Area-Selective Atomic Layer Deposition of Two-Dimensional WS2 Nanolayers.

With downscaling of device dimensions, two-dimensional (2D) semiconducting transition metal dichalcogenides (TMDs) such as WS2 are being considered as promising materials for future applications in nanoelectronics. However, at these nanoscale regimes, incorporating TMD layers in the device architecture with precise control of critical features is challenging using current top-down processing techniques. In this contribution, we pioneer the combination of two key avenues in atomic-scale processing: area-selective atomic layer deposition (AS-ALD) and growth of 2D materials, and demonstrate bottom-up processing of 2D WS2 nanolayers. Area-selective deposition of WS2 nanolayers is enabled using an ABC-type plasma-enhanced ALD process involving acetylacetone (Hacac) as inhibitor (A), bis(tert-butylimido)-bis(dimethylamido)-tungsten as precursor (B), and H2S plasma as the co-reactant (C) at a low deposition temperature of 250 °C. The developed AS-ALD process results in the immediate growth of WS2 on SiO2 while effectively blocking growth on Al2O3 as confirmed by in situ spectroscopic ellipsometry and ex situ X-ray photoelectron spectroscopy measurements. As a proof of concept, the AS-ALD process is demonstrated on patterned Al2O3/SiO2 surfaces. The AS-ALD WS2 films exhibited sharp Raman (E 2g 1 and A 1g) peaks on SiO2, a fingerprint of crystalline WS2 layers, upon annealing at temperatures within the thermal budget of semiconductor back-end-of-line processing (≤450 °C). Our AS-ALD process also allows selective growth on various TMDs and transition metal oxides while blocking growth on HfO2 and TiO2. It is expected that this work will lay the foundation for area-selective ALD of other 2D materials.

The downscaling of dimensions in nanoelectronic devices has led to the exploration of alternate, two-dimensional (2D) semiconductors for future nanodevice applications.[1−6] In this regard, semiconducting transition metal dichalcogenides (TMDs) such as WS2 have attracted much interest, because of their high carrier mobility and direct bandgap in the monolayer regime.[3,7] In ultra-scaled nanoelectronic devices, it is crucial to precisely incorporate TMD layers at desired locations in the device architecture with consistent layer characteristics. Conventionally, lithography-based top-down processing techniques are used to pattern films into device features. Precise patterning and alignment of critical device features will become challenging using the current top-down processing schemes especially in sub-5 nm technology nodes.[8−10] At the same time, patterning 2D TMD layers with resist-based lithography risks inducing contaminants into the layers, that can significantly impact their functional characteristics.[11−14] In this regard, selective growth of TMD layers on predetermined locations (also referred to as area-selective deposition) via bottom-up approaches has attracted significant interest from both academia and industry.[14] Rather than performing deposition, lithography, and etching for each layer in a device stack, some of the layers can be incorporated in a bottom-up manner using this approach. By directly depositing TMD layers on prepatterned surfaces, area-selective deposition can enable the integration of TMD layers in multilayer device stacks without the need for resist-based patterning of the delicate TMD layers. Therefore, this approach decreases the total number of lithography and etching steps in the fabrication of nanoelectronic devices and, thus, can enable cost-effective device fabrication schemes for future technology nodes.

Area-selective deposition of TMD layers has been sparingly reported in the literature. Flakes of MoS2 have been deposited at predetermined locations by using lithography-patterned nucleation seeds or topological features in chemical vapor deposition (CVD) processes.[15,16] In these reported methods, the removal of seeds after fabrication is not straightforward.[14,17] The control over thickness and shape of the seeded TMD structures is also a point of concern.[17] Area-selective growth of CVD MoS2 has also been enabled by enhancing the reactivity of predetermined areas to CVD surface reactions through chemical treatments[17] and mechanically induced triboelectric effects.[14] The use of seed layers were mitigated in these processes. However, the longevity and stability of the enhanced surface reactivity of predetermined areas is uncertain with these methods. Gallium ion beams have been used to control the density of surface hydroxyl groups on the initial substrate (SiO2) and achieve patterned growth of MoS2 in a CVD process.[18] Recently, WS2 was selectively deposited using prepatterned a-Si sacrificial layers in a pulsed CVD process.[19] Conceivably, to date, area-selective growth of TMD layers has been restricted to CVD-based processes[14−16] that typically involve high-temperature processing (≥700 °C ). Moreover, these methods do not necessarily offer atomic-level thickness control over the TMD layer being deposited.

