Reaction Mechanisms during Atomic Layer Deposition of AlF3 Using Al(CH3)3 and SF6 Plasma.

Metal fluorides generally demonstrate a wide band gap and a low refractive index, and they are commonly employed in optics and optoelectronics. Recently, an SF6 plasma was introduced as a novel co-reactant for the atomic layer deposition (ALD) of metal fluorides. In this work, the reaction mechanisms underlying the ALD of fluorides using a fluorine-containing plasma are investigated, considering aluminum fluoride (AlF3) ALD from Al(CH3)3 and an SF6 plasma as a model system. Surface infrared spectroscopy studies indicated that Al(CH3)3 reacts with the surface in a ligand-exchange reaction by accepting F from the AlF3 film and forming CH3 surface groups. It was found that at low deposition temperatures Al(CH3)3 also reacts with HF surface species. These HF species are formed during the SF6 plasma exposure and were detected both at the surface and in the gas phase using infrared spectroscopy and quadrupole mass spectrometry (QMS), respectively. Furthermore, QMS and optical emission spectroscopy (OES) measurements showed that CH4 and CH y F4-y (y ≤ 3) species are the main reaction products during the SF6 plasma exposure. The CH4 release is explained by the reaction of CH3 ligands with HF, while CH y F4-y species originate from the interaction of the SF6 plasma with CH3 ligands. At high temperatures, a transition from AlF3 deposition to Al2O3 etching was observed using infrared spectroscopy. The obtained insights indicate a reaction pathway where F radicals from the SF6 plasma eliminate the CH3 ligands remaining after precursor dosing and where F radicals are simultaneously responsible for the fluorination reaction. The understanding of the reaction mechanisms during AlF3 growth can help in developing ALD processes for other metal fluorides using a fluorine-containing plasma as the co-reactant as well as atomic layer etching (ALE) processes involving surface fluorination.

Introduction

Metal fluorides typically have a low refractive index (1.3–1.6) and a wide band gap (>10 eV), and due to these properties, they are widely used in optical devices.[1−6] Recently, metal fluorides, AlF3 and LiF in particular, have also received considerable attention for application in Li-ion batteries, either as a protective layer on the electrodes or as an active electrode material.[7−14] Furthermore, metal fluorides are of interest as a catalyst for the synthesis of chlorofluorocarbons and hydrofluorocarbons[15] and have been employed in photovoltaics, e.g., as an electron-selective contact.[16−24] A variety of methods have been used for the deposition of metal fluorides, including evaporation,[1,25,26] sputtering,[27−29] and ion-assisted deposition[30−32] as well as atomic layer deposition (ALD).[33−41] Since ALD is a chemical vapor deposition technique that relies on sequential and self-limiting surface reactions, it typically results in highly uniform and conformal ultrathin films.[42,43] These characteristics are often relevant for the aforementioned applications of metal fluorides.[42,44] ALD of metal fluorides has been reported using various co-reactants, such as NH4F, the volatile metal fluorides TaF5 and TiF4, and HF.[33−39,41] Moreover, Lee et al. demonstrated the ALD of AlF3, ZrF4, MnF2, HfF4, MgF2, and ZnF2 using HF from a HF–pyridine solution.[45,46] For the thermal ALD of AlF3 using HF, it was proposed that the reaction occurs according to the following two reaction equations:[45]In the precursor subcycle (eq ), the Al(CH3)3 (trimethylaluminum, TMA) precursor reacts with HF that remained adsorbed to the surface after the preceding co-reactant subcycle. Upon this reaction with HF, one or more CH4 molecules are released, resulting in AlF(CH3)3– (x = 1, 2) surface species. On the basis of quartz crystal microbalance (QCM) measurements, it was found that mainly AlF(CH3)2 is formed.[45] In the following co-reactant dose (eq ), the remaining CH3 ligands combine with H (from HF) and are eliminated as CH4, accompanied by the fluorination of the surface to AlF3. The amount of AlF(CH3)3– species adsorbed to the surface was found to decrease with temperature, explaining why this ALD process is strongly dependent on the deposition temperature.[45,47] Interestingly, the temperature-dependent desorption of AlF(CH3)3– species makes that TMA and HF can also be used for atomic layer etching (ALE) of Al2O3 at temperatures >250 °C or AlF3 at >150 °C.[45,47−53] The main reactions that govern the Al2O3 ALE process are given by[48,50]In eq , HF is responsible for the self-limiting fluorination of the Al2O3 surface. Subsequently, the formed AlF3 surface layer can be removed by dosing Al(CH3)3, leading to the formation of volatile AlF(CH3)2,[50] as described by eq . The reaction in eq can be considered a ligand-exchange transmetalation reaction, where TMA accepts F and donates CH3 to the surface.[47,54] As mentioned, the volatility of AlF(CH3)3– species is dependent on the temperature, and the desorption of these species at high temperatures enables ALE. In addition to the reaction of eq , Al(CH3)3 can react with Al2O3 after the AlF3 layer is etched, resulting in AlCH3 species on the surface, which are responsible for the self-limiting behavior of the surface reactions.

