Surface-enhanced Raman spectroscopy (SERS) has become a powerful analytical tool due to its high sensitivity and non-invasive nature since its discovery in 1974^1-3. Thus far, there are two main origins of SERS: electromagnetic mechanism (EM) and chemical mechanism (CM)^{4, 5}. EM is derived from the amplified electromagnetic field induced by localized surface plasmon resonance (LSPR) around noble metal nanoparticles triggered by incident light, thereby increasing the Raman scattering cross section of nearby molecules^{6, 7}. Although the Raman signal of the adsorbed molecules can be significantly enhanced depending on EM, the negative effects brought by noble metals cannot be ignored, such as steep price, poor biocompatibility and distortion of Raman signals caused by strong metal-molecule interactions^{8, 9}. On the other hand, CM is closely related to charge transfer (CT) between the substrate and molecules¹⁰. CT results in further separation of positive and negative charges, increasing the polarizability of molecules and thus heightening the Raman scattering cross section¹¹. This means that the key to improving the CM effect is to facilitate the interfacial CT process. In addition, CM-based two-dimensional (2D) materials for SERS offer some irreplaceable advantages over EM-based noble metals, such as low cost, high specificity, superior biocompatibility, excellent signal homogeneity and strong maneuverability^12-16.

If the energy gap between the substrates and probe molecules is tuned to align with the excitation laser energy, photo-induced charge transfer (PICT) will be effectively enhanced through the “borrowing effect” of the Herzberg-Teller coupling term, and consequently improving the SERS signal. To this end, various strategies, such as electric field regulation^{17, 18}, chemical doping^19-22, and gas molecular adsorption or desorption²³, have been developed to improve CM enhancement by manipulating Fermi level (E_f) or valence band (VB) and conduction band (CB) of substrates to match the highest occupied molecule orbital (HOMO) and the lowest unoccupied molecule orbital (LUMO) of the analyte. Feng et al. designed a nitrogen-doped graphene substrate where the E_f of graphene can be shifted by controlling the nitrogen doping level, thus achieving modulation of the CT efficiency with certain organic fluorescent molecules¹⁹. Seo et al. optimized the complementary resonance effects between the substrate and probe molecule by varying the oxygen concentration in the ReO_xS_y thin film²⁰. Liang et al. developed a two-dimensional (2D) borocarbonitride (BCN) SERS platform with adjustable CB level by changing the carbon atomic ratio for detection of specific molecular with appropriate band structure²¹. Zhou et al. developed a molecular adsorption strategy to regulate the CT efficiency between rGO and target molecules by controlling the adsorption or desorption of gas molecules on rGO surface²³. However, an inevitable problem of these similar approaches is that the characteristic of a SERS substrate is fixed after preparation, and only the probe molecules with specific band structure are efficient in detection. In addition, these methods are cumbersome to operate, which is not conducive to the convenient collection of the SERS signals of multispecies molecules with different energy band structures. Currently, it is of great significance to provide a solution with convenient adjustment of substrate’s electronic property to flexibly regulate the CT efficiency between the substrate and probe molecules with different band structures.

Lead magnesium niobate-lead titanate (PMN-PT) is a ferroelectric material in which the asymmetric distribution of positive and negative charges leads to the formation of electric dipoles^{24, 25}. At a certain temperature T (T < T_c ~ 135 °C), the self-polarized electric field generated by the stacking of electric dipoles induces an equal number of opposite charges on each side. In this case, holes or electrons are induced in a plane perpendicular to the direction of polarization, providing an excellent opportunity to modulate the electronic properties of the material adsorbed on PMN-PT surface²⁶. As a two-dimensional material with convenient preparation and good biocompatibility, graphene and its derivatives (graphene oxide (GO) and reduced graphene oxide (rGO) ) have attracted intense interest in the field of biosensors and photodetector^27-32. Since the proposal of graphene-enhanced Raman spectroscopy (GERS) by Zhang et al. in 2010³³, a great deal of work is reported on studying the mechanism and application of GERS^34-36. As we all know, graphene is an excellent candidate for studying CM enhancement owing to the fact that it does not support plasmonic excitation in the visible range³⁷. Furthermore, the linear dispersion of graphene for electrons with the density of states disappearing at the Dirac point allows the injection of relatively small numbers of charge carriers to alter significantly its electronic properties^{38, 39}.

