Efficacy of boron nitride encapsulation against plasma-processing of 2D semiconductor layers

The isolation of layered van der Waals crystals into atomically thin, two-dimensional (2D) structures has led to significant new insights into condensed matter physics,^1,2 which have in turn led to fundamentally new electronic device designs.^3,4 Significant effort has been spent on the study of synthetic routes, fundamental physical phenomena, and device properties in these systems. However, there has been less focus on device processing. Ultimately, the applications of all 2D materials will require precise control over crystalline quality, thickness (layer number), and, thus, over device processing conditions. With this background, we address an old but relevant problem associated with a ubiquitous process in microelectronics: plasma processing.

Plasma processing is widely used to clean, functionalize, and passivate surfaces, as well as to etch materials.^5–11 It has been applied to 2D materials since the early days of graphene device research.^12–15 However, it was soon realized that energetic plasmas could affect the structural and chemical stability of 2D materials and thereby degrade lateral transport in electronic devices.^16–18 Therefore, high-quality, stable, and scalable encapsulants are needed to protect 2D device channels, and they continue to be an object of research. However, it is unclear if plasma processing leads to charge incorporation in encapsulating layers that degrades carrier transport in a structurally intact 2D material or if it directly damages the material. Often charge transport measurements are used to infer the role of defect formation in these cases, but transport measurements only provide indirect evidence of defect introduction. These facts motivate us to explore how plasma etching conditions affect both the encapsulation layer and the active channel using a direct approach.

In all high-performance 2D semiconductor devices, it is essential to isolate or encapsulate the active semiconductor layers to limit charge inhomogeneities and exposure to processing chemicals. Encapsulation is needed to protect electronic devices such as transistors, where the channel is buried under a dielectric insulator, and optoelectronic devices, where the junction is buried under contacts and barrier layers. Charge inhomogeneity results from trapped charges, dangling bonds, and dipoles of ionic bonds, all of which impede electronic transport and inhibit radiative recombination in 2D layers. Organic layers (polymers) and flat, highly crystalline, and nearly covalent materials such as hexagonal boron nitride (h-BN)¹⁹ have proven to be effective substrates and encapsulants for 2D channels and active layers.^20–22 However, polymers and small organic molecules are susceptible to thermal damage as well as swelling and dissolution upon solvent exposure. This means that they are unsuitable as permanent encapsulants during semiconductor device fabrication and processing. However, h-BN possesses high chemical and thermal stability, making it a potentially superior encapsulant. Several schemes for direct growth and transfer of h-BN encapsulated graphene and 2D semiconductor devices have been developed.^23–26 In these studies, a postlithography etching step is essential for defining channels and contacts. Although several studies assume that h-BN is adequately protecting the underlying active 2D layer,^27–30 there have been no systematic studies that provide mechanistic insight into its effectiveness. In this study, we perform a systematic investigation of the efficacy of the h-BN layer as an encapsulant. We correlate optical spectroscopy and atomic-resolution imaging analysis to understand how plasma dose variation, sequential plasma exposure, and encapsulant and underlayer thickness affect the rates of damage accumulation.

Mechanically exfoliated 2D MoS₂ layers were prepared using the conventional scotch tape method, as described elsewhere.³¹ The thickness of the exfoliated layers was intentionally chosen such that each sample could be reproduced for a number of repetitive analyses. Exfoliated h-BN and MoS₂ layers were transferred to the SiO₂/Si substrate by dry transfer technique utilizing a poly-dimethyl siloxane (PDMS) stamp. The dry transfer technique uses a motorized micromanipulator stage (X-Y-Z axes), attached with an optical microscope along-with with a home-made heating stage (based on pyroelectric material). After transferring MoS₂ layers to SiO₂/Si substrates, samples were annealed in a quartz tube furnace in a closed gas (Ar + H₂) environment to remove all PDMS contaminants. Annealing was performed at 300 °C for 4 h to clean contaminants as well as release the strain developed during the pressure-based dry transfer method. It is worth noting that the 2H phase MoS₂ is extremely stable at 300 °C in a reducing atmosphere. This has been confirmed by several prior studies and no evidence of structural phase changes or defect formation has been observed at our chosen annealing temperatures.^32,33 Similarly, h-BN and MoS₂ layers were transferred to a dedicated SiNx TEM grid (Norcada Inc., 3 × 3 array of 100 μm diameter holes) and annealed in the same manner to remove PDMS contaminations prior to irradiation and subsequent scanning transmission electron microscopy (STEM) characterization. Two different plasma irradiation systems are used here for the different samples at different exposure times. Ultra-pure Ar gas (99.995%) was used in all the plasma exposure analysis. A dedicated plasma cleaning system was available for holding TEM holders such that we can treat samples for multiple exposure times while the TEM grid installed into TEM holder. Plasma exposure time for TEM analysis (sample on TEM grid) vs Raman analysis (sample on SiO₂/Si substrate) is different as sample positioning distance from the plasma ignition source is different in both the systems based on the available configuration.

