene.

Bis(1,4-ditert-butyl-1,3-diazabutadienyl)cobalt(II).

Measured on the Ru substrate.

Cyclopentadienyl isopropyl acen class="Chemical">tamidinato-cobalt.

Bis(N,N′-diisopropylacetamidinato)n class="Chemical">cobalt(II).

In this work, a detailed study of the use of n class="Species">H2-, N2-, and NH3-based plasmas as co-reactants for the ALD of Co using CoCp2 as the precursor is presented. As illustrated in Figure , three ALD processes with different co-reactants were investigated: an AB-type process with NH3 plasma (referred to as “AB-NH3 process”, Figure a), an AB process with a mixed H2/N2 plasma (“AB-H2/N2 process”, Figure b), and an ABC process with subsequent N2 and H2 plasmas (“ABC-N2-H2 process”, Figure c). As will be shown, the separation of the H2 and N2 plasmas in an ABC-type cycle provides an insight into the role of NH species that are present in both the NH3 plasma and the H2/N2 plasma (but not in the N2 or H2 plasmas).

Figure 1

Schematic overview of the three Co ALD processes investigated in this study: (a) AB-NH3, (b) AB-H2/N2, and (c) ABC-N2-H2 process. The ABC process uses separate N2 and H2 plasmas exposures. Note that each purge step is followed by a pump step (see Section under Experimental Section), which is not shown in the figure for simplicity.

Schematic overview of the three n class="Chemical">Co ALD processes investigated in this study: (a) AB-NH3, (b) AB-H2/N2, and (c) ABC-N2-H2 process. The ABC process uses separate N2 and H2 plasmas exposures. Note that each purge step is followed by a pump step (see Section under Experimental Section), which is not shown in the figure for simplicity.

This work is structured as follows. First, the experimentn class="Chemical">al conditions related to the film deposition, the plasma studies, and the film analysis are discussed. In Section , the species present in the NH3 and H2/N2 plasmas are identified and the role of the H2/N2 ratio on the plasma composition is investigated. This is followed in Section by a study on the reaction products released during the plasma exposures of the three different ALD processes. In Section , the obtained material properties are compared. In addition, the effect of the H2/N2 ratio and NH concentration in the plasma on the material properties are addressed. Next, the role of NH species and a possible reaction mechanism are discussed in Sections and 4.2, respectively. Finally, the main conclusions of this work are summarized.

Experimental Section

ALD Reactor and Conditions

Co films were depon class="Chemical">sited in a home-built ALD reactor, as described in the previous work.[49] In short, the reactor is equipped with a remote inductively coupled plasma source and a turbo pump reaching a base pressure of ∼10–6 Torr. During all experiments, the temperature of the substrate table was set to 300 °C, whereas the walls were heated to 100 °C. Prior to all experiments, the reactor wall was covered by Co by running at least 200 cycles of the AB-NH3 process. The CoCp2 precursor (98%, Sigma-Aldrich) was contained in a stainless steel bubbler. The bubbler and the dosing line were heated to 80 and 120 °C, respectively, as was previously found to be appropriate for the deposition of CoO in the same reactor.[50] The ALD recipe consisted of precursor dosing for 6 s in the first subcycle, using Ar as a carrier gas, which resulted in a chamber pressure of approximately 15 mTorr. Subsequently, the reactor was purged with Ar for 3 s and pumped down for 6 s. All plasma exposures were performed at a power of 100 W for 11 s and were followed by a purge and a pump step of 1 and 11 s, respectively. The NH3, N2, and H2 plasmas were started after flowing the source gas into the reactor for 3 s. For the AB-H2/N2 process, the N2 flow was started 2 s before the H2 flow, and subsequently after 5 s, the H2/N2 plasma was ignited. This was done to stabilize the gas flows and minimize overpressures. The precursor dosing and plasma exposure times were based on saturation studies as shown in the Supporting Information (Figure S1).

The pressure used for the NH3 plasma was 1.5 mTorr. For the sn class="Chemical">tandard H2/N2 plasma, the N2 and H2 pressures were separately set to 1.5 and 15 mTorr, respectively. Because of the addition of N2 to the H2 gas, the pumping speed increases (as compared to only H2), leading to a lower total pressure of approximately 13 mTorr for the H2/N2 mixture. Moreover, the actual H2/(H2 + N2) mixing ratio is approximately ∼0.77 (for the H2 and N2 pressures of 1.5 and 15 mTorr, respectively) because of the shorter residence time of H2 as compared to that of N2 (see also Section ). The results for different H2/(H2 + N2) ratios in Sections and 4.1 were obtained by varying the partial pressures of H2 and N2, while keeping the total pressure of the mixture constant at 13 mTorr. For the ABC-N2-H2 process, a pressure of 7.5 mTorr was used for both the N2 and H2 plasmas, and both plasma exposures were 11 s long.

