AVS 72 Session AP+EL+PS+TF-WeM: Thermal and Plasma Enhanced Atomic Layer Etching

As logic devices dimensions scale down⁽¹⁾ and quantum gates advance in scalability and environment resilience⁽²⁾, the demand for gentle, selective, and highly controlled fabrication processes has become critical. Atomic-scale processing—particularly cyclic approaches like Atomic Layer Etching (ALE) and quasi-ALE—has emerged as a potential solution. By leveraging fluorocarbon-based chemistries in sequential deposition/removal cycles, these methods have already demonstrated superior performances in terms of SiO₂ to Si₃N₄ selectivity, such as in Self-Aligned Contact (SAC) etching⁽³⁾. While the fundamental mechanisms of these cyclic processes are increasingly well understood⁽⁴⁾, their industrial applicability—especially for advanced integration schemes—still faces challenges in versatility, robustness, and throughput. In this presentation, we will highlight recent advances in adapting cyclic etching strategies to selectively etch dielectric materials (SiO₂, SiON, Si₃N₄) for two cutting-edge applications: Contact multipatterning in next-generation FDSOI CMOS logic devices and intertwined gate patterning for cryogenic FDSOI quantum chips. This study explore the critical role of process parameters, including fluorocarbon precursor selection, multi-step cyclic strategies, substrate temperature and plasma power modulation, and their collective influence on etch per cycle stability, selectivity, uniformity, throughput, and pattern transfer fidelity.

By bridging fundamental insights with integration-driven optimization, this work clears the path to highly scalable, atomic-precision etching—a key enabler for both classical and quantum semiconductor technologies.

The research work presented in this paper was carried out in the framework of the FAMES Pilot Line of the Chips JU, funded by Horizon Europe grant 101182279 and the ANR NextGen project ANR-22-NEXTG001 of the France 2030 initiative. Part of this work, carried out on the Platform for Nanocharacterisation (PFNC), was supported by the “Recherche Technologique de Base” and "France 2030 - ANR-22-PEEL-0014" programs of the French National Research Agency (ANR).

(1)C. Fenouillet-Beranger, et.al., Solid-State Electronics, 2026, 231, 109264, https://doi.org/10.1016/j.sse.2025.109264

(2)B. Bertrand, et.al., International Electron Devices Meeting, IEDM 2023, DOI: 10.1109/IEDM45741.2023.10413763

(3)M. Honda, et.al, 2017, J. Phys. D: Appl. Phys. 50 234002 [tel:50 234002], https://doi.org/10.1088/1361-6463/aa6f27

(4)R.J. Gasvoda, J. Vac. Sci. Technol. A 38, 050803 (2020); https://doi.org/10.1116/6.0000395

Titanium nitride (TiN) has become a widely adopted material for both logic and memory devices due to its appropriate work function with high-k gate stacks and excellent barrier properties against fluorine and oxygen diffusion. Atomic layer etching (ALE) of TiN was performed by a combination of surface modification by oxidation during the O₂/Ar plasma half-cycle and release of volatile etch products in the SF₆/H₂/Ar plasma half-cycle. [1,2] For the ALE experiments, we used low-resistivity TiN deposited by atomic layer deposition. In the O₂/Ar plasma half-cycle, TiN was oxidized to TiO_x with O radicals. For the SF₆/H₂/Ar plasma cycle, the SF₆-to-H₂ flow ratio was tuned to minimize continuous etching of TiN. Quadrupole mass spectrometry (QMS) was used to identify conditions that maximize the production of HF in the SF₆/H₂/Ar plasma, which acts as the etchant for surface modified TiO_x layer, and does not spontaneously etch TiN. Using QMS, we also identified the primary reaction products produced in each half-cycle. Our data show that NO was the primary nitrogen-containing product in the Ar/O₂ plasma half-cycle. TiF₂⁺ and TiF⁺ ions were detected by the QMS in the SF₆/H₂/Ar plasma half-cycle, which most likely originate due to cracking of TiF₄ parent molecules in the ionizer. The likely O-containing product during the SF₆/H₂/Ar plasma half-cycle is H₂O, which could not be detected due to the background signal in the QMS. In addition, we also detected SO⁺ and SO₂⁺ ions during both half-cycles due to the presence of S on the substrate and/or the chamber walls.

