Enhanced interlayer electron transfer by surface treatments in mixed-dimensional van der Waals semiconductor heterostructures Open Access

Since the isolation of graphene,¹ a wide variety of two-dimensional (2D) van der Waals (vdW) materials have been studied over the past decades. A library of 2D vdW crystals has been expanded to include semiconducting transition metal dichalcogenides (TMDCs),^2–5 insulating hexagonal boron nitride (hBN),⁶ and even magnetic CrI₃⁷ and Cr₂Ge₂Te₆,⁸ as well as ferroelectric and piezoelectric SnTe,⁹ SnS,^10,11 WTe₂,¹² and α-In₂Se₃.^13–15 Importantly, such 2D vdW materials generally require a substrate to provide mechanical support, which at the same time significantly influences their properties. For example, the significant enhancement of the electronic properties of graphene encapsulated in high-quality hBN⁶ emphasizes the importance of the substrate material. This means that the interface between 2D materials and substrates, which are typically three-dimensional (3D) bulk, becomes crucial for harnessing the potential characteristics of 2D materials.

Such 2D/3D heterostructures are a promising platform for novel artificial material design. This is because dangling bond-free surfaces for 2D materials allow the stacking of constituent layers at desired sequences and angles, which cannot be realized by epitaxial growth of 2D materials on 3D substrates. In particular, vdW heterostructures based on 3D III–V semiconductors and 2D TMDCs are among the most promising platforms, taking advantage of high mobility and strong light absorption for III–V semiconductors and spin–valley coupling for TMDCs. Indeed, a variety of state-of-the-art devices have been realized, including helicity-controllable spin light-emitting diodes,¹⁶ highly efficient solar cells,¹⁷ and ultrasensitive photodetectors.^18,19

As has already been explored in 2D/2D heterostructures,^20–23 for these applications, charge and/or spin transfer between these 2D and 3D materials is critical to the efficiency of device operation. Interestingly, as will be shown in this paper, such interconnection across the interface is drastically modified only by controlling the surface of 3D materials. However, the effect of surface treatment itself on both the electrical and optical properties of the 2D/3D heterostructures has not been discussed intensively. Therefore, in this study, we shed light on the surface treatment techniques and comprehensively discuss their effects on the optoelectrical properties of the 2D material.

We focused on monolayer WS₂/III–V semiconductor heterostructures and demonstrated that reducing the native oxides of the III–V semiconductor enhances the electron transfer from the WS₂ to the III–V substrate. Furthermore, passivating the reduced III–V substrate with sulfur effectively reduces the spectral weight of localized excitons.

The III–V semiconductors used in this study were heavily n-doped In_0.04Ga_0.96As and semi-insulating GaAs structures. The reason for choosing GaAs and WS₂ is based on a small conduction band energy offset of less than 0.4 eV: the electron affinity is reported to be 4.07 eV²⁴ and 3.7–3.96 eV^20,25,26 for GaAs and monolayer WS₂, respectively. Given the small conduction band offset, carrier transfer from WS₂ to GaAs should be possible. The n-doped structure is composed of a 20 nm n⁺-In_0.04Ga_0.96As layer (n = 2 × 10¹⁹ cm⁻³), a 500 nm n-In_0.04Ga_0.96As layer (n = 6 × 10¹⁶ cm⁻³), and a 300 nm semi-insulating GaAs buffer layer, epitaxially grown on a (001) semi-insulating GaAs substrate. The bottom of the conduction band (CB) in the n⁺-In_0.04Ga_0.96As layer is below the Fermi level, while that in the n-In_0.04Ga_0.96As layer is above the Fermi level. As shown in Fig. 1(a), since the surface Fermi level is at 0.72 eV above the top of the valence band (VB),²⁷ the Fermi level difference between the surface and bulk of the n⁺-In_0.04Ga_0.96As layer induces carrier depletion near the surface and significant band bending. On the other hand, the semi-insulating structure consists of nominally undoped (001) GaAs. As shown in Fig. 1(b), the surface Fermi level of this structure is also at 0.72 eV above the top of the VB.²⁷ Hereafter, the n-type and semi-insulating structures are denoted as n-GaAs and i-GaAs, respectively.

