Raman Enhancement of Molecules Sandwiched between van der Waals Materials Authors: Xielin Wang, Nguyen Tuan Hung, Shengxi Huang Published: 2026 ABSTRACT Raman enhancement has been widely observed for molecules adsorbed on van der Waals (vdW) materials including graphene, MoS2, hexagonal boron nitride, and transition-metal dichalcogenides. This enhancement is often dominated by a chemical mechanism. However, how the Raman response evolves when a molecule is sandwiched between two vdW layers remains unclear. Here, using copper phthalocyanine (CuPc) as a probe, we compare three sandwich structures: graphene/CuPc/TiS2, graphene/CuPc/TiSe2, and graphene/CuPc/graphene, spanning asymmetric to near-symmetric dual-interface geometries. We find that adding a second interface in contact with the molecule can suppress, moderate, or enhance the Raman response of CuPc depending on the contacting material pair and vibrational symmetry. The graphene/CuPc/graphene sandwich shows stronger enhancement than single-interface CuPc/graphene, with an excitation wavelength dependence. These results show that the dual-interface Raman response is influenced by interface-dependent charge-transfer effects and increased nonradiative damping. This work provides an experimental framework for understanding Raman enhancement in molecule-sandwiched structures. Raman spectroscopy probes molecular vibrations through inelastic light scattering and provides a chemical fingerprint for molecular identification. − Its sensitivity is often limited by the intrinsically weak Raman cross sections, , motivating the development of surface-enhanced Raman spectroscopy (SERS). While conventional noble-metal SERS is dominated by the electromagnetic mechanism (EM), − Raman enhancement on two-dimensional (2D) materials relies primarily on the chemical mechanism (CM) with strong molecular and vibrational-mode selectivity. − Since the discovery of graphene-enhanced Raman scattering, other van der Waals (vdW) materials such as hexagonal boron nitride (h-BN) and transition-metal dichalcogenides (TMDCs) such as MoS2, WSe2, and MoTe2, have also been reported to exhibit excellent SERS effects. − vdW materials possess atomic-level smooth surfaces and great chemical stability and exhibit strong chemical interactions with molecules. Comparative studies using copper phthalocyanine (CuPc) on different 2D materials have shown that the Raman enhancement depends strongly on the contacting material and is dominated by the charge-transfer CM. Hybrid 2D materials/plasmonic platforms have combined chemical enhancement with electromagnetic enhancement to achieve single-molecule detection. − Related studies on metallic 2D materials have suggested that differences in enhancement among 2D substrates are linked to their electronic structure, especially the density of states near the Fermi level, which can increase the probability of charge-transfer-mediated Raman enhancement. One advantage of 2D materials is that they can be deliberately engineered into layered structures, enabling controlled variation of molecule–2D contact geometries. Prior studies on 2D heterostructures, where a neighboring 2D layer modifies the electronic environment of the first molecule–substrate contact without directly contacting the molecule, can modify the molecule–2D interaction and have achieved enhanced SERS performance. , Tan et al. showed that a WSe2/graphene heterostructure enhances the Raman signal of CuPc relative to either component alone. Later work on nanostructured TMDC/graphene heterostructures reported much larger sensitivity gains. Chen et al. further showed that the Raman enhancement of graphene/MoS2 heterostructures depends strongly on the interlayer distance. However, these cases only involve one direct molecule–2D interface with indirect modulation by a neighboring layer and are often mixed with the EM. Zhang et al. studied a molecule–2D–molecule bifacial geometry using CuPc and showed that the second molecular layer can be fully or partially added to the total Raman response, depending on its electronic character. For graphene, whose high carrier density screens the two faces from one another, the two molecular layers produced an approximately additive 2× Raman intensity, whereas h-BN showed subadditive and mode-dependent enhancement due to partial