AVS 70 Session AP+PS+TF-TuA: Atomic Layer Etching III: Plasma Processes

The critical dimensions of semiconductor devices are continuously shrinking in nanometer and atomic scale with 3D device structure. The demand for dimension control in angstrom level is drastically increasing also in etching processes. Atomic layer etching (ALE) processes are being actively studied and developed for various metals, semiconductor, and dielectric materials. In this talk, plasma processes for atomic layer etching will be discussed for both isotropic and anisotropic patterning of metals and dielectric materials including molybdenum, ruthenium, cobalt, titanium nitride, tantalum nitride, hafnium oxide, zirconium oxides. [1-9] Typical ALE processes consist of surface a modification step and a removal step. For the surface modification, various fluorination, chlorination and oxidation schemes were applied including fluorocarbon deposition, halogenation, oxidation with radicals generated plasmas. For the removal or etching step, various schemes were applied including ion-bombardment, heating, ligand volatilization, ligand exchange, and halogenation. The surface characteristics such as surface roughness and surface residue after plasma-enhanced ALE processes will be also discussed.

1) K. Koh, Y. Kim, C.-K. Kim, H. Chae, J. Vac. Sci. Technol. A, 36(1), 10B106 (2017)

2) Y. Cho, Y. Kim, S. Kim, H. Chae, J. Vac. Sci. Technol. A, 38(2), 022604 (2020)

3) Y. Kim, S. Lee, Y. Cho, S. Kim, H. Chae, J. Vac. Sci. Technol. A, 38(2), 022606 (2020)

4) D. Shim, J. Kim, Y. Kim, H. Chae, J. Vac. Sci. Technol. B., 40(2) 022208 (2022)

5) Y. Lee, Y. Kim, J. Son, H. Chae, J. Vac. Sci. Technol. A., 40(2) 022602 (2022)

6) J. Kim, D. Shim, Y. Kim, H. Chae, J. Vac. Sci. Technol. A., 40(3) 032603 (2022)

7) Y. Kim, H. Chae, Appl. Surf. Sci., 619, 156751 (2023)

8) Y. Kim, H. Chae, Appl. Surf. Sci., 627, 157309 (2023)

9) Y. Kim, H. Chae, ACS. Sustain. Chem. Eng., 11, 6136 (2023)

This talk will report an experimental investigation of Cl₂ vs. HBr for plasma atomic layer etching (ALE) of silicon. An inductively coupled plasma (ICP) source with a constant flow of Ar (and sometimes He) carrier gases, and HBr or Cl₂ as a dosing gas was used for etching Si (100) samples. Optical emission spectroscopy was used to follow relative yields of SiCl, SiCl₂, SiBr and SiBr₂, and scanning electron microscopy and profilometry were used to measure etching rates. HBr and Cl₂ residence times in the chamber were determined by measuring time-resolved pressure during gas dosing. It was found that the pressure rise and fall times were much longer for HBr compared to Cl₂, suggesting that HBr hangs up on the chamber wall after gas dosing. The effect of the delay time between gas dosing and the start of ICP power on the etching rate was also investigated. When HBr or Cl₂ were injected into the reactor with the plasma on, etching occurs for both sources, with Cl₂ having a higher etching rate. When HBr or Cl₂ were fed to the reactor with the plasma off, only Cl₂ etches. This indicates that the HBr does not chemisorb on Si, and bromination of the surface requires the plasma to form Br atoms, which do adsorb.

Molybdenum disulfide (MoS₂) is a two-dimensional (2D) transition metal dichalcogenide (TMD) with applications in catalysis and nanoelectronics.^1–3 To enable integration of 2D materials in nanoelectronics, highly controlled and low-damage etching processes are required. One example of such a process is the atomic layer etching (ALE) of WSe­­₂.⁴ In this work, we present a plasma ALE process to etch crystalline MoS₂. The process involves plasma fluorination and oxygenation, targeting Mo-O-F as a volatile species. Previous work using the approach of fluorination and oxygenation involved thermal chemistries with an etch-per-cycle (EPC) of 0.5 Å for amorphous films and 0.2 Å for crystalline films.⁵ As plasmas are more reactive than their gaseous counterparts, they are expected to allow for etching crystalline materials.

