Ultrasensitive, light-induced reversible multidimensional biosensing using THz metasurfaces hybridized with patterned graphene and perovskite

2.1 Design of the PGPP@MS biosensor

A schematic layout and all geometric parameters of the proposed PGPP@MS biosensor are depicted in Figure 1. Figure 1(a) shows an overview of the fabrication process for the PGPP@MS biosensor, which begins with a 5 μm-thick PI film spin-coated on a 300 μm-thick quartz substrate (1.0 cm × 1.0 cm) (Figure 1(a), ①). An array of 0.2 μm-thick aluminum microstructure units is fabricated atop the PI layer via standard photolithography methods (Figure 1(a), ②). The dimensions of the unit cell are shown in the photomicrographs in Figure 1(b)–(d). It consists of external U-shaped (parallel long sides (PLS) and vertical long side (VLS)) and internal asymmetric spiral structures (ASS).

In the next step, a 0.25 μm-thick MAPbI₃ film is spin-coated onto the metasurfaces, covering the microstructures (Figure 1(a), ③). This film shows good compactness and uniform morphology, with large grain sizes and no pinholes, as shown by the top-view SEM image in the inset of Figure 1(e). Figure 1(e) also shows the x-ray diffraction (XRD) pattern of the perovskite film. Diffraction peaks are located at 14.3°, 20.3°, 23.8°, 24.7°, 28.7°, and 32.1°, corresponding to the (110), (112), (210), (202), (220), and (310) crystal planes, respectively. No characteristic peaks associated with PbI₂ or other redundant phases were observed.

Next, a PI film was spin-coated atop the MAPbI₃ layer (Figure 1(a), ④), which protects and isolates the MAPbI₃. A 1.0 cm × 1.0 cm graphene layer was then transferred onto the top surface of the PI film (Figure 1(a), ⑤). Finally, for patterning of the graphene, zinc was sputtered onto the uppermost layer of graphene in 5 nm-thick stripes to remove a carbon layer selectively from those areas (Figure 1(a), ⑥). The device was placed in dilute HCl solution (about 0.05 M) for 1 min to dissolve the zinc. The device was then rinsed with water and dried in air. This series of steps removes one layer of carbon, and the procedure can be repeated to remove additional carbon layers. Ultimately, a graphene layer with a striped pattern is obtained. A photomicrograph of the striped graphene is shown in Figure 1(b) and as an inset to Figure 1(f).

Raman spectroscopy was performed at 514 nm excitation to confirm the quality of the graphene (Figure 1(f)). The ratio between the G peak (∼1578 cm⁻¹) and the G′ peak (∼2699 cm⁻¹) intensities is about 1.28. The full width at half maximum of the G′ peak is 54 cm⁻¹. These features indicate that the graphene is of good quality [35]. The conductivity of graphene is about ∼1.7 × 10⁶ S/m (square resistance ∼600 Ω).

The resonant features of the PGPP@MS sample were experimentally characterized with an 8f confocal terahertz time-domain spectroscopy system (see detail in Figures S1 and S2). A 405 nm/532 nm/808 nm laser with a 3 mm spot diameter was used as an optical pump (Figure 1(i)).

Whey protein was chosen as a probe analyte to verify the performance of the PGPP@MS biosensor (Figure 1(h) and (i)). We prepared five different CWP in suspension, C₁ = 6.25 ng/mL, C₂ = 2.52 μg/mL, C₃ = 46.8 μg/mL, C₄ = 316 μg/mL, and C₅ = 1.25 mg/mL. All the experimental data presented in this work are averaged after 3 tests.

2.2 Simulation of the PGPP@MS biosensor

We performed full-wave simulations with a frequency-domain solver based on finite-differential time-domain methods using CST Microwave Studio. The electric and magnetic boundary conditions were along the x and y directions (Figure 1(d)), respectively. The polarization direction of incident THz radiation was set along the z direction. In the simulation model, the quartz substrate was treated as a lossless dielectric with dielectric permittivity ε = 3.84 [36]. We used the classical Drude model to describe the conductivity of the graphene [37, 38]. The thickness of the graphene was set at 1 nm. The refractive index of polyimide was 3.1. The tangent loss was 0.05. The transmission T(ω) was defined as T(ω) = |E _s(ω)/E _r(ω)|², where E _s(ω) and E _r(ω) are the THz electric field amplitudes of the sample and reference, respectively, after fast Fourier transformation of the THz pulses.

