Auramine O is a water-soluble organic dye commonly used in industrial products such as cotton textiles, paper, and leather [1]. It is also utilized as a fluorescent dye for staining microorganisms in biological experiments. However, it poses serious health hazards to humans, with prolonged exposure increasing the risk of cancer. It is difficult to degrade naturally and is classified as a human carcinogen by the International Agency for Research on Cancer (IARC) [2]. In the food and pharmaceutical industries, Auramine O has been strictly prohibited by many countries [3]. Unfortunately, driven by commercial interests, Auramine O is still being abused in the food industry, posing a serious threat to human health. Therefore, there is an urgent need to develop an effective method for detecting it. Currently, various platforms have been designed for the analysis and determination of Auramine O, such as high-performance liquid chromatography (HPLC), gas chromatography, Raman spectroscopy, and carbon dot sensing methods [4–6]. However, these techniques often require cumbersome experimental steps, bulky instruments, and specialized operators, which greatly restricts their promotion and development. In recent years, research has been devoted to the development of a simple, fast, and economical method of detecting Auramine O.

Metasurfaces are artificial structures composed of periodically arranged subwavelength units, exhibiting unique electromagnetic responses such as negative permeability, negative permittivity, and negative refractive index [7–9]. Material types and unit structures can be reasonably selected to achieve desired resonant properties [10]. Meanwhile, strong local field enhancement effects occur near the resonance frequency, effectively enhancing the interaction between trace analytes and incident electromagnetic waves, endowing metasurfaces with extreme sensitivity to changes in the surrounding medium [11]. Traditional metallic metasurfaces, while significantly enhancing the interaction between light and matter through local field enhancement effects, also encounter issues such as inherent metal losses and high background noise [12–14]. In recent years, all-dielectric metasurfaces have attracted widespread attention from scholars due to their lower inherent losses and higher damage thresholds [15–17]. However, they still face challenges in controlling radiation losses, which not only reduces the quality factor (Q factor) but also lowers the efficiency of light–matter interactions [18–20].

Nowadays, the emergence of bound states in the continuum (BIC) in metasurfaces holds the promise of addressing the above problems. BIC can achieve resonances with infinite Q factors, holding great potential for applications in some fields such as matter sensing [21], lasers [22], and imaging and nonlinear optics [23]. BIC can be classified into two major categories. One type is symmetrically protected BIC, where the structure exhibits in-plane inversion symmetry and is often found in periodic systems [24]. The other type is accidental BIC, achieved by adjusting parameters of the target system to achieve radiation cancellation in the continuum [25]. In practical applications, people generally prefer to break in-plane symmetry to transform ideal BIC into quasi-BIC (QBIC) [26–30]. QBIC possesses a high Q factor, capable of detecting tiny frequency shifts caused by trace substances, thus meeting the requirements of sensing applications. The reported QBIC metasurfaces mostly employ metallic materials in the design [31–33]. Metals exhibit significant losses under terahertz wave interactions. Therefore, in pursuit of lower losses, researchers are exploring QBIC metasurfaces under an all-dielectric.

In the visible to near-infrared spectral range, the unit cell thickness size of all-dielectric QBIC metasurfaces typically ranges from tens to hundreds of nanometers. These samples are usually fabricated using thin film deposition techniques and have achieved experimental results with Q factors reaching the order of 10³–10⁴ [34–36]. By contrast, it is difficult to obtain experimentally all-dielectric QBIC metasurfaces operating at the terahertz (THz) band, due to high thickness of the unit cell. The unit cell thickness of THz metasurfaces is generally in the range of tens to hundreds of micrometers. Such high thickness cannot be achieved through thin film deposition techniques and can only be realized by thinning thick dielectric materials through etching processes, which significantly increases the difficulty of fabrication. Hence, most research on all-dielectric THz QBIC metasurfaces has remained at the numerical simulation stage [37–39]. Only a few studies have reported experimental research on all-dielectric THz QBIC [40–42], but the reported sample preparation still faces some challenges such as high cost. It is necessary to design more optimized device structure and explore more simple preparation techniques for all-dielectric THz QBIC metasurfaces.