Fluorination was performed by a SF₆-based plasma, and as removal step various oxygen sources were tested. To avoid continuous etching of the MoS₂, the plasma composition of the fluorination step had to be optimized.⁶ Addition of H₂ to the plasma mixture reduces F radical concentration by creating HF species in situ.⁷ With an adequate H₂ content using a ratio of SF­₆:(SF₆+H₂) < 0.3, continuous etching is suppressed, and only modification of the top surface takes place. As second half-cycle, H₂O, O₂ gas and O₂ plasma exposures were tested. Of these only an O₂ plasma resulted in etching, while dosing H₂O or O₂ gas resulted in no measurable thickness change. Raman measurements showed a strong decrease in the characteristic E¹_2g and A_1g peaks, indicating etching of crystalline MoS­₂. The complete ALE recipe with saturated SF­₆-based plasma and O₂ plasma exposures had an EPC of 1.1 ± 0.2 Å at 300 °C table temperature.

1. Cao, Y. ACS Nano15, 11014–11039 (2021).
   
2. Mahlouji, R. et al. Adv. Electron. Mater.8, (2022).
3. Deijkers, J. H. et al. Adv. Mater. Interfaces 4–9 (2023).
4. Nipane, A. et al. ACS Appl. Mater. Interfaces13, 1930–1942 (2021).
5. Soares, J. et al. Chem. Mater.35, 927–936 (2023).
6. Xiao, S. et al. Sci. Rep.6, 1–8 (2016).
7. Hossain, A. A. et al. J. Vac. Sci. Technol. A41, (2023).

Any future physics-based, rather than purely data-driven, digital twin of a plasma tool and process will need a reduced order model (ROM) of the effects of the plasma on the surface being processed. In this talk, I present one possible version of a ROM for simulating a plasma atomic layer etching (ALE) process. The ROM is based on a transient version of a surface site balance model that was first applied in the 1980s for plasma etching. Classical molecular dynamics (MD) simulations of ALE of Si using Cl₂/Ar⁺ cycles are first validated with experimental measurements and then used to provide values for parameters (e.g., ion energy dependent etch yields) for a transient site balance model of the process. The ROM is computationally much faster than the MD simulations and is shown to reproduce additional, and previously unexplained, experimental results. Future applications of MD to plasma-surface interactions in semiconductor device fabrication will likely use another type of ROM. MD methods require interatomic potentials or force fields for all simulated atomic interactions. The force field parameters can be inferred from higher level methods such as density functional theory (DFT), coupled with neural net data-driven algorithms. I will conclude with preliminary results using one such approach, namely Deep Potential MD (DeePMD), to develop illustrative force fields.

Atomic layer etching (ALE) is a critical technology in semiconductor device fabrication, including for interconnect metals such as Cu and Co. Previously, thermal ALE of Cu has been reported using alternating half-cycles of hexafluoroacetylacetone (hfacH) and O2 or O3 at ~275 °C. It has also been shown that hfacH cannot spontaneously etch Cu, and peroxidation of the Cu surface is required. In this presentation, we will report on ALE of Cu at a much lower temperature of 150 °C using hfacH and O2/Ar plasma half-cycles.