3.1 Properties of the PGPP@MS biosensor

To investigate the underlying resonant mechanism of the PGPP@MS metasurfaces, the transmission spectra of the PGPP@MS biosensor were both simulated and measured, as shown in Figure 2(a). The experimental and simulated results agree well except for a slight difference in intensity at some resonance frequencies. This deviation may have been caused by fabrication errors or differences between the simulation conditions and the actual conditions. It is also interesting to note that a transmission peak with 60% efficiency appears at around 0.63 THz and 1.16 THz, as shown in Figure 2(a). This weak transparency phenomenon reasonably suggests that there may be an EIT-like mode-coupling effect in our metasurfaces [13, 39].

Figure 2:

Simulated electric field intensity distribution of the metasurfaces: (a) experimental (blue line) and simulated (red line) transmission spectra of the PGPP@MS biosensor sample. (b)–(d) Simulated electric field intensity distributions at 0.63 THz of the PLS (b), VLS (c), and ASS (d) arrays. (e) Simulated electric field intensity distribution of the EIT mode at 0.63 THz. (f) Simulated electric field intensity distribution of the EIT mode at 1.16 THz. The dashed circles highlight field intensity that is indicative of dark mode excitation. The entire incident light is x-polarized.

To further understand the physical mechanism of the EIT-like resonances, the electric field intensities of the PLS array, VLS array, and ASS array were simulated and shown in Figure 2(b)–(d), respectively. The field is mainly distributed at the four ends of the long sides of the PLS (Figure 2(b)) and localized around the starting and ending points of the ASS (Figure 2(d)). The VLS has no electric field distribution (Figure 2(c)). Therefore, the PLS acts as the bright mode, the VLS acts as the dark mode, and the ASS acts as the quasi-dark mode, simultaneously [40]. Combining the fields of the three coupled resonators results in two transmission windows around 0.63 THz and 1.16 THz, shown in Figure 2(a).

Figure 2(e) clearly shows that the electric field intensity in the PLS is entirely suppressed at these two frequencies as a result of their destructive interference. At 0.63 THz, the field intensity is mainly localized around the starting and ending points of the ASS, indicating the excitation of the quasi-dark mode. However, at 1.16 THz, the electric field intensity is mainly concentrated around the starting point, the right corner of the ASS, and two regions on the right side (black dotted circles in Figure 2(f)), suggesting that the dark mode is now excited. Therefore, our metamaterial can produce these two EIT-like modes under x-polarized incident light as a result of the destructive interference of the PLS (the bright mode), the ASS (the quasi-dark mode), and the VLS (the dark mode).

3.2 Sensing performance of the PGPP@MS biosensor

To investigate the sensing performance of the PGPP@MS biosensor, whey protein at various concentrations in aqueous solution was dropped onto the surface of the device, and the experimental test was performed after the water evaporation was completed.

Concentrations of 0 ng/mL (C₀), 6.25 ng/mL (C₁), 2.52 μg/mL (C₂), 46.8 μg/mL (C₃), 316 μg/mL (C₄), and 1.25 mg/mL (C₅) were tested. The transmission spectra of the PGPP@MS biosensor covered by different CWP are shown in Figure 3(a). The overall trend of the results is that transmission value decreases with increasing CWP.

Figure 3:

Experimentally measured and theoretically fitted for transmission spectra of the PGPP@MS biosensor at different CWP.

