Next generation nanopatterning using small molecule inhibitors for area-selective atomic layer deposition

Currently, the fabrication of most nanoelectronics such as those in the semiconductor industry employs top-down processing techniques that are generally based on repeated steps of deposition, photolithography, and etching. However, with continued device downscaling, this top-down approach is beginning to face challenges in pattern resolution and alignment.^1,2 To circumvent this bottleneck in device fabrication, alternative processing techniques are being developed. One promising alternative relies on atomic layer deposition (ALD), which is a thin film deposition technique that provides atomic-level thickness control and high conformality on nanostructured samples. ALD processes are based on alternating exposures of a precursor and a coreactant that react with the surface in a self-terminating fashion. By repeating these alternating reactions in a cyclic manner, ALD results in the layer-by-layer deposition of a desired material on the substrate.^3–5 Due to the surface-dependent nature of ALD chemistry, the differences in local surface properties can be exploited to deposit a film in an area-selective manner, in which deposition occurs on some regions of a patterned substrate and not others. As such, area-selective deposition (ASD) represents a self-aligned, bottom-up fabrication technique that can overcome key challenges facing the semiconductor industry.^6–9 

In a selective process, there must be a surface or surfaces upon which deposition does occur (the growth surface, GS) and one or more upon which no deposition occurs (the nongrowth surface, NS). This selectivity can be quantified using the following equation based on a formalism introduced by Gladfelter:

where S_x represents the selectivity after x number of ALD cycles and R represents the atomic composition of the deposited material on either GS or NS.¹⁰ S_x can range from 0 (no selectivity) to 1 (perfect selectivity). Numerous factors related to an ASD process will affect what degree of selectivity is achievable, and much effort has gone into maximizing selectivity in ASD. There are many different means of tuning a process to be selective. One method of achieving ASD is to rely directly on the aforementioned differences in substrate character and the corresponding variations in how reactants adsorb at those substrates, in order to achieve selective precursor or coreactant adsorption. This technique relies on the inherent selectivity of the system.^11–13 Albeit useful to consider how the inherent selectivity can be leveraged to impede undesired deposition, generally only a short growth delay on the NS and thus a small selectivity window is observed for such systems. For more industrially relevant applications, which require high selectivity for thicker films, more proactive techniques will likely be required.^1,6

One of the most prevalent and versatile approaches for ASD on patterned substrates relies on surface functionalization with self-assembled monolayers (SAMs), which are used to deactivate the surface to ALD growth. SAMs are organic films composed of amphiphilic molecules that possess three key constitutive parts: a reactive headgroup that facilitates binding to the surface potentially in a selective manner, a tail group that makes the film inert to ALD chemistry, and a backbone that promotes the formation of a densely packed monolayer through van der Waals dispersion forces.^14–19 SAMs block ALD by preventing diffusion of ALD precursors to the substrate surface and by effectively replacing the active sites with which ALD precursors would react—thereby inhibiting nucleation and growth.^14–17,19 SAMs can be chosen to block deposition on a variety of different surfaces by altering the reactive headgroup functionality, and their success in ASD is well-established.^14,19–21

However, SAMs possess certain disadvantages that limit their compatibility with high-volume manufacturing (HVM) and may make them unsuitable for sub-10 nm device applications. Not all SAMs can be readily deposited on the substrate by vapor phase delivery because they have low volatility due to their large size. As such, many SAMs must be applied via wet chemistry, which limits their applicability and impedes their integration with vapor phase based-HVM processes.^8,22–24 Additionally, the drive to continue device downscaling potentially requires the utilization of materials with sub-10 nm, high aspect ratio patterned features. However, due to the relatively large size of SAM molecules, SAMs face challenges facilitating ASD around these features, as they may block growth not only on the NS but also on neighboring regions of the GS as well. Furthermore, around these small features, the SAMs may not block deposition at the NS due to edge or corner effects that limit their ability to crystalize and establish an effective inhibitor layer.^18,22,25 As such, SAMs may not provide the precision necessary for ASD purposes in sub-10 nm device applications.^1,26 Therefore, efforts have been growing in finding an alternative to SAMs that can address these issues while still successfully enabling ASD.

One solution to these challenges is to employ small molecule inhibitors (SMIs) in lieu of SAMs. Due to their high volatility, SMIs can be delivered in the vapor phase for adsorption on the substrate. Thus, the formation of the inhibitor layer on the NS does not require wet chemistry and can be more easily integrated into the overall ALD process, which is more conducive to HVM. Furthermore, any damage that the inhibitor layer incurs during the ALD process may, in principle, be repaired by vapor phase reapplication of the SMI in vacuo.^23,27,28 This SMI reapplication process has two primary benefits to the overall ALD process. First, by mending the inhibitor layer by reapplication, the selectivity can potentially be extended to higher ALD cycle numbers. Second, ALD processes that would have been deemed incompatible with a static inhibitor layer—such as those that involve the use of plasma or ozone, which are known to damage inhibitor layers—may be reconsidered if in vacuo regeneration of the inhibitor layer can be achieved. In this way, using SMIs may open up more ALD processes and new materials for ASD.²³ Finally, because they lack the steric bulk that SAM molecules possess, SMIs may better support ASD on substrates with sub-10 nm, high aspect ratio patterned features by providing a higher degree of deposition precision.¹ Therefore, SMIs hold great potential in the field of ASD.

