The role of plasma in plasma-enhanced atomic layer deposition of crystalline films

Atomic layer deposition (ALD) is a material growth process that is based on a pair of sequential, self-limiting, surface-mediated reactions that lead to a nominal layer-by-layer growth process for a wide range of oxides, nitrides, and metals and has become a key enabling technology in semiconductor manufacturing and a rapidly expanding number of other applications where ultrathin, conformal coatings can bring performance advantages. The process generally involves four steps: (1) pulse of the precursor (e.g., metalorganics); (2) purge of excess precursors and reaction by-products from the gas phase; (3) pulse of the reactant; and (4) purge of the reaction products and excess reactants. The reaction between the precursor and reactant forms the nominal monolayer of choice and constitutes one ALD cycle, which can be repeated to build thickness of the desired material in a linear fashion.

The method was discovered independently in the late 1960s and early 1970s (Refs. 1 and 2) and has undergone a rapid expansion in research and application over the last 15 years. This inception and its expansion have been well-reviewed in many recent papers. One such review by Ahvenniemi et al. offers a history of ALD through development of a recommended reading list on early publications of ALD.³ In this review article, the authors highlight the inception of ALD, alternatively coined atomic layer epitaxy or molecular layering, by providing a list of 22 critical articles that provide a balanced overview regarding the early history of ALD. Pertinent to the present article is an extensive review on the crystallinity of inorganic films grown by ALD in general, but with an emphasis on thermally driven processes.⁴ This thorough review covers crystallinity of inorganic ALD materials from oxides to nitrides to chalcogenides and others through very extensive and well-referenced tables in each of these material collections. It also provides a perspective on general trends regarding the effects of temperature, impurities, plasma-enhancement, substrate, and film thickness on crystallinity of grown layers.⁴ 

The vast majority of the early ALD processes were based on temperature induced reactions, processes that are now identified as thermal ALD. More recently, there has been an explosion of research effort in plasma-assisted ALD or plasma-enhanced ALD (PEALD) to expand the properties and application of thermal ALD materials and the number and types of materials beyond the material set available to thermal ALD. PEALD generally uses a plasma to create the reactant “pulse” in the ALD sequence. Relevant to the present article is a thorough review of plasma-enhanced ALD research written in 2011,⁵ which provides a brief review of plasma basics, brief descriptions of PEALD reactor configurations, advantages and challenges of PEALD processes, and a table highlighting and referencing the many materials grown with such processes—from metals to oxides and nitrides. This subject area has very recently been updated in a review by Knoops et al.,⁶ highlighting advances since the 2011 review in the area.

The plasma sources employed in PEALD come in many types from relatively low electron density, capacitively coupled RF to high electron density, inductively coupled RF and hollow cathode (HC) plasmas. These different plasma sources vary significantly in terms of the total and relative densities of species they produce, and, as a result, they can provide very different flux of reactant species to the surface. The use of plasmas in ALD processes presents both advantages, such as lower growth temperatures and more dense films, and challenges like the potential loss of conformality and damage. These processes have been well-described in the aforementioned review by Profjit et al.⁵ 

Knoops et al. updated the status and prospects of PEALD in 2019 by focusing on research published between 2011 and 2018.⁶ In this review, it is noted that the PEALD process has obtained a prominent position in the field of ALD through a rapid increase in the number of processes reported, including high-volume manufacturing and a growing number of PEALD reactors from which to choose. Most useful is their Figure 1, and related Table III, which presents a collection of periodic tables depicting the vast number of pure elements (mostly metals) and binary (and by digital alloy extension, ternary and quaternary) oxides, nitrides, sulfides, fluorides, and phosphides demonstrated through PEALD. The many challenges for PEALD are also discussed including complexity of the process that involves not only radicals and ions but also energetic electrons and photons; plasma/surface interactions including damage and redeposition; and conformal deposition.

Although much of the early work in ALD was focused on amorphous materials, there has been a history of, and a rapidly increasing interest in, the growth of crystalline materials including oxides (TiO₂, ZrO₂, Ga₂O₃, etc.) and nitrides (TiN, GaN, AlN, InN, NbN, etc.). Crystalline materials offer better chemical and electrical properties for many applications. A very thorough review of crystallinity of inorganic films grown by ALD, predominantly thermal ALD, was published in 2013.⁴ 

The interest in the growth of crystalline materials has also extended to PEALD. This is not surprising as the energy available from the plasma in the form of reactive radicals, charged particles (ions, anions, and electrons), as well as photons can promote surface diffusion and better bonding at the surface of the growing film. This surge of research in PEALD of crystalline materials is evident in Fig. 1.

In the present paper, we attempt to increase awareness of the many aspects of plasmas that need to be considered in controlling crystallinity, both overall and for phase selection and purity, of films grown by PEALD. We start with a review of basic plasma physics focusing first on the generation and transport of energetic and reactive species in the plasma and then on the plasma-surface interactions that most directly impact film qualities. We highlight key considerations from these physicochemical processes that can markedly influence the PEALD processes and key diagnostics that can provide real-time insights into the plasma properties that influence the growth process. We present both current experimental data and a review of previously published works to describe how variations in the approach to plasma generation and the interactions between plasma-produced species with the growth surface influence the plasma reactant step in PEALD processes. We then highlight two case studies to demonstrate these relationships: (1) controlled phase purity of crystalline titanium dioxide (TiO₂) films and (2) crystalline growth of semiconducting indium nitride (InN). We conclude by summarizing these observations and offering a perspective on future prospects of PEALD for crystalline material growth.

The inclusion of plasma in ALD processes generally offers the benefit of substantially reduced growth temperatures and greater flexibility in tailoring the gas-phase chemistry to produce specific film characteristics. The benefits plasmas provide, however, come at the cost of a complex array of process variables that often challenge the ability to predict, a priori, the influence of any one input parameter. In this work, we attempt to provide some clarity as to how plasmas are formed and controlled, and how they can most optimally be employed within the framework of atomic layer deposition. To begin, we cover some of the fundamentals of plasma generation along with the production of energetic and reactive species and their transport within the plasma. We then focus on how different plasma generation schemes and geometries, including capacitively coupled plasmas (CCPs), inductively coupled plasmas (ICPs), and RF-HC plasmas, differ in their production of energetic and reactive species. Finally, we address the plasma-surface interactions that are critical for film growth and control of crystallinity. The details of low temperature plasmas used in the synthesis and processing of materials has been the subject of many textbooks,^7–9 and as such, we will limit this discussion to a cursory treatment of plasma phenomena. Interested readers are directed to the aforementioned texts for more details.

Plasma is an ionized gas consisting of roughly equal numbers of negatively and positively charged species—generally electrons and ions. The class of plasmas most common to ALD systems is weakly ionized, low temperature, nonequilibrium discharges. The term describes three critical aspects of the plasma state. Here, weakly ionized indicates charged-to-neutral particle ratio is low, typically <10⁻³. Low temperature refers to the fact that, compared to astrophysical and fusion plasmas, the electron and ion temperatures in these systems are quite modest (0.1–10 eV for low temperature plasmas, compared to 10²–10⁴ eV for astrophysical and fusion plasmas). Finally, the term nonequilibrium indicates that the population of electrons within plasma are much more energetic than the population of heavy particles (ions and neutrals). In fact, the average energy of the electrons within the plasma production volume is above 1 eV, representing an effective temperature greater than 10000K. The ions and neutrals are comparatively much cooler, with a temperature generally less than 1000 K.

Plasma discharges rely on externally imposed electric fields that serve to energize the population of electrons. This energy absorption is balanced against energy loss processes within the bulk plasma such as electron-neutral collisions, electron-electron collisions, electron-ion collisions, and wall losses to sustain the plasma. The interplay between the electric field imparting energy to plasma electrons and inelastic processes removing energy from plasma electrons gives rise to a population of energetic electrons described by the electron energy distribution function (EEDF). The breadth of the EEDF is increased due to the rapid response of electrons to the imposed electric field. This energy transfer between the electric field and electrons is efficient due to the fact that the fraction of kinetic energy lost in collisions between the electrons and the more numerous heavy particles in the system (atoms or molecules) is very small (∼10⁻⁵)⁸ due to the large mass difference between these species. Moreover, at low pressures (<10 Torr), electron collisions are infrequent compared to the time scale on which electrons gain energy from the electric field. Consequently, the electrons within the plasma are generally not in thermal equilibrium with the heavy particles in the system.

