Characterizing fluorocarbon assisted atomic layer etching of Si using cyclic Ar/C₄F₈ and Ar/CHF₃ plasma

Advanced semiconductor manufacturing sets high demands for etch precision and material selectivity.^1,2 Consequently, the field of atomic layer etching (ALE) has seen a great increase in interest during recent years.^3–6 Many different types of processes aiming to achieve atomic resolution are being investigated.⁷ While not all methods are true atomic layer processes, they all aim for atomic layer precision. Atomic layer deposition (ALD) is an established and widely employed process for device fabrication.^8–11 Several advances made and insights gained in the field of ALD are being exploited to drive forward the development of ALE methods.¹² For example, spatially resolved ALE, based on existing processing methods for ALD, is being investigated to negate low wafer throughput, one of the obstacles of ALE processing.¹³ 

Many ALE processes offer a larger parameter space and more process adaptability than continuous etching processes. It is therefore important to characterize the influence of process parameters and surface chemistry on ALE mechanisms. One attractive approach is a flux controlled etching process. This process offers high flexibility because of the possibility to balance reactive surface chemistry and low energy ion bombardment. Controlled, precise etching of SiO₂ and Si based on such a flux controlled, cyclic Ar plasma has been described previously.^14,15 Additionally, the process has shown promise for device patterning on a manufacturing scale.¹⁶ In this article we further characterize this process, especially with regard to process parameter impact on etch behavior, material etching selectivity, and surface chemistry. Additionally, several interesting phenomena have been observed and will be presented here.

Processing conditions employed closely match our previous work^14,15 and are briefly described here. This work was performed using an inductively coupled plasma (ICP) system. A planar, water-cooled brass coil above a quartz window was powered by a 13.56 MHz power supply with an L-type matching network. The source power, processing pressure, and Ar flow were kept constant at 200 W, 10 mTorr, and 50 SCCM, respectively. The plasma was confined within a 195 mm diameter anodized Al confinement ring. A 125 mm diameter silicon substrate is located 150 mm below the top electrode on an electrostatic chuck. Independently biasing the bottom electrode at 3.7 MHz allowed a RF self-bias potential of −5 to −15 V, creating maximum ion energies of about 20–30 eV. The base pressure achieved prior to processing was in the 5 × 10⁻⁷ Torr range. The temperature of the samples (25 × 25 mm²) was stabilized at 10 °C by substrate backside cooling during plasma processing. In order to minimize impacts of atmospheric exposure of both the sample and processing chamber, a load lock and vacuum transfer were used for all experiments. A standard O₂ plasma-based cleaning process and an Ar plasma-based conditioning process between each experiment ensured processing conditions were as comparable as possible for each experiment. Additional details of the plasma system have been described previously.^17–19 

Material removal was studied by in situ ellipsometry.²⁰ The ellipsometer was an automated rotating compensator ellipsometer working in the polarizer-compensator-sample-analyzer (PCSA) configuration at a ≈72° angle of incidence. Measurements were performed in Ψ-Δ-space, corresponding to changes in phase and relative amplitude of the polarized laser light components (He–Ne laser, λ = 632.8 nm). Optical multilayer modeling was used for interpretation of recorded data and to establish real-time thickness changes of various films. The materials studied were SiO₂–Si–SiO₂ stacks deposited on a silicon substrate by PECVD techniques with various thicknesses. Using SiO₂–Si–SiO₂ stacks allowed for precise thickness measurements as well as investigating the transition from SiO₂ to Si etching and the potential to achieve SiO₂ over Si etching selectivity.

To obtain insights on surface chemistries, X-ray photoelectron spectroscopy (XPS) has been performed at various characteristic points after the ALE process after vacuum transfer of the sample. All measurements were performed when the etch behavior was nearly identical from cycle to cycle in order to study quasi-steady-state conditions. Measurements were performed by a Vacuum Generators ESCALAB MK II surface analysis system after vacuum transfer to avoid exposure to air. Narrow scan spectra of the Si2p, C1s, O1s, and F1s binding energy regions were obtained at 20 eV pass energy at an electron take-off angle of 20° (shallow probing depth ≈20–30 Å) and 90° (deep probing depth ≈80 Å) with respect to the sample surface. Results presented and discussed here typically focus on shallow probing depths to emphasize the very surface. Spectra were surface charge compensated by calibrating the binding energy position of the Si–Si peak to 99.3 eV and fitted using a least square fit after Shirley background subtraction.^21,22 Si2p spectra were fit with peaks corresponding to Si–Si, SiF, SiF₂, SiF₃, SiC, SiO₂, and fluorinated silicon oxide (SiO_xF_y). C1s spectra were fit with peaks corresponding to C–C, SiC, C–CF_x (x = 1, 2, and 3), CF, CF₂, and CF₃. O1s spectra were fit with peaks corresponding to SiO₂ and fluorinated silicon oxide. F1s spectra were fit with peaks corresponding to SiF_x (x = 1, 2, and 3), fluorinated silicon oxide, and CF. All fittings were required to show consistency across all individual spectra, i.e., the chemical information extracted from Si2p, C1s, O1s, and F1s was internally consistent. Additional information about this analysis method can be found in previous publications.^15,22–28

