Nano-optical Visualization of Interlayer Interactions in WSe2/WS2 Heterostructures.

The interplay between excitons and phonons governs the optical and electronic properties of transition metal dichalcogenides (TMDs). Even though a number of linear and nonlinear optical-, electron-, and photoelectron-based approaches have been developed and/or adopted to characterize excitons and phonons in single/few-layer TMDs and their heterostructures, no existing method is capable of directly probing ultralow-frequency and interlayer phonons on the nanoscale. To this end, we developed ultralow-frequency tip-enhanced Raman spectroscopy, which allows spectrally and spatially resolved chemical and structural nanoimaging of WSe2/WS2 heterostructures. In this work, we apply this method to analyze phonons in nanobubbles that are sustained in these heterobilayers. Our method is capable of directly probing interlayer (de)coupling using our novel structurally sensitive nano-optical probe and the interplay between excitons and interlayer/intralayer phonons through correlation analysis of the recorded spectral images.

Single- and few-layer 2D transition metal dichalcogenide (TMD) crystals exhibit unique electronic and optical properties. Their early applications in valleytronics[1−4] and field-effect transistors[5] and more recent implementations that take advantage of TMDs in hybrid 2D/3D architectures[6] have motivated quests to characterize excitons and phonons in TMDs as well as TMD-featuring devices.[7,8] Ultralow-frequency Raman spectroscopic measurements (<50 cm–1) that track interlayer breathing (out-of-plane) and shearing (in-plane) modes in few-layer TMDs are particularly relevant to this work.[9,10] To date, the spatial resolution in these measurements is nonetheless diffraction-limited. Increasing the spatial resolution in such measurements would allow us to better understand the interplay between excitons and local nanostructural motifs that are well-known to significantly affect the properties of TMDs. This is bolstered through tip-enhanced Raman spectroscopy (TERS) and tip-enhanced photoluminescence (TEPL) measurements of TMD monolayers and heterostructures, where nanoscale structure–function relationships become possible.[11−15]

Stacking different TMD monolayers leads to a host of novel properties in type II heterobilayers, including long-lived interlayer excitons and ultrafast (phonon-mediated[16]) interfacial charge transfer.[17−21] Moiré superlattices in such constructs also pave the way toward on-demand band gap/electronic structure engineering with an ultimate aim of unlocking a host of quantum phenomena that can be harnessed in modern optoelectronic devices.[17,22] Another approach to tailoring the electronic states of few-layer TMDs is through localized strain.[23,24] Recent studies exploited nanobubbles in monolayer WSe2 and heterobilayers composed of combinations of MoS2, MoSe2, WS2, and WSe2 to examine how strain affects the electronic and optical response of these systems on the nanoscale[23,25] using TEPL. Here we take advantage of a similar platform to understand the interplay between interlayer phonons and excitons in a WSe2/WS2 heterobilayer.

Excitons in WSe2/WS2 heterobilayers have been previously investigated using diffraction-limited hyperspectral optical microscopy.[26,27] Compared with its monolayer constituents, the bilayer sample exhibited spectral broadening and exciton resonance shifts in the visible region of the electromagnetic spectrum. Broadening was associated with faster charge separation, which was also confirmed by femtosecond pump–probe spectroscopy.[26] The lowest exciton resonance (A) in monolayer WSe2 (746 nm) shifts to 774 nm in the WSe2/WS2 heterobilayer,[26] which is close to the excitation wavelength of 785 nm that we use in this work. The 774 nm band in the heterobilayer is nonetheless complex in character, particularly when the two heterolayers are closely aligned,[17,22] as is the case herein (see Figure S1). In particular, the formation of moiré superlattices is accompanied by the emergence of a host of excitonic states that preclude simple assignments of the 774 nm band on the basis of a direct comparison between monolayer and homo/heterobilayer spectra. Moreover, diffraction-limited measurements average over spatially varying excitonic resonances, e.g., localized excitons previously observed in TEPL mapping of defects in TMD monolayers and heterobilayers.[23,25] Prior to an introduction of our distinct approach to the problem, we begin by describing our sample and its general characteristics.