The potential for SMIs has already been demonstrated for ASD purposes in both chemical vapor deposition (CVD) and ALD literature. For instance, Suh et al. recently investigated the use of 4-octyne as a coadsorbate acting as an SMI in the CVD of ZrO₂ thin films on SiO₂ in the presence of Cu and showed that selective ZrO₂ deposition on the SiO₂ growth surface could be achieved for upward of 30 nm.²⁹ In ALD literature, Khan et al. showed that short chain aminosilanes could be used to facilitate the selective deposition of Ru and Pt on H-terminated Si in the presence of SiO₂.³⁰ Additionally, Merkx et al. has explored the use of two different SMIs for ASD processes: the beta-diketonate acetylacetone (Hacac) for the selective deposition of SiO₂ on various dielectrics in the presence of other dielectrics and the aromatic amine aniline for the selective deposition of TiN on a dielectric in the presence of a metal.²⁸ 

With their promising potential, SMIs merit further investigation. As of now, little work has been conducted to understand the mechanisms by which these small molecules adsorb and inhibit deposition. However, based on what is known about the current SMIs being studied, in conjunction with the extensive knowledge held about SAMs, it is possible to derive some guiding principles for choosing SMIs to facilitate ASD in a particular system. In this Perspective, the two key interfacial phenomena that impact ASD will be discussed: interactions at the inhibitor-ALD precursor interface and interactions at the inhibitor-substrate interface, both shown in Fig. 1.

By looking closely at these interfacial phenomena, a better understanding of the role that SMI composition and structure plays in adsorption and inhibition can be attained, which will ultimately inform SMI design for future ASD systems. Additionally, because it represents an important feature of the small molecule inhibitors, SMI reapplication will also be discussed. Finally, this Perspective will conclude with an outlook on the use of SMIs and their role in new ASD processes.

The mechanism by which SMIs work is informed by understanding how SAM molecules function in SAM-based ASD. In SAM-based ASD, the SAMs deactivate surface reactive sites and replace them with inert termination groups. For instance, alkylsilane SAM molecules have been shown to readily react with surface hydroxyl groups on SiO₂ substrates, thereby inhibiting reactions between those surface groups and ALD precursors.³¹ Similarly, alkylphosphonic acid SAM molecules react with various metal oxide substrates, again replacing potential reactive sites with their associated inert alkyl groups.¹⁴ While deactivating surface reactive sites by reaction with the SAM headgroup has inspired many studies of SAM molecules with varied chemisorption functionalities (phosphonic acids, trifunctional silanes, thiols, etc.) in ASD, inhibitor selection based on chemisorption functionality is just one factor that goes into preventing undesired interactions at the inhibitor-precursor interface.

Another key factor that influences the ability of SAMs to block ALD is their capacity to form a dense organic layer. In fact, the ability of SAMs in the systems described above to restrict precursor access to the inhibitor-precursor interface is primarily attributed to the formation of semicrystalline organic regions comprised of tightly packed alkyl chains.^15,18 Undesired growth (“defect sites”) caused by interstitial or substrate-level ALD nucleation has been shown to more easily develop when the backbone chains are unable to form a well-ordered organic layer.¹⁴ Such sites are often pinholes or regions of poor packing. The formation of the semicrystalline blocking layer is in turn affected by the length and type of the SAM backbone, the inertness of its tail, and the areal density of the SAM molecules. Many studies have shown that the degree of order of a SAM is strongly dependent upon the coverage and on the chain length of the backbone.³² For example, a recent theoretical study by Clerix et al. of copper thiolate SAMs showed that while surface saturation of smaller chain thiolates (n = 1, 2) is purely limited by steric hindrances, longer chain thiolate SAMs (n = 6, 12) undergo a phase change from lying parallel to the surface of Cu[100] at low coverages, to tightly packed, upright structures at full surface saturation.³³ Furthermore, a number of ASD studies using SAMs have collectively drawn attention to aliphatic chain packing, which leads to the production of the semicrystalline layer, as a major factor in promoting selective deposition.^18,34–36 For example, a study by Chen et al. supported the idea that increasing the SAM order led to increased selectivity when the authors showed that longer chain alkyltrichlorosilane molecules, which are known to form more well-packed SAMs, provided better inhibition of HfO₂ ALD.¹⁸ Connections between SAM order and subsequent inhibitive performance were also shown in the same work by Chen et al., specifically that alkyl chain backbones in the SAMs blocked HfO₂ ALD better than bulkier backbones such as those with phenyl group constituents.¹⁸ However, to be effective in blocking, the organic layer must not only contain the right type of backbone to achieve dense packing, but it must also have an effective terminal functionality and sufficiently high areal density.

A substantial body of work has supported the use of inert methyl groups for the SAM termination in order to achieve good blocking of ALD.^31,37,38 Killampalli et al. demonstrated this concept experimentally when they explored reactions between tetrakis(dimethylamido)titanium (TDMA-Ti) and alkyltrichlorosilane SAMs with -OH, -NH₂, and -CH₃ termination.³¹ In that work, they showed the facility of TDMA-Ti to react with OH- and NH₂-terminated SAMs over those with CH₃ termination. Furthermore, their work also suggested that ALD nucleation on SiO₂ substrates with CH₃-terminated SAMs occurred at the SAM-substrate interface, indicating that ALD growth occurred when TDMA-Ti was more easily able to diffuse between the SAM layer and access underlying reactive sites. Other work showed the importance of maximizing the areal density of SAM molecules for increased blocking efficiency. For example, Hong et al. studied the connection between formation time for chlorosilane SAMs and blocking efficiency for HfO₂ ALD, in which longer formation time led to more densely packed SAM molecules.³⁹ Their work showed a positive correlation between these two effects, up to a saturation point where increased formation time for the octadecyltrichlorosilane (ODTS) SAM no longer led to increased blocking ability. These results suggest that high blocking performance is inherently tied to the formation of an inhibitor layer with maximized areal density, where inert methyl groups on inhibitors replace hydrophilic reactive sites on the original substrate.