cancellation of interfacial dipoles. Another potential way to modify Raman enhancement is to create a 2D–molecule–2D sandwich structure, in which the molecule is simultaneously contacted by two vdW layers. Wei et al. predicted from first-principles calculations that certain vdW sandwich structures can produce larger Raman tensors for adsorbed molecules. However, the computations only considered simple molecules and have not yet been experimentally verified. A systematic experimental Raman study that isolates the effect of a second direct molecule–2D interface is still lacking. Filling this gap is important for understanding how chemical enhancement combines across multiple interfaces. In this work, using CuPc as a probe molecule, we study three vdW sandwich structures that vary the electronic character of the two interfaces: graphene/CuPc/TiS2 as an asymmetric sandwich in which the bottom TiS2 interface provides a much weaker charge-transfer enhancement than graphene, graphene/CuPc/TiSe2 as an asymmetric sandwich in which the bottom interface TiSe2 provides stronger charge-transfer enhancement of CuPc than TiS2 but still smaller than graphene, and graphene/CuPc/graphene in which both interfaces are chemically identical, approaching a near-symmetric dual-interface geometry. By comparing sandwich regions with single-interface reference within the same devices, we find that a second direct molecule–2D interface can reshape the Raman response in different ways, as a function of the interfacial identity, vibrational symmetry, and excitation wavelength. Raman response of the graphene/CuPc/TiS2 sandwich structure is suppressed and exhibits a mode-symmetry dependence. The graphene/CuPc/TiSe2 sandwich exhibits an intermediate response between the TiSe2 and graphene single-interface limits for all modes. In graphene/CuPc/graphene, the sandwich response exceeds that of single-interface graphene and shows a clear excitation-wavelength dependence. These results provide a systematic experimental picture of how a second direct molecule–2D interface modifies Raman enhancement and establish molecule-sandwiched vdW structures as a platform for understanding and engineering multi-interface chemical enhancement. To understand how CuPc Raman scattering changes when the molecule forms one or two direct interfaces with vdW materials, we measure Raman spectra of systems that contain both single- and dual-interface regions. Three representative systems, including graphene/CuPc/TiS2, graphene/CuPc/TiSe2, and graphene/CuPc/graphene, are fabricated using the method shown in Figure a. In all cases, the vdW material flakes were mechanically exfoliated onto Au-coated SiO2/Si substrates, CuPc was deposited by thermal evaporation, and graphene was transferred where required to form the target layered structure. Thermal evaporation was used to ensure uniform CuPc coverage across different regions of the same sample, as described in Section S1. Consider the graphene/CuPc/TiS2 sandwich structure as an example, in which CuPc is in direct contact with both graphene (on the top) and the underlying TiS2 (on the bottom), as shown in Figure b. The measured Raman response reflects contributions from two distinct molecule–2D charge-transfer pathways simultaneously. Because the transferred graphene covers only part of the CuPc-coated bottom flake, the sample contains both a dual-interface region (graphene/CuPc/TiS2) and single-interface regions (CuPc on TiS2 and graphene on CuPc), enabling internally referenced comparisons under identical molecular deposition and processing conditions. Dual-interface SERS in vdW sandwiches. (a) Sample fabrication workflow. (b) Sandwich structure geometry and schematic of the Raman process. (c) Raman spectra of CuPc on TiS2, graphene/CuPc/TiS2, and graphene/CuPc with 633 nm excitation. Solid and dashed vertical lines denote the selected TS and NTS modes, respectively. Modes marked with a triangle and an asterisk are used to compute the intensity ratio R and the degree of charge transfer ρCT. The mode marked with a circle corresponds to the 1T-TiS2 phonon mode. Figure c shows the measured Raman spectra of graphene/CuPc/TiS2, CuPc on TiS2, and graphene on CuPc with 633 nm excitation. CuPc vibrational modes appear across the 500–1600 cm–1 range. The vertical solid and dashed lines mark selected