The etching process was performed at 200 °C. While the oxidation of TiN was quasi-self-limiting during the O₂/Ar plasma half-cycle, subsequent etching in the SF₆/H₂/Ar plasma half-cycle was not self-limiting. The synergy parameter, S, [3] is an indicator of the ideality of the ALE process. We show that the ALE end point could be determined by the time decay of the reaction products measured by the QMS, and the synergy parameter could be optimized by limiting the duration of the SF₆/H₂/Ar plasma half-cycle. We were able to obtain a synergy parameter of ~82% under optimized conditions.

[1]Lee et al., Chem. Mater. 29, 8202 (2017);

[2]Hossain et al., J. Vac. Sci. Technol. A 41, 062601 (2023);

[3] Kanarik et al.,J. Vac. Sci. Technol. A 35, 05C302 (2017);

Precise, damage-free etching of InGaAs with Angstrom-scale control is a critical requirement for next-generation III-V semiconductor device fabrication, including gate recess etching in InGaAs HEMTs, which require very precise etching of ~20 nm of InGaAs. Conventional plasma etching is insufficiently precise for this application and introduces subsurface damage and surface roughness, motivating the development of atomic layer etching (ALE) processes that achieve controlled material removal through sequential, self-limiting surface reactions.

In this work, we investigate a plasma-based ALE process for InGaAs using (CH₄+H₂)/Ar⁺ chemistry, consisting of alternating surface modification and removal half-cycles. The expected volatile products are In(CH₃)₃, Ga(CH₃)₃, and AsH₃. The effects of chamber conditioning, process pressure, and Ar flow rate are systematically investigated to achieve self-limiting ALE behavior. Real-time optical emission spectroscopy (OES) measurements are used to monitor plasma species and investigate the role of residual CH-related species in the transition between self-limiting ALE and RIE-like behavior. For a process at 10 mTorr with 200 W ICP power, 5 sccm Ar, a 1 sec dose of 5 sccm CH₄ and 20 sccm H₂, and 5 sec etch time, an ALE window is present from 10-15 W RF bias, with EPC of 0.25 nm/cycle. The etched surfaces are characterized using scanning electron microscopy (SEM), energy dispersive spectroscopy (EDS), atomic force microscopy (AFM), and profilometry to evaluate etch per cycle (EPC), surface morphology, roughness evolution, and compositional uniformity of the InGaAs surface. Synergy is also measured.

Aluminum is a light, conductive metal used in an incredible assortment of device fabrication processes, including in superconducting devices, where performance is highly sensitive to material impurities, interface damage, and critical dimensions. Most aluminum reactive ion etch processes use Cl₂ or BCl₃, which have the dual disadvantages of toxicity and post-etch corrosion. Fluorine is not typically used as an etchant for Al films because of the high melting point of AlF₃. It is known, however, that fluorinated AlO_x has a lower sputter threshold than untreated AlOx[1]. We present a novel 3-step plasma Atomic Layer Etch (ALE) process for Al using O₂, CF₄, and Ar⁺. This room-temperature cyclical process starts by oxidizing Al in O₂ in Step A. Step B consists of a tetrafluoromethane dose to create an aluminum oxyfluoride modified layer. In Step C, this modified layer is removed via argon ion bombardment, using the oxide as an etch stop for the removal of the more easily sputtered oxyfluoride layer.

EPC of 0.08 nm/cycle and synergy of >99% were achieved with 100 W ICP power and 7 W table bias during step C. Etch control is governed by the thickness of the grown oxide and the depth of fluorine diffusion. We also present AFM, SEM, and XPS to show the effect of this process on the Al surface. This process could provide an avenue toward improved performance in aluminum-based superconducting devices.

[1] Chittock, N. J. (2024). Plasma processes for atomic layer etching. [Phd Thesis 1 (Research TU/e / Graduation TU/e), Applied Physics and Science Education]. Eindhoven University of Technology.

Superconducting materials are foundational to many quantum technologies, yet surface and interfacial disorder remain important barriers to device performance and scalability. In particular, native oxide formation on superconducting metals is believed to contribute to loss and decoherence, creating a need for processing methods that provide precise surface control while maintaining the integrity of the underlying material. Niobium is an attractive superconducting material because of its relatively high critical temperature and broad relevance to quantum circuit platforms. However, complex native oxide formation and near-surface chemical disorder complicate its integration into devices such as Josephson junctions and other superconducting circuit elements. Here, we investigate an atomic-scale surface treatment and passivation approach for niobium thin films aimed at improving control over the metal/oxide interface. Our results show that thermal atomic layer etching can effectively remove native surface oxides from niobium while preserving the underlying metallic surface. Subsequent atomic layer deposition prevents reformation of the native oxide layer on the surface. Together, these findings establish a promising pathway for interface control in superconducting niobium and support the development of improved fabrication strategies for quantum circuit technologies.