In this work, we provide an innovative SERS substrate of graphene oxide coupled ferroelectric PMN-PT (PMN-PT@GO). The relationship between the doping level of GO and ferroelectric polarization was investigated and more quantitative information about the E_f of GO is characterized by employing a Kelvin probe force microscopy. We found that the E_f of GO could be well manipulated by the control of the direction and degree of ferroelectric polarization, and the effect on the chemical Raman enhancement of GO was studied. The SERS signal of rhodamine 6G (R6G) is further enhanced under stronger p-type doping, and consequently, the detection line is reduced from 10⁻⁷ to 10⁻⁸ M. In particular, the E_f could be changed actively by altering the temperature of the substrate owing to the variation of the ferroelectric polarization of PMN-PT. As a result, the detection line of R6G is again reduced to 10⁻⁹ M due to the stronger p-type doping triggered by the stronger ferroelectric polarization at low temperature. More importantly, the SERS signals of crystal violet (CV), methylene blue (MB) and p-nitrothiophenol (PNTP) molecules with different energy band structures from R6G are also successfully enhanced by selectively setting the temperature of substrate without the need of redesigning the substrate structure. The energy band regulation strategy proposed in this paper not only provides a new way to deeply understand the CM enhancement effect, but also is expected to be extended to other 2D materials due to the feasibility of ferroelectric modulation of semiconductor energy levels^40-44.

In order to determine the type of charge induced on the surface of PMN-PT, the short-circuit current was measured firstly by applying temperature fluctuations. As shown in Fig. S1(b), under forward connection, a positive current was received when the temperature rose from 20 to 100 °C (dT/dt = 2.5 °C/s), and a corresponding negative current was observed when the temperature drops from 100 to 20 °C (dT/dt = −2.5 °C/s). Notely, under reversed connection, the obtained current signals were switched in sign (Fig. S1(d)), which is evidence that the electric signals are coming from the pyroelectric effect of ferroelectric PMN-PT^{45, 46}. This is because the inherent spontaneous polarization (P_s) of PMN-PT leads to the existence of opposite free charges on the same electrode when it is forward and reversed connected to the measurement system, separately (Fig. S1(a) and S1(c))⁴⁷. Therefore, the polarization direction inside PMN-PT can be determined according to the pyroelectric current direction (see Section 1 in Supplementary information for details).

Next, based on the above detailed analysis of the polarization, GO was spin coated on both surface (perpendicular to the direction of P_s) of PMN-PT to prepare PMN-PT@GO (P_s⁺, the polarization direction points to GO film) and PMN-PT@GO (P_s⁻, the polarization direction departs from GO film) substrates, respectively. In addition, Al₂O₃ without ferroelectric effect was also selected to synthesize Al₂O₃@GO as a control group using the same method. The probe molecule R6G was employed to further investigate their SERS properties. As shown in Fig. 1(a), PMN-PT@GO(P_s⁻) exhibits the best SERS activity with an enhancement factor (EF) of 7.0×10³, followed by Al₂O₃@GO with an EF of 4.2×10³, while PMN-PT@GO (P_s⁺) is the weakest with an EF of 3.8×10³ (Fig. S2), which may be caused by the doping effect of PMN-PT on GO. SERS spectra taken from five random points of PMN-PT@GO(P_s⁻), Al₂O₃@GO and PMN-PT@GO (P_s⁺) were collected to evaluate the uniformity of R6G on the three samples, respectively. The three spectral gradients show nearly identical signal intensity on their respective samples (Fig. S3), indicating the uniform SERS effect of GO due to the flat two-dimensional surface and the homogeneous adsorption of R6G on GO. It has been reported that the frequency of the G band is shifted by changes in carrier concentration of graphene. N-type or p-type doping causes opposite shifts of the G band, due to stiffening or softening of phonons⁴⁸. It is noteworthy that, compared to the Raman frequency (1587 cm⁻¹) of the G band in Al₂O₃@GO, the Raman frequencies of the G band in PMN-PT@GO (P_s⁺) and PMN-PT@GO(P_s⁻) are red-shifted and blue-shifted to 1584 and 1593 cm⁻¹, respectively (Fig. 1(b)), corresponding to an increase and decrease in carrier concentration of ~2 × 10¹² cm⁻² and ~4 × 10¹² cm⁻²⁴⁹. This suggests that the ferroelectric effect of PMN-PT can indeed change the carrier concentration of adsorbed GO (Fig. 1(d−f)), leading to a shift in the corresponding GO Fermi level, thus further affecting the PICT process between the substrates and the probe molecule, as described below. In addition, when R6G is adsorbed on GO, the change of absorption peak (~230 cm⁻¹) also indicates experimentally that SERS enhancement is related to the CT interaction between them (Fig. 1(c))⁵⁰.