In this study, we propose a silicon based THz metasurface composed of hollow silicon arrays, supporting QBIC. The in-plane symmetry of the structure is broken with a simple hollow design, achieving a dynamic transition from BIC to QBIC. Subsequently, suitable structural parameters were selected based on simulation results and fabricated into samples. Experimental testing confirmed the presence of a high-Q resonance peak at 0.33 THz, achieving QBIC in the THz regime with all-dielectric hollow structures. Subsequently, a detection scheme for Auramine O was designed based on this resonance peak. Through simulation analysis, an estimation of the detection performance and trends of the metasurface was completed. Subsequent experimental results indicate that the designed metasurface can identify changes in the concentration of Auramine O based on refractive index variations. Furthermore, the metasurface passed reliability tests for temperature and humidity, verifying its feasibility in various environments. These results indicate that our study not only provides a new approach for Auramine O detection but also offers new insights into the realization and application scenarios of all-dielectric THz QBIC.

The designed metasurface consists of a periodic array of hollow high-resistivity silicon structures, as shown in the schematic diagram in Fig. 1(a). The substrate-free all-dielectric metasurface not only simplifies the manufacturing process but also effectively avoids the loss of the substrate material on the incident wave. The schematic diagram of a single unit cell structure is shown in Fig. 1(b), with the following geometric parameters: the thickness of the full silicon metasurface is $H=200\mathrm{\mu m}$, and the period is $P=600\mathrm{\mu m}$. In the center, there is a hollow structure with a length of $m=420\mathrm{\mu m}$ and a width of $n=340\mathrm{\mu m}$. However, directly above this hollow structure, a solid rectangle with period $g$ is still retained. Here, $g$ is an adjustable parameter to break the in-plane rotational symmetry of the overall structure. The silicon is lossy; its material type is selected as normal, with a permittivity of 11.9, permeability of 1, and conductivity of 0.00025 S/m. Figure 1(c) shows the optical path of the terahertz time-domain spectroscopy (TDS) system. The metasurface is horizontally placed on the test platform. Terahertz radiation is generated by the photonic antenna at the emission end and focused onto the metasurface sample through lenses, and then reaches the detector through another lens. The time-domain waveform of the terahertz pulse is obtained by scanning the relative delay between the THz pulse and the detection pulse. After Fourier transformation, the spectral information of the sample under test can be obtained. The appearance of a single metasurface is shown in Fig. 1(d). The sample consists of $20\times 20$ units, with actual size of $1.6\mathrm{cm}\times 1.6\mathrm{cm}$. The imaging of the metasurface under an optical microscope is shown in Fig. 1(e). Figure 1(f) is used to verify the actual etching depth of the metasurface. It is placed on a platform with a 45° tilt angle for scanning electron microscopy imaging. From the image, the measured depth is approximately 140 μm. Calculating the actual depth as $H=140\mathrm{\mu m}\text{/}\cos 45°=197\mathrm{\mu m}$, it falls within an acceptable error range.

The specific fabrication process is illustrated in Fig. 1(g). First, the 200 μm thick, 10 kΩ high-resistivity silicon wafer is cleaned with acetone and isopropanol. Next, a layer of aluminum film is deposited on the bottom surface of the silicon wafer. This is because after the silicon wafer is etched through, the equipment may issue an error warning. Adding aluminum ensures the normal operation of the equipment. Then, a photoresist is spin-coated on the top surface of the silicon, and the mask pattern is exposed and developed using ultraviolet light to create the designed structure. Subsequently, inductively coupled plasma (ICP) etching technology is employed. After repeated cycles of passivation and etching, the desired etching depth is achieved. Finally, after removing the photoresist and aluminum film, the designed hollow metasurface is obtained. Compared to traditional etching techniques, ICP offers advantages such as fast etching rate, high selectivity, high anisotropy, low etching damage, good uniformity over large areas, high controllability of the etched surface profile, and smooth etched surface. In recent years, ICP etching technology has been widely used in the etching of materials such as silicon and silicon dioxide.

The resonant spectrum analysis of the metasurface was performed using the CST Microwave Studio software designed for terahertz metasurfaces and employing the time-domain solver. In the simulation, periodic boundary conditions were set in the $x$ and $y$ directions, while open boundary conditions were applied in the $z$ direction. In the background properties, the material type is selected as normal. The excitation source is chosen as a plane wave, utilizing linear polarization. Additionally, a probe is placed at 3000 μm from the metasurface for observation. To enhance the resonance response, the incident electromagnetic waves propagated along the $z$ direction, with the electric field oriented along the $y$ direction. While keeping other parameters constant, the period $g$ of the retained rectangle was gradually increased from 0 to 150 μm. The simulated transmission spectrum is shown in Fig. 2(a). To ensure the accuracy of the simulation, we chose a long time-domain signal length, but this also introduced echo signals. The overlay of echo signals and the original signal leads to oscillations in the transmission spectrum. Shortening the length of the time-domain signal can reduce this oscillation to some extent.

