AVS 68 Session PS1+AP-WeA: Plasma Assisted Atomic Layer Etching

Interest in atomic scale processing techniques, such as atomic layer etching (ALE), is growing rapidly as accurate thickness control and low damage etching have become necessities in IC fabrication. In this work the mechanism for isotropic plasma ALE of Al₂O₃ using the diketone species hexafluoroacetylacetone (Hhfac) and H₂ plasma is investigated. This alternative process for isotropic plasma ALE could provide improved selectivity for etching metal oxides over Si-based materials.^1,2

ALE behavior was confirmed using Al₂O₃ planar substrates. Etching is only observed when alternating between Hhfac and H₂ plasma pulses, with the EPC determined to be 1.7 Å/cycle at 350 ˚C, which is higher than the EPC observed for Hacac/O₂ plasma ALE of Al₂O₃.² No thickness change was observed on thermally grown SiO₂ with this ALE chemistry, indicating near perfect selectivity between Al₂O₃ and SiO₂.

This process exhibits ALE characteristics; however, the self-limiting mechanism is not fully understood. To investigate the reaction mechanism Fourier transform infrared spectroscopy (FTIR) was employed to observe the different binding configurations of hfac on an Al₂O₃ surface. Two configurations can be determined from FTIR: chelate where both oxygen atoms bind to the same metal atom, or monodentate when only one oxygen is bound.³ Chelate is the preferred binding configuration; however, for low dose times the majority of measured hfac surface species are in monodentate configuration. It was concluded that chelate species are lost from the surface as volatile metal-diketonate etch products. As dosing continues, the monodentate coverage increases and inhibits chelate binding, and thus etching, which explains the self-limiting behavior. A H₂ plasma pulse in half-cycle B removes the adsorbed hfac, resetting the surface for the next ALE cycle.

In summary the two binding configurations of hfac adsorbates play important and distinct roles: the chelate configuration generates volatile etch products, while the monodentate configuration ensures that the etch is self-limiting. Together these reactions, combined with surface cleaning during half-cycle B, yield an ALE process for Al₂O₃ with excellent selectivity to SiO₂ and a high level of etch control.

1. Elham Mohimi, ECS J. Solid State Sci. Technol. 7 P491, 2018
2. A. Mameli, ACS Appl. Mater. Interfaces, 2018, 10
3. M Merkx, Chem. Mater. 2020, 32, 8

Atomic Layer Etching and Quasi Atomic Layer Etching processes have gained a lot of traction and application throughout the last years. However, not everything described as ALE is a true atomic layer process in its strict definition. We will first discuss the use and definition of various cyclic process and compare to pulsed processes, the process regime of which can differ significantly. In a more general consideration, the approach of reactant control, as the core of ALE, is additionally employed in various other applications that are not QALE. Subsequently, we will highlight and describe specific use cases for which such processes have been explored and implemented at IBM Research in Albany. Fin patterning is a major example of QALE yielding major improvements in profile and CD control during RIE. Additionally, QALE allowed enhanced selectivity during self-aligned contact etches, crucial to its success. For S/D recess possible benefits are expected in a similar fashion. While many of the applications focus on the FEOL, there are also applications of cyclic processes, or QALE-like, in the BEOL, such as Photoresist protection and CD control.

Copper is widely used for interconnect lines in semiconductor devices, and the resistivity of copper is rapidly increasing as the critical dimension shrinks below 30nm. [1] Dual damascene process based on copper have reached limits due to the continuous shrinkage of critical dimensions. Semi damascene process is an alternative technique, and metal is patterned by reactive ion etching (RIE). [2] Conventional plasma etching process has limitations in uniformity, surface roughness. Atomic layer etching (ALE) process has advantages of good uniformity, smoother surfaces, and precise thickness control than conventional reactive ion etching process.

In this work, ALE process was performed for molybdenum in an inductively coupled plasma (ICP) reactor. The ALE process consists of two steps: surface modification with fluorine contained plasma and ion bombardment with Ar plasma. In the first step, the fluorocarbon layers are deposited on molybdenum surface with C₄F₈ or CHF₃ plasmas. In the second step, the modified layers are etched with ions from Ar plasma. Molybdenum ALE process was performed with changing Ar plasma bias voltage at 50~350V, and the ALE window that constant etch rate region was confirmed with bias voltage of 100~230V. Etch rate of cobalt was self-limited at the Ar plasma time of 3 minutes. The etch rate of molybdenum controlled below 1 nm/cycle. The surface roughness was compared RIE process and ALE process, and the surface roughness of the ALE process was lower than that of the RIE process.

