Atomic layer deposition of thin films: from a chemistry perspective

The ligand-exchange mechanism (figure 2(a)) is the most common mechanism for ALD, and it can be described as the following general reaction scheme:

$$\begin{align}\left| {\left| { - a + M{L_n}\left( g \right){\text{ }} \to {\text{ }}} \right|} \right| - M{L_{n - 1}} + aL\left( g \right),\end{align} \tag{ 1 }$$

Zoom In Zoom Out Reset image size

where || denotes the surface, a is a surface group, M is a metal atom, and L is a ligand of the precursor. During the ligand-exchange reaction, a ligand (L) of the gaseous precursor molecule (ML_n) is displaced by the surface species (||−a), forming the volatile byproduct of aL, and meanwhile, the remainder of the precursor molecule bonds to the surface, forming a chemisorbed new surface species of ||−ML_n−1. The ligand-exchange reaction may continue to occur between the chemisorbed ||−ML_n−1 and another nearby ||-a surface group, and therefore, multiple chemical bonds may be formed between the precursor and surface as the following:

$$\begin{align}\left| {\left| { - M{L_{n - 1}} + {\text{ }}} \right|} \right| - a \to \left| {\left| { - M{L_{n - 2}} - } \right|} \right|{\text{ }} + aL\left( g \right).\end{align} \tag{ 2 }$$

A common example of the ligand-exchange mechanism is the proton transfer scheme, in which a Brønsted base abstracts a proton from a Brønsted acid. As the scheme shown in equation (1), the metal precursor (ML_n) serves as a Brønsted base, and it can abstract protons (a = H) from the Brønsted acidic surface groups, such as hydroxyls. In this respect, the metal precursor with a stronger ionic character of its metal–ligand bond is more favorable for the proton transfer reaction, because in this case the ligand is more analogous to a free anion and hence more effective to abstract a positively charged proton. The acidity of the ligand itself is also very important; a stronger basic ligand is more effective to abstract a proton, and this effect is well manifested when comparing the precursors of metalorganics and metal halides.

The ALD of Al₂O₃ from the precursors of trimethylaluminum (TMA) and H₂O is a typical process that follows the proton-transfer ligand-exchange mechanism. This ALD process has been extensively studied both experimentally and theoretically [33–35]. In the TMA half-cycle, the surface terminal –OH groups react with the TMA via the proton-transfer ligand-exchange reaction, forming the O–Al bonds and releasing the volatile byproduct of CH₄, and then, in the subsequent H₂O half-cycle, the remaining surface methyl ligands are displayed by hydroxyls via another proton-transfer ligand-exchange reaction with the incoming H₂O molecules:

$$\begin{align}\left\| { - {\text{OH}} + {\text{Al}}{{\left( {{\text{C}}{{\text{H}}_{\text{3}}}} \right)}_{\text{3}}}\left( {\text{g}} \right) \to } \right\| - {\text{O}} - {\text{Al}}{\left( {{\text{C}}{{\text{H}}_{\text{3}}}} \right)_{\text{2}}} + {{\text{C}}{{\text{H}}_{\text{4}}}}\left( {\text{g}} \right),\end{align} \tag{ 3 }$$

$$\begin{align}\left\| { - {\text{Al}} - {\text{C}}{{\text{H}}_{\text{3}}} + {\text{ }}{{\text{H}}_{\text{2}}}{\text{O}}\left( {\text{g}} \right) \to } \right\| - {\text{Al}} - {\text{OH }} + {\text{ C}}{{\text{H}}_{\text{4}}}\left( {\text{g}} \right).\end{align} \tag{ 4 }$$

The formation of the strong Al–O bonds is the main driving force for the surface reactions, and the reaction enthalpy was reported to be one of the highest among all ALD reactions [4]. Both the above reactions are efficient and self-limiting, which makes the ALD of Al₂O₃ an almost ideal case for ALD.

The halogen transfer is another example of the ligand-exchange mechanism. The halogen transfer appears in the thermal ALD of W and Mo metals [36, 37], where Si₂H₆ is typically used with the metal fluorides of WF₆ and MoF₆. In these ALD processes, the silane can abstract the fluorine atoms from the surface, forming volatile fluorosilanes, such as SiF₄ and SiHF₃, along with H₂ and HF. For instance, the chemical reaction for the ALD of W from WF₆ and Si₂H₆ were found to be [38]

$$\begin{align}& \left\| { - {\text{W}} - {\text{SiH}}{{\text{F}}_{\text{2}}}^*{\text{ }} + {\text{ W}}{{\text{F}}_{\text{6}}}\left( {\text{g}} \right) \to } \right\| - {\text{W}} - {\text{W}}{{\text{F}}_{\text{5}}}^*{\text{ }} \nonumber\\ & \quad + {\text{ SiH}}{{\text{F}}_{\text{3}}}\left( {\text{g}} \right),\end{align} \tag{ 5 }$$

$$\begin{align} & \left\| { - {\text{W}} - {\text{W}}{{\text{F}}_{\text{5}}}^*{\text{ }} + {\text{ S}}{{\text{i}}_{\text{2}}}{{\text{H}}_{\text{6}}}\left( {\text{g}} \right) \to } \right\| - {\text{WW}} - {\text{SiH}}{{\text{F}}_{\text{2}}}^*{\text{ }} \nonumber\\ & \quad + {\text{ SiH}}{{\text{F}}_{\text{3}}}\left( {\text{g}} \right){\text{ }} + {\text{ 2}}{{\text{H}}_{\text{2}}}\left( {\text{g}} \right).\end{align} \tag{ 6 }$$

These reactions are highly exothermic, owing to the formation of the strong Si–F bonds. Simultaneously, the H₂ gas is produced by the reductive elimination of the hydrogen atoms of the unstable intermediate metal hydrides, which are formed by the fluorine-transfer ligand-exchange reaction. The reductive elimination also affords the desired metal films, and further details of the reduction mechanism will be discussed in section 2.4.