The EEDF, or f(E), can be thought of as a function describing the fraction of the total electron density, n_e, with a particular energy, E + δE, where $ne=∫0∞f(E)dE$⁠. The mean electron energy, 〈E〉, and electron temperature, T_e, associated with the EEDF are of particular importance and can be defined as

The shape of the EEDF in energy space and the number density of electrons in the plasma are the two principal drivers for the generation of all the energy-carrying and reactive species within the plasma source. As such, the EEDF and its shape are of critical importance to optimizing the reactant pulse in a plasma-enhanced ALD process. Figure 2 illustrates a typical EEDF in a low temperature plasma along with the cross sections relevant to reactive and energetic species generation within the plasma. The reaction rate, k_x producing the reactive or energetic species of interest is proportional to the product of the EEDF, f(E), the density of the heavy particle reactant, n_A, with the electron collision cross section, σ_x(E), a measure of the probability that the species of interest will be created during a collision of an electron with a heavy particle. The production rate is written as follows where only the energy of the electrons is considered important since they are generally far more energetic than the heavy particle species:

As can be seen from Eq. (2) and Fig. 2, the production of species within a plasma is also highly dependent on the shape of electron impact cross sections. While it is beyond the scope of this manuscript to detail the quantum mechanical principles behind the shapes of various cross sections, the following resources are suggested to the interested reader.^7,10–14

For most purposes, the state of a discharge plasma can be characterized by the densities of heavy particles N_j, where j corresponds to the jth species, the electron density n_e and the electron energy distribution function f_e(E).¹⁵ A final constraint, so-called quasineutrality, is also important for describing the plasma state. Quasineutrality implies that the densities of positive, n⁺, and negative, n⁻, charged particles are approximately equal throughout the bulk of the plasma, n⁻ ≈ n⁺. For plasmas of industrial relevance, positive species include ions (or cations), while the negative particles include both electrons, negative ion (or anions), and, in the case of high throughput industrial processes, charged dust particles.¹⁶ 

The aforementioned difference between the average energy of the electron population and the ion population in the plasma, in conjunction with the vast mass difference between electrons and ions, makes sustaining the plasma problematic, since the electron flux to any enclosing surface will be much greater than the ion flux. To equalize these fluxes and maintain quasineutrality, a space charge layer, known as the sheath, forms at all surfaces in contact with the plasma. Except for cases involving very high current densities to powered surfaces, the sheath will contain primarily the low-mobility ion species, and it is generally referred to as a positive space charge sheath or simply an ion-sheath.^7,8 The ion-sheath gives rise to a global potential difference between the bulk plasma and surfaces enclosing the plasma, which is referred to as the plasma potential, V_p. V_p is positive with respect to adjacent surfaces and serves to equalize the flux of plasma electrons and ions leaving the plasma, due the large thermal velocity difference between ions and electrons. As such, the magnitude of V_p is linked to the shape of the EEDF and in particular, the high-energy tail of the EEDF. This is illustrated in Fig. 3. In essence, most electrons in the EEDF will be confined by the plasma potential except a small, high-energy fraction, energetic enough to overcome the potential of the ion-sheath. The flux at the wall of this high-energy electron subgroup is necessarily equal to the flux of ions at the wall. For a Maxwellian EEDF, the magnitude of the plasma potential relative to the wall potential (usually ground), V_w, is given by

where M is the ion mass and m_e is the electron. For plasmas produced in commonly used gas mixtures in PEALD, the mass ratio leads to the simplified expression in Eq. (3). This relation should be treated as a “rule of thumb” that is less meaningful when the EEDF is highly non-Maxwellian (see Fig. 3) or in more complex plasmas such as those containing a significant fraction of negative ions (i.e., O₂ or SF₆) or in systems operating at high neutral pressures (>10 Torr).

Figure 3 also illustrates the spatial variation of charge within the sheath, which is governed by the screening of space charge by the electrons within the plasma. The spatial extent of the sheath is governed by a plasma property known as the Debye length, λ_D, defined as the distance over which a space charge potential is reduced by a factor of $1/e$ or, $λD=ϵ0Tenee$⁠, where ɛ₀ is the permittivity of free space and e is the electronic charge. Under typical conditions for laboratory plasmas, the sheath width is several λ_D.

Importantly, the magnitude of the plasma potential and the characteristics of the sheath govern the interactions between the bulk plasma and surfaces in contact with the plasma, since the sheath and plasma potential determine the energy of ions impacting surfaces. Consider, for example, the case of a Maxwellian discharge with a T_e = 2–5 eV (i.e., Fig. 2). According to Eq. (3), the ion energy at a grounded surface will range between 10 and 25 eV, which can have profound implications for the crystallinity of the films being grown.^17,18

In Fig. 3, the limitations of Eq. (3) are demonstrated when a Maxwellian and a non-Maxwellian EEDF are considered. For Maxwellian EEDF with a T_e = 5 eV, V_p is 25 V. In the case of a non-Maxwellian EEDF commonly observed in capacitively coupled plasmas, despite the large population of low energy electrons, leading to an overall T_e of 2 eV, V_p remains 25 V due to the effect of the high-energy tail electrons.

In addition to determining the number of ions, V_p, and the ion energy at surfaces, the shape of the EEDF also governs the generation of excited and reactive species within the plasma. These species include reactive atoms, such as N and O, as well as metastable atoms and molecules [e.g., Ar[1s₅] and $N2(A3Σu+)$]. Chemically reactive species such as atomic radicals can significantly enhance the chemical reaction rate at the growth surface leading to better film quality.^17,19,20 Metastable species are long-lived electronically excited species that can also enhance film growth by transferring their excess electronic potential energy (1–20 eV) to the growth surface via Auger processes.^21–23 The delivered energy can drive the impact site out of thermal equilibrium with the surrounding material, which can help overcome activation barriers and enhance film growth at low growth temperatures. At the same time, excitation processes driven by the EEDF also produce high-energy photons. High-energy (5–20 eV) photons have been implicated as a means of driving the reactant step of the ALD cycle;²⁴ however deleterious²⁵ effects on film properties and device function²⁶ have also been widely reported.

The densities of these important plasma species are profoundly influenced by the shape of EEDF as is illustrated in Fig. 4. In Fig. 4, typical EEDFs for an ICP and CCP at moderate pressure (≈50 mTorr) are shown.^27,28 Since ICPs are generally 1 to 2 orders of magnitude higher in electron density than CCPs, the EEDFs have been appropriately scaled to show the ICP EEDF with significantly higher density and a more Maxwellian shape. By examining the overlap between the electron impact dissociation and excitation cross sections for N₂ and Ar (shaded area beneath EEDF), one can clearly see how the differing EEDF shapes and the differing total electron density for these exemplary cases will affect the generation of excited species and radicals in these respective plasmas. Here, the ICP and CCP have nominally the same T_e and thus V_p. However, the shape of the ICP indicates significantly more radicals and excited species can be produced relative to the CCP. We will elaborate on this example further in upcoming sections of this article. Table I includes a number of the reactions driven by the electron collisions that occur as part of plasma generation. While this list is not comprehensive, it gives the reader a sense of the most important species generation processes driven by energetic electrons as well as the minimum electron energies necessary to drive these processes. Heavy particle collisions (such as charge transfer between ions and neutrals) are also an important source of energetic and reactive species relevant to PEALD; however, they will be dealt with separately in Sec. II A 4.

It is important to note that many of the reactors used for PEALD processes have geometries that employ plasma sources that are spatially separated, or remote, from the growth substrate. In these cases, there is no significant electric field heating the EEDF in the region between the plasma generation volume and the growth substrate. As such, the species that are delivered to the growth substrate can be significantly different from species generated within the plasma source. The gas pressures and chamber dimensions typical in PEALD systems allow collisions, even multiple collisions, between charged particles and neutral particles and so, one should not expect the relative concentrations of charged and reactive species within the source region of the plasma to necessarily resemble the relative concentrations of charged and reactive species that reach the growth surface. In particular, charge-exchange and recombination reactions significantly alter the relative charged particle flux as plasma diffuses from the generation volume to the growth substrate. In this section, we briefly cover some of the important reaction and loss channels that can alter species densities in the transport region between the plasma source volume and the substrate. In addition to the electron-ion recombination reactions in Table I, Table II gives an overview of many of the heavy particle reaction pathways that exist in low temperature plasmas. As was the case for Table I, Table II should not be considered a comprehensive list of possible reaction mechanisms. The interested reader is directed to the works referenced in Tables I and II for a more thorough treatment of the gas-phase chemistry in materials processing systems.

Charge transfer. Resonant (⁠$A+A+→A++A$⁠) and nonresonant (⁠$A+B+→A++B$⁠) charge transfer reactions are important phenomena in the transport of heavy particles in plasmas, particularly in plasmas that contain a mixture of noble gases and reactive gases, such as Ar/O₂ and Ar/N₂. In such mixtures, Ar⁺ is rapidly converted into molecular ions of the reactive component via charge transfer. Thus, even in cases where the reactive gas fraction is low (<10%) one can expect nearly all the ions incident at the growth surface to be molecular ions of the reactive component (N₂⁺ or O₂⁺). While this effect is obviously important in the context of plasma-surface interactions, in that energetic molecular ions will likely lead to different surface effects than noble gas ions, it also has important implications toward total ion flux since molecular ions generally undergo electron/ion recombination much more rapidly than atomic ions.

When the gas pressure is high enough, and/or plasma density low enough, charge transfer processes can also occur within the sheath region described previously. Under these conditions, where the ions impacting the neutrals have a large directional velocity component, fast neutrals can result with a velocity close to that of the original ion. That is, the ions are neutralized via charge transfer, but retain nearly their original velocity. This effect occurs when the mean free path for charge transfer reactions, λ_CT, is less than the width of the sheath. For example, at 1 Torr, λ_CT ≈ 5 mm,³¹ which is on the order of the sheath thickness if n_e < 10¹⁰ cm⁻³. While fast neutrals generated in sheaths will generally not have the full kinetic energy imparted from the sheath potential drop, they nonetheless can deliver, as would ions, a significant energy to the surface around their impact sites.