As illustrated in Fig. 1(c), the n- and i-GaAs structures underwent three different surface treatments: Types A, B, and C. Type A samples were rinsed with acetone, ethanol, and deionized water, leaving their surfaces covered with amorphous native oxides. Type B samples underwent surface reduction through the self-cleaning effect of trimethylaluminum (TMA) vapors during atomic layer deposition (ALD) of Al₂O₃^28,29 after being rinsed with acetone, ethanol, and deionized water. The self-cleaning process consisted of 60 cycles of TMA pulse–purge sequences, which effectively removed a significant portion of the native oxides.^28,29 Subsequently, a buffered oxide etchant (BOE) was used to etch the Al₂O₃ layers, leaving the III–V substrates unetched. Type C samples underwent sulfur passivation by immersion in a CH₃CSNH₂/NH₄OH solution,^30,31 following degreasing with acetone, ethanol, and deionized water, as well as the above-mentioned surface reduction process.

The sulfur passivation used for the type C samples provides the following two benefits: The first is that it prevents re-oxidation of the GaAs surface. When the dangling bonds existing even after the reduction of the GaAs surface are exposed to the air for a long time, amorphous oxides are once again formed on the GaAs surface, which makes it challenging to integrate 2D/3D structures. To circumvent this issue, sulfur passivation is employed to saturate dangling bonds in the type C sample. For III–V semiconductors such as GaAs and InGaAs, it has been established that chalcogen atoms, such as sulfur, are highly effective in terminating dangling bonds on the surface,³² thereby preventing re-oxidation.³³ The second benefit of the sulfur passivation is that it enhances the affinity between WS₂ and GaAs. Since the surface of monolayer WS₂ exhibits a 2D network of sulfur atoms, through the sulfur passivation on the GaAs surface, two adjacent atoms between the WS₂ and GaAs become sulfur, which could pretend to have a similar interface to multilayer WS₂. It is, therefore, anticipated that the sulfur passivation would result in a strong affinity between the WS₂ and GaAs.

Highly crystalline WS₂ monolayers synthesized using a NaCl-assisted chemical vapor deposition (CVD) method were then transferred onto the n-GaAs and i-GaAs structures using a polycarbonate-based clean transfer method. The samples were then encapsulated with an Al₂O₃ layer deposited by ALD to prevent the physisorption of H₂O and O₂ molecules on the WS₂ surface, which could otherwise destabilize the carrier density in WS₂.³⁴ 

As a control substrate, we used a silicon substrate consisting of thermally oxidized 300 nm SiO₂/Si, where the 300-nm-thick SiO₂ layer prevents interlayer charge transfer between WS₂ and Si.

We first confirm that the self-cleaning effect occurs during the type B and C surface treatments by comparing the samples processed with and without the multiple TMA purge sequences before Al₂O₃ deposition. Figures 2(a) and 2(b) show the transmission electron microscopy (TEM) images of the Al₂O₃/n-GaAs interfaces with one cycle and 60 cycles of TMA pulse–purge sequences, respectively. Here, while a brighter region corresponds to the Al₂O₃ layer, a darker region corresponds to the n-GaAs structure. Since the one pulse–purge sequence reduces only a small fraction of the native oxides,²⁹ the interface shown in Fig. 2(a) is blurrier than that in Fig. 2(b).

For quantitative analysis of the interfacial roughness, we used TEM-coupled energy dispersive x-ray (EDX) spectroscopy to obtain atomic composition profiles for the regions shown in Figs. 2(a) and 2(b). The EDX line profiles are presented in Figs. 2(c) and 2(d), respectively. The sums of the EDX signals from the Ga and As L shells are represented by the black circles. The interfacial roughness is extracted using fits to a step-like function $a/2erf[(x−μ)/(2σ)]+b/2$⁠, shown by the red curves. Here, $erf(z)$ is the error function with $z$ as a variable, $a$ is the step height, $b$ is the background signal, $x$ is the position, $μ$ is the center position, and $σ$ is the interfacial roughness. The interfacial roughness $σ$ was improved from 2.2 to 1.1 nm after 60 cycles of TMA pulse–purge sequences, as shown in Fig. 2(d). This decrease in the interfacial roughness indicates the effective removal of the amorphous native oxides, which is consistent with the self-cleaning effect.^28,29

