Enhancement of Conformality of Silicon Nitride Thin Films by ABC‐Type Atomic Layer Deposition

Atomic layer deposition (ALD) is utilized for the fabrication of miniaturized electronic devices with nanometer‐scale features. However, the conventional ALD process on high‐aspect‐ratio (HAR) substrates often results in the deposition of thin films with suboptimal conformality over the depth of the trench structures. This study introduces an ABC‐type ALD of silicon nitride (SiNx) employing vapor‐deliverable tert‐butyl chloride (TBC) as a surface inhibitor. Herein, density functional theory (DFT) calculations elucidate the surface reaction mechanisms, confirming the inhibition of the Si precursor by the t‐butyl moiety. Notably, the ABC‐type ALD exhibits reduced growth per cycle relative to the conventional process while avoiding detectable carbon contamination in the SiNx thin films. Applying this process to substrates with trench structures revealed a substantial improvement in conformality following the introduction of TBC. Furthermore, the step coverage of the deposited SiNx can be modulated by adjusting TBC exposure, enabling greater film thickness at the bottom than that at the top of the trench. The proposed method of modulating ALD processes holds potential for the high‐volume manufacturing of semiconductor devices with HAR structures.


Introduction
[3][4][5] In nanoscale electronic devices, such as vertical NAND (V-NAND) and dynamic random access memory (DRAM), high-quality thin-film deposition on high-aspect-ratio (HAR) substrates is essential.Atomic layer deposition (ALD) is a technique enabling the selflimiting growth of thin films through alternating adsorption DOI: 10.1002/aelm.202300722[8] Utilizing sequential gassurface reactions, ALD achieves thin films with exceptional conformality, offering precise control and tunable composition at the atomic scale.Consequently, ALD is widely used for thinfilm deposition over HAR structures.However, during ALD on HAR substrates with extreme aspect ratios, the thin film deposition rate often diminishes along the depth, resulting in undesirable seams or voids. [9,10]Such nonconformal deposition could result from the diffusion-limited flux of precursor molecules inside the HAR structures, or the parasitic chemical vapor deposition (CVD) reactions that are not self-limiting.Therefore, methods to increase the conformality of ALD thin films are actively being investigated.
Moreover, area-selective ALD (AS-ALD) is an emerging technique that utilizes variations in initial film growth due to surface chemical reactions in ALD on substrates with varying surface properties.[16] Therefore, understanding the molecular interactions among the inhibitor, precursor, and counter-reactant on the surface is crucial for their sequential adsorption and removal steps.Interestingly, AS-ALD can improve the ALD thin film step coverage by applying inhibitors with a gradient in density from the entrance. [17]By modulating the exposure of inhibitors within the sub-saturation regime, a controlled gradient of adsorbed inhibitors can be achieved from the top to the bottom within HAR trenches.Then, the deposited thin film results in enhanced conformality, thereby facilitating seamless gap-fill.However, the presence of the inhibitor at the substrate and the ALD thin-film interface, unless removed during the process, can affect the interfacial properties.[20][21][22][23] For AS-ALD, the inhibitor (step A) can hinder the adsorption of the precursor (step B), and if removed during the counter-reactant pulse (step C), it will not remain at the film interfaces.26][27][28][29][30][31]  Silicon nitride (SiN x ) thin films have garnered significant interest in various applications owing to their exceptional mechanical, electrical, and chemical properties.[34][35][36][37] However, the increasing miniaturization and structural complexity of electronic devices pose challenges in depositing SiN x thin films, including limitations in conformality and step coverage over substrates with extreme HAR.
This study introduces an ABC-type thermal ALD process to enhance the conformality of SiNx films on HAR substrates.Herein, tert-butyl chloride (TBC), hexachlorodisilane (HCDS), and ammonia (NH 3 ) are employed as the inhibitor, precursor, and counter-reactant, respectively.By leveraging the gradient adsorption of the inhibitor, the ALD growth rate can be modulated, thereby yielding improved conformality of ALD SiNx, with no detectable impurities in the films.Furthermore, density functional theory (DFT) calculations were utilized to investigate the chemical mechanisms in each reaction step of the ABC-type ALD processes.Therefore, this study presents a novel method for enhancing conformality on HAR substrates during ALD of silicon nitride-a technologically crucial material.

