Predicting synergy in atomic layer etching

The field of plasma etching is more critical than ever to the formation of nanometer-sized features in advanced chip technology. The structures going into our smartphones and other connected devices have become so small that they require atomic-scale processing. Atomic layer etching (ALE) is a multistep process for removing ultrathin layers of material and is used today in the fabrication of 10 nm logic.¹ The basic concept involves splitting the etch process into at least two separate steps: first modifying the surface and second removing the thin modified layer. The cycle is repeated until the desired depth is reached. Previously, we published an overview of ALE focused on Cl/Ar etching of Si.² Here, we expand our study to other materials relevant to the semiconductor industry: Ge, C, W, GaN, and SiO₂. We focus on directional (anisotropic) ALE, which is needed for a majority of critical applications. Plasma is important, as the radicals it produces are more reactive and thereby accelerate modification. Plasma is equally important in the removal step as the directional ion bombardment enables anisotropy and low-temperature processing. We study ALE in different materials to understand fundamental mechanisms.

The history of ALE is closely associated with atomic layer deposition (ALD), which has been studied since the 1960s and used in production for well over a decade.^3,4 The study of ALE originated in the late 1980s apparently as an extension of ALD, given that much of the early literature involved researchers with ALD backgrounds.⁵ The contributions and influence from ALD are still apparent; for example, a new ALE scheme using thermal reactions was discovered in 2014 during an ALD experiment.⁶ 

In parallel, our understanding of ALE is also deeply rooted in the field of plasma etching. In the late 1970s, plasma etching found widespread use in manufacturing semiconductor devices.⁷ The basic mechanism became better understood with the “ion-neutral synergy” concept credited to the 1979 article by Coburn and Winters.⁸ They showed that etching requires cooperative interactions at the surface by two different species: ions and reactive neutrals. In their famous experiment, they found that Si etching was not significant when exposed to only an energetic Ar ion beam or to only a reactive neutral (e.g., either XeF₂ or Cl₂ gas). It was with exposure to ions and reactants in combination that produced an enhanced etch rate. They conclude that plasma etching is synergistic because it “cannot be easily explained by simply superimposing a physical sputtering process onto the chemical etching process.” Although their successors only studied the ALE field years later, we find their insights relevant to the basic ALE principles today. As such, we devote this article to the memory of Harold F. Winters.

We expand the synergy concept to ALE by generalizing surface interactions as step A and step B, corresponding to the reactive neutrals and inert ions in plasma etching. In conventional etching these steps are operated continuously and simultaneously, while in ALE they are in rapid sequence. Synergy means that favorable etching occurs due to the interaction of steps A and B. While the concept was originally intended for conventional etching, it is straightforward to apply to ALE as steps A and B are already separated in either time or space (Fig. 1). The separation helps decouple A–B interactions, control the order of surface reactions, and enable self-limiting behavior. In fact, the synergistic interaction between steps captures the underlying ALE self-limiting mechanism for why etching in step B stops after reactants from step A are consumed. It is this self-limiting behavior that provides the idealistic benefits of aspect ratio independence, uniformity, smoothness, and selectivity. We originally suggested use of an ALE synergy test in our overview article.² Here we propose to quantify ALE synergy by defining it as a percentage relative to the total amount of material removed per cycle

EPC is “etch per cycle” representing the total thickness of material removed in one cycle, typically averaged over many cycles. The values of “α” and “β” are (undesirable) contributions from the individual steps A and B, respectively, and also have units of nm/cycle. Ideally, synergy will approach 100% with no etching from either step alone. In practice, nonzero contributions from α and β may be present due to photon-induced etching,⁹ physical sputtering, step contamination, and/or competing reactions of conventional etching. To experimentally measure the quantities of α and β, steps A and B can be performed as independent processes. By way of example, consider the Si ALE case study using a chlorination step A and Ar ion bombardment step B. If the overall ALE process is found to remove 0.70 nm/cycle, but 0.03 nm/cycle etches in step A alone and 0.04 nm/cycle sputters in step B alone, then S = 90%. Synergy is a simple test that captures many important aspects of ALE behavior, and is well-suited to compare different ALE conditions or systems. Ultimately, it will need to be considered in conjunction with other metrics such as electrical performance and throughput.