Raman spectroscopy as well as photoluminescence measurement is carried out using the LabRAM HR Evolution HORIBA system. A 633 nm laser with a spot size ∼0.5 μm, 1% laser power was used for diagnosed defect related analysis in h-BN as well as MoS₂ layers. For the case, PL analysis, 405 nm laser, was used with 0.1% laser power (0.25 μW) with 1 s acquisition time. Optical microscope from Olympus, USA was used to capture and analyze all the samples before and after plasma treatment. The plasma irradiation system (Tergeo Plasma Cleaner, PIE Scientific) was used with 65 sccm of Ar (99.995% purity) flow and 50-W transmitted RF power under 0.35 Torr base vacuum environment. The same parameters were utilized in the case of a dedicated TEM plasma cleaner system (Gatan, Solarus 950). High-Angle Annular Dark Field (HAADF)-STEM has been used to directly visualize all the samples for defects evolution. An aberration-corrected JEOL NEOARM STEM, operating at an accelerating voltage of 200 kV and convergence angle of 25–29 mrad, is used for all samples. For the JEOL NEOARM STEM, the condenser lens aperture was 40 μm with a camera length of 4 cm for imaging and the probe current was 120 pA. All of the captured STEM images were collected using gatan gms software and associated Gatan bright-field and high-angle annular dark field detectors. Experimentally acquired STEM images are smoothed using the adaptive gaussian blur function (with a radius of 1–2 pixels) available in imagej.

During sample preparation, device fabrication, and postprocessing, a 2D material or heterostructure is likely to encounter several forms of energetic radiation sources. Radiation sources that are used include electron, ion (Ar, He, and Xe), and photon (laser, UV) fluxes. However, ion beams and UV light are the most routinely adopted in etching and lithography processes, respectively. Here, we investigate the etching behavior of h-BN, MoS₂, and their heterostructures when subject to Ar⁺ ion plasma exposure. Argon ion plasmas are the most commonly used as argon does not form chemical bonds with the sample due to its inherent inertness. Argon is also a heavy enough ion that can provide sufficient kinetic energy to etch samples at reasonable accelerating voltages.

Figure 1 shows a schematic of the physically stacked 2D heterostructure samples used in this study. Thin flakes of h-BN and MoS₂ were mechanically exfoliated from the bulk using Scotch tape. They were subsequently transferred onto oxidized silicon wafers or dry PDMS stamps. Both heterostructures and samples transferred onto STEM heating platforms were created using direct transfer via viscoelastic PDMS stamps. Further details of the sample preparation and transfer methods are provided in Sec. II A.

Raman spectroscopy is used to provide a global measure of defect formation as a function of plasma exposure for both h-BN shield and unshielded MoS₂ (Fig. 2). Figure 2(a) presents an optical microscopy (OM) image of a stacked heterostructure of h-BN (circled). There are also regions of pure h-BN and pure MoS₂ adjacent (arrows). Height images and corresponding h-BN thickness profiles are presented for the same heterostructure (h-BN/MoS₂) in Fig. 2(b). Raman spectra as a function of Ar⁺ exposure time from the marked unshielded and shielded MoS₂ regions shown in Fig. 2(c). The distinct peak at 226 cm⁻¹ [labeled LA (M)] has been associated in prior work with the appearance of defects arising from the scattering of phonons at the Brillion zone edge.^34–36 Similarly, E_2g and A_1g mode intensities decrease with increase of plasma exposure time, confirming that the lattice experiences increasing damage with exposure time. Control data from the h-BN layer is presented as Figs. S1 and S2 in the supplementary material,⁴⁹ and changes in optical contrast in the OM image are presented in Fig. S3 in the supplementary material.⁴⁹ Pristine MoS₂ has a very small LA (M) Raman mode as a monolayer and is negligible for few-layer MoS₂: this LA mode most likely originates from defects created during the exfoliation and transfer process. Two and half minutes of plasma exposure leads to a significantly enhanced LA (M) signal, indicating the formation of a significant quantity of defects. After 5 min of plasma exposure, the intensity of the LA (M) Raman mode is of the same magnitude as the lattice A_1g(M) – LA (M) phonon mode, indicating significant damage accumulation.