(a) Experimentally measured THz transmission spectra of the PGPP@MS biosensor at different CWP. C₀ = 0 ng/mL, C₁ = 6.25 ng/mL, C₂ = 2.52 μg/mL, C₃ = 46.8 μg/mL, C₄ = 316 μg/mL, and C₅ = 1.25 mg/mL). Peaks f ₁–f ₅ are indicated to illustrate the sensing effect. (b) and (c) The transmission curves obtained by theoretically fitted at each CWP for two different EIT-like frequency bands, 0.25–1.0 THz (b) and 0.62–1.58 THz (c). (d) The CWP dependence of the oscillator parameters γ ₁, γ ₂, δ, and κ, extracted by fitting the transmission curves in (b) with Eq. (2). (e) As in (d) for the transmission curves in (c).

Based on the THz transmission spectra of the PGPP@MS biosensor, we chose to examine shifts in the transmission spectrum at five frequencies, f ₁ = 0.42 THz, f ₂ = 0.81 THz, f ₃ = 1.16 THz, f ₄ = 1.39 THz, and f ₅ = 1.58 THz. Variations in both the frequency and the magnitude of the EIT peak occurred in the transmission spectrum as a function of CWP. At frequency points ƒ₃, ƒ₄, and ƒ₅, the changes in transmission and peak frequency shifts were relatively large. In particular, ultra-sensitive detection of a very low CWP of 6.25 ng/mL was possible. We compare the sensitivity and sensing type of previous work mainly based on the interaction of external molecules with graphene (see Table 1). WP molecules have no benzene ring like structures and without π-electrons. Therefore, it cannot form π–π stacking with graphene. First, we compared the case without π–π stacking, and from Table 1, it can find that the sensitivity of our designed sensor is two orders of magnitude higher than the other two sensors. And, our biosensor realizes three-dimensional sensing. Generally, if an external molecule has a benzene ring like structure with π-electrons, it strongly interacts with the π-electrons of graphene through π–π stacking [19]. Therefore, the graphene-based sensor show higher sensitivity to the external molecules containing π-electrons [19]. However, the sensitivity of our designed sensor to detect WP molecules without π-electrons is higher than that of detecting CMM which form π–π stacking with graphene in previous work. In summary, our designed sensor has higher sensitive sensing performance.

The effect of the analyte on the behavior of the PGPP@MS biosensor can be explained by the coupled harmonic oscillator model, which describes the near-field coupling between bright- and dark-mode resonators. The coupled differential equations can analytically describe the interference [43]:

Here, x ₁, x ₂, γ ₁, and γ ₂ are the resonant amplitudes and losses of the bright and dark modes, respectively. ω ₀ is the resonance frequency of the bright mode oscillator (γ ₂ « γ ₁ « ω ₀), δ denotes the detuning of the resonant frequency of the dark mode oscillator from the bright mode (δ « γ ₁), and κ is the coupling coefficient between the two oscillators. By solving Eq. (1) with the approximation ω − ω ₀ « ω ₀, the susceptibility χ is obtained:

where χ _r is the real part of the susceptibility, and χ _i is the imaginary part, which is proportional to the energy losses. The transmission T can be obtained through T = 1 − gχ _i [44], where g is the geometric parameter that indicates the coupling strength of the bright mode with the incident electric field E.

We defined two frequency bands, from 0.25 THz to 1 THz and from 0.62 THz to 1.58 THz, as two distinct regions of EIT-like effects. The measured transmission spectra of the PGPP@MS biosensor in these regions were theoretically fitted with Eq. (2), as shown in Figure 3(b) and (c). The fitting results show reasonable consistency with the experiments.

The fitting parameters as a function of the CWP are shown in Figure 3(d) and (e) for the lower- and higher-frequency bands, respectively. It can be observed from Figure 3(d) that δ, κ, and γ ₂ are almost constant in the low-frequency band, whereas γ ₁ changes slightly with increasing CWP. The radiative damping of the bright mode resonator, γ ₁, decreases to 4.1 THz at C₄, compared with 4.5 THz for a bare sensor.