Recently, we presented an ALD process for AlF3 as well as an ALE process for Al2O3,[55,56] using SF6 plasma as an alternative, easy-to-handle, and readily available co-reactant. SF6 is a stable and nontoxic gas, which is widely used in the industry. Notably, F-containing plasmas such as SF6, NF3, CF4, and C2F6 are employed for etching of Si-based materials (e.g., Si, SiO2, Si3N4).[57] An SF6 plasma is characterized by high concentrations of F radicals and F– ions, which are known to be approximately 103 times higher than the concentrations of S and S+.[58] SF6 gas can form electronegative plasmas where significant concentrations of positive ions (mainly SF5+ and SF4+) are balanced by negative ions (mainly F– and SF6–) and electrons.[58−60] In an inductively coupled SF6 plasma, the dominant neutral species are reported to be F2, SF6, and SF4.[58−60] It is noted that F is very reactive due to a very high electronegativity and electron affinity. The high concentration of reactive F radicals makes SF6 plasma a suitable co-reactant for ALD of metal fluorides.[61]

The ALD process using SF6 plasma as co-reactant yields high-purity films[56] with material properties similar to those reported in the literature for AlF3 films deposited using various methods.[1,39,62,63] Since an SF6 plasma etches Si or SiO2 substrates, ALD-grown Al2O3 was used as a protective coating.[64] A decrease in the growth per cycle (GPC) was found from 1.50 Å at a deposition temperature of 50 °C to 0.55 Å at 300 °C, similar to previous findings for the thermal ALD process using TMA and HF.[39,45,56] This decrease of the GPC was believed to be related to a lower density of HF surface species at high temperatures (see eq ). To extend the use of SF6 plasma as co-reactant to the ALD of other metal fluorides, understanding of the chemistry behind the process is required. Moreover, it is valuable to investigate the competition between ALD and ALE reactions during processes involving surface fluorination. For these reasons, this Article provides insight into the reaction mechanisms taking place during TMA–SF6 plasma cycles.

This Article is structured as follows. First, the experimental conditions are described in Section . In Section , surface infrared spectroscopy results are discussed, revealing the surface reactions during Al(CH3)3–SF6 plasma cycles and a transition to etching of Al2O3 at higher temperatures. This is followed by an identification of the plasma species and gas-phase reaction products in Section . On the basis of the insights obtained from these experiments, a detailed description of the reaction mechanisms of AlF3 ALD is given in Section .

Experimental Section

ALD Reactors and Conditions

The quadrupole mass spectrometry (QMS) and optical emission spectroscopy (OES) experiments were performed using an Oxford Instruments FlexAL reactor.[65] For the surface infrared (IR) studies, a home-built ALD tool was used, which has a very similar pumping system and plasma source as the FlexAL reactor.[66] Both reactors reach a base pressure of ∼10–6 Torr and are equipped with an inductively coupled plasma (ICP) source, operated at a radio frequency of 13.56 MHz. The conditions for the AlF3 ALD process in the FlexAL reactor have been reported elsewhere.[56] Briefly, a TMA dose of 80 ms, followed by a purge step of 6 s, an SF6 plasma exposure of 10 s (300 W power), and a final purge step of 4 s were used. The TMA precursor (Sigma-Aldrich, >99.9999%) was contained in a stainless steel canister, kept at a temperature of 30 °C. The chamber pressure was set to 15 and 50 mTorr during the TMA dose and SF6 plasma exposure, respectively. The substrate table and reactor walls were kept at temperatures of 200 and 120 °C, respectively. Since the surface area of the reactor wall is significantly larger than of the substrate table, a large part of the QMS signal originates from reactions occurring on the wall at a temperature of 120 °C.