Subsequently, the further band structure analysis was carried out to more clearly illustrate the PICT process. For R6G molecule, the HOMO and LUMO levels are −5.7 and −3.4 eV, respectively, with an energy interval (∆E_h-l) of 2.3 eV⁵¹. Molecular resonance (μ_mol) occurs because ∆E_h-l almost perfectly matches the energy of 532 nm excited laser (2.33 eV). Kelvin probe force microscopy (KPFM) was used to determine the E_f of GO for each sample using the gold probe with the E_f of −5.10 eV as a reference (see Section 4 in Supplementary information for details)⁵². The contact potential differences (V_CPD) between Au tip and GO in Al₂O₃@GO, PMN-PT@GO (P_s⁺) and PMN-PT@GO(P_s⁻) are −155, −225 and +250 mV, respectively, as shown in Fig. 1(g–i). Therefore, the Fermi levels of GO in Al₂O₃@GO, PMN-PT@GO (P_s⁺) and PMN-PT@GO(P_s⁻) are −4.95, −4.85, and −5.41 eV (vs. E_vac), respectively. According to the above results, the corresponding SERS enhancement mechanism is proposed. Owing to the fact that the transition energy (∆E_f-l = 1.55 eV) from the E_f of GO in Al₂O₃@GO to the LUMO of R6G is mismatched with incident photon energy, the PICT process is obstructed and thus cannot provide a satisfactory SERS effect, as depicted in Fig. 1(j). Even weaker SERS signal appears on PMN-PT@GO (P_s⁺) due to further suppression of PICT by a narrower band gap (∆E_f-l = 1.45 eV) than incident photon energy (Fig. 1(k)). However, the E_f of GO is closer to the HOMO of R6G due to the enhanced p-type doping of PMN-PT (P_s⁻) on GO. The energy of PICT (∆E_f-l = 2.01 eV) is better matched with incident photon energy, and consequently higher SERS activity is obtained (Fig. 1(l)). This phenomenon can also be easily understood by utilizing the polarization tensor formula: α_σρ = A + B + C, where A represents molecular resonance, which is associated with inter-band transitions within molecules, B refers to the surface plasmon resonance (SPR), and C denotes the PICT resonance at the Fermi level between substrate and molecule (see Section 5 in Supplementary information for details)⁵³. Term B is not considered in this paper because the plasmonic excitation of graphene is powerfully confined to the range of terahertz/infrared spectral⁵⁴. Among all samples, only PMN-PT@GO(P_s⁻) can satisfy both the molecular resonance A and optimal PICT resonance C, where μ_mol ≈ μ_PICT. In this case, the polarization tensor of molecular is significantly amplified through the Herzberg-Teller vibration coupling effect⁵⁵. Therefore, PMN-PT@GO(P_s⁻) exhibits the largest SERS activity. We also tested the enhancement properties of three samples against R6G molecules using a 785 nm laser (Fig. S5). Interestingly, only the Al₂O₃@GO exhibits characteristic peak of R6G due to the PICT resonance from the E_f of GO to the LUMO. However, the extremely weak signal compared to the obvious characteristic peaks under 532 nm laser excitation is due to the inability of 785 nm to trigger resonance Raman scattering (RRS or called molecular resonance).