Synergy can be understood in terms of the energetics of underlying microscopic surface interactions. Our analysis is based on the kinetics of the process, for which we consider the energy barriers (i.e., activation energies) of key reactions that may take place in each step. As the relevant energy barriers will generally depend on the ALE scheme, we focus on the scheme using alternating chemical modification and inert ion removal. For processes to provide high synergy, the system and conditions must meet these energy requirements

E_mod is the energy barrier for surface modification, while E_des and E_O are the energy barriers for removing the chemically modified and bulk material, respectively. The energies ε_A and ε_B are delivered to the surface in steps A and B to overcome these barriers based on chosen process conditions and tool configuration. The relation derives from the synergy concept that material removal should not take place in either step independently. In other words, chemical modification should occur without etch product desorption in step A (E_mod < ε_A < E_des), and modified material (etch product) should desorb without removal of bulk material in step B (E_des < ε_B < E_O). The requirements encompass the ALD and ALE process energy windows previously known.^2,3 In general, E_des will be different in steps A and B especially if new chemistry is introduced into step B. However, since we use inert ions in the removal step, we obtain this relation by cautiously assuming that E_des is similar in steps A and B. We also take each energy as a single rate-limiting value, knowing that in reality the surfaces and energies are not monotonic and the order of reactions may not be elementary. Thus, while more details could be added, the five energies provide a simple framework for understanding ALE synergy.

The relation can be used to predict which reactant–substrate combinations could be synergistic if the relative energies are known. Although energy barriers may be difficult to obtain, values may be available from ab initio calculations or estimated from chemical–physical tables and other published literature. The energy barrier E_mod is attributed to surface rearrangement and dissociation as the reactant adsorbs to the surface (e.g., Cl₂ onto Si). For plasma-enhanced modifications, the barrier is considered negligible as plasma does the work of dissociating reactants into radicals prior to adsorption.¹⁰ Next, the energy barrier E_des is attributed to surface rearrangement and bonds breaking as the etch product desorbs from the surface. Values may be inferred from thermal desorption temperatures, volatility measurements, and/or heats of vaporization. In the case of Si ALE, ab initio calculations estimate E_des ∼ 2.3 eV for thermal desorption of SiCl₂(g) occurring at 650 °C.¹¹ The use of plasma enhances desorption and enables low-temperature processing. Finally, the barrier E_O is the bulk surface binding energy, the cohesive force that opposes atoms from leaving the surface. These values are commonly estimated from heats of sublimation, typically ranging from 2 to 10 eV and with E_O = 4.7 eV for Si.¹² The ALE synergy then depends on choosing conditions for ε_A and ε_B preferably within the energy windows. The amount of energy delivered depends on the energy-transfer method. For example, for thermal energy, the Arrhenius equation can be used to determine substrate temperatures that modulate ε_A within an energy window. When using plasma, the energy may come from photons, electrons, and/or ion irradiation energy. For accelerated ions, energy is primarily transferred via collisions with the surface and yields increase with the square root of ion energy above a threshold.¹³ Appreciable yields require ion energy about an order of magnitude higher than the surface binding energy due to the significant energy lost during the collision cascade.¹² The remainder of this article is devoted to synergy analysis of the six systems studied, in particular, to the trends with the surface binding energy.

We studied various materials comprising semiconductors, metals, and dielectrics each of which is relevant to the microelectronics industry: Si, Ge, C, W, and the binary materials GaN, and SiO₂. Reactants contained either halogens and/or carbon, chosen based on expected synergy with the material under study. Wafers were processed using a Lam Research Kiyo^® Series etch system equipped with Advanced Mixed Mode Pulsing technology for repeatable process step switching. Before entering the chamber, the Si and Ge samples were dipped in dilute hydrofluoric acid (HF) solution to remove native oxide. The other materials were directly placed into the chamber. EPC was determined using ellipsometry (Woollam M-2000 or KLA-Tencor ASET F5x) to measure the total amount of material etched divided by the number of cycles (typically ∼100). Atomic force microscopy (AFM, Veeco Vx310) was used to measure surface roughness. Values for α and β were determined by running the cyclic process with only step A or B, respectively. All experiments were conducted at a substrate temperature of 60 °C. Step A used no applied bias. In step B, Ar ions were accelerated by applying bias voltage in the range 20–100 V, which gives the ion energy. Synergy measurements were made at roughly the center of the ion energy window based on an ion energy scan. Figure 2 summarizes the systems studied and synergy analysis.