To determine the effect of the H2/(n class="Species">H2 + N2) ratio on the NH3 production in Section , a constant pressure of 75 mTorr was used for the gas mixture. This pressure was higher than the “standard” 13 mTorr to allow for more accurate variation of the gas flows and to enable mixing ratios higher than 80 vol %. For a pressure of 13 mTorr, it is not possible to keep the pressure constant for mixing ratios higher than ∼80 vol % because of the low gas flows used and because of changes in pumping speed upon mixing gas flows.

Plasma Studies

Quadrupole mass spectrometry (QMS) measurements were performed un class="Chemical">sing Pfeiffer Vacuum Prisma QME-200 (mass-to-charge ratio m/z = 1–200), attached to the side of the ALD chamber. Measurements were done with the substrate table, and reactor walls were kept at the standard temperatures of 300 and 100 °C, respectively. Note that a considerable part of the QMS signal can originate from the reactions at the reactor walls because the surface area of the wall is significantly larger than the surface area of the substrate table. It was confirmed that growth also occurs at a deposition temperature of 100 °C, albeit at a lower GPC (∼0.13 Å as compared to ∼0.29 Å) and with a higher impurity content. Because the aim is to compare the three ALD processes with one another, the temperature difference between the wall and the table is considered to be of minor influence.

For determination of the main species in the n class="Chemical">NH3 and H2/N2 plasmas, mass scans (i.e., ion current as a function of m/z) for masses 1–30 were used. These mass scans were collected after stabilization of the gas flows and plasma. The H2/(H2 + N2) mixing ratios were determined using the QMS ion currents at m/z ratios 2 and 14 (corresponding to H2+ and N+, respectively) from such mass scans. For a complete description of this method, see the Supporting Information.

The procedure for time-resolved QMS measurements was n class="Chemical">similar to the method as previously described by Knoops et al.[51] In short, for m/z ≤ 40, four m/z ratios were measured simultaneously, of which one was always m/z = 40. This value corresponds to Ar+ and is used as reference. For m/z > 40, besides m/z = 40, only one other m/z ratio was followed per measurement, in order to keep the signal-to-noise ratio optimal while maintaining a reasonable time resolution. Three different cycles were studied using the QMS measurements: a “normal” (AB- or ABC-type) ALD cycle, a cycle without CoCp2 dosing (but with Ar carrier gas dosing), and a cycle without igniting the plasma(s) (see Figure S2). This was done to discern reaction products from the species present because of the precursor dosing, source gas exposure, or plasma ignition. For each type of recipe, 10 cycles were performed, and only the signals over the last nine were averaged, assuming the first cycle can deviate because of the recipes performed previously. To further minimize the influence of previous cycles, every set of cycles was preceded by a cleaning step consisting of an O2 plasma for 90 s, followed by a NH3 plasma for 120 s. Moreover, the purging and gas stabilization times were extended as compared to the standard ALD cycle, in order to separate the effects of pressure overshoots from the reaction products. See the Supporting Information (Figure S3) for a more detailed description and an example of the raw data that is collected using this procedure.

Opticn class="Chemical">al emission spectroscopy (OES) was performed using a USB4000 spectrometer from OceanOptics, with a wavelength range of 180–1100 nm, mounted horizontally to the side of the plasma source.

Film Analysis

For characterizan class="Chemical">tion of the deposited material, Co films were grown on Si(100) coupons with 450 nm thermal SiO2. Prior to deposition, the samples were cleaned in situ with an O2 plasma for 2 min. It was found that unloading the samples after the deposition at a table temperature of 300 °C led to significant oxidation of the Co film. Therefore, the substrate table was cooled down from 300 to 100 °C after each deposition to minimize the oxidation. Although the effect of the table temperature was not investigated in detail in this study, the GPC and film purity were found to decrease when the sample temperatures were lowered, which will be addressed in a follow-up publication. The depositions for generating the saturation curves (Supporting Information, Figure S1) were performed on an ALD-grown Co seed layer to avoid nucleation effects. This Co seed layer was deposited by performing 400 cycles of the standard recipe using NH3 plasma on a thermal SiO2 wafer, resulting in a film thickness of approximately 12 nm. Coupons of this seed layer were loaded into the reactor with the table temperature set to 100 °C. After heating the substrate table in vacuum to the standard deposition temperature of 300 °C, the coupons were treated with a NH3 plasma for 3 min to reduce the surface oxidation.

The ALD growth was monitored in n class="Chemical">situ by spectroscopic ellipsometry (SE) using a J.A. Woollam, Inc. M2000U ellipsometer.[52] The dielectric function of the deposited films was parameterized using a B-spline model.[53] The Co film microstructure was studied using transmission electron microscopy (TEM) using JEOL ARM 200F, operated at 200 kV. For the TEM analysis, a lamella was prepared using a focused ion beam (FIB) in a FEI Nova600i NanoLab. X-ray photoelectron spectroscopy (XPS) was performed with a Thermo Scientific KA1066 spectrometer, using monochromatic Al Kα X-rays with an energy of 1486.6 eV. For XPS depth profiling, sputtering was carried out using Ar+ ions with an energy of 500 eV. In addition, four-point probe (FPP) resistivity measurements were done using a Keithley 2400 Sourcemeter and Signatron probe.

