Gallium nitride (GaN) is a powerful and promising semiconductor that perfectly fills and enhances the function obtained with silicon in technology development and in the fabrication of next generation of high-power devices as for example, the normally-off High Electron Mobility Transistors (HEMTs). The development and optimization of a damage-free etching process is still a research topic. Therefore, the Atomic Layer Etching (ALE) process has become predominant for the development of GaN-based HEMT and the conservation of its electric characteristics since it provides more precise etching depth and is less damaging compared to continuous plasma etching.

Up to today, the largest part of the research on the GaN ALE process has been carried outusing chlorine-based chemistries for the modification step, considering that the GaCl₃ etch by-product is volatile under ion bombardment ^1,2. However, the fluorinated chemistry is less investigated as compared to the chlorinated chemistry for the ALE of GaN ^3,4. This is primarily due to the non-volatility of GaF_x etch by-products as the GaF₃, which has a boiling temperature of 1000°C at atmospheric pressure ⁵. Among the few fluorine based GaN ALE research reports, K. Nakazawa et al ³ have performed a two-step ALE process that involves a CF₄/O₂/Ar plasma adsorption step and an Ar plasma with ion bombardment activation step. A separate study of thermal ALE of GaN using fluorine chemistry was carried out by another group where each cycle is initiated by an exposure to XeF₂ gas, followed by an Ar purge step, and lastly an exposure to BCl₃ gas ⁴.

In our work, the ALE of GaN process consists of a cyclic modification (fluorination) step using SF₆ plasma and an Ar plasma removal step performed in an Inductively Coupled Plasma (ICP) reactor. In this paper we report on the results of energy scan, the synergy and the self-limiting regime of ALE process using fluorinated chemistry. The influence of different plasma parameters such as the ICP power and the pressure on the Etch Per Cycle (EPC) and the RMS surface roughness was also investigated using Atomic Force Microscopy.

The authors would like to thank the European Union (IPCEI standing for Important Project of Commun European Interest) and Région Centre Val de Loire for their support through the Nano2022 project. We also acknowledge CERTeM 2020 platform, which has provided the majority of the equipment.

1. Ohba, T et al, Jpn. J. Appl. Phys. 56, 06HB06, 2017
2. Le Roux, F et al, Microelectron. Eng. 8, 2020
3. Nakazawa, K et al, DPS, 2019
4. Johnson, N. R et al, Appl. Phys. Lett. 114, 243103, 2019
5. Pearton, S. J et al, J. Appl. Phys. 86, 1–78, 1999

GaN-based hetero structured high-electron-mobility transistors (HEMTs) are key components for the next generation of RF (radio frequency) and high power applications due to their high breakdown field, high-temperature operation, and their strong spontaneous and piezoelectric polarization-induced 2D electron gas (2DEG) of high carrier density and mobility. For safety considerations, normally-off operation HEMTs are vital for power applications. Due to their high gate voltage sweep and low leakage current, gate recessed metal insulator semiconductor (MIS)-HEMTs have been intensively investigated for normally-off operation. But their manufacturing has proven so far too challenging. Among the most significant challenges is sub-nanometer etch monitoring and controlling, to leave a thin (typically under 5 nm) layer of AlGaN, obtaining repeatable device performance.

In this work, we have realized of precision and repeatability of 5nm and 2nm AlGaN left on GaN by in-situ etch depth monitoring. we demonstrate how complementing a relatively slow, low damage atomic layer etch process with monitoring of the remaining AlGaN thickness allows excellent etch depth control of the AlGaN. The AlGaN layer thickness is determined by an AlGaN/GaN-optimized LayTec TRIton^TM in-situ UV reflectance metrology tool with ±0.5nm. The accuracy and repeatability of the 2 and 5 nm target depths of AlGaN have been qualified by TEM. All ALE processes were controlled by monitoring the AlGaN depth through the UV reflectance interferometry traces. Furthermore, the UV reflectance results imply that GaN cap and neighbouring AlGaN etch at a varying rate, lower than the reproducible etch rate of bulk AlGaN. The UV reflectance traces clearly resolve the etch steps, which depends on the details of the cyclic ALE process. We have verified the good correlation between AlGaN thicknesses determined by TEM, and in-situ by real-time UV reflectance.