Our samples were prepared by mechanical exfoliation of WSe2 and WS2 monolayers on polydimethylsiloxane (PDMS) and their subsequent transfer onto a gold substrate using all-dry viscoelastic stamping.[28] Moiré patterns were observed across the sample (Figure S1) in topographic images, which documents the intimate alignment between the two layers in the heterostructure. An atomic force microscopy (AFM) image of a selected area of the Au/WSe2/WS2 sample is shown in Figure . Besides nanometrically flat regions, the AFM image in Figure also shows nanobubbles and wrinkles. This is consistent with the general topographic features that were recently observed in similarly prepared TMD heterobilayers[25] and are common for van der Waals heterostructures.[29] A simultaneously recorded AFM image and nano-Raman map along with a far-field Raman map of an isolated nanobubble are shown in Figure . The topographic image (Figure A) reveals an asymmetric morphology of the nanobubble. The same is evident in the nano-Raman map (Figure B), where a horseshoelike scattering profile with a dim response toward the center of the protruded structure is observed. Our observed TERS image profile is reminiscent of TEPL maps of nanobubbles of monolayers and heterobilayers of TMDs.[23,25] There, the observed patterns were faithfully reproduced using theoretical confinement potentials obtained from atomistic models based on nanobubble strain calculations and Harrison’s rule.[23] We similarly associate our observed profiles (also see Figure S2) with strain in our protruded nanostructural motifs. However, our structures are more complicated than nanobubbles composed of single layers,[23] in part because of the contributions of multiple resonances[25] herein arising from WS2, WSe2, and their combination in the heterobilayers. As we illustrate below, two-dimensional correlation analysis aids in identifying the operative local resonances through their correlations with intralayer and interlayer phonons.

Figure 1

Representative AFM image of our WS2/WSe2/Au sample. Flat regions, bubbles, and a wrinkle are visible in this image and all throughout the sample used in this work.

Figure 2

Simultaneously recorded (A) AFM image and (B) 22 cm–1 TERS intensity map of a nanobubble along with (C) a far-field Raman map (also at 22 cm–1) of the same region. Experimental conditions: 785 nm laser excitation, 100 μW/μm2, 50 ms time integration. The lateral step sizes used in the near-field and far-field maps were ∼1.7 and 5 nm, respectively.

The TERS spectral image of the nanobubble is analyzed in detail in Figure . Spectra averaged over three distinct areas (marked 1–3) in the image in Figure A are shown on the same plot in Figure D. Several observations in these spectra are important to highlight. First, with the exception of the center of the bubble (region 1), spectra taken at peripheral regions of the bubble (region 2) as well as at flat regions of the substrate (region 3) are dominated by the interlayer phonon signature at 22 cm–1.[9,30] This is evident in Figure D,E, where normalized spectra from regions 1–3 are shown along with the spatially averaged far-field response taken from Figure C. Second, the disappearance of the interlayer phonon signature is accompanied by a rise in the relative intensity of the A1g mode of WS2 at ∼416 cm–1. As discussed below, the appearance of the WS2 mode is accompanied by the appearance of a local optical resonance that is distinct from its analogue in the flat regions of the substrate. Taken together, our first and second observations suggest that the two layers are decoupled toward the center of the bubble, which is marked by the disappearance of the interlayer phonon signature. The mutual exclusivity of the 22 cm–1 interlayer phonon and the 416 cm–1 WS2 A1g mode is visualized along with the bubble topography for the cross section in Figure B. The interlayer phonon is present only up to approximately half of the bubble height, and then its intensity abruptly vanishes; the 416 cm–1 mode appears instead, peaking in intensity exactly at the top of the bubble. Identical behavior was also observed for a different bubble (see Figure S2). The decoupling of the layers is schematically illustrated in Figure C. The appearance of the WS2 signature further suggests that a localized optical resonance contributes to this optical signature.[25] The spectroscopic features arising from WSe2 intralayer phonons,[9] which are resonantly enhanced at our excitation wavelength, otherwise dominate the nano-optical response away from the bubbles. This brings us to our third observation: the contribution of intralayer photoluminescence seems to be significantly dimmed, which is consistent with prior observations,[26,30] wherein this effect was associated with strong interlayer coupling. Overall, the results in Figure allow us to infer a spatial resolution on the order of ∼10 nm in our measurements.

Figure 3

(A) Combined TERS map showing the peak intensity (background-corrected) of the 22 cm–1 (blue) and 420 cm–1 (red) Raman bands. (B) Horizontal cross section of the nanobubble height and the TERS intensities along the dashed line in (A). The same color coding as in (A) is used. (C) Schematic illustration of the inferred local structure of the nanobubbles. (D) TERS spectra averaged over the correspondingly enumerated areas in (A). (E) Details of the spectra in (D), normalized and expanded for clarity. Blue and red stripes in (D) and (E) highlight the TERS peaks whose intensities are plotted in (A).