For the in situ infrared spectroscopy studies in the home-built reactor, an SF6 plasma pressure of 15 mTorr and a plasma power of 100 W were used. Measurements were performed on Aerosil OX50 SiO2 powders, pressed on a tungsten grid. As will be discussed in Sections and 4, a transition from AlF3 ALD to Al2O3 ALE was observed in the infrared spectroscopy experiments, whereas this was not observed for the temperatures studied in the FlexAL reactor.[56] This difference can likely be attributed to the higher temperature of the SiO2 powder as compared to the Si wafer coupons. For depositions on the SiO2 powder, the tungsten grid was heated directly by resistive heating using DC current, and the temperature was monitored using a thermocouple. This method of heating is significantly different from the heating employed for the deposition on blanket wafers in the FlexAL reactor, where wafer coupons are placed on a carrier wafer, which is loaded onto a heated substrate table. Due to reduced thermal contact at low working pressure between the wafer coupons, the carrier wafer, and the table heater, the actual sample temperature is significantly lower than the substrate table.[67] For instance, the sample temperature is estimated to be 200–220 °C for a table temperature of 300 °C. This is in contrast to the depositions on SiO2 powder where the actual temperature of the powder surface is expected to be in agreement with the set temperature. Moreover, it was found that the SF6 plasma exposure resulted in slight heating (∼20–30 °C) of the tungsten grid, while this effect is expected to be less pronounced on wafer coupons.

Reaction Mechanism Studies

The in situ surface spectroscopy studies were done with a Bruker Vector Fourier transform infrared spectrometer, as described in previous work.[68] Prior to AlF3 ALD, the SiO2 powder was coated with Al2O3 using 50 thermal (TMA + H2O) ALD cycles at ∼275 °C. Sets of five TMA–SF6 plasma cycles were performed at temperatures between 100 and 250 °C, while collecting an infrared spectrum after each subcycle. After each set of the TMA–SF6 plasma cycles, 10 Al2O3 ALD cycles were repeated at ∼275 °C to obtain the same starting surface. The TMA dose and SF6 plasma exposure time for these experiments were 1.1 and 30 s, respectively, which was confirmed to be sufficient to reach saturation of the surface reactions on the Al2O3-coated SiO2 powder. To correct for changes in surface area of the powder due to film growth, the absorbance spectra were normalized on the basis of the signal for Al(CH3)3 adsorption on Al2O3.[69] For this correction, the normalized integrated infrared absorbance corresponding to CH3 stretching vibrations in the range of 3060–2800 cm–1 was used. This absorbance peak can serve as a measure for the quantity of adsorbed Al(CH3)3 and was determined by calculating the difference spectrum using measurements before and after dosing Al(CH3)3 on the Al2O3-coated powder.

Quadrupole mass spectrometry (QMS) measurements were performed using a Pfeiffer Vacuum Prisma QME-200 (mass-to-charge ratio m/z = 1–200) connected to a flange on the side of the FlexAL reactor chamber. The selection of measured m/z ratios was based on initial screening and mostly corresponds to SF and CHF4– species. The procedure for time-resolved QMS measurements was similar to a method described previously.[70] In short, six m/z ratios were measured simultaneously, of which one was always m/z = 40. This m/z value corresponds to Ar+ and is used as a reference. For each type of recipe, five cycles were performed and averaged. OES was done using a four-channel AvaSpec ULS2048 spectrometer from Avantes (200–1100 nm) as a line-of-sight measurement perpendicular to the substrate table (i.e., the fiber is installed vertically at the top of the plasma source). For the OES measurements, an SF6 gas flow of 50 sccm was used instead of the standard 100 sccm to increase the residence time of the plasma species.

Results

Surface Reactions

In situ surface infrared spectroscopy was performed to investigate the surface reactions taking place during the subcycles. Figure shows absorbance spectra in the range of 4250 to 2500 cm–1 for the TMA and the SF6 plasma subcycles performed at temperatures between 100 and 250 °C. For all temperatures, an absorbance gain is observed in the spectrum for the TMA subcycle in the region of 3060–2800 cm–1, which corresponds to the formation of CH3 stretching vibrations.[71−73] The presence of CH3 can be explained by AlF(CH3)2, AlF2(CH3), or AlCH3 species formed upon the reaction of TMA with the surface.[47] With increasing temperature, the absorbance gain decreases, indicating less Al(CH3)3 adsorption at high temperatures, which is in line with the decrease of the GPC as a function of temperature.[56] This temperature dependence of the CH3 stretching vibrations is similar to what was observed by DuMont and George for thermal ALD/ALE using HF.[47] Furthermore, an absorbance loss can be seen in the range of 3700–3060 cm–1 in the spectra for 100 and 150 °C, which is attributed to the consumption of HF species on the AlF3 surface.[47,74] These HF species are formed during the SF6 plasma exposure, as evidenced by the positive absorbance peaks in the spectra for the plasma subcycle at 100 and 150 °C. Note that HF was also detected as a gas-phase reaction product during the SF6 plasma exposure using QMS, as will be addressed in Section . At temperatures of 200 and 250 °C, negligible peaks corresponding to HF are observed, which reveals that no significant amount of HF is consumed or adsorbed during the TMA or SF6 plasma subcycle, respectively. Although the removal of HF surface species is only observed at 100 and 150 °C, the spectra for the TMA subcycle indicate the formation of CH3 groups at all investigated temperatures. This implies that, at higher temperatures, Al(CH3)3 reacts directly with the AlF3 film, likely by accepting F from the surface and forming CH3 surface groups.[48]

Figure 1

Difference spectra of the infrared absorbance for the TMA subcycle and for the SF6 plasma subcycle for deposition temperatures between 100 and 250 °C. Spectra were collected after TMA dosing or SF6 plasma exposure and were referenced to the spectrum for the preceding subcycle. The pairs of spectra were given an offset on the vertical axis for clarity.