To confirm the working mechanism, first-principles density functional theory (DFT) calculations were performed to study the interaction properties of graphene with ferroelectric PMN-PT polarization. Figure 2(a−c) show the constructed atomic structure model. The initial structure of PMN-PT was optimized with the polarization direction normal to the graphene surface, including graphene on both the up-polarized (P_s⁺) and down-polarized (P_s⁻) surfaces. The opposite net surface charges appear on each side of the bulk ferroelectric PMN-PT owing to the polar stacking. These surface charges will be passivated by surface reconstruction, mobile charges, and in this case by graphene. In other words, graphene is responsible for providing additional compensation⁴⁹. In intrinsic graphene, the Fermi level is located at the Dirac point energy (Fig. 2(d)). However, as a result of contact with the P_s⁺ and P_s⁻ surfaces, the Dirac cone of graphene on the P_s⁺ surface is shifted downward due to the increased electron density (Fig. 2(e)), and the Dirac cone of graphene on the P_s⁻ surface is shifted upward due to the decreased electron density (Fig. 2(f)), with respect to the Fermi level of intrinsic graphene. This indicates effective n-type or p-type doping in graphene. We emphasize that the simplified model is designed to capture the physical nature of the problem, not to provide quantitative consistency with experiments. The theoretical results indeed prove that the self-polarized ferroelectric PMN-PT could realize the regulation of the Fermi level of graphene adsorbed on its surface, which in turn is responsible for the further enhancement of SERS signals of probe molecules.

It is well known that the polarization of ferroelectric materials varies with ambient temperature^{24, 56}. The variation of polarization affects the free charge density of the material surface⁵⁷. Inspired, we carefully investigated the temperature-dependent SERS response behavior of PMN-PT@GO(P_s⁻) due to the optimal SERS activity. As depicted in Fig. 3(a), the signal intensity of R6G decreases or increases as the temperature of PMN-PT@GO(P_s⁻) rises or drops. This may be attributed to smaller or larger P_s⁻ inside PMN-PT caused by higher or lower temperature, and consequently changing the surface charge density (Fig. 3(b)). At room temperature, PMN-PT possesses a certain strength of P_s⁻, which will induce a certain number of holes in GO (Fig. 3(b-i) and 3(c-i)). At higher temperature, the smaller P_s⁻ inside PMN-PT induces fewer holes in GO, leading to the E_f to ascend, which attenuates the PICT effect (Fig. 3(b-ii) and 3(c-ii)). Contrarily, at lower temperature, the larger P_s⁻ inside PMN-PT induces more holes in GO, making the E_f fall and further closer to the HOMO of R6G (Fig. 3(b-iii) and 3(c-iii)). Notably, there is no significant change in SERS strength on Al₂O₃@GO, despite the presence of possible interfacial strain (Fig. 3(d)). To assess the sensitivity of PMN-PT@GO(P_s⁻), the concentration gradient experiment was performed. As shown in Fig. 3(e), the detection limit of R6G is only 10⁻⁸ M at 20 °C due to the weak chemical enhancement compared to that of EM^{58, 59}. However, the characteristic peaks of R6G with the concentration of 10⁻⁹ M are still distinguished at 0 °C (Fig. 3(f) and Fig. S6). Differently, the Al₂O₃@GO substrate can only detect SERS signal of 10⁻⁷ M either at 0 °C or 20 °C (Fig. 3(g)). In a word, PMN-PT at lower temperature enables the E_f of GO better match with the HOMO of R6G, which significantly improves the SERS performance of PMN-PT@GO(P_s⁻). In addition, temperature changes of substrate may trigger degradation of the molecule (Fig. S7). The SERS signal intensity did not decay significantly until after 150 cycles, which was not enough to affect the accuracy of the other temperature-related experiments results (within one or two cycles).