In addition to the areal density of inert tail groups, the stability of the SAMs is another factor that has been shown to impact selectivity, with literature studies specifically considering the conditions for a particular ALD process, the ALD precursors being used, and the inherent ability of the SAM to resist defect formation.^40,41 Instability may lead to induced disorder of the organic layer as well as desorption of the SAM in some cases. Thiol SAMs are notably susceptible to issues related to thermodynamic stability, with alkanethiols such as octadecanethiol and dodecanethiol (DDT) shown to decompose or desorb at temperatures (110–120 °C) that are relatively mild compared to most ALD processes.^42,43 ALD precursor choice can also negatively affect SAM stability depending on the SAM molecule being used. For instance, using ALD precursors with chloride ligands may destabilize SAMs that adsorb to metal oxides because HCl gas, which is a primary by-product when water is used as a coreactant with chloride precursors, can etch the surface. For SAMs to adequately resist displacement under ALD conditions, they must form strong surface bonds in high density on the underlying substrate, such that surface defects are difficult to produce.

The extensive research carried out on SAMs as inhibitors for ASD ultimately helps identify two important factors in precursor-inhibitor interactions for promoting selectivity. First, preventing access points that lead to defect formation within the inhibitor layer is essential for maintaining blocking performance. Second, this defect prevention is most readily realized by maximizing the areal density of inert blocking groups at the inhibitor-precursor interface.

From the understanding gleaned from the SAM-based ASD literature, it is clear that a primary consideration in designing an inhibitor is the blocking of active sites at the substrate. The inhibitor layer must successfully prevent the ALD precursor from reacting with the substrate surface by blocking the sites with which it may react. Additionally, this inhibitor layer must not be prone to defect formation during the ALD process. While active site blocking is the primary means of preventing deposition for both SAMs and SMIs, it has been observed that the act of blocking can be more challenging for SMIs due to interactions between the ALD precursor and the inhibitor that can lead to degradation of the passivation layer.⁴⁴ 

The disparity in blocking performance between SAMs and SMIs underscores the key differences between these inhibitor types. Here, we define an SMI as a molecule that is easily vaporizable, limited in size by subnanometer dimensions, and characterized by a chemisorptive reactive moiety as well as an inert functional group with negligible intermolecular interactions. These aspects of SMIs differentiate them from SAM molecules, primarily in the fact that self-assembly of most SAMs is supported by van der Waals interactions between large aliphatic chains (Fig. 2).³² Although a typical SMI can be described analogously to an SAM in that the small molecule may contain both one or more reactive moieties (like the SAM headgroup) that bind to the surface, and other inert functionalities (like the SAM tail group) that are not involved in surface bonding, with SMIs, we assume a negligible contribution from attractive intermolecular forces on monolayer formation, primarily due to the size and structure of the inert group.

For both types of inhibitors, some reactive groups may remain even after binding to the surface and, therefore, serve as a potential site for unwanted interaction with ALD precursors. For example, the headgroups of an organosilane SAM on SiO₂ may contain some free Si-OH groups even after surface attachment and hydrolysis. However, in the case of a SAM, even if the headgroup retains some reactive functionality, the well-packed tails on the SAM can effectively block the ALD precursor from accessing those sites. On the other hand, SMIs lack the relatively thick monolayer of alkyl chains possessed by SAMs (Fig. 2) and thus do not benefit from the mechanism by which ALD precursors are prevented from diffusing toward and interacting with both the substrate surface and the reactive moieties of the inhibitor. One potential consequence is that attractive interactions may occur between ALD precursors and the reactive moiety of the inhibitor. These interactions could inadvertently lead to undesired nucleation and a loss of selectivity in two different ways. First, the interaction between the ALD precursor and the reactive moiety may increase the likelihood that the inhibitor is displaced by the precursor—causing defect formation—and second, the attractive forces may increase the lifetime of the precursor at the surface—increasing the probability that the precursor will react with the SMI itself or a nearby reactive site of the surface.²⁷ 

A noteworthy example of the first phenomenon can be seen in the work by Merkx et al. in which Hacac was used as an SMI.^23,27 They observed that the SMI could assume two distinct chemisorbed binding configurations on the substrate at high surface coverages due to steric interactions—a chelate configuration and a monodentate configuration, each of which can be seen in Fig. 3. The molecules in the monodentate configuration were found to be responsible for defect formation in two ways, both attributed to the depletion of the SMIs from the surface. First, the monodentate species were not as strongly bound to the surface as those in the chelate configuration, and thus thermally desorbed under ALD conditions more readily than those in the chelate configuration. Second, the SMIs in the monodentate configuration were shown to possess higher attractive interaction energies with the ALD precursor than those in the chelate configuration due to the presence of a free O or OH group on the SMI. It was proposed that these relatively strong attractive interactions between the monodentate adsorbates and the precursor were the starting point for a process in which the precursor displaced the inhibitors from the surface.

