1. Introduction

Active optical metasurfaces have demonstrated significant potential in dynamic, real-time, and reversible light manipulation, as evidenced by their use in a wide range of applications, such as dynamic holograms [1], beam steering [2], optical modulator [3], and biosensing [4]. For active metasurfaces, the structural transformation or changes in material properties constitute are two primary methods to manipulate the amplitude, phase, and polarization of light. The formal method to encode the propagation or field localization properties of the light by carefully changing the shapes, sizes and orientations of the structure [5–7]. However, for a host of more advanced applications, such as dynamic holographic displays [8,9], beam scanning [10–12] and on-chip optical communication systems [13,14], there is a greater need for the development of post-fabrication tunable metasurfaces. To enable post-fabrication tuning of metasurfaces, the integration of a static metasurface with an active material platform, whose refractive indices can be altered by external stimuli, such as electricity, heat and lasers, might be an efficient method of active light manipulation.

A variety of active materials, including liquid crystals [15,16], transparent conductive oxides (TCOs) [3,17], 2D materials [18,19], and semiconductor multi-quantum wells [20], have been explored for many optical and electronic applications. However, most of these active materials possess volatile properties, implying that their functionality changes upon the removal of external stimuli and necessitates continual energy input to maintain performance. Recently, phase change materials (PCMs) have shown great potential in dynamic spectral manipulation [21,22]. Consequently, some researches have focused on the incorporation of phase change materials (PCMs) into metasurfaces, attributed to their capacity for inducing near-unity refractive index changes throughout their volume. Among existing PCMs, the Ge₂Sb₂Te₅ (GST) has exhibited attractive intrinsic features including non-volatility, fast switching speed, high switching robustness, good thermal stability, and compatibility with CMOS processes, garnering significant interest in nonvolatile tunable optical devices.

The design of PCM-based active metasurfaces mainly adopt two structural configurations. One configuration is to combine GST with high refractive index nanodisks, and manipulate the magnetic dipole (MD) or electric dipole (ED) mode of the structure by changing the crystallization fractions of GST [23–26]. This type of metasurfaces can indeed achieve significant spectral frequency tuning, while the full width at half maximum (FWHM) of the resonance is limited by the significant radiation loss of dielectric nanodisks, resulting in quality factors (Q = λ₀ / FWHM) is less than 15. Moreover, integrating PCMs with resonant plasmonic nanostructures is another effective configuration for active metasurfaces [27–31]. This type of metasurfaces can significantly enhance the interaction between light and PCMs, thereby expanding the dynamic range of spectral wavelength shift, while the significant inherent Ohmic losses of metals limit the improvement of their Q factors. However, the realization of high-resolution on-chip spectrometers places higher demands on miniaturized tunable narrowband filters based on active metasurfaces, thus the further research is essential to achieve sharp absorption peaks. The bound state in the continuum (BIC) emerges as a promising candidate for realizing a narrow linewidth metasurfaces [32]. In BIC configurations, light of a specific wavelength becomes confined within the structure owing to the interference among one or more scattering paths, which cannot couple with the far-field and therefore has an infinite quality factor. However, it is extremely susceptible to external perturbations and by changing the structural parameters or excitation wave vectors the radiation coupling between light and far-field is achieved, the BICs degenerate into Quasi BICs (QBICs) [33]. The quality factor of QBIC can achieve modulation from infinity to a certain value. By combining PCM with structures based on BIC or QBIC modes, this will provide new opportunities for precise spectra control [34]. Therefore, we also use the QBIC to construct our active metasurfaces.

In this work, we propose an active metasurface using non-volatile GST to achieve near-perfect multi-resonant absorption, a multilevel tuning range, and narrowband performance in the infrared band. The metasurface possesses two magnetic modes in the near-field, and the strong interaction between the light and GST layer can generate a wider tuning range for the structure. Moreover, it features Friedrich-Wintgen (FW) QBIC, which guarantees the narrowband performance of the spectrum. To further explore the physical mechanisms of the structure, we numerically investigated the effects of various structural parameters on the spectrum. In addition, by combining the active metasurfaces (which serve as an excellent tunable filter) with compressive sensing algorithms, we have achieved accurate spectral reconstruction for various scenarios using only a single nanostructure. It realized an average fidelity rate exceeding 0.99 based on 51 measurements. And a spectral resolution of 0.5 nm at a center wavelength of 2.538 µm is predicted when the crystallization fractions of GST change from 0 to 20%. This work has significant potential for on-site matter inspection and point-of-care testing.