Another type of the ligand-exchange mechanism is based on the exchange of the ligands between two metal precursors, which appears in the ALD of Cu and Co metals [39, 40]. As an example, Cu metal can be deposited from bis(dimethylamino-2-propoxy)copper(II) [Cu(dmap)₂] and diethylzinc (DEZ). In this ALD process, the surface dmap groups from Cu(dmap)₂ are displaced by the ethyls from DEZ via the ligand exchange reaction, forming volatile Zn(dmap)₂. The afforded –Cu(C₂H₅) surface species are not stable and readily forms Cu metal and butane through the reductive elimination as the following:

$$\begin{align} & 2\left\| { - {\text{Cu}} - {\text{dmap }} + {\text{ Zn}}{{\left( {{{\text{C}}_{\text{2}}}{{\text{H}}_{\text{5}}}} \right)}_{\text{2}}}\left( {\text{g}} \right) \to {\text{2}}} \right\| - {\text{Cu}} - {\text{ }}\left( {{{\text{C}}_{\text{2}}}{{\text{H}}_{\text{5}}}} \right){\text{ }} \nonumber\\ & \quad + {\text{ Zn}}{\left( {{\text{dmap}}} \right)_{\text{2}}}\left( {\text{g}} \right),\end{align} \tag{ 7 }$$

$$\begin{align}2\left\| { - {\text{Cu}} - {\text{ }}\left( {{{\text{C}}_{\text{2}}}{{\text{H}}_{\text{5}}}} \right) \to {\text{2}}} \right\| - {\text{C}}{{\text{u}}^{\text{0}}} + {\text{ }}{{\text{C}}_{\text{4}}}{{\text{H}}_{{\text{10}}}}\left( {\text{g}} \right).\end{align} \tag{ 8 }$$

Notably, for all these ligand-exchange mechanisms, it is important that one of the exchanged products is sufficiently volatile to be liberated from the surface, as otherwise, nonideal issues would occur, as will be discussed later. Generally, the ligand-exchange mechanisms are essential for the majority of the ALD processes.

The chemisorption of the precursors can also follow a dissociative or non-dissociative mechanism. In the dissociative chemisorption scheme (figure 2(b)), the precursor molecule dissociates into two or more parts when it is chemisorbed on the surface,

$$\begin{align}\left| {\left| {{\text{ }} + M{L_n} \to {\text{ }}} \right|} \right|\left. { - M{L_{n - 1}} + } \right\|{\text{ }} - L.\end{align} \tag{ 9 }$$

This mechanism often occurs on the surfaces with catalytic activity. For instance, noble metals, such as Ru, have good catalytic nature, and therefore, the ALD processes to deposit Ru often involve the dissociative chemisorption of the ruthenium precursors [41, 42]. Similar processes were also observed on Ni [43, 44] and Cu [45] surfaces. The dissociative chemisorption can also occur on the SiO₂ surface, where strained siloxane (Si–O–Si) bridges are present. For instance, Al(CH₃)₃ can be dissociatively chemisorbed on SiO₂ via the following reaction:

$$\begin{align}& \left| {\left| { - {\text{Si}} - {\text{O}} - {\text{Si}} - } \right|} \right|{\text{ }} + {\text{ Al}}{\left( {{\text{C}}{{\text{H}}_{\text{3}}}} \right)_{\text{3}}} \to \left| {\left| { - {\text{Si}} - {\text{C}}{{\text{H}}_{\text{3}}} + {\text{ }}} \right|} \right| \nonumber\\ & \quad - {\text{Si}} - {\text{O}} - {\text{Al}}{\left( {{\text{C}}{{\text{H}}_{\text{3}}}} \right)_{\text{2}}}.\end{align} \tag{ 10 }$$

In comparison, in the non-dissociative chemisorption mechanism (figure 2(c)), the precursor molecule is adsorbed as a whole on the surface:

$$\begin{align}\left\| { - a + M{L_n} \to } \right\| - a - M{L_n}.\end{align} \tag{ 11 }$$

This mechanism usually occurs when the precursor and surface functional group can form a Lewis acid–base complex. It should be noted that the non-dissociative chemisorption is different from the physisorption, where only weak intermolecular force is present. Nevertheless, the physisorption is often required for a molecule to enter into the chemisorption potential well, especially for the cases without prior activation (e.g. plasma). In the non-dissociative chemisorption, a strong dative bond is formed, and it also has the surface saturation feature. For instance, Al(CH₃)₃ has an empty Al 3p orbital, which is capable of accepting additional electrons, and therefore, it can associatively chemisorb on the surface hydroxyl or siloxane bridge, forming a dative bond between the Al and O atoms [46, 47]:

$$\begin{align}\left\| { - {\text{OH }} + {\text{ Al}}{{\left( {{\text{C}}{{\text{H}}_{\text{3}}}} \right)}_{\text{3}}} \to } \right\| - {\text{OH}} - {\text{Al}}{\left( {{\text{C}}{{\text{H}}_{\text{3}}}} \right)_{\text{3}}},\end{align} \tag{ 12 }$$

$$\begin{align} & \left| {\left| { - {\text{Si}} - {\text{O}} - {\text{Si}} - } \right|} \right|{\text{ }} + {\text{ Al}}{\left( {{\text{C}}{{\text{H}}_{\text{3}}}} \right)_{\text{3}}} \nonumber\\ & \quad \to \left| {\left| { - {\text{Si}} - {\text{OAl}}{{\left( {{\text{C}}{{\text{H}}_{\text{3}}}} \right)}_{\text{3}}} - {\text{Si}} - } \right|} \right|.\end{align} \tag{ 13 }$$

The experimental identification of the species in the non-dissociative chemisorption is challenging, and most of the previous research focused on the theoretical calculations. Nevertheless, the non-dissociative chemisorption is often considered as the first step of the ALD surface reactions, and thus, it may have a profound effect to influence the subsequent reaction pathways.

The oxidation mechanism is closely related to the combustion reactions, which appear in many ALD processes to deposit metal oxides and noble metals. In this respect, strong oxidants, such as O₂, O₃, and O₂ plasma, are often used to combust the surface ligands, which normally produces CO₂ and H₂O as the byproducts [48, 49],

$$\begin{align}\left\| { - M{L_n} + {\text{ O*}} \to } \right.\left| {\left| { - M{\text{* ( or }}} \right|} \right| - M{\text{O*) }} + {\text{ C}}{{\text{O}}_{\text{2}}}{\text{ + }}{{\text{H}}_{\text{2}}}{\text{O}}.\end{align} \tag{ 14 }$$

The combustion naturally forms metal oxides on the surface. However, for the noble metals, their oxides are either unstable or can be efficiently reduced by the metal precursor in the subsequent half-cycle [50, 51]. Therefore, this approach has also been applied to the ALD of Ru [50, 52], Os [53], Rh [54, 55], Ir [56, 57], Pd [58, 59], Pt [60, 61], and Au [62, 63] metals.

The oxidation approach can also be applied to deposit metal oxides with high oxidation states. Some metal precursors have low oxidation states for their metal ions, and by using this oxidation approach, the oxidation states of the metals can be increased in the deposited oxides. One example is the ALD of In₂O₃ from cyclopentadienyl indium(I) (InCp), where indium is monovalent in the precursor and trivalent in the target oxide. To achieve the oxidation, H₂O and O₂ can together be used as the oxygen sources [64]. In this process, the surface –Cp groups are first released by H₂O, with no change on the indium oxidation state, and then O₂ oxidizes the indium ions from +1 to +3 (figure 3(a)). Another example is the ALD of SnO₂ from N²,N³-di-tert-butyl-butane-2,3-diamido-tin(II) and H₂O₂ [65, 66]. In this case, the stannylene molecule is a π-acid, and it can associate to a surface –OH group in the tin precursor half-cycle. Upon the H₂O₂ dose, the ligand bonded to the tin atom is first displaced by a hydroxyl group from H₂O₂, and then additional H₂O₂ can further oxidize tin(II) to tin(IV) via oxidative addition. It is worth noting that the final oxidation state depends on the exposure of the oxidant.