The cyclic process used for the fluorocarbon (FC) based atomic layer etching (ALE) discussed here has been described in detail previously and is schematically shown in Fig. 1.^14,15 Each cycle consists of two steps, the reactant deposition step and the etch step. During the reactant deposition step a short, precise precursor injection allows for controlled, thin FC film deposition. Sufficient time is given to allow for the deposition to saturate and any precursor in the gas phase to be pumped out. Subsequently, a bias potential is applied to induce low energy Ar⁺ ion bombardment during the etch step. The two steps are repeated in a cyclic fashion enabling precise control over material removal. Etch rates are chemically enhanced as long as a chemical etchant, F in this case, is present on the surface but will decrease once it is depleted.^29–31 This leads to time-dependent etch rates, i.e., material etch rates change throughout each cycle. The instantaneous etch rate, i.e., etch rate at a given point in time, has an impact on the etch depth per cycle, i.e., amount of substrate material removal per cycle. If instantaneous etch rates are overall lower during a given cycle, the total removal per cycle will be lower. However, the etch depth per cycle is also dependent on the etch step length (ESL).

The FC film composition and thickness deposited during each cycle play an essential role in the etching of SiO₂ and Si. The film provides a limited amount of F as a chemical etchant to the surface during each cycle, and it is important to characterize the FC film deposition. The deposited film thickness per cycle based on the precursor injection, i.e., the number of precursor molecules in each pulse, is shown in Fig. 2(a). C₄F₈ has a higher deposition yield than CHF₃, leading to an easier control of very thin FC deposition by CHF₃. Lower deposition yields make resulting film thickness less dependent on small fluctuations in the number of precursor molecules injected. A shift in deposition behavior was observed when comparing deposition on the substrate and on an already existing FC film. Previous work has shown a change in FC film composition based on its thickness.¹⁴ Figure 2(a) additionally shows an initial onset for FC film deposition. A linear increase in film thickness can be observed when injecting at least 0.15 × 10¹⁹ molecules per pulse. A relatively sharp drop off in deposited thickness is seen for small precursor injections.

CHF₃ showed a deposition behavior not observed for C₄F₈. For CHF₃, the deposition yield showed a dependence on whether a SiO₂ or Si substrate was used for the FC film deposition. Figure 2(b) shows the deposited FC film thickness on Si as compared to deposited thickness on SiO₂ for several process conditions. Each data point was taken as an average over several cycles within the same experiment. The deposited thickness increases significantly for CHF₃ when depositing on Si instead of SiO₂. This increase in deposition has also been confirmed via XPS surface analysis.¹⁵ An example of the evolution of FC deposition from cycle to cycle can be seen in Fig. 3. A SiO₂–Si–SiO₂ stack was etched using CHF₃ and an ion energy of 25 eV. The first 15 cycles etched through the top SiO₂ layer, depositing about 4 Å of FC film each cycle. The SiO₂–Si interface was reached around cycle 15. The etching of the silicon interlayer is also shown and provides information on the etching depth in the SiO₂/Si/SiO₂ stack. The FC deposition can be seen to rapidly increase to ≈7.5 Å per cycle within 5 cycles. It is noteworthy that the external processing parameters, such as the precursor injection, have been kept constant throughout the whole experiment.

