Compositional engineering of interfacial charge transfer in van der Waals heterostructures of graphene and transition metal dichalcogenides

In recent decades, two-dimensional (2D) materials, including graphene (Gr) and monolayer transition metal dichalcogenides (TMDs: MoS₂, MoSe₂, WS₂, and WSe₂) have captured the attention of researchers in the realms of optics,^1,2 electronics,³ and electrochemistry.^4,5 Artificial stacking of TMDs and Gr can form van der Waals heterostructures (vdWHs) without considering lattice-matching conditions. Gr-TMD vdWHs have the potential to combine excellent charge carrier mobility with high optical absorption, providing new avenues to pursue high-performance optoelectronic applications.^6,7

A general requirement for many optoelectronic applications is that photogenerated charge carriers must be separated at the interface and migrate toward selective electrodes to establish an electric potential or generate a current to make devices functional. Therefore, understanding the fundamentals of photo-induced charge transfer (CT) and recombination processes in Gr-TMD vdWHs is paramount for determining the emerging properties of these bilayer systems and their device performance. For example, Massicotte et al. have demonstrated that thermalized hot electrons in Gr with excess energy above the interfacial CT energy barrier can be injected into the WSe₂ layer after below-bandgap photoexcitation.⁸ This so-called photo-thermionic emission (PTE) mechanism has recently been validated in Gr-WS₂ vdWH using optical-pump terahertz-probe (OPTP) spectroscopy^9,10 and transient absorption spectroscopy.^11,12 On the other hand, in the same system, Chen et al. have proposed a modified PTE model, in which hot carriers can be efficiently extracted from Gr concurrently following individual electron and hole intraband heating, before the occurrence of interband electron–hole thermalization.¹¹ In addition, Yuan and colleagues have proposed direct excitations of CT states by promoting charge carriers from the valence band of Gr to WS₂.¹³ The controversy on the CT mechanism across Gr-TMD vdWHs calls for further investigations. A relatively unexplored landscape to test the different models is by tuning the CT energetics.¹⁴ 

Along with efficient charge separation, interfacial charge recombination through back charge transfer (BCT) is another important process that critically affects device performance. Interestingly, the reported BCT times also remain highly debated even in the same hybrid system, ranging from picoseconds (ps) to nanoseconds (ns): for instance, a short BCT time of ∼1 ps has been claimed in Gr-WS₂ vdWH based on the excited state dynamics of WS₂ revealed by transient absorption and time- and angle-resolved photoemission spectroscopies.^11,13,15 In contrast, employing OPTP measurements to monitor photoconductivity caused primarily by charge carrier gain/loss in graphene, we recently showed that BCT can be very long-lived, beyond 1 ns owing to the presence of mid-gap states that can capture and “store” charge carriers before the interfacial BCT recombination takes place.⁹ A recent follow-up work by us further unveiled the role of defect occupancy in tuning CT and BCT dynamics in Gr-WS₂ vdWH under electrochemical gating conditions. Compelling evidence shows that the involved defect states likely originated from sulfur vacancies.¹⁰ Further elucidating the BCT mechanism in different Gr-TMD hybrid systems is important, given that compositional engineering can substantially affect the defect nature and interfacial energetics, which are critical for tuning BCT rate and efficiency.

In this work, we aim to explore the CT and BCT mechanisms in different Gr-TMD vdWHs using OPTP spectroscopy. The variations in chemical composition and interfacial energetics allow us to gain comprehensive and universal rules describing CT and BCT processes. For the CT process, we provide consistent evidence that (i) PTE mechanism prevails in all hybrids and (ii) the CT direction following below-bandgap excitation is universal with Gr donating electrons to TMDs. On the other hand, the interfacial CT efficiency has been shown to critically depend on the chalcogen type (S or Se) in TMDs. The CT efficiency of S atom-based Gr-TMD vdWHs is 3–5 times larger than that of Se atom-based Gr-TMD vdWHs due to their lower injection barriers. For the BCT processes, our results suggest that the BCT time is transition-metal dependent and critically depends on the energetics of the mid-gap defects in TMDs: owing to their much deeper defect energetic position, the charge separation lifetimes in W transition metal-based Gr-TMD vdWHs (≫800 ps) are significantly longer than those in Mo-based Gr-TMD vdWHs (i.e., ∼24 ps for Gr-MoS₂ vdWH and ∼8 ps for Gr-MoSe₂ vdWH). These fundamental findings facilitate the optimization of device operation and performance through compositional and/or defect engineering.