Multiple cycles were performed, and the collected spectra for each cycle were referenced to the first cycle to gain insight into the progression of the film with the number of cycles. From these measurements, the data collected after two and after five cycles are displayed in Figure for temperatures between 100 and 250 °C. The spectra for 100 and 150 °C show a small positive absorbance peak starting around 775 cm–1, which can be attributed to the Al–F stretching vibrations in AlF3.[75,76] Unfortunately, the end of the detector range lies at ∼700 cm–1, meaning that only a part of the peak corresponding to the Al–F stretching vibrations could accurately be measured. However, the positive absorbance between ∼775 and ∼750 cm–1 increases when additional cycles are performed, in line with the growth of the AlF3 film. Interestingly, an absorbance loss can be observed in the region below 950 cm–1 for temperatures of 200 and 250 °C. The region around ∼1030–550 cm–1 corresponds to the Al–O stretching vibrations in amorphous Al2O3, and the absorbance loss therefore suggests the removal of Al2O3. For a temperature of 250 °C, the absorbance loss is larger than for 200 °C, indicating that more Al2O3 is etched, in agreement with what can be expected on the basis of the literature.[45,47] The data in Figure thus show that a transition from AlF3 ALD to Al2O3 ALE occurs at higher temperatures.

Figure 2

Absorbance spectra collected after performing two and five TMA–SF6 plasma cycles, both referenced to the spectrum collected after performing one TMA–SF6 plasma cycle on Al2O3-coated SiO2 at the corresponding temperature. The pairs of spectra were given an offset on the vertical axis for clarity.

The ALE process using TMA and SF6 plasma developed in the same reactor showed an etch rate of approximately 3 Å at 250 °C and 2 Å at 200 °C, while no etching was observed for temperatures below 150 °C, which indicates a strong dependence on the temperature and corroborates the observed transition from AlF3 ALD to Al2O3 ALE.[55] In addition, it was confirmed that an SF6 plasma can be used for fluorination of Al2O3, similar to the use of HF for thermal ALE. Al2O3 films (∼10 nm thick) were exposed to SF6 plasma and subsequently analyzed by spectroscopic ellipsometry (SE; see Figure S1). It was found that exposing Al2O3 to SF6 plasma leads to a small increase in thickness and a decrease in refractive index, due to self-limiting fluorination of the top surface. Furthermore, XPS results presented in Figures S2 and S3 demonstrate that most of the F is located in the top layer of the Al2O3 and that AlOF bonds are present in addition to Al2O3. Similar self-limiting behavior can be expected for fluorination of MgO, HfO2, and ZnO films, since their fluorides are nonvolatile and have been deposited by ALD using HF.[46]

Gas-Phase Reaction Products

QMS was employed to study the formation of gas-phase reaction products during AlF3 ALD. Figure shows time-resolved QMS data for selected mass-to-charge (m/z) ratios 15, 16, 19, and 89, which correspond to CH4 and key plasma species (F and SF). In addition, Table lists the most relevant m/z ratios and the assignment of these m/z ratios to corresponding parent species. The increase in signals for m/z = 15 (CH3+) and 16 (CH4+) upon TMA dosing (at t = 0 s) in Figure can be explained by the release of CH4 as a reaction product as well as by dissociation of the precursor in the QMS detector. The small increase in ion current for m/z = 89 (SF3+) is likely due to an increase in chamber pressure after TMA dosing. Subsequently, at the start of the SF6 gas flow around t = 6 s, the chamber pressure increases from 15 to 50 mTorr, leading to an increase in ion current for all m/z ratios. After ignition of the SF6 plasma (at t ≈ 12.5 s), the ion currents for F+ (m/z = 19) and SF3+ (m/z = 89) decrease slightly, attributed to consumption and dissociation of F radicals and SF6, respectively. Meanwhile, the signals for m/z = 15 (CH3+) and 16 (CH4+) go up, indicating the formation of CH4 as the reaction product. After the initial increase, ion currents for m/z = 15 and 16 decrease quickly, which suggests that CH4 is being evacuated from the reactor or dissociated in the plasma.