Excitation transfer. Charged particles are not the only energetic species of interest in low temperature plasmas. Neutral atoms and molecules can also carry excess energy, in the form of excited electronic states, and in the case of molecules, excited vibrational and rotational states. Long-lived electronically excited neutral species, often referred to as metastables, frequently have densities equal to or exceeding that of the ions and electrons. Metastables are atoms and molecules which are “trapped” in an excited electronic state due to quantum mechanical selection rules that prevent de-excitation in the absence of a collision with another body. As such, these atoms and molecules remain in this excited state for long periods of time (milliseconds to seconds) and often release their energy upon colliding with a surface. Since they are not charged, their transport is not impeded by electric and magnetic fields and they generally follow Fick’s law of diffusion in the same manner as ground state neutral species.

In addition to imparting their energy to surfaces, metastables can also produce ion-electron pairs and atomic neutrals through gas-phase reactions. The ionization process, known as the Penning ionization, occurs when either two metastables or one metastable and one ground state molecule collide, resulting in the ionization of one of the colliding species. Similarly, these collisions can result in the dissociation of the molecules if the stored energy within the metastable is high enough. As such, metastable species can be a source of both ions and reactive atomic neutrals in the transport region downstream from the plasma generation volume.

Particle loss. Finally, it is important to take account of how reactive and energy-carrying particles are lost in low temperature plasmas. In the case of charged particles, losses occur either at surfaces or through electron-ion recombination collisions in the gas phase. Molecular ions are lost far more rapidly to electron-ion recombination reactions than atomic ions, which can only be destroyed at surfaces due to momentum and energy transfer considerations. As such, plasma densities in molecular gas mixtures will decay much more rapidly, both spatially and temporally, than plasmas sustained in pure noble gases. Moreover, the creation of molecular ions through charge transfer reactions in mixtures of noble and molecular gases (e.g., Ar/O₂ mixtures) discussed above, combined with their loss to dissociative recombination, can often be an important source of atomic neutral species.

The loss of neutral species generally occurs at surfaces within the reactor. These loss processes are generally some type of either chemical or physical process. Since reactor walls are such an important loss mechanism for neutral species, reactor geometry and the proximity of the plasma generation volume to the substrate surface can often play a critical role in determining the flux of reactive neutral species at the growth surface. Moreover, if energetic particle flux to the walls is not carefully managed, it can be a source of contamination, as sputtered material from the walls could participate in film growth.

Particle loss at surfaces also has important implications for deposition on structured substrates with small features. This effect is particularly important in the context of reactive and energetic species loss during deposition on high aspect ratio structures. This topic has been dealt with extensively in recent work by Arts et al.³² 

It is clear from this section that there is a wide variety of chemically active and energetic species generated within a low temperature plasma that is relevant to driving surface chemistry and film growth in the context of PEALD. Considering this section’s focus on transport combined with the focus of Secs. II A 1 and II A 3 on species generation, one should expect the characteristics of the plasma source and its location relative to the substrate, coupled with chamber geometry, to have a critical effect on the relative flux of reactive and energetic species at the substrate surface. This inherent complexity makes it challenging to optimize the plasma conditions without a robust understanding of the aforementioned mechanisms of species generation and loss, which are heavily dependent on both the plasma generation method used and the chosen reactor geometry.

The manner in which power couples to the plasma can have significant impacts on the plasma density and the shape of the EEDF within a plasma source and, ultimately, the delivery species to the growth surface. The mechanisms through which electrical power couples to the plasma in RF-driven plasma sources are often quite complex and dependent on process conditions such as neutral pressure, gas composition, and RF frequency.^27,30,33 As such, it is important to understand these mechanisms to be better equipped to recognize different power coupling regimes for a given PEALD system.

In this section, we summarize how different power coupling methods including capacitive coupling of RF power and inductive coupling of RF power affect plasma production, as well as how changes in electrode geometry, such as changing from a flat plate to a hollow cathode geometry, affect plasma production. We will also mention the different power coupling modes that are observed for different operating conditions and how those power coupling modes change the production of energetic and reactive species within the plasma source.

Capacitively coupled plasmas are generally operated between 10 and 20 MHz (nominally 13.56 MHz) in many PEALD reactors, although a far broader frequency range can be found in literature if one includes other plasma processing applications such as etching and other deposition techniques.^24,31,33 At frequencies above the ion plasma frequency, nominally ∼1 MHz, only the electrons can follow the temporal variations in applied electric field.⁸ Thus, the simplest description of these systems involves an electron gas that oscillates at the applied frequency among an equal population of relatively stationary ions. As the electrons are drawn toward or repelled from a powered electrode, an increasing or decreasing volume of positive space charge forms at the opposing electrode resulting in an oscillating positive ion-sheath. This means that the magnitudes of the sheath potential drops at each electrode are 180° out of phase with each other. The electric fields within these sheaths directly heat the electron population within the CCP and sustain the EEDF. This describes ohmic heating, or the α-mode regime, which typically results in a relatively low electron density (<10¹⁰ cm⁻³) with an EEDF that can be reasonably approximated by a single electron temperature (usually 3–5 eV).

The shape of the EEDF in CCP systems can be greatly affected, however, by varying operating pressure, RF voltage, and/or RF current applied to the powered electrode, as well as driving frequency and waveform shape. These effects have been extensively detailed in the literature^34–38 and can result in dramatic operating mode transitions for CCP sources. Under certain conditions, a higher electron density mode, the so-called γ-mode, can be accessed, where the heating mechanism can be described as electron acceleration via the temporally and spatially varying electric field of the high voltage sheath on the powered electrode. This mode transition at varying neutral pressure and varying input power is shown in Fig. 5, which is reprinted from earlier work by Godyak and Piejak³⁴ and Graham et al.³⁹ Figure 5 shows measurements of the electron energy probability function (EEPF), which is determined by dividing the EEDF by $E$⁠. A great deal of work has been done over the years to understand the nuances of this phenomenon. In general, γ-mode operation requires high voltages at the powered electrode and results in substantially increased plasma density (10¹⁰–10¹¹ cm⁻³) and lower average electron energies (1–2 eV).⁴⁰ This high plasma density is largely the result of a high-energy, beamlike population of electrons produced at the powered electrode and accelerated within the oscillating RF sheath,^7,41 which enhances ionization within bulk plasma. From a practical stand-point, the transition from α-mode to γ-mode operation can be detected by monitoring light output from the plasma or the RF-induced, self-bias on the powered electrode. An abrupt increase in photon emission and/or self-bias is a reliable indicator of the transition to γ-mode operation.

In PEALD systems, CCPs are usually operated at relatively high pressures (∼1 Torr) and low power (<200 W), which places them typically in the α-mode power coupling regime.^7,34 Because the resulting plasma densities tend to be lower (∼10⁸–10⁹cm⁻³), CCP sources can be placed relatively close to the growth substrate without concerns for excessive ion flux at the surface. However, α-mode operation also means that power coupling to the plasma tends to be less efficient. Recent work by Napari et al.⁴² has shown that operating an indirect CCP in the α-mode leads to more uniform film growth than γ-mode operation, while operation in the γ-mode leads to films of higher crystalline quality.

CCPs are highly sensitive to electrode geometry. In most cases, the electrode geometry in CCPs are asymmetric, meaning the two electrodes do not have the same surface area. This is due to the fact that in nearly all practical plasma reactors, there is far more grounded surface area than there is powered electrode surface area. The consequence of this asymmetry is that the sheath voltage drop will be shared unevenly between the powered and grounded electrodes. Typically, the lower capacitance of the smaller powered electrode will force its sheath voltage drop, averaged over the RF cycle, to be much larger than the corresponding voltage drop at the larger, higher capacitance grounded surface. This has the very important effect of lowering the average ion energy at the grounded electrode relative to the powered electrode.⁷ Accordingly, CCP-based PEALD systems utilize an asymmetric electrode design and usually ground the growth surface so as to minimize the energy of ions impacting the substrate. While we do not address the topic of additional substrate biasing in this section, it should be noted that this is possible and may be advantageous in the context of crystalline film growth.

CCP-based systems used for PEALD usually fall into two categories, direct and indirect plasmas geometries. These two varieties are illustrated in Fig. 6. Direct exposure CCP systems, as the name implies, place the substrate on the grounded electrode of the CCP. This results in exposure of the growth substrate to charged particles and reactive species generated directly within the CCP volume. Alternatively, indirect exposure CCP systems utilize a semitransparent grounded electrode located between the CCP plasma volume and growth substrate. The reader is again directed to the work of Napari et al.,⁴² where a direct CCP configuration was shown to produce crystalline oxide films whereas the indirect CCP configuration resulted in amorphous film growth when operated in the α-mode.

The semitransparent ground electrode in the indirect configuration serves to isolate the growth substrate from plasma generation volume, while still allowing a significant fraction of the plasma-produced species generated within the CCP volume to interact with the growth surface. It is important to note that while the electrode does separate the substrate from the region of high electric field where plasma generation occurs, it may still permit charged particles to interact with the substrate. Charged particle transport through the electrode will depend on both the optical transparency of the electrode and geometry of the apertures in the electrode. These semitransparent electrodes are often perforated metal sheets, comprised of a pattern of holes with a uniform diameter and spacing. The optical transparency is then simply defined as the ratio of open area to total electrode area, with the former determined by multiplying the surface area of each hole by the number of holes and the latter, the area of the electrode, as though there are no holes. The hole diameter and aspect ratio are important geometric considerations, as they determine both the ability to confine the plasma and the flux of particles that pass through the electrode. If the hole diameter is too big and/or the electrode is too thin, the plasma can leak through the electrode. Practically, the hole diameter should be significantly smaller than the length of the sheath that forms at the electrode surface.⁴³ Under these conditions, the plasma behaves as though there is not a hole present. As such, ions leave the plasma and transit the hole following trajectories that are normal to the electrode surface. Neutrals, of course, also diffuse through. When these conditions are met and the electrode is thin, optical transparency is the primary consideration. When the electrode is thick and the holes have a large aspect ratio (length to diameter ratio), the flux of both charged and neutral species can be reduced via collisions with hole walls. As has been demonstrated in the literature,^17–19,42 there are situations, particularly in the context of growing crystalline films, where it is advantageous to allow energetic plasma species to interact with the substrate, and so, it is important to consider the transparency of the electrode.