Results

Species in NH3 and H2/N2 Plasmas

The similn class="Chemical">arities and differences between the NH3 plasma and H2/N2 plasma were identified by collecting mass spectra in the range m/z = 1–30. As can be seen in Figure , both plasmas mainly contain H2 (m/z = 2), N2 (m/z = 14 and 28), and NH3 (m/z = 15–17). However, the ratio between these species differs for the two plasmas, with the relative amount of NH3 being larger for the NH3 plasma. The mass-to-charge ratios 15 and 16 could correspond to NH (x < 3) species formed in the plasma as well as NH species formed by the dissociation of NH3 in the QMS analyzer. However, NH (x < 3) radicals present in the plasma are likely recombined before being detected in the QMS, indicating that the signals for m/z = 15–17 can mainly be attributed to NH3. Although NH (x < 3) species cannot directly be detected using the QMS, it can be assumed that they are present in the plasma as a consequence of dissociation of NH3.[54] See also the Supporting Information (Table S2) for the assignment of species to corresponding mass-to-charge ratios.

Figure 2

QMS spectra for a NH3 plasma and a H2/N2 plasma. The main plasma species (H2, N2, and NH3) are indicated in the figure. The NH3 pressure was 1.5 mTorr, whereas the H2/N2 pressure was 13 mTorr.

QMS spectra for a NH3 plasma and a n class="Species">H2/N2 plasma. The main plasma species (H2, N2, and NH3) are indicated in the figure. The NH3 pressure was 1.5 mTorr, whereas the H2/N2 pressure was 13 mTorr.

By compn class="Chemical">aring the QMS spectrum for the source gas with the spectrum for the corresponding plasma, it becomes visible which species are formed upon plasma ignition (see Figure S4). When a NH3 plasma is ignited, the signals for m/z ratios 15, 16, and 17 decrease, whereas the signals at m/z = 2, 7, 14, and 28 increase. These observations indicate that part of the NH3 is dissociated, leading to the formation of both N2 and H2. Similarly, in a H2/N2 plasma, N2 and H2 are dissociated upon plasma ignition, followed by the formation of NH3 (see Figure S4b). NH3 production using a H2/N2 plasma occurs mostly at the reactor walls because a three-body reaction in the gas phase is unlikely for the pressures used in this work.[55,56]

The two plasmas were further compn class="Chemical">ared using OES measurements (see Figure S5).[57] The emission spectra for the NH3 and H2/N2 plasmas were found to be very similar. Moreover, the emission peak at ∼336 nm corresponds to the A3Π → Χ3Σ transition of NH and was identified in the spectra for both plasmas (Figure S5b), corroborating the presence of NH species.[58,59]

To study the compon class="Chemical">sition of the H2/N2 plasma as a function of the mixing ratio between the H2 and N2 gases, QMS spectra were collected for different H2/(H2 + N2) ratios. The amount of NH3 species produced in the plasma was found to depend on the mixing ratio. Figure shows the QMS ion currents at m/z ratios 16 and 17 as a function of the H2/(H2 + N2) ratio for a constant chamber pressure of 75 mTorr. This pressure is higher than the standard 13 mTorr used for the ALD process, as explained in Section . The m/z ratios 16 and 17 correspond to NH2+ and NH3+, and their ion currents are a measure for the amount of NH3 produced in the plasma. Figure indicates a maximum in NH3 production around 60–80% H2 in the H2/N2 mixture, in agreement with the previous work.[51,60] Interestingly, the optimum is found close to the ratio between N and H atoms in the NH3 molecule (0.75).[60,61] Because the plasma composition depends strongly on the mixing ratio, selecting the H2/(H2 + N2) ratio is highly important when using a H2/N2 plasma for ALD, as will also be discussed later. On the basis of the optimum found in Figure , a H2/(H2 + N2) mixing ratio of ∼0.77 was employed for further QMS studies and depositions using the AB-H2/N2 process, unless specified otherwise.

Figure 3

QMS ion current at m/z ratios 16 and 17 for H2/N2 plasmas as a function of H2 fraction in the H2/N2 mixture. The H2/(H2 + N2) mixing ratios on the horizontal axis were determined using the ion currents at m/z ratios 2 and 14, corresponding to H2+ and N+ (see the Supporting Information), before igniting the plasma. The total chamber pressure was kept constant at 75 mTorr.

QMS ion current at m/z ran class="Chemical">tios 16 and 17 for H2/N2 plasmas as a function of H2 fraction in the H2/N2 mixture. The H2/(H2 + N2) mixing ratios on the horizontal axis were determined using the ion currents at m/z ratios 2 and 14, corresponding to H2+ and N+ (see the Supporting Information), before igniting the plasma. The total chamber pressure was kept constant at 75 mTorr.