Recently, area-selective atomic layer deposition (AS-ALD) has attracted much attention as a promising alternative pathway to realize selective growth of materials on predetermined locations.[20,21] AS-ALD is an advanced atomic layer deposition (ALD) process that limits growth to predetermined areas by taking advantage of the surface chemistry differences between the growth and non-growth areas.[22,23] The chemospecific characteristic of the process coupled with the atomic-level accuracy of ALD is exploited to enable bottom-up growth of materials at precise locations without requiring any additional patterning of the film being deposited.[20,22,24] To date, especially AS-ALD of metals[21,25−30] and dielectrics[21,23,31] has been reported in the literature. These studies are being considered promising in terms of patterning ALD grown films for applications in nanoelectronics[22,32,33] and catalysis.[34,35] However, area-selective ALD of 2D TMD layers has remained unexplored. Ideally, an AS-ALD process should enable area-selective growth and, at the same time, offer the merits of conventional ALD processes including angstrom-level thickness control, uniform and conformal film growth over large-area substrates and high-aspect-ratio 3D structures, and low-temperature processing (typically T ≤ 450 °C).[36−41] Therefore, there is considerable interest in developing AS-ALD processes for selectively depositing TMD layers.

Selective growth via AS-ALD can be realized by deactivating the non-growth area for a specific ALD chemistry.[20,21] Select areas of the substrate can be functionalized with a blocking layer that allows growth only on the non-functionalized areas of the substrate.[20] Self-assembled monolayers (SAMs) and small inhibitor molecules have been investigated as blocking agents in AS-ALD processes. Recently, Mameli et al. showcased the efficacy of vapor-phase dosed acetylacetonate inhibitor molecules (referred to as Hacac) in blocking precursor adsorption and thereby, blocking film growth on Al2O3 surfaces during AS-ALD of SiO2.[23] The Hacac molecules were dosed every cycle, unlike the one-time application of SAMs in other AS-ALD processes, which makes the approach compatible with a wider range of processes, including plasma-enhanced ALD (PEALD). Taking a cue from these results, in this work, we use Hacac as an inhibitor to selectively deposit mono-to-multilayers of WS2. Selective growth was realized on (1) SiO2 (commonly used substrate for the growth of 2D TMD layers), (2) 2D TMDs such as MoS2, NbS2, and TiS2 and (3) transition metal oxides such as MoO3, Nb2O5, and WO3 while effectively blocking growth on Al2O3 and HfO2 surfaces at a low deposition temperature of 250 °C. We expect this first AS-ALD process for 2D WS2, to enable exploration of area-selective ALD processes for other functional 2D materials.

A schematic of our AS-ALD approach is shown in Figure for prepatterned Al2O3 (non-growth area) and SiO2 (growth area) surfaces. It is based on a three-step (i.e., ABC-type) ALD process. In step A, we dose Hacac molecules that adsorb only on the Al2O3 surface and not on the SiO2 surface. In step B, the precursor adsorption is blocked by the adsorbed Hacac inhibitor molecules on the Al2O3 surface, while the tungsten precursor (bis(tert-butylimido)-bis(dimethylamido)-tungsten) readily adsorbs on the SiO2 surface. In the final step (step C), the H2S plasma functions as the ALD co-reactant, enabling the growth of WS2 on SiO2. On the Al2O3 surface, it removes the adsorbed inhibitor molecules from the Al2O3 surface. Such ALD cycles are repeated to deposit WS2 films selectively on SiO2 surfaces. The BC steps of the ABC cycles used in this work were adopted from our previously reported WS2 ALD process, which resulted in the growth of WS2 films.[42]

Figure 1

Schematic illustration of the WS2 area-selective ALD process using ABC-type ALD cycles. The Al2O3/SiO2 patterned surface is shown before, during, and after ALD. The individual ALD steps of the ABC-type ALD cycle are indicated: Step A, Hacac dose; Step B, bis(tert-butylimido)-bis(dimethylamido)-tungsten precursor dose; and Step C, H2S plasma exposure. Using this process, WS2 is selectively deposited on SiO2 in the presence of Al2O3.