2. Phase-change metasurface and mechanism

Our reprogrammable phase-change plasmonic metasurface design comprises a periodic array of shallow metallic nanoantennas and a continuous metal sheet, separated by an ITO/GST multilayer stack, as shown in Fig. 1(a). The GST film can transition between the amorphous and crystalline phases under specific external excitation conditions, as depicted in the inset of Fig. 1(a). This characteristic exhibits a strong refractive index contrast but relatively low absorption losses in the infrared regime (Fig. S1a, Supporting Information), making it well-suited for efficiently manipulating the resonance frequency and intensity of metasurfaces. Furthermore, owing to the non-volatile nature of the GST, the intermediate crystallization states can be accessed and maintained (Fig. S1a, Supporting Information), endowing the active metasurface with multi-level manipulation and reprogrammable characteristics. The corresponding structural parameters are labeled in Fig. 1(b). P and w_g represent the period of the meta-unit and width of nanoantenna, respectively. The h_g, h₁ and h₂ represent height of the nanoantenna, the thickness of GST layer and ITO layer. The complex refractive index of gold is obtained from Johnson and Christy [35], and the complex refractive index of ITO is described by Drude model [17], as shown in Fig.S1 (b) (Supporting Information). Here, we set P = 1.35 µm, h_g = 30 nm, w_g = 0.8 µm, h₁= 20 nm, h₂ = 40 nm, respectively.

 

Fig. 1. (a) Schematic of the three-dimensional reprogrammable phase-change metasurfaces, which contain a periodic array of shallow metallic stripes and a continuous metal sheet separated by an ITO/GST multilayer stack. Insert: a scheme of atomic distribution of fully crystalline GST (cGST) and amorphous GST (αGST). (b) Cross sectional view (x-z plane) of the metasurface unit cell. (c) Simulated color-coded absorbance spectra of the active phase-change metasurface with different crystallization fractions of GST. (d) The central wavelength λ₀ of the resonance peaks and quality (Q) factors vary with different crystallization fractions of GST.

Download Full Size | PDF

Numerical simulation of the structure’s optical response was conducted using the finite-difference time-domain (FDTD) method [36]. When transverse-magnetic (TM) polarized light normally illuminates the structure, multi-resonances are excited, leading to multiple near-perfect absorption peaks in the spectra, as shown in Fig. 1(c). Owing to the variations in the refractive index of the GST layer under different crystallization fractions, the multiple resonance peaks of the metasurfaces can be simultaneously manipulated, realizing a reprogrammable multifunctional metasurfaces. As shown in Fig. 1(c), there are three significant resonance absorption peaks (marked as Peak 1, Peak 2 and Peak 3) within the wavelength range of 1 to 2 µm. As the crystallization fractions of the GST layer increase, the three resonance peaks undergo varying degrees of redshifts, and the Q factors of the absorption spectra correspondingly increase, as shown in Fig. 1(d). When the crystallization fraction exceeds 40%, the resonance peak gradually vanishes as the imaginary part of the refractive index increases in the crystalline state. Peak 1 experienced a shift of approximately 46 nm as the GST transitioned from an amorphous state to a 40% crystalline state. Peak 2 experienced a shift of approximately 123 nm, with the Q factor decreasing from 36 to 8 as GST transited from an amorphous state to a fully crystalline state. Peak 3 experienced a shift of approximately 200 nm, with the Q factor decreasing from 14 to 10 as GST transited from amorphous state to fully crystalline state. Consequently, by transforming the 20 nm-thick GST layer from an amorphous to a crystalline phase, the overall spectrum of the structure shifts around 369 nm, which is of great significance for high-performance active photoelectronic applications.