This example suggests that an SMI possessing multiple possible binding configurations (e.g., monodentate versus chelate) and multiple adsorbing moieties (e.g., two carbonyls) may be particularly susceptible to those types of precursor-inhibitor interactions that can be detrimental to the inhibitor layer and thus selectivity. However, such cross-interactions may be expected even if only one bonding configuration is possible, simply because of the reactivity of the SMI. Thus, it is clear that to create an effective inhibitor layer using SMIs, consideration needs to be given to the reactive moieties of the inhibitor in order to limit precursor-inhibitor interactions. Furthermore, even though its size is small relative to that of an SAM, the SMI must possess an inert moiety that can serve to passivate the substrate surface upon successful adsorption and to isolate the adsorbing component from the incoming ALD precursor as much as possible.

In addition to tuning the SMI to limit precursor-inhibitor interactions, it is important that the ALD precursor itself be tuned to limit such interactions. In an investigation of the selectivity of metal, metal nitride, and metal oxide ALD on dimethylamino-trimethylsilane (DMA-TMS) passivated SiO₂, Soethoudt et al.⁴⁴ demonstrated that the ability of the SMI to inhibit growth depended strongly on the ALD chemistry. In particular, they observed that Ti(OCH₃)₄ exposure led to a degradation of the DMA-TMS passivation layer, while TiCl₄—another Ti precursor used for TiO₂ deposition—did not.⁴⁴ 

When selecting an SMI-precursor pairing that will possess limited interactions and facilitate ASD, the steric interactions of both the SMI and the precursor should be considered. Kim et al. recently studied the use of two different precursors—trimethylaluminum (TMA) and dimethylaluminum isopropoxide (DMAI)—for the ASD of Al₂O₃ on SiO₂ in the presence of ethanethiol (ET)-inhibited Co and Cu.²² They discovered that ET was able to inhibit deposition much better when DMAI was used than when TMA was used. Importantly, density functional theory results indicated that DMAI dimerizes more preferentially than TMA. Based on Monte Carlo simulations, the authors suggested that in its larger dimer form, DMAI could not nucleate on the ET-inhibited surfaces due to steric effects on the relatively densely packed thiol layer, whereas TMA in the smaller monomer form was more easily able to adsorb and nucleate in small gaps present in the inhibitor layer. Their work illustrates that while it is important to design an SMI that will provide sufficiently high areal density to prevent the precursors from reaching the surface, analogous considerations must also be made in ALD precursor selection. Nevertheless, sterics are only one of the design factors to be considered. How the SMIs bind to the surface and how that binding affects the molecular interactions between the ALD precursors and SMIs are also important. Hence, a holistic approach must be taken when selecting the SMI chemistry and the ALD chemistry to carry out ASD in a particular system.

In addition to the precursor, it is also important to understand how the coreactant may interact with the SMI. During their study of metal, nitride, and oxide ASD processes facilitated by the DMA-TMS SMI, Soethoudt et al. observed that the exposure of the SMI layer to the coreactants O₂, NH₃, and H₂O did not alter the inhibitor layer, indicating minimal interactions.⁴⁴ However, such inertness is not always the case. In the works of Mameli et al.²³ and Merkx et al.,²⁷ a plasma coreactant, particularly O₂ plasma, was used in the low temperature ALD of SiO₂, and it was observed that the plasma coreactant simultaneously served to remove the inhibitor layer on the NS—thereby necessitating the reapplication of the SMI. In another recent study by Merkx et al. using Hacac as an SMI for ASD, it was determined that the O₂ plasma step only partially removed the Hacac from the NS.⁴⁵ The authors observed that the O₂ plasma left fragments, such as formates and carbonates, on the surface that impeded Hacac reapplication and created defect sites that led to a loss of selectivity. However, when an additional H₂ plasma step was included in the process prior to the coreactant step, the surface was entirely cleaned of the inhibitor species, which ultimately led to a more successful reapplication of the inhibitor and a further prolonged nucleation delay. As such, it is critical to understand if and how the coreactant is affecting the inhibitor layer in order to better design an ASD process that can address the issues associated with the evolving nature of the system.

By considering precursor-SMI interactions, a few key lessons regarding SMI design can be introduced. The SMI should possess reactive moieties that only facilitate attractive interactions between the SMI and the substrate. To this same end, an ALD precursor that is not highly attractive to the reactive moieties of the SMI should be chosen. Another means of minimizing the probability that an ALD precursor will interact with the SMI is to use a precursor that is relatively large compared to the packing density of the SMI layer, such that the precursor is sterically inhibited from reaching the reactive moieties of the SMIs.

One strategy for choosing an SMI that will pair well with a particular precursor for carrying out ASD is to consider chemistries already present in the ALD process of interest. The mechanism by which SMIs prevent deposition is by adsorbing to and blocking those surface sites that are particularly reactive with the deposition precursor. Thus, one way to facilitate this specific blocking is to use an SMI that closely resembles either the ligands of the deposition precursor or one of the by-products of the precursor-surface reaction, because these species are already known to react at the substrate. Furthermore, because ALD precursor adsorption is self-limiting, it is implied that precursor adsorption can be blocked when the surface is saturated by the ligands of the precursor. This strategy has been applied successfully for CVD purposes. For instance, while growing TiB₂ from the precursor Ti(BH₄)₃(dme), Kumar et al. used one of the film growth by-products, 1,2-dimethoxyethane (dme), in order to inhibit growth.⁴⁶ Additionally, Babar et al. selectively blocked the deposition of Cu films on thermal SiO₂ and porous carbon doped oxide by using vinyltrimethylsilane (VTMS) to inhibit the chemisorption of the precursor, Cu(hfac)VTMS.⁴⁷ This strategy has also been proven effective in at least one ALD case; Yanguas-Gil et al. showed that Hacac, a beta-diketonate, almost completely inhibited Er₂O₃ ALD when Er(tmhd)₃—a beta-diketonate based precursor—was used.⁴⁸ Further examples are sure to be developed in the coming years.