Zoom In Zoom Out Reset image size

The reduction mechanism is commonly involved in the ALD of metals, where the surface groups bonded to the metal atoms are removed by a reducing agent. H₂ gas is a typical reducing agent, and it has been used in depositing a variety of metals [67–73]. The corresponding reductive mechanism involves the elimination of the surface ligands by H₂, as expressed in the following:

$$\begin{align}\left\| { - M{L_n} + {\text{ }}n{\text{/2 }}{{\text{H}}_{\text{2}}} \to } \right\| - M{\text{* }} + {\text{ }}n{\text{ H}}L{\text{ }}\left( {\text{g}} \right).\end{align} \tag{ 15 }$$

Similarly, nitrogen-containing reducing agents, such as NH₃ [71, 72, 74] and N₂H₄ [75], can also be used to deposit metal films, but nitrogen impurities are often induced. Alcohols can also be used with metal β-diketonates to deposit the transition metals of Ni [76], Cu [77], and Pd [78]. During these processes, the alcohols produce reductive hydrogen radicals via the dehydrogenation reactions as catalyzed by the transition metal surfaces [79, 80].

Another type of the reduction mechanism is related to the reductive elimination of the surface species. For instance, the reduction of Ti upon the adsorption of the TiCl₄ was reported in previous literature, but the involved mechanisms were not shown [81]. More evidence of the reductive elimination was found for the metal amides. It was proposed that the ALD of TiN using Ti(NMe₂)₄ involves the reductive elimination of the dimethylamino ligands to afford hydrazine [82],

$$\begin{align}2\left\| { - {\text{Ti}}{{\left( {{\text{NM}}{{\text{e}}_{\text{2}}}} \right)}_{\text{2}}} \to {\text{2}}} \right\| - {\text{Ti}}\left( {{\text{NM}}{{\text{e}}_{\text{2}}}} \right){\text{ }} + {\text{ M}}{{\text{e}}_{\text{2}}}{\text{NNM}}{{\text{e}}_{\text{2}}}\left( {\text{g}} \right),\end{align} \tag{ 16 }$$

and the oxidation state of Ti changes from +4 to +3. The reductive elimination may also be accompanied with the β-hydrogen elimination to produce amine [82, 83]. As another example, in the aforementioned ALD process to deposit Cu (equations (7) and (8)), the intermediate surface ethyl groups are unstable and prone to the coupling to form the pure metal and volatile butane. A similar reductive elimination mechanism was suggested for the ALD of Co from CoCl₂(TMEDA) (TMEDA = N,N,N',N'-tetramethylethylenediamine) and Zn(DMP)₂ (DMP = dimethylaminopropyl). According to the DFT calculation, the reductive elimination pathway was most viable through the concerted C–C bond formation, and the β-hydrogen elimination may also take place in the metal reduction (figure 3(b)) [39]. However, many proposed pathways of the reductive elimination are not explicitly proved, and further investigations are still needed.

Plasma is sometimes used to assist the ALD, and such method is also known as plasma-assisted ALD (PALD). PALD utilizes the energetic species of radicals, ions, and electrons generated from the plasma as the reactants, and using the plasmas of O₂, NH₃, H₂S, and H₂, one can deposit oxides, nitrides, sulfides, and metals, respectively. The surface chemistry mechanisms of PALD are significantly complex, and they may consist of multiple mechanisms as discussed above. Owing to the high reactivity of the species from the plasmas, the surface reactions in PALD are usually easier to reach completion, and many low-reactivity metal precursors can be employed for PALD even at a low deposition temperature. For instance, with the O₂ plasma, some precursors (e.g. cyclopentadienyls and β-diketonates) that show low reactivity with H₂O or O₂ can also be used to deposit oxides; with the H₂ plasma, the hydrogenation of the precursor ligands is more efficient, which is helpful to deposit dense pure metal films. The diversity of the plasma also offers more tunability of the film composition. For instance, the N element can be incorporated into the films by adding N₂ to the O₂ plasma straightforwardly [84, 85]. As another example, Guo and Wang [86] employed the H₂S plasma to deposit the sulfur-rich phases of the metal pyrites of FeS₂, CoS₂, and NiS₂. The energetic reactive species in the plasma provided sufficient driving force to form the dimeric S–S moiety in the pyrites, which can hardly be achieved by the thermal ALD using molecular H₂S. On the other hand, the radicals and electrons may also activate additional reactions on the surface. For instance, the electrons from plasma can promote the desorption of hydrogen from the –NH₂ species on the surface [87]. It is worth noting that the presence of large amounts of gas-phase and surface species make the chemistry of PALD very complex. In some cases of using two reactant gases in the plasma, new molecules can be formed through gas-phase reaction or recombination [88]; the byproducts from the surface reactions may also be dissociated by the plasma when they are desorbed from the surface [17].

Notably, a concern about PALD is that the radicals generated from the plasma are prone to recombine. Therefore, compared to thermal ALD, the conformality of the film coating on high-aspect-ratio trench structures is sometimes compromised in PALD, owing to the severe radical recombination on the walls of the trench, as shown in figure 4(a). In addition to radicals, of which the diffusion is isotropic, the charged ions also play an important part in the PALD process. The charged ions can be accelerated in the direction perpendicular to the substrate surface within the plasma sheath, thus leading to the ion bombardment [90]. The ion energy is susceptible to the mean free path, plasma sheath thickness, and applied bias. Additional energy can be provided to the substrate surface, which may enhance the surface reaction or diffusion. Also, a recent study showed that the ions with low energy under mild plasma condition could significantly influence the deposition. As shown in figure 4(b), conformal deposition in the cavity with a high aspect ratio of 900:1 was successfully achieved [89]. However, in the area exposed to the anisotropic ions, a comparatively lower growth rate was observed. These results show the promise of using radicals to achieve good film step coverage. More methods, such as substrate biasing [91] and waveform tailoring [92], have been studied to modulate the ion energy. These methods are promising for better control of the film properties (e.g. crystallinity and conformality).

Zoom In Zoom Out Reset image size

Despite numerous merits of ALD to fabricate high-quality films, a large fraction of the reported ALD processes do involve nonideal issues, and therefore, a good understanding of these issues is critical for better film quality control. To this end, we elaborate some typical nonideal factors in ALD as follows, including nucleation delay, byproduct adsorption, ligand decomposition, agglomeration, and ion diffusion.