We observed that the peaks exhibit differential changes in wavelength shifts and FWHMs corresponding to the increase in the crystallization fraction of GST. As depicted in Fig. 2(a), the emergence of multiple resonance peaks in the PCM-metasurface indicates the presence of various near-field coupled optical modes within the system. To investigate the physical mechanism of the system, we simulated the normalized electric field distribution at the different resonance wavelengths when the GST was in its amorphous phase, as illustrated in Fig. 2(b-d). We can see from Fig. 2(b) and (d) that they exhibit similar electric field distributions of the resonant modes, which the light is intensely concentrated in the gaps between the two metallic layers, forming different numbers of ‘hotspots’ [37]. Further, we use the Fabry–Pérot resonance conditions to quantitatively explore this type of resonant characteristics of the metasurfaces, as follows [38]:

Here, λ₀ is the free space wavelength, φ is the reflection phase pickup for the gap plasmons reflecting from the cavity truncations. The n_eff represents the effective refractive (mode) index for the gap surface plasmon. The m is an integer denoting the order of resonance. According to this expression, Fig. 2(b) and (d) represent the first-order magnetic mode (marked as M1) and the second-order magnetic mode (marked as M2) of the structure, respectively. Because the electric field is primarily localized at the ITO layer and the two edges of metallic nanoantennas. Our structural configuration significantly enhances the interaction between light and the GST layer, and the n_eff also undergoes a significant change when GST layer is under different crystallization fractions, thereby expanding the wavelength range of spectral tuning.

 

Fig. 2. (a) the absorbance spectra of the active metasurface with αGST. The normalized electric field distribution magnitude at the different resonance wavelengths in the x–z plane of a meta-atom. (b) λ₀ = 1.175 µm, (c)1.419 µm and (d)1.736 µm, respectively.

Download Full Size | PDF

Furthermore, it is noteworthy that a relatively narrow resonance peak with a Q factor of up to 36 at the wavelength of 1.49 µm, which is not the typical characteristic of plasmonic resonance. Consequently, we also simulated the normalized electric field distribution at the wavelength of 1.49 µm, as shown in Fig. 2(c). This simulation demonstrates that the electric field is localized simultaneously in the gap between the two gold layers and on the top surface of the metallic nanoantennas. This phenomenon appears to result from the coupling of the second-order magnetic mode in the metal-isolator-metal (MIM) configuration with the first-order surface lattice resonance (SLR) of the nanoantennas. It is known that the Q factor is indicative of the energy loss of the structure during resonance and serves as a crucial parameter for quantifying the energy dissipation capacity of the structure. In this analysis, the losses of the structure are calculated using the coupled mode theory (CMT), and the single-port model is employed for the sufficient thickness of the bottom metal layer. The reflection coefficient of the structure is expressed as follows [37]:

Here, r represents the reflection coefficient of the structure; ω₀ denotes the resonance frequency; and τ_a and τ_γ represent the mode lifetimes attributable to the internal absorption of the structure and far-field radiation, respectively. The absorption and radiation losses of structures under different crystallization fractions are shown in Fig. 3. It is apparent that both the radiation and absorption loss of Peak 2 are both smaller than those of peak 3, indicating less energy dissipation, and therefore peak 2 is sharper. Additionally, when 1/τ_a equals 1/τ_γ, the structure attains perfect absorption, as the radiation and the absorption loss are close for both peaks, enabling near-perfect absorption to be achieved.

 

Fig. 3. Absorption and radiation losses of structures under different crystallization fractions of GST.