Yet another strategy for choosing an SMI that will possess limited interactions with the ALD precursor is to use a compound that lacks a relatively rich electron density. As will be discussed further in Sec. II D, certain moieties on an inhibitor molecule are more attractive to incoming precursor molecules than others—particularly those with a high electron density or more electron lone pairs. This strategy of using SMIs that lack electron lone pairs has already been observed in both the CVD and ALD literature. Suh et al.^29,49 has shown that both 3-hexyne and 4-octyne can be used as SMIs for the ASD of ZrO₂ on SiO₂ in the presence of Cu by adsorbing to Cu through an sp → sp² rehybridization process and blocking active sites for reaction. In these cases, the alkynes do not possess any electron lone pairs; rather, the electron density of the reactive moiety is entirely held within the covalent triple bond. During the rehybridization process, the electron density from one of the bonds within the triple bond is donated to the undercoordinated Cu atoms of the substrate surface. Using SMIs with this type of adsorbing moiety, where the electron density is less exposed, may serve to limit SMI-precursor interactions. Interestingly, both of those alkyne-based SMIs enabled ASD by competitive adsorption, meaning they needed to be coflowed with the incoming precursor. It was noted during the case of CVD that predosing the SMI (4-octyne) did not lead to ASD.²⁹ The authors concluded that the adsorption of the SMI was reversible and that the residence time was too short to block the growth of the film on the NS if it were only predosed rather than coflowed. These examples highlight the trade-off that exists between SMI adsorption strength and SMI reactivity. While reactive moieties on an SMI may encourage undesirable precursor-inhibitor interactions, they may also be necessary to ensure the SMI layer is strongly bound to the surface with a low tendency toward desorption and thus defect site creation.

The use of inert organic functional groups that isolate the reactive component of the SMI molecule from the ALD precursor, and thus stabilize that interface, is primarily informed by extensive work with SAM molecules used in ASD. With well-ordered SAM passivation layers, the precursor is assumed to interact primarily with the aliphatic chain at the inhibitor surface, due to the distance from the more reactive SAM headgroup, as described in Sec. II A. Methyl groups have been used extensively as terminating sites for SAMs at this interface due to their inertness to a wide variety of chemistries.

The choice of methyl groups as inert units for ASD is supported by rate equations defined by Xu and Musgrave³⁷ for the reactions between TMA, a common Al₂O₃ precursor, and various SAM terminating groups. As shown in Fig. 4, the reactivity of hydroxyl, amine, and methyl groups all increase as a function of temperature, as expected. However, terminating methyl groups are ∼40 orders of magnitude less likely to react with TMA than are both primary amine and hydroxyls groups at the most elevated temperatures, and are even less reactive at lower temperatures.

Activation barrier data from Xu and Musgrave³⁷ and Patwardhan et al.⁵⁰ for the reaction between TMA and various functional groups are plotted in Fig. 5. It is clear from the data that the activation barrier is largest for methyl groups, in agreement with the relative inertness of methyl-terminated inhibitors against ALD. Also shown in Fig. 5 is the number of lone pairs in each of the functional groups. The ease with which TMA reacts with the various terminating group can be seen to increase with the density of delocalized lone pair electrons, with the carboxyl group (four associated lone pairs) exhibiting virtually no barrier to react with TMA and the methyl group (no lone pairs) exhibiting the highest barrier. Hydroxyl and thiol groups (two lone pairs) also have relatively low activation barriers toward reaction with TMA, while a primary amine (one lone pair) exhibits a barrier intermediate between that of OH/SH and methyl. Further studies have also reported that electron-deficient ALD precursors (such as the Lewis-acidic TMA) have a higher probability of chemisorption in the presence of accessible lone pairs,⁵¹ an assertion consistent with the trend shown here. Thus, the density of lone pairs on a functional group on the SMI can be used to qualitatively predict the facility of chemisorption between the electron-deficient TMA species and the SMI.

The analysis presented above indicates that methyl termination is desirable for minimizing reactions with an ALD precursor such as TMA, consistent with the well-established results of SAM blocking in ASD described in Sec. II A. However, with a SAM, choosing a tail group with methyl termination may be sufficient to block ALD independent of the SAM headgroups. This is because even if the headgroup contains functionality with high lone pair density, the tail groups in a well-packed SAM can block access to the ALD precursor. On the other hand, SMI blocking agents are likely to be much more influenced by the consequences of increased lone pair density near the substrate-inhibitor interface because it is relatively easy for the precursors to reach that interface. Hence, SMIs inherently support a more reactive environment near the inhibitor-precursor interface, and careful consideration must be given to the functionality of the reactive moiety of SMIs when choosing a selective system. Whereas headgroup reactivity with precursors is often less important than the degree of packing when using SAMs, considering the reactivity of an entire SMI will likely prove much more critical for stabilizing the inhibitor-precursor interface.

