(image) (image) [ Home ]

Phenomenological Single-Particle
Modeling of Reactive Transport
in Semiconductor Processing

Chapter 5 Modeling of Low-Bias SF6 Plasma Etching of Si

Conventionally, plasma etching techniques exploit the presence of a plasma sheath which vertically accelerates ions, to enable high aspect ratio (AR) structures [131]. Nevertheless, there are many situations which do not involve high AR features, instead requiring isotropic or near-isotropic etching conditions. This is the case in, for example, optical devices such as microlenses [49] or microcavity resonators [168], for on-chip cooling microchannels [169], or for microelectromechanical systems (MEMS) membranes [170]. In these situations, plasma etching is still an invaluable tool as long as the electric field between the plasma and the wafer, i.e., the voltage bias, is kept low [31]. The plasma is then employed as a low-temperature source of etchants, leading to etch characteristics similar to those obtained by isotropic wet etching [171].

Low-bias plasma etching offers several advantages with respect to isotropic wet etching. Its results are usually more controllable, reproducible, and uniform [49], as well as leading to better surface cleanliness which is key for optical applications [168]. However, it is known that the shape of features resulting from low-bias plasma etching is not the same as those from ideal isotropic wet etching [171]. Therefore, the development of this technology requires a deep understanding of the mechanisms leading to non-ideal isotropy. This is particularly necessary for optical applications, where exact control over the shape is paramount.

To bridge this gap, this chapter presents a modeling study of low-bias plasma etching of silicon (Si) from the most established gas source: sulfur hexafluoride (SF6) [45, 172]. The underlying plasma complexity is condensed into a single particle aggregating etchants with similar properties, i.e., "neutrals" [43, 134]. This work used measurements obtained from a collaboration with experimentalist colleagues from the University of Vienna and TU Wien to evaluate the reactant flux modeling approaches from Section 2.3. Having determined the most suitable approach, the reactive transport process is interpreted with respect to the involved phenomenological model and applied to experimental data reported in the literature [49, 171]. The issue of reactor loading is addressed, and several reported trenches are reproduced with the model. Finally, a phenomenological relationship between the most critical model parameter, the Si sticking coefficient, and a measurable degree of isotropy is proposed.

Own contributions: The evaluation of the flux models in the context of topography simulation of low-bias SF6 plasma etching of Si are conducted, as well as the interpretations and analyses of these simulations. The most important analysis is the developed empirical relationship between a phenomenological parameter and experimental topographies. This work has been originally presented at the EUROSOI-ULIS 2021 conference [173] and as an invited research article [88]. The analyses stemming from the experimental collaboration has been originally published in [174] and is discussed in further detail in Chapter 6.

5.1 Low-Bias Etching of Si from SF6 Plasma

To avoid the issues arising from wet etching, a vapor-phase, plasma-less, process can be used to obtain isotropic etch characteristics. This process has been studied since the 1960s after the synthesis of xenon difluoride (XeF2) [175] which is the most studied etchant [170] even though other reactants such as fluorine gas (F2) have also been investigated [176]. Their widespread adoption has been hindered by the complexity of the involved equipment and chemicals. For instance, XeF2 reacts with water vapor and forms hydrogen fluoride (HF) [177]. The latter spontaneously etches silica (SiO2), which is a common masking material in complementary metal-oxide-semiconductor (CMOS) processing, leading to selectivity challenges. This complexity has led to vapor-phase processes being restricted to niche applications [171].

Therefore, low-bias plasma etching processes have emerged as the main CMOS-compatible alternatives to isotropic wet etching. After the investigation of the etch characteristics of several fluorinated gases, SF6 has emerged as the preferred process due to its inert handling characteristics and higher etch rates [172]. In fact, this anomalous high etch rate is an active field of research [45, 178], and it is thought that the sulfur atoms play an active role in catalyzing the etch reaction. However, this increased reaction probability and etch rate themselves play a role in limiting the achievable isotropy. A more efficient reaction likely increases the sticking coefficient \(\beta \) and, as discussed in Section 2.3.1, an ideally isotropic reaction has \(\beta \to 0^+\).

Low-bias plasma etching of SF6 then stands in a very challenging intersection of technological trade-offs. The industrially-established status of SF6 etching technology, notably with inductively couple plasma (ICP) reactors, makes it very desirable to achieve isotropic etch characteristics. Simultaneously, the selfsame qualities which have made it an established process thwart the desired etch characteristics. Nonetheless, this process has been shown to lead to structures with desired traits, even though the etching process is not perfectly isotropic. Originally, the process was introduced for Si microlens fabrication [49], where the lower isotropy of a masked etch step is mitigated by a second, maskless, etch step. More recently, this process has been refined by controlling the roughness of the final surface [179, 180], ultimately leading to high-finesse optical microcavity resonators [168].

Nonetheless, the challenge of controlling the isotropy, and thus the precise final shape of the structure, remains. Increased attention has been placed on experimentally investigating and quantifying the isotropy in a masked etch step [171, 181]. Additional insight to this open question can be brought by accurate topography simulation, which will be discussed in the subsequent sections. To obtain physically meaningful simulations, they must be compared and calibrated to relevant experimental data. The ICP reactor setups used in the experiments are shown in Tab. 5.1.

The evaluation of the reactive transport models, discussed in Section 5.2, is done with respect to experimental data obtained from collaboration partners from the University of Vienna and TU Wien, originally described by Wachter et al. in [168]. This two-step (first step with a photoresist mask, second step maskless) etching process is described in more detail in Section 6.1. The remaining analyses in this chapter are performed in comparison to single-step (with photoresist) etched profiles reported by Larsen et al. [49] and Panduranga et al. [171].

et al. 2019 [168]
et al. 2019 [171]
et al. 2005 [49]
Pressure (\(\si {\milli \torr }\)) - 30 10
Flow rate (\(\si {\sccm }\)) 100 50 200
Coil power (\(\si {\kilo \watt }\)) 2 2 3
Table power (\(\si {\watt }\)) 15 0 0
Chuck temperature (\(\si {\celsius }\)) 30 20 20
Table 5.1: Reported ICP reactor configurations of experiments to which the topography simulation is compared.