Download Full Size | PDF

Metasurfaces with narrowband tunable characteristics hold significant value in research fields such as high-resolution spectral sensing. To further investigate the narrowband resonance peak and understand the underlying physical mechanism, we simulated a series of reflection spectra of the metasurface when GST is under amorphous state. The reflection spectra with varying period, thickness of ITO h₂, width and height of nanoantenna are shown in Fig. 4. For detailed simulation, the initial parameters were set as follows: P= 1.35 µm, h_g = 30 nm, w_g = 0.8 µm, h₁= 20 nm, h₂= 40 nm, respectively. The reflection spectra were calculated for each varying parameter while keeping other parameters constant. As observed in Fig. 4(a), with an increase in the period, the first-order magnetic resonance mode remains almost unchanged. Similarly, the second-order magnetic resonance mode remains nearly constant at a fixed wavelength, but the absorption intensity gradually decreases with the increase of the period. More noteworthy is that when first-order SLR resonance mode approaches the second-order magnetic resonance mode, the M2 suddenly vanishes. When the first-order SLR resonance mode approaches both magnetic resonances modes, it leads to a pronounced degradation in absorption spectra. This means that by harnessing the destructive interference mechanism between two leaky resonances the novel Friedrich–Wintgen (FW) bound states in the continuum (BIC) [39,40] is obtained. In the system, one resonance mode transfers all its losses to another, thereby becoming a FW-BIC. However, in practice, even minor perturbations in the structural parameters can destroy the ideal lossless system, leading to BICs transition into Quasi-bound states in the continuums (QBICs). Thus, it can be inferred that the field distribution depicted in Fig. 2(c) corresponds to Quasi-bound states in the continuum (QBICs).

 

Fig. 4. (a) The simulated reflection spectra of the PCM metasurface with (a) the variable period ranging from 1 to 1.8 µm; (b) the variable thickness of ITO h₂ from 20 to 80 nm; (c) the variable width of nanoantenna ranging from 0.3 to 1.2 µm; (d) the variable height of nanoantenna ranging from 20 to 100 nm.

Download Full Size | PDF

Furthermore, to enhance the practical applicability of the structure, we investigated the impact of various structural parameters on the reflection spectrum. Figure 4(b) demonstrates that variation in the thickness of ITO layer significantly affects the magnetic resonance mode of the structure. As the thickness of the ITO layer increases, both magnetic resonance modes undergo significant blue shifts, while the peaks of FW-QBIC remain almost unchanged. This occurs primarily because the effective index of gap plasmon (n_eff) of the structure decreased, and since the resonance wavelength is directly proportional to the n_eff, it results in a blue shift of magnetic resonance wavelength. Figure 4(c) shows that variations in the width of the nanoantenna significantly impact the magnetic resonance modes of the structural, where the modes undergo a significant redshift, which matches the quantity relationship shown in Eq. (1). Additionally, it is important to note that whether the first-order SLR mode couples with the first-order or second-order magnetic mode, both of these situations will lead to the construction of BICs, manifested as the disappearance of resonance in the spectra. As expected, there is no observable local field enhancement at the BIC, as shown in Fig.S2 (Supporting Information). Finally, we investigated the effect of height variations on the structural reflection spectra of nanoantennas, as shown in Fig. 4(d). Although the height of the nanoantenna has little effect on the resonance peak position of the spectrum, it can affect the FWHM of the spectra.

In this section, we systematically studied the relationship between structural parameters and reflection spectra, thereby understanding the relationship of different optical modes of light in relation to structural parameters. This understanding enables more efficient design of corresponding structures tailored to specific spectral working range requirements. As depicted in Fig. S3 (c), the active metasurface operates effectively within the 2∼4.5 µm (Supporting Information). We can learn that the FWHM of QBIC is 42 nm at the wavelength of 2.6 µm, and it has a corresponding Q factor up to 62. This is largely attributed to the negligible imaginary part of the complex refractive index of GST in the amorphous phase within the wavelength range, as shown in Fig.S3 (a). The Q factor significantly exceeds that of traditional plasmonic active metasurfaces, making it crucial for high-precision applications.

3. Miniaturized spectroscopy based on active metasurfaces

In terms of portable precise matter inspection fields, spectroscopy is an effective method to recognize the spectral fingerprints of target molecules. It is well-known that the infrared band has plenty of spectral fingerprint information. However, current commercial infrared spectroscopy systems tend to be bulky, complex and expensive. Recently, the development of miniaturized or chip-scale spectrometers based on metasurfaces has been notable as they enable ultra-small device sizes, relatively simple optical paths, high accuracy, and integration with CMOS technology. To evaluate the performance of the spectrometer, the operating bandwidth and the spectral resolution are two critical parameters. And filters play an important role in spectrometers and have a significant impact on the critical parameters. Here, the reprogrammable metasurfaces we designed combines near-perfect light absorption, a wide spectral tuning range, and a narrow resonance peak, which can be regarded as an excellent tunable filter and serve as the one of the key components for a compact spectrometer. This helps to create a compact and cost-effective spectral sensing platform.

