In recent years, the critical dimension of semiconductor device has been reduced by a few nanometers, and the aspect ratio of patterns has been increased [1–4]. Accordingly, the need for technologies with atomic-level control in semiconductor manufacturing processes has grown as well. In response to the demands of current and next semiconductor industry, atomic layer etching (ALE), which enables a precise control without damaging the bottom layer, has emerged.

ALE is a cyclic process that repeats a modification step for surface reactions and a removal step to remove the surface reaction layer [5]. As shown in Fig. 1(a), in the modification step, the substrate surface is chemically modified by the reaction between the reactive gas and the substrate. In the next (removal) step, only the modification layer is removed using a specific energy corresponding to the bonding energy between the reaction layer and the substrate. Under these specific energy conditions, a ‘self-limiting’ characteristic appears in which the removal thickness saturates to a monolayer with increasing time, and this energy region is called the ‘ALE window’ [5].

Figure 1. (a) Schematics of cycles in the ALE process. (b) Plasma reactions and plasma parameters in the modification and removal steps in the ALE process using Ar/C₄F₆. Reproduced with permission from [4], Copyright 2021, American Institute of Physics.

Methods for applying specific energy to the substrate in the ALE process include the thermal method [6,7], electron beams [8,9], laser beams [10], ion beams [11], neutral beams [12], and plasmas [13–16]. Among them, a plasma has a fast reaction rate [5] and can be easily applied to existing semiconductor equipment without additional equipment [4]. ALE studies using various plasma sources such as capacitively coupled plasma [17], helicon plasma [18], inductively coupled plasma (ICP) [4,14,17,19,20], and electron cyclotron resonance plasma [21,22]. Among these sources, ICP is widely used as a source for ALE because of its low plasma potential and low electron temperature. Besides, the ion energy and electron density of an ICP can be independently controlled using an additional bias system [23,24]. Chlorine [18,25,26] or fluoroc-arbon-based gases [4,14,27,28] for the ALE of silicon-based materials are generally used as reactive gases during the modification step. In ALE with high-aspect-ratio (HAR) patterns, polymer-rich etching gases, such as C₄F₈ [13,28,29] and C₄F₆ [4] are used to protect the sidewalls. Fluorocarbon gases with high C concentrations dissociate in the plasma to form carbon radicals [30,31]. The polymer film deposited on the sidewall of the HAR pattern by these carbon radicals prevents distortion of the sidewall etch profile owing to ion bombardment during etching [32,33]. Compared with C₄F₈, C₄F₆ has a lower global warming potential and excellent etch selectivity [30], it can be a candidate for next-generation etching gases. In the removal step, compared to the modification step, the modified layer is removed by applying an inert gas and specific ion energy. Therefore, the external conditions such as gas species and bias voltage are different for each step, and plasma characteristics also change accordingly. Thus, the plasma parameters considered for each step are different, as shown in Fig. 1(b).

Optimizing the ALE process requires a comprehensive understanding of plasma characteristics not only in each step, but throughout the entire cycle. Plasma parameters such as electron density, electron temperature, and ion energy depend on external parameters such as ICP power, bias power, and type of gas, and process results depend on these plasma parameters. Therefore, process optimization based on the plasma characteristics enables precise process control. Although there have been many reviews of the ALE process using plasma, the plasma characteristics and process results of ALE using a polymer-rich etching gas have not been actively reviewed. In this paper, we introduce and discuss the characteristics of radiofrequency (RF)-biased ICP and C₄F₆ plasma, and the ALE process.

2.1. Considerations for the entire ALE process

In the modification step, fluorocarbon gases are widely used as reactive gases to protect the sidewalls of HAR patterns during etching. Fluorocarbon-based plasma is an electronegative plasma that forms negative ions in the plasma [34] and has different plasma characteristics compared to electropositive plasmas, such as Ar plasma.