When $g=0\mathrm{\mu m}$, the all-silicon metasurface is in a completely symmetric state. At the location marked by the blue pentagram in the figure (0.348 THz), there exists an ideal BIC state with an infinitely high Q factor, considered to be a resonance with zero linewidth, which cannot be directly observed in the spectrum. This state can only be infinitely approached through simulation. As $g$ increases continuously, the overall C2 symmetry of the structure is broken, and resonant peaks with gradually increasing linewidth appear. This is because BIC couples with extended waves, generating channels for outward radiation, thereby converting the ideal BIC mode into a QBIC mode. Figure 2(b) provides a more intuitive illustration of the relationship between resonance linewidth and resonance frequency, indicating that by changing the asymmetry parameter, the resonance frequency and intensity of QBIC can be adjusted. The transmission spectrum at $g=90\mathrm{\mu m}$ is fitted using the typical Fano formula [43]$$T_{\mathrm{Fano}}(\omega )={\left|a_1+j{a}_2+\frac{b}{\omega -{\omega }_0+j\gamma }\right|}^2.$$

In the formula, $a_1$, $a_2$, and $b$ are constant real numbers, ${\omega }_0$ is the resonance frequency, $j$ denotes the imaginary unit, and $\gamma$ represents the radiation loss of the resonant mode, which is proportional to the linewidth of the resonance. For the asymmetric nonlinear Fano resonance, the Q factor can be calculated using $Q={\omega }_0/2\gamma$. The simulated transmission curve and the corresponding fitting results are shown in Fig. 2(c). The overall fitting is satisfactory, with a central frequency of 0.339 THz, and the calculated Q factor currently is 70.7.

The intuitive relationship among the Q factor, resonance frequency ${\omega }_0$, and the tunable parameter $g$ is shown in Fig. 2(d). It can be observed that with the increase of $g$, the Q factor sharply decreases, and simultaneously, the resonance frequency shifts towards the red direction along the black dashed line, consistent with the trend of transition from BIC to QBIC. Figure 2(e) illustrates the deeper relationship between the Q factor and the asymmetry parameter $\alpha$. Here, $\alpha =\mathrm{\Delta }S/S$ represents the asymmetry parameter, where $\mathrm{\Delta }S=\mathrm{\Delta }{g}^2$ is the area of the retained rectangle, and $S=P^2-\mathit{mn}$ is the area remaining after excluding the hollow rectangle in the initial BIC state. Through linear fitting, it can be found that the Q factor and the asymmetry parameter satisfy the relationship $Q\propto {\alpha }^{-2}$, meeting the basic conditions of symmetrically protected BIC.

To further investigate the impact of breaking structural symmetry, Fig. 3(a) shows the transmission spectrum from 0.33 to 0.36 THz at $g=0\mathrm{\mu m}$ and $g=60\mathrm{\mu m}$. When $g=0\mathrm{\mu m}$, the overall structure exhibits in-plane C2 symmetry, corresponding to an ideal BIC mode. When $g=60\mathrm{\mu m}$, the in-plane symmetry of the structure is broken, and the ideal BIC mode transitions to a QBIC mode, resulting in a sharp resonance peak at 0.344 THz. To investigate the physical mechanism behind the QBIC resonance in depth, Fig. 3(b) performs a multipole decomposition. Compared to the magnetic dipole (MD), magnetic quadrupole (MQ), toroidal dipole (TD), and electric dipole (ED), the electric quadrupole (EQ) exhibits the highest scattering power, indicating that the QBIC resonance excited at this frequency is EQ. In Fig. 3(c), we investigated the magnetic field distribution characteristics for different $g$ values. As $g$ increases from 30 to 120 μm, the asymmetric parameter $\alpha$ also increases, leading to a gradual enhancement of the leakage of QBIC resonance, ultimately weakening the magnetic field distribution in the $z$ direction and the electric field distribution in the $y$ direction.