CuPc totally symmetric (TS) and nontotally symmetric (NTS) modes, respectively, assigned following work by Basova et al. These modes are quantified in Figure . Full mode assignments are summarized in Table S1, and additional mode-resolved data for CuPc on SiO2/Si and for the graphene/CuPc/TiS2 sandwich are shown in Figures S1 and S2. The asterisk and triangle indicate the TS mode at 680 cm–1 and the NTS mode at 1528 cm–1, which are used as the reference pair to calculate the degree of charge transfer. The peak with a shoulder near 330 cm–1, marked by a circle, corresponds to the 1T-TiS2 phonon mode, which appears in both the TiS2 reference and the sandwich spectra. These spectra define clear single-interface limits: graphene/CuPc shows substantially larger enhancement of all modes than CuPc/TiS2 due to stronger charge transfer between graphene and CuPc. In particular, the NTS modes at 1448 and 1528 cm–1 show much larger enhancement on graphene/CuPc than CuPc/TiS2. The most enhanced NTS mode at 1528 cm–1 reaches an enhancement factor of 42.5 on graphene/CuPc compared with 1.1 on CuPc/TiS2. The mapping of the CuPc 1528 cm–1 signal across the graphene/CuPc/TiS2 region is shown in Figure S3. For the graphene/CuPc/TiS2 sandwich spectrum, the TS modes at 830 and 1106 cm–1 fall below both single-interface references (solid vertical lines), whereas the NTS modes at 952, 1448, and 1528 cm–1 remain between them (dashed vertical lines). This symmetry-dependent separation shows that the sandwich does not behave like a simple average of the two single-interface responses. In contact with two interfaces, CuPc participates in charge-transfer interactions with both the graphene and TiS2 simultaneously, and these two pathways reshape the Raman polarizability in a symmetry-dependent manner. Mode-resolved SERS in graphene/CuPc/TiS2 sandwiches. (a) Optical image and (b) graphene G′-band peak-intensity map of the graphene/CuPc/TiS2 stack, with a red scale bar of 10 μm. Yellow and black dashed lines outline the TiS2 and graphene flakes, respectively. (c) Within-spectrum intensity ratio R and the corresponding degree of charge transfer ρCT. Enhancement factors of selected CuPc (d) NTS modes and (e) TS modes. The optical image (Figure a) and graphene G′-band intensity map (Figure b) identify the graphene/TiS2 overlap region and define the areas used for extracting spectra and enhancement factors. Yellow and black dashed lines outline the TiS2 and graphene flakes, respectively. Enhancement factors were calculated by normalizing the CuPc peak intensity in each region to the corresponding CuPc intensity on SiO2/Si measured under the same acquisition conditions, as described in the SI Method. Error bars represent the standard deviation of values extracted from spectra within each mapped region. This symmetry-selective behavior is consistent with prior CuPc Raman and SERS studies, in which NTS modes become more prominent with charge-transfer-enabled vibronic channels. In the Herzberg–Teller vibronic-coupling description commonly used for 2D-material chemical enhancement, the Raman polarizability contains a Franck–Condon background (A term) and charge-transfer-enabled vibronic contributions (B and C terms). , The A term contributes only to TS vibrations: for CuPc (D 4h ), it is the A1g modes, whereas NTS modes such as B1g, B2g, and Eg are forbidden in this channel. [We note that all CuPc modes analyzed in this work correspond to in-plane vibrations (A1g, B1g, and B2g), whereas the Raman-active out-of-plane Eg modes are intrinsically weak and are not resolved with a sufficient signal-to-noise ratio. The second interface could affect in-plane and out-of-plane modes differently with vdW forces and out-of-plane electrostatic asymmetry, which is an interesting direction for future studies.] Consequently, TS modes contain A, B, and C terms, while NTS modes only contain B and C terms. As a result, TS and NTS modes can respond differently to interfacial charge transfer. To track the charge-transfer (CT) contribution across the three structures, we calculate the within-spectrum intensity ratio R and the corresponding degree of charge transfer ρCT, which are defined by , R(λ)=Ik(λ)I0(λ),⁣ρCT(λ)=R(λ)1+R(λ) 1 where I k is the intensity of the selected mode of interest and I 0 is the intensity of the chosen TS