In situ ellipsometry allows for real-time measurement of the thickness evolution during ALE. An optical model has been established for this work to interpret measured values of Ψ and Δ in terms of changes in layer thicknesses. The model has been chosen based on a thorough consideration of various parameters and stringent testing and confirmation. Single wavelength ellipsometry is not capable of distinguishing between materials of similar optical properties.²⁰ The optical model employed therefore combines FC, SiO₂, and fluorinated substrate material in a mixed layer on top of elemental Si and is shown schematically as an inset in Figure 4. Figure 4 shows an example thickness trajectory of a single cycle of ALE of Si obtained with the methods explained above. While ALE of SiO₂ showed strongly time-dependent instantaneous etch rates, leading to a truly self-limited process, ALE of Si using the same process conditions showed a significantly reduced variation of instantaneous etch rate during a single cycle.¹⁵ One possible explanation is the lower physical sputtering threshold for Si, leading to comparable amounts of physical sputtering and thus preventing strongly time-dependent instantaneous etch rates.⁵ Additionally, it has been suggested that Si shows a significantly larger, chemically reacted, near-surface region than SiO₂.²⁴ Since etching primarily takes place in the reacted, mixed top layer, the presence of a relatively thick reacted layer can reduce the time-dependence of instantaneous Si etch rates for the etch step lengths explored here.

The impact of several processing parameters on the etching behavior is discussed here. The process parameters explored are the FC film deposition thickness per cycle, the ion energy during the etch step, and etch step length, i.e., the duration of the ion bombardment step during each cycle. The range of FC film thicknesses investigated has been limited to thicknesses of up to 12 Å. The lower part of this range enables processing under F-starved surface conditions and time-dependent instantaneous etch rates based on chemical reactant depletion. Similarly, ion energies were kept at or below 30 eV to prevent significant physical sputtering. The resulting etch step lengths leading to FC depletion are on the order of 20–60 s.

Precise FC film deposition is the source of chemical etchant during the etch step. Similar to continuous etch approaches, the FC film is actively involved in the etching mechanism of the substrate material.^21,29,32 Thicker FC film depositions provide more F reactant but also inhibit energy deposition into the substrate. It can be seen in Fig. 5 that thicker FC film depositions led to more substrate removal for SiO₂ and Si at all ion energies. The etch step length has been kept constant at 40 s for this consideration. A saturation in etch depth at thicker FC films can be observed. If the FC film deposition per cycle becomes relatively large, a large fraction of each etch step is required to remove the FC film deposited. Little additional interaction with the substrate occurs, leading to a saturation in etch depth per cycle.

The FC film removal during each cycle plays a significant role for these processes. The removal can be controlled via the ion energy but also via the time of ion bombardment in each cycle, i.e., the etch step length. One would expect longer ion bombardment times per cycle to lead to more removal. This is seen for SiO₂, whereas the opposite behavior was observed for Si. Figure 6 summarizes the etch depth per cycle for SiO₂ and Si at 25 and 30 eV for a 5 Å FC deposition per cycle. An ion energy of 30 eV leads to a faster FC removal but is still below the physical sputtering threshold of SiO₂. A saturation in etch depth per cycle with etch step length for ALE of SiO₂ at 30 eV can therefore be seen, and the substrate removal is limited by the amount of FC admitted during the deposition step. This can be generally referred to as a reactant-limited regime. Ion energies of 25 eV did not show this saturation yet due to a slower FC removal compared to 30 eV. It is expected, however, that extending the etch step length beyond the 60 s explored in this work will lead to a similar saturation.

The Si removal per cycle showed a slight decrease with increasing etch step length. In order to explain this, the instantaneous Si etch rate during the etch step has to be considered. Figure 7 shows that longer etch step lengths led to a significantly reduced instantaneous etch rate for Si. The strong reduction in instantaneous etch rate can be explained by an overall F poorer environment combined with a thicker reacted Si layer on the surface. Surface chemistry analysis showed a higher signal intensity at 104 eV in the Si2p spectra, related to the fluorinated oxide layer, for longer etch step lengths. The additional ion bombardment during each cycle can cause ion enhanced oxidation of the Si surface by very low levels of oxygen present in the ICP chamber, e.g., due to erosion of the quartz coupling window.^21,33–35 This oxide layer inhibits the etching of the underlying Si substrate, thus reducing the instantaneous etch rate.³⁶ The reduction in instantaneous etch rate offsets the longer ion bombardment time, leading to an overall comparable or even slightly decreased removal per cycle. Additionally, the change in instantaneous etch rate within one cycle, i.e., the variation from the initial to the final instantaneous Si etch rate, was increasing with increasing etch step length. This shows a strong depletion of F species, leading to time-dependent etch rates similar to those seen for SiO₂ etching. In addition, the stronger change in instantaneous etch rate within one cycle shows the overall F poorer environment, offering less chemical etchant to enhance substrate etching.

For many applications, the material etching selectivity, i.e., etching one material selectively to another, is essential. The following describes how each process parameter can control material etching selectivity for SiO₂ and Si.

