A two-step plasma-thermal atomic layer etching (ALE) process that is capable of etching Ni with high selectivity with respect to the SiO2 hard mask and high anisotropy is evaluated in this work with a reactive ion etching (RIE) process to highlight the contrast between these two processes and the advantages of combining these two processes to tailor the sidewall profile with greater processing efficiency. The RIE chemistry leveraged the enhanced volatility of chlorinated nickel in the presence of hydrogen atoms. The hybrid RIE and ALE process achieved the desired sidewall profile, with no measurable residual halogen.
The continuous device miniaturization in the semiconductor industry calls for patterning of an increasing variety of materials on the nanometer level. Atomic layer etching (ALE) is under rapid development for atomic level precision patterning.1 ALE has been developed for a number of materials such as semiconductors,2 dielectrics,3 polymers,4 metals,5 metal oxides,6 and metal nitrides.7 One main concern of many reported ALE chemistries, however, is that the atomistic-level control of material removal that ALE provides comes at the cost of prolonged processing time. In an integrated process for mass production, low throughput steps could bottleneck the entire process and should be circumvented if possible. The trade-off between throughput and precision would depend on the specific application scenarios as the target amount of material removal is largely dictated by the preceding deposition step.
As the industry embraces the extreme ultraviolet (EUV) lithography technology, which relies on reflection optics rather than transmission optics for photolithography, the patterning of the absorbers on EUV masks (based on Si/Mo multilayers) becomes a real challenge. A thick layer of absorbers (e.g., 60 nm Ta) was used to ensure the complete absorption of EUV in designated regions on the mask. However, a thick absorber layer can create a mask shadowing effect as shown in Fig. 1(a). With the EUV light source directed to the surface at an angle of θ off normal, the shadow created by the height of the absorber resulted in a change in the critical dimension (CD): CDactual = CDdesigned ± 2dMtan θ, where d is the absorbing layer thickness, M is the EUV scanner reduction factor, and θ ≈ 6° for the current generation EUV lithography system. The plus-minus sign corresponds to the CD change for the patterned lines and spaces, respectively. Assuming that all other optical properties remain comparable, reducing the absorbing layer thickness generally improves the fidelity of the mask absorber pattern. However, a thinner absorbing layer would require a metal that has an index of refraction (n) close to 1 and a large extinction coefficient (k). It has been measured experimentally8 and simulated computationally9 that Ni is such a candidate material where a much thinner Ni (∼30 nm) is as effective as the state-of-the-art absorber layers such as Cr- and Ta-based materials, commonly ranging from 50 to 70 nm,10 as shown in Fig. 1(b). The amount of deviation would be halved with the replacement of Ta by Ni (highlighted in thicker line). The improvement is especially noticeable for extreme dimension features, where the deviation amount could be non-negligible to the actual feature sizes.
The main challenge in implementing the Ni EUV absorber is in its patterning. If an ALE process was used to pattern a 30 nm thick Ni film, a rather long process time would be required. For example, based on Paper I,11 using laboratory scale reactors with no specific processing optimization, a processing time of 8 h would be required. It is, therefore, a motivation to develop a process where the majority of the nickel can be removed quickly, while the final remaining layer is removed by ALE that simultaneously tailors the sidewall profile.
Conventional ion beam etching by noble ions or reactive ion etching (RIE) processes using halogen chemistries have been used to etch Ni. Both have shown some effectiveness, but the former resulted in sidewall redeposition and the latter suffers from both sidewall redeposition and corrosion concerns due to residual halogen. Most reported halogen-based plasma etching techniques of metals12 and other materials13–15 leverage the halide compound's vapor pressure,16 which is reasonable but not high. To completely remove the halide compounds from the surface, inert ions such as Ar are commonly used to sputter the surface. This approach, while generally applicable to a wide selection of materials, suffers low selectivity. In order to develop an effective RIE process to remove the bulk of Ni thin films while alleviating the concern of residual halogen, a cyclic RIE process is proposed in this work as shown in Fig. 2. Specifically, a chlorine plasma is used to facilitate the formation of chlorinated nickel, while a hydrogen plasma was used subsequently to remove the chlorinated nickel. Note that the sidewall profile from this cyclic RIE process is not expected to be vertical due to possible redeposition of etch products. The imperfect sidewall profile is then tailored by a plasma-thermal ALE process to achieve the desirable angle.