All Gr-TMD vdWHs used in this work were obtained from SixCarbon Technology Company (Shenzhen, China), produced by using the chemical vapor deposition (CVD) method. We first performed UV–Vis absorption and Raman measurements to investigate the optical and vibrational properties of Gr-TMD vdWHs. As an example, the UV–Vis absorption and Raman spectra of Gr-MoSe₂ vdWH are shown in Figs. S1 and S2, respectively. Two significant exciton resonance peaks appear at 1.59 and 1.79 eV in the UV–Vis absorption spectrum, attributed to the A- and B-exciton transitions of monolayer MoSe₂, respectively. Moreover, owing to the absorption of monolayer graphene, a nearly constant absorption offset of ∼2% is noticed in the near-infrared range. For the Raman spectrum, we observe two distinguished Raman peaks located at 1590 and 2694 cm⁻¹, corresponding to the G band and 2D band of monolayer Gr, respectively. Considering the location of the G band and the p-doping nature of CVD-graphene on sapphire, we estimate the Fermi level in Gr to be ∼0.23 eV below the Dirac point in Gr-MoSe₂ vdWHs.

To investigate the CT process in Gr-TMD vdWHs, we employ OPTP spectroscopy to track the pump-induced photoconductivity dynamics of hybrids, as shown in Fig. 1. In this study, an optical pump pulse with a center wavelength of 1.55 eV (∼50 fs duration; below the optical bandgaps of all the TMDs studied) is used to selectively excite only the Gr layer; a single-cycle THz pulse is employed to measure the photo-induced conductivity changes (i.e., photoconductivity Δσ = σ_pump − σ₀, where σ_pump and σ₀ are the hybrid’s conductivity with and without excitations, respectively). The photogenerated free charge carriers interact with THz radiation, resulting in THz field attenuation (ΔE = E_pump − E₀). We then obtain the time-dependent photoconductivity dynamics by monitoring the ΔE evaluation following Δσ ∼ −ΔE/E₀.¹⁶ In Gr-TMD hybrids, as charge carriers in Gr are orders of magnitude more mobile than those in TMDs, the photoconductivity dynamics are, therefore, dominated by the carrier dynamics in Gr. For supported and doped graphene without TMD, one typically observes negative photoconductivity due to the reduction in carrier mobility upon photoexcitation. CT processes modulate the charge carrier density around the Fermi surface in Gr, further altering the conductivity dynamics.^17–21 

We have previously shown that PTE dominates the interfacial hot carrier transfer from Gr to WS₂.⁹ To test its applicability to other Gr-TMD systems, we conducted fluence-dependent photoconductivity measurements on Gr-WS₂, Gr-MoSe₂, and Gr-MoS₂ vdWHs, as shown in Figs. 2 and S3. In Fig. 2(a), taking the Gr-MoS₂ vdWH as an example, below-bandgap excitation leads to a short-lived negative photoconductivity decay within ∼3 ps, followed by switching to the positive side. The transient negative photoconductivity in Gr and Gr-based vdWHs has been previously shown by us^9,18,19 and others²¹ to stem from the hot carrier response in Gr: photoexcitation of doped Gr with high intrinsic carrier density generates hot carriers whose mobility is substantially lower than their cold counterparts due to enhanced scattering. This photo-induced transient reduction in charge mobility gives rise to the observed negative photoconductivity. On the other hand, the long-lived positive photoconductivity can be seen as a result of CT from Gr to MoS₂: for p-doped Gr, the “leakage” of hot electrons from Gr to TMDs causes the Fermi level to shift away even further from the Dirac point, thereby increasing its conductivity.

To elucidate the CT mechanism and how chemical composition tuning impacts CT efficiency, we summarize the maximum positive photoconductivity (averaged between 2 ps around the maximum value), as shown in Fig. 2(b). Two prominent features are present.

• The maximum positive photoconductivity shows a universal superlinear dependence on the pump fluence. We fit the data by A · N_abs^α (with A as the prefactor and α as the power index) and find that α of Gr-MoS₂, Gr-MoSe₂, and Gr-WS₂ vdWHs are 1.16 ± 0.07, 1.15 ± 0.02, and 1.13 ± 0.08, respectively. This result is fully consistent with the previously proposed PTE model⁸ and spectroscopic results^9,10,22 describing thermalized hot carrier transfer across Gr-based vdWHs interfaces: excitation of Gr results in the generation of nascent nonequilibrium carriers. Following efficient carrier–carrier (∼10 fs) scattering, these energetic carriers can be thermalized to form “hot carriers,” which are characterized by a well-defined, elevated electronic temperature, following the Fermi–Dirac distribution. The highly energetic hot carriers in the distribution tail can cross the Schottky barrier from Gr to TMDs. The superlinear trend originates from the nonlinear dependence of the hot carrier population above the Schottky barriers on the pump energy. Simulations can capture this behavior well (see the supplementary material in Ref. 9). Our results here thus confirm the universality of PTE effects for Gr-TMDs vdWHs.