Indium gallium oxide (IGO) and Indium gallium zinc oxide (IGZO) can be used as channel materials in metal oxide semiconductor thin-film transistors (TFTs). These TFTs have high carrier mobility, low leakage current and a wider bandgap compared with silicon. Etching the channel material to a precise and controlled thickness is important for device performance. Thermal atomic layer etching (ALE) of these metal oxide semiconductors is possible using surface modification and volatile release reactions.

In this study, the thermal ALE was performed with sequential exposures of boron trichloride (BCl3) and hydrogen fluoride (HF). In situ spectroscopic ellipsometry (SE) measurements on flat films determined etch rates for IGO of 2.1 Å/cycle at 225°C and 3.0 Å/cycle at 255°C. The etch rate for IGZO was 2.5 Å/cycle at 205°C and 5.6 Å/cycle at 255°C. The etch rates were observed to be nearly self-limiting versus the BCl3 and HF exposures.

Quadrupole mass spectrometry (QMS) investigations identified the volatile products during the BCl3 and HF exposures on In2O3, Ga2O3 and ZnO powders. The initial BCl3 exposure reacted with the In2O3, Ga2O3 and ZnO powders and produced InCl3, GaCl3 and ZnCl2, respectively. These products are consistent with conversion to B2O3. The next HF exposure then produced BF3 and H2O from the spontaneous etching of B2O3. The HF then proceeded to yield additional H2O from the fluorination of the metal oxide. The subsequent BCl3 exposure first yielded BF3 from ligand-exchange with the metal fluoride. The BCl3 exposure then proceeded to produce metal chloride by conversion of the underlying metal oxide. The QMS results for Ga2O3 powder are displayed in Figure 1.

Additional studies on flat films explored the effect of IGO composition on the etch rate. Figure 2 shows that the IGO etch rates were 2.8, 3.6, and 4.2 Å/cycle at In:Ga ratios of 1:1, 2:1 and 4:1, respectively at 230 °C. The etch rate then decreased to 0.8 Å/cycle at an In:Ga ratio of 8:1. The thermal ALE was also observed to smooth the initial roughness of the IGZO surface.

The fabrication of gate-all-around nanosheet transistors employs multi-layer stacks of crystalline SiGe and Si. Completely removing sacrificial SiGe and precisely recessing SiGe may be accomplished by chemical vapor etching (CVE) and atomic layer etching (ALE), respectively. Facile Si CVE has been shown previously using anhydrous HF vapor above 150°C. This Si CVE is ceased by co-dosing H₂O+HF. The present work explored selective CVE and ALE pathways for Ge compared with Si using HF only or with oxidants.

Quadrupole mass spectrometry (QMS) measurements identified the volatile etch products during different reactant gas exposures at ramped or isothermal temperatures. Exposing Ge nanopowder to HF during an up-ramp from 25 to 400°C discovered that facile Ge CVE produced volatile GeF₄ at 50-75°C and then turned off due to non-volatile GeF₂ polymer formation. The GeF₂ polymer could then be volatilized at >300°C. Additional experiments at varied ramp rates, a reversed down-ramp, and isothermal HF exposures corroborated this non-volatile GeF₂ polymer formation.

Exposing a mixture of Ge & Si nanopowders to HF during a temperature ramp showed that the GeF₄ etch product in turn etched Si, producing volatile SiF₄ at 75-110°C (Figure 1a). Co-dosing H₂O+HF still etched Ge (or GeO₂) and produced volatile GeF₄. However, co-dosing H₂O+HF eliminated the parasitic Si etch pathway via GeF₄ (Figure 1b). Introduced H₂O oxidized Si and formed an ultra-thin chemical oxide. This SiO₂ was not etched by GeF₄.

Repeatedly exposing GeO₂ nanopowder to HF at 100°C revealed that GeO₂ etched spontaneously without the self-termination observed for Ge. This finding inspired a unique Ge ALE mechanism: A first oxidant exposure may oxidize Ge to form an ultra-thin GeO₂ surface layer. A subsequent HF exposure may volatilize this GeO₂ surface layer but self-terminate on the underlying Ge due to GeF₂ polymer buildup. The following oxidant exposure may remove the GeF₂ polymer and again oxidize Ge.