It has been shown previously that choosing appropriate process conditions allows for selective etching of SiO₂ over Si.¹⁵ Etching selectivity was achieved using a FC film build-up condition, meaning process conditions were chosen so as not to entirely remove the deposited FC film from Si during each cycle, while still etching a SiO₂ substrate. Due to an increase in FC film deposition on Si, the chosen etch step conditions are not sufficient to remove the deposited FC film from Si substrates, leading to a growing FC film from cycle to cycle. Eventually, the FC is thick enough to inhibit etching of the underlying Si substrate. While this method yields technically infinite selectivity of SiO₂ over Si, there is a significant Si loss during the FC build-up phase, about 45 Å for conditions examined, and a thick FC film is left on the surface after processing. Etching selectivity can in principle also be achieved through different approaches. ALE of Si has shown an oxidized layer at the surface which can also inhibit substrate etching, somewhat similar to a thick FC film.^15,36 Additionally, intrinsic etch properties, e.g., bond breaking energies, can lead to differences in etch depth per cycle, enabling material etching selectivity.

The FC film thickness deposited per cycle has a great impact on etch depth per cycle. Since the dependence of etch depth per cycle on FC film thickness is quite similar for SiO₂ and Si, the FC film thickness deposited per cycle is not an ideal parameter to tune etching selectivity. FC film thickness deposited per cycle is instead useful to control the overall amount of substrate material removed per cycle. Strong changes in the deposited film thickness can additionally shift the process into different parameter regimes, such as F-rich versus F-starved conditions.

The etch step length (ESL) is a crucial parameter of flux controlled ALE processes. It has a great impact on instantaneous etch rates and material removal per cycle. Short etch step lengths can lead to insufficient FC film removal. Prolonged etch step lengths can cause potential surface oxidation and physical sputtering, especially for Si substrates, by extended ion bombardment. Significant differences between SiO₂ and Si are seen and described above. SiO₂ etch depth per cycle is increasing, whereas the Si etch depth per cycle is fairly constant or even slightly decreasing with increasing etch step length. This opposite behavior allows for material etching selectivity control. On the one hand, increasing etch step lengths leads to a strongly F-starved surface chemistry regime enabling selective etching of SiO₂ over Si. On the other hand, reducing etch step lengths leads to a selective Si over SiO₂ etch.

In addition to etch step length, ion bombardment energy during the etch step is useful for control of material etching selectivity. The ion bombardment energy impacts FC film defluorination and removal, energy transport to the substrate, and ion-enhanced surface oxidation.^21,29 An overview of material removal per cycle at 5 Å FC film deposition per cycle is presented in Fig. 8. A strong increase in removal per cycle with ion energy can be seen for SiO₂. Si removal does also increase but less than SiO₂. A possible explanation for this is based on ion-enhanced surface oxidation. Higher ion energies can lead to a stronger surface oxidation of Si. The oxidized surface layer then, in turn, inhibits substrate etching.³⁶ Therefore, higher ion energies show selective SiO₂ over Si etching, while lower ion energies show selective Si over SiO₂ etching. The same process can be tuned for selectively etching either material. However, it should be kept in mind that other aspects, e.g., SiC material at the surface, can also contribute to the ion energy dependence.

In continuous etch processes the etch yield is defined as the number of atoms removed per incident ion per unit time. Due to the strongly time-dependent conditions of this cyclic ALE process, the etch yield is averaged over a cycle and defined as the number of atoms removed per cycle divided by the number of incident ions per cycle. The time-averaged amount of FC available per incident ion (F/Ar⁺) can be calculated by the FC thickness deposited per cycle divided by the number of incident ions per cycle and provides a measure of F availability. A high F/Ar⁺ ratio describes a F-rich condition, i.e., high amounts of F are available per incident ion, while a low F/Ar⁺ ratio describes F-starved conditions. The etch yield dependence on the F/Ar⁺ ratio for ALE of SiO₂ and Si is shown in Fig. 9 for 25 eV and 30 eV ion energy. The etch step length ranged from 20 s to 60 s, and FC thickness deposited per cycle ranged from 3 Å to 11 Å. ALE of Si shows a stronger increase in etch yield than SiO₂. Additionally, the etch yield saturates at higher values for ALE of Si compared to SiO₂. This suggests that ALE of Si is depending stronger on the availability of F than SiO₂ while the limiting factor for etching of SiO₂ is the ion bombardment. The physical sputtering yield for SiO₂ is lower than for Si,⁵ suggesting that more ion bombardment is required to etch SiO₂ substrates. This agrees with the observation above that ALE of SiO₂ is limited by the ion bombardment, while ALE of Si does not require the same ion bombardment and is therefore limited by the availability of chemical etchant, i.e., F.