• The CT efficiency shows a clear dependence on the chalcogenide type [as shown in Fig. 2(c)]: the CT efficiency of Gr-MoS₂ and Gr-WS₂ vdWHs is about 3–5 times higher than that of Gr-MoSe₂ and Gr-WSe₂ vdWHs. This result suggests a substantially reduced Schottky barrier for CT in S-atom-based Gr-TMD vdWHs than that of Se-atom-based Gr-TMD vdWHs. Although quantification of interfacial energetics remains challenging, we follow the standard practice in the field, that is, using the energetics of bare layers to discuss the relative energy offset [i.e., the energy difference between the Fermi level in Gr and the conduction band minimum (CBM) of TMDs in our case] for CT. For that, we have listed the energetics, i.e., the position of CBMs and valence band maximum (VBMs), of bare TMDs presented in Table I. In line with our expectations, the energy barriers for electron transfer (ΔE_e) in S-atom-based Gr-TMD vdWHs are lower than those in Se-atom-based Gr-TMD vdWHs. The smaller ΔE_e, in general, the easier and higher efficiency for CT from donor to acceptor under similar interfacial coupling strength. Finally, for the Gr-WS₂ system, the relative band alignment has been widely reported by Kelvin probe force microscopy²³ and spectroscopic studies.^24,25 By correlating the interfacial energetics and the measured CT signals in the Gr-WS₂ vdWH, we can back-extract the interfacial CT energetics for other vdWHs (in bracket and marked in ), as presented in Table I and supplementary material note. Compared to bare TMDs, the inferred energetics show a substantial shift up to 390 meV. This result indicates that while the common practice of using the energetics of bare TMDs for understanding the interfacial CT dynamics in hybrid systems is useful, extreme care needs to be taken when quantitative discussions are necessary.

We now focus on the photoconductivity dynamics on a time scale beyond 10 ps in Gr-TMD vdWHs, as shown in Fig. 3. Remarkably, we show that the BCT dynamics strongly correlate with the type of transition metal used: Gr-WS₂ and Gr-WSe₂ vdWHs exhibit long-lived positive photoconductivity dynamics with lifetimes beyond our probe window (≫800 ps). In contrast, the photoconductivity dynamics show a significant decay on a timescale of ∼10 ps for Gr-MoS₂ and Gr-MoSe₂ vdWHs, as presented in Table II.

Our previous work revealed that two unoccupied mid-gap defect states in WS₂, likely originating from S-vacancies, play a critical role in the BCT process in Gr-WS₂ vdWH.¹⁰ Combining THz and transient absorption studies, we showed that photoinjected electrons in TMDs are trapped and stored in these defect states, giving rise to photogating effects in Gr.⁹ Given this scenario, the key question is how defect energetics from chalcogenide vacancies are affected by compositions. We have summarized the defect energetics based on chalcogenide vacancies from the literature presented in Table III. Schuler et al. inferred two defect states located at ∼0.60 and ∼0.78 eV below the CBM of WS₂ using a scanning tunneling microscope (STM);²⁷ Zhao et al. revealed that there were two narrow mid-gap defect states located at ∼0.23 and ∼0.63 eV below the CBM of MoS₂;²⁸ and the defect states of MoSe₂ have been predicted by theoretical calculations.²⁹ The mid-gap defect density of the Gr-WS₂ vdWH was estimated to be around 10¹² cm⁻² by combining operando OPTP spectroscopy and simulation methods, originating from sulfur vacancies in WS₂.¹⁰ The estimated sulfur vacancy density is also in good agreement with that measured by imaging techniques, such as STM (10¹¹–10¹³ cm⁻²).^27,30 Furthermore, the defect densities of CVD-grown monolayer WSe₂ and MoSe₂ were both estimated to be about 10¹² cm⁻² by STM and optical pump–probe spectroscopy, respectively.^31,32 As presented in Table III, the mid-gap defect levels of W-atom-based TMDs are deeper than that of the Mo-atom-based TMDs because of their heavier d orbitals. This substantially reduces the energetics for BCT and thus prolongs the charge separation lifetime for W-based TMDs. Hence, the significant difference in BCT times can already be qualitatively explained alone using the defect energetics involved.