The difference due to the surface treatments can also be seen in the Raman spectra measured for the n-GaAs structure. Figure 3(a) shows the Raman spectra for the n-GaAs structure after transferring the WS₂ monolayer at room temperature in the backscattering configuration under 532 nm excitation. The laser penetration depth was estimated to be 124 nm, and a phonon signal near the n-GaAs surfaces can be observed. Two distinct peaks at 288 and 268 cm⁻¹ are attributed to the longitudinal optical (LO) phonon mode and the lower branch of the LO-phonon–plasmon-coupled (L⁻) mode, respectively.^35,36 The L⁻ mode originates from an un-depleted region and becomes prominent at high carrier densities, where the coupling between LO phonons and plasmons becomes important.^35–37 Based on its position, the L⁻ mode originates from the un-depleted region in the n⁺-In_0.04Ga_0.96As layer, not in the n-In_0.04Ga_0.96As layer. This suggests that charge depletion occurs in the heavily doped n⁺-In_0.04Ga_0.96As layer, allowing us to analyze the depletion region width using the intensity ratio between the LO and L⁻ modes. It is important to note that Raman scattering from transverse optical phonons, which could spectrally overlap with the LO mode,³⁸ is forbidden due to the (001) crystallographic orientation and the backscattering configuration.³⁹ This selection rule allows for a safe comparison of the intensity ratio $r=IL−/ILO$⁠, where $IL−$ and $ILO$ represent the integral intensities of the L⁻ and LO modes, respectively. The integral intensities were determined using a Voigt fit for each peak. The values of $r$ decrease from 0.51 for type A to 0.28 and 0.26 for type B and C, respectively. This indicates that the self-cleaning and sulfur passivation processes have reduced the thickness of the depletion region. The reduction is attributed to the decrease in the number of surface states. When the density of surface states increases, more free electrons are trapped on the surface.

Figure 3(b) shows the PL spectra for i-GaAs at room temperature under 2.21 eV excitation with a fixed excitation power. The PL centered at 1.42 eV is attributed to radiative recombination from the CB to the VB.⁴⁰ Importantly, the peak intensity of type B showed a 1.25-fold increase compared to that of type A. After the additional sulfur passivation process, type C showed a substantial 13.3-fold increase in PL intensity compared to type A. These facts suggest that non-radiative recombination pathways were reduced due to a decrease in the number of surface states. These results are consistent with those obtained by Raman spectroscopy [Fig. 3(a)]. It is worth noting that, unlike the top layers of the n-GaAs structure, the i-GaAs structure is nominally undoped, and thus, the number of free carriers is insufficient to observe the L⁻ mode. In the case of the n-GaAs structure, the aforementioned sharp band bending repels photoexcited electrons into the bulk region. Thus, PL intensity is insensitive to the number of surface states.

Polarization-resolved PL measurements were performed to assign the excitonic species in WS₂ of the WS₂/silicon, WS₂/n-GaAs, and WS₂/i-GaAs samples. Both circularly and linearly polarized PL were measured under 2.21 eV excitation at a cryogenic temperature, T, of 4 K. The excitation energy was near-resonant to the exciton PL energy. In the circularly polarized PL measurements, right-handed (σ₋) and left-handed (σ₊) circularly polarized PL were measured upon photoexcitation of the K valley by σ₋ polarized light, following the spin–valley optical selection rule.^41–46 In the linearly polarized PL measurements, on the other hand, horizontal (H) and vertical (V) linearly polarized PL were measured upon photoexcitation of both K and K′ valleys by H polarized light. To ensure reliable analysis of neutral and charged excitons, a combination of circularly and linearly polarized PL measurements was used instead of peak deconvolution. This approach excludes the effect of peak shifts that may be caused by changes in dielectric screening due to surface treatment of the III–V substrate.^47,48