In the high-frequency band (Figure 3(e)), δ is almost constant while γ ₁, γ ₂, and k change slightly with increasing CWP. The radiative damping of the bright mode resonator, γ ₁, varies for CWP between C₂ and C₅. The non-radiative damping of the dark mode resonator, γ ₂, changes from 3.8 THz at C₀ to 4.1 THz at C₅. At the same time, κ decreases from 5.2 to 5 THz² with increasing CWP. These variations in fitting parameters indicate that the physical properties of the analyte may affect the local field on the surface of the PGPP@MS biosensor, changing both the self- and mutual coupling of the bright and dark modes.

To characterize the sensing performance quantitatively, ΔV _T = V _T0 − V _Ti (V _T0 is transmission value at C₀, V _Ti is transmission value at C_i) and Δƒ_n = |f _ni − f _n0| (f _ni/f _n0 is for frequency resonances with and without protein) were calculated between the bare sample and each CWP from C₁ to C₅ at frequencies f ₁–f ₅. ΔV _T is shown in Figure 4(a), and Δƒ is shown in Figure 4(b).

Figure 4:

Quantitative analysis of multi-dimensional sensing performance.

(a) and (b) The transmission value difference ΔV _T (a) and frequency shift Δƒ (b) between the bare sensor and each CWP between C₁ and C₅ at frequencies f ₁–f ₅. (c) Phase difference ΔP between the bare sensor and each CWP tested. (d) Summary of the CWP dependence of the physical properties in (a)–(c). Blue axis and markers: The sum of transmission value difference ΔV _Ttotal for all frequency points. Purple axis and markers: total frequency shift Δƒ _total. Red axis and markers: fitted slopes of the phase difference, ΔP.

With increasing CWP, the value of ΔV _T at ƒ ₁ is almost unchanged, ΔV _T at the ƒ ₂ peak increases slightly, and ΔV _T at ƒ ₃–ƒ ₅ increases significantly. ΔV _T reaches a maximum value of 0.4 at ƒ ₄ for a CWP of 316 μg/mL (C₄).

We selected the total transmission change at all frequency points, ΔV _{T total}, for each CWP as the first biosensing index of the PGPP@MS platform. It is shown in Figure 4(d) (blue axis and markers). A significant change in ΔV _{T total} was observed with increasing CWP. When the concentration was 6.25 ng/mL (C₁), ΔV _{T total} was significantly different from the 0 to at C₀. ΔV _{T total} increases remarkably between concentrations C₁ and C₂ (2.52 μg/mL). When the concentration further increases to 316 μg/mL (C₄), ΔV _{T total} reaches about 82%.

The peaks and minima of the transmission spectrum also change with CWP. At the low end of the wave band, terahertz waves have low energy and are insensitive to environmental changes. Even at high CWP, almost no shift of ƒ ₁, ƒ₂, or ƒ ₃ takes place (Figure 4(b)). However, when the CWP is C₄, Δƒ ₄ increases, then reaches a maximum value of 96 GHz at C₅. Δƒ ₅ increases with increasing CWP to a maximum value of 224 GHz at C₄, then decreases.

The total frequency shift Δƒ _total of all frequency points is shown in Figure 4(d) (purple axes and markers) as the second biosensing index of the PGPP@MS platform. Δƒ _total increases with increasing CWP, reaching 350 GHz at 316 μg/mL whey protein (C₄). The result indicates that the PGPP@MS biosensor can detect frequency shifts as a useful sensing dimension.

The main sensing mechanism may be a change in the ultrasensitive photoconductivity properties of the MOSLS layer of the device based on the whey-protein-induced ED effect. This effect occurs when free electron/hole charges induced by electrostatic field excitations replace those normally supplied by a donor/acceptor dopant species. The distinct merit of ED is that the carrier concentration and polarity are tunable via external excitation.