2.6.1. Nucleation delay.

The initial nucleation of the film growth is an important aspect in ALD. As shown in figure 5(a), the nucleation delay is a phenomenon that the apparent film growth rate was significantly slower in the initial stage than that in the later steady-growth stage, owing to a limited number of the nuclei formed initially on the surface. In ALD, the precursor usually needs to react with a functional group on the surface, and therefore lacking such functional groups on the initial substrate surface would result in significant nucleation delay for the film growth. For an inert surface, the nucleation usually starts from the surface defects, and in such cases, the initial ALD film growth is usually nonuniform and tends to form discontinuous islands [93]. These islands continue to grow in the follow-up ALD process and eventually coalescence to form a continuous film. It is worth noting that the nucleation delay is quite common for the ALD of oxides on the inert H-terminated silicon [94, 95]. Another example is the ALD of the platinum group metals on inert substrates using O₂ as the co-reactant [53, 96]. The lack of the activation of O₂ on the initial substrate hinders the chemisorption of metal precursor through oxidation mechanism, which results in the nucleation delay. Notably, because of the steric hindrance of the organic ligands, not all the active surface sites can react with the precursor molecule in each ALD cycle, and therefore, an ALD cycle usually affords less than one monolayer of the deposited material [35, 97, 98], and the island growth is sometimes inevitable.

Zoom In Zoom Out Reset image size

2.6.2. Byproduct adsorption.

In an ideal ALD process, the byproducts from the surface reactions should be sufficiently volatile to be completely liberated from the surface; however, this is not always the case in actual ALD processes. For instance, Zhao et al [99] found that for the ALD of NiS_x from bis(N,N'-di-tert-butylacetamidinato)nickel(II) [Ni(amd)₂] and H₂S, the reaction byproduct (N,N'-di-tert-butylacetamidine, Hamd) was not released from the surface during the H₂S half-cycle but formed a nonvolatile acid-base complex with the sulfhydryl on the surface:

$$\begin{align}\left\| { - {\text{Ni}} - {\text{amd }} + {\text{ }}{{\text{H}}_{\text{2}}}{\text{S}} \to } \right\| - {\text{Ni}} - {\text{SH}} \ldots {\text{Hamd}}.\end{align} \tag{ 17 }$$

Another common case is the ALD process with metal chlorides and H₂O [100]. The released HCl during the ligand-exchange can further react with the surface hydroxyl groups and even corrode the deposited films. Generally, the adsorbed byproducts can occupy the reactive sites on the surface (figure 5(b)) and therefore reduce the film growth rate [5]. Also, the adsorbed byproducts may be trapped in the deposited films or at the interfaces as impurities [101].

2.6.3. Ligand decomposition.

2.6.4. Agglomeration.

2.6.5. Ion diffusion.

In some ALD processes, the diffusion of the ions in the substrates or deposited films cannot be overlooked. For instance, Zhu et al [107] found that the ALD growth of NiS_x on a Co₉S₈ film could significantly reduce the Co amount in the Co₉S₈ film, and this phenomenon was because of the simultaneous occurrence of the gas–solid metal exchange reaction as

$$\begin{align}\left\| { - {\text{Co* }} + {\text{ Ni}}{{\left( {{\text{amd}}} \right)}_{\text{2}}}\left( {\text{g}} \right) \to } \right\| - {\text{Ni* }} + {\text{ Co}}{\left( {{\text{amd}}} \right)_{\text{2}}}\left( {\text{g}} \right),\end{align} \tag{ 18 }$$

together with the fast diffusion of the metal ions inside the Co₉S₈ film (figure 5(e)). The similar gas–solid metal exchange reactions were also observed in the ALD growth of Al₂O₃ on a ZnO film [108, 109]. This effect can be very important for the ALD of multi-element compounds (e.g. ternary or quaternary compounds) because the most common approach to deposit the multielement compounds is to alternately deposit binary compounds in supercycle manner. In another case, significant out-diffusion of the oxygen ions from the substrate was observed for the ALD of Al₂O₃ on the Sn-doped In₂O₃ (ITO) substrate [110]. The out-diffusion of the oxygen ions resulted in non-self-limited chemisorption of TMA, which substantially enhanced the initial growth of Al₂O₃ on ITO. Therefore, the ion diffusion effect should be carefully considered if the involved ions are possibly mobile in the substrates or deposited films.

2.6.6. Etching effect.

The etching effect may occur during the film deposition. This effect is common in PALD, in which the energetic ions from plasma can sputter off the films and thus etch the materials [91, 92]. The etching effect mainly depends on the plasma power and exposure in PALD, and therefore, the plasma parameters should be carefully tuned when depositing films. The etching effect may also occur through the chemical reaction to form volatile products (figure 5(f)), and a well-known example is the pulsing of Al(CH₃)₃ on the AlF₃ film surface. Al(CH₃)₃ would react with AlF₃ through the ligand-exchange to form the volatile AlF(CH₃)₂ [111]:

$$\begin{align}{\text{Al}}{{\text{F}}_{\text{3}}} + {\text{2Al}}{\left( {{\text{C}}{{\text{H}}_{\text{3}}}} \right)_{\text{3}}} \to {\text{3AlF}}{\left( {{\text{C}}{{\text{H}}_{\text{3}}}} \right)_{\text{2}}},\end{align} \tag{ 19 }$$

and AlF₃ is removed during the reaction. It is worth noting that the self-limiting etching of Al₂O₃ can be achieved by the fluorination followed by the etching reaction, and this strategy is extensively used in atomic layer etching [112, 113].

Metal halides, such as chlorides and fluorides, have been used as the precursors since the early days of ALD. The halides usually have a good thermal stability, which is desirable for ALD. As an example, TiCl₄ has a long history of use in the IC manufacturing for the deposition of TiN [201], which is the electrode material of DRAM. Because DRAM has a 3D structure with high aspect ratio, the excellent conformality feature of ALD is indispensable for this application. However, the deposition temperature was often high (>300 °C), and the deposited films usually contained halogen impurities. As another example, HfCl₄ can be used with H₂O to deposit HfO₂, which is a high-k dielectric material for CMOS devices. In the deposited HfO₂ films, the chlorine impurity was reported to be ca. 5% when the deposition temperature was 225 °C, and increasing the deposition temperature to 500 °C could significantly reduce the chlorine level [133, 202]. Also, the nucleation delay was observed in this process, and the HfO₂ films grew as islands on the H-terminated Si surfaces at the initial stage, which could cause large electrical leakage for the gate dielectric applications. To this end, additional surface treatments many be needed to remedy this issue. The silicon surface can be oxidized or nitrided to provide enough surface groups, which are typically reactive with the precursor [203, 204].

On the other hand, except for a few halides, such as WF₆, which is a gas at room temperature, the majority of the metal halides are not quite volatile and therefore require high-temperature heating to produce sufficient vapor for ALD. For example, InCl₃ and CuCl need to be heated to 275 °C and 340 °C, respectively, for ALD [205, 206]. In addition, the reaction byproducts of the halides include corrosive HF or HCl, which may further etch the substrates and deposited films, thereby deteriorating the uniformity and conformality of the ALD films. For instance, WF₆ and H₂ can be used to deposit tungsten metal, and the tungsten metal is used for the local interconnects in ICs. However, the HF generated by the ALD process of WF₆ and H₂ can corrode the SiO₂ dielectrics and cause electrical shorting between devices [207]. Despite of these issues, the halide precursors still have a significant role in manufacturing, and one of the most important considerations is their low cost.