Figure a shows the nucleation curves for the ABC-type WS2 ALD process on SiO2 (growth area) and Al2O3 (non-growth area), as determined from in situ spectroscopic ellipsometry (SE). The plot shows that WS2 grows readily on SiO2 surfaces without any growth delay. The film thickness increases linearly with the number of ALD cycles, a characteristic of the ALD growth behavior, with a growth per cycle (GPC) of ∼0.6 Å. On the other hand, for the same ABC-type WS2 ALD process on Al2O3, a growth delay of ∼20 ALD cycles is observed. These results indicate that ∼1 nm of WS2 (at least one monolayer) can be selectively deposited on SiO2 while blocking the growth on Al2O3. When performing BC cycles (without the use of Hacac), no growth delay was observed on Al2O3, as evidenced by a linear increase in thickness with the number of ALD cycles (see Figure S1 in the Supporting Information).

Figure 2

(a) Film thickness as a function of the number of ALD cycles for the ABC-type WS2 process on SiO2 (growth area) and Al2O3 (non-growth area), as determined from in situ SE. (b) Integrated area of XPS W peaks after various numbers of ALD cycles determined for the Al2O3 and SiO2 surfaces. (c) Raw XPS spectra of the W 4f core level after 20 ALD cycles on the Al2O3 and SiO2 surfaces. XPS measurements were performed on samples deposited at 250 °C.

The nucleation behavior of our process was corroborated with X-ray photoelectron spectroscopy (XPS) measurements (Figure b and 2c). Consistent with the in situ SE data, the integrated area of the XPS W 4f peaks increased linearly with the number of ALD cycles on SiO2, while a growth delay of ∼20 ALD cycles was observed on Al2O3 (see Figures S2a–S2d in the Supporting Information for W and S XPS peak evolution). As shown in Figure c, distinct W 4f doublet peaks corresponding to WS2 (W4+ oxidation state, binding energy ≈ 32.1 eV and 34.2 eV ) were observed on the SiO2 surface, whereas no W 4f core level (W+4) signals were observed on the Al2O3 surface after 20 ALD cycles. Note that the XPS detection limit for W on top of SiO2 and Al2O3 was estimated to be <0.01 monolayer or ∼2 × 1012 atoms/cm2.[43] A film stoichiometry (S:W) of 2.3 was determined from XPS measurements on SiO2 similar to our previous work.[42] The binding energy of the XPS W 4f core levels (W 4f7/2 and W 4f5/2) appears to match those commonly observed for the 1T phase of WS2.[44] However, we have previously[42] established the growth of a 2H phase of WS2 layers for the same ALD process at a similar temperature (at 300 °C using XRD measurements). Therefore, we expect these layers to be in the 2H phase.

On SiO2, the adsorption of Hacac molecules is known to be very minimal relative to Al2O3,[23] and, consistent with this, no growth delay was observed in this work as discussed above (see Figures a and 2b). Furthermore, XPS depth profiling of the WS2 films revealed no carbon impurity incorporation from Hacac molecules (Figure S2e in the Supporting Information). These results confirm that the addition of Hacac (step A) to the WS2 ALD process (steps B and C) does not influence the WS2 deposition on a SiO2 starting surface and on the WS2 itself. Therefore, our process can be used to selectively deposit pure WS2 (∼2 monolayers) with angstrom-level thickness control on SiO2 in the presence of Al2O3.