Reaction Products during Plasma Subcycle

A further insight into the use of n class="Chemical">NH3 and H2/N2 plasmas was obtained by studying the reaction products formed during the ALD cycles using time-resolved QMS measurements. First, QMS signals were collected for m/z ratios 40 (Ar+), 59 (Co+), and 66 (HCp+) during the precursor subcycle (Figure S6). On the basis of these results, it can be concluded that HCp (C5H6+, m/z = 66) is released as a product during the precursor half-reaction. Second, QMS signals were recorded during the co-reactant subcycles of the AB-NH3 process (Figure a), the AB-H2/N2 process (Figure b), and the ABC-N2-H2 process (Figure c). Measurements were done for a normal ALD cycle and for a reference cycle without CoCp2 dosing, using plasma exposures of 11 s (see Figure S3). Differences between the signals for the two cycles indicate the formation of species as a consequence of the ALD reactions.

Figure 4

Time-resolved QMS signals for m/z ratios 17 (NH3+), 28 (N2+), 27 (C2H3+, HCN+), 39 (C3H3+, HCN+), and 66 (C5H6+), collected during the plasma subcycle for the (a) AB-NH3 process, (b) AB-H2/N2 process, and (c) ABC-N2-H2 process. A normal ALD cycle and a reference cycle without the CoCp2 precursor dosing were measured, with plasma ignition for 11 s during both cycles (indicated with an arrow in the panels for m/z = 28). The H2/(H2 + N2) mixing ratio of the H2/N2 plasma was ∼0.77.

Time-resolved QMS n class="Chemical">signals for m/z ratios 17 (NH3+), 28 (N2+), 27 (C2H3+, HCN+), 39 (C3H3+, HCN+), and 66 (C5H6+), collected during the plasma subcycle for the (a) AB-NH3 process, (b) AB-H2/N2 process, and (c) ABC-N2-H2 process. A normal ALD cycle and a reference cycle without the CoCp2 precursor dosing were measured, with plasma ignition for 11 s during both cycles (indicated with an arrow in the panels for m/z = 28). The H2/(H2 + N2) mixing ratio of the H2/N2 plasma was ∼0.77.

Figure a shows the results collected during the plasma subcycle of the n class="Chemical">AB-NH3 process. The signals for m/z ratios 17 and 28 are very similar for the ALD cycle and the corresponding reference cycle and are related to the main plasma species, namely, NH3 and N2. The increase in ion current for m/z = 28 and a decrease for m/z = 17 after plasma ignition correspond to the formation of N2 (m/z = 28), which is a consequence of the dissociation of NH3 (m/z = 17). The current for m/z = 17 demonstrates a transient behavior, as NH3 is a “sticky” molecule and the NH3 flow does not stabilize within the time of the exposure.[62] Meanwhile, the initial rise in ion currents (at ∼0 s) for m/z ratios 27, 39, and 66 upon plasma ignition for the (normal) ALD cycle can be attributed to the release of reaction products (see Table S2). This rise is not observed for the reference cycle without CoCp2 dosing. The increase for m/z = 66, assigned to HCp+ (C5H6+), upon plasma ignition indicates the elimination of the Cp ring from the surface. A similar increase in ion current was observed for m/z = 65 (C5H5+, data not shown). The detection of HCp+ reveals that some of the Cp ligands are still present on the surface after the CoCp2 subcycle. The mass-to-charge ratio 27 corresponds to C2H3+ or HCN+ and m/z = 39 to C3H3+ or C2HN+. The presence of, for example, HCN and C2HN might be caused by the reaction of CH and NH species in the plasma. The detection of C2H3+ and C3H3+ can be explained by dissociative ionization of HCp in the QMS (see the cracking pattern in Figure S7) and/or by the formation of C2H4 and C3H4 in the plasma because of dissociation of HCp. Such production channels can unfortunately not be distinguished using the current experimental setup.

The QMS results for the AB-H2/N2 process n class="Chemical">are shown in Figure b. The ion currents for m/z ratios 17 and 28 behave very similar for the ALD cycle and the reference cycle and indicate the formation of NH3 (m/z = 17) and consumption of N2 (m/z = 28) in the H2/N2 plasma. These findings are in line with the QMS measurements discussed in Section (and as shown in Figure S4). Note that the signal for m/z = 17 continues to increase during the plasma exposure because of the “sticky” nature of NH3 and/or ongoing stabilization of the NH3 production.[62] However, the current for m/z = 17 starts to drop after the plasma exposure, accompanied by a small increase in the signal for N2, indicating that no more N2 is being consumed. The ion currents for m/z ratios 27, 39, and 66 for the ALD cycle increase when the plasma is started (at t ≈ 0 s), similar to the data shown in Figure a. This increase can be explained by the release of reaction products, as was discussed for the AB-NH3 process.