This behavior is quite different from continuous etch approaches due to very different surface chemistries and etching conditions. Continuous FC-based etching of Si-based materials shows a decrease in etch yield with increasing steady-state FC film thickness.^23,35,37 Additionally, the etch yield presented here is averaged over a full ALE cycle and significantly lower due to the comparably F-starved condition. Furthermore, a lot of previous ALE of Si work focused on a Cl/Ar chemistry. The Si-Cl-system showed a self-limited behavior with similar trends and phenomena regarding the cycle averaged etch yield dependent on the Cl/Ar⁺ ratio.^38–41 

The available chemical etchant is based on the deposited FC film, and changes in the properties of the various surface layers will strongly influence the etch behavior. X-ray photoelectron spectroscopy (XPS) was used to characterize the typical surface chemistry at crucial points of the ALE cycle for various conditions of Si etching. Surface chemistry during ALE of SiO₂ has been discussed previously and will therefore not be discussed again here.¹⁴ High resolution spectra of the Si2p, C1s, O1s, and F1s core levels were taken and decomposed based on a thorough analysis. Owing to the complexity of these surfaces, all results were checked for agreement across all spectra for a given surface to provide an overall consistent depiction of the surface chemistry. Exemplary decompositions of each spectrum are shown in Figs. 10–13.

The Si2p spectra (examples shown in Fig. 10) are composed of a large variety of moieties. The two main peaks observed originated from elemental Si, at 99.3 eV, and oxidized Si, around 104 eV. Fluorination of the oxide slightly shifts the binding energy to higher levels.²² Fluorinated Si intensities are located between these two main peaks, around a binding energy of 100–103 eV. Generally, the signal of these fluorinated Si species was low in comparison to the two main peaks.

The deposited FC film is primarily characterized by the C1s spectral signatures. Figure 11 shows an exemplary decomposition of the C1s spectrum after deposition of an 11 Å thick FC film and after completion of the etch step. The high binding energy region in the C1s spectrum is attributed to CF_x (x = 1, 2, and 3) species. The more F atoms a C atom is bonded to, the larger the chemical shift from the C–C binding energy.^21,23,24 In addition to bonding to F, bonding to O causes a very similar shift in the C binding energy. It is therefore difficult to distinguish between CF_x and C–O species solely by the C1s spectra. As will be noted below, a cross-reference to the O1s and F1s spectra is established to account for C–O species. The F/C ratio is a common characteristic used for FC films and can be calculated using the decomposed C1s spectra.^23,24 Essentially, the CF_x species intensity is calculated relative to the total C1s intensity. A strong removal in CF_x species and, therefore, reduction in the F/C ratio, were typically seen during ion bombardment, consistent with the ellipsometric measurements.

Typical O1s spectra are shown in Fig. 12. The main signal intensity originated from O bound to Si, in the form of SiO₂ and fluorinated oxide. Additionally, a peak at lower binding energy was observed and attributed to C–O species, and related to the C–O intensities seen in the C1s spectra, as described above. Additionally, O1s intensities were compared to oxide intensities in the Si2p spectra. This relationship provides a measure of Si oxidation.

Figure 13 shows typical decompositions of an F1s spectrum after FC film deposition and after completion of the etch step. The signal was a combination of F associated with either the FC film or the substrate. The C1s spectra were used to determine the intensity of CF_x species and cross-referenced to the F1s spectra. Furthermore, F1s intensities were compared to fluorinated Si intensities in the Si2p spectra, providing a measure of substrate fluorination. Comparing the amount of F relative to C with the F/C ratio of the FC films allows estimation of the amount of F located in the substrate.²³ To this end, we calculated the F/C ratio of the FC films based on the C1s spectra and the ratio of F1s/C1s intensity. If both ratios are similar, the F is primarily located in the FC film. If the F1s/C1s ratio is significantly greater than the F/C ratio, significant amounts of F must be located in the substrate. The difference ΔF/C, i.e., F1s/C1s – F/C, is therefore an indication of substrate fluorination.