The hollow cathode enhanced capacitively coupled plasma (HC-CCP), more commonly referred to simply as a hollow cathode source,^6,44 operates similar to the standard asymmetric CCP-based PEALD tools; however, the powered electrode is augmented by multiple openings through which the working gas can flow. A schematic of this configuration is shown in Fig. 7(a).

RF driven hollow cathode based plasma sources have been used for some time as tools for materials processing,^20,45 although this type of source has only been adopted within the last decade for use in PEALD. In general, hollow cathodes, both DC and RF, enable significantly higher plasma densities to be achieved for a given applied voltage, due to the so-called hollow cathode effect. While there have been less work done to understand the operation of this configuration, it well known that the hollow cathode effect is enabled by the geometry shown in Fig. 7(b), due to electron trapping. Electrons, which are accelerated in the cathode sheath and enter the bulk plasma, cannot escape once they have lost an amount of energy that is about equal to their initial ejection energy (which is only a few electron volts)^46,47 since they encounter a repulsive ion-sheath whenever they approach the cathode wall. In this configuration, electrons are only lost out the ends of the tubes and ionization is much more efficient due to the longer residence time of energetic electrons within the gas volume. In addition to this electron confinement effect, relatively high concentrations of doubly ionized species are also thought to contribute to the hollow cathode effect⁴⁸ due to their comparably high secondary electron coefficients upon collision with the cathode wall. This being the case, hollow cathode sources typically produce high plasma densities,⁴⁹ with the hollow cathode based PEALD systems reportedly attaining plasma densities of 10¹² cm⁻³ near the cathode electrode.^50,51

Due to these high densities, hollow cathode systems are generally operated with a significant standoff between the plasma generation region and substrate, ∼10s of centimeters. Because of this large standoff distance, reactive and energetic plasma species generated near the hollow cathode will undergo many collisions with background neutral species before reaching the substrate. As such, accounting for heavy particle collisions and losses are important in understanding the flux of species at the growth surface, particularly for the relatively high pressures at which these systems are typically operated (0.3–10 Torr).⁵¹ 

It is also important to note that these systems are rich in plasma physics. The combination of the hollow cathode geometry, with RF power coupling, and strong plasma density gradients creates an environment that is well suited for the generation of spatiotemporal variations in the EEDFs and the plasma potential,⁵² as well as the plasma-wave behavior.^27,53,54 Indeed, a significant portion of the hollow cathode literature is dedicated to understanding the plasma turbulence around hollow cathodes.^54–56 The presence of broadband plasma oscillations can make EEDF measurements difficult. However, DC hollow cathode EEDF measurements show a large population of a low energy electrons with a beamlike high-energy tail that sustains ionization.⁵⁷ Time-resolved optical emission studies performed by Schmidt and co-workers^58,59 on HC-CCPs illustrate how the inclusion of hollow cathode geometries increases high-energy electron population near the RF electrode by more than a factor of 4 (Fig. 8).

Simulations of hollow cathode plasmas indicate that the high-energy electrons will attain energies comparable to the cathode bias voltage.^59–61 An exemplary EEPF from an HC-CCP particle-in-cell simulation⁶² at conditions of P = 258 mTorr and 13.56 MHz RF voltage of ±400 V is shown in Fig. 9. Note that the EEPF contains a high-energy component extending to 300 eV.

Electrons in this energy regime can have a significant range in low pressure gases. As such, to avoid high-energy electron impact at the growth surface reactor, operation at pressures >1 Torr may be necessary. Under certain conditions, however, high-energy electron impact at the growth surface could be advantageous. Recent work has indicated that high-energy electrons can aid in the deposition processes through electron-stimulated desorption and dissociation of adsorbed surface species.⁶³ 

Another widely used method for generating high-density plasmas is through the use of ICPs. In these systems, the powered element is not in direct contact with the plasma but rather separated from the plasma by a dielectric material. In PEALD systems, this is often a coil wrapped around a quartz tube that also serves as a structural part of the vacuum system. Whereas CCPs couple power into the plasma through the electric field in the sheath, ICPs rely on electromotive forces to generate strong time varying electric fields capable of ionization.⁶⁴ These plasmas also operate in two distinct modes: (1) a lower efficiency, low-density mode characterized by capacitive coupling, commonly referred to as the E-mode, and (2) a higher efficiency, high-density mode characterized by inductive coupling, called the H-mode. In the H-mode, when power transfer from the amplifier to the plasma is properly matched, the system is mostly inductive in nature and power is deposited through a so-called “skin depth layer” at the boundary between a dielectric window and the plasma.⁷ The plasma density in the H-mode can be as high as 10¹³cm⁻³ and is easily distinguishable from the E-mode by increased light emission from the plasma and increased power transfer efficiency.^7,18,64,65 E-mode generally produces 1–2 orders of magnitude lower plasma density and is consequently less effective at generating reactive neutrals.⁶⁶ As such, it can be generally regarded as an undesirable operating condition when using remote plasma geometries in PEALD. The noncapacitive nature of the power transfer in the H-mode allows low voltages to be achieved at all the sheaths around electrodes and walls in these systems. Generally, the plasma potential in inductive discharges is between 10 and30 V, and the EEDF is highly Maxwellian,^7,27 as can be seen in Fig. 10. ICPs in PEALD systems are usually operated with 200–600 W of RF power leading to high plasma densities (10¹¹–10¹³ cm⁻³). Due to these conditions, most reactor designs locate these plasma sources at a significant distance (20–30 cm) from the growth substrate (see Fig. 11). However, this remote plasma configuration may not be necessary under low power (<50 W) operating conditions provided the precursor deposition on the dielectric window could be mitigated.

Also of note from Fig. 11 is the feature of RF biasing such that a negative self-bias is achieved on the substrate. The negative self-bias of the substrate serves to elevate ion energies at the surface and has been recently explored to enhance the crystallinity of films grown using ICP-based PEALD systems.^67,68 While this technique could also be easily employed in other PEALD geometries, it seems to have had the most use in ICP-based systems. Substrate biasing can be achieved by using additional RF voltage source applied to an electrically isolated stage or by manipulating the RF impedance between an isolated stage and ground.^5,25,69–71 Both approaches can achieve substantial (>50 eV) positive ion energies at the substrate, although somewhat higher ion energies and greater waveform flexibility can be achieved using an additional RF biasing system on the stage.

It is important to recognize that the application of RF power to an electrode immersed in a downstream plasma can also generate a “secondary discharge,” which will enhance the production of both charged and reactive species locally and thus elevate the flux of both at the growth surface. Abrupt changes in RF-induced self-bias or changes in power coupling will serve as a good indicator that a secondary discharge has been initiated.

Similarly to the hollow cathode source, the large standoff distance between plasma source and growth surface implies that reactive and energetic plasma species generated within the ICP will undergo many collisions with background neutral species before reaching the substrate. As such, accounting for the heavy particle collisions and loss mechanisms described in Sec. II A 4 is important for understanding the evolution of reactive and energetic species fluxes between the plasma source and the substrate. These changes in the relative flux of ionic species can be particularly important when substrate biasing is applied as the ion species flux at the growth surface may be very different from those generated within the ICP source.

In Sec. II B 3, the sources described can be divided into two categories: remote plasma sources and direct plasma sources. The ICP, the gridded CCP, and hollow cathode enhanced CCP can be considered remote sources, either due to a physical barrier between plasma generation region and substrate or a large (10s of cm) spatial standoff between the plasma generation region and substrate. CCPs utilizing a direct plasma configuration are also described above. In Fig. 12, we summarize the salient features of these plasma sources in the context of PEALD. While all of them generally act as a source of electrons, ions, reactive, neutrals, and photons, there are differences between the various sources that are important to consider.

Direct CCPs are usually operated at low power (<200 W), with high neutral pressures, such that they operate in the α-mode, and thus exhibit low electron densities (10⁸–10⁹cm⁻³). The asymmetric electrode configuration typical of these systems coupled with high pressure operation is generally sufficient to ensure relatively modest, low energy ion fluxes at the substrate.

Remote CCP configurations typically operate with relatively small standoff distances between semitransparent grid and the substrate (∼1 cm or less) and compared to direct CCPs will result in a lower flux of reactive and energetic species to reach the substrate. It is important to consider the attributes of the physical barrier that separates the source and substrate, since it serves to reduce the flux of all plasma generated species leaving the source. An important consideration for these sources is that their low electron density will result in less effective dissociation of the background gas compared with higher density PEALD sources. Dissociation will still occur though, and having the growth surface relatively close to the source volume may make up for some of the deficiency in atomic radical generation of these lower density plasma sources. One important benefit of CCP configurations that has led to its adoption in industry is the relative ease with gas exchange can be handled in these systems. CCPs allow one to confine discharge reaction volumes easily, whereas ICPs’ use of induction to heat the plasma makes confining the discharge difficult.