Integral to the process of ASD is a requirement that the inhibitor bond to only specific regions of the substrate. Various SMIs have been shown to preferentially bind to several different substrate materials, with a variety of proposed adsorption mechanisms.^{22,23,28–30} However, all of the reported SMI systems have ultimately relied on some form of Lewis or Bronsted acid-base surface chemistry. As defined, the Bronsted acid-base theory focuses on the transfer of protons, while the Lewis acid-base theory places emphasis on the migration of electron lone pairs.^52,53 With this distinction, we can begin to conceptualize SMI reactivity in terms of either type of acid-base classification. For instance, Suh et al. studied 4-octyne adsorption on bare copper, where they hypothesized that dispersion interactions and a Lewis basic electron transfer led to a molecular rehybridization (sp to sp²).²⁹ The aminosilane inhibitors studied by Khan et al. required surface acidic hydroxyls on SiO₂ to react with basic amide ligands to facilitate chemisorption.³⁰ Still further, the weakly acidic acetylacetone studied by Merkx et al. was shown to preferentially bind to more alkaline substrates, with the energetically favored chelate conformation displaying both Bronsted acidic and Lewis basic characteristics.²⁷ From these examples, we can infer that the SMI and the substrate must be mutually chosen to facilitate the desired adsorption via acid-base reaction.

A convenient method for thinking about SMI and substrate compatibility is to consider their relative acidities with a quantitative scale, with the guiding principle that SMIs with strongly acidic adsorption groups will have strong interactions with basic substrates and limited driving force for reaction with acidic substrates. Such an analysis was also applied by Mameli et al. to interpret the results of their Hacac SMI study.²³ For metal chalcogenide surfaces, Jeong et al. proposed a quantitative scale for surface acidity (SA), given by

where $Nm$ represents the formal oxidation state of a metal in a compound and $δm$ the Sanderson partial charge on the metal ion. $δm$ is further defined in Eq. (3) for metal chalcogenides of structure M_xE_y,

where $Smn+$ and $SE$ represent Sanderson's electronegativities of the metal ion of oxidation state n as well as that of the chalcogenide, respectively.⁵⁴ This metric is helpful because it accounts for how Lewis acidity and overall electron density allocation is affected by oxidation state of the metal, as well as the presence of stoichiometric chalcogens. The vast majority of ASD literature has involved oxide-based metal chalcogens; thus, we choose to focus on metal oxide substrates and compare various surface acidities in this analysis. We can determine the following tabulated values for surface acidity of common metal oxides based on the two equations above (Table I).

In this table, higher values of SA indicate a substrate with higher surface acidity, while lower values for SA suggest a more alkaline surface character. With these scaled values for surface acidity, the adsorption behavior of many inhibitors that have been studied becomes clearer. As shown in the literature, phosphonic acids, Hacac, and thiols all have primarily acidic characteristics and thus would be expected to bond more favorably with more alkaline substrates.^20,23,34 For Hacac, the SA values support the observations made by Mameli et al.²³ regarding the disparity in chemisorption behavior for the SMI on various alkaline and acidic substrates. Specifically, their work indicated that Hacac adsorbed favorably on Al₂O₃ and HfO₂ substrates (more alkaline surface character), whereas there were more limited interactions with SiO₂, WO₃, and GeO₂ substrates (more acidic surface character). For the well-studied SAM molecules, octadecylphosphonic acid (ODPA) and DDT, comparable behavior can also be seen, as previous reports have indicated preferential adsorption of both inhibitors to CuO_x and CoO_x (lower SA) at room temperature over SiO₂ (higher SA).^14,35,59 The work of Kim et al. further corroborates the adsorptive behavior of thiols; they found that an ethanethiol SMI selectively adsorbed on CuO_x and CoO_x over SiO₂.²² 

While useful, the approach described above for predicting substrate-inhibitor compatibility has limitations. For one, it is unable to completely describe what are often both acidic and basic inhibitor adsorption mechanisms. For instance, while the hydroxyl group of the ODPA gives that inhibitor some Bronsted acidic character, the strong electron withdrawing nature of the oxygen in the phosphonic group imparts a significant Lewis basic character to the inhibitor overall via the phosphoryl groups. In this case, it is has been suggested that interactions between the Lewis basic phosphoryl groups and cationic metals centers can promote the eventual reaction between the acidic hydroxyl group of the ODPA and surface hydroxyl groups of metal oxides.⁶⁰ A similar effect could likely be observed for Hacac, but to a lesser extent, due to the fact that the oxygen in the carbonyls have a lower degree of Lewis basicity than those of ODPA. Furthermore, this approach is inherently limited to metal chalcogens. Similar acid-base considerations may be used to individually explain comparable SMI interactions with reduced metals or elemental semiconductors. However, a greater degree of electron delocalization and structural complexity make the formation of analogous acidity scales difficult. Ultimately, however, it is clear that the acidic and basic properties of both the substrate surface and of the adsorbing groups will impact the strength of the interactions at the substrate-inhibitor interface. The combinations of these interactions are, therefore, important, and although it is based on a simple analysis, the semi-quantitative approach can be helpful for predicting SMI and substrate compatibility, which may in turn expedite the design process for inhibitors in ASD.

When addressing the compatibility of the SMI to a given substrate, it is important to consider that the substrate character may not be static but rather may evolve over time. In the study by Suh et al. on the use of 3-hexyne as an SMI for the ASD of ZrO₂ on SiO₂ in the presence of Cu, it was determined through density functional theory (DFT) calculations that the alkyne molecule adsorbed strongly to elemental copper by means of an sp → sp² rehybridization.⁴⁹ However, a loss in selectivity was observed experimentally after just ten ALD cycles of the precursor and the coreactant, O₂. X-ray photoelectron spectroscopy data revealed the formation of a Cu₂O film, suggesting that successive exposure of the coreactant to the metal surface led to oxidation. Because the alkyne molecule could not adsorb as effectively to this newly oxidized surface, a loss of selectivity occurred. Therefore, it is critical to take into account how the ASD process conditions may alter the substrate character and thus the stability and performance of the inhibitor layer. These dynamic systems evolve over time and thus the process must be designed accordingly.