Additionally, the impact of structural parameter variations on the transmission spectrum was investigated, as shown in Fig. 4. As shown in Fig. 4(a), with the gradual increase of period $P$, the resonance frequency exhibits a red shift. This is because the increase in $P$ enlarges the volume of the solid part of the metasurface, leading to an increase in the effective permittivity of the overall structure, thus lengthening the resonance wavelength and ultimately resulting in the red shift of the resonance frequency. In Fig. 4(b), as $P$ increases, the quality factor Q gradually increases. This change can be explained by the variation of the asymmetric parameter $\alpha$, as mentioned earlier. According to the previous discussion, $\alpha =\mathrm{\Delta }S/S=\mathrm{\Delta }{g}^2/(P^2-\mathit{mn})$; an increase in $P$ reduces the $\alpha$, and since Q is inversely proportional to $\alpha$, the Q value will gradually increase. In Figs. 4(c)–4(f), a similar analysis approach can be adopted. With the increase in the length $m$ and width $n$ of the rectangular aperture, the volume of the solid part of the metasurface decreases, while also causing an increase in the $\alpha$. This ultimately results in a blue shift of the resonance frequency and a gradual decrease in the Q value. As for the full width at half maximum (FWHM) in the figure, it exhibits a trend opposite to that of Q, which exactly corresponds to the relationship $Q=f_0/\mathrm{FWHM}$.

Considering the limited detection resolution of the current terahertz time-domain spectroscopy (TDS) systems, to ensure that the resonance peak can be fully scanned, the fabricated samples need to ensure a relatively high FWHM. Inevitably, the Q value will decrease accordingly. But too small Q value is also detrimental to practical applications. Therefore, after comprehensive analysis based on Figs. 2 and 4, the structural parameters of the actual fabricated samples were determined to be $g=90\mathrm{\mu m}$, $P=600\mathrm{\mu m}$, $m=420\mathrm{\mu m}$, and $n=340\mathrm{\mu m}$.

To test the sensing performance of the metasurface, we placed a layer of polyimide (PI) film with a thickness of 20 μm beneath the metasurface. The PI is lossy; its material type is selected as normal, with a permittivity of $3.5+0.0095i$ and a permeability of 1. The PI film exhibits excellent transmission in the terahertz frequency range, which means it barely absorbs or reflects terahertz incident waves. Additionally, in sensing applications, PI film demonstrates a good chemical stability and thermal stability, enabling it to cope with complex and variable practical application scenarios. Here, we use PI film as the substrate of the metasurface, allowing more analytes to enter the internal of the structure, enhancing sensing efficiency. A schematic diagram of the simulation model is shown in Fig. 5(a). Considering that in practical applications, the analyte tends to distribute on the surface of the metasurface, we placed 50 μm thick films on both the upper and lower surfaces. Considering that the refractive index of most biological molecules in nature falls within the range of 1–2, such as DNA, RNA, cells, proteins [44–46], we also increased the refractive index of the film from 1–2, as shown in Fig. 5(b). With the gradual increase of the refractive index, the center frequency of the resonance peak also undergoes a corresponding red shift. Figure 5(c) is presented to demonstrate the relationship more intuitively between the resonance peak linewidth and the refractive index variation. It is evident that the linewidth of the resonance peak remains essentially unchanged with the refractive index variation.

Sensitivity is an important indicator for measuring the sensing performance of the metasurface, and it can be expressed as $S=\mathrm{\Delta }f/\mathrm{\Delta }n$, where $\mathrm{\Delta }f$ is the change in the resonance peak center frequency and $\mathrm{\Delta }n$ is the change in the refractive index, with units of RIU. Figure 5(d) shows the corresponding linear fitting result, from which the sensitivity of the metasurface is extracted as $S=20\mathrm{GHz}/\mathrm{RIU}$, which is based on the slope of the fitting curve, indicating that the designed structure exhibits good sensing performance.

To further explore the sensing mechanism of the designed metasurface for Auramine O, transmission spectra were obtained by varying the thickness of the Auramine O film, with an extracted refractive index of 1.748 from experiments. The transmission spectra are shown in Fig. 6(a). When the thickness is 0 μm, which is equivalent to no Auramine O added, only a layer of PI film is present, resulting in a center frequency of 0.332 THz. By varying the thickness of the Auramine O film, the effects of Auramine O concentration variation in experiments can be simulated. It can be observed from the graph that with the increase in the thickness of the Auramine O film, the center frequency of the resonance peak shifts towards lower frequencies. The fitting results in Fig. 6(b) show that within the range of 0–50 μm, the resonance frequency shift is linearly related to the film thickness. Due to surface tension, the tested substances tend to adhere more to the upper surface. To better approximate the actual distribution, simulations were conducted based on the thickness ratio $r=t_{\mathrm{lower}}/t_{\mathrm{upper}}$. The abscissa represents the thickness of the upper layer film $t_{\mathrm{upper}}$ and the specific results are shown in Figs. 6(c)–6(e). It can be observed that with the increase in film thickness, resonance peaks in all three scenarios experience a red shift phenomenon. Moreover, as the value of $r$ increases, the black arrows rotate clockwise, indicating that resonance shift becomes more sensitive to thickness increase.