Another case reviewed in Fig. 12 is the HC-CCP which, as noted previously, produces the highest energy electron population of any of the sources discussed here. As such, it is very efficient at ionizing the background gas since election impact ionization cross sections continue to increase until energies in excess of a few hundred eV are reached. It is important to keep in mind that this high-energy electron population can also interact with the substrate under low pressure conditions. Whether this is good or bad will likely be material dependent.

Finally, the remote ICP due to its high electron density and Maxwellian-like EEDF shape will be the most effective source at dissociating molecular gases via electron impact. As such, it will generate a high-density reactive atomic neutral within the source volume. However, these systems must be remote if one wants to isolate the inductive discharge from the substrate. The remote nature of the system though means that many of the atomic species generated in the source region could be lost to recombination at the walls during transport to the substrate. A potential solution to this problem is spatial ALD, which is employed in industry.^72,73

Within the field of PEALD, there are numerous works that identify the important plasma-produced species in terms of how they interact with surfaces and their role in enhancing ALD.^{5,6,25,67,68,74} Concisely, plasmas deliver both reactive species and energy to a surface during growth. Reactive species such as radicals (or molecular fragments) provide not only the desirable species (e.g., N atoms for metal-nitride growth) but also help lower the activation energy for chemical reactions compared to purely thermal processes.⁷⁴ Ions, fast neutrals, electrons, and photons can deliver a significant amount of energy, and certain ions (e.g., O⁺ ions) are also chemically active. Of those, energetic ions are generally recognized as an important vehicle for delivering energy to a growth surface. Ions deliver their kinetic energy by virtue of their acceleration through the sheath, and momentum exchange between these ions and the lattice atoms produces a thermal spike in the material near the point of contact. Ions will also deliver energy associated with their ionization state. This potential energy is often overlooked but should be considered.⁷⁵ For example, the ionization potential of argon is about 15 eV, which can be a significant fraction of the kinetic energy when the sheath is collisional. Fast neutrals, or those neutrals with kinetic energies well above the background gas, are typically created via charge exchange collisions in the sheath and will deliver their kinetic energy to the surface like the ions. Energetic electrons will also carry kinetic energy to the surface. Although the large difference in mass between the electrons and lattice atoms precludes them from transferring significant momentum to the lattice, electrons can serve to heat the material. When electron energies are high enough, they can also drive chemical reactions at the surface⁷⁶ through electron-stimulated desorption.⁷⁷ Additionally, photons are produced in abundance in plasmas and have been seen to influence ALD processes.^78,79 Their interaction with the substrate will depend on their energy, with vacuum ultraviolet (VUV) (<200 nm) photons producing the most significant effects, including surface heating, photoemission, and the potential for driving chemistry through desorption processes similar to those produced by a flux of electrons.^6,80 However, as noted earlier, photons can also produce adverse effects in film properties²⁵ and device structures.²⁶ 

The abundance of energy delivered to and absorbed by the substrate heats the material and serves a similar function to substrate temperature in thermal ALD. This, of course, comes with a significant distinction: the energy is deposited and absorbed within a limited depth, causing the surface to be driven out of thermal equilibrium with the bulk material. For example, Graves and Humbird⁸¹ estimate a 200 eV Ar⁺ ion incident to a silicon surface will deposit its kinetic energy within about 2.5 nm from the impact site. Analogous to the role of temperature in the transition from amorphous to crystalline growth in many CVD processes, it is possible that the flux and energy of the arriving species will result in a range of morphologies in many materials deposited by PEALD. Certainly, the role of energetic particle fluxes in governing film morphologies during growth has been of long-standing interest to the PVD community.^76,82,83

With these ideas in mind, we can construct a PEALD window—considering both amorphous (a-ALD) and crystalline (c-ALD) growth—based on the delivery of both reactive species and energy during the plasma half-cycle. This is illustrated in Fig. 13. In this construct, it is assumed that the requisite metal precursor is delivered during the first half-cycle and that substrate temperature alone is not sufficient to drive the needed chemistry. Here, the ordinate represents the fluence of plasma-produced reactive species and the abscissa is the energy flux density, Γ, delivered by all species (i.e., ions, fast neutrals, electrons, photons, etc.). The latter is determined by summing over all energy-carrying species impacting a surface. For each species, $Γi=fi⋅Ei$⁠, where f_i is the fluence and E_i is the energy. The result, shown in Fig. 13, are growth windows bound by a range of reactive species fluences and energy flux densities that is analogous to growth windows previously developed for thermal ALD.^84,85

The lowest energy flux density (Γ_o) is the minimum required to achieve amorphous ALD. Below this value, there is insufficient energy delivered, regardless of the reactive species flux, resulting in incomplete reaction with the precursor delivered during the first half-cycle. The second (Γ₁), represents the transition between a-ALD and c-ALD. That is, between Γ₀ and Γ₁ precursor conversion or reduction is achieved during the plasma half-cycle, but the film remains amorphous. Between Γ₁ and Γ₂, the energy flux density is sufficient to promote crystalline growth. The shaded region around Γ₁ indicates the likelihood that the transition between a-ALD and c-ALD is not abrupt, where one can assume an energy flux density range over which mixed phases are possible and the film is not homogenous. Over the energy flux range for either a-ALD or c-ALD, the fluence of reactive species must be maintained within a reasonable bound. When the plasma-produced reactive species fluence is too low, there is an insufficient amount available to react with the metalorganic precursor. When the fluence of plasma-produced reactive species is too high, an excess of physisorbed species and/or unwanted surface reactions could retard or prevent the desired precursor reduction reactions, potentially leading to film impurities. Though, such concerns are likely only relevant to plasmas produced in complex gas mixtures where the possibility of unaccounted for reaction pathways is higher. A properly chosen precursor and plasma background gas is expected to alleviate many of these concerns regarding excess chemical reactivity.

Beyond Γ₂ and independent of reactive species flux, the energy delivered to the surface becomes excessive, producing deleterious effects such as diminishing growth rates. For example, when reactive species fluence is low and the energy flux density is high, the precursors can be decomposed and/or desorbed from the surface. Similarly, structural damage that can lead to suboptimal material properties at high-energy flux densities is possible. However, annealing can minimize certain types of damage (e.g., defect generation). At Γ₃ and above, the energy flux density becomes so high that the per cycle etch rate exceeds the deposition rate.

If one considers substrate heating, as is often the case in PEALD, the required energy flux density values and subsequently the widows will shift. Generally, one presumes that the energy flux density thresholds will decrease or increase when the substrate temperature increases or decreases, respectively. In addition, the creation of all plasma species is generally linked together by virtue of the fact that their production rate is dependent on the EEDF. That is, as reactive species generation increases or decreases, one can expect a corresponding increase or decrease in the production of energy-carrying species. Thus, a change in the fluence of reactive species at the surface should proceed with comparable change in energy flux density at the surface. Practically, the delivery of reactive neutral species and energy to the surface can be somewhat decoupled by pressure, gas mixture, source-to-substrate distance, and, perhaps, artificial confinement of the plasma with electrodes or meshes. Still, it would be challenging to access the extreme regions of the figure in typical reactor configurations utilizing reactive gas mixtures, where the fluence of reactive species far exceeds the fluence of energy-carrying species and vice versa.

The previous schematic was agnostic to the means of energy delivery. We now consider only ions, as they represent an important species in the delivery of energy to a surface during plasma-based deposition processes.⁷⁵ Ions possess both significant energy and mass, allowing them to efficiently deliver energy to the surface of the growing film. If we assume the fluence of plasma-produced reactive species during the plasma half-cycle falls within the range discussed above, we can construct the a-ALD and c-ALD windows in terms of the ion fluence at the surface and the associated ion energy. The result is shown in Fig. 14.

The energy flux density delivered by ions impacting a surface is $Γ=f⋅E$⁠, where f is the ion fluence (y-axis) and E is the kinetic energy (x-axis) of the ions. For simplicity, we have assumed that all ions are the same and have not considered the potential energy of the ions. However, the latter contributes to the local heating of the material^75,86,87 and could likely be added by considering an offset in the energy axis similar to that discussed above for substrate heating. The important energy flux density thresholds in Fig. 13 (Γ₀, Γ₁, Γ₂, Γ₃), when accounting for only the ions, are shown in this figure as diagonals when the ordinate and abscissas are plotted using log scales. As such, the origin in the plot is not zero but rather an arbitrary minimum fluence (f_min) and energy (E_min) and the scale for both would likely span several orders of magnitude.