To quantify the selectivity of our process, we use the accepted definition of selectivity in the field of area-selective deposition:[45,46]where, θGA and θNGA represent the amount of material present (WS2 in this case) on the growth and non-growth areas, respectively. Using this definition, the selectivity of the ABC-type WS2 ALD process was calculated using the number of W XPS counts in Figure b. After 20 ALD cycles, a high selectivity value of ∼0.95 was obtained. This corresponds to a selective deposition of WS2 with a thickness of ∼1 nm, which is more than one monolayer (one monolayer of WS2 has a thickness of ∼0.65 nm). After 30 cycles, the selectivity was determined to be ∼0.82. This corresponds to a selective deposition of ∼3 monolayers of WS2. As the number of ALD cycles increases, the selectivity starts to decrease drastically (0.45 after 50 ALD cycles). The loss in selectivity primarily arises from ineffective blocking of precursor adsorption by Hacac with increasing number of ALD cycles. A degraded blocking of the precursor adsorption can occur, because of the introduction of surface defects that influence Hacac adsorption negatively and/or incomplete Hacac coverage that allows the precursor molecules to access certain surface reactive sites.[23]

To explore the versatility of our process on various surfaces, the ABC-type WS2 ALD cycles were performed on several starting surfaces including TMDs and transition metal oxides, as shown in Figure . The integrated area of the XPS W 4p3/2 peaks on these surfaces were used to compare the ALD growth. The W 4p3/2 peaks were used to establish the nucleation curves in Figure instead of the W 4f peaks as the W 4f peaks overlaps with some of the elemental XPS peaks of the surface constituents (e.g., Hf 5p3/2 and Ti 3p3/2; see Figure S3 in the Supporting Information). Among various starting surfaces investigated, a growth delay of ∼10 ALD cycles was observed on HfO2 (Figure ). Even after 50 ALD cycles, the integrated W 4p3/2 peak area was significantly lower on HfO2, when compared to the peak area determined for a monolayer of WS2 deposited on SiO2 (dotted black line). The integrated W 4p3/2 peak area was observed to be also significantly lower on TiO2. On the other hand, characteristic XPS W 4p3/2 peaks with relatively large integrated peak areas were observed on 2D TMDs including MoS2, NbS2, and TiS2 starting surfaces, which indicated WS2 film growth without any significant growth delay. As WS2 grows readily on several TMDs surfaces, our process can be also used to selectively grow 2D TMD vertical heterostructures (e.g., WS2 on MoS2) in the presence of the non-growth areas (i.e., Al2O3 and HfO2). Immediate film growth was also observed on transition metal oxides such as MoO3 and Nb2O5.

Figure 3

Integrated area of XPS W 4p3/2 peaks as a function of number ALD cycles for the ABC-type WS2 process on various starting surfaces. The dotted black line serves as a reference to the integrated area of XPS W 4p3/2 peak (2.7 × 104 counts/(s eV)), corresponding to a monolayer of WS2 deposited on SiO2 (prepared using ∼12 ALD cycles; see Figure a).

Raman spectroscopy is a widely used technique to establish and characterize the growth of crystalline TMD layers. Raman measurements revealed that our as-deposited AS-ALD WS2 films on SiO2 at 250° C were amorphous in nature, because signature WS2 Raman fingerprints were not observed. Cross-sectional TEM images (Figure S4a in the Supporting Information) indicated the growth of an amorphous WS2 film matrix embedded with nanocrystalline regions. The growth mechanism for ALD MX2 layers (both crystalline and amorphous) is not well understood in the literature. In our work, we believe that the growth mechanism for amorphous WS2 layers is expected to occur similar to that of crystalline WS2 layers with the exception of long-range ordering of the layers. A model for the WS2 crystalline film growth during plasma-ALD is described in our previous works.[42,47] The growth mode in amorphous WS2 films (layer by layer or island growth, etc.) could vary, depending on the processing conditions and the defects present on the growth surface.