The discharge mode of ICP can be divided into the E mode (low electron density) by capacitive coupling and the H mode (high electron density) by inductive coupling [35]. The H mode of ICP is more suitable as a discharge mode for ALE because it is easier to control low ion energy owing to its lower plasma potential than the E mode and has a higher reaction rate owing to its higher electron density than the E mode [1]. The discharge mode of ICP depends on external parameters such as the ICP power, gas species, gas pressure, and chamber geometry [36,37]. As the injection and purging of the reactive gas are repeated in the ALE process, the ICP power that satisfies the H mode should be selected in the plasma with and without the reactive gas. Figure 2 shows the electron density measured by the cut-off probe [38] with increasing ICP power for pure Ar and the Ar/C₄F₆ mixture at 10 mTorr. In both plasmas, a transition from E mode to H mode was observed as the ICP power increased. The electron density of the Ar/C₄F₆ mixture was lower than that of pure Ar over the entire ICP power region. This is a result of the collision energy loss for electronion pair creation by the many inelastic reactions of C₄F₆ [39,40]. The Ar/C₄F₆ mixture not only had a lower electron density than Ar, but also required higher ICP power to maintain the H mode. Skin depth is a criterion for understanding the discharge mode of ICP and must be shorter than $\frac{2}{\sqrt{3}}\left(\frac{\omega }{{\nu }_m}\right)l$ (where l, v_m, and ω are the length of the chamber, the electron–neutral collision frequency, and driving frequency, respectively) [37] to maintain a stable H mode. Because the electron– neutral collision frequency of C₄F₆ [41] is higher than that of Ar [42], the skin depth of C₄F₆ is shorter than that of Ar. Furthermore, because the skin depth is inversely proportional to electron density, more ICP power is required to maintain the H mode [4]. Therefore, the ICP power for ALE must consider the E-H mode characteristics of the reactive gas, and the ICP power that can maintain the H mode when both the reactive gas is injected and purged must be selected.

Figure 2. (a) Electron density with increasing ICP power in a pure Ar and Ar/C₄F₆ mixture at 10 mTorr. (b) IFEDF with decreasing gap distance in Ar plasma at 10 mTorr. Reproduced with permission from [4], Copyright 2021, American Institute of Physics.

Chamber geometry is an external factor that affects plasma parameters such as electron density, electron temperature, and ion energy. The gap between the ICP antenna and the electrode can change the ion energy. Additionally, the ALE process is sensitive to ion energy, and the gap between the antenna and the electrode must be considered. Figure 2(b) shows the ion flux energy distribution function (IFEDF) for the antenna-electrode gap in the Ar plasma at 10 mTorr. The peak ion energy increased as the gap distance decreased. As the gap distance decreases, the effective plasma size decreases, increasing the electron temperature from the particle balance equation given by $\frac{K_{iz}\left(T_e\right)}{u_B\left(T_e\right)}=\frac{1}{n_g{d}_{eff}}$ [35], where $K_{iz}\left(T_e\right)$, $n_g$, and $d_{eff}$ are the rate constant of ionization, neutral number density, and effective plasma size, respectively. In ICP with Ar, the floating wall potential can be expressed as $V_p-V_f=5.2T_e$, where $V_p$ is the plasma potential, $V_f$ is the floating potential, and $T_e$ is the electron temperature [35]. Thus, the ion energy increased with increasing electron temperature as the gap distance decreased in the ICP without additional bias. Therefore, to precisely control the ion energy during ALE, the chamber geometry must be considered.

2.2. Modification and removal steps

Even with ICP power, where the H mode is maintained, instability occurs in fluorocarbon-based plasmas owing to negative ions and dust [43,44]. Figure 3(a) illustrates the electron density with increasing ICP power and the instability region of each E and H mode in the Ar/C₄F₆ mixture. The discharge mode transition occured at 335 W, and, as shown in the pictures observed through the chamber window [4], the discharge volume repeated contraction and expandsion over time in the ICP power range of 335−600 W even in H mode. However, above 600 W, the discharge volume remained stable over time. These periodic instabilities can also be observed as oscillations of emission light intensity and fluctuations in plasma parameters, such as electron and ion densities [45–47]. This instability is a phenomenon caused by negative ions and dust in electronegative plasma [45], and it causes changes in plasma parameters and can reduce process reproducibility. Therefore, the ICP power of the H mode, where instability does not occur, as well as the ICP power for ALE must be considered.