To examine the role of the NH species in the plasma in the reaction mechanism, the n class="Species">H2/N2 plasma was replaced by separated N2 and H2 plasma steps in an ABC-type cycle (see Figure ). The results for the ABC-N2-H2 process in Figure c show that no NH3 was present during the N2 exposure, as can be expected. Moreover, upon ignition of the N2 plasma (at t ≈ 0 s), a minimal amount of HCp+ (m/z = 66) is detected (revealed by the small difference with the reference cycle), which is much smaller than for the AB-NH3 and AB-H2/N2 processes. Upon ignition of the subsequent H2 plasma, a rise in ion currents for both m/z = 17 and 28 (at t ≈ 32 s, observed for the ALD cycle and also for the reference cycle) indicates that NH3 and N2 are released. However, the amounts are almost negligible and are limited by the amount of nitrogen-containing species adsorbed to the substrate and reactor wall after the N2 plasma exposure. The H2 plasma mostly leads to the detection of C2H3+/HCN+ (m/z = 27) and C3H3+/C2HN+ (m/z = 39) species and no significant amount of HCp+. The limited amount of HCp+ detected during both plasma exposures indicates that the Cp ring is not eliminated as a whole but rather dissociated because of the interaction with the plasmas.

Compn class="Chemical">arison of the results for the three different ALD processes provides an insight into the similarities and differences in reaction mechanisms. Except for the differences in plasma species (m/z = 17 and 28), the results in Figure a,b show very similar reaction products for the AB-NH3 and AB-H2/N2 processes. These analogies between the two AB processes suggest a similar reaction pathway, where Cp ligands are eliminated from the surface during both the precursor and plasma subcycle. QMS measurements for the ABC-N2-H2 process show significant differences in terms of plasma species and reaction products (see Figure c), as compared to the AB processes, suggesting a different reaction pathway.

Film Properties

Before characterizan class="Chemical">tion of the material properties, the ALD behavior of the two AB processes was studied by determining the GPC as a function of the CoCp2 dosing and the plasma exposure times. As can be seen in Figure S1, both the precursor and co-reactant subcycles demonstrated a self-limiting behavior for the NH3 plasma as well as for the H2/N2 plasma processes. Moreover, the saturation curves for the two processes look very similar, in line with the finding that the two AB processes show similarities in terms of plasma composition and reaction pathways as discussed in Sections and 3.2. The GPC saturates to a value of 0.29 ± 0.02 Å, which is slightly lower than that reported by Kim and co-workers (0.48 Å).[26,37]

The material propern class="Chemical">ties for the three different ALD processes were investigated for the films deposited using 1000 cycles. A film deposited using the AB-NH3 process was investigated using TEM after preparation of a lamella using a FIB. The cross-sectional images in Figure reveal that the film is polycrystalline and the crystal grains can clearly be observed. The film forms a closed layer of approximately 29 nm thick and has a low roughness. SE modeling yielded a film thickness of ∼32 nm. The difference between the thicknesses derived from SE and TEM is thought to be due to the film roughness, which is not included in the SE modeling.

Figure 5

(a,b) Cross-sectional TEM images of a Co film deposited by performing 1000 ALD cycles of the AB-NH3 process.

(a,b) Cross-sectionn class="Chemical">al TEM images of a Co film deposited by performing 1000 ALD cycles of the AB-NH3 process.

The material propern class="Chemical">ties obtained using the different processes are shown in Table . As can be seen, for both the AB-NH3 and the AB-H2/N2 process, the film thickness (determined using SE) is ∼25 nm after 1000 cycles, corresponding to an average GPC of approximately 0.25 Å. It is noted that the sample prepared for the TEM analysis was different from the sample listed in Table and the film thickness was slightly higher. The ABC-N2-H2 process resulted in an average GPC as high as 0.44 Å. The films deposited using the AB-NH3 and AB-H2/N2 processes both demonstrate a low resistivity (41–42 μΩ cm, as compared to a Co bulk resistivity of 6.24 μΩ cm) and have a similarly low impurity content.[63] The resistivity values obtained for the AB processes are slightly higher than the best reported value for Co ALD and lie within the range of values obtained for processes using NH3 or H2/N2 plasma as the co-reactant (10–140 μΩ cm, see Table ). This is in contrast to the film deposited using the ABC-N2-H2 process, which has a high resistivity (>1000 μΩ cm), likely caused by considerable amounts of impurities found in the film (O, N, and C add up to 25 at. %).

Table 2

Material Properties of Co Films for the Three Different ALD Processes As Determined from SE, FPP, and XPSa

ALD process	d (nm)	ρ (μΩ·cm)	[O] (at. %)	[N] (at. %)	[C] (at. %)
AB-NH3	25	41	0.5 ± 0.3	2.3 ± 0.5	0.6 ± 0.6
AB-H2/N2	25	42	1.0 ± 0.4	2.8 ± 0.5	0.7 ± 0.7
ABC-N2-H2	44	1 × 103	10.0 ± 0.5	8.4 ± 0.5	7 ± 1

1000 ALD cycles were performed. The impurity contents were determined using XPS after sputtering with Ar+ for 6 min.

1000 ALD cycles were performed. The impurity n class="Chemical">contents were determined using XPS after sputtering with Ar+ for 6 min.