A strong difference in surface chemistry evolution was observed between 5 Å and 11 Å FC film depositions per cycle. Figure 14(a) shows that the F/C ratio of thin FC films is decreasing rapidly upon ion bombardment, and that the film is being defluorinated. At the same time ΔF/C is increasing, i.e., relatively more of the F becomes associated with the Si substrate than the FC film. Therefore, F is preferentially removed from the FC film and mixing with the Si substrate takes place. It has been shown that F can be “recycled” in an advancing reacted layer, i.e., the F is propagating through a reacted layer, rather than being removed from the substrate.³² Additionally, there could be a chamber wall interaction. FC material on the chamber walls from prior cycles can be an additional source of F during the etch step. At the end of the etch step, the FC film deposition introduces F in the form of a FC film, therefore reducing ΔF/C. The evolution of the F1s and CF_x species intensity confirms the behavior of the F/C ratio and ΔF/C described above and is shown in Fig. 14(b). For 5 Å FC film depositions per cycle the CF_x species, as measured in the C1s, decrease rapidly to a stable low level, but the F1s intensity does not follow this behavior.

Thicker FC film depositions of 11 Å, however, show a different behavior. The F/C ratio and ΔF/C are shown in Fig. 15(a). The F/C ratio of the FC films decreases more slowly and over the entire etch step. ΔF/C is stable and close to 0 throughout the entire cycle. Additionally, the F1s and CF_x intensities are very similar, as seen in Fig. 15(b). The substrate is therefore in a fairly steady state throughout the etching. The FC film is being removed from the surface, together with substrate material, while a mixed, reacted substrate layer is propagating further down. The steady state of the substrate is additionally reflected in the observation of minimal changes in the Si2p and O1s spectra throughout one cycle for FC film depositions of 11 Å.

Thicker FC depositions of 11 Å show a relatively lower intensity of reacted Si in the Si2p spectrum than thinner FC depositions of 5 Å. Thicker films inhibit the ion bombardment onto the Si surface, thus reducing ion-enhanced surface oxidation. It has been shown before that a thicker FC film during continuous etching leads to a thinner fluorinated Si reaction layer underneath.²³ The relative amount of reacted Si in the Si2p spectra is fairly stable throughout the cycle.

In order to confirm the XPS surface analysis, SIMS measurements have been performed with samples processed for a 5 Å FC film deposition per cycle and 25 eV ion energy for 40 s condition. Good agreement between results obtained by XPS and SIMS was observed. The very surface layer is a mix of C, F, O, and Si species. This mixed layer extends up to 15 Å into the sample. F and O species are predominately located at the surface with a decreasing concentration up to 25 Å into the sample.

In addition to C₄F₈, CHF₃ has also been employed as a precursor. These two precursors have shown several significant differences in continuous etching and are therefore of interest to compare when used for the ALE approach.^23,29 C₄F₈ and CHF₃ show significantly different steady-state FC films during continuous etching, based on monomer dissociation and H scavenging F from the FC film.⁴² The presence of H in CHF₃ can have a strong impact on etching behavior. H species can penetrate deep into and bond to the Si substrate.^43,44 Enhanced FC deposition has been observed for H₂ addition to FC based Si etching.⁴² However, since C₄F₈ and CHF₃ show many intrinsic differences in addition to the H, further work is required to precisely determine the specific impact of H on the ALE process.

The dependence of the SiO₂ and Si thickness removed per cycle on FC film thickness deposited per cycle is shown in Fig. 16 for CHF₃. Similar to C₄F₈, an increase in etched thickness per cycle with increasing FC film deposition is seen, along with a saturation for thick deposited FC films. Higher ion energies of 30 eV show a very strong dependence on the FC film thickness with a rapidly increasing etch depth for thin FC films. The etch depth per cycle shows a similar dependence on ion energy for CHF₃ and C₄F₈, as described above and shown in Fig. 8.

When considering material selectivity for CHF₃, it is important to keep the change in deposition yield discussed above in mind. FC film deposition using CHF₃ showed an increased deposited thickness on Si compared to SiO₂ substrates. Therefore, selective etching of SiO₂ over Si can be achieved via a FC build-up regime, as described previously. If process conditions not showing a FC build-up are chosen, the etch depth per cycle will be higher for Si compared to SiO₂ based on the thicker FC film deposition per cycle.

CHF₃ behaves similarly in terms of surface chemistry, shown in Fig. 17, when compared with C₄F₈. Thicker films show a fairly steady state substrate condition, while thinner films show a change in substrate condition throughout the ALE cycle. The F/C ratio of films deposited by CHF₃ has been shown to be slightly higher compared to C₄F₈ during this ALE process.¹⁵ This results in a slightly larger substrate removal per cycle for CHF₃, based on additional available F.³⁵ The F/C ratio is not only dependent on the precursor but also on the thickness of the deposited films at these low thicknesses. Figure 17 shows that 11 Å thick FC films have a F/C ratio of 0.6 while 5 Å thick films show a F/C ratio of only 0.25. The evolution of the F/C ratio as well as ΔF/C is similar for CHF₃ and C₄F₈. However, it is noteworthy that ΔF/C is significantly larger for 5 Å FC film depositions per cycle using CHF₃ compared to using C₄F₈, potentially due to H interactions. The oxidized top layer present during ALE of Si depends more strongly on the FC film thickness than on the choice of precursor. For instance, for both C₄F₈ and CHF₃ thinner FC film depositions of 5 Å show a stronger oxidized layer than thicker FC film depositions of 11 Å.