Additionally, the phase change ΔP with and without whey protein was obtained for CWP ranging from C₁ to C₅, as shown in Figure 4(c). It is interesting to note that the dependence of phase difference on frequency is quasilinear. Consequently, the slope of the linear dependence can be considered a third biosensing index for PGPP@MS platforms. This is shown in Figure 4(d), where it can be seen that the change in slope is larger between lower concentrations. However, when the concentration grows from 46.8 μg/mL (C₃) to 316 μg/mL (C₄), the slope changes less rapidly. These results reveal that the phase is also altered after the introduction of protein, mainly because of the change in conductivity of perovskite. The maximum phase difference was 360° at 1.25 mg/mL (C₅) with a slope of ∼400. Additionally, to verify this explanation, similar devices without perovskite were synthesized and the phase was measured under the same conditions. (See the Supplementary information for details.) The phase difference is almost zero as the CWP increases. Therefore, our claim that the phase change of the PGPP@MS biosensor is mainly caused by perovskite has been verified.

These results show that the PGPP@MS biosensor can function as a multidimensional biosensor capable of detecting ΔV _T, Δƒ and ΔP in the presence of proteins.

3.3 Physical rationale for the sensing mechanism

To provide a physical rationale for the PGPP@MS device’s sensitivity, we propose a mechanism by which the ED effect [33] can be induced by protein molecules, illustrated in Figure 5. Without whey protein (Figure 5(a)), the energy band of perovskite is in its intrinsic state, and the initial Fermi level (E _F0) of the p-type graphene is in the valence band (Figure 5(c) and (e)) [45, 46]. Usually, protein aggregates have a positive net charge because amino acids exist in the form of ions in aqueous solution [47, 48]. When a positively charged protein is dropped onto the upper graphene surface as an analyte, its electrostatic field will cause charge accumulation in the graphene and perovskite layer (see Figure 5(b)). Consequently, the Fermi level shifts from E _F0 to E _F1 (compare Figure 5(c) and (d)), and the conductivity of graphene gradually increases, resulting in lower transmission, as in Figure 4(d). The electric field increases with CWP, increasing the strength of the ED response.

Figure 5:

Schematic of the proposed sensing mechanism of the PGPP@MS biosensor.

(a) and (b) Schematic diagram of the balanced state and charge distribution in PGPP@MS without (a) and with (b) whey protein. (c) and (d) Schematic diagrams of the Fermi level of graphene for PGPP@MS without (c) and with (d) whey protein. (e) and (f) Energy band of the perovskite layer for PGPP@MS without (e) and with (f) whey protein.

Simultaneously, the protein-induced ED effect may also cause changes in the conductivity of the perovskite layer by changing its charge polarity and carrier concentration; or in other words, by changing its Fermi level (see Figure 5(f)).

To understand the physical mechanism of the frequency shifts, we used modified perturbation theory to analyze the relationship between the Δƒ values and changes in the dielectric environment. The relative change in the angular frequency Δω can be calculated by [11, 49], [50], [51]:

where E _AM is the electric field in the original metasurface, and ε is the equivalent dielectric constant of PGPP@MS. Because the electric field decays exponentially along the direction normal to the metamaterial, the frequency shift calculated from Eq. (3) scales with the amount of analyte as follows:

where Δε is the difference in the equivalent dielectric constants with and without whey protein, h is the thickness of the monolayer of graphene (∼1 nm), and L is the penetration depth of the THz field. It can be concluded that the conductivity of graphene and perovskite changes because of the incorporation of protein, leading to an alteration of the dielectric environment of the metasurface and an increase in Δƒ with increasing CWP. Both the mobility and concentration of charge carriers in graphene will reach a maximum and saturate even if the CWP increases further, hence why Δƒ ₅ reached a maximum at C₄, then decreased.

3.4 Light-induced reversible sensing

Until now, reuse of, or reversible sensing by, THz metasurfaces biosensors has been seldom reported. However, by tuning the conductivity of the MOSLS with a laser, the transmission spectrum can in principle be restored to the state without protein. Our work provides an opportunity to achieve reversible sensing in the THz region. Because of the poor thermal stability of whey protein, laser illumination may degrade it, reducing the net charge and weakening the ED effect on the MOSLS.