reference mode at excitation wavelength λ. Here, I k and I 0 follow the mode-index notation of the degree-of-charge-transfer analysis. In this work, I k is taken from the CT-sensitive NTS B1g mode at 1528 cm–1 and I 0 from the TS A1g mode at 680 cm–1, following prior CuPc-on-2D analyses. , We use ρCT here as a comparative CT-sensitive metric; see Section S4. The extracted ρCT values (Figure c) are 0.95 for graphene/CuPc, 0.93 for graphene/CuPc/TiS2, and 0.52 for CuPc/TiS2. Thus, the sandwich has a stronger CT contribution than CuPc/TiS2 but a weaker one than the graphene/CuPc single-interface limit. The enhancement factors in Figure d,e reveal a clear symmetry dependence. The NTS modes at 952, 1448, and 1528 cm–1 in Figure d all remain between the two single-interface limits, consistent with the ordering of ρCT. Because the NTS intensity is set entirely by the B- and C-term CT contributionswith no A-term backgroundthese modes more directly track the net CT amplitude at the interface, which in the sandwich lies between the two single-interface values, and so does the enhancement. The enhancement factor of the CT-sensitive 1528 cm–1 NTS mode increases from 1.1 on TiS2 to 19.2 in the sandwich while remaining below the graphene/CuPc value of 42.5. By contrast, for the TS modes at 830 and 1106 cm–1 in Figure e, the enhancement factor in the sandwich case is smaller than that of both single-interface cases. For example, the 830 cm–1 mode has an enhancement factor of 2.3 in the sandwich, smaller than both graphene/CuPc (4.0) and CuPc/TiS2 (3.0). A similar suppression has been reported for the CuPc/h-BN/CuPc system, where partial cancellation of oppositely oriented interfacial dipoles reduces the modes that are most strongly enhanced on a single face. The CT contributions from the two interfaces can enter the polarizability with mismatched phases, partially canceling the CT-enabled component, and the second contact introduces additional nonradiative damping that broadens the intermediate-state linewidth. Both effects reduce the net TS polarizability, which carries an A-term background alongside the CT-enabled terms. Additional mode-resolved spectra, 532 nm excitation comparisons, extracted peak intensities, and enhancement factors for graphene/CuPc/TiS2 are provided in Figure S2 and Tables S2 and S3. To further understand the charge-transfer mechanism, we also study a graphene/TiS2 heterostructure in which CuPc sits on graphene and TiS2 lies underneath the graphene. In this structure, CuPc directly contacts graphene, while TiS2 modifies the electronic environment of graphene without directly contacting the molecule. The CuPc/graphene/TiS2 region shows intermediate enhancement between graphene/CuPc and CuPc/TiS2. The graphene G′ peak shifts in the regions contacting TiS2, showing that the adjacent TiS2 layer changes the graphene Fermi level. The reduced Raman enhancement in CuPc/graphene/TiS2 compared with CuPc/graphene is therefore attributed to Fermi-level detuning of the graphene/CuPc charge-transfer condition. The corresponding Raman mapping and mode-resolved analysis are provided in Section S5 and Figure S4. We next examine how the Raman response changes when the bottom interface is TiSe2. TiSe2 is chemically similar to TiS2 but shows a stronger chemically enhanced single-interface Raman response for CuPc. The difference is consistent with the semimetallic character of TiSe2 and its higher density of states near the Fermi level relative to TiS2, which facilitates stronger charge-transfer coupling to adsorbed molecules. A representative CuPc-coated TiSe2 flake partially covered by graphene is shown in Figure a, and the Raman intensity map of the CuPc 1528 cm–1 band (Figure b) defines the regions used for extracting spectra and enhancement factors. Within the same field of view, CuPc/TiSe2, graphene/CuPc, and graphene/CuPc/TiSe2 regions can be compared directly. CuPc/TiSe2 gives R = 2.1 and degree of charge transfer ρCT = 0.67, larger than CuPc/TiS2 (ρCT = 0.52), confirming that TiSe2 provides stronger charge-transfer coupling to CuPc than TiS2 as a single-interface substrate (Figure c). Figure d shows the Raman spectra, vertically offset for clarity. Unlike the TiS2 case, both TS and NTS modes in the graphene/CuPc/TiSe2 sandwich follow the same ordering across the three regions. The