Several metal alkyls are well-known for their excellent properties as ALD precursors, such as Al(CH₃)₃ and Zn(C₂H₅)₂. Al(CH₃)₃ is a typical metal alkyl precursor, which has widely used and comprehensively studied. Al(CH₃)₃ is highly volatile and can be used at room temperature for ALD. Al(CH₃)₃ is also highly reactive and can be used with H₂O [33, 35], H₂O₂ [208], O₃ [209, 210] and O₂ plasma [211] to deposit Al₂O₃. The involved gas–solid surface reactions are efficient and self-limiting, although some methyl ligands could persist on the surface when the deposition is below 300 °C using H₂O as the co-reactant [34]. Al(CH₃)₃ can also be used to deposit the III–V compounds of AlN [212], AlP [213], and AlAs [214] with the use of NH₃, PH₃, and AsH₃ as the co-reactants, respectively. Similar to Al(CH₃)₃, Zn(C₂H₅)₂ is also a desirable precursor, and it has been used to deposit ZnO [137], ZnS [215], ZnSe [216], and ZnTe [217] with the use of H₂O, H₂S, H₂Se, and H₂Te as the co-reactants, respectively.

As for the heavier IIIA elements of gallium and indium, the corresponding alkyl compounds, such as In(CH₃)₃ and Ga(CH₃)₃, have been used in semiconductor industry for the CVD of III–V materials (e.g. InP, InAs, GaN, and GaAs). However, compared to Al(CH₃)₃, In(CH₃)₃ and Ga(CH₃)₃ are less reactive with H₂O, and therefore, a stronger oxidant, such as O₃ [140, 141] or O₂ plasma [218, 219], is often needed for the ALD of the oxides (figure 6(a)). Particularly, for the ALD of Ga₂O₃ from Ga(CH₃)₃ and O₂ plasma, the crystal phase of the deposited Ga₂O₃ films could be controlled by the gas composition (Ar-diluted O₂), flow rate, and pressure during the plasma step, together with the deposition temperature (figure 6(b)) [218]. Sn(CH₃)₄ is also a reported ALD precursor; but its reactivity with H₂O is low, and therefore, a strong oxidant co-reactant, such as N₂O₄, is needed to deposit SnO₂ [142].

Zoom In Zoom Out Reset image size

It should be noted that although the abovementioned metal alkyls are successful for ALD, most of the other metal alkyls do not have sufficient volatility or thermal stability. Therefore, the use of metal alkyl precursors is still limited to only a few metals.

Metal cyclopentadienyls (Cp) have been used in a large number of ALD processes, and many of these compounds are fairly stability. For instance, the metal cyclopentadienyls (and their derivatives) have been quite successful for the ALD of the IIA-element compounds. As an example, MgCp₂ has been used with H₂O to deposit MgO, and the film growth rate was reported to be ca. 1.5 Å/cycle at 150 °C [220]. Similarly, Sr(ⁱPr₃Cp)₂ and Ba(Me₅Cp)₂ have been used with H₂O to deposit SrTiO₃ [221] and BaTiO₃ [222], respectively, and these compounds have also been used with H₂S to deposit SrS and BaS [223]. Notably, in these cases, the Cp rings are isopropyl- and methyl-substituted to prevent the oligomerization of these compounds, and this substitution strategy is fairly important for the bulky metal ions. Likewise, La(ⁱPrCp)₃ and Ce(Me₄Cp)₃ have been used for the ALD of La₂O₃ [224] and CeO₂ [225], respectively. A few late-3d-metal cyclopentadienyls, such as CoCp₂ [149, 226] and NiCp₂ [227, 228], have also been used for ALD, but they usually need a strong oxidant of O₃ or plasma (O₂, H₂, or NH₃ plasma) as the co-reactant to activate the surface reactions [229–231]. For instance, Co metal films can be deposited from CoCp₂ and NH₃ plasma, and in this process, the NH₃ plasma exposure generates a N-terminated surface, which can further react with CoCp₂ to form the Co–Co bonds and a series of volatile organic byproducts (figure 7(a)) [230]. As for the noble metals, their cyclopentadienyls (and derivatives) have been quite successful for ALD. Using O₂ as the reactant, Ru [151], Pt [232], and Ir [233] metals can be successfully deposited from their cyclopentadienyl precursors.

Zoom In Zoom Out Reset image size

It is worth noting that it is sometimes difficult to completely remove the cyclopentadienyl ligands on the surface by a Brønsted-acid co-reactant, because the hydrogenation of cyclopentadienyl involves the breakup of its aromatic conjugation. Therefore, carbon residue could be an issue for the deposited films [234], and even for some cyclopentadienyl compounds, H₂O is not sufficiently reactive to serve as the co-reactant [152].

Volatile metal alkoxides are also suitable ALD precursors. Ti(OⁱPr)₄ is a representative example and has been used for the ALD of TiO₂. Using H₂O as the co-reactant, the ALD growth rate of TiO₂ was found to be 0.2 Å/cycle at 150 °C–250 °C [235]. It is worth noting that the metal alkoxides can self-oligomerize in solution, and therefore, the deposition may be CVD-like at the low temperature near the precursor condensation point [236]. Ti(OMe)₄ is a more thermally stable precursor than Ti(OⁱPr)₄, and the decomposition does not occur significantly until 350 °C; using H₂O as the co-reactant, the ALD growth rate for TiO₂ was 0.55 Å/cycle at 250 °C–325 °C [237]. In addition, Ti(OⁱPr)₄ can be used in combination with TiCl₄ to directly deposit TiO₂ without using any other oxygen source. The reaction follows an alkyl-transfer mechanism, and the TiO₂ growth rate was 0.7 Å/cycle at 150 °C−250 °C [238].

Notably, the alkyl part of the alkoxide can be further engineered for the ALD applications. For instance, with an additional terminal dimethylamino group, the isopropoxide-derived ligand of dimethylamino-2-propoxide (dmap) can form chelated structures with metal ions (figure 7(b)), and the chelation can substantially enhance the stability of the formed metal–organic compounds. In this respect, Cu(dmap)₂ is a good example; as shown previously, Cu(dmap)₂ can be used with Zn(CH₂CH₃)₂ for the ALD of Cu metal [40]. In addition, it is worth noting that it is sometimes difficult to completely remove the surface alkoxide ligands during ALD, because the M–O bonds in the alkoxides are often quite strong. Therefore, oxygen and carbon impurities may be incorporated in the deposited films, and this issue can be critical for the ALD of pure metals [74].