Figure 4(a) shows representative circularly polarized PL spectra for the WS₂/silicon samples (black) and the WS₂/n-GaAs samples with type A, B, and C treatments (purple, orange, and green). In each pair of spectra, the solid and dashed curves represent σ₋ and σ₊ polarized PL spectra, respectively. The degree of circular polarization $ρc=(I−−I+)/(I−+I+)$ for each sample is summarized in Table I, where I_∓ is the intensity of σ_∓ polarized PL averaged within ±20 meV around the peak energy. The large $ρc$ independent of the substrate conditions means all samples realized excitonic valley polarization.^41–46 In other words, depolarization pathways through the electron–hole (e–h) Coulomb exchange interaction^46,49 and simultaneous intervalley scattering of electrons and holes by two longitudinal acoustic (LA) phonons^50–52 were effectively suppressed. This is indeed plausible because the broken inversion symmetry $E↑(k)≠E↓(−k)$⁠, where $k$ represents the crystal momentum, should still persist. This is in contrast to the case of bilayer TMDCs.⁴² 

Figure 4(b) shows linearly polarized PL spectra for the WS₂/silicon (black) and the WS₂/n-GaAs samples with type A, B, and C treatments (purple, orange, and green) measured at the same spot as in Fig. 4(a). In each pair of spectra, the solid and dashed curves represent H and V polarized PL spectra, respectively. The degree of linear polarization $ρl=(IH−IV)/(IH+IV)$ for each sample is summarized in Table I, where $IH(V)$ is the intensity of H(V) polarized PL averaged within ±20 meV around the peak energy. By analyzing changes in $ρl$⁠, we assign the dominant excitonic species that contribute to the observed PL lines as follows:

Previous studies have shown that in low carrier densities, the PL spectrum of WS₂ is dominated by a Coulomb-bound electron–hole pair, known as a neutral exciton (X⁰).^44,45 At $T=4 K$⁠, where our measurements were performed, PL emission from X⁰ under linearly polarized excitation is considered to show $ρl≠0$ as follows: A linear superposition of K and K′ valleys created by linearly polarized light emits the same linearly polarized light if the recombination lifetime is shorter than the valley decoherence time. In our case, this situation is plausible since the valley depolarization was effectively suppressed until recombination, as confirmed by $ρc≠0$ under the circularly polarized excitation [Fig. 4(a)]. In higher carrier densities, on the other hand, X⁰ captures the excess electron or hole and forms negatively or positively charged excitons (X⁻ or X⁺), respectively. The PL from both X⁻ and X⁺ is not linearly polarized (i.e., $ρl=0$⁠), even if valley depolarization is suppressed. This is because the optical transition allows for circularly polarized light rather than linearly polarized light due to the eigenstate mixing of excitonic states induced by the exchange interaction.^44,46 Therefore, a larger $ρl$ indicates that X⁰ is dominant.

For the WS₂/silicon samples, X⁻ or X⁺ dominate the spectral weight since $ρl$ = 0% ± 1% was observed. The small $ρc$ values agree well with the n-type behavior of CVD-grown WS₂ monolayers due to the energetically preferred formation of sulfur vacancies.⁵³ Therefore, we assign X⁻ as the dominant excitonic species in the WS₂/silicon samples.

For the WS₂/type A n-GaAs samples, a relatively small $ρl$ = 4% ± 3% suggests that either X⁻ or X⁺ dominates the spectral weight. In addition to the fact that the CVD-grown WS₂ monolayers behave n-type, as mentioned above,⁵³ the energy gap of native oxides of ∼4 eV⁵⁴ is large enough to disturb the electron transfer from the WS₂ to the substrate even after the heterostructure fabrication. Furthermore, the heavily n-doped substrate cannot provide sufficient holes for WS₂. Therefore, X⁻ still dominates the spectrum from the WS₂/type A n-GaAs samples, similar to the WS₂/silicon samples.

