Glucose molecule is one of the most important molecules in the process of the life activities¹⁻³, and accurately obtaining its structural information is of great significance for the development of life, disease treatment, and molecular science⁴⁻⁶. So far, many researchers have attempted to analyze the glucose molecules by the Raman scattering with the advantages of non-contact and unlabeled properties^7,8. Unfortunately, examining glucose molecules using Raman scattering has been proven to be extremely challenging, due to their small Raman scattering cross-section and weak adsorption/non-adsorption on bare metals⁹.

Raman scattering cross-section of glucose molecules (5.0±1.1 – 8.9±0.9×10⁻³⁰ cm²·molecule⁻¹·sr⁻¹) is about five times smaller than that of the benzene molecules¹⁰, which have a large scattering cross-section of 2.8×10⁻²⁹ cm²·molecule⁻¹·sr⁻¹. Theoretically, the surface-enhanced Raman scattering (SERS) can amplify the Raman scattering intensity of the glucose molecules to detectable levels¹¹⁻¹⁴. Nevertheless, the weak/non-adsorption of glucose molecules on bare metal surfaces hinders efficient interaction between the glucose molecules and the localized electric field on the metal surface. Although surface modification or functionalization methods have been proposed to facilitate the adsorption of glucose molecules on bare metal surfaces¹⁵⁻¹⁷, the static plasmonic nanocavity characteristics of SERS lead to the random detection of only partial Raman modes of glucose molecules, accompanied by interference from Raman scattering of the bridging molecules.

Tip-enhanced Raman scattering (TERS) overcomes the limitations of the static plasmonic nanocavity in SERS¹⁸⁻²¹, but current TERS techniques still appear inadequate for the glucose molecule examination. Scanning tunneling microscopy (STM)-based TERS requires the tunneling junctions between the plasmonic tip and a conductive substrate. However, the clustering characteristics of the glucose molecules hinder the formation of these tunneling junctions²². Atomic force microscope (AFM)-based TERS techniques can overcome the obstacles of the glucose molecular clusters²³⁻²⁵, but the electric-field enhancement of the AFM-TERS is limited under the axial/side illumination and accompanied by strong background noise. This makes it difficult for the AFM-TERS to detect the weak Raman signal of the glucose molecules.

Herein, we developed a TERS platform that integrates both a feedback mechanism and a novel illumination method. The shear-force feedback of the scanning near-field microscopy (SNOM) adopted to control the tip-substrate distance does not depend on the flatness and conductivity of the substrate, offering greater flexibility in substrate selection for our SNOM-TERS. Furthermore, using the fiber-based radial vector mode (RVM) internally illuminating the plasmonic fiber tip (PFT) significantly enhances the energy conversion and suppresses the background noise. This innovation enables the creation of a background-free tip nanofocusing hotspot with the electric-field intensity amplified by two orders of magnitude. Approaching the tip hotpot to the glucose molecules located on the hydrophilic silicon, all Raman vibrational modes of the glucose molecules have been successfully visualized within the Raman window of 400–3200 cm⁻¹, consistent with the density functional theory (DFT) calculations.

The home-built SNOM-TERS platform adopted an architecture of the fiber RVM internal excitation and the lateral spectral collection is shown in Fig. 1(a1). A He-Ne laser at λ=632.8 nm was used as the excitation source. The incident direction, line width, intensity, and linearly polarized direction of the excitation light source was managed through a series of components of M, LL, A, HWP, and then coupled into a four-mode fiber (FMF) via a fiber adopter (FA). The high-order vector modes in FMF were filtered via a mode stripper (MS)²⁶, leaving only the fundamental vector mode (HE^x₁₁) in the fiber core. The HE^x₁₁ mode was high efficiently converted to the RVM (TM^OM₀₁), as shown in Fig. 1(a2), via an acoustically-indued fiber grating (AIFG)²⁷, which was produced using an acoustic transducer (AT) driven by a radio frequency source (RFS). As shown in Fig. 1(b), the fiber RVM was used to internally illuminate the PFT to achieve the generation of the background-free tip hotspot^28,29, which was bunded on the tuning fork to approach the target analytes dispersed on the substrate (Supplementary Section 1). The tip-substrate distance was adjusted using the shear-force feedback of SNOM to enhance the Raman scattering intensity of the tip hotspot-molecules interaction process. Figure 1(c) is the scanning electric microscopy (SEM) image of a typical PFT with the tip curvature radius of 20 nm, and the background-free tip hotspot formatted at the tip apex of the PFT was photographed in Fig. 1(d).