We have investigated the role of TMD composition on interfacial CT dynamics and the effect of mid-gap defect states level on the interfacial charge recombination process in Gr-TMDs vdWHs employing OPTP spectroscopy. The interfacial CT direction is universal and independent of the energetics of Gr-TMD vdWHs, with electron transfer from Gr to TMDs after photoexcitation. We further find that the interfacial CT efficiency in Gr-TMD vdWHs is determined by the chalcogen (S or Se) atoms in TMDs. The CT efficiency of S atom-based Gr-TMD vdWHs is 3–5 times higher than that of Se atom-based Gr-TMDs vdWHs. In contrast, the BCT process from TMDs to Gr is dominated by the mid-gap defect level of TMDs and closely related to the nature of transient metals in the TMDs. The lifetime of charge separation (≫800 ps) in W transition metal-based Gr-TMD vdWHs is significantly longer than that of Mo atom-based Gr-TMD vdWHs (∼24 ps for Gr-MoS₂ and ∼8 ps for Gr-MoSe₂). This can be attributed to deeper mid-gap defect states in W-based TMDs. Our results shed light on the role of energetics and defects on the interfacial CT and recombination dynamics in Gr-TMD vdWHs, which are pivotal for optimizing optoelectronic devices, particularly for photodetection.

All the Gr-TMD vdWHs on a sapphire substrate are bought from the SixCarbon Technology Company (Shenzhen, China) and produced by using the CVD method. The monolayer Gr is grown on polished copper foil substrates using methane as the carbon source. The copper foil is first annealed at 1000 °C for 1 h in an atmosphere of 10% hydrogen in argon to remove surface oxides. Subsequently, methane is introduced at a flow rate of 5 sccm, and the hydrogen is turned off. Graphene growth is done at 1000 °C for 10 min, followed by natural cooling to room temperature. TMDs are also grown by using CVD methods. Taking MoSe₂ as an example, it is synthesized using 99.999% pure MoO₃ and selenium pellets as sources in a dual-zone tube furnace with 90% argon and 10% hydrogen as carrier gases. MoO₃ is heated to 750 °C and selenium to 260 °C, with the target substrate placed 5 cm from the MoO₃ source. MoSe₂ is grown on sapphire substrates for 10 min. Finally, the monolayer MoSe₂ is transferred onto the monolayer Gr using a polymer stamp technique. The other heterostructures are achieved by using a similar method. The Fermi level in graphene is inferred from Raman spectroscopy.

We employ a commercial mode-locked Ti: sapphire laser (Spectra Physics Spitfire Ace) to operate our optical pump-THz probe spectrometer. Laser pulses with a central wavelength of 800 nm, 1 kHz repetition rate, and ∼50 fs duration are generated by using this laser system. THz pulses are generated by optical pumping to zinc telluride (ZnTe) crystal (1 mm thick) with a bandwidth of up to 2.5 THz. The THz signal is detected based on the electric-optical sampling effect in a second ZnTe crystal. The time-resolved THz photoconductivity dynamics are recorded by fixing the sampling beam to the peak of the THz field and varying the time delay between the pump and sampling beam with a motorized delay stage. All the measurements were performed in a dry N₂-purged environment at room temperature.

See the supplementary material for details on the simulation of net electron loss in Gr, the UV–Vis absorption, Raman, and OPTP characterization.

G.W. acknowledges the fellowship support from the China Scholarship Council (CSC). We acknowledge funding support from the Max Planck Society.

The authors declare no conflicts of interest.

Guanzhao Wen: Data curation (equal); Formal analysis (equal); Funding acquisition (equal); Visualization (equal); Writing – original draft (equal); Writing – review & editing (equal). Shuai Fu: Conceptualization (equal); Data curation (equal); Formal analysis (equal); Writing – review & editing (equal). Mischa Bonn: Data curation (supporting); Formal analysis (supporting); Funding acquisition (lead); Project administration (equal); Supervision (supporting); Writing – review & editing (supporting). Hai I. Wang: Conceptualization (lead); Data curation (lead); Formal analysis (lead); Funding acquisition (supporting); Project administration (equal); Supervision (lead); Visualization (lead); Writing – original draft (lead); Writing – review & editing (lead).

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