On the other hand, the WS₂/type B n-GaAs and WS₂/type C n-GaAs samples exhibited relatively large $ρl$⁠, 16% ± 3% and 18% ± 1%, respectively. This indicates X⁰ emission increased, suggesting a reduction in electron density compared to the cases of the WS₂/silicon and WS₂/type A n-GaAs samples. Therefore, the self-cleaning effect due to the TMA pulse–purge sequences successfully removed the native oxides on the substrate surface and, thus, lowered the energy barrier between WS₂ and the substrate.

It should be noted that the above consideration is consistently confirmed from the A_1g Raman mode and the PL peak shifts (see 7 and 8 of the supplementary material). For the A_1g Raman mode, the type B and C samples show a blue-shift accompanied by a line narrowing compared to the case of the type A samples. The blue-shift and narrowing of the A_1g peak reflect the reduction in the electron density,^55,56 consistent with the above identification of the excitonic species: X⁻ for the type A samples and X⁰ for the type B and C samples. In addition, the PL peak energy in the type A samples was ∼10 meV smaller than that in the type B and C samples. The sign and magnitude of the PL peak shift do not contradict the previous report,⁴⁴ where X⁻ PL shows up a few tens meV lower than X⁰ PL. The blue-shift of the A_1g mode and the PL peak shift also support the above identification of the excitonic species.

The impact of sulfur passivation can be seen in a PL shoulder at around 1.95 eV and is present in the type B samples but absent in the type A and C samples [Figs. 4(a) and 4(b)]. This shoulder was reproduced in all type B samples measured, not just the type B sample shown in Fig. 4. As has been reported in MoS₂/GaAs heterostructures,⁵⁷ the PL shoulder is considered to originate from exciton localization or interfacial charge trap. The presence of the PL shoulder in the WS₂/type B n-GaAs samples and the absence of such a shoulder in the WS₂/type C n-GaAs samples suggest that sulfur passivation plays a significant role in reducing exciton localization or interfacial charge trap in monolayer WS₂. This is consistent with the broadest LO peak for type B n-GaAs in Fig. 3(a) since the dangling bonds can be the origin of surface states, which are in principle disordered and lead to the broadening of the Raman peak. Consequently, the use of III–V substrates with well-controlled surface states plays a significant role in both the localization and delocalization of electrons in monolayer WS₂/III–V semiconductor heterostructures.

The underlying mechanism behind the change in the excitonic species is discussed below. The topmost n⁺-In_0.04Ga_0.96As layer of the n-GaAs structure is n-doped and has a thin depletion region. Modifications in both surface states and depletion regions may be dominant factors. To investigate these mechanisms, the semi-insulating i-GaAs structures were used to exclude the effect of the depletion regions.

Figure 5(a) shows representative circularly polarized PL spectra for the WS₂/silicon samples (black) and the WS₂/i-GaAs samples with type A and C treatments (purple and green). All samples show a clear difference between the σ₋ and σ₊ components, indicating effective suppression of the e–h Coulomb exchange^46,49 and LA phonon^50–52 depolarization pathways. Similar to the n-GaAs cases, this fact allows us to identify the excitonic species through the degree of linear polarization.

Figure 5(b) shows the linearly polarized PL spectra measured at the same spot as in Fig. 5(a), where H (V) polarized PL are indicated as solid (dashed) curves. As summarized in Table II, $ρl$ is larger in the WS₂/type C i-GaAs samples (green) compared to the WS₂/silicon samples (black) and the WS₂/type A i-GaAs samples (purple). This suggests a lower electron density in the WS₂/type C i-GaAs samples compared to the WS₂/silicon and WS₂/type A i-GaAs samples, similar to the n-GaAs case. It should be noted that the batches used to synthesize WS₂ were different in Figs. 4 and 5. This must lead to the batch-to-batch difference in the quality of the WS₂ monolayers, including defect and carrier density, resulting in the difference in PL polarization degrees, for example, as seen in the WS₂/silicon samples. However, a quantitative comparison within the same figure should make sense.