Metal β-diketonates have been widely used as the ALD precursors. Many metals, including alkaline earth metals, transition metals and lanthanides, can form suitable β-diketonate compounds for the ALD applications. In a typical metal β-diketonate structure, the two ketone oxygen atoms coordinate to the center metal, forming a chelated six-member ring structure. The chelation can substantially enhance the compound thermal stability, and the side chains can be further tailored to promote the compound volatility. Common β-diketonate ligands include acetylacetonate (acac), 2,2,6,6-tetramethylheptane-3,5-dionate (thd), and 1,1,1,5,5,5-hexafluoroacetylacetonate (hfac) (figure 7(c)). The acac ligand is suited for small-size metals; whereas the bulky thd ligand is suited for large-size metals for preventing dimerization [239, 240]. The hfac ligand is a fluorine-substituted acac ligand, and the fluorine substitution can significantly enhance the compound volatility; however, it may also cause fluorine contamination in the deposited films.

Metal β-diketonates have been studied for the deposition of oxides [241–244], nitrides [245], sulfides [246–248], and metals [55, 59, 166, 167]. For the oxide deposition, owing to the stable chelated structures of the metal β-diketonates, the ALD processes usually need either high deposition temperature [249, 250] or a stronger oxidant, such as O₃ or O₂ plasma, as the co-reactant [241–244]. Also, pure nitrides and sulfides are not easy to achieve, because the surface β-diketonate ligands are difficult to be completely removed in these ALD processes [248, 251]. Noble metals, such as Rh [55], Pd [59], Ir [166], and Pt [167], can be deposited from the metal β-diketonates based on the combustion chemistry. In some cases, organic reducing agents, such as formalin [252], methanol [76], and hydroquinone [253], can be used with the metal β-diketonates to deposit metal films. For instance, Pd metal could be deposited from Pd(hfac)₂ and formalin, and the growth rate was found to be 0.35 Å/cycle at 200 °C [252]. As another example, Ni metal films could be deposited from Ni(acac)₂ with a primary alcohol at a temperature below 300 °C [76]. As shown in figure 7(d), three primary alcohols of ethanol, propanol, and methanol were evaluated, and methanol gave out the best results, as the deposited films were high-purity metallic Ni with low resistivity. In addition, the thd-based metal β-diketonates can also be used with TiF₄ or TaF₅ to deposit metal fluorides, such as MgF₂, CaF₂, and LaF₃ [254].

Metal amides generally have a high reactivity, and therefore, if they are volatile, they can be used for low temperature ALD (<100 °C). The low ALD temperature can allow for the films to grow on thermally unstable substrates, such as polymers, plastics, and photoresists [255]. As an example, M(NMe₂)₄, M(NMeEt)₄, and M(NEt₂)₄ (M = Hf, Zr) have been used as the hafnium and zirconium precursors to deposit ZrO₂ and HfO₂ [172]. V(NMe₂)₄ has been used with H₂O or O₃ to deposit VO_x films at 50 °C–200 °C (figure 8(a)) [256, 257]. All these precursors are highly volatile and reactive. When using H₂O as the co-reactant, the ALD surface reactions follow a proton-transfer ligand-exchange scheme, where the alkylamide ligand accepts a proton from the surface –OH, affording the alkylamine as the byproduct, while the metal ion bonds to the surface oxygen. This exchange reaction is highly exothermic, as the metal–oxygen bond is stronger than the metal–nitrogen bond and the amine is more basic than the alcohol. However, the thermal stability of the metal alkylamides is generally limited, and therefore, high deposition temperature may lead to the partial decomposition of the metal alkylamides and therefore increase the carbon and nitrogen impurity levels in the deposited films.

Zoom In Zoom Out Reset image size

The thermal stability can be improved using a diamide ligand to form a chelated cyclic structure with the metal ion [258, 260]. As shown in figure 8(b), (N,N'-di-t-butyl-2-methylpropane-1,2-diamido)tin(II) has been used as the tin precursor for depositing SnS. Using H₂S as the co-reactant, the SnS growth rate was approximately 0.12 nm/cycle at the deposition temperatures of 75 °C–175 °C [258].

Metal amides have also been applied to the ALD of metal nitrides. Using Ti(NEt₂)₄ and NH₃ plasma, TiN films with a low resistivity of ca. 250 μΩ cm were deposited at 300 °C; however, higher carbon concentration was observed at lower deposition temperatures [261]. Ta(NMe₂)₅ was used with H₂ plasma to deposit TaN [201]. The TaN film growth rate was 0.56 Å/cycle at 225 °C, and the film electrical resistivity was 380 μΩ cm. Long H₂ plasma pulse time was found to be important to remove the carbon residue in the films. ALD of GaN was achieved by using [Ga(NMe₂)₃]₂ and NH₃ plasma [262]. The self-limiting growth behavior was observed at 130 °C–250 °C, and the film growth rate was 1.4 Å/cycle. The GaN films deposited on the Si(100) substrate were crystalline with an atomic Ga/N ratio of 0.97, and the carbon and oxygen impurity concentrations were 2.8 at% and 3 at%, respectively.

Metal silylamides have also been used for ALD to deposit various materials, such as Li₃N [263], La₂O₃ [264], ZrO₂ [265], and PbI₂ [266]. The metal silylamides are highly reactive but not quite thermally stable, and the decomposition of the metal silylamides would leave silicon impurities in the films. For instance, in the ALD ZrO₂ films deposited from ZrCl₂[N(SiMe₃)₂]₂ and H₂O, the silicon concentration was found to be 1.1–5.4 at% for the deposition temperatures of 150 °C–350 °C [265]. On the other hand, silicon-containing W_x Si_y N thin films could be deposited from W(N^t Bu)₂[N(SiMe₃)₂]₂ and H₂ plasma (figure 8(c)) [259].

Metal amidinates have been widely used as the ALD precursors. The amidinate ligands can bidentately coordinate to the metal ions to form chelated four-member ring structures. The chelation can considerably enhance the thermal stability, and therefore, the metal amidinates are generally more thermal stable than the metal amides. The chelation of the amidinates resembles that of the β-diketonates, but the nitrogen–metal coordination bonds are much weaker than the oxygen–metal bonds in β-diketonates, and therefore, the metal amidinates are usually of better reactivity. The amidinate side chains and backbones can be further substituted to modulate the volatility and reactivity of the complexes. Particularly for large-size metals, introducing bulky side alkyl chains is important to prevent oligomerization.

Metal amidinates have been used for the ALD of transition metals [73, 267], oxides [181–184], sulfides [86], and selenides [268]. For instance, Cu, Co, Fe, and Ni metals can be deposited from their N,N'-dialkylacetamidinate compounds together with the use of H₂ as the co-reactant [73]. As another example, cobalt oxides can be deposited from Co(ⁱPr₂AMD)₂ by using H₂O or O₃ as the co-reactant [267]. As shown in figure 9(a), using H₂O, rock-salt CoO films could be deposited at 150 °C–200 °C with a growth rate of 0.45 Å/cycle; whereas using O₃, spinel Co₃O₄ films could be deposited at 200 °C–225 °C with a growth rate of 0.5 Å/cycle. As for the metal sulfides, M(^tBu₂AMD)₂ (M = Fe, Co, Ni) could be used with molecular H₂S to deposit FeS_x [269], Co₉S₈ [185], and NiS_x [103], and they could also be used with H₂S plasma to deposit the metal disulfides of FeS₂, CoS₂, and NiS₂ [86]. Recently, these compounds were also used with diethyldiselenide (DEDSe) and Ar/H₂ plasma to deposit the metal selenides of FeSe₂, CoSe₂, and NiSe₂ [268].