Γ₀ is defined as the product of f_min and E₀, where E₀ is the minimum ion energy needed to achieve a-ALD at the minimum fluence, f_min. Γ₁, Γ₂, and Γ₃ are defined similarly, where E₁ is the minimum energy required to drive crystallization, E₂ is the threshold at which suboptimal growth characteristics begin to emerge, and E₃ is the energy required to produce a net etch rate per cycle. In the context of c-ALD, the window between E₁ and E₂ has some precedence. Brice et al.⁸⁶ have identified a similar widow for ion-enhanced deposition processes, where E₁ is the kinetic energy required to create surface displacements, E₂ is the kinetic energy required to produced bulk displacements, and energies between E₁ and E₂ are well suited to stimulate crystalline growth. It should be noted that when invoking a kinetic energy threshold, it is not unreasonable to expect a discontinuity along the lines of constant energy flux density, Γ. This is because the energy thresholds represent a transition between processes driven by kinetic effects and those related to energy deposition that serve to heat the material, where the material’s ability to dissipate the energy should be considered. This could cause the energy flux density curves Γ₁ and Γ₂ to transition to vertical for ion fluences below f_min. That is, when the ion kinetic energy is capable of driving a chemical reaction or morphology change, any ion fluence is sufficient.

Figure 14 considers energy flux densities dictated by the material system of interest and a generic plasma. Next, we consider the plasma properties and how they can be adjusted to meet the demands defined by the material system. The results are shown in Fig. 15.

The flux of ions and range of energies at a surface are determined by the intrinsic plasma properties including the plasma density and EEDF, along with the operating pressure and gas composition, as discussed previously. The ion fluence scales with plasma density by f_i = n_i v_b, where n_i is the ion density and v_b is the Bohm velocity. Hence, f_i max will scale directly with applied power and is arbitrarily located along the y-axis in Fig. 15. In the case of remote PEALD systems, the density falls as the plasma expands toward the substrate by an amount determined by the pressure and standoff distance. As such, the range in fluence will vary accordingly. For a fixed standoff or direct plasma geometries, applied power and pressure thus become operational control “knobs” to adjust the ion fluence over a range of values defined by f_i max and f_i min in Fig. 15. Note that the response to changes in pressure is not as clear as the response to changing power input. In reactive gas plasmas and remote geometries, the wealth of gas-phase reactions discussed earlier will likely lead to a reduction in fluence at the surface. However, in noble gases and direct plasma systems, an increase in pressure—or neutral density—can lead to an increase in plasma density and thus fluence at the surface.

The EEDF determines the plasma potential, which in turn, defines the highest ion energy possible, E_{i max}, at a grounded surface. It is arbitrarily positioned on the abscissa in the example of Fig. 15. The minimum ion energy, E_{i min}, is determined by the collisionality of the sheath. For a given plasma density, an ion will undergo many collisions while it transits the sheath at high pressure [see Fig. 15(b)], thus losing a significant amount of energy compared to one that transits the sheath at low pressure. So, one can expect a possible range of ion energies, depending on operating pressure. Here again, the pressure can be used as a possible knob to control ion energy. Alternatively, substrate biasing can be used to increase the ion energy above the inherent maximum dictated by the plasma potential. When using RF biasing, the energy range will include a contribution associated with the driving frequency.⁸⁸ An example of this effect is illustrated in Fig. 16 where by raising the ion energy via RF biasing, one moves from an amorphous/low crystallinity ALD regime to a high crystallinity c-ALD regime, assuming the proper ion flux conditions are met. The measurements shown in Fig. 16(b) illustrate the breadth and structure of the ion energy distribution function (IEDF) at RF biased substrates. It is important to note that it remains unclear which attribute of the IEDF is most important in PEALD. Parameters such as the minimum, maximum, and mean ion energies, as well as the shape of IEDF (e.g., bimodal or unimodal) can influence film growth. Understanding these relationships provides an opportunity for research.

Taken together, the parameters of the plasma along with the use of substrate biasing define an operational window that indicates the limitations and possibilities in terms of accessing the appropriate ion fluence and energy conditions required to meet the demands for a-ALD and c-ALD as defined by the material system.

While these plots provide some guidance, it is critically important to note their shortcomings when compared to actual systems. Specifically, the values needed to determine the energy flux densities in Fig. 13 are not well known for the broad range of materials synthesized using PEALD. Those would require an accounting of all species and reactions that contribute to the energy balance at the surface.⁸⁹ Even if simplified by only considering ions (as in Figs. 14–16), the task remains challenging and would be best accomplished using simpler processes, such as fundamental studies that employ ion beam assisted deposition²⁵ processes. However, the range of applicable material systems is likely limited in such processes. An alternative to empirical approaches include modeling efforts that incorporate kinetics considerations.⁹⁰ 

From an applied perspective, there are challenges associated with meeting the demands of energy deposition. It is thought, for example, that ion energies can be too high in industrial tools.⁹¹ Indeed, the inherent ion energy in a plasma, E_{i max} can be higher than E₁, E₂, and E_s for a given material system, while the relative values of E₁, E₂, and E_s can be small or indistinguishable, with regard to the spread in ion energies at surfaces. This can limit the level of control over the growth process, suggesting an important need in the field: tighter control over the plasma parameters. If one were to craft the ideal plasma processing system, it would include the ability to provide independent control over ion and reactive species production; tight control over the ion fluence at surfaces; and a narrow distribution of low energy ions, whose energy can be increased without broadening. There are a number of potential paths toward increasing the level of control over plasma parameters in PEALD systems. A few possibilities include more sophisticated RF waveform control using multiple frequencies and nonsinusoidal waveforms,^92–94 pulsed plasma systems that utilize time varying RF power to both plasma source and substrate,^95,96 the use of neutral beam sources,^97,98 and also flexible low temperature plasma sources that offer enhanced control over ion and radical production.^99–101 All of the aforementioned approaches are areas of active research, and, as such, require further study by researchers spanning the fields of plasma physics and thin film growth.

Due to the complex nature of plasmas, it is often useful to monitor plasma processes with a variety of different diagnostics to better ascertain the identities and densities of the energetic and reactive species being generated within the reactor. This section aims to provide an overview of the different methods typically used to measure various plasma species and provide the reader with useful references if they believe it useful to attempt some of these measurements in their deposition systems.

Langmuir probes. Langmuir probes are deceptively simple diagnostics as they are essentially small pieces of a wire that collect positive and negative charged particles from the plasma at varying voltages. Larger, planar geometry probes can also be used to calculate the flux of ions at the substrate; however, they tend to extensively perturb the plasma when biased above the floating potential of the system.^18,68 From the collected current versus applied voltage trace generated with the probe, a variety of plasma properties can be calculated, including the EEDF, plasma potential, plasma density, and the ratio of negative ions to electrons. Unfortunately, the interpretation of Langmuir probe data is often not straightforward as there are potential sources of error, particularly in plasmas with complex, depositing chemistry driven by RF power input. As such, it is advisable to familiarize oneself with the available literature^102,103 before undertaking such a measurement, particularly in deposition plasmas. Of particular interest are methods for RF compensation,^104,105 proper construction materials for Langmuir probes,¹⁰³ and probe heating techniques to minimize or eliminate deposition on the probe surface. Since many PEALD processes are performed at relatively high pressure (>100 mTorr), an understanding of how collisional transport to the probe affects data interpretation is also important.^102,103,106 As an alternative, commercial Langmuir probe systems are available from several vendors that take into account these considerations.^107–109 Finally, due to the remote nature of the plasma sources in many PEALD reactor configurations, it is critical that measurements were performed as close to the substrate location as possible due to the significant spatial variation in plasma parameters throughout the reactor.

RF impedance probes are a class of probes that are particularly useful in depositing environments where Langmuir probe performance can suffer from surface contamination. These probes are generally well suited for electron density measurements as they all exploit the resonance effects that occur when AC electric fields are applied near the plasma electron frequency, $ωpe=nee2/meϵ0$⁠. Here e is the electronic charge, m_e is the electron mass, and ε₀ is the permittivity of free space. The operation of this class of probes is detailed in many works within the low temperature plasma physics literature, and there are many variants including the plasma absorption probe or plasma impedance probe,^110–112 the wave cutoff probe,¹¹³ the hairpin resonator,^114,115 and the multipole resonance probe.^116,117 While these systems offer considerable promise, one drawback is that they employ sophisticated high frequency electronics, and there are, currently, no commercial vendors of these systems.

Gridded energy analyzers or retarding field energy analyzers (RFEA) are a class of more sophisticated charged particle measurement tools that can be used to obtain both EEDFs and IEDFs. This is done by constructing a series of isolated transparent grids¹¹⁸ that allow electrostatic discrimination of positive and negative charge species while also providing energy analysis of the collected species. Construction of these instruments are somewhat more involved than Langmuir probes and similarly a number of commercial vendors provide them.^119,120 An important consideration in the use of RFEAs is operating pressure. Because diagnostic fidelity depends on the collisionless transport of ions between the grids, neutral collisions within the device are a major source of error. Generally, these diagnostics are limited to ∼50 mTorr operating pressures^119,121 although this can be increased by adding differential pumping to the system.¹²² 

Energy-resolved mass spectrometers configured for positive and negative ion analysis offer an even more comprehensive diagnostic tool than RFEAs. Devices of this nature are generally composed of an energy analyzer in series with a mass spectrometer and provide the ability to separate the flux charged particles according to mass and energy, thus producing energy-resolved, mass distributions or mass-resolved, energy distributions. Typically, a flux of ions is sampled through a small orifice at the front end of the device, which then passes through ion optics before mass and energy separation. Because the trajectories through the system are long and critical for analysis, the systems are differentially pumped to ensure that the final measured flux of species’ results represent the flux of species entering the orifice. Process operating pressure and pumping speed along with sampling orifice dimensions are important considerations that influence measurement efficacy.¹²³ We also note that the orifice size considerations discussed earlier concerning the ability to prevent plasma leaking through an electrode mesh in remote CCP sources also apply here. While the ability to discriminate based on both mass and energy provides a wealth of information concerning the flux of species sampled with these tools, they are complex and require a detailed understanding of the system operation by the user, particularly as it pertains to properly measuring ion energy distributions that encompass a large range.^124,125 That said, they are well suited for understanding the flux of charged species incident to the substrate in complex plasma chemistries.