A two-step plasma-thermal ALE of Ni has been detailed in Paper I,11 which shows promising selectivity and anisotropy. This plasma-thermal ALE process utilizes one plasma process and one thermal process and distinguishes itself from other established and richly discussed approaches in the field, namely, plasma ALE, where both activation and removal steps are plasma-based, and thermal ALE, where both activation and removal steps are thermally activated. Plasma ALE is considered directional, but the removal step, usually Ar-based, is an energy limited process but not a self-limiting reaction and can induce mask damage. Thermal ALE only targets the modified region but is omnidirectional due to the isotropic chemical reactions. This plasma-thermal approach is novel in that a plasma-driven oxidation step introduces an anisotropic conversion of the surface layer. The isotropic component of the process comes from the removal step utilizing an isotropic organic vapor—in this step, the organic vapor (intrinsically isotropic) leaves behind an anisotropic profile, because the organic vapor only reacts with the nickel oxide but not the nickel. In other words, the anisotropy introduced by the oxidation is only revealed in the subsequent isotropic chemical vapor exposure. A number of published works suggested the realization of various degrees of anisotropy using the alternative plasma exposure and thermochemical reactions, but no microscopic image of the patterned final features was shown,6,17–19 especially not with detailed cross-sectional images of etched metal with nanoscale patterns.
Another important point to note is related to the concept of the ALE window.20 In a plasma ALE process, where ion-based removal of the modified layer is in effect and the ion energy is chosen to be high enough to remove the activated/modified layer but not the bulk material. The region for energies below the ideal ALE window is characterized by ion energies that are insufficient to completely remove the modified surface layer. A third process regime designates a region where the ion energy is high enough to remove the bulk material. In this work, the oxidation process has a dependence on oxygen ion energy, which is controlled by the plasma potential and applied bias in the oxidation half-cycle. Within the energy range studied, a plateau of the modified depth was not observed. However, the self-limiting nature of formic acid reacting with only the metal oxides guarantees a complete removal of the modified layer, which is not an energy limited process.
In this proposed hybrid process, leveraging well characterized etch rates, a non-self-limiting RIE process is used to remove the majority of the material, within 1–3 nm of the etch end point, achieving a less-desirable sidewall but at a higher efficiency. A slower and more precise ALE process is then used to trim the final feature with greater precision to the vertical angle.
To determine the reaction chemistry in the RIE process, the reaction between Ni and chlorine was first assessed by thermodynamics. The possible reactions with a negative Gibbs free energy are summarized in Table I. Indeed, chlorine would react with both metallic and chlorinated Ni to form gas phase nickel chloride.21 Due to the low vapor pressure of NiCl2 (99.1 kPa at 25 °C), it is expected that some NiClx would redeposit on the surfaces, especially at a sidewall of the feature, due to its limited volatility. The presence of hydrogen, especially hydrogen atoms, has been shown to enhance the removal of metal chlorides, such as CoClx.22 This effect is evaluated by considering the reactions of hydrogen with NiClx, where potential reactions with a negative Gibbs free energy is summarized in Table I. The driving force of the reactions in this ternary system is the propensity of H atoms to form NiH(g). A stable by-product such as HCl(g) also contributes to the Gibbs free energy of a reaction being negative, thereby generating NiH(g) with a higher partial pressure.
|Reaction .||ΔG (kJ/mol) .|
|Ni(c) + Cl2(g) → NiCl2(c)||−258.8|
|Ni(c) + Cl2(g) → NiCl2(g)||−86.4|
|NiCl2(c) + 3H(g) → NiH(g) + 2HCl(g)||−190.9|
|Reaction .||ΔG (kJ/mol) .|
|Ni(c) + Cl2(g) → NiCl2(c)||−258.8|
|Ni(c) + Cl2(g) → NiCl2(g)||−86.4|
|NiCl2(c) + 3H(g) → NiH(g) + 2HCl(g)||−190.9|
A blanket 40 nm Ni thin film on an Si wafer was deposited via sputtering. One patterned sample was prepared to evaluate the etching anisotropy and profile control: a 30 nm thick Ni film has a hard mask pattern formed by SiO2 lines with an initial sidewall angle of 84° (SiO2 thickness = 50 nm, linewidth = 65 nm, and line pitch = 200 nm). The Ni layer was deposited over a 3-nm Ta adhesion layer on a 30 nm SiO2 layer. These samples were cleaved into square pieces with dimensions of 1 × 1 cm2 for processing.
For the RIE process, samples were placed on a carrier wafer and treated in a commercial etcher (Unaxis 770). A chlorine plasma was first generated at an RF power of 400 W and an applied bias of 20 W, for a duration of 20 s. A 90 s nitrogen purge followed before the introduction of hydrogen. An hydrogen plasma was generated at an RF power of 200 W and an applied bias of 60 W for 15 s for stabilization, followed by an increase of the RF power to 400 W for 90 s. This cyclic process could be repeated as needed. Optical emission spectroscopy (OES) is used in tandem with optical emission database tabulated by NIST for identifying reaction products during the plasma surface reactions.
