TiO2-coated guided-mode resonance gratings for polarization-selective green light filtering Authors: Khizzra Aslam, Nouraiz Mushtaq, Yuwei Chai, Guoguo Kang Published: 2026 ABSTRACT Dielectric gratings with guided mode resonance (GMR) are integral to narrow-band filtering due to their high resonances, enhancing optical metrics like diffraction efficiency and sensitivity. Recent advancements have introduced polarization-sensitive GMR filters, enabling selective control of TE and TM mode resonances based on polarization. This work presents theoretical and experimental studies on a green filter using polarization-sensitive GMR gratings. These devices are pivotal in applications such as telecommunications, spectroscopy, and imaging. We demonstrate the fabrication of TiO2-coated one-dimensional (1D) GMR gratings with polarizing resonance characteristics at normal incidence. A trapezoidal-profile grating, created via holographic lithography, achieves excellent polarization filtering for green light. Additionally, TiO2-coated photoresist (PR) gratings on SiO2 substrates are fabricated using magnetron sputtering, showcasing high performance and potential for cost-effective mass production. Nanoimprinting these gratings provides a scalable, economical method for producing reproducible GMR filters.Supplementary InformationThe online version contains supplementary material available at 10.1186/s11671-026-04699-z. Introduction Over the past two decades, dielectric waveguide gratings with guided mode resonance (GMR) have garnered significant attention due to their ability to generate high-Q resonances, making them highly effective for narrow-band filtering applications [1]. Research efforts have focused on enhancing their optical performance, including improvements in diffraction efficiency, sideband suppression, passband frequency, and Q factor. These advancements have broadened the applications of GMR filters, ranging from laser systems to optical communications [2]. A major breakthrough in GMR technology is the development of polarization-sensitive GMR filters [3]. Unlike conventional GMR filters that interact uniformly with all incident light regardless of its polarization, these advanced filters are engineered to respond differently based on the polarization state of the incoming light [4–7]. This innovation has significantly expanded the versatility and functionality of GMR-based optical devices. This polarization sensitivity enables these filters to selectively control and manipulate the spectral properties of light based on its polarization state. By engineering distinct resonance characteristics for transverse electric (TE) and transverse magnetic (TM) modes, these filters provide an added layer of precision in optical filtering [8–10]. The fabrication of GMR filters typically involves high-precision lithography and advanced materials to create the periodic structures essential for guided mode resonance [11]. One-dimensional (1D) periodic gratings, in particular, exhibit stable polarization dependency, making them highly effective for polarizing filters. These filters can be tailored to selectively respond to the polarization of incident light, such as transverse electric (TE) or transverse magnetic (TM) modes. Polarization sensitivity is achieved by leveraging the anisotropic properties of the grating or waveguide materials or by designing the grating geometry to preferentially couple to specific polarizations [12, 13]. The increasing demand for device integration has driven a growing focus on actively tunable filters. Techniques such as modulating voltage to adjust the refractive index (RI) of the material or altering the polarization of incident light allow dynamic tuning of the filter’s functional wavelength [14, 15]. Additionally, researchers have introduced advanced methods for controlling guided-mode resonance configurations. These include designing gradient grating periods to create linear-variable filters, regulating the incident angle, adjusting the azimuthal angle of incident light, and utilizing dielectric-coated gratings [16–18]. These approaches expand the possibilities for designing adaptable filters based on GMR gratings. Dielectric-coated gratings, as classical grating patterns, have widespread applications in optical systems. However, studies have shown that the spectral characteristics of grating filters (GFs) can vary significantly with changes in the incident angle and polarization [19]. TiO2 was chosen for its high refractive index, low optical loss in the visible range, and excellent stability, enabling strong guided-mode resonance and high filter performance. It is also compatible with scalable deposition techniques, such as magnetron sputtering and nanoimprinting. However, the present work introduces a distinct design-fabrication combination. The proposed trapezoidal-profile 1D GMR grating achieves polarization-sensitive filtering at normal incidence by breaking TE/TM symmetry through geometric asymmetry-typically requiring oblique incidence in conventional structures. Integration of nanoimprinting with TiO2 coating via magnetron sputtering enables low-cost, large-area fabrication compared with electron-beam or interference lithography. The fabricated grating exhibits a Q-factor for green, blue, and red filters ranges from 142.75 to 451.64, 93.87–187.01, and 131.83–331.46, respectively, as