References

[1] S. Decoster, E. Camerotto, G. Murdoch, S. Kundu, Q. T. Le, Z. Tokei, G. Jurczak, and F. Lazzarino, J. Vac. Sci. Technol. B 40, 032802 (2022)

Keywords: Atomic layer etching, Molybdenum etching

Anisotropic etching of silicon with active fin heights of 45nm or greater is critical for fin patterning and continuous CMOS scaling. Tight control of fin CD and taper is critical for the device, in particular channel control. In addition, uniform etching of iso/dense features and accurate pattern transfer across various fin dimensions require Si etching to enter atomic level control. In this IBM and Lam collaborative study, we evaluated the impact of He and Ar sputter as well as silicon nitride and silicon oxide hard mask (HM) on selectivity and Si profile. We find that nitride HM tends to be less resilient than oxide HM as sputter energy increases. For both He and Ar ALE, we explored the influence of ion-angle-scattering on fin profile as well as various surface protection methods to address ion-angle-scattering side-effect. We will also examine Ar ALE at low bias.

Optical emission spectroscopy was used as a real-time monitor of atomic layer etching (ALE) of Si in an Ar inductively-coupled (ICP) plasma. Pulses of Cl₂ gas were repetitively injected into a continuous flow of Ar, followed by simultaneous ignition of the ICP and application of substrate rf bias power (either continuous, or rapidly modulated) for 10s. Optical emissions in the UV from Si, SiCl, SiCl₂, SiCl₃ and Ar⁺, and in the far-red from Ar and Cl were monitored as a function of time during the bias period, as well in the ICP after the bias power was extinguished. By monitoring emission along a line parallel and close to the substrate surface, and operating at high Ar flow rates, and from an analysis of the time dependencies of the decays of emissions during the bias period, it is argued that emissions are mainly from the primary products sputtered by the energetic Ar ions. Products decay with different, non-exponential time signatures. Cl rapidly decays to undetectable levels early in the bias period, indicating some sputtering of Cl (and/or Cl₂) from the top surface, while Si emission decays more slowly to a nearly constant level, indicative of physical sputtering after sub-surface Cl is nearly depleted. SiCl₂ emissions decay to nearly undetectable levels toward the end of the bias period. SiCl emission follows a decay profile between that of Si and SiCl₂. Relative emission intensities increased strongly with increasing bias power (self-bias voltages of -30 to -200 V_DC), but only moderately with Cl₂ pulse durations (1 to 16s). The ALE products are depleted in SiCl₂ and especially SiCl₃ relative to etching with constant Cl₂/Ar flows and continuous ICP and bias powers.

Cu/Cu hybrid bonding for 3D integration relies on nanometer level control of copper pad recess depth to create proper and reliable electrical contacts. While the recess can be achieved through chemical mechanical planarization (CMP), as feature size shrinks both Cu and dielectric polish at the same rate, resulting in planar contact pads which severely limits the process window. In this work, we report the atomic layer etch (ALE) of Cu with high selectivity to silicon oxide to supplement CMP for tight recess control across the wafer. Each cycle consists of a H₂ plasma modification step followed by directional removal of the hydrogenated layer by low energetic Ar ions. Precise ion energy control was achieved by applying a DC pulse waveform at a given voltage (energy) to the target (1µs ON time and 2µs OFF time per period). In situ ellipsometry measurements reveal self-limiting removal of material at an etch rate of ∼3–4 Å/cycle for the first ALE cycle. As the number of the ALE cycles increases, the amount of material etched per cycle decreases, which is due to the modification of the initial surface roughness condition. After modification of the initial surface, the Cu film is less sensitive to the H₂ plasma and leads to decrease in etch amount per cycle. This can be recovered by physical sputtering of the surface by energetic Ar ions or by introduction of an additional modification step prior to each ALE cycle, resulting in process linearity restoration. Real time mass spectral analysis of the plasma of each etch cycle demonstrates the majority of volatile by-product is in the form of CuH_x and shows excellent correlation with the ellipsometry trends. Additional surface characterization by in vacuo XPS and AFM will be discussed.

Copper is the primary material used for the interconnects of an integrated circuit (IC). The dual damascene process was initially developed to deposit and pattern copper due to the difficulties of using traditional dry etching techniques, but the dual damascene process is reaching its practical limits. Atomic layer etching (ALE) is emerging as a process that could help replace the dual damascene process. In this work, we examine a plasma-thermal ALE process that uses plasma modification followed by a formic acid (FA) vapor phase removal step. Ambient pressure X-ray photoelectron spectroscopy (AP-XPS) was used to examine the FA vapor exposure step in particular detail. These in operando measurements aided in proposing the surface reaction mechanism of the modified copper surface during the vapor phase removal.

In addition, a nitrogen plasma was used to develop an ALE process based on nitridation of the copper film. Previous ALE of copper has used oxidation as the first modification step—however, copper spontaneously oxidizes on ambient exposure, making it difficult to decouple the effects of the process from ambient exposure. On the other hand, copper nitride does not form spontaneously on exposure to ambient conditions, so all nitridation is a direct result of processing. Plasma nitridation of copper has been accomplished using an ICP reactor with a nitrogen plasma, and the resulting copper nitride film has been characterized by XPS. The copper nitride film has an approximate composition of CuN_0.3.