In situ ellipsometry revealed a very interesting phenomenon during FC film deposition on Si substrates. A representative individual cycle of ALE of Si is shown in Fig. 18. To highlight this phenomenon, a 10 s delay was introduced between each cycle, i.e., after the etch step. No bias potential was applied during this time, but the Ar plasma was sustained. The thickness trajectory shows that Si removal stops once the bias potential is no longer applied, and the Si thickness is constant. Once the precursor is injected into the plasma, marked with a dashed line at 10 s, a decrease in Si thickness is observed, along with FC film deposition. The initially faster Si reaction rate is seen to decrease with time during the deposition step. This reduction in elemental Si thickness can be based on several phenomena. One possibility is spontaneous etching of Si based on an interaction with F and/or H.⁴⁵ Another possibility is the reaction of Si forming fluorinated silicon or silicon carbide. This reaction of Si causes a change in optical properties, which is detected as an apparent loss of elemental Si in ellipsometry. XPS surface analysis does not show evidence of a change in substrate fluorination upon FC film deposition; however, the magnitude of these effects is small, i.e., on the order of 1 Å or less, and it may be difficult to detect these changes after the ALE process. Due to the limitations of single wavelength ellipsometry, it was not possible to determine whether such a reaction takes places during FC film deposition on SiO₂ substrates.

Figure 19 shows the FC film deposition and amount of reacted Si per cycle for ALE of a SiO₂–Si–SiO₂ stack. The reaction of Si during the deposition steps appears to be connected to the change in deposited film thickness, as both are increasing during the transition from SiO₂ to Si. The amount of Si reacted, i.e., via spontaneous etch and/or formation of fluorinated silicon and silicon carbide, based on the processing conditions is shown in Fig. 20. The following trends can be observed. Thicker FC film depositions, i.e., stronger precursor injections, lead to a stronger Si reaction. CHF₃ causes a stronger Si reaction than C₄F₈. Longer etch step lengths as well as higher ion energies reduce the amount of Si reacted during the deposition step. This holds true for both precursors, C₄F₈ and CHF₃, as well as thinner and thicker (5 and 10 Å) FC film depositions. Therefore, generally speaking, a F-richer processing environment leads to a stronger Si reaction, i.e., a larger change in elemental Si thickness. It is noteworthy that the etch step conditions strongly impact this Si reaction, which is occurring during the deposition step. This shows the importance of surface chemistry and a form of history effect of the substrate.

The FC based ALE process is schematically shown in Fig. 21(a) for a thin FC deposition on the order of 5 Å and in Fig. 21(b) for a thicker FC deposition on the order of 10 Å. It has to be noted that the surface is not comprised of clearly separate layers as depicted in this schematic, but, especially for thin FC films, rather a mix of several moieties. Surface analysis via XPS and SIMS has shown that the various species are intermixed. The main interactions involved during the deposition step are the FC deposition on the surface and a reaction of the Si substrate with F and/or H. Interactions during the etch step include reaction/intermixing of F with the substrate and the removal of FC film and mixed, reacted substrate material. The process was controlled via the composition and thickness of the deposited FC film, as well as the time and energy of ion bombardment during the etch step.

The etch mechanism for FC based etching of Si materials, e.g., Si, SiO₂, Si₃N₄, has been intensely investigated.^30,46–49 During typical continuous etching approaches, a constant FC flux is given, leading to a balance of FC film deposition and etching. This results in a steady-state FC film during etching, which heavily influences the etch rate and mechanism.^21,24,50 A key difference between this cyclic approach and continuous etching is the limitation of FC and the temporal separation of FC film deposition and etching, in addition to lower ion energies. The FC film is never in a steady-state condition, although similar from cycle to cycle, but changing within each cycle. These continuous changes lead to several key differences in the mechanism and characteristics of cyclic ALE.