As shown in Fig. 1(a1), the SNOM-TERS spectrum was laterally collected via a long work distance micro-objective (MO, 50×, NA=0.42, Working distance: 4 mm) and then coupled into the spectrometer via a FA after filtering the Rayleigh line by an edge filter (EF). The included angle between the PFT and the MO was determined based on the far-field scattering calculation of the tip hotspot (Supplementary Section 2). A charge coupled device (CCD) assisted by a beam splitter (BS) was adopted to monitor the position relationship between the collection light path and the PFT.

The electric-field intensity enhancement factor of the tip hotspot was calculated using the finite-time domain differenced (FDTD) method³⁰. As sketched shown in Fig. 2(a), the RVM of FMF was obtained by a boundary mode source at λ=632.8 nm and then was adopted as the internal illumination light source. The PFT has a curvature radius of 20 nm, a conic angle of 15°, and an Ag film thickness of 20 nm. The dielectric constant of Ag material was obtained from ref.³¹, and the refractive index of the FMF and the silicon substrate were taken from the Palik handbook³². To ensure the calculation resolution, the tip apex region of PFT was meshed to be 0.1 nm, while gradually increased mesh scale of other areas and the transition metal boundary conditions were used to reduce the computation time.

The shear-force feedback makes it difficult to accurately obtain the variation of the tip-substrate distance, which was selected based on the data provided in the SNOM usage instructions³³. Although the nanocavity-plasmonic mode is not formed in the nanogap, but the electric-field enhancement of the tip hotspot remains highly sensitive to the variations of the tip-substrate distance (Supplementary Section 2). Figure 2(b) is the calculated electric-field intensity enhancement factor |E_tip/E₀|² of the tip hotspot in the cased of d=10 nm, revealing that the electric-field intensity of the tip hotspot undergoes drastic changes from the tip apex to the dielectric substrate. Additionally, it still has two orders of magnitude enhancement on the dielectric substrate surface, and the effective suppression of the background noise is sufficient to compensate for the shortcomings of the electric field intensity enhancement.

When the PFT was attached to the tuning fork, the mass and rigidity distribution of the "tuning fork+PFT" underwent significant changes, resulting in a decrease in the Q-factor of the low- and high-frequency resonance modes of the tuning fork, and the Q-factor of the high-frequency resonance mode was greater than that of the low-frequency resonance mode (Supplementary Section 3). Therefore, the high frequency resonance mode was adopted to perform the shear-force topography to better resist external environmental interference. More importantly, the transverse displacement of the "tuning fork+PFT" working in the high-frequency resonance mode was one order of magnitude smaller than that of the low-frequency resonance mode (Supplementary Section 3), ensuring that the background-free tip hotspot drove as few molecules as possible to undergo Raman scattering process. Background noise suppression of the tip hotspot was examined using the single-wall carbon nanotubes (SWCNT, Diameter: 1.2–1.7 nm) dispersed on the silicon substrate (Supplementary Section 4). Figure 2(c) is the shear-force topography of the SWCNT dispersed on the silicon substrate with the performance of the "tuning fork+PFT" being of previously validated by the standard samples (Supplementary Section 5). By increasing the excitation power from 0 mW to 0.12 mW, a time series of the SNOM-TERS spectra was obtained, as shown in Fig. 2(d). The signal-noise ratio (SNR) of the extracted spectra (Fig. 2(e)) was quantitatively analyzed to be SNR=23 (1556 cm⁻¹), which was three times higher than that of the far-field Raman spectrum of the SWCNT (Supplementary Section 4). With the excitation power being of maintained at 0.2 mW, the time-series of SNOM-TERS spectra of the SWCNT at 1556 cm⁻¹ was examined, as shown in Fig. 2(f). The relative standard deviation of the Raman signal intensity was estimated to be 4.7%, exhibiting that the SNOM-TERS platform retained excellent time stability.