For the ALE process, specifics of the process conditions are detailed in Paper I.11 In this work, the oxidation step was performed only by plasma processing, as thermal oxidation was not feasible due to hardware limitation. To characterize the etch rate, surface composition, and the etched profile sidewall angle, scanning electron microscopy (SEM), X-Ray photoelectron spectroscopy (XPS), energy dispersive X-Ray spectroscopy (EDS), and high-resolution transmission electron microscopy (HRTEM) were utilized, and the experimental details are the same as detailed in Paper I.11 To quantify the sidewall profile, an average sidewall angle is reported by performing a linear fit to the digitized HRTEM sidewall contour in eight equally partitioned sections of the entire nickel film, which translated to 3.75 nm thick sections.
III. RESULTS AND DISCUSSION
A. Reactive ion etch of nickel
Using the proposed cyclic process with chlorine and hydrogen plasmas, an etch rate of 3.8 ± 0.5 nm/cycle is extrapolated from Fig. 3, where the etched Ni thickness is shown as a function of the cycle number. Comparing with the etch rate of 3 nm/cycle (2 min of oxygen plasma with 500 W RF power and 0 W applied bias and 1 h of 350 Torr formic acid exposure at 80 °C) reported in Paper I,11 etch rate per cycle of the RIE chemistry is ∼1.5 times higher. Considering that each cycle is about 70 min for ALE and about 5 min for RIE, reported RIE chemistry is about 20 times more efficient in etching Ni comparing to the reported ALE chemistry.
OES was performed to quantify the reactants (ionic and atomic chlorine and hydrogen) as well as to detect the reaction products. The data shown in Fig. 4 included the OES scan during the hydrogen plasma, immediately following a chlorine plasma step. A baseline for comparison is an identical process where there was no preceding chlorine plasma. A major difference between these two spectra was at around 340 and 415 nm. NIST database of atomic spectra22 listed peaks at 341.57 and 413.39 nm are characteristics of singly charged Ni and Cl atoms, respectively. Though OES was not able to directly detect the presence of NiClx, the data nonetheless confirmed the removal of nickel and chlorine from the surface. These results are in line with previous work done on other transition metals, where an enhanced removal of metal-chloride from the surface was observed with the addition of hydrogen atoms.21
Surface composition analysis by XPS also confirmed the formation and removal NiClx, as shown in Fig. 5, where patterned Ni samples were etched by (a) only a chlorine plasma, (b) one cycle of chlorine followed by hydrogen plasma, and (c) six cycles of chlorine followed by hydrogen plasma. The Ni 2p, Cl 2p, and Ta 4f spectra are deconvoluted to show the chemical bonding configurations on the etched surface. Ta was deposited underneath Ni as the adhesion layer. After the hydrogen half-cycle, signal intensity of the metallic Ni peak at 852.7 eV increased from 3% to 15% and signal intensity of the Ni-Cl peak at 854.4 eV decreased from 5% to 0%, again confirming the etching effectiveness of hydrogen chemistry. Repeating the treatment for six cycles resulted in a decrease in signal intensity of both Ni and Cl peaks. A maximum remaining Ni layer thickness of 7 nm could be deduced from the final spectrum as previously undetectable Ta signal is measured.
Figure 6(a) shows the schematic diagram of the patterned Ni sample. The TEM image for a feature from a sample after six cycles of the aforementioned RIE treatment is shown in Fig. 6(b), while Figs. 6(c)–6(g) showed the corresponding EDS elemental maps of Ni, Si, O, Cl, and Ta to allow for the inspection of the spatial distribution of each element. 30 nm of exposing Ni is completely removed in the vertical direction in the exposed regions; at the same time, 20 nm of the SiO2 hard mask is etched by the chlorine and hydrogen plasmas. An etch selectivity of 1.5 is thus calculated for Ni over SiO2. Sidewall angles of 45° and 70° are measured at the top and bottom of the feature, respectively. The RIE step resulted in a trapezoidlike feature. The convex curvature near the top of the feature indicates redeposition of the reaction products during the cyclic etching by chlorine and hydrogen plasmas. The EDS scan of Ni confirmed this and also showed Ni deposition on the sidewall of the SiO2 hard mask. The angle discontinuity at the SiO2-Ni interface implies that SiClx is more volatile compared to NiClx, which is consistent with the measured vapor pressure of 167.2 kPa for SiCl2 at 25 °C.23 This RIE-etched profile, though not directional, serves as an ideal starting profile to assess the effect of the plasma-thermal ALE process, as detailed in Paper I.11 It is worth mentioning that six cycles of the aforementioned treatment took about 30 min, with about half of the time used for residual gas pumping and nitrogen purging, leaving considerable room for throughput improvements.