Figure 3. (a) Electron density with increasing ICP power in a Ar/C₄F₆ mixture. (b) Residence time with increasing total flow rate for 10 and 50 mTorr. Reproduced with permission from [4], Copyright 2021, American Institute of Physics.

During the modification step, molecular gases, such as chlorine and fluorocarbon, are injected for surface reaction of substrate. Subsequently, they are purged to proceed to the removal step. A sufficient purging time is required because insufficient residual gas purging can affect the next removal step. When selecting the purging time of reactive gas, the gas residence time can be a criterion and is given by, $\tau =\frac{pV}{Q}$in steady state [48,49] where p, V, and Q are the pressure, chamber volume, and total gas flow rate, respectively. Figure 3(b) shows the gas residence time as a function of the total flow rate at 10 and 50 mTorr. The gas residence time decreased with decreasing pressure and increasing total flow rate. Because process time is related to process yield, shorter purging times are more efficient in the ALE process.

After the modification step, the reacted surface layer is removed via ion bombardment in the removal step. The ion bombardment energy can be controlled by applying an additional RF bias to the electrode. Figure 4(a) shows the IFEDF with an increasing RF bias power of 12.56 MHz in Ar plasma. As the bias power increased, the peak ion energy of the IFEDF increased, and the shape of the IFEDF also changed from unimodal to bimodal. These distributions are determined by the RF frequency, ion mass, and ion transit time in the sheath [35]. A sharp and unimodal distribution is required for precise ion energy control during the ALE process. In addition, when selecting the bias power for the ALE, the sputtering threshold energy of the material to be etched must be considered. The sputtering threshold energy for silicon is below 40 eV [50,51], and that for silicon oxide and silicon nitride is below 50 eV [52–54]. As shown in Fig. 4(b), when the bias power is applied in the removal step, the H mode must still be maintained, and a region where the ion energy and electron density can be independently controlled must be selected.

Figure 4. (a) IFEDF with increasing RF bias power in Ar plasma at 50 mTorr. (b) Electron density with and without RF bias in Ar plasma at 50 mTorr. Reproduced with permission from [4], Copyright 2021, American Institute of Physics.

Figure 5 shows the ALE results of the patterned wafer consisting of an amorphous carbon (a-C) mask/SiO₂/Si performed using the process recipe set based on the plasma characteristics of the Ar/C₄F₆ mixture. The ALE recipe includes the purging time considering the gas residence time, the ICP power of the H mode without plasma instability by C₄F₆, and bias power less than the sputtering threshold energy of silicon dioxide. The ALE process time depends on process conditions such as ICP power, pressure, and gas flow rate and is typically 40–300 s per cycle [13–17,28,55]. As shown in Fig. 5, 91 s were required per cycle: modification for 1 s, gas purging with ICP off for 30 s, ICP on for 10 s, and removal for 50 s in one cycle. The etch per cycle (EPC) was 0.60 nm/cycle, and the EPC was constant even when the number of cycles increased, as shown in Figs. 5(c) and 5(d). Compared with the continuous etching in Fig. 5(b), the ALE results in Figs. 5(c) and 5(d) show etch profiles with vertical sidewalls and flat bottoms.

Figure 5. Field emission scanning electron microscopy images of the patterned wafers (a) before etching, (b) after continuous etching, (c) after ALE for 30 cycles, and (d) after ALE for 100 cycles using an RF-biased ICP in Ar/C₄F₆. Reproduced with permission from [4], Copyright 2021, American Institute of Physics.

In this paper, the characteristics of C₄F₆ plasma and the corresponding ALE process were briefly reviewed. This review will aid in developing processes and equipment for next-generation semiconductor devices that require precise control.

This research was supported by the Material Innovation Program (Grant No. 2020M3H4A3106004) of the National Research Foundation (NRF) of Korea and was funded by the Ministry of Science and ICT and the R&D Convergence Program (Grant No. CRC-20–01-NFRI) of the National Research Council of Science and Technology (NST), Republic of Korea; Korea Evaluation Institute of Industrial Technology (Grant No. 1415181740); and Korea Research Institute of Standards and Science (Grant No. KRISS GP2023-0012-08, GP2023-0012-09).

The authors declare no conflicts of interest.