Silicon is the prototypical ALE system studied since 1990.¹⁴ It has a moderate surface binding energy of E_O = 4.7 eV.¹² The system analyzed here is the ALE case study using alternating Cl₂ plasma and Ar ions, with presumed etch products of SiCl₂(g) and/or SiCl₄(g). We previously reported ALE behavior characterized by uniform etching across the wafer within 0.4 nm (σ), surface smoothness within 0.1 nm RMS, and an energy window at 40–60 eV.² At 50 eV, we measured S = 90% [Fig. 2(a)]. For comparison, nearly perfect synergy was observed in laboratory conditions by Park et al. in 2005.¹⁵ This shows that ideal synergy is possible for this system in a laboratory setting, although at the expense of cycle time which was rather long ∼10 min. With plasma in the chlorination step and saturation time <1 s, our synergy value of 90% has 3–4 orders of magnitude improved productivity rate while still maintaining ALE behavior.¹⁶ We use Si ALE as our benchmark for the following material systems.

Germanium is a candidate to replace the Si transistor channel due to its superior hole mobility. Ge is a group IV element similar to Si, however with surface binding energy ∼20% weaker at E_O = 3.8 eV.¹² Also, Ge is more reactive to chlorine as GeCl₂(g) thermally desorbs at ∼350 °C versus SiCl₂(g) at 650 °C.¹⁷ In terms of previous ALE reports, an isotropic Ge ALE in 1997 used thermal desorption.¹⁸ Also in 1997, a directional Ge ALE was reported using Cl₂ gas (without plasma) followed by Ar ions.¹⁹ More recently, an isotropic wet Ge ALE approach was offered by Dorp et al. using oxidation with removal in wet HF solution.²⁰ We did not find previous Ge ALE studies using chlorine plasma. Here, we report on Ge ALE using alternating Cl₂ plasma and Ar ions.²¹ The presumed etch products are GeCl₂(g) and/or GeCl₄(g). The Ge ALE window was observed at 20–30 eV ion energy, which is lower in energy than the 40–60 eV for Si ALE, consistent with the lower activation energies involved. Also, there was significant background etching with Cl₂ plasma. Reducing power in step A resulted in α ∼ 0.2 nm/cycle, which still gives ∼3× more background etching of Ge compared to Si (α = 0.03 nm/cycle). At 25 eV in the Ge ALE window, we measured S = 66% for EPC = 0.80 nm/cycle [Fig. 2(b)]. Promising ALE behavior was observed. For example, AFM measurements indicated smoothness is maintained to within 0.1 nm RMS even after 70 nm of etching. Also, by inserting an oxide passivation step to manage the background etching of sidewalls, patterned test wafers of Ge show smooth etch front and aspect ratio independence of the etch depth for trench widths of 55 and 77 nm (Fig. 3). Overall, Ge ALE can be achieved at reduced operation energies but is highly prone to “background” etching in the chlorination step making it more challenging to achieve high synergy than the Si ALE benchmark.