XPS measurements showed that the surface of the Co films is slightly oxidized (Figure S8). After n class="Chemical">Ar+ sputtering, the O contents of the films deposited using both AB processes (NH3 and H2/N2) are however found to be close to 0 at. %, and metallic Co 2p peaks were detected at around 780.2 eV.[64] Apart from minimal amounts of O, C, and N, no other impurities were detected in the Co films grown using the AB-NH3 and AB-H2/N2 processes. For the film deposited using the ABC-N2-H2 process, significant amounts of O, C, and N were detected in the bulk of the film (see the XPS results in Figures S8 and S9). It was found that exchanging the H2 and N2 plasma exposures, corresponding to an ABC-type cycle first with the H2 plasma followed by the N2 plasma, led to a comparable impurity content of approximately 25 at. % (see Table S3).

The significant higher GPC and ren class="Chemical">sistivity for the ABC-N2-H2 process can be explained by the impurity incorporation, leading to a lower film density and/or a higher surface roughness. The difference between the two AB processes on one hand and the ABC-N2-H2 process on the other hand can most likely be attributed to the absence of NH species in the plasmas for the ABC-N2-H2 process (see Section ), as will also be discussed later.

As described in Section , the amount of n class="Chemical">NH3 produced in the H2/N2 plasma depends on the ratio between the two source gases. To study the effect of the NH3 concentration in the plasma, a set of Co films was deposited for various H2/(H2 + N2) ratios, using a constant pressure of 13 mTorr for the gas mixture. The properties of the Co films were studied using SE, FPP, and XPS, and the results are shown in Table . The thicknesses of the films varied slightly as a function of H2 fraction, with higher film thicknesses (∼20 nm) obtained for intermediate ratios of 0.35 and 0.42. The films deposited using low mixing ratios of 0.13 and 0.23 had resistivity values that were too high to be measured (>109 μΩ cm). For higher mixing ratios, the resistivity of the films decreased with increasing H2 fraction, which correlates to a significant decrease in O impurity content. In addition to the O content, differences in film structure (e.g., crystallinity, density, or porosity) might affect the film resistivity, and further research would be required to obtain a more detailed understanding. The O incorporation is likely due to the dissociation of background species (such as H2O) in the plasma during film growth or post-deposition oxidation of the films.[65,66] Interestingly, the C and N contents of the films are relatively constant and do not show a clear trend as a function of H2/(H2 + N2) ratio. The relation between the H2/(H2 + N2) ratio and the material properties will be further discussed in Section .

Table 3

Material Properties of Co Films for Different H2/N2 Mixing Ratiosa

H2/(H2 + N2)	d (nm)	ρ (μΩ·cm)	[O] (at. %)	[N] (at. %)	[C] (at. %)
0.13	17.9	>109	7.0 ± 0.2	9.5 ± 0.5	4.0 ± 0.9
0.23	19.9	>109	6.2	9.5	3.8
0.35	20.3	3.6 × 108	5.6	9.3	3.8
0.42	20.3	2.5 × 103	4.2	9.6	3.5
0.52	19.6	1.5 × 103	4.4	8.8	3.8
0.77	17.5	78	0.2	8.4	4.6

800 ALD cycles were performed. The total pressure was kept constant at 13 mTorr. The impurity contents were determined using XPS after sputtering with Ar+ for 3 min. Typical errors in the impurity content are indicated in the top row.

800 ALD cycles were performed. The totn class="Chemical">al pressure was kept constant at 13 mTorr. The impurity contents were determined using XPS after sputtering with Ar+ for 3 min. Typical errors in the impurity content are indicated in the top row.

Discussion

Role of NH Species

The results addressed in Section can provide an inn class="Chemical">sight into the role of different plasma species during the Co film growth. Specifically, the data in Table show that the impurity content decreases with the H2 fraction in the H2/N2 plasma. Meanwhile, the ion current at m/z = 17, indicating NH3 in the plasma, increased when varying the H2/(H2 + N2) ratio up to ∼0.77 (see Figure S10). This finding is in line with the dependence of the NH3 content on the mixing ratio as was discussed in Section . In Figure , the Co concentration of the films is plotted as a function of the ion current for m/z = 17. Interestingly, this graph shows a linear trend, suggesting that a higher amount of NH species in the plasma leads to a higher film purity. Note that with a higher NH3 concentration in the plasma, also the amount of NH radical species increases, as a consequence of NH3 dissociation. The increase in Co purity is mostly due to the decline in O content, suggesting that NH or NH3 species facilitate the removal of O from the material or lead to a film which is less prone to post-deposition oxidation. Additional research is required to distinguish between these two possible explanations.