A profoundly critical fact to note is that the CM effect strongly depends on the energy level arrangement of the substrate-molecule system and the wavelength of the excited laser. The different choices of excited laser and molecules to be tested greatly affect the CM enhancement factor of SERS substrate⁶⁰. The SERS signal of R6G excited by 633 nm laser is much weaker than that excited by 532 nm laser (Fig. S8(a)). This is attributed to inactivated molecule resonance due to the insufficient energy of 633 nm laser (1.96 eV), indicating that molecule resonance Raman scattering (RRS) plays a pivotal role in the total signal enhancement (Fig. 3(c-i) and Fig. S8(c-i))⁶¹. Interestingly, different from the case under 532 nm laser excitation, PMN-PT@GO(P_s⁻) shows the strongest SERS activity at 20 °C under 633 nm laser excitation (Fig. S8(b)). That's because the energy required to jump from the E_f of GO to the LUMO of R6G at 20 °C already matches the energy of 633 nm laser. The upturn or downturn of the E_f would result in a mismatch of levels (Fig. S8(c-ii) and S8(c−iii)). In addition, CV was also employed to study the CM effect. As displayed in Fig. 4(a), the SERS signal of CV excited by 633 nm laser is stronger than that excited by 532 nm laser. For CV molecule, the HOMO and LUMO levels are −6.0 and −4.1 eV, respectively. Although the electron transition from HOMO to LUMO in CV can be activated by either 532 or 633 nm laser, the energy of 633 nm laser is better matched to the energy interval (1.9 eV) than that of 532 nm laser, as shown in Fig. 4(c-i) and 4(d-i). Thus, the molecule resonance of CV is enhanced with 633 nm laser excitation. Figure 4(e) and 4(f) show the temperature-dependent SERS responses of CV molecule under excitation of 532 and 633 nm laser, respectively. Owing to the more effective PICT process, the strongest SERS signal is obtained at 0 °C in both cases (Fig. 4(c-iii) and 4(d-iii)). In synergy with the molecular resonance and the optimized PICT resonance, the detection limit of CV on PMN-PT@GO(P_s⁻) substrate can also be optimized to 10⁻⁹ M (Fig. S9).

To further investigate the universality of the CM enhancement strategy proposed in this paper, MB and PNTP were also used to conduct SERS measurement. MB exhibits SERS modulation behavior similar to that of CV due to similar energy level structure (Fig. S10). However, since the large energy gap between the LUMO and HOMO in PNTP, molecule resonance cannot be activated with the insufficient excitation energy of either 532 or 633 nm laser (Fig. 5(d-i) and 5(e-i)). It is noteworthy that PNTP is better detected by PMN-PT@GO(P_s⁻) at 40 °C, rather than at 0 °C with either 532 or 633 nm laser excitation (Fig. 5(a) and 5(b)), due to the more suitable energy position (Fig. 5(d-ii) and 5(e-ii)). For mixed solutions, as shown in Fig. S11, the SERS spectra of R6G and PNTP were selectively enhanced at 0 and 40 °C compared to that at 20 °C, respectively. Therefore, the optimized SERS responses of molecules with different band structures can be obtained by flexibly adjusting the E_f of GO through simple temperature changes without the need to redesign the SERS substrate. Also, it should be pointed out that PMN-PT@GO (P_s⁺) exhibits higher SERS activity than PMN-PT@GO(P_s⁻) for PNTP (Fig. S12(a) and S12(b)). Higher E_f of PMN-PT@GO (P_s⁺) is conducive for electrons in it to transition to the LUMO of PNTP (Fig. S12(c-ii) and S12(c-iv)). The corresponding temperature-dependent SERS spectra were also measured, and adjustable PICT efficiency was shown (Fig. S13). Although the PMN-PT@GO (P_s⁺) substrate exhibited the highest SERS activity toward PNTP at 20 °C, its detection limit was only able to reach 10⁻⁷ M (Fig. S14), which may be due to the inactivated molecular resonance under 532 nm excitation laser (Fig. S13(d-i)).

In summary, we have presented a flexible energy band regulation strategy to manipulate the E_f of GO utilizing the ferroelectric PMN-PT for SERS detection of molecules with different band structures. The SERS signal is significantly enhanced after the energy gap between the E_f of GO and the LUMO of molecules is tuned to align with the energy of excitation laser, due to the enhanced PICT efficiency. More importantly, this strategy provides new perspective for the SERS substrate design based on CM effect.