We have also examined the Ar⁺ plasma exposure effects for shorter intervals of time (within a 1-m duration, using 20 s step sizes) across different layer thicknesses of unshielded MoS₂ and report those results in Fig. S4 in the supplementary material.⁴⁹ We find that defect formation occurs at levels measurable via Raman Spectroscopy upon Ar⁺ plasma exposure for time scales as short as 30 s for an unprotected monolayer MoS₂ sample. In contrast, Ar⁺ plasma exposure times of 2.5 and 5 min showed a less significant rise in the LA (M) Raman signal for the region of the MoS₂ flake that was shielded under a few layers of h-BN, as shown in Fig. S5 in the supplementary material.⁴⁹ While Raman spectroscopy is a reliable way to ascertain lattice damage, Raman signals from 2D materials are inherently inefficient. Photoluminescence is a far more sensitive measure of crystal quality and defect density for a high-quality, direct band gap semiconductor. The smallest deviation from perfect crystalline order can induce nonradiative recombination that reduces the primary PL efficiency.³⁷ Deviations from crystalline order can also introduce trap states that lead to photoluminescence that is red-shifted from the original PL peak location.^37,38

We have recorded the PL response after plasma exposure for shielded and unshielded monolayer MoS₂. Figure 2(a) presents an OM image of the heterostructure sample from which we obtained PL spectra. Region 1 has a monolayer of MoS₂ that is shielded under a thin h-BN flake (∼11.2 nm thin), and region 2 is unshielded monolayer MoS₂. The PL data are presented as Figs. 2(d) and 2(e). The primary peak at ∼652 nm (∼1.9 eV) is caused by X₀-exciton emission,³⁹ and the peak at ∼740 nm (∼1.67 eV) is known to be a defect bound exciton peak.^40,41 Similar PL spectra for the low incident light intensity have observed previously in the literature for MoS₂.^42–44 The PL intensity of the unshielded MoS₂ region before irradiation (“pristine”) is shown in red in Fig. 2(d). The PL intensity of the unshielded MoS₂ flake is comparatively weaker than the shielded region for pristine samples. This is an optical effect related to the formation of the heterostructure: the h-BN has a higher index of refraction, while the SiO₂ wafer has a lower index of refraction. The higher index medium thus enables increased light extraction. With increasing plasma exposure, we see two effects, which we summarize in Fig. 2(e). First, with increasing Ar⁺ plasma exposure, the peak associated with defect bound exciton emission (X_B) increases, clearly indicating the increase in defect content in the layer as the incident energetic Ar₊ ions penetrate the h-BN shield. Second, there is a concurrent, correlated decrease in the primary neutral exciton (X₀) peak intensity. Interestingly, the h-BN/MoS₂ heterostructure shows a sudden increase in the MoS₂-X_B intensity with 3.5 min and greater plasma exposure time. These data suggest that the h-BN layer is an effective shield in the heterostructure up to a certain threshold of plasma exposure only.

Optical and vibrational spectroscopies are an effective way to track general trends in damage accumulation. However, they do not provide atomic-scale information about the accumulation of individual defects. To complement these spectroscopies, we have performed extensive characterization using aberration-corrected high-angle annular dark field (HAADF) scanning transmission electron microscopy (STEM) imaging. Flake samples were prepared using the same methodology described in Fig. 1 and transferred onto a SiNx TEM grid that contains separated 100 μm diameter holes. This allowed the heterostructures to be irradiated with Ar⁺ plasma without damaging the underlying SiNx membrane. Details of the plasma irradiation system used for analyzing all STEM samples are described in Sec. II.