To detect Auramine O more accurately, it needs to be configured into a solution with a certain concentration, air-dried, and then placed in the TDS system for detection. This treatment method enables Auramine O to be more uniformly dispersed on the sensor surface, thereby improving the accuracy of sensing. Auramine O powder was purchased from Hefei Chisheng Biotechnology Co. We prepared multiple sets of Auramine O solutions with different concentration gradients. Enough solution was pipetted onto the center of the sample using a pipette gun to ensure that the coverage area of Auramine O was larger than the terahertz spot. Then, it was placed in a drying oven at 50°C for 25 min to wait for the solution to air-dry into a film. During this period, keeping the blower function of the drying oven one can ensure that the formed film is uniformly distributed on the metasurface. The detailed preprocessing process is shown in Fig. 7(a). Figure 7(b) shows the comparison between the simulated and experimental transmission curves of the metasurface sample without analyte. From the graph, the resonant frequency measured in the experiment is 0.331 THz, which is basically consistent with the simulation results. And the experimentally measured Q is 30.1. However, in the experiment, there is a certain attenuation in the right half of the resonance peak, which may be caused by differences in material parameters, limitations in the fabricating process, influences from the surrounding environment, and inherent errors in the testing system. Overall, the curve characteristics and trends of the resonance peak observed in the experiment are in good agreement with the simulation results. The transmittance spectra at different concentration gradients are shown in Fig. 7(c). With the increase of Auramine O solution concentration, the resonance peak undergoes a red shift, corresponding exactly to the trend of increasing film thickness in the simulation. An enlarged view of the resonance peak is provided in the inset. For ease of observation, we conducted regression analysis, as shown in Fig. 7(d). The result is consistent with the predicted trend in the simulation, and the lowest sensitivity to Auramine O was achieved at a concentration of 1 mg/mL.

In addition, to verify the stability of the metasurface sensor, we conducted reliability experiments under two indicators: temperature and humidity. The corresponding results are shown in Figs. 7(e)–7(h). For all experimental groups, 1 mg/mL Auramine O solution was used, and the remaining processes remained unchanged. During the collection process, for each set of results, the spectra were averaged by scanning three times. From Figs. 7(e) and 7(g), it can be observed that the characteristics of the transmission curves remain essentially unchanged, and there is no significant shift in the position of the resonance peak. Figures 7(f) and 7(h) show the statistical distribution of the transmission spectra, and the ranges of the transmission spectrum data are relatively close. Connecting the median lines of each group of data with red lines, the overall results are relatively stable. Through the series of experiments described above, it is demonstrated that the designed metasurface sensor exhibits good stability in both humidity and temperature, making it more adaptable to complex scenarios in practical applications. Table 1 compares the sensor proposed in this work with the reported terahertz metasurface sensors. Utilizing a hollowed all-dielectric structure for sensing is quite novel. Even in the low-frequency range, we still achieve a good sensing performance.Table 1.

Comparison with Reported Terahertz Metasurface Biochemical Sensors^a

We have designed an all-silicon metasurface based on the QBIC mechanism. The metasurface is composed of periodic arrays of hollow silicon, and by breaking the in-plane symmetry of the structure, a sharp QBIC mode resonance peak is generated at 0.33 THz. In this mode, the constrained electromagnetic energy can be effectively enhanced, thereby promoting the interaction between incident electromagnetic waves and the test substance. Through simulation, it is found that the resonant frequency and Q factor of the QBIC mode can be adjusted by modifying the structural parameters. Additionally, we selected a set of structural parameters, then fabricated and processed them into samples, and observed resonance peaks under the QBIC mode in experiments. The processed metasurface samples can also serve as a sensing platform for Auramine O, achieving Auramine O detection with a precision of 0.1 mg/mL, and passing reliability tests for humidity and temperature. The designed all-dielectric hollow metasurface, compared with other QBIC metasurfaces with complex structures and diverse materials, is convenient to manufacture and has low loss. It is not only suitable for the terahertz band but can also be extended and applied to the infrared band through structural scaling. Our work can provide new references and inspirations for the design and application of QBIC modes in the future.