Figure 3

Time-resolved QMS signals for selected m/z ratios 15, 16, 19, and 89. Five AlF3 ALD cycles were averaged to improve the signal-to-noise ratio. The pressure in the reactor increases from 15 to 50 mTorr when the SF6 gas flow is started (at t ≈ 6 s), leading to an increase in ion currents.

Table 1

Relevant m/z Ratios, Their Assigned Ions, and Their (Main) Assigned Parent Moleculesa

m/z	assigned ion(s)	assigned parent species
15	CH3+	CH3F, CH4, Al(CH3)3
16	CH4+	CH4
19	F+	HF, SF6, F2, CHyF4–y
20	HF+, Ar2+	HF, Ar
26	C2H2+	C2H2, C2H4, C2H6
31	CF+	CH3F, CH2F2, CHF3, CF4
32	S+, CHF+	SF6, CH3F, CH2F2, CHF3
33	CH2F+	CH2F2, CHF3
40	Ar+	Ar
50	CF2+	CH2F2, CHF3, CF4
51	SF+, CHF2+	SF6, CH2F2, CHF3
69	CF3+	CHF3, CF4
70	SF2+, CHF3+	SF6, CHF3
89	SF3+	SF6

Most m/z ratios correspond to SF6 or CHF4– species (y ≤ 4). The assignments for CHF4– species are based on cracking patterns taken from the NIST database.[77]

Figure shows time-resolved QMS data for additional m/z ratios collected during the co-reactant subcycle. At the start of the SF6 plasma exposure, the signals for m/z ratios 16, 31, 33, 50, and 69 increase, all of which correspond to CHF4– (hydrofluorocarbon) species. See Figure S4 for the cracking patterns of the CHF4– species. In addition, the signal for m/z = 26 indicates the production of C2H2 (or possibly C2H4 or C2H6), which can be explained by the reactions of, for instance, CH4 (m/z = 16) in the plasma (in the literature, it is reported that C2H2 forms in CH4 plasmas).[78] Several other m/z ratios (18, 28, 32, 34, 35, 48, and 52) were also investigated, but their ion currents were found not to be affected by TMA dosing. Upon closer inspection of Figure , it can be seen that the temporal behavior is not the same for all of the investigated species. Signals related to CH4 and C2H2 in Figure a show the highest ion currents at the beginning of the plasma exposure (t ≈ 14 s) and a rapid decrease, while others (especially in Figure b) show the highest intensity later on. More specifically, the higher the concentration of F atoms in the species, the later the maximum intensity occurs. This suggests that the F concentration of surface species increases due to the interaction with F radicals from the plasma before they are released from the surface or that CH4 and CHF4– plasma species react with F in the plasma. The signal for m/z = 20 is also displayed in Figure a and corresponds to Ar (Ar2+ in this case) as well as HF. Between t ≈ 8 and 12 s, the ion current for m/z = 20 decreases, which can be attributed to Ar being evacuated from the chamber (Ar is used as the background gas during the TMA dose). At the start of the SF6 plasma (t ≈ 12 s), the ion current quickly rises, indicating the formation of HF, after which the ion current stabilizes (at t ≈ 17 s). The formation of HF can be explained by the recombination of H and F radicals (likely at the surface) and can lead to an additional thermal component in the reaction mechanism (see eq ). This reaction between HF and CH3 surface groups is likely responsible for the release of CH4 groups, as will be addressed in Section .

Figure 4

Time-resolved QMS signals collected during the plasma subcycle corresponding to (a) CH4, HF, and C2H2 (m/z = 16, 20, 26) and (b) hydrofluorocarbons (m/z = 31, 33, 50, 51, 69). Note that the time for which the species reach the maximum intensity correlates to the F-content and that signals corresponding to species with a higher F-content in (b) (m/z = 31, 50, and 69) reach the maximum later than H-rich species in (a). The data for m/z = 16 is also shown in Figure .

To confirm that the detected species correspond to reaction products instead of SF6 plasma species, the ion currents for the selected masses were measured during an SF6 plasma exposure of an ALD cycle and during an SF6 plasma exposure (without preceding TMA). Signals of key reaction products (m/z = 15, 16, 20, 25, 26, 30, 33, 50, and 69) in Figure are significantly higher for the ALD cycle as compared to the signals for only SF6 plasma. The higher ion currents prove that all these species are formed due to the interaction of the SF6 plasma with the precursor ligands remaining after the precursor dose or due to interaction of the SF6 plasma with formed reaction products. For instance, CH4 species can be formed in a surface reaction and subsequently undergo dissociation reactions in the plasma. Certain signals (e.g., m/z = 32, 51, and 70) simultaneously correspond to hydrofluorocarbons and SF and do not show a higher current for the ALD cycle than for the SF6 plasma without TMA dosing. Probably, the production of CHF4– species does not affect the currents for those m/z ratios due to the high background signal from SF6. It is noted that the ion currents for m/z = 15–16 (CH3+, CH4+), m/z = 25–29 (C2H+, x = 1–5), and m/z = 31, 50, and 69 (CF+) also increased slightly during the SF6 plasma exposure without preceding TMA dosing, which indicates that some C-containing species were present on the reactor wall due to previously performed ALD cycles.