A basic MoS₂ structural model and low magnification as well as atomically resolved HAADF-STEM imaging for pristine and different plasma exposed MoS₂ regions are shown in Figs. S6 and S7 in the supplementary material.⁴⁹ HAADF-STEM imaging was performed after successive plasma exposures with varying time intervals (5, 10, and 15 s). This leads to four sample conditions: “pristine” (0 s exposure) and 5, 15, and 30 s of cumulative exposure, respectively, as shown in Fig. 3. We compare the defect evolution analysis for two different regions of the same sample, an h-BN shielded MoS₂ heterostructure versus an unshielded (bare) MoS₂ region. It is worth noting that even though a vertical van der Waals heterostructure of h-BN over MoS₂ is being imaged in transmission in Figs. 3(a)–3(d), nearly all of the image signals are from the MoS₂ layer. There are two reasons for this. First, aberration-corrected STEM images have a very small depth of field, and we maintained the focus on the MoS₂ layer.⁴⁵ Second, the intensity in HAADF images scales with Z^1.65,⁴⁵ and both Mo and S are significantly heavier than B and N. Therefore, nearly all the signal that comprises images in Fig. 3 arises from the MoS₂ layer. Nonetheless, a fast Fourier transformation (FFT) diffractogram can detect the periodicity in the image from both the MoS₂ and h-BN lattices, and it indicates that there is a 12.2° twist angle between them. Figures 3(a)–3(c) show that the h-BN shielded MoS₂ heterostructure experiences negligible defect formation until a cumulative 15 s of plasma exposure. However, after 30 cumulative seconds of plasma exposure, the h-BN/MoS₂ heterostructure shows visible lattice damage (circled).

In contrast, bare MoS₂ shows signatures of visible lattice damage after merely 5 s of exposure [Fig. 3(f)]. Contrast in the different HAADF-STEM images (from pristine to varying plasma exposure time) is not quantitatively comparable since each individual image is self-normalized by the image acquisition software (gatan gms) during image acquisition to its maximum intensity. Gray-scale histograms for the images in Fig. 3 are presented in Fig. S8 in the supplementary material.⁴⁹ Each histogram has been normalized to the respective maximum image intensity and scaled to a common distribution based on the mean intensity and standard deviation. All the image processing and normalization carried out using “Sci-kit” as well as “Fiji” to read and process the images.^46,47 Lattice damage is induced by the plasma in both the bilayer and the few-layer portion of the MoS₂ flake following just 5 s of exposure. Additional plasma exposure [Figs. 3(f)–3(h)] leads to progressively more severe lattice damage, resulting in near amorphization of the bilayer region after 30 s of plasma exposure [Fig. 3(h)]. These images also indicate that the damage grows in spatially localized regions with increasing exposure. In other words, damage accumulates at defects introduced at earlier times not through continued renucleation. We summarize that the defective regions created by that plasma are passivated by ambient atmospheric species that accumulate during transfer from the plasma chamber to the TEM column. This chemical passivation stabilizes these regions, and subsequent plasma exposure leads to increased damage at these undercoordinated sits. Additional low magnification STEM images for both the heterostructure and bare sample regions are shown in Fig. S6 in the supplementary material,⁴⁹ and another set of atomic-scale STEM image showing clustered patches of defected areas after 30 s of plasma exposure are presented in Fig. S9 in the supplementary material.⁴⁹ We have performed electron energy loss spectroscopy (EELS) measurements after 15 s of plasma exposure time (Fig. S10 in the supplementary material).⁴⁹ The atomically resolved EELS data showed the presence of adsorbed oxygen, which are found in defective 2D MoS₂ basal planes.⁴⁸ Mo and S signals are weaker since the thickness of the MoS₂ layer (∼4–5 nm) is very thin in comparison with the total sample thickness which includes a 20 nm SiNx membrane underneath as well as the h-BN layer (∼10 nm) covering MoS₂ on the top. We have also studied EELS for a heterostructure (h-BN/MoS₂) which undergoes continuous plasma exposure up to 20 s. We have first acquired the core loss EELS spectra for oxygen in the pristine state and then after 20 s plasma exposure from the same region for unshielded as well as the shielded region of MoS₂ flake, as presented in Fig. S11 in the supplementary material.⁴⁹ A slight increased amount of oxygen has only observed. In the process of transferring the samples from plasma processing system into the TEM vacuum column, our samples get exposed to ambient air for a maximum of 10 min. This observation supports our hypothesis that ambient exposure stabilizes defects during transfer in and out of the microscope.

To understand the effect of air exposure during sequential plasma bombardment, we subjected a different heterostructure stack to a continuous plasma exposure for 15 s. We observed widespread damage and etching of the basal plane in the unshielded MoS₂, as shown in Fig. 4. This is in stark contrast with the sequential plasma exposure combined ambient air exposure [vide supra, Fig. 3(g)], where the lattice damage is more uniform. We see that increasing the total plasma exposure creates a large number of defects in the h-BN as well as the unshielded MoS₂ layer, as shown in Figs. 4(a) and 4(b), respectively. Furthermore, the defects created here are much larger and are not individual point defects, instead, holes of 3–5 nm diameter form. This suggests that without the stabilization provided by oxygen during the transfer process, there is an accelerated accumulation of defects during irradiation. Following the nucleation of an individual point defect, the undercoordinated atoms can readily be knocked off their lattice sites, which, combined with extended migration during the continuous plasma exposure, could lead to the growth of substantially large voids.