B. Hybrid RIE-ALE of nickel
The two-step plasma-thermal process in Paper I11 showed promising selectivity and anisotropy. When compared to previously published work on patterning Co,24 the results are different since plasma oxidation as zero applied bias led to an isotropic etching profile in Co but an anisotropic etching profile in Ni. This difference is likely due to the different reactivity of Co and Ni with the oxidants (oxygen ions and atoms) in different plasma reactors and with the formic acid at different pressures.
This section discusses the final feature profile by combining high-throughput RIE with high-precision ALE. RIE-etched samples were then processed by ALE cycles that are based on 1-min plasma oxidation and 30-min formic acid vapor exposure at 550 Torr. To capture the trend of profile evolution, TEM image of features after six cycles of RIE, six cycles of RIE and one cycle of ALE, and six cycles of RIE and two cycles of ALE were taken and shown in Figs. 7(a)–7(c). Comparing images of samples before and after one cycle of ALE treatment, it is observed that the remaining Ni from RIE in the exposed regions were mostly removed, leaving footing at the edge of the feature. Another ALE cycle effectively removed the footing and left with a vertical sidewall of 90° measured at the bottom of the feature as shown in Fig. 7(c).
The trend of feature width and sidewall angle after each processing step is plotted in Fig. 8 with targeted values highlighted by the shading (≈65 nm initial hard mask width, ≥87° sidewall angle for anisotropy). The decrease in the sidewall angle at the bottom of the feature after one cycle of ALE is due to the thickness of the convex shaped sidewall reaching maximum at those regions, as the oxygen plasma did not convert the entire thickness to oxide, leaving metallic layer underneath, which was then not removed. The final sidewall angle returned to vertical as expected, once the footing was oxidized and removed. It is noted that the expansion of the Ta adhesion layer during oxidation half-cycles covered some of the Ni footing, making part of the regions inaccessible for formic acid. It is expected that more complete removal could be achieved without the adhesion layer or with an adhesion layer less prone to expansion during oxidation.
Feature outlines for three TEM images were digitized and superimposed in Fig. 9. It is noted that a vertical etch rate of 20 nm/cycle is measured, which is considerably higher than the 6 nm/cycle upper bound recorded in Paper I.11 This increase could be attributed to the effect of RIE process preceding the ALE process: the characteristic concave shape after the RIE process indicates redeposition of reaction products due to the low volatility of the nickel halides. This concave sidewall shape is due to RIE processing because patterned samples studied in Paper I11 by ALE processing alone yielded different sidewall profiles. The redeposited layer is expected to have a different density and composition compared to the pristine layer and thus could be more prone to reaction with the ALE chemistry, leading to an increased removal rate.
In addition, the digitized features also show a consistent lateral etch rate of 2.5 nm/cycle. This undercut could be due to the presence of oxygen atoms in the plasma, which is isotropic and would oxidize the sidewall laterally. Overall, the chemistry is proven to be highly directional as the vertical etch rate is about eight times the lateral etch rate. Further improvement in anisotropy could be achieved via the implementation of the ion source with low neutral to ion ratio, or with a directional ion source.
In this work, the ALE portion of the process took 1.5 h to complete, combined with 0.5 h of the RIE process; the overall processing time is about 2 h. Compared to the final feature profile achieved in Paper I,11 this hybrid RIE-ALE approach resulted in a partially removed hard mask and about 5 nm of undercut, but the overall processing time is less than 25% of that of the ALE-only process. This drastic improvement provides a new perspective for patterning materials with thickness higher or on the order of tens of nanometers, opening up new etching pathways for high-throughput and high-precision etching.
Achieving nanometer-scale anisotropy is a common patterning goal on various materials, the realization of which is nonetheless not trivial as directional reactivity is difficult to control on the submicrometer level. To address this problem, directional modification is applied via plasma exposure, where the intrinsic directionality of charged species enables subsequent selective removal. With the reduced layer thickness requirement of the EUV mask absorber, the inclusion of RIE could improve the overall efficiency.
IV. SUMMARY AND CONCLUSIONS
An RIE chemistry of Ni is developed to remove the majority of the exposed Ni layer for a higher processing throughput without sacrificing too much SiO2 hard mask thickness, the result of which is then used as the starting point of the subsequent ALE treatment. Ni features with a vertical sidewall were obtained from this RIE-ALE approach with an acceptable selectivity to SiO2 hard mask.
The authors acknowledge the financial support from the Semiconductor Research Corporation (SRC 2802.001), the National Science Foundation (1805112), Lam Research, and the Center of Design-Enabled Nanofabrication (C-DEN). The authors thank Changju Choi and Tristan Tronic at Intel as well as Nathan Marchack at IBM for providing patterned samples. The authors thank Noah Bodzin at the UCLA for TEM sample preparation and imaging.