Figure 5

QMS mass spectra showing selected masses during the SF6 plasma exposure of an ALD cycle and during an SF6 plasma exposure (without TMA dosing). The spectra were collected at t ≈ 13.5 s, which is approximately 1.5 s after igniting the plasma.

The excited species present in the plasma were analyzed using OES to corroborate the findings from the QMS measurements. Figure a shows OES spectra for an SF6 plasma during an ALD cycle (after TMA dosing) and an SF6 plasma (without preceding TMA dosing). Various key excited species can be identified in both spectra, including F, S, H, and H2.[79,80] Emission related to F and S can be attributed to dissociation of SF6 in the plasma, whereas H and H2 are likely related to residual background species (e.g., H2O). For the plasma during an ALD cycle, additional emission peaks corresponding to C2H2 and CF2 are detected at wavelengths of 247.8, 257.7, and 259.2 nm (see the inset of Figure a).[81−84] The OES spectrum for the ALD cycle thus corroborates the presence of CHF4– species as reaction products.

Figure 6

(a) OES spectra of an SF6 plasma during an AlF3 ALD cycle (“ALD, after TMA”) and an SF6 plasma (“no ALD”), which serves as a reference. Signals corresponding to highlighted species (i.e., C2H2, CF2, S, F, Hα, and Ar) are plotted as a function of time in (b) and (c). Inset: zoom of the range between 245 and 275 nm, indicating emission peaks corresponding to C2H2 and CF2. (b, c) Time resolved OES signals for selected wavelengths during (b) the SF6 plasma exposure in an ALD cycle, after TMA dosing, and (c) an SF6 plasma (“no ALD”). For clarity, the signals related to S (467.5 nm) and F (685.2 nm) have been scaled. Note that the SF6 gas flow was decreased from the standard value of 100 to 50 sccm in order to increase the residence time of the species in the plasma.

Intensities corresponding to relevant species were followed as a function of time during an SF6 plasma exposure of an ALD cycle (Figure b) and during the SF6 plasma exposure without TMA (Figure c). Figure b shows that the signal corresponding to F radicals (623.9 nm) increases gradually after the plasma is started. As compared to the reference signals (Figure c), there is initially less emission related to excited F species, indicating that F is consumed in the surface reactions. At the end of the plasma exposure, the emission signal at 623.9 nm approaches the same intensity as in an SF6 plasma (Figure c), which can be explained by the saturation of the surface reactions between CH3 ligands and F radicals. Emission peaks at other wavelengths related to F (e.g., 712.8, 720.2, and 733.1 nm) were found to show a similar behavior as function time, corroborating saturation. When comparing Figure b,c, it becomes clear that signals assigned to reaction products (257.7, 259.2, and 656.2 nm) demonstrate a quick rise and subsequent decrease during the ALD cycle, while they are not observed for the reference measurement of the SF6 plasma (without TMA dosing). The additional emission in Figure b can be explained by the formation of reaction products (i.e., CHF4– and HF species) due to the reaction of plasma species, most likely F radicals, with the surface groups. The gradual decrease in emission from these reaction products reveals the elimination of precursor ligands by F radicals and their removal from the reactor chamber. Note that a small amount of excited Ar is observed for the ALD cycle, which is apparently still present in the reactor after the precursor purge step but decreases soon after plasma ignition. Overall, the OES results are in line with the QMS data, indicating the consumption of F radicals and the formation of CHF4– species as reaction products.