Results and Discussion
DFT calculations probed the potential chemical reaction mechanisms during each step of the ABC-type modulated ALD.First, the behavior of the TBC inhibitor upon its introduction into the chamber with the SiN x substrate (step A) was examined (Figure 1).Given the high deposition temperature of 600 °C for SiN x , the molecular self-decomposition and adsorption of TBC on SiN x were compared.Figure 1a depicts a reaction energy diagram illustrating TBC adsorption on an NH 2 -terminated SiN x surface.The Cl ligand of TBC binds to the H of the surface NH 2 , yielding gaseous HCl byproducts and binding the tBu (tertbutyl) group to the surface.This adsorption reaction exhibits a transition state 126.1 kJ mol −1 above the reactants' energy, with activation energy (E a ) of 154.5 kJ mol −1 measured from the intermediate state of the TBC physisorption.Conversely, the gas-phase self-dissociation reaction of the TBC is known to proceed via intramolecular hydrogen transfer, resulting in HCl and isobutene (C 4 H 8 ) formation. [38]Our DFT calculation on such self-decomposition displayed an E a of 178.1 kJ mol −1 (Figure 1b), aligning with the experimentally reported value of 191 kJ mol −1 . [39]E a for TBC adsorption on SiN x surfaces is ca.24 kJ mol −1 lower than that of its thermal decomposition, suggesting a kinetic preference for substrate adsorption over selfdecomposition.
Second, considering that alkyl termination on surfaces can diminish the reactivity toward the adsorption of ALD precursors, [40][41][42] a reduction in the adsorption reactivity of the Si precursor, HCDS is anticipated as the surface becomes partially passivated by the tBu moiety.To examine the adsorption reactivity of HCDS on a tBu-modified nitride surface, a surface model featuring one tBu and an ─NH 2 terminating group was employed (Figure 2).[45] The activation energy for this reaction is 60.7 kJ mol −1 on the ─NH 2 terminating group, increasing to 260.3 kJ mol −1 on the tBu-modified surface.Consequently, this indicates that surface passivation by TBC adsorption effectively inhibits HCDS adsorption.
[48] Concurrently, the tBu moiety must be eliminated via reaction with NH 3 to ensure nitride deposition without carbon contamination.Various pathways for tBu group removal using NH 3 were investigated (Figure S1, Table S1, Sup-plementary Information).Among these, the catalytic elimination of isobutene via -hydrogen transfer with the assistance of NH 3 [49,50] exhibited the lowest transition state energy of 250.3 kJ mol −1 (activation energy of 297.2 kJ mol −1 ), as illustrated in Figure 3a.Intramolecular hydrogen transfer can facilitate isobutene removal, leaving ─NH 2 on the surface (Figure 3b); however, the activation energy for tBu removal without NH 3 was significantly higher at 302.8 kJ mol −1 , indicating the necessity of NH 3 for efficient hydrocarbon elimination.
It is noted that currently reported DFT calculation results are purely electronic energy values at 0 K, while considerable thermal effects may be present at the experimental process temperature as high as 600 °C.The finite-temperature entropy may be extrapolated from the DFT-optimized structures by hindering translational and vibrational motions of the adsorbates, [51] whose contribution to the Gibbs free energy may be as large as several hundred kJ mol −1 . [52,53]However, such an effect likely would not change the relative trends in the above discussion comparing differently functionalized surfaces.