Ruthenium is a potential replacement for copper in metal interconnects below 10 nm, where the grain boundary scattering and the need for a barrier layer increases the effective resistivity of Cu. Unlike Cu, Ru can be directly etched using O₂-plasma-based processes, allowing a subtractive metal patterning to be used. Current Ru etching processes largely rely on O₂/Cl₂-based RIE. While this process is able to anisotropically etch Ru, it requires Cl₂ and damages the hard mask, resulting in ballooning and low selectivity. A cyclic process that is halogen-free, anisotropic, and selective has been proposed. The process consists of a nitrogen plasma passivation step, hydrogen plasma reduction step, and oxygen plasma etching step. The passivating layer of RuN formed by N₂ plasma exposure can be reduced in H₂ plasma. Bias applied to the substrate during the H₂ step results in the selective reduction of RuN on the vertically-exposed surface, leaving a layer of RuN on the sidewalls. During the O₂ plasma step, RuN passivates the sidewall from O radicals that would otherwise etch the sidewall. The two surface modification steps cause the normally isotropic O₂ plasma etch to become anisotropic, while maintaining a high selectivity vs. SiO₂ and SiN used as the hard mask. The sequential process has been experimentally shown to be anisotropic and selective, and results in lower resistivity and surface roughness vs. O₂ plasma alone. To understand the effects of the surface modification steps on the overall process, each step of the sequential process has been examined, including the surface composition and structure of the Ru film after each plasma exposure. RuO₂ and RuN thin films were deposited as references, allowing the presence of RuN after nitridation to be confirmed using XPS and XRD. The etch rate of nitridated Ru films was found to be 0.4 nm per minute in O₂ plasma, more than 15 times less than the etch rate of Ru in O₂ plasma at the same conditions. Removal of the N₂ plasma step from the process resulted in isotropic etching, confirming that nitridation is responsible for sidewall passivation. XPS was used to confirm that most, but not all of the RuN surface layer is reduced after exposure to H₂ plasma. H₂ plasma was also found to significantly affect the etching of hard mask SiO₂. Reduction of the plasma density by lowering the power and increasing the pressure was necessary to reduce SiO₂ etching and corner faceting during the H₂ step. The combination of surface modification techniques results in a process that has been demonstrated for features down to 32 nm pitch.

A novel plasma process design called transient-assisted plasma process (TAPP) has been recently introduced to tackle multiple patterning challenges brought by the introduction of 3D devices, new ultra-thin films, and compound materials without suffering from the poor throughput of Atomic layer etching. In the realm of dry-etching, it has exhibited promising patterning capabilities at etch rates compatible with high-volume manufacturing. In deposition applications, It also has demonstrated superior control over precursor dosage and fragmentation for in-situ hard-mask deposition compared to conventional methods. Moreover, from a sustainability perspective, Transient-assisted processing (TAP) presents considerable advantages by notably reducing the consumption of problematic gases. TAPE operates in cycles, involving at least two phases: Time-limited injection of the reactant (with or without plasma), and when the injection is stopped, a gas transient happens in the plasma, where the reactant concentration diminishes over time, as shown in Figure 1. The fluence of reactive species is regulated by the gas pulse characteristics (partial pressure, etc.) and its associated plasma transient, while ion fluence is governed by the duration of the plasma phase. This segregation yields significantly enhanced control over plasma-surface interactions compared to conventional plasma etching techniques. This design is compatible with any gas mixture, and energy sources (ions, photons, electrons, fast neutral species/clusters, etc.).

Enhancing control over plasma-surface interactions holds promise in rectifying the uneven etching observed in multicomponent materials like alloys_. This imbalance in etching induces a compositional drift, thereby impeding the process and deteriorating material characteristics as shown in Figure 2a. Typically, one component undergoes predominantly chemical etching, while the other experiences primarily physical etching. In TAPE, most chemically-driven etching occurs early during the plasma step when a substantial amount of etchant is present. The modified surface/profile will then be exposed to a reduced etchant quantity and a continued ion bombardment. Each cycle is thus capable of providing the necessary species for a balanced etching of the compound’s elements. Figure 2b compares the chemical composition of InGaZnO after conventional etching and TAPE, while Figure 2c compares the etch rate, and profile relative to a previously published ALE process. Meanwhile, TAPE consumes 25 times less CH₄ than the ALE process.