Figure 6

Co content from XPS as a function of QMS ion current at m/z = 17. The QMS ion current is a measure for the NH3 production in the H2/N2 plasma and was varied by changing the H2/N2 mixing ratio of the source gas. The pressure of the H2/N2 gas mixture was kept constant at 13 mTorr. The Co content was determined using XPS on films obtained by performing 800 ALD using the various H2/N2 mixing ratios. XPS was carried out after sputtering with Ar+ ions for 3 min. The resistivity values of the Co films are indicated in the figure, in which the dashed line represents a linear fit through the data. It is noted that the decrease of the film resistivity as a function of NH concentration might be related to the changes in the film structure, aside from the increased film purity.

Co n class="Chemical">content from XPS as a function of QMS ion current at m/z = 17. The QMS ion current is a measure for the NH3 production in the H2/N2 plasma and was varied by changing the H2/N2 mixing ratio of the source gas. The pressure of the H2/N2 gas mixture was kept constant at 13 mTorr. The Co content was determined using XPS on films obtained by performing 800 ALD using the various H2/N2 mixing ratios. XPS was carried out after sputtering with Ar+ ions for 3 min. The resistivity values of the Co films are indicated in the figure, in which the dashed line represents a linear fit through the data. It is noted that the decrease of the film resistivity as a function of NH concentration might be related to the changes in the film structure, aside from the increased film purity.

Overall, the findings show that cn class="Chemical">areful consideration is needed when a H2/N2 plasma is employed as the co-reactant for ALD. Moreover, the dependence on the NH concentration explains why H2 plasmas or N2 plasmas are not suitable as co-reactants, mainly because of the lack of NH species. In addition, the fact that NH3 gas cannot be used as the co-reactant in a thermal ALD process substantiates the hypothesis that NH (x < 3) plasma species are necessary. The results thus suggest that NH species play a crucial role in the Co growth. This is further elaborated on in Section , where a reaction mechanism is proposed.

Interestingly, the dependence of the film ren class="Chemical">sistivity on the H2/N2 ratio is in agreement with the work of Yoon et al. and Hong et al. for the deposition of Co and Ru, respectively.[12,26] Both studies observed high-resistivity values for high N2 fractions and suggested that NH species in the plasma play an important role in the growth.[12,26] In addition, in the work of Ten Eyck et al., the H2/N2 ratio was optimized in order to maximize the NH generation in the plasma.[10] Finally, ALD processes for Ru, Ni, and Ag have been reported to improve when NH3 plasmas are used instead of H2 plasmas (in terms of a higher GPC and lower resistivity), which also suggests an important role of NH species present in the NH3 plasmas.[18−20]

Reaction Mechanism

The QMS studies addressed in Section can help to unravel the reacn class="Chemical">tion mechanisms for the ALD growth of Co. Before reviewing the main QMS results, it is relevant to obtain an understanding of the surface groups present on Co after plasma exposure (i.e., prior to precursor dosing). Although the interaction of a NH3, H2, or N2 plasma with a Co surface has not been studied in detail, the surface science literature on Co provides some insights. For instance, Kizilkaya et al. investigated the stability of NH groups on Co after NH3 exposure and found that NH3 adsorption on Co is followed by decomposition, resulting in NH groups remaining on the surface.[67] In addition, Wang et al. studied the nitridation of transition-metal surfaces and calculated that Co is not prone to nitridation, although a N-covered surface is stable after formation.[68] On the basis of these reports, it can be expected that NH groups are present on the Co film after NH3 or H2/N2 plasma exposure. H adsorbed on the Co surface seems less likely because H is reported to desorb (as H2) from Co between the temperatures of approximately 300 and 400 K, which is lower than the sample temperature during ALD (300 °C = 573 K table temperature).[69−71] Notably, the role of N-containing surface species has previously been addressed for the ALD of Pt and Ag using N2 or NH3 plasmas.[19,72,73] Specifically, N-containing species were proposed to adsorb during the N2 or NH3 plasma exposures and to react with the precursor during the subsequent precursor dose.

In short, the QMS studies in Section reven class="Chemical">aled the following. During the precursor and co-reactant subcycles, HCp (C5H6) is released for both the AB-NH3 and AB-H2/N2 processes. In addition, fragments of the Cp ring (e.g., CH+ and CHN+) were detected during the plasma exposure. Furthermore, variation of the H2/N2 ratio showed that the NH3 species (NH) play an important role in the growth mechanism. These observations suggest that during the precursor subcycle, the CoCp2 molecules chemisorb to the surface and that part of the Cp rings react with H atoms present at the surface, followed by the release of HCp. This is illustrated in step “A” in Figure . Considering that chemisorbed H is not stable on a Co surface, it is speculated that NH surface groups present after the plasma exposure provide the source of H (NH with the subscript y is used to indicate the distinction from the NH species in the plasma). Additional research such as surface Fourier-transform infrared spectroscopy (FTIR) is required to identify the surface groups to which the precursor binds. Possibly the precursor chemisorbs directly to unoccupied Co surface sites or to the NH species after the release of HCp.