Mass spectrometers configured for neutral detection are often referred to as residual gas analyzers (RGAs). Like the mass spectrometer discussed above, these devices provide the ability to measure a wide range of species present in a plasma by separating the species that enter the system by mass. The neutrals that enter the tool are first ionized, so that that can be passed through a series of ion optics and then separated by mass in a quadrupole mass analyzer. The process of ionization via thermal emission from hot filaments requires a low operating pressure—particularly when the operating environment includes reactive or corrosive gases—to ensure the filament lifetime is reasonable. The requisite pressure is typically much lower than the common operating pressure in PEALD systems. Accordingly, RGAs must be differentially pumped when used to sample neutrals species during the plasma step and the user should apply the same considerations of sampling orifice size discussed in Sec. II D 1.

Optical emission spectroscopy (OES) is a useful tool for measuring light emission from the plasma source region of a reactor where the EEDF is energetic enough to excite the background gas into radiative states. The light emission from these radiative states can be useful for identifying reactive species—both atomic and molecular—generated in the plasma source ^18,42,126 and broadband spectrometers of appropriate resolution can be obtained at relatively low cost. In particular, the methods of actinometry^127,128 and self-actinometry¹²⁹ are useful in determining the relative and absolute densities of reactive species within the plasma source.¹³⁰ These methods rely on the introduction of a known density of rare gas atoms (e.g., Ar, Kr) to determine the density of an unknown species within the plasma (i.e., O, N, F). Process monitoring and time-resolved measurements of precursor (e.g., trimethyl aluminum)/reactant (e.g., reactive oxygen species) interactions are additional powerful capabilities enabled by OES.¹³¹ 

VUV emission spectroscopy is similar to OES described above but differs in the wavelength range that is observed (30–200 nm). As such, VUV spectroscopy is a somewhat more difficult technique to implement as it requires an in vacuo spectrometer to observe these photons. Photons in this energy range (6–40 eV) have the ability drive photoemission from the surface and also have a significantly reduced range¹³² in the material compared with the lower energy photons in the visible to near infrared range. As such, they are not only a useful neutral species diagnostic but are also an important player in plasma-surface interactions.¹³³ Because of this, monitoring their flux as a function of plasma conditions is a useful piece of information for process development.

Absorption spectroscopy is another technique that utilizes gas excitation to measure the density of neutral species of interest in plasma reactors. In absorption spectroscopy, a broadband light source or laser is supplied to induce a desired excitation indicative of the presence of a particular atom or molecule. By comparing the transmission spectrum in the presence of the plasma to the case without plasma, one can obtain a quantitative measure of relative or absolute atom or molecule densities in the reactor.¹³⁴ This technique is applicable in the VUV range as well as the UV-NIR range if proper equipment is available.¹³⁵ Broadband VUV spectroscopy can be particularly useful in afterglow plasmas, such as those frequently encountered in PEALD systems, where the level of gas excitation is low and absorption only occurs for photons capable of exciting a ground state atom or molecule into its first excited state.

In order to explore the influence of plasma characteristics on the properties of ALD films, two demonstrations are presented: (1) TiO₂ films on sapphire substrates and (2) InN films on either GaN templates or a-plane sapphire. Even though plasma characteristics are often convoluted with interdependent properties, correlating these with resulting ALD film properties can elucidate which knobs should be considered during deposition.

Atomic layer deposition of TiO₂ has been widely explored in recent years due to its promise in nonvolatile resistive switches, high-k gate dielectrics, solar cell, nonlinear optics, and photocatalytic applications. This method has become increasingly useful as device dimensions are reduced and nonplanar complexity is increased. Traditionally, the low ALD growth temperature (T_g), adventitious for integrating dissimilar materials or accessing metastable phases, yields amorphous TiO₂ films below 300 °C, which can inhibit the viability of these films in device applications. Additionally, there are many cases where depositing a certain TiO₂ phase, anatase or rutile, is critical to achieving the best device performance. Recently, there have been a limited number of reports of deposition of crystalline films using PEALD.^{17,18,67,78,136,137} Moreover, depositing the rutile phase instead of the anatase phase has been demonstrated by increasing the growth temperature or postdeposition annealing, using nucleation layers, or substrate biasing. In this case study, we present the impact of remote RF plasma properties on the ability to achieve crystalline films at temperatures less than 200 °C with phase selectivity.

For this particular investigation, 40 nm TiO₂ films were deposited on c-plane sapphire substrates in a Veeco Fiji G2 reactor with tetrakis(dimethylamido)titanium and either 20 SCCM Ar/O₂ or pure O₂ plasma at 100–350 °C. Here, the focus will be primarily on low growth temperatures (≤150 °C) where this work represents the first example of crystalline growth and phase selectivity of TiO₂ at temperatures where photoresist patterning is possible. To achieve these results, a remote ICP plasma was operated at 300 W and a pressure ≤10 mTorr in all cases. In addition, multiple gas-phase ratios and flow rates of Ar/O₂, and pure O₂, were investigated using optical emission spectroscopy to determine the effect of gas-phase chemistry on atomic O production.^17,18 Total flow rates through the plasma source of 20 SCCM (both for Ar/O₂ and pure O₂) were found to be optimal for the aforementioned conditions of pressure and RF power.

Figure 17 shows that varying the gas composition from an Ar-diluted O₂ to a pure O₂ condition had a dramatic effect on the phase of TiO₂ films, deposited on c-plane sapphire substrates at various temperatures. It is of interest to note that all films exhibited crystallinity independent of deposition conditions down to 100 °C where films are typically amorphous^136,138,139 even when substrate biasing has been employed. At temperatures less than 200 °C, using an Ar/O₂ plasma (ratio = 4) produced anatase TiO₂ films with (004) orientation similar to previously reported,¹⁷ while a pure O₂ plasma with similar total flow and pressure resulted in a rutile film with a (200) orientation.

While phase selectivity was achieved by varying gas composition within the plasma source, it is important to realize that these changes are only effective due to the changes in reactive and energetic species flux that are enabled at the substrate. Other works have detailed the relationship between varying ICP gas-phase composition and its effect on ion and reactive neutral production;^17–19 however, we will briefly review them here to clarify what changes in the plasma flux are most relevant to growing crystalline TiO₂ films.

The most notable change in species production that results from varying gas flow composition is that the pure O₂ plasma produces at least a factor of 3 higher atomic oxygen concentration than the Ar-diluted plasma.¹⁸ O₂ flow rate through the ICP also plays an important role in optimizing atomic O production. Since PEALD processes in general usually require an Ar purge gas in addition to the plasma gas flow, maintaining a high enough plasma gas flow rate to prevent Ar backstreaming into the plasma source is an important consideration;¹⁹ this was found to be in the range of 20–40 SCCM. Increasing O₂ flow to ≈100 SCCM, however, was also found to be suboptimal for atomic O generation, implying that the residence time of O₂ within the ICP impacts the dissociation fraction.¹⁸ It may also impact the production of other plasma generated species such as O⁻,^140,141 which are known to be strong oxidizers. Studies of other oxide materials have also shown the flow rate to be another important parameter in controlling crystallinity.¹⁹ Finally, the pure O₂ case also exhibited an elevated plasma potential relative to Ar-diluted case, which, at these pressures, implies higher ion energies at the surface. In the Ar-diluted case, the maximum ion energy was estimated, based on plasma potential measurements, to be 32 eV, whereas in the pure O₂ case, it was 46 eV.¹⁸ This change in ion energy would have the effect of moving the growth condition to the right in the c-ALD growth window shown in Fig. 14. Achieving these elevated ion energies seems to be the result of a high-energy population of electrons emanating from the ICP and raising the plasma potential downstream in the growth reactor.¹⁸ It is notable, however, that these ion energies are substantially lower than those typically achieved using substrate biasing,^{5,25,32,67,68,136} indicating that gas-phase chemistry is at least as important as ion energy and flux to the growth of crystalline films.

As can be seen from Fig. 17, both the Ar-diluted and pure O₂ conditions result in highly oriented, high-quality crystalline films. Figure 18 shows that the growth rate and index of refraction (n, at 633 nm), which is related to film density, is also changed with gas composition. The growth rate increased from 0.55 ± 0.02 to 0.71 ± 0.03 Å/cycle while the n increased from 2.45 ± 0.02 to 2.49 ± 0.01. These values compare well with the theoretical refractive index of 2.45 for TiO₂ at 633 nm. The trends are similar to those attained with substrate biasing except at lower overall ion energies.^5,6,67

These results suggest that the ion energy (30–50 eV) and ion flux (1 × 10¹⁵ cm⁻²s⁻¹)¹⁸ obtained at these low pressures place such growth conditions within the c-ALD window discussed previously. While both films exhibit crystallinity, the XRD intensity is higher and the full-width-at-half-maximum (FWHM) is much smaller, and thus, the quality of the film is higher for the rutile film deposited with O₂ only plasma.