extracted enhancement factors of selected CuPc modes (680, 744, 1106, 1340, 1448, and 1528 cm–1) in Figure e place the sandwich response between the CuPc/TiSe2 and graphene/CuPc single-interface limits. For example, the NTS mode at 1340 cm–1 has enhancement factors of 2.4, 8.7, and 17.0 for CuPc/TiSe2, graphene/CuPc/TiSe2, and graphene/CuPc, respectively, while the corresponding values for the TS mode at 1106 cm–1 are 1.8, 2.2, and 4.4. This indicates that increasing the bottom-interface CT contribution reduces the strong symmetry-selective redistribution that is observed in graphene/CuPc/TiS2. A full mode-resolved graphene/CuPc/TiSe2 comparison, the extracted peak-intensity table, and enhancement factors are provided in Figure S5 and Tables S4 and S5. Mode-resolved Raman response in the graphene/CuPc/TiSe2 sandwich. (a) Optical image and (b) Raman intensity map of the CuPc 1528 cm–1 mode for a TiSe2 flake partially covered by graphene with a red scale bar of 10 μm. Gray and black dashed lines outline the TiSe2 and graphene flakes, respectively. (c) Within-spectrum intensity ratio I 1528/I 680 and the corresponding degree of charge transfer, ρCT. (d) Raman spectra with vertical offset for clarity of CuPc/TiSe2, graphene/CuPc/TiSe2, and graphene/CuPc. Gray vertical solid and dashed lines mark the selected CuPc TS and NTS modes used for the enhancement-factor comparison in part e. (e) Enhancement factors of selected CuPc modes for graphene/CuPc/TiSe2 and the corresponding single-interface references. The charge-transfer metric shows the same trend (Figure c). The graphene/CuPc/TiSe2 sandwich gives R = 16.2 and ρCT = 0.94, which is slightly larger than graphene/CuPc/TiS2 (ρCT = 0.93) but still below graphene/CuPc (ρCT = 0.95). Because TiSe2 provides a stronger CT contribution than TiS2, the two CT amplitudes in graphene/CuPc/TiSe2 are more evenly matched than those in graphene/CuPc/TiS2, reducing the symmetry-selective redistribution. Together, these measurements show that graphene/CuPc/TiSe2 remains an asymmetric sandwich whose response and charge transfer stay intermediate between the two single-interface limits. In graphene/CuPc/TiS2 and graphene/CuPc/TiSe2, the two molecule–2D interfaces are different, so the sandwich response reflects an unequal combination of two CT contributions. Thus, to examine the case in which the two contacts are chemically identical, we next study the graphene/CuPc/graphene sandwich structure, with the same top and bottom interfaces. This near-symmetry geometry minimizes uncertainty associated with interface substitution and provides a clean experimental platform for probing how dual contact modifies chemical enhancement. Single-interface graphene already provides strong chemical enhancement for CuPc through favorable Fermi-level alignment with the CuPc highest occupied molecular orbital (HOMO). , First-principles studies of vdW sandwich structures suggest that an additional contact can open extra resonant channels and provide strong Raman enhancement of molecules. In the graphene/CuPc/graphene sample, the sandwich region is shown by the optical image and by the spatial mapping of the CuPc 1528 cm–1 band under 532 nm excitation (Figure a,b). The top and bottom graphene flakes are marked with black and white dashed lines. The overlap region, where CuPc is sandwiched between two graphene flakes, shows the strongest CuPc Raman signal compared with nearby single-interface regions. Bright features at the lower left of the map arise from folded or twisted graphene introduced from transfer artifacts and are excluded from the quantitative analysis. Near-symmetric dual-interface Raman enhancement in graphene/CuPc/graphene. (a) Optical image and (b) Raman intensity map of the CuPc 1528 cm–1 mode under 532 nm excitation, with a white scale bar of 10 μm. Black and white dashed lines outline the top and bottom graphene flakes, respectively. (c and d) Raman intensities with error bars of selected CuPc modes under (c) 633 and (d) 785 nm excitation, normalized to the Si Raman peak. Compared with single-interface graphene/CuPc, the graphene/CuPc/graphene sandwich shows systematically larger CuPc Raman intensities across multiple modes after normalization to the Si peak (Figure c,d). At 633 nm, the 1340 cm–1 mode has enhancement factors of 25.4 and 29.4 for