Based on the home-built SNOM-TERS platform, the "fingerprint information" of the glucose molecules was examined. To eliminate the interference of the bridging molecules, the noble metallic substrate was replaced by the hydrophilic silicon (Method 2) with the hydroxyl group effectively adsorbing the glucose molecules. Figure 3(a) is the shear-force topography of the glucose molecules on the hydrophilic silicon (Method 3), revealing that the hydroxyl groups of the glucose molecules result in clusters on the hydrophilic silicon, rather than monolayers. Figure 3(b) is a typical three-dimensional morphology distribution of one glucose molecule cluster with a height of ~25 nm (Fig. 3(c)). Furthermore, the far-field Raman scattering spectra of the glucose molecules dispersed on the silicon and the hydrophilic silicon, as show in Fig. 3(d), furtherly proving that the effectively adsorption of the glucose molecules on the hydrophilic silicon. With the background-free tip hotspot approaching the top of the glucose molecules cluster, the time-series of the SNOM-TERS spectra was obtained within the spectral window of 400–2000 cm⁻¹, as shown in Fig. 3(e), with the excitation power increasing from 0.1 mW to 0.25 mW. The visualized Raman modes gradually increased as the intensity of the tip hotspot gradually increased. With the excitation power being of maintained at 0.25 mW, a SNOM-TERS spectrum of the glucose molecules within the spectral window of 400–3200 cm⁻¹ was examined, as shown in Fig. 3(f), and the inset shows a partial enlargement of the spectral recording window (400–2000 cm⁻¹) with weak Raman scattering intensity. Note that, the stretching vibration of C=O functional group (~1750 cm⁻¹) of the chain glucose molecules has been clearly observed, consistent with the DFT calculations (Fig. 3(g)). The D-(+)-glucose molecules can be formed by the chain glucose molecules that undergo internal reactions, where the interaction between C=O and –OH groups cause C=O change to C–O, leading to the inability to observe Raman peak of C=O functional group (~1750 cm⁻¹).

Because of the low resolution of the spectrometer during the broadband spectrum operation, the SNOM-TERS spectrum of the chain glucose molecules exhibited a wide peak with multiple sharp peaks (Fig. 3(f)). To clearly distinguish the multiple Raman vibrational modes, multiple peaks Gaussian fitting on the SNOM-TERS spectrum was performed within twelve segmental Raman windows and compared detailly with the DFT-Raman calculation results for each segmental Raman window, as shown in Fig. 4. The comparison between the DFT-calculation and the SNOM-TERS Raman vibrational modes of the chain glucose molecules was summarized in Supplementary Table S1. All Raman vibrational modes (thirty-four) of the chain glucose molecules have been effectively visualized within a spectral recording window by our developed SNOM-TERS platform, while this visualization was not achievable with the conventional far-field excitation Raman, SERS, and STM-TERS (Supplementary Section 6). Unlike previous studies that randomly observed the vibrational modes of glucose molecules^34,35, our SNOM-TERS approach allows for the simultaneous visualization of Raman vibrational modes in glucose molecules. Compared with SERS technology, the SNOM-TERS can avoid the interference of bridging molecules in Raman spectrum, and by adjusting the spatial relationship between the background-free tip hotspot and the glucose molecules (Supplementary Seciton 7), the excitation efficiency and SNR of Raman spectra of the glucose molecule can be significantly improved. In addition, compared with STM-TERS technology, the shear-force feedback mechanism of SNOM-TERS can position the background-free tip hotspot on the surface of the glucose molecules cluster, thereby achieving efficient excitation of TERS spectrum of glucose molecules.