Our earlier research demonstrated SiO₂ etching by HF using a H₂O liquid-layer in a vacuum environment. The current work characterized the dependence of SiO₂ etch rate on the H₂O and HF pressures. We have observed a dramatic increase in SiO₂ etch rates at higher H₂O and HF pressures. This behavior suggests that the thickness of the HF/H₂O liquid-layer may be increasing rapidly at higher pressures.

SiO₂ etch rates may be sensitive to the HF/H₂O liquid-layer thickness when the thickness is close to the size the hydrated F^-, HF₂^- and SiF₆^2- ions that are involved during SiO₂ etching. For example, the size of the hydrated F^- ion is 7 Å. Lower SiO₂ etch rates may result from incomplete hydration of these ions. SiO₂ etch rates may approach the SiO₂ etch rates observed in solution for thicker liquid-layers.

Investigations were performed to determine the thresholds for HF/H₂O condensation versus H₂O and HF pressure. Spectroscopic ellipsometry was employed to measure the liquid-layer thickness, the mean squared error (MSE) in the thickness model, and the transmitted light intensity. Condensation was assumed to occur when the increase in liquid-layer thickness, the increase in MSE, and the decrease in light intensity all occurred rapidly at higher H₂O and HF pressures. Figure 1 shows that rapid changes in the liquid-layer thickness, MSE and transmitted light intensity are observed at a total pressure of 22.9 Torr (18 Torr H₂O and 4.9 Torr HF) at 30.4°C.

In GAAFET fabrication, Si_1-xGe_x is etched selectively from Si. Wet, plasma, and thermal chemistries have been investigated to etch Si_1-xGe_x with a range of selectivity [1]. Thermal fluorine-based etch processes, including F₂ and XeF₂, show high selectivity with etch rates exhibiting a volcano-shaped dependence on x [2, 3].

In this study, we report Si_1-xGe_x (x = 0 to 1) etch by a XeF₂ beam. XeF₂ was sublimed into an expansion chamber to a target pressure, then pulsed into the reaction chamber. In each pulse, ∆P_XeF2 was proportional to the XeF₂ dose. Ex situ x-ray photoelectron spectroscopy (XPS) identified the chemical bonding states on the Si_1-xGe_x surface from XeF₂, while scanning electron microscopy (SEM) measured the etch depth for etch rate (ER) calculations.

XeF₂ exposure on as-received Si_1-xGe_x samples showed no etch, confirmed by XPS and SEM. This was attributed to a native oxide which prevented XeF₂ from interacting with the underlying Si_1-xGe_x. Prior to XeF₂ exposure, in situ Ar ion sputtering was used to remove the native oxide for all sequential etch conditions.

XPS results of XeF₂-processed Si_0.75Ge_0.25 , Si, and Ge identified Ge-F, Si-F, and C-F bonding in the F 1s region at binding energies of 685.0 eV, 687.1 eV, and 689.6 eV, respectively, consistent with literature values [4-6]. Ge-F, Si-F, and C-F made up 20.8%, 37.4%, and 41.8% of the total fluorine content on the XeF₂-processed Si_0.75Ge_0.25. Quantitative analysis of fluorination by ex situ XPS was limited by air exposure between etch and measurement.

Synergistic etching, with Ar ions and XeF₂ simultaneously exposed, showed a Si ER of 6 nm/pulse at an ion beam voltage of 200 V and ∆P_XeF2 =0.5 Torr, showing a higher ER than the sequential Si ER of 0.4 nm/pulse at the same ion beam voltage and ∆P_XeF2.

The peak sequential Si_1-xGe_x ERs for the following Ar ion beam voltage and ∆P_XeF2 conditions, 100 V and 0.5 Torr, 200 V and 0.5 Torr, and 100 V and 0.2 Torr, were 7.8 nm/pulse, 9.9 nm/pulse, and 4.1 nm/pulse, respectively. For all conditions, ER increased from x = 0 to 0.1, followed by a decrease or plateau past x = 0.15. Higher pretreatment ion beam voltage and higher ∆P_XeF2 both led to an etch rate increase for x = 0 to 0.25. The peak Si_1-xGe_x/Si etch selectivity ranged from 20 to 30 under all conditions.

This research was supported by Lam Research Corporation. The development of the experimental system was partially supported by the U.S. DOE, Office of Science, as part of the ELMIC MSRC through the Plasma-enabled 2D Materials project at PPPL, under contract No. DEAC02-09CH11466.

References