single-interface CuPc/graphene and graphene/CuPc/graphene, respectively; for 744 cm–1, they are 2.2 and 2.8. At 785 nm, the 744 cm–1 mode shows a much larger increase, from an enhancement factor of 3.4 for single-interface CuPc/graphene to 13.5 in the sandwich, corresponding to a sandwich-to-single ratio of nearly 4. We also performed density functional theory (DFT) calculations for the graphene/CuPc/graphene structure (Figure S6). Compared with graphene/CuPc, the binding energy of the sandwich is slightly smaller, whereas the charge transfer analyzed by Mulliken populations is slightly larger, indicating that introducing a second graphene contact modifies the interfacial electronic environment. In the sandwich, the top and bottom molecule–graphene interfaces can experience different local electrostatic conditions and interface dipoles. The sandwich therefore expands the range of accessible resonance conditions compared with a single interface. A distinctive feature of the near-symmetric graphene/CuPc/graphene sandwich is its pronounced excitation-wavelength dependence. We define the sandwich-to-single ratio as the CuPc Raman intensity in the graphene/CuPc/graphene region divided by that in the single-interface CuPc/graphene region, with both intensities normalized to the Si Raman peak. Under 633 nm excitation, the enhancement of the graphene/CuPc/graphene sandwich relative to single-interface graphene is modest (Figure c), while under 785 nm excitation, the sandwich-to-single ratio becomes much larger and reaches nearly 4 times for selected modes (Figure d). This nearly 4 times ratio arises because the two interfacial contributions enter the molecular Raman polarizability at the amplitude level, while two spatially separate molecular layers contribute approximately additively in the CuPc/graphene/CuPc geometry with a 2 times response. The mode-resolved sandwich-to-single ratios are summarized in Table S7. For single-interface CuPc/graphene, the excitation dependence is expected for charge-transfer-mediated Raman enhancement and has been reported previously. , Let us consider a possible Raman process named MMS, in which an electron initially in the HOMO state is scattered to the lowest unoccupied molecular orbital (LUMO) state of the molecule (M) by absorbing an excitation laser energy, and this excited electron is then scattered to an unoccupied state of the 2D surface material (S) by creating a phonon and is finally scattered back to the initial HOMO state of the molecule by the emission of a photon. The Raman scattering amplitude, RMMS , for the MMS process is given by RMMS=∫dE[1−f(E−EF)]g(E)×MHSerMSLepMLHer(EL−EH−Elaser−iγ)(E−EH+Eph−Elaser−iγ) 2 where M HS er and M LH er are the electron–photon matrix elements between the HOMO (H) and 2D surface (S) states and the LUMO (L) and H states, respectively, M SL ep is the electron–phonon matrix element between the L and S states, E laser, E ph, and E F are the excitation laser, vibrational, and Fermi energies, respectively, E L, E H, and E are the LUMO, HOMO, and surface state energies, respectively, g(E) is the electron density of states of the 2D material, f(E–E F) is the Fermi–Dirac distribution function, and γ represents line width broadening of the molecular states. For simplicity, we consider M HS er, M LH er, and M SL ep as constant values in eq . Due to the 2D nature of the metal, g(E) can also be treated as constant over the entire energy range of interest. Therefore, the Raman intensity, which is proportional to the squared Raman scattering amplitude, depends on both E laser and E F. In particular, the Raman enhancement can occur when the resonance condition holds in the denominator. It is noted that Barros and Dresselhaus identified 36 different Raman processes in the molecule/graphene system that can contribute to the Raman enhancement. These Raman processes can make a complete analysis of the Raman-enhancement process in 2D materials. They also showed that all Raman processes depend on both the excitation laser energy and the Fermi energy, in which the Raman enhancement occurs for the following four conditions: (1) E laser = E L – E H (or E laser = E L – E H + E ph), (2) E F = E H ± E ph (or E F = E L ± E ph), (3) E laser = E F – E H (or E laser = E F – E H + E ph), and (4)E laser = E L – E F (or E laser = E L – E F – E