Figure 7

Schematic representation of the proposed reaction mechanisms during the ALD of Co using NH3 or H2/N2 plasma as the co-reactant. During the precursor subcycle (“A”), CoCp2 binds to the surface, accompanied by the release of HCp. The site to which CoCp2 chemisorbs is indicated as “X” because it remains to be identified. In the co-reactant subcycle (“B”), the NH species from the plasma cause the release of HCp and fragments thereof and lead to the formation of NH surface groups. NH with the subscript y is used to indicate the distinction from NH species in the plasma.

Schematic represenn class="Chemical">tation of the proposed reaction mechanisms during the ALD of Co using NH3 or H2/N2 plasma as the co-reactant. During the precursor subcycle (“A”), CoCp2 binds to the surface, accompanied by the release of HCp. The site to which CoCp2 chemisorbs is indicated as “X” because it remains to be identified. In the co-reactant subcycle (“B”), the NH species from the plasma cause the release of HCp and fragments thereof and lead to the formation of NH surface groups. NH with the subscript y is used to indicate the distinction from NH species in the plasma.

In the co-reacn class="Chemical">tant subcycle, the Cp ligands remaining after precursor dosing are eliminated by the NH3 radical species from the plasma (NH). This reaction leads to the formation of HCp and C-, H-, and N-containing fragments, as illustrated in step “B” in Figure . The release of HCp was also observed during the NH3 plasma exposure in the work of Oh et al.[74] Moreover, Shimizu et al. and Yuan et al. used NH3 as the co-reactant gas for Co and Ni hot-wire ALD using CoCp2 and NiCp2 as precursors, respectively, and claimed that the NH2 gas-phase species formed on the hot wire are needed for the dissociation of the metal–Cp bond.[38,75] In the reaction mechanism shown in Figure , the NH species generated in the plasma thus fulfill this role. Furthermore, the NH species as well as NH3 lead to the formation of NH surface groups, which react with the precursor molecules in the next cycle. Although NH3 can dissociate after adsorption on a clean Co surface, it is believed that the NH plasma species are essential for complete ligand removal.

Note that a small amount of N (∼2–3 at. %) is present in the n class="Chemical">Co films deposited using the AB-NH3 and AB-H2/N2 processes and that this content is slightly higher (∼4.5 at. %) in the (sub)surface region (see Figure S9). This corroborates the expectation based on the surface science literature that N-containing species (e.g., NH) are present and that they are more stable on the surface than in the bulk. However, a remaining question is how the surface NH species leave the film, such that they are not incorporated in the film. Because the films contain only a small amount of N, it is speculated that most of the N diffuses out of the surface region and desorbs in the form of either NH3 or N2.

Interestingly, an ann class="Chemical">alogy appears to exist between the proposed reaction mechanism and the ALD growth of noble metals using O2 as the co-reactant. During noble metal ALD, chemisorbed O plays a crucial role in the reaction mechanism, while the stability of noble metal oxides is limited.[76] This is similar to the ALD of Co, where NH species appear needed for chemisorption of the precursor and removal of the Cp ligands but are not built into the film.

Conclusions

The use of H2-, N2-, and n class="Chemical">NH3-based plasmas as co-reactants for the ALD of Co using CoCp2 was investigated. A direct comparison was made between ALD processes with a NH3 plasma and a combined H2/N2 plasma as the co-reactant. It was shown that the NH3 and H2/N2 plasmas contain comparable plasma species, including NH, although the relative concentrations are different. Moreover, the reaction products detected during the plasma subcycle are very similar, suggesting analogous reaction pathways. Variation of the mixing ratio of H2 and N2 in the H2/N2 plasma showed that the lowest resistivity is achieved for the ratios with the highest NH concentration. In addition, the films deposited using plasmas with a lower NH concentration contained significant amounts of O impurities. Deposition using an ABC-type cycle with consecutive N2 and H2 plasma steps resulted in a Co film with significant amounts of impurities and a high resistivity. These insights indicate that the NH species present in both the NH3 and H2/N2 plasmas are necessary for eliminating the precursor ligands and for obtaining high-purity films, which explains why H2 plasmas or N2 plasmas are not suitable as co-reactants. Furthermore, on the basis of the QMS results and surface science literature, a reaction mechanism was proposed where NH play an essential role, leading to the release of HCp in both subcycles.

Overall, the work shows that the plasma n class="Chemical">composition can strongly affect the obtained material properties and that the co-reactant of a metal ALD process should be carefully selected. Specifically, when using a H2/N2 plasma for Co ALD, selecting the H2/(H2 + N2) ratio is crucial. Other literature reports further illustrate the importance of choosing a suitable co-reactant and demonstrate that NH3 or H2/N2 plasmas can be preferred over other co-reactants.[17−20,72] This suggests that the findings of this work can be generalized and can also apply to other metal ALD processes.