Zoom In Zoom Out Reset image size

Metal amidinates have also been used as the precursors for varieties of lanthanides and basic metals. La(ⁱPr₂FMD)₃ is a typical example. This compound is fairly volatile and has been extensively used for the ALD of the lanthanum-based oxides [193, 270, 271]. As for the basic metals, In(ⁱPr₂FMD)₃ [187] and Sn(ⁱPr₂AMD)₂ [260] have been reported as the precursors. For instance, In(ⁱPr₂FMD)₃ could be used with H₂O to deposit high-purity In₂O₃, and the film growth rate was 0.55 Å/cycle at 150 °C–275 °C (figure 9(b)) [187].

Heteroleptic metal precursors are rapidly emerging in recent years. Two or more different ligands are bonded to the center metal atom in heteroleptic metal precursors, and a combination of the best properties of the respective parent homoleptic compounds can possibly be obtained [272]. For instance, the thermal stability of the metal precursors may be improved by incorporating the halide ligands; ZrCp₂Cl₂ [195] and HfCp₂Cl₂ [196] are thermally stable, without losing the self-limiting growth nature even at 300 and 350 °C, respectively. Cyclopentadienyl-alkylamido compounds have also been synthesized for zirconium [273], hafnium [274] and titanium [197]. Compared to their homoleptic metal amides, better thermal stability was obtained and the reactivity was comparable when using O₃ as co-reactant to deposit the oxide films. A recent study suggested that the cyclopentadienyl-alkylamido compounds chemisorbed to the film surfaces mainly through the reactions of the alkylamide ligands, and therefore, H₂O was difficult to be used as the co-reactant to deposit the oxide films [197].

The asymmetry of the heteroleptic compounds also helps the precursors to remain in liquid form. For instance, M(ⁱPrCp)₂(ⁱPr₂AMD) (M = Y, La, Pr, Gd, and Dy) have been used to deposit the rare-earth oxide films [198, 199]. All these precursors are liquid and highly volatile. Using H₂O as the co-reactant, the oxide films could be deposited at no higher than 200 °C. It is worth noting that the hygroscopicity of the lanthanide oxides may lead to the loss of the self-limiting growth, and this problem can be solved by introducing O₃ as the oxygen source [199].

Notably, in some cases, the heteroleptic metal precursors do not exhibit improved properties, or even worse, some undesired properties of the respective homoleptic precursors may be introduced. Generally, the understanding about the chemistry of the heteroleptic compounds is still very limited, and the surface chemistry investigation is much needed.

Aside from the metal precursors, the precursors for the non-metal elements are also important for ALD. For instance, to deposit metal oxides, a variety of oxygen sources can be chosen. If the metal precursors are sufficiently reactive, H₂O is the most common oxygen source, and alcohols can sometimes be used. However, if the reactivity of the metal precursors is low, a strong oxidant, such as H₂O₂, O₂, O₃, or O₂ plasma, is usually needed. Notably, H₂O₂ and O₃ are prone to decompose at high temperature, and the deposited films may also catalyze this process [152], which would deteriorate the uniformity and conformality of the deposited films. As for the O₂ plasma, the O radicals in the plasma are reactive species, but the radicals are prone to recombine at the substrate surface and therefore possibly difficult to reach the bottom of a very deep trench. Therefore, conformal coating on high-aspect-ratio trenches may not be easy to realize. The O₂ plasma also contains oxygen ions, of which the motion can be anisotropic and therefore affect the deposition conformality [89].

The nitrogen precursors for the ALD of metal nitrides include NH₃ [212, 275], hydrazine [276], N₂ plasma [277, 278], and NH₃ plasma [261, 262]. NH₃ is commonly used, given that the metal precursors are sufficiently reactive [212, 275]. Some ALD processes use hydrazine [276], which is more reactive than NH₃, but this compound is toxic and explosive. N₂ or NH₃ plasma can be used to enhance the surface chemistry reactivity [262, 278]. In some cases, particularly for the deposition of low-resistivity metallic TiN and TaN_x , H₂ plasma or mixed H₂:N₂ plasma can be used to reduce the carbon impurity level in the deposited films [201, 278, 279].

As for chalcogenides, H₂S and H₂S plasma are often used to deposit the metal sulfides, despite that the H₂S gas is toxic and explosive. Recently, an organosulfur precursor of di-tert-butyl disulfide (TBDS) was reported to replace H₂S for the sulfide ALD [280]. Also, the alkylsilyl compounds of (Et₃Si)₂Se and (Et₃Si)₂Te are often used for the ALD of metal selenides and tellurides, respectively [134].

As for pnictogenides, PH₃ and AsH₃ have been used to deposit the metal phosphides [213] and arsenides [214], respectively. As for fluorides, HF is extremely toxic and corrosive, and therefore, alternatives, such as TiF₄ and TaF₅, have been used to deposit the metal fluorides [254].

With the fast development of artificial intelligence, the demand for computing power has increased significantly. In the past decades, the critical size of the transistors has been continually shrinking to boost the performance of the ICs. However, as the traditional Si-based transistor approaches its limitation of the critical size downscaling, new materials are needed to be introduced to the Si technology to further improve the ICs. The desired materials should have good immunity to short-channel effect, high mobility, and low contact resistance to achieve high-performance transistors. More importantly, the fabrication of the materials should have good compatibility with the BEOL process and the advanced device structures, such as FinFET and GAAFET. Oxide semiconductors, such as In₂O₃ [291–293], InGaO (IGO) [294–296], InZnSnO (IZTO) [297], and InGaZnO (IGZO) [14, 298, 299], have outstood as the promising materials for the new transistor and IC technology. The FETs made of oxide semiconductors have extremely low off-current and a reasonably high electron mobility (>10 cm² V⁻¹ s⁻¹). More importantly, the fabrication of these oxide-based transistors needs fairly thermal budge, which is compatible with the BEOL requirement (⩽400 °C), and therefore, the oxide-based transistors are possible to be built directly on the traditional Si-based CMOS circuits to achieve 3D circuit integration. The ALD technique has been considered as a pivotal approach to fabricate these oxide-based transistors for its low process temperature and precise control of the film thickness and composition. With careful engineering of the involved surface chemistry and precursor chemistry, high-performance oxide transistors can be achieved by ALD. For instance, Si et al [291] demonstrated that using the ultrathin In₂O₃ (0.50 nm) channel layers made by ALD, the FETs with the channel length of 8.0 nm were able to achieve a very high on/off ratio of 10¹⁰ (figure 10(a)). Also, Cho et al [299] showed that the transistors using the amorphous IGZO (a-IGZO) channel layers made by ALD could achieve a high electron mobility of 36.6 cm² V⁻¹ s⁻¹ and a good bias stress stability.