ph). Because the 633 nm excitation lies closer to the CuPc Q-band resonance at 1.98 eV, the Raman response is more sensitive to line width broadening γ, whereas 785 nm lies farther from resonance and is less affected by additional damping in the sandwich. Accordingly, the effect of the second graphene contact is expressed more clearly at 785 nm. The sandwich-to-single ratio of the CuPc 1528 cm–1 mode increases from about 1.1 at 633 nm to about 3 at 785 nm. To constrain the resonance parameters for CuPc on graphene, we adopt the excitation-profile parametrization reported by Ling et al. for graphene-enhanced Raman scattering of CuPc under multiple excitation wavelengths. An illustrative fit applying eq is shown in Figure S7. Figure a shows the sandwich-to-single ratio for six CuPc modes plotted against the absolute energy detuning from the Q-band resonance at three excitation wavelengths. Near resonance (633 nm), all six modes cluster between 1.1 and 1.5, indicating that damping limits the gain from the second contact. At the largest detuning (785 nm), the sandwich-to-single ratio increases substantially: the 744 and 680 cm–1 modes reach nearly 4 and above 3, respectively, while the higher-frequency modes show more modest gains. The larger ratios for low-frequency modes may be due to their smaller intrinsic line widths and weaker coupling to phonon-bath dephasing, making them more sensitive to the additional CT amplitude from the second contact at large detuning. The intermediate detuning at 532 nm shows an intermediate sandwich-to-single ratio between 633 and 785 nm. Thus, the second graphene contact enhances Raman scattering more effectively when damping is reduced, depending on the excitation energy and vibrational mode. Figure b summarizes this dual-interface picture, in which charge-transfer-enabled B- and C-term contributions arise from both molecule–vdW contacts. In the asymmetric graphene/CuPc/TiS2 and graphene/CuPc/TiSe2 sandwiches, the two interfaces contribute unequally with mismatched phases and damping. For the graphene/CuPc/graphene sandwich, the two contributions are more comparable, and favorable amplitude-level addition can produce an intensity approaching 4 times the single-interface response. More broadly, this dual-interface chemical enhancement provides a controllable design parameter that could be combined with plasmonic field confinement, offering a promising direction for pushing chemically enhanced Raman scattering toward higher sensitivity. Mode-resolved excitation dependence of graphene/CuPc/graphene and schematic of the dual-interface picture. (a) Sandwich-to-single Raman intensity ratio for six CuPc modes (680, 744, 1140, 1340, 1448, and 1528 cm–1) plotted versus the absolute energy difference between the excitation and the CuPc Q-band resonance. All modes remain above unity, indicating a stronger Raman response in the sandwich than in single-interface graphene/CuPc. The ratios are grouped closely near resonance and become both larger and more dispersed at greater detuning, with the largest mode-dependent spread observed for 785 nm excitation. (b) Schematic of the dual-interface Herzberg–Teller picture for representative graphene and TMDC interfaces. In summary, we experimentally study how a second direct molecule–2D interface modifies chemical Raman enhancement in molecule-sandwiched vdW structures. We find that adding a second contact can redistribute, limit, or enhance the Raman response depending on the electronic character of the two interfaces and the balance between the CT contribution and additional damping. The asymmetric TiS2 and TiSe2 sandwiches reveal how unequal interfaces reshape the Raman response in different ways, whereas the near-symmetric graphene/CuPc/graphene structure shows the strongest enhancement together with pronounced wavelength dependence. These results clarify how multiple direct molecule–material interfaces shape chemical Raman enhancement in vdW sandwich structures. The work establishes molecule-sandwiched vdW structures as a controlled platform for multi-interface chemical-enhanced Raman scattering and suggests that interfacial stacking can be used as a practical design parameter for tuning molecule–material coupling in vdW heterostructures, molecular spectroscopy, and sensing applications. Supplementary Material