To demonstrate reversible sensing, we measured the transmission value (V _T) of the biosensor at frequencies f ₁–f ₅ under illumination at wavelengths λ = 405 nm, 532 nm, and 808 nm and three different optical fluxes, F _{op(1, 2, 3)}. Pentagonal radar maps of the change in V _T are plotted in Figure 6. V _T is shown in gray for the bare sensor (C₀) and in red for the highest CWP tested, C₅, because it showed a clear response at all five frequencies (see Figure 4(a)). (The corresponding transmission spectra are shown in Figures S3–S5.) V _T after optical exposure is shown in yellow for 405 nm, green for 532 nm, or purple for 808 nm. As F _op increases (across rows in Figure 6), the optically induced V _T gradually increases from the initial value marked in red (corresponding to C₅) until it recovers to the value without protein, marked in gray.

Figure 6:

Description of reversible sensing.

(a)–(c) Schematic diagrams of the PGPP@MS biosensor under THz beams and optical pumping at 405 nm (a), 532 nm (b), and 808 nm (c). (a1)–(a3) Pentagon radar maps of the transmission value (V _T) at frequency peaks f ₁–f ₅ under different optical fluxes, F _op(i), at 405 nm. Every vertex in the radar maps represents V _T at a different frequency peak. Gray shapes represent V _T for the bare sensor, and red represents V _T when the protein concentration is C₅. V _T after 405 nm illumination is shown in yellow. i = 1: F _op(1)–λ_405 nm = 1.1 mW/cm². i = 2: F _op(2)–λ_405 nm = 11.5 mW/cm². i = 3: F _op(3)–λ _405 nm = 103.5 mW/cm². (b1)–(b3) Radar maps for V _T following 532 nm illumination (shown in green; other colors are as in (a)). i = 1: F _op(1)–λ _532 nm = 1.9 mW/cm². i = 2: F _op(2)–λ _532 nm = 42.4 mW/cm². i = 3: F _op(3)–λ _532 nm = 162.7 mW/cm². (c1)–(c3) Radar maps for V _T following 808 nm illumination (shown in purple; other colors are as in (a)). i = 1: F _op(1)–λ_808 nm = 2.2 mW/cm². i = 2: F _op(2)–λ _808nm = 28.3 mW/cm². i = 3: F _op(3)–λ _808 nm = 132.6 mW/cm².

At 405 nm and F _op(1) = 1.1 mW/cm², only the V _T of ƒ ₅ increases. The main reason is that the laser power is too low to significantly influence the conductivity of the MOSLS. The V _T of all peaks increases for F _op(2) = 11.5 mW/cm². At F _op(3) = 103.5 mW/cm², the transmission values of all peaks return almost completely to their original values. The V _T values behave similarly at 532 nm and 808 nm, indicating that protein-induced transmission changes are indeed reversible.

It is known that the absorption wavelength of perovskite is less than 780 nm, so while it may respond to light at 405 nm and 532 nm, it should not respond to illumination at 808 nm. However, the device shows responses to all three wavelengths, implying that graphene may be the main contributor to the conductivity of the MOSLS under optical excitation.

Theoretically, the conductivity of graphene (σ _Gr) is determined by the carrier densities, i.e., by the Fermi energy µ, and their effective temperature T [52]. Assuming that σ _Gr is a DC, low-signal-frequency conductivity determined by short-range scattering on defects and acoustic phonons, the scattering time τ = τ ₀ (T ₀/рʋw) ∝ 1/p [53], where τ ₀ is the characteristic short-range scattering time [54]. Then one can obtain the following formula:

Where σ 0 = e 2 T 0 τ 0 / π h 2 is the intrinsic conductivity under without any excitation conditions, T = T ₀ and µ = 0. At relatively weak irradiation, |µ| « T ₀, T, and Eq. (5) yields

Laser excitation can drive electrons from the photoexcited electron-hole pairs to accumulate in the graphene layer, leading to a shift of the Fermi level from E _F1 (Figure 5(d)) in the valence band toward the Dirac point caused by a reduction in μ. Meanwhile, the carriers inevitably cause a temperature increase. From Eq. (6), the conductivity of graphene decreases with enhanced optical pumping, and the transmission values of all the frequency peaks decrease. In summary, V _T increases after laser excitation, permitting reversible sensing.