In Figure , we further analyze the recorded TERS map using 2D correlation analysis.[31,32] The goal is to understand the correlations between the different phenomena that are captured in our recorded image cube without interpretation bias that could potentially interfere in multipeak fitting procedures. The correlation map is shown in Figure A, whereas correlation slices taken at different resonances are shown in Figure B. The cross-sectional cuts may be used to infer the correlations between the different sharp peaks we observe that arise from interlayer (22 cm–1) and intralayer (>100 cm–1) phonons and, more importantly, the correlations between the phonons and the broad photoluminescence signatures that track excitonic states in our heterobilayer. The 22 cm–1 correlation slice shown in Figure B (black spectrum) reveals a positive correlation between the interlayer phonon and a background centered at ∼780 nm assigned to the WSe2 exciton. The central wavelength of the background is consistent with previously reported WS2/WSe2 reflectance spectra that exhibited a broad resonance at 774 nm.[26] The red-shifted resonance we capture in our case is attributed to the effect of the Au substrate here, in contrast to sapphire in the previous work.[26] The 22 cm–1 slice also shows (i) a weak correlation between the interlayer phonon and WSe2 intralayer phonons (at ∼250 cm–1) and (ii) a negative correlation between the interlayer phonon and the A1g mode of WS2. The fact that the interlayer phonon is correlated with the ∼780 nm resonance and noncorrelated with the WSe2 phonons emphasizes the complex nature of the transition that we excited using our driving laser, as discussed elsewhere[17] and mentioned in our introductory remarks. The negative correlation of the 22 cm–1 interlayer phonon with the A1g intralayer phonon of WS2 emphasizes the picture painted by the analysis of the TERS images above: the center of the nanobubble enhanced/dimmed WS2/interlayer phonon signatures. In the same vein, the cross-correlation slice taken at the WS2 resonance maximum (416 cm–1) reveals a negative correlation with the 22 cm–1 interlayer phonon as well as the ∼780 nm exciton of WSe2. Interestingly, the A1g mode of WS2 is strongly correlated to a broad peak centered at ∼810 nm that seems to arise from a localized indirect WS2 exciton.[33,34] Although a definite assignment of the underlying background to a localized WS2 exciton is difficult, the observed (i) correlation between the broad band and the WS2 intralayer phonon, (ii) disappearance of the interlayer phonon indicating that the two layers are decoupled toward the center of the protrusion, and (iii) proximity of the inferred exciton resonance to an indirect transition in WS2 bilayers all support our tentative assignment.

Figure 4

(A) Cross-correlation map (ρ = σ2/σ·σ) of the TERS spectral image of the nanobubble in Figure . (B) Cross-correlation slices (v ≠ v) taken at the interlayer phonon resonance (22 cm–1) and the intralayer A1g WS2 phonon resonance (416 cm–1).

In conclusion, this work describes a novel approach based on ultralow-frequency tip-enhanced Raman spectroscopy that is capable of tracking phonons on the nanoscale. In the model system used herein to describe our new capability, this approach may be used to directly track interlayer (de)coupling in a TMD bilayer. Further analysis of the recorded hyperspectral Raman nanoimages allowed us also to track the interplay between excitons and phonons (both intralayer and interlayer) on the nanoscale. In nanobubble structures, we found that local resonances that vary on the nanometer scale govern the recorded TERS images. We also observed the decoupling of the two layers, as marked by the disappearance of the ultralow-frequency interlayer phonon TERS signature. Beyond the scope of this work, our approach should be generally capable of correlating nanostructural motifs with electronic and optical properties in both model systems and state-of-the-art solid-state devices.

Methods

Samples were prepared using a previously described procedure.[25] Briefly, we used mechanical exfoliation of bulk crystals (HQ graphene). The WS2 and WSe2 monolayers were separately exfoliated on PDMS stamps and consecutively transferred onto 50 nm sputtered Au-coated SiO2/Si substrates using a dry-transfer technique. Figure S1 shows topographic AFM and photoluminescence spectra from the samples used in this study.

Nano-optical characterization was performed using a custom LabRam-Nano AFM-Raman system (HORIBA Scientific) equipped with a 785 nm laser and a 300 lines mm–1 grating. To resolve ultralow-frequency Raman signatures, three consecutive Bragg notch filters (OptiGrate) with total laser line suppression of OD 11 were used in the collection path. For TERS mapping, the laser power at the sample was kept at ∼350 μW, and the signals were time-integrated for 50 ms with a lateral step size of ∼2 nm. We used OMNI-TERS-SNC-Au probes from Applied Nanostructures Inc. for these measurements. To extract Raman peak intensities, linear background was subtracted from the relevant part of the spectra within the range of ±20 cm–1 from the expected peak frequency, and the peak was fitted with a Lorentzian line shape.