Continuing in the group IV column, amorphous carbon has various ratios of sp³-, sp²-, and sp-bonding depending on the fabrication method and conditions. The surface binding energy is approximated as E_O = 7.4 eV based on graphitic carbon.¹² Amorphous C films are used as hard masks due to their relative resistance to chemical etching compared to photoresists. In terms of ALE history, diamond was the subject of Yoder's first ALE report in 1988 using alternating oxygen adsorbed from an NO₂(g) source followed by ion-induced removal.⁵ Another relevant ALE study in 2013 was based on C-containing polymer material using cycles of O₂(g) and Ar ions.²² The presumed etch products are CO(g) and CO₂(g). In our study on amorphous C ALE, we used alternating O₂ plasma and Ar ions. For C ALE at 50 eV, we measured S = 97% with EPC = 0.31 nm/cycle [Fig. 2(c)]. The carbon was exposed to O₂ plasma to form a CO_x layer. The use of low plasma power in the O₂ plasma prevented spontaneous etching (α = 0). Sputtering was minimal at 50 eV with only β = 0.01 nm/cycle at 50 eV roughly in the middle of the synergy process window. Figure 4 shows the characteristic ALE window at 35–75 eV ion energy where the C ALE process is self-limiting and has correspondingly high synergy. Also, AFM measurements indicated smooth surfaces as the initial value improved from 0.4 nm RMS (before ALE) to 0.3 nm RMS (after ALE) even after 50 nm of etching. Overall, this C ALE system more readily achieves higher synergy than the Si ALE benchmark; it is the closest to ideal material system studied here.

Tungsten is a metal used for making electrical contact at the transistor level, as well as for electrical interconnects in DRAM and advanced memory devices. Tungsten is notable among the elements for its strong surface binding energy (E_O = 8.9 eV),¹² over twice as strong as that of Si. Few reports exist for metal ALE, and in particular we could not find any reports for directional ALE of W. Here we report on W ALE using alternating Cl₂ plasma and Ar ions. The presumed etch product is WCl₆(g). For W ALE at 60 eV, we measured S = 95% [Fig. 2(d)]. Tungsten had no detectible background etching with Cl₂ plasma such that α = 0 and sputtering was measured as only β = 0.01 nm/cycle. Figure 5 shows self-limiting behavior for both steps A and B as the etch amount saturates at EPC = 0.21 nm/cycle. As compared to the Si ALE benchmark, W ALE required higher ion energy and a longer chlorination time of ∼2 s in order to saturate versus ∼0.5 s. Overall, W ALE has been realized, and its synergy is more ideal than the benchmark Si ALE system.

Gallium nitride is used in high electron mobility transistors for RF and high frequency communications devices. GaN is a binary III–V compound semiconductor with a high surface binding energy of E_O ∼8.6 eV.²³ ALE of GaN and other III–V materials was first studied in 1999 with the “digital etching” isotropic approach, in which the surface was oxidized and then removed with a dilute acid bath.²⁴ In our directional GaN ALE study, we used alternating Cl₂ plasma and Ar ions.²⁵ The presumed etch products are GaCl₃(g) and N₂(g). The observed ALE window at ∼50–90 eV is at relatively high ion energies compared to Si and Ge. For GaN ALE, at 70 eV (roughly the center of the window) we measured S = 91% with an EPC = 0.33 nm/cycle [Fig. 2(e)]. No background etching was detected with exposure to Cl₂ plasma such that α = 0. Minor sputtering was observed at 70 eV (β = 0.03 nm/cycle). AFM measurements confirmed smooth surfaces as the initial value improved from 0.8 nm RMS (before ALE) to 0.6 nm RMS (after ALE). Also consistent with an ALE behavior, uniformity of the thickness of material removed across the 200 mm wafer had low variation of 0.4 nm (σ). Further, XPS data confirms stoichiometry did not change after ALE. This is in agreement with previous studies on binary ALE materials.²⁶ Maintaining stoichiometry indicates self-limiting behavior as the chemically modified layer is removed in each cycle. Overall, synergy in GaN ALE was more readily achieved than the Si ALE benchmark.