Discussion

The insights obtained from the infrared, QMS, and OES data of Section help one to understand the reaction mechanism of AlF3 growth using Al(CH3)3 and SF6 plasma. Before the reaction mechanisms are discussed, the main results can be summarized as follows:The role of HF in the reaction mechanism requires some more discussion before the complete mechanism can be postulated. QMS measurements revealed that HF is produced as a reaction product during the SF6 plasma, which can adsorb to the AlF3 surface. The surface infrared spectroscopy data presented in Figure show the adsorption of HF surface species during the SF6 plasma exposure subcycle and the consumption of HF in the TMA subcycle (for 100 and 150 °C). Since the GPC for the plasma process is similar to the GPC for the thermal process, it is possible to compare the relative absorbance values for the two processes. For the thermal ALD process, DuMont and George also reported the consumption of HF, as indicated by a negative absorbance peak in the range of ∼3700–3000 cm–1 (corresponding to HF), which was similar in intensity to the absorbance peak in the region of ∼3000–2800 cm–1 corresponding to AlCH3 species.[47] However, when comparing the data shown in Figure with the data reported by DuMont and George, it becomes clear that the relative absorbance corresponding to adsorbed HF is much smaller for the SF6 plasma process than for the thermal process using HF as the co-reactant. Although some HF is produced during the SF6 plasma of the co-reactant subcycle, the HF exposure and resulting surface coverage is likely much lower than when dosing HF directly as the co-reactant. The smaller relative absorbance corresponding to HF therefore suggests that Al(CH3)3 also reacts directly with the AlF3 in a ligand-exchange reaction during the plasma process (forming AlF(CH3)2; see the discussion of eq ).

Al(CH3)3 dosing in the precursor subcycle results in CH3 groups on the surface, which are removed in the co-reactant subcycle.

At low temperatures (≤150 °C), adsorbed HF is consumed during the TMA dose and formed during the SF6 plasma exposure, while it does not play a role in the surface reactions at higher temperatures. The presence of HF in the gas phase was observed during the SF6 plasma exposure (at a temperature of 120 °C).

CH4 and CHF4– gas-phase species are formed during the SF6 plasma exposure, indicating the reactions of CH3 groups with plasma species and HF.

During the SF6 plasma step, F radicals are consumed in surface reactions with CH3 groups (and possibly other surface species) and lead to the fluorination of the surface.

On the basis of the results, it can be concluded that, during AlF3 ALD using Al(CH3)3 and SF6 plasma, Al(CH3)3 reacts with AlF3 as well as with HF adsorbed on the surface and that part of the CH3 ligands are released as CH4. This is illustrated in Figure and described by eqs and 3.2. Al(CH3)3 reacts with the AlF3 film in a ligand-exchange reaction, forming AlF2(CH3) or AlF(CH3)2 (eq ). In addition, at low temperatures (≤150 °C), adsorbed HF serves as a reactive site for Al(CH3)3 to bind (eq ). HF is formed during the preceding co-reactant subcycle and partly remains adsorbed on the surface. Saturation of the reactions between Al(CH3)3 and the surface likely occurs due to steric hindrance effects between CH3 groups. It is noted that in QCM measurements for thermal ALD using HF it was found that mainly AlF(CH3)2 is formed upon dosing Al(CH3)3.[45]Equations and 3.2 therefore describe the formation of AlF(CH3)2, although AlF2(CH3) surface species might probably also be present.[45]

Figure 7

Schematic diagram of the proposed reaction mechanism for AlF3 ALD when using Al(CH3)3 and SF6 plasma. The orange and blue colors indicate elements introduced during the precursor or co-reactant subcycle, respectively. In the precursor subcycle (“A”), Al(CH3)3 reacts with HF and AlF3, resulting in AlF(CH3)2 surface species. During the plasma subcycle (“B”), F radicals from the plasma remove the CH3 ligands and fluorinate the film to AlF3. HF, which is formed during the SF6 plasma exposure, can adsorb to the surface, serving as an additional reactive site at low deposition temperatures. The formed reaction products can undergo dissociation and recombination reactions in the SF6 plasma and at the reactor wall. Note that the reaction between AlF3 and Al(CH3)3 results in three AlF(CH3)3– groups, although only two are displayed. For simplicity, only AlF(CH3)2 species are drawn after TMA dosing, whereas AlF2(CH3) can also be present.

The CH3 ligands remaining after the precursor subcycle are eliminated by F radicals during the following SF6 plasma exposure, resulting in the formation of CHF4– species (see eq and Figure ). These CHF4– species can be formed either in surface reactions or in the plasma. Furthermore, HF is produced as a reaction product in eq , leading to an additional thermal component, which results in the production of CH4, as described by eq . It is was found that at low temperatures (≤150 °C) a small part of the formed HF remains adsorbed on the surface after the SF6 plasma, providing reactive sites for the next TMA dose (eq ).The combination of the reactions described by eqs , 3.2, 4.1, and 4.2 results in the following overall reaction equation for the complete AlF3 ALD cycle:In the mechanism described by eq and Figure , Al(CH3)3 reacts with both AlF3 and HF species, while F radicals play the dominant role in the ligand removal and the fluorination reaction. These mechanisms differ from the thermal ALD process on two aspects. During the thermal ALD process, Al(CH3)3 is believed to only react with HF, resulting in the formation of CH4.[45] For the plasma-based process, the main pathway is likely the ligand-exchange reaction between Al(CH3)3 and AlF3 due to the lower coverage of adsorbed HF. The second difference between the two processes is which products are formed in the co-reactant subcycle. Whereas only CH4 is released in the thermal process, the SF6 plasma process is characterized by the formation of CHF4– species due to the interaction of F radicals from the SF6 plasma with CH3 surface groups. Furthermore, the reaction products released from the surface can undergo a wide range of reactions in the plasma.