[56] Treatment of such dynamic situations through various advanced simulation techniques can be an interesting future research direction.Experiments were conducted to investigate the fundamental process characteristics of the ALD processes.A typical growth characteristic of ALD is the saturation of the growth per cycle (GPC), indicative of self-limiting surface reactions for each precursor and reactant, confirming their suitability for ALD processes.Although the process temperature of 600 °C slightly exceeds the reported ALD temperature window for the HCDS precursor, [57,58] the currently observed GPC of the HCDS-NH 3 ALD process in response to varying exposures of each reactant, with a fixed exposure of the other parameter, exhibited reasonable saturation behavior (Figure S2a,b, Supporting Information).Figure S2c (supporting information) illustrates that with the temperature rise from 500 to 600 °C, the GPC of HCDS-NH 3 ALD increased from 0.8 to 1.1 Å cycle −1 , aligning with previously reported HCDS processes for SiN x ALD.Nonetheless, as each reactant attains self-limiting saturation conditions, the overall process demonstrates linearity in film thickness relative to cycle numbers, as depicted in Figure S2d (supporting information).
Then, the GPC of the proposed ABC-ALD process can be modulated according to TBC exposure.Figure 4a presents the GPC of the ABC-ALD process as a function of TBC exposure time.With-out TBC, the ALD SiN x GPC was 1.1 Å cycle −1 , which decreased to a saturation value of 0.5 Å cycle −1 following 10 s of TBC exposure.A slight further reduction in GPC was observed with 30 s of TBC exposure; however, this increased the surface roughness of 10-nm thick SiN x thin films from 0.14 to 0.26 nm, as depicted in Figures 4b,c.A large dosage of TBC may lead to decreased nucleation density, resulting in larger grain size and increased roughness of the SiN x films. [59]egardless of the TBC introduction, the ABC-ALD thin films exhibited no carbonaceous impurities.XPS analysis of the asdeposited 10-nm thick SiN x thin films revealed no detectable carbon contamination in all samples examined (Table S2, Figure S3, Supporting Information).It is noted that the elemental detection limit in XPS measurements is often in the range of 0.1 atomic %. [60] The stoichiometry including consistent N/Si ratios of ≈1.1 remained unaffected by TBC exposure.Additionally, film stoichiometry was uniform across their thickness, as evidenced by XPS depth profiles (Figure S4, supporting information).
The conformality of SiN x thin films deposited via the ABC-ALD process was assessed using a wafer with a trench structure, featuring top and bottom critical dimension (CD) values of 84 and 65 nm, respectively, and an aspect ratio of 22:1 (Figure 5a).To calculate the relative thickness (t/t 0 , where t 0 denotes the thickness measured at the top of the thin film), the thickness (t) of the ALD SiN x thin films was measured at the top, middle, and bottom of the trench using cross-sectional TEM.Initially, two-step ALD SiNx exhibited a decrease in relative thickness along the trench depth, with t/t 0 reducing to 0.88 at the bottom.Subsequently, the relative thickness within the trench was enhanced by increasing TBC exposure time.With 5 s of TBC exposure, near-perfect step coverage was achieved, with t/t 0 values of 1.02 in the middle and 1.05 at the bottom.Furthermore, with 10 s of TBC exposure, the deposited thickness increased along the trench depth, resulting in t/t 0 values of 1.10 and 1.22 at the middle and bottom, respectively.Thus, the current process offers the potential for deliberate control of step coverage, ranging from superconformal thin film formation (t/t 0 = 1) to bottom-up gap-fill (t/t 0 > 1) within HAR trenches via adjusting the exposure of the ABC-type inhibitor.The time required for reactants to reach saturation in trench structures considerably exceeds that on flat substrates. [61,62]herefore, by adjusting process conditions, such as TBC exposure time, the degree of inhibition within the trenches can be regulated.This can result in more substantial inhibition at the top regions of the holes, while the inhibitor coverage at the bottom of the holes is relatively lesser.Such gradient inhibition can have multiple effects.First, deeper in the trenches, the remaining surface density of reactive sites for HCDS precursor adsorption would be higher, leading to increased precursor adsorption.Second, the entrance region of the trench, with reduced reactivity, enables a larger fraction of the precursor flux to penetrate deeper into the trenches.Consequently, either approach potentially enhances the conformality of ALD thin films within HAR trenches through ABC-type ALD.