In order to improve the crystallinity of WS2 layers, the samples were annealed at 450 °C in a H2S gas atmosphere for 30 min in the same ALD reactor (pressure during annealing = 300 mTorr). WS2 is considered to be a promising material for applications such as low-power devices in the back-end-of-line (BEOL).[3] Hence, the films were annealed within the thermal budget of BEOL-compatible processing (≤450°C). Upon annealing, signature Raman vibration modes for crystalline WS2 were observed on SiO2 (Figure b). The two characteristic Raman modes at 356 cm–1 and 418 cm–1 wavenumbers correspond to the WS2 in-plane (E21) and out-of-plane (A1) vibrations.[48−50] Cross-sectional TEM imaging (Figure S4b in the Supporting Information) showed a significant improvement in the crystallinity of the WS2 layers upon annealing. The approximate grain size was ∼10 nm, as deduced from top-view TEM (see Figures S4c and S4d in the Supporting Information). A detailed study on the fabrication and characterization of electronic devices such as field effect transistors (FETs), using the annealed WS2 layers (including temperature-dependent resistivity measurements), will be performed in a separate study. The root-mean-square (rms) surface roughness of the WS2 films increased from ∼0.1 nm to ∼0.4 nm upon annealing, as determined from AFM measurements (Figure S5 in the Supporting Information). No Raman peaks were observed after performing WS2 ABC cycles on Al2O3 as expected (see Figure b).

Figure 4

(a) Scanning electron microscopy (SEM) images of the Al2O3/SiO2 patterned samples. (b) Raman spectra showing the characteristic in-plane (E21) and out-of-plane (A1) Raman modes of WS2 on SiO2 after annealing at 450 °C and (c) the corresponding Raman E21 peak intensity line scans over the Al2O3/SiO2 patterned samples after 20 ABC-type WS2 ALD cycles. (d) XPS elemental W and Al line scans after 20 ABC-type WS2 ALD cycles on the Al2O3/SiO2 patterned samples.

As a proof-of-concept, we tested our ABC-type AS-ALD process on patterned Al2O3/SiO2 surfaces. ALD-grown Al2O3 was patterned on ALD-grown SiO2 (Figure a), using a regular lift-off process. After performing 20 ALD cycles, the patterned samples were annealed at 450 °C in a H2S atmosphere. Raman spectroscopy line scans were performed to investigate the selectivity over the patterned surfaces. The line scan of the E21 Raman mode over the patterned surface revealed very clear and sharp transitions at the SiO2/Al2O3 interfaces with steep slopes (Figure c).

XPS line scans were also performed on the annealed patterns to investigate the selectivity (see Figure S6 in the Supporting Information for XPS W 4f core-level spectra). The XPS line scans in Figure d show strong W signals in the SiO2 regions, whereas no W signals were observed in the Al2O3 regions. The overlap of the W and Al line scans at the interface can be attributed to the large spot size of the X-ray beam (∼70 μm), which is comparable to the region of overlap and much larger than the spot size of the Raman laser (∼5 μm).

In conclusion, we have demonstrated the area-selective deposition of 2D WS2 nanolayers using ALD in a bottom-up processing approach. AS-ALD of WS2 was achieved using acetylacetone (Hacac) inhibitor (A), bis(tert-butylimido)-bis(dimethylamido)-tungsten precursor (B), and H2S plasma (C) pulses in an ABC-type ALD process at a low deposition temperature of 250 °C. With this approach, WS2 nanolayers are readily deposited on SiO2, various 2D TMDs and transition metal oxides, while growth on Al2O3 and HfO2 surfaces is effectively blocked. On the growth areas, pure WS2 is deposited with angstrom-level thickness control. The AS-ALD WS2 films exhibited sharp Raman peaks, a fingerprint of crystalline film growth, upon annealing at BEOL-compatible temperatures (≤450 °C). As a proof of concept, the AS-ALD process has been demonstrated on patterned Al2O3/SiO2 surfaces. Raman line scans over the SiO2/Al2O3 patterns showed very sharp peak intensity transitions at the interface. The selectivity of our process was quantified, and, after 20 ALD cycles (at least one monolayer), a high selectivity of 0.95 is obtained. The results obtained in this work can be used as a platform to further explore the area-selective deposition of other 2D TMD materials.