Note that some analogies can be identified between the reaction mechanism described by eq and plasma ALD of Al2O3 using Al(CH3)3, where the formation of H2O, CH4, and C2H2 species is reported.[85] During plasma ALD of Al2O3, H2O is formed due to interaction of O radicals with CH3 surface groups, which triggers a thermal component in the reaction mechanism, resulting in the formation of CH4. The formed CH4 can subsequently be dissociated in the plasma and lead to the formation of C2H2 species. Furthermore, the OH group serves as the reactive surface site, which can be seen in analogy with adsorbed HF during AlF3 ALD.

On the basis of the reaction mechanisms that were found, it can be speculated that SF6 plasma as co-reactant can be combined with metal organic precursors containing carbohydrate-based ligands (i.e., metal alkyls) for the growth of metal fluorides. Similar to AlF3 ALD, this would lead to the release of CH species in the precursor subcycle. The reaction with F radicals during the plasma exposure would result in the elimination of the remaining ligands and the formation of CHF4– species. Note that Lee et al. demonstrated ALD of MnF2, MgF2, ZnF2, and AlF3 using metal organic compounds with carbohydrate-containing ligands as the precursors and HF as the co-reactant.[46] In addition, they employed precursors containing alkylamide and alkoxide groups, suggesting that also these types of precursors might be combined with SF6 plasma.

Interestingly, the chemistry based on fluorination and TMA dosing can also be used for ALE of Al2O3 at higher substrate temperatures,[48] which was demonstrated using an SF6 plasma as the reactant in our previous work.[55] The transition between AlF3 ALD and Al2O3 ALE is governed by the substrate temperature and is believed to depend on the desorption temperature of AlF(CH3)2– species.[47] At low sample temperatures, AlF(CH3)2– species remain adsorbed to the surface, leading to growth, while at higher temperatures, they desorb from the surface; hence, etching is obtained. As described by eqs and 2.2, two steps are essential for the ALE mechanism. The top surface of Al2O3 should be fluorinated, followed by the formation of volatile AlF(CH3)2– species upon TMA dosing. The data on the fluorination of Al2O3 using SF6 plasma in Figures S1–S3 showed that an SF6 plasma converts the top surface to AlOF in a self-limiting manner, similar to when HF is used as the F-source for ALE. Furthermore, the results in Figure showed negative absorbance peaks in the region below ∼950 cm–1 at temperatures ≥200 °C, indicating the removal of Al2O3. As discussed in the Experimental Section, the actual sample temperature during the experiments on the wafer coupons deviates from the sample temperature during infrared spectroscopy experiments. In our previous work, ALE was already observed at a substrate temperature of 155 °C by employing a high TMA dose of 0.5 s.[55] This indicates that, besides the temperature, the TMA dose has an influence on the competition between deposition and etching.

Conclusions

Following up on previous work where the use of an SF6 plasma was demonstrated, the reaction mechanisms during AlF3 ALD using TMA and SF6 plasma were studied. On the basis of the infrared spectroscopy results and the information on gas-phase reaction products from QMS and OES, a pathway for AlF3 ALD was proposed where Al(CH3)3 reacts with AlF3 and HF surface species during the precursor subcycle. During the subsequent SF6 plasma exposure, F species fluorinate the surface and eliminate the CH3 surface groups, resulting in CHF4– species as well as HF. The infrared spectroscopy data showed the growth of AlF3 at low temperatures (≤150 °C) and etching of Al2O3 at high temperatures (≥200 °C), which is explained by the desorption of AlF(CH3)3– species at high temperatures. The postulated reaction mechanisms can be generalized to plasma ALD of metal fluorides using metal alkyl compounds as the precursor. It can be speculated that CH species are released upon adsorption of such precursors on the metal fluoride surface and that F radicals from the SF6 plasma exposure result in the elimination of the remaining ligands and the formation of CHF4– species. Furthermore, it is also expected that an SF6 plasma can be employed for self-limiting fluorination of a variety of metal oxides. The understanding provided in this work can thus help to develop ALD processes for other metal fluorides as well as ALE processes for metal oxides using a fluorine-containing plasma as the F-source.