Conclusion
An ABC-type ALD process for SiN x using TBC as a partial inhibitor was investigated.DFT calculations were conducted to clarify the surface reaction mechanisms of each step, encompassing the adsorption of both the inhibitor and HCDS precursor and the removal of the alkyl group by NH 3 .The results demonstrated that TBC adsorbs onto the SiN x surface as a t-butyl moiety, thereby inhibiting HCDS precursor adsorption.NH 3 then catalytically removes the hydrocarbon, producing isobutene.Experimentally, the ABC-type ALD process exhibited a considerably lower GPC compared to the conventional HCDS-NH 3 ALD process, without any detectable carbon impurities in the deposited film.Additionally, when applied to substrates with trench structures, the introduction of TBC enabled modulation of step coverage, contingent upon the extent of TBC exposure.Therefore, this ALD process presents potential applications in the super-conformal formation of thin films or bottom-up gap filling within HAR substrates, suggesting its suitability for high-volume manufacturing of 3D integrated circuits.

Experimental Section
DFT calculations were conducted using the Gaussian 09 software package. [63]The M06-L meta-GGA functional, [64] recognized for its high performance toward main group thermochemistry including noncovalent interactions, [65,66] was employed with a def2-SVP basis set.The surface model for the nitride substrate was represented by a Si 9 N 7 H 21 cluster with two NH 2 termination groups. [67][70][71][72][73] After optimization, the transition state geometries were verified by identifying an imaginary vibrational mode along the reaction coordinate.The energy variations according to the reaction were determined using the equation below: where E complex denotes the energy of the cluster-reactant complex when the reactant is the inhibitor, precursor, or counter-reactant.E byproduct , E cluster , and E reactant represent the energies of the isolated byproduct, cluster, and reactant, respectively.The activation energy (E a ) is defined as the deviation in the energy between the transition and physisorption states.
For the growth of SiN x , hexachlorodisilane (HCDS, 99.9%, Soulbrain Co.) was utilized as the Si precursor, and NH 3 (99.9%)served as the counter-reactant.The substrate comprised a p-type B-doped singlecrystalline Si(100) wafer with native oxide.A commercial showerhead-type ALD system (Atomic Premium, CN1) was employed, maintaining the process temperature at 600 °C.An ALD cycle was constituted by 30 s of source exposure, followed by a 30 s N 2 purge, then a 30 s reactant dose, and another 30 s N 2 purge.During deposition, the chamber pressure was consistently maintained at ≈1 Torr.For the ABC-type ALD, an additional step involving tert-butyl chloride (TBC, 99.9%, Soulbrain Co.) feed with subsequent N 2 purge was incorporated prior to each precursor exposure.
SiN x films with a thickness of 10 nm were utilized for characterization.Film thicknesses were determined using spectroscopic ellipsometry (Ellipso Technology, Elli-SE-aM8).X-ray photoelectron spectroscopy (XPS, Ulvac-PHI 5000 VersaProbe) was conducted employing Al K radiation.To assess the conformality of the thin films deposited on trench wafers, transmission electron microscopy (TEM, Jeol, JEM-ARM200F) was used to acquire cross-sectional and top images.A focused ion beam (Thermo Fisher, Helios 5 UX) tool facilitated the preparation of TEM samples.Surface roughness measurements were performed using atomic force microscopy (AFM, Park Systems, XE-100).For the HAR substrates, the critical dimension (CD) values at the top and bottom holes measured 84.3 and 64.7 nm, respectively.

Figure 1 .
Figure 1.Energy diagram and corresponding structures for Step A of ABC-type ALD.a) Adsorption of TBC on SiN x cluster, b) gas-phase self-dissociation reaction of TBC.Phy and TS refer to Physisorption and transition state, respectively.Blue = N, Brown = C, Gray = Si, Green = Cl, and White = H.

Figure 2 .
Figure 2. Surface reaction under HCDS exposure for Step B of ABC-type ALD.Energy diagram for adsorption of HCDS on a) tBu-passivated surface and b) ─NH 2 functional group, respectively.Blue = N, Brown = C, Gray = Si, Green = Cl, and White = H.

Figure 3 .
Figure 3. Surface reaction for step C of ABC-type ALD.Energy diagram and corresponding structures for tBu moiety removal reaction from the surface.a) Reaction follows the catalytic mechanism, which is the most preferred pathway under NH 3 exposure.b) A reaction where tBu moiety spontaneously dissociates without NH 3 exposure.Blue = N, Brown = C, Gray = Si, Green = Cl, White = H.

Figure 4 .
Figure 4. a) Average GPC and b) surface roughness as functions of exposure time of TBC, and c) corresponding AFM images.

Figure 5 .
Figure 5. a) Schematic of thin film deposition over HAR trench.b) TEM images of the 10-nm thick SiN x ALD film with exposure to TBC inhibitors of 0, 5, and 10 s at the top, middle, and bottom positions.Dark regions indicated by arrows represent the deposited SiN x films.c) Variations in relative thickness of the thin films inside the trench according to varying amounts of TBC exposure.