Traditional grating spectrometers measure different spectral components at relevant positions through photodiode line arrays. The number of units in these arrays and the grating constant essentially determine the spectrum’s resolution. However, our approach utilizes a phase-change metasurface, wherein the absorption spectrum is tunable across different crystallization fractions of GST. This method overcomes the limitations of spectral resolution imposed by the detector. Additionally, the metasurface exhibits three significant resonance peaks within the 1 to 2 µm wavelength range, effectively covering the entire operational range of the structure, as shown in Fig.S4.

Firstly, we investigated the spectral reconstruction performance of the proposed reprogrammable metasurface within the 1 to 2 µm wavelength range using 51 filters. The response generated by each modulation in Fig. 5(a) can be written as [41,42]:

 

Fig. 5. (a) Schematic of the wavelength-shift spectral analysis. (b-d) Reconstructed different types of target spectra (dotted lines) using our proposed active metasurface absorption filter responses, as shown in Fig. S4. The black lines represent the reference spectra.

Download Full Size | PDF

Then, spectral reconstruction algorithms are needed to extract original spectral information. The L₁ minimization algorithm is an important method to implement compressed sensing theory, by minimizing the L₁ norm of the signal, the target spectral signal can be efficiently recovered from a small number of measurements [43].

Here, the target spectrum (L) was sampled by several different crystallization fractions using the structure with a pre-calibrated response matrix network (A), which was finally solved using the L₁ algorithm.

Through the application of a reconstruction algorithm to extract spectral information, we enhance the accuracy of spectral reconstruction. As shown in Fig. 5(b-d), the reconstruction accurately locates the information of various spectra, while a little deviation exists in the spectral profile. To quantify these deviations, we employ fidelity for evaluation. The calculation of fidelity can be written as [44,45]:

Finally, in order to demonstrate the spectral resolution of this technique, reconstruction of a spectrum with two close narrowband peaks (Δλ=0.5 nm) is conducted with a single peak of our proposed structure, as shown in Fig. 6. The detailed structural parameters can be obtained from Supporting Information S3. Here, we calculated the absorption spectra of our metasurfaces under QBIC mode vary with the crystallization fractions of GST change from 0% to 20%, as shown in Fig. 6(a). Moreover, we employed the techniques described above for spectral reconstruction. It can be seen that the distinct dual-peak spectral profile is fully resolved by only 21 times measurement, in Fig. 6(b). And the predicted spectral resolution is comparable to the commercial spectrometers (such as Anritsu MS9740A). All these results confirm the decent performance of such a novel miniaturized spectroscopy technique based on active metasurfaces and show promising potential for applications in on-site matter inspection and POC diagnosis.

 

Fig. 6. (a) Spliced simulated absorption spectra of the active metasurfaces versus the crystallization fractions of GST. (b) Reconstructed spectrum of a narrowband target spectrum with the spectral responses in (a).

Download Full Size | PDF

4. Conclusion

To summarize, we have proposed a reprogrammable PCM metasurface based on FW-QBIC states and MD modes, which can achieve multilevel continuous manipulation of multiple resonance absorption in the infrared band by changing the crystallization fractions of GST. Benefiting from the significantly localized electromagnetic fields and large refractive index changes of GST, this structure achieves a 369 nm wavelength shift with only a 20 nm GST layer. Moreover, the FW-QBIC of the structure guarantees the narrowband performance of the spectrum. In addition, metasurfaces can serve as ideal tunable filters and, when combined with compressive sensing algorithms, we have achieved accurate spectral reconstruction for various scenarios. The average fidelity rate up to 0.99 only through 51 times measurements and a spectral resolution of 0.5 nm is also predicted. This work is a potential tool for complex on-site rapid detection and lightweight application scenarios.