Experimental Procedures

All depositions were performed in a commercial FlexAL ALD reactor from Oxford Instruments. In essence, the reaction chamber is equipped with a remote inductively coupled plasma (ICP) source, a 200 mm substrate table, and a turbo molecular pump that enables a base pressure of 10–6 Torr. The reaction chamber wall temperature was set to 150 °C (maximum possible value), and the substrate table temperature was set to 250 °C. With these settings, the substrate temperature was estimated to be ∼200 °C, using a thermocouple on reference samples. The mismatch between the set and the estimate temperature can arise from limited contact under vacuum. Table S1 in the Supporting Information compares the set temperature (referred to as deposition temperature) and estimated temperatures. Throughout this work, deposition temperatures are used for discussion. The WS2 AS-ALD process was primarily tested and characterized on SiO2 (growth area) and Al2O3 (non-growth area). Both SiO2 and Al2O3 (∼30 nm) were deposited on c-Si with 450 nm thermal oxide using well-established ALD processes. Other starting surfaces reported in this work (various 2D TMDs and transition metal oxides) were also deposited using ALD processes. All substrates were subjected to a 20 min pre-heating step in a 200 mTorr Ar environment to stabilize the substrate temperature. The substrates were then subjected to a H2 plasma for 5 min prior to the AS-ALD process. The H2 plasma power was set to 500 W and the pressure in the chamber was maintained at 50 mTorr. The following optimized exposures were used in our PEALD recipe: 3 pulses of 5 s each for the Hacac dose (step A), 10 s for the bis(tert-butylimido)-bis(dimethylamido)-tungsten precursor dose (step B), and 60 s for the H2S plasma exposure (step C). The Hacac inhibitor (Sigma–Aldrich, ≥99% purity) was stored in a canister at room temperature and was vapor-drawn into the reaction chamber. The two-step (BC) WS2 PEALD recipe, employing the tungsten precursor and H2S plasma pulses reported in our previous work, was used as a starting point to deposit WS2 films.[42] The growth inhibition on Al2O3 and thereby, the selectivity of our process was observed to be significantly dependent on ALD processing conditions including H2S plasma exposure time, H2 plasma pretreatment of Al2O3, and deposition temperature. This is further described in the Supporting Information (Figures S7 and S8). Frequent usage of H2S plasma exposures beyond 60 s led to flaking of deposited material on the substrate table. Thus, H2S plasma exposures were seemingly limited to 60 s during most of our processing.

In situ spectroscopic ellipsometry (SE) was used to measure the apparent WS2 film thickness using a B-spline function or Cauchy-based parametrization to model the experimental SE data. All in situ SE measurements were performed using a J.A. Woollam Model M2000F ellipsometer. X-ray photoelectron spectroscopy (XPS) was used to detect the presence of W on SiO2 and Al2O3 surfaces. The XPS detection limit for W on top of SiO2 and Al2O3 was determined to be below ∼0.01 monolayer (∼2 × 1012 atoms/cm2).[43] All XPS studies were performed using a Thermo Scientific KA1066 spectrometer with monochromatic Al Kα X-ray source (hν = 1486.6 eV). The spot size of the incident X-rays was ∼70 μm. XPS data analysis was performed using the Avantage XPS software.

Scanning electron microscopy (SEM) images were obtained using a Zeiss Sigma microscope with an in-column, secondary electron detector. The acceleration voltage of the electron beam was 2 keV. Raman spectroscopy was performed to investigate the characteristic vibrational modes in WS2 films. Raman spectra were obtained using a Renishaw Invia Raman microscope equipped with a 514 nm laser at a power of ∼0.5 mW. The laser spot size was ∼5 μm. High-angle annular dark field (HAADF) scanning transmission electron microscopy (STEM) images were obtained using a probe-corrected JEOL Model JEM-ARM200F transmission electron microscope (TEM) operated at 80 kV. For top-view STEM imaging, WS2 layers were deposited on Si3N4 windows coated with 5 nm of ALD SiO2. For cross-sectional TEM studies, WS2 layers deposited on Si3N4 windows were coated with an additional SiO2 protective layer on top. A focused ion beam was used to create a cross-sectional sample using the standard lift-out method. Atomic force microscopy (AFM) measurements were performed using a NT-MDT Solver P47 AFM system.