Zoom In Zoom Out Reset image size

Aside from the channel layer, the gate-dielectrics can also be made by ALD, and the afforded dielectric–channel interface can be achieved with very low trap density. For instance, Li et al [300] used ALD to deposit 4 nm Al₂O₃ as the gate-dielectric for the a-IGZO transistors, and owing to the good quality of the Al₂O₃/a-IGZO dielectric–channel interface, the afforded transistors showed a fairly low subthreshold swing of only 60.9 mV dec⁻¹ and a very good bias stress stability (figure 10(b)). Cho et al [298] also reported high-performance transistors based on the ALD-prepared IGZO channel and stacked HfO₂–Al₂O₃ gate-dielectric, and the transistors exhibited a high electron mobility of 74.0 cm² V⁻¹ s⁻¹ with negligible hysteresis (figure 10(c)).

As a step forward, both the oxide channel and gate-dielectric layers can be made by ALD. With the unique feature of ALD for conformal coating, both the layers can be deposited on vertical side walls or in hole structures to realize vertical-structure transistors. For instance, Baek et al [297] showed that vertical oxide transistors could be fabricated on side walls based on the ALD-prepared IZTO channel layer and Al₂O₃ dielectric layer (figure 10(d)). Also, Chen et al [15] demonstrated that vertical channel-all-around (CAA) FETs could be fabricated in hole structures by ALD. Compared to the traditional planar FETs, the vertical CAA FETs have a much smaller lateral size, which can be utilized to fabricate high-density 2T0C DRAM with 4F² cell structure. All these results have shown the great promise of ALD to prepare high-performance oxide-based transistors.

As the traditional transistor reach its physical limitation, the novel computing architectures, such as in-memory computing, are also attractive to boost the performance of the ICs beyond the Moore's law. Ferroelectricity is featured for the presence of two states of spontaneous electric polarization that can be reversibly switched by an applied electric field. The two-state behavior can be utilized for the binary data processing in digital computing, which perfectly meets the requirements of the in-memory computing. The HfO₂-based ferroelectric materials can be deposited by ALD, which is well compatible with the contemporary IC fabrication, and therefore, the HfO₂-based materials are highly promising for building new devices based on the ferroelectricity [301]. The ferroelectricity of the ultrathin HfZrO (HZO) films deposited by ALD was confirmed by Cheema et al [302]. The composition of the HZO films could be tuned by the subcycle numbers for HfO₂ and ZrO₂, and switchable polarization was observable even for the HZO thickness downscaling to 1 nm (figure 11(a)).

Zoom In Zoom Out Reset image size

The HfO₂-based ferroelectric materials are well suited for non-volatile memories for their low power consumption to switch the polarity. This new type of memories has a higher density than the traditional DRAM. Luo et al [303] fabricated the ferroelectric HZO diodes by ALD and implemented these diodes in an eight-layer 3D array for memory applications. The diode memory devices showed a high operation speed of 20 ns and robust endurance of >10⁹ (figure 11(b)). Another attractive application is the ferroelectric FETs, in which the transistor threshold voltage can be controlled by the polarization state of the ferroelectric layer. For instance, Halid et al [304] showed that the ALD Si:HfO₂ could be implemented to fabricate ultrascaled ferroelectric FETs, and domain-level ferroelectric switching could be directly observed (figure 11(c)). Liao et al [305] demonstrated that 3D GAA nanosheet ferroelectric FETs could be made based on the ALD HZO, and the devices showed a large memory window of 1.3 V, robust endurance cycles of >10¹¹, and stable data retention of >20 000 s. All these results indicate that, with the ALD technology, the dimensional scaling of the ferroelectric devices is highly promising for future advanced memory and computing technologies.

As the technology node of integrated ICs continues to shrink down to 3 nm and beyond, the increasing density of the transistors has led to a significant increase in the interconnect complexity, and this complexity presents serious challenges to the BEOL interconnect fabrication. In particular, the linewidths of the local interconnects in the first few interconnect layers are less than 20 nm and their aspect ratios are more than 4:1. Hence, it is necessary to achieve uniform sub-nm thicknesses for the barrier/liner/seed layers in trenches and vias to ensure enough space for electroplating the metal wires. This is obviously a great challenge for the traditional PVD because of its tendency to form overhangs, as shown in figure 12(a) [306]. Severe overhangs leave only a small gap space to fill the metal, which would result in the increase of the line resistance and resistance–capacitance delay, and also reduce the interconnect reliability. ALD has the advantages of high uniformity, excellent step coverage, and atomic-scale thickness control, which makes it one of the key technologies for advanced interconnect fabrication, as shown in figure 12(a). To be compatible with the process temperature limit of BEOL (⩽400 °C), the thin-film coatings can be prepared by PALD. In the damascene Cu process, TaN, Ta, and Cu metals are used as the barrier, liner, and seed layers, respectively [307]. All these metals can be prepared by PALD. For instant, low-resistivity TaN films can be deposited from tertbutylimidotris(diethylamido)tantalum (TBTDET) and H₂ plasma [308]. The Ta films can be deposited from TaCl₅ and H₂ plasma [128, 309]. The high-quality Cu seed layers can be deposited from Cu(acac)₂ for interconnects [310]. In another case, Guo et al reported that the Cu thin films could be deposited from [Cu(ⁱPr₂AMD)]₂ and H₂ plasma at temperature as low as 50 °C, and the properties of the ALD Cu films, such as continuity, uniformity, and resistivity, were comparable to the PVD Cu films (figure 12(b)) [186].

Zoom In Zoom Out Reset image size

As the Cu interconnect technology advances, Co and Ru have been developed to replace Ta as the liner material. Co and Ru have a better wettability with Cu, which can facilitate the filling of Cu during the electroplating, and they can also improve the overall integrity of the TaN barrier [312, 313]. For example, Park et al demonstrated that the Co films deposited from dicobalt hexacarbonyl tert-butylacetylene (CCTBA) and H₂ plasma could be used for direct Cu electroplating (figure 12(c)) [311]. Swerts et al reported that the Ru films could be deposited from (MeCpPy)Ru and N₂/NH₃ plasma at 330 °C for interconnect applications [314]. In the advanced interconnect technology, Cu wires have been gradually replaced by Co or Ru for high-conductivity narrow interconnects [315]. The Co interconnects have been used in the advanced nodes of chip fabrication, while the Ru interconnects are still at the laboratory research stage. The barrier and liner materials of these two interconnects are TiN and Ti, respectively, which can also be obtained by PALD [128, 316]. Therefore, ALD is a critical fabrication technology for the advanced IC interconnects.