In Fig. 17, it is shown how growth temperature variation at constant plasma conditions, as well as gas-phase chemistry at a particular temperature, could be used to fine tune the strain in the film and even be tailored to promote either compressive or tensile strain in deposited films. Figure 17(b) illustrates that films deposited with pure O₂ plasma are compressively strained only at 100 °C and tensile strained at any elevated temperature with the magnitude of the tensile strain relatively independent of temperature. However, by changing the gas-phase chemistry at elevated temperatures, as shown in Fig. 17(c), the tensile strain in the rutile phase can be drastically reduced. Additionally, the crystallinity and quality of the films deposited in pure O₂ increase as the deposition temperature decreases. This counterintuitive finding illustrates the important role that both ion energy and reactive neutral concentration can play in c-ALD film growth.

In general, changing the plasma gas composition from a diluted oxygen to pure oxygen plasma produces higher quality films. As discussed above, this suggests that the added ion energy and higher atomic oxygen concentration can enhance crystalline quality and vary strain states, sometimes in nonintuitive ways, while promoting various phases is still reliant on other conditions such as substrate material or orientation.¹⁷ 

Finally, Fig. 19 shows a high-resolution XRD (HRXRD) scan taken with CuKa1(α) radiation of the rutile TiO₂ film deposited on the c-plane sapphire substrate with a pure O₂ plasma at 100 °C from Fig. 17. Typical TiO₂ films are amorphous at these temperatures even when employing PEALD with substrate biasing. However, the clear Pendellosung fringes present in the spectra are indicative of an atomically smooth interface and fully strained, high-quality, epitaxial film with 0.154% compressive strain. In addition, the FWHM of the 40 nm film is only 389 arc sec, further validating the highly crystalline quality of this thin film. Using the Scherrer equation, the crystallite size was found to be 41.6 nm, similar to the 44 nm thickness of the films determined by spectroscopic ellipsometry measurements, suggesting that the film is nearly single crystal. These results demonstrate the ability to obtain epitaxial, metastable films at very low temperatures by adjusting the plasma parameters to provide both flux of reactive atomic oxygen and ion energy to the surface to overcome thermal equilibrium barriers.

In this section, we turn to nitrides and how changes in plasma properties can influence their growth and crystallinity. InN and its alloys with GaN and AlN (III-N) have made a significant impact on electronics and optoelectronics including solid-state lighting, laser diodes, RF amplifier technology, and power switching.^142,143 However, the relatively high growth temperature required in III-N growth techniques, such as molecular beam epitaxy (MBE) and metal-organic chemical vapor deposition (MOCVD), has impeded further development due to challenges with miscibility gaps and strain limitations, preventing the realization of high In-content alloys of InGaN, InAlN, and InGaAlN.^142–144 Phase segregation of InN and GaN or AlN in the growth of III-N alloys is particularly problematic at the high growth temperatures (600–1200 °C) that are required in MBE and MOCVD. The low growth temperatures offered by PEALD provides a new approach to III-N growth.^143,145 In the following, we examine the role of the plasma for the epitaxial growth of InN, which has previously been grown at <260 °C using alternating pulses of trimethylindium and argon/nitrogen plasma.^144–146 

To study the effect of varying gas-phase chemistry, InN films were grown at ≈250 °C, for various N₂ flow fractions, at P > 200 mTorr, and characterized in situ with grazing incidence small angle x-ray scattering (GISAXS) using a portable c-ALD reactor¹²⁶ at the G3 hutch of a Cornell High Energy Synchrotron Source. Details of experimental conditions can be found in Refs. 102 and 147. In Fig. 20, plasma source OES data are combined with two GISAXS contour plots, one for the low N₂ flow fraction (5%) and one for the high N₂ flow fraction (32%). It is notable that the low N₂ fraction produces 10× more atomic N than the high N₂ flow fraction. Importantly, although the previous example for oxide growth utilized a low pressure growth process (P ≈ 10 mTorr), these growth conditions are of higher pressure, and as such collisions and transport play a larger role. The details of this measurement can be found in Ref. 126, but, in brief, this result is related to how the EEDF changes with varying concentrations of Ar and molecular N₂. At higher concentrations, the vibrational states of N₂ act as an energy sink that cools the EEDF and thus limits the plasmas’ ability to dissociate N₂ via electron impact. These plasma chemistry effects are reflected in the GISAXS data. In these InN growths, correlation peaks form, indicating island-type growth where the islands are correlated in size and position.¹²⁶ For low N₂ flow fraction (higher N atom concentration), the size of the features is larger than for the higher N₂ flow. In all cases, the films are continuous and growth evolves to an island morphology. For the 32% N₂ by flow condition, the island spacing is ∼11.5 nm, whereas for the 5% N₂ by flow condition, the island spacing is ∼12.5 nm. This change is surface morphology illustrates that growth kinetics can be influenced by plasma gas-phase chemistry. In Fig. 21, ex situ AFM analysis further supports this connection by demonstrating that markedly smoother films were grown under high atomic N conditions. It is also notable that while 1% carbon content was detected for the film grown in the high N₂ condition, no carbon content was detected via XPS (30 s sputter, 2 keV medium current Ar ion beam) for the 5% N₂ growth condition.

Figure 22 shows the effect of the addition of hydrogen to an N₂:Ar flow of 1:1 during plasma generation. The addition of a 4% admixture of hydrogen is accompanied by substantial emission from the NH*(A³Π→X³Σ⁻) transition¹²⁶ indicating the formation of NH_x within the source. Typically, the addition of hydrogen in InN growth is avoided^148,149 as it has deleterious effects on MOCVD processes. The results in Fig. 22, however, indicate that at low temperature (250 °C) under c-PEALD conditions, adding hydrogen to the plasma pulse leads to significantly improved film quality. Figure 21(b) shows a high-resolution XRD image of 20–40 nm thick InN on GaN/a-sapphire grown with and without hydrogen. Without hydrogen, oriented InN on GaN/sapphire was not measurable. With hydrogen, c-oriented (0002) InN films were observed. Based on XPS (30 s sputter, 2 keV medium current Ar ion beam, 10⁻⁷ Torr base pressure), films deposited with H₂ have much lower (1%–2%) oxygen content with no detectable carbon.

These results could be due to the synergistic process in which enhanced surface chemistry due to the presence of H and NH_x reactive species remove oxygen in conjunction with the plasma ion flux, providing energy to crystallize the film. The reduction in carbon could be related to mechanisms proposed by Erwin and Lyons¹⁵⁰ in which hydrogen was theoretically shown to effectively remove methyl radicals leftover from metalorganic precursors. This again points to the unexpected and often beneficial results that can be obtained from properly tuned plasma chemistry. In addition to reactive species in plasmas, the selection of process gases (e.g., precursors, reducers, etc.) as well as their combination is critical for carbon reduction.

Plasma-enhanced ALD is rapidly becoming a method of choice in the materials processing industry, maintaining key elements of the conventional ALD process (surface-mediated, self-limiting growth, etc.) while expanding the materials possible and the substrates upon which they can be grown. A key to advancing PEALD further includes a better understanding and control of plasma properties that are coordinated to the resulting film properties.

In this work, we have conducted a cursory review of relevant plasma physics in order to review the most common plasma sources employed in PEALD and discuss the pros and cons of each. Important elements of this review are the nature of the electron energy distribution function, which governs nearly all aspects of the plasma’s properties from energy of ions to distributions of reactive species as they are generated in the plasma source. We then discussed key considerations in the transport of species from the source region to the substrate surface where films are grown, including the impact of process conditions on key plasma parameters. We then focus our attention on the most critical region of the PEALD process: plasma-surface interactions. Here, issues of energy deposition to the growth surface and the resulting non-thermal-equilibrium environment of the near surface, compared to the bulk of the substrate are considered. We note that energy comes in many forms to that surface and can contribute to surface chemical reactions, surface bond breaking, and subsurface bond breaking depending on the nature of the source and magnitude of the energy. Finally, we demonstrate the importance of plasma property control in the PEALD of crystalline materials (epitaxial and not) through two case studies: TiO₂ growth on sapphire and InN growth on GaN and sapphire. In each case, the importance of controlling ion flux and reactive neutral fluxes on the crystalline quality of films are exemplified.

The overarching message is that, although plasmas create a much larger range of possible materials by ALD through PEALD methods, the capability comes with significant complexity in both the plasma physics and chemistry of these systems, where numerous interdependent mechanisms are involved. Understanding can be advanced through multiple, careful plasma characterizations that can guide process development to achieve improved/preferred materials properties. This work points, perhaps, to a promising area of future research in which engineered plasma sources that allow independent control of ion and neutral/radical fluxes enable process tuning for crystalline materials growth by PEALD. This increased level of plasma control, if realized, should lead to further expansion of PEALD’s footprint in industry and research, particularly in cutting edge areas of development such as area selective deposition and the integration of ALD with atomic layer etching.

Importantly, we would be remiss not to mention the connection of this work to Dr. John Coburn. Research and development in areas of plasma-based materials processing has been strongly influenced by the works of John Coburn and his co-workers. The direct citations of his work as well as the citations of his that can be found in the references in this manuscript are substantial and cover topics ranging from plasma characterization to materials processing. To merely say that his contribution to our understanding of low temperature plasmas and their application to materials processing is significant feels almost insufficient without also extending our gratitude.

The authors acknowledge the support of the Naval Research Laboratory Base Program through the Office of Naval Research.