The h-BN/MoS₂ heterostructure remains remarkably intact, as we can see in Figs. 4(b) and 4(c). This is consistent with the ability of few-layer h-BN to be an effective shield, as shown in the PL data of Fig. 2. We have also performed STEM imaging for a continuous plasma exposure of 20 s. Figure S12 in the supplementary material⁴⁹ presents AFM height images and corresponding thickness profiles from the different section of h-BN/MoS₂ heterostructures. Furthermore, the same flake was analyzed using Raman spectroscopy and the defect mode is clearly seen in the case of unshielded MoS₂ (Fig. S13 in the supplementary material).⁴⁹ Afterward, STEM imaging from the same region of two different sample configurations—unshielded and shielded MoS₂—is presented after 20 s continuous plasma exposure in Fig. S14 in the supplementary material.⁴⁹ We have also separately examined the defect evolution, and corresponding lattice damage within just the h-BN layers that are being used as a shield in the heterostructure the contrast on bare h-BN in Fig. 4(a) is not as clear. However, the voids are visible and marked. We have observed similar void formation and damage patches in h-BN as well upon continuous plasma exposure, see Figs. S15–S17 in the supplementary material⁴⁹ for further details. We hypothesize that these localized voids act as a channel for further penetration of plasma ions through layer by layer etching, ultimately reaching the underlying layer, which then causes localized damage in the MoS₂ (Fig. S9 in the supplementary material).⁴⁹ Again, this is in stark contrast with sequential exposure to plasma in multiple small steps with ambient air contact between steps. The creation of defective patches and corresponding voids can be easily seen in the AFM height image of h-BN layers (Fig. S18 in the supplementary material).⁴⁹ 

We have studied the dynamics of Ar⁺ plasma-induced defect generation and etching in atomically thin van der Waals heterostructures of h-BN and MoS₂. We observe that h-BN effectively shields underlying layers from plasma damage. An atomic-scale imaging suggests that plasma-induced lattice damage is instantaneous for unshielded MoS₂, whereas shielded MoS₂ is protected by the h-BN until a certain extent of exposure, and that the extent of damage as a function of the time of exposure depends upon the h-BN thickness. Finally, we conclude that continuous plasma exposure is more damaging as opposed to via sequential exposure. These results indicate that h-BN encapsulation does provide limited protection to underlying MoS₂ layers during plasma processing but that the level of protection varies on a range of parameters.

This work was carried out in part at the Singh Center for Nanotechnology at the University of Pennsylvania, which is supported by the National Science Foundation (NSF) National Nanotechnology Coordinated Infrastructure Program Grant No. NNCI-1542153. D.J., E.A.S., and P.K. acknowledge primary support for this work from via the NSF DMR Electronic Photonic and Magnetic Materials (EPM) core program (Grant No. DMR-1905853) as well as the University of Pennsylvania Laboratory for Research on the Structure of Matter, a Materials Research Science and Engineering Center (MRSEC) supported by the National Science Foundation (No. DMR-1720530). K.S.F. was supported by the LRSM MRSEC REU and the Penn-UPRH Partnership for Research and Education in Materials (PREM), Program No. NSF-DMR-1523463. N.A. and D.J. acknowledge support from Vagelos Integrated Program for Energy Research at the University of Pennsylvania as well as the Center for Undergraduate Research and Fellowships at Penn. D.J. also acknowledges support for this work by the US Army Research Office under Contract No. W911NF1910109. A.C.F. and E.A.S. would like to acknowledge the Vagelos Institute for Energy Science and Technology at the University of Pennsylvania for a graduate fellowship to A.C.F. The authors thank James Horwath for help/assistance in the STEM image histogram analyses. The authors also thank Douglas Yates and Jamie Ford in the Singh Center for Nanotechnology for help with the TEM/STEM measurements. There are no conflicts of interest to declare.

Most of the data sets generated or analyzed during this study are included in this article and its supplementary material. Additional findings of this study are available from the corresponding author upon reasonable request.