Due to the small Raman scattering cross-section of the glucose molecules, a longer integration time was required to obtain the effective SNOM-TERS signals, which placed higher demands on the time stability of the SNOM-TERS platform. The Raman window (2800–3000 cm⁻¹) with strong Raman activity was selected to assess the time stability of the SNOM-TERS platform. As shown in Fig. 4(l1), the decomposed Gaussian fitting results of the SNOM-TERS spectrum involve five vibrational modes, coinciding with the DFT calculations (Fig. 4(l2)). The highly active Raman mode (~2906 cm⁻¹) associated with C1,2,3,4,5–H stretching vibration was selected to time mapping. The relative standard deviation of the Raman scattering intensity was calculated to be 2.8% (Supplementary Section 8), indicating the excellent time stability of the SNOM-TERS platform.

In summary, the vibrational modes of glucose molecules have been effectively visualized using our developed SNOM-TERS platform. The shear-force feedback overcomes the clustering characteristics of the glucose molecules and drives the PFT to approach the surface of the molecular cluster. The background-free tip hotspot with significant electric-field intensity enhancement effectively solves the small Raman scattering cross-section of the glucose molecules and increases the SNR of Raman scattering, resulting in all vibrational modes of the glucose molecules being simultaneously visualized within the Raman recording window. The examination of vibrational modes of the glucose molecules is of great significance for the development of life, disease treatment, and molecular science. Our SNOM-TERS technology provides an effective and promising platform for accurately identifying the near-field Raman scattering of molecules and expands the applicability of TERS technique.

The model of a glucose molecule was constructed using the Chem 3D software, and the Raman spectrum of a glucose molecule in the gas phase was calculated using the DFT³⁶, B3LYP functional, and the 6-31G (d) group set in Gaussian 09 software. The vibrational modes corresponding to each Raman peak of the glucose molecules were obtained from the Result-Vibrations function window of Gaussian View software and scaled by a frequency scaling factor of 0.9613. The Gaussian output file was imported into the Multiwfn software to calculate the Raman intensity of each vibrational mode at 300 K and 632.8 nm excitation³⁷.

Hydrophilic treatment of silicon originated from ref.³⁸. A silicon wafer was soaked in a 100 °C piranha solution (concentrated H₂SO₄:30% H₂O₂ volume ratio=7:3) for 30 minutes for hydrophilic treatment, and then the silicon wafer was placed in the ultrapure water for ultrasound of 5 minutes to remove the surface residues. Figure S15 shows the Raman spectra of the silicon (black curve) and the hydrophilic-treated silicon (red curve), revealing that the hydrophilic silicon does not introduce additional Raman vibrational modes.

Glucose powder was purchased from Aladdin. 180 mg of the glucose powder was added to 10 mL of the ultrapure water to obtain the glucose solution (10⁻¹ mol/L). The hydrophilic treated silicon was soaked in the glucose solution for 24 hours, and then the remaining moisture on the silicon wafer was blown dry with nitrogen gas. Figure 3(d) is the Raman spectrum of the glucose molecules deposited on the silicon (black curve) and the hydrophilic silicon (red curve). Compared with the silicon substrate, more vibrational modes of glucose molecules can be examined on the hydrophilic silicon substrate, which indirectly indicates the effective adsorption of the glucose molecules on the hydrophilic silicon.

This work was supported by National Natural Science Foundation of China (12374358, 91950207), Guangdong Basic and Applied Basic Research Foundation (2024A1515010420).

Z.L.X. and C.M. built the experimental configuration, performed the experiments. L.X. and T.M. analyzed the data. Z.L.X. performed electromagnetic field calculations. C.M. performed DFT calculations. D.H.Y. performed Investigation. W.D.Z. designed and supervised the work. All authors discussed the results and contributed to the writing of the manuscript.

The authors declare no competing financial interests.

Supplementary information for this paper is available athttps://doi.org/10.29026/oes.2025.240027