Extreme ultraviolet (EUV) lithography is entering a new era with high numerical aperture (NA) EUV, increasing the importance of integrating new absorber layer materials like Ni₃Al to mitigate mask 3D effects. However, Ni₃Al adoption has been limited by difficulties in anisotropically etching Ni₃Al with high selectivity to the underlying Ru capping layer. A previously reported ALE based on plasma oxidation was shown effective at etching Ni₃Al but had poor selectivity to Ru, which etches spontaneously in oxygen plasma. An atomic layer etch (ALE) based on plasma nitridation and formic acid (FA) vapor exposure was previously demonstrated to etch Ni¹, and in this work the nitridation-based ALE process is extended to the more complex and industrially relevant Ni₃Al.
X-ray photoelectron spectroscopy (XPS) is used to analyze the surface composition of blanket Ni₃Al films after plasma nitridation, showing the nitrided films have a stoichiometry of Ni_2.4AlN. The effects of plasma power, from 200 to 700 W, and pressure, from 8 to 50 mTorr, on plasma nitridation are examined, with current results indicating that high power and low pressure lead to the greatest amount of surface nitridation. Langmuir probe measurements are used to correlate these effects with changes in plasma density and electron temperature. For the removal half-cycle, FA vapor was shown to remove the nitride layer as indicated by the disappearance of the characteristic metal nitride peak in the N 1s spectrum at 397.2 eV. XPS also shows depletion of Ni as successive ALE cycles are applied, an effect that can be mitigated using basic etchants due to their selective removal of Al over Ni. Selectivity to Ru is measured by etching blanket Ru films and measuring the thickness with SEM, with a current selectivity of 2.4. To assess anisotropy, Ni₃Al is first conformally deposited over a patterned Si substrate by physical vapor deposition (PVD), and the samples are subsequently processed by ALE. The ion energy of the nitrogen plasma is varied by changing the voltage applied to an ion extraction grid from -100 to +100 V, with the sidewall Ni₃Al etch compared to the vertical Ni3Al etch using scanning electron microscopy (SEM).

¹ T.G. Smith, A.M. Ali, J.F. de Marneffe, J.P. Chang, J. Vac. Sci. Technol. A 42, 022602 (2024).

Power electronics are essential for our daily lives, driving everything from consumer gadgets to industrial machinery. Gallium Nitride (GaN) stands out in this domain for its superior electrical properties, offering higher efficiency and power density compared to traditional silicon-based devices. Vertical GaN nanowire (NW) high-electron-mobility transistors (HEMTs) represent the forefront of GaN-based technologies, boosting advantages such as increased power density, reduced switching losses, improved heat dissipation, enhanced breakdown voltage, scalability, and radiation hardness. Despite these benefits, conventional fabrication techniques for GaN NW transistors face challenges. Non-uniform mask etching can lead to variations in NW diameters, impacting device performance. Reactive ion etching (RIE), which is often used for mask etching, may damage the substrate surface, compromising interface quality and device functionality. Surface irregularities exacerbate electron scattering, increasing energy dissipation and degrading electrical interfaces. Addressing these issues, atomic layer etching (ALE) emerges as a solution. ALE operates at the atomic level, removing one layer at a time without harming the substrate. Unlike traditional methods, ALE ensures a high-quality interface for improved device performance.

In our study, we utilize ALE to enhance the fabrication of GaN NW-based HEMT transistors. Our method employs fluorine-based chemistry for uniform SiN mask etching, enabling precise pattern transfer, and facilitating uniform NW growth. Additionally, we introduce chlorine-based ALE to refine substrate surfaces, reducing roughness and improving device performance. AFM characterization reveals a substantial decrease in surface roughness, directly contributing to enhanced electrical performance. We use Retarding Field Energy Analyzer (RFEA) data to optimize ALE parameters for different etching steps, and enhancing device fabrication.

In conclusion, our study highlights the potential for ALE processes to improve GaN NW-based HEMT transistor fabrication. By addressing mask uniformity and substrate surface roughness issues, ALE enables the production of GaN devices with superior interface quality and enhanced electrical performance.

Aknowledgemtn The ALL2GaN Project (Grant Agreement No 101111890) is supported by the Chips Joint Undertaking and its members including the top-up funding by Austria, Belgium, Czech Republic, Denmark, Germany, Greece, Netherlands, Norway, Slovakia, Spain, Sweden and Switzerland.