Regardless of the substrate doping concentrations, the type A samples exhibit lower values of $ρl$⁠, i.e., a dominance of X⁻ over X⁰. In contrast, the type B and C samples exhibit an increased spectral weight of X⁰. Namely, the reduction in the electron density in WS₂ is observed regardless of the presence or absence of depletion regions near the surface of the III–V semiconductors. This suggests that the space charges in the depletion regions do not play a dominant role in reducing the electron density in WS₂. In addition to that, the WS₂/type B n-GaAs and WS₂/type C n-GaAs samples exhibit similar $ρl$ regardless of the reduction of surface states. This also suggests that the space charges in the surface state do not play a significant role. Because the space charges in the depletion regions and at the surface do not play a role in changing the carrier density in monolayer WS₂, we conclude that the electron transfer from the CB of monolayer WS₂ to that of III–V semiconductor plays a role in reducing the carrier density in the WS₂/type B n-GaAs and WS₂/type C n-GaAs samples.

We focused on WS₂/n-GaAs and WS₂/i-GaAs heterostructures and investigated the effect of surface treatments for the III–V substrates on the excitonic species in the WS₂ monolayer. The PL emission from the WS₂ monolayers was dominated by the X⁻ feature in the WS₂/untreated n-GaAs and WS₂/untreated i-GaAs samples, regardless of the substrate doping concentrations. However, in the case of the WS₂/surface-reduced n-GaAs, WS₂/sulfur-passivated n-GaAs, and WS₂/sulfur-passivated i-GaAs samples, the X⁰ feature dominated the PL spectra, indicating a reduced electron density in the WS₂ monolayers. The removal of the native oxides may enhance the electron transfer from the WS₂ monolayer to the III–V substrates. This means that the relatively low spin injection efficiency, for example, from the (Ga,Mn)As substrate to the WS₂ monolayer can be improved by removing the native oxides of the (Ga,Mn)As substrate.¹⁶ Furthermore, for the WS₂/surface-reduced n-GaAs samples, an additional PL shoulder appears at ∼50 meV lower than the X⁰ feature, suggesting exciton localization or an interfacial charge trap. However, such a PL shoulder was absent in the WS₂/sulfur-passivated n-GaAs samples. Therefore, the surface passivation of the III–V substrate may play an important role in suppressing exciton localization or an interfacial charge trap in the WS₂ monolayer, which can lead to high mobility devices.⁶ These findings could lead to novel device concepts, including gate-free doping and deterministic carrier localization or delocalization for 2D materials.

The supplementary material contains detailed procedures for the surface treatments, the WS2 growth and transfer, the PL measurements, and the analysis of the A1g Raman mode and PL peak shifts in WS2.

T.O. and K.K. acknowledge the financial support from the Graduate Program in Spintronics (GP-Spin) at Tohoku University. T.O. was financially supported by the JST-SPRING program (Grant No. JPMJSP2114). T.O. acknowledges the technological support from Dr. Kosei Kobayashi at the Graduate School of Engineering, Tohoku University. This work was supported by the JSPS KAKENHI program (Grant No. 21H04647), the JST-FOREST and CREST programs (Grant Nos. JPMJFR203C and JPMJCR22C2), and the Asahi Glass Foundation.

The authors have no conflicts to disclose.

Takeshi Odagawa: Data curation (lead); Formal analysis (lead); Investigation (lead); Methodology (equal); Resources (lead); Validation (lead); Visualization (lead); Writing – original draft (lead); Writing – review & editing (equal). Sota Yamamoto: Formal analysis (equal); Validation (equal); Writing – review & editing (equal). Chaoliang Zhang: Formal analysis (equal); Methodology (equal); Resources (equal); Validation (equal); Writing – review & editing (equal). Kazuki Koyama: Resources (equal); Validation (equal); Writing – review & editing (equal). Jun Ishihara: Formal analysis (equal); Validation (equal); Writing – review & editing (equal). Giacomo Mariani: Validation (equal); Writing – review & editing (equal). Yoji Kunihashi: Validation (equal); Writing – review & editing (equal). Haruki Sanada: Validation (equal); Writing – review & editing (equal). Junsaku Nitta: Conceptualization (equal); Project administration (equal); Writing – review & editing (equal). Makoto Kohda: Conceptualization (lead); Methodology (lead); Project administration (lead); Writing – review & editing (lead).

The data that support the findings of this study are available from the corresponding author upon reasonable request.