ALE is becoming increasingly popular due to the high level of etch control and low damage etching it can provide. The use of self-limiting half-cycles for ALE is well-documented, however why and how half-reactions self-limit is less well understood. Such fundamental insight into sputter thresholds and yields is vital information that is missing from the literature for many ALE chemistries. In this work we examine (i) the self-limiting nature of the removal step of anisotropic ALE, (ii) the required energy window and (iii) how the etch rate changes as the material is sputtered.

The sputtering of fluorinated Al₂O₃ and HfO₂ is investigated, where the fluorination is performed by a SF₆ plasma and removal by directional Ar ions. Exploration of surface fluorination goes beyond the widely investigated case of Si ALE by chlorination,¹ providing insight into possible alternative ALE chemistries. To accurately evaluate the minimum energies required for material removal, we utilize tailored waveform (TW) biasing,² which allows for improved ion energy control over conventional radio-frequency biasing. The sputter thresholds for Al₂O₃ and HfO₂ were determined to be 59 ± 5 eV and 81 ± 5 eV, respectively. Following modification with an F-based plasma both films were shown to have a decreased sputter threshold, 23 ± 4 eV for Al₂O₃ and 37 ± 4 eV for HfO₂, which opens an ALE window. Additionally, the sputter yield of the modified films is increased as compared to bulk material aiding in removal of the modified layer.

Another important consideration for ALE processing is ensuring that the modified layer can be fully removed. By performing a SF₆ plasma exposure followed by multiple Ar plasma removal steps, we investigated how the layer is removed as a function of Ar plasma exposure time. Initially the etch rate of the modified layer is high, but as the modified layer is removed, the etch rate reduces back to that of bulk material. In our case this suggests the presence of an intermixed AlF_xO_y layer between the surface AlF₃ and bulk Al₂O₃.

Improved understanding of material removal, both in terms of ion energy windows and rate, will facilitate future development of anisotropic ALE processes, and thereby the integration into advanced IC fabrication.

1. Berry, I.L., et al. J. Vac. Sci. Technol. A 36, 01B105 (2018)

2. Faraz, T., et al. J. Appl. Phys. 128, 213301 (2020)

Isotropic ALE processes are being considered for a range of applications such as controlled thinning of material, smoothing of surfaces, and removal of contaminated or damage surface layers. Although most reported isotropic ALE processes use thermal processes, also plasma-based ones are of recent interest, due to a promise of easier to handle precursors and expanded process parameter space. In our recent work, isotropic plasma ALE of Al₂O₃ was performed using SF₆ plasma and AlMe₃, which was demonstrated on a small-scale homebuilt reactor.¹ The AlMe₃ dose in the process needs to be sufficient to cause net etching since otherwise AlF₃ deposition is observed.² The required AlMe₃ dose for net etching will furthermore depend on the degree of fluorination of the various surfaces in the reactor. Therefore, a long and extensive fluorination by the SF₆ plasma could cause challenges in achieving uniform etching with a given AlMe₃ dose.

In this contribution we will discuss the transfer of this ALE process to a commercial ALD tool (Oxford Instruments FlexAL) which would also allow process sequences combining ALD and ALE. Depending on the used substrate temperature, plasma dose and precursor dose, either deposition or etching was found to occurred. After obtaining understanding of these parameters, a decent ALE uniformity was achieved (~2% 1-sigma thickness non-uniformity over 150 mm wafer area). Possible applications (e.g., in quantum and power electronics) and parameters to keep in mind for process transfer will be discussed in this contribution.

1. Chittock et al.,Appl. Phys. Lett. 117, 162107 (2020)
2. Vos et al., Appl. Phys. Lett. 111, 113105 (2017)