SiO₂ is widely used as an electrical insulator in transistor, interconnect, and advanced memory applications. SiO₂ has a moderate surface binding energy of E_O ∼ 5 eV.²⁷ SiO₂ ALE has been reported with a variety of methods as described in our previous article.² The approach most thoroughly studied is the directional ALE using alternating fluorocarbon plasma and Ar ions. The presumed etch products are SiF₄(g), CO(g), CO₂(g). The concept of using fluorocarbons in SiO₂ ALE goes back simulations in 2007 (Refs. 28 and 29) and experimental validation in 2014.³⁰ This type of ALE has since moved to production for logic devices at the 10 nm technology node.³¹ We analyze a version of the SiO₂ ALE using hydrofluorocarbon polymer deposition on blanket wafers, alternating between CHF₃ plasma and Ar ions. For SiO₂ ALE at 50 eV, we measured S = ∼80% based on EPC = 0.5 nm/cycle [Fig. 2(f)]. The presence of deposition makes this ALE case more complicated than the other systems in this study. Since step A uses polymer deposition, we utilized an O₂ plasma after the tests to remove residual polymer in order to measure actual SiO₂ loss. The ellipsometry of the fluorocarbon deposition step indicated α ∼ 0.02 nm/cycle. We attribute this to surface modification during the initial deposition, rather than to etching, as the amount did not appreciably change with increasing cycles. This is consistent with previous understanding that while the fluorocarbon deposition in step A does not saturate, the surface modification and removal step can be synergistic and self-limiting. Overall, the results show reasonably high synergy performance in the etching behavior of this type of process.

In this section, we compare ALE results for the six materials, which were studied with comparable methodology and reactor configuration. Process conditions were similar apart from the reported adjustments to measure at the center of the window. The materials are characterized by their surface binding energy, E_O, found independently in literature from heats of sublimation and/or vaporization measurements, as referenced in Sec. IV. Figure 6(a) plots the upper edge of the ALE window. This value is the maximum ion energy in step B which provides a self-limiting etch with minimal sputtering, and is therefore related to the sputtering threshold for the bulk material. The correlation between this energy and E_O is consistent with previous sputtering reports, in which the sputtering yield is inversely proportional to the surface binding energy.^32,33

The ALE synergy is plotted in Fig. 6(b) showing correlation to the bulk surface binding energy. This trend is explained by the energy relation in Sec. II, in which E_O is the upper bound of the overall ALE window. Larger E_O values are thought to relax the energy requirements and widen the process window to accommodate more reactant choices. Although synergy could be optimized further, the energy relation and experimental data both suggest that materials with large E_O have an inherent advantage in this ALE scheme. This means that although Si is historically the case study ALE material, other materials such as C and W exhibit behavior closer to the ideal. Further support of the importance of surface binding energy is given in Fig. 6(c), which plots “net” EPC (i.e., α and β subtracted from EPC) to estimate the reactive layer thickness. We previously reported that plasma-enhanced Si ALE has an EPC value of a few monolayers, attributed to a mixed chlorinated layer that forms in the presence of plasma.² The correlation in Fig. 6(c) suggests thickness of this reactive layer depends on the material itself. An explanation is that stronger materials have thinner reactive layers because their bonds resist penetration by chemical species even with plasma.

Overall, the trends in Fig. 6 lead us to propose that other materials with high surface binding energies may also be amenable to this ALE approach. Other candidates include refractory metals such as Ta (E_O = 8.1 eV), Re (E_O = 8.0 eV), and Nb (E_O = 7.5 eV)¹² provided suitable reactants and process conditions are available. In contrast, softer metals such as Mn (E_O = 2.9 eV) or Ag (E_O = 3.0 eV) may uncontrollably etch making synergy more difficult to achieve in directional ALE. While the ALE synergy concept may be universal, the relevant energy barriers and scaling laws may differ especially if new chemistry is introduced. Therefore, the methodology used here could be applied to other ALE schemes with careful consideration of the relevant energy barriers. We are pleased to find there are also efforts in ALD to measure relative energy barriers,³⁴ highlighting the continued connection between the fields. Altogether, the insights will be vital for designing successful ALE material systems in the future.

Conclusions based on the directional, plasma atomic layer etching experiments and analysis reported in this paper can be summarized as follows:

1. ALE synergy is a quantifiable metric of ideality

2. ALE is applicable to a wide range of materials each showing unique behavior

3. Synergy relates to the underlying microscopic mechanisms through an energy criteria involving energy barriers of the system and energies delivered by the process.

4. Materials with strong bulk surface binding energies are most likely to be amenable to the directional ALE approach

We are grateful for the work of Harold F. Winters and his colleagues dating back more than 40 years in helping us understand the basics of plasma etching that have led us now to the advantages of ALE.

The authors would like to thank Steve Grantham from Lam Research; and Matthew A. Jacobs and Eran Rabani for their contributions.