The small size and high volatility of SMIs present an opportunity for in vacuo regeneration of the inert inhibitor layer. Most earlier studies of ASD with inhibitors typically relied on an initial passivation of reactive sites by the inhibitor to extend the inherent nucleation delay of the subsequent ALD process. This single-dose inhibition step is particularly convenient for solution-processed SAMs, due to ex vacuo processing requirements. The expected behavior of an ALD system following predosing of an inhibitor is illustrated schematically in Fig. 6, which plots deposition thickness as a function of ALD cycle number. The plot shows that whereas an ALD process in the absence of any inhibitor may exhibit a small nucleation delay, addition of an inhibitor pushes out the onset of sustained growth. Moreover, the figure shows that while well-packed SAMs can often extend a nucleation delay far beyond that of a typical ALD process,¹⁴ SMIs reported to date have produced much shorter nucleation delays.^22,23,29 Nevertheless, for both SAMs and SMIs, the onset of ALD nucleation eventually occurs when inhibitor dosing is limited to a single step.^14,23,61

Due to the reduced capacity of inhibitors, particularly SMIs, to effectively block deposition for high ALD cycle numbers, regenerating the inhibition layer by redosing inhibitor molecules is an attractive option. Hashemi et al. demonstrated this concept for SAM inhibitors when they showed redosing DDT could extend the nucleation delay of ZnO ALD to allow for selective deposition of ZnO more than three times thicker than an approach that did not regenerate the SAM.^17,62 The outcome of redosing the inhibitor molecule repeatedly throughout the ALD process is illustrated schematically in Fig. 7. Compared to deposition without an inhibitor [Fig. 7(a)] and with just a single predose of the inhibitor [Fig. 7(b)], deposition on the nongrowth surface is attenuated if the inhibitor is repeatedly dosed [Fig. 7(c)]. As shown in Fig. 7(c), the inhibitor is redosed with a frequency such that breakthrough nucleation (inferred from the slowed yet still increasing growth) is minimized. This type of ASD scheme further reinforces the benefit of using easily vaporized inhibitor molecules for selective processes in vacuum conditions, since redosing inhibitors from the vapor phase is a process that is easily integrated into existing ASD processes. Moreover, such a technique could prove particularly useful for SMIs that have been shown to have limited blocking capabilities beyond a small number of cycles.^23,29

One challenge with the reapplication of inhibitors is maintaining selective growth while limiting adsorption of the inhibitor onto the growing ALD layer. Such unwanted inhibitor adsorption could hinder the ALD reactions on the GS as well as introduce impurities to the film that compromise its material properties. On the other hand, some processes can withstand limited uptake of the SMI onto the growth surface yet still lead to satisfactory properties. In a recent study by Merkx et al. that used aniline as an SMI in the selective deposition of TiN on a dielectric in the presence of a transition metal, the authors observed that the SMI did, in fact, adsorb on the growing material. However, they found that this adsorption had minor effects on the growth behavior and the electrical properties of the film.²⁸ This observation is particularly important for two reasons. First, it demonstrates that inhibitors may indeed adsorb on the growing material and consequently one needs to monitor such adsorption and its impact on the ALD process. Second, their work shows that although the inhibitor may adsorb on two different substrates, different growth dynamics and blocking abilities may still ensue.

Although introducing more frequent doses of inhibitor molecules can help promote higher selectivity, redosing the SMI may simply reintroduce cyclic nucleation delays without removing undesired deposited material on the nongrowth surface. Thus, nucleation may be slowed, but not completely inhibited, as illustrated in Fig. 7(c), which shows the deposited film thickness ultimately rising with increasing cycle number, albeit at a slower rate. In addition, as more material begins to be deposited on the NS, the inhibitor may be less able to adsorb on the NS since its surface character becomes increasingly dominated by the ALD material rather than the underlying substrate. Introducing a cyclic etching step into ASD cycles is one way to mitigate these issues, as shown in Fig. 7(d). Curve D shows that if the undesired deposits on the NS are cyclically removed, together with reapplication of the inhibitor, the NS can, in principle, be kept free of deposits for a larger number of cycles. Note that the chemistry for this process must be carefully developed to proceed without fully removing the desired deposits from the GS. In the ideal scheme, an etchant would act to remove unwanted deposits on the NS and also remove the defected inhibitor layer and allow for the production of a fresh surface for reapplication of the SMI. Toward the etching of inhibitors, Hashemi et al. and Prasittichai et al. showed that both acid etches and electrochemical reduction, respectively, could remove ODPA SAMs.^40,61 Although those processes were solution based and thus not compatible with in vacuo systems, their proof-of-concept demonstration clearly showed that unwanted deposits could be removed in the process together with the SAM, leading to increases in overall selectivity.