In continuous etch approaches, the steady-state FC film thickness has proven to be of great importance.^23,51–53 Thick films can inhibit etching of the underlying substrate, similar to the material selectivity considerations described above. Interaction with the substrate, e.g., O from a SiO₂ substrate, can heavily impact the steady-state FC film thickness and therefore enhance material selectivity. Since cyclic ALE processes do not have a steady-state FC film, the mechanisms of etching are different. FC film thickness deposited per cycle yields good control over material removal per cycle for ALE. Since the FC films are removed during each etch step, the deposited film thickness does not provide good control over selectivity, except for FC build-up conditions, which were not investigated in detail here. A saturation of etch depth per cycle with deposited FC film thickness is seen. This is reminiscent of the etch-inhibiting properties of thick steady-state FC films in continuous etch approaches. Thick films inhibit ion bombardment of and reactant transport to the underlying substrate and require a substantial energy dose to maintain a steady-state thickness, thus taking away energy otherwise delivered to the substrate.^21,23 The surface characterization results are consistent with the overall idea that the FC film is strongly involved in the etch mechanism of the substrate material.⁵⁰ F from the FC film interacts with Si bonds as a result of ion bombardment, and a mixed layer is formed, while the FC film is simultaneously defluorinated.^32,54–56

The process parameters were changed over a fairly small range in this work, e.g., the FC deposition thickness per cycle ranged from 3 to 11 Å, and the ion energy ranged from 20 to 30 eV. However, a significant impact on etch behavior and etch depth per cycle was observed. This suggests the high sensitivity of ALE and the potential impact of even small levels of contamination. As discussed above, even a small amount of oxygen from the quartz window can lead to surface oxidation. In addition, FC films on the chamber walls can be a source of chemical etchant and affect process stability.⁵⁷ This shows that very high control over process and equipment is essential for the success of ALE.

Throughout this work, ALE of SiO₂ was generally easier to control than ALE of Si. This general difference can be attributed to the high reactivity of Si surfaces, in contrast to relatively unreactive SiO₂ surfaces. Since flux-controlled ALE is heavily dependent on surface chemistry, a highly reactive surface will be significantly more sensitive to any form of process fluctuation or impurity compared to an unreactive surface. Additionally, a fluorinated, oxidized surface layer is present during ALE of Si. This layer was shown to greatly impact the etching behavior of Si substrates.

Overall, the FC film deposition per cycle provides strong control of material removal, while the etch step length and ion energy provide strong control of material selectivity. Thicker FC depositions lead to more material removal, based on a higher F availability for chemically enhanced etching. The material removal per cycle saturates with deposited FC thickness. Higher ion energies and longer etch steps provide SiO₂ over Si selectivity, while lower ion energies and shorter etch steps provide Si over SiO₂ selectivity. For Si etching, with a lower physical sputtering energy threshold than SiO₂, the additional ion bombardment seen for extended etch step lengths causes physical sputtering and potential oxidation of the surface. It should be noted that a FC build-up regime needs to be considered separately.

XPS analysis has shown that for FC depositions of 11 Å per cycle, surface chemistry changes during one cycle mostly occur in the FC film. The substrate chemistry is considerably stable throughout one cycle, resulting in a quasi-steady-state. For FC depositions of 5 Å per cycle, however, ΔF/C is increasing during one cycle, indicating that F is preferentially removed from the FC film. Defluorination of the FC film can be seen in the strong reduction of the F/C ratio. Thinner FC depositions of 5 Å show a relatively higher intensity of reacted Si in the Si2p spectrum than thicker FC depositions of 11 Å.

A reaction of Si is seen during the deposition step. This reaction occurs upon injection of the precursor and is believed to be based on a chemical reaction with the fluorinated substrate. Higher ion energies during the etch step and longer etch steps reduce the amount of Si reacted upon injection.

CHF₃ has an overall similar etch behavior to C₄F₈. The deposition yield for CHF₃ is lower than C₄F₈ and does depend on substrate material. The Si reaction seen during FC film deposition is stronger for CHF₃ and believed to be related to the change in deposition yield.

We thank Andrew Knoll, Dr. Nick Fox-Lyon, Adam Pranda, Pingshan Luan, and Dr. Elliot Bartis for collaboration and helpful insights and discussion on this project. We thank Dr. Eric Hudson, Dr. Steven Lai, Dr. Michal Danek, and Dr. Alexander Dulking from Lam Research for helpful discussions. We thank Michael Saccomanno and Marinus Hopstaken for help with the SIMS results. We gratefully acknowledge financial support of this work by the National Science Foundation under Award No. CBET-1134273 and US Department of Energy (Grant No. DE-SC0001939).