Achieving a vacuum-compatible etching process as described above necessitates a vapor phase etchant that can produce volatile by-products from the reaction with the inhibitor molecules or the undesired deposit in order to remove these species from the surface. Plasmas (e.g., H₂ or O₂) have been shown to serve the first purpose well. Mackus and co-workers have explored several ASD processes involving SMIs, each deposited in an ABC-cyclic fashion and using a plasma that is integral to both deposition and inhibitor removal. In their process, SMIs were selectively deposited in an initial step (A), before the ALD precursor was also introduced (B), followed by a plasma coreactant (C) that both developed the ALD layer on the desired growth surface while also removing the deposited SMI on the nongrowth surface. With this scheme, Mackus et al. selectively deposited SiO₂, WS₂, and TiN.^23,28,63 Their work showed the potential of removing and reforming inhibitor layers with each cycle of a selective process. This reformation process is helpful because if defects are introduced into the inhibitor layer, e.g., by decomposition of the inhibitor molecules, they can act as potential sites for undesired ALD growth. It is worth noting, however, that even with this process, ALD nucleation still eventually occurred at high numbers of cycles. Thus, fully correcting defects by way of vapor phase etchants and SMIs is a feat yet to be realized, and a challenge for future systems is to determine compatible SMIs and etchants that allow for continual selectivity at high cycle numbers.

The above examples illustrate that when choosing an SMI, one must take a holistic approach that includes considerations of each major component of the ALD process—the precursor, the SMI, the surface, and each interface of interaction. The choice of inhibitor thus becomes an intricate process of choosing an SMI based not only on the surface character of the GS and NS but also on that of the growing ALD film as well. This complex selection process can be assisted by an understanding of acid-base interactions between the SMI and the various surfaces and will be a function of how selective a particular process needs to be. The choice of inhibitor may also be facilitated by computational studies to provide high-throughput screening of potential SMIs.

Area-selective ALD represents a powerful technique for the fabrication of next-generation nanoelectronics. However, the current methods of achieving ASD possess certain limitations that could impede their application in the industry. Using SMIs for ASD can overcome some of the issues in selective deposition that act as a bottleneck in device fabrication. Currently, SMIs have only just begun to be used in ALD processes, and little is known about their mechanism of facilitating ASD. Using case examples, some understanding is elucidated about SMI behavior, particularly how they interact with ALD precursors and with substrate surfaces. With this information, guiding principles can be established to better aid in proper SMI design.

Because SMIs cannot benefit from the thick monolayer that SAMs possess to prevent precursor diffusion to and reaction with the substrate surface, much more attention needs to be given to both the reactive moieties and the general structure of the SMI for multiple reasons. Regarding the reactive component of the SMI, in general, it is important to minimize the attractive interactions between the ALD precursors and the SMI in order to prevent undesired effects such as potential SMI displacement and defect formation. This minimization of interaction should be considered both when designing the SMI and when selecting a deposition precursor. At the same time, the reactive component must also facilitate strong attractive interactions between the inhibitor and the substrate surface. These interactions prevent thermal desorption and maintain an effective inhibitor layer. Thus, there are certain trade-offs to consider when designing the reactive component of the SMI, because an SMI that selectively and strongly binds to a substrate may also exhibit more prominent interactions with the ALD precursor. Regarding the general molecular structure, it is helpful that the SMI be designed such that the maximum number of reactive surface sites will be blocked against chemisorption of the ALD precursor. This requirement demands consideration of SMI size, areal coverage, and binding configuration. By tuning the SMI structure, one aims to create a densely packed inhibitor layer with minimal defects, thereby reducing the possibility for undesired nucleation.

Developing SMI-based ASD processes can be challenging due to numerous competing considerations. Additionally, ASD processes are inherently dynamic, as the substrate character, inhibitor layer, and deposited material may be changing over time. Therefore, a holistic approach must be taken when creating ASD processes that incorporate the use of SMIs. This type of approach, combined with opportunities for process tuning and optimization, presents many new routes for potential exploration in the field ASD.

We gratefully acknowledge funding from a Merck KGaA, Darmstadt, Germany 350 Research Award.

Data sharing is not applicable to this article as no new data were created or analyzed in this study.

 Stacey F. Bent is the Jagdeep and Roshni Singh Professor at Stanford University, where she is Professor of Chemical Engineering and Professor, by courtesy, of Chemistry, of Materials Science and Engineering, and of Electrical Engineering. Bent obtained her B.S. degree in chemical engineering from UC Berkeley in 1987 and her Ph.D. degree in chemistry from Stanford in 1992. After carrying out postdoctoral work at AT&T Bell Laboratories, she joined the faculty of the Chemistry Department at New York University in 1994, before moving to Stanford University in 1998. At Stanford, she has held various leadership roles, including chair of the Department of Chemical Engineering (2015–2016), Senior Associate Dean in the School of Engineering (2016–2019), and Director of the TomKat Center for Sustainable Energy (2010–2019). She has served as Vice Provost for Graduate Education and Postdoctoral Affairs at Stanford since 2019. Her research interests are in the understanding of surface chemistry and materials synthesis and the application of this knowledge to a variety of problems in sustainable energy, semiconductor processing, and nanotechnology. Her group’s research on ALD has ranged from fundamental mechanistic studies to applications in solar cells, fuel cells, catalysts, and batteries. She and her group have been active in the development of area-selective ALD. Bent has published over 280 papers and holds 6 patents. She has supervised nearly 50 Ph.D. students and 20 postdoctoral scholars. Bent was elected to the U.S. National Academy of Engineering in 2020. She is also a Fellow of the American Chemical Society (ACS) and the American Vacuum Society (AVS). She received the ACS Award in Surface Chemistry in 2018, the SRC Technical Excellence Award in 2020, and the ALD Innovator Award in 2021. She was the recipient of the Peter Mark Memorial Award in 2000.