the ambient refractive index (RI) increases from 1.00 to 1.10. The green filter displays a nearly linear Q-factor rise, indicating reduced energy loss. The corresponding FOM values are 301.11 for green and ~ 107.98 for red/blue, while sensitivities vary from 1.79 to 454.70 nm/RIU for green, confirming its higher precision and environmental stability. Overall, the proposed TiO2-coated trapezoidal GMR grating combines high optical performance with cost-efficient scalability, demonstrating strong potential for integrated photonic and display applications. Scheme of the polarizing GMR gratings The studied GMR gratings are schematically illustrated in Fig. 1. The GMR trapezoidal photoresist grating, coated with TiO2, is fabricated on a SiO2 substrate. This purpose of this scheme is to confirm the effect of polarizing filters by using the uncomplicated single-layer 1D GMR grating configuration, which has rarely been reported before. It also seeks to determine the reproducibility of the 1D polarizing GMR grating [20]. The grating is designed to operate in the visible region, specifically as a green filter. These filters utilize periodic structures to diffract light, with the grating’s spacing and structure determining which wavelengths and polarizations are affected. The resonance wavelength is produced around 549 nm for TM polarization and 578 nm for TE polarization. The vector of electric field for the incident light can be analogous with TE or vertical along TM to the grating positions. Fig. 1Schematic of the polarizing GMR grating filter Schematic of the polarizing GMR grating filter The RI of the fabricated coated film, which is delicate to the deposition technique, was studied with ellipsometry and tailored with the Sellmeier formula:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ n^{2} (\lambda ) = 1 + A\lambda ^{2} /\lambda ^{2} - B $$\end{document} where A = 4.316, B = 3.846* 104 nm2, and λ is the wavelength in vacuum (in nm). The RI of the substrate and gratings are comparatively stable and less diffusing. The small dispersion of the dielectric coating material was considered in our calculations, while its impact on resonance sites is marginal. GMR gratings, including Giant Reflection to Zero Order (GIRO) mirrors, function based on the interference of two propagating modes within the grating region for both TE and TM polarizations [21]. The resulting outcoupled wave depends on the phase difference between these interfering waves. In non-polarizing GMR grating configurations, modifying the structural parameters allows precise tuning of the dispersion relations for TE and TM modes, enhancing the guided modes within the grating layer. This adjustment can lead to a scenario wherever both polarizations exhibit the same constant spreading, occurring at the intersection point of the spreading graphs of TE and TM modes under normal incidence. In this specific geometry, grating configurations are used to selectively permit the propagation of individually the zeroth diffused diffraction order at resonance. The resonance wavelength λres is associated with the grating period d, allowing GMR to be achieved at any desirable wavelength by selecting an appropriate d < λres < nsub [9]. Thus, optimizing other structural parameters achieves the resonance effect, enabling excitation under normal incidence. Results and discussion Through an optimization process considering fabrication feasibility, the optimal geometrical parameters for the GMR gratings were established: grating height (h) = 200 nm, period (p) = 360 nm, photoresist duty cycle (f) = 0.65, and thickness (t) = 36 nm. Simulations were performed using the commercial software Lumerical FDTD. A normally incident plane wave polarized along the grating direction (y-axis) was applied, with periodic boundary conditions along the X-axis and perfectly matched layer (PML) boundary conditions along the Y-axis. The simulated transmittance spectra of the GMR gratings for various polarization states, shown in Fig. 2, demonstrate the polarizing resonance effect suitable for green filtering. Fig. 2Numerically calculated TE and TM transmittance T of the GMR gratings h = 200 nm, p = 360 nm, f = 0.65 and t = 36 nm Numerically calculated TE and TM transmittance T of the GMR gratings h = 200 nm, p = 360 nm, f = 0.65 and t = 36 nm The strong GMR resonance effect was further validated by analyzing the electric field distribution for TE and TM polarizations, as shown in Fig. 3. These figures illustrate the electric and magnetic field responses for both polarization states, highlighting the polarization-specific behavior of GMR gratings. This behavior is primarily attributed to the asymmetric arrangement of the grating structure. In the visible spectrum, narrow resonant peaks with low full-width at half-maximum (FWHM) values are observed, generated by both magnetic and electric dipoles [22]. For TE polarization, the resonance effect is significantly influenced by the alignment and intensity of the electric field (E), which is perpendicular to the plane of incidence. This alignment enhances the interaction between the electric field and the dipoles within the grating material, contributing substantially to the observed resonance effect. Structural parameters such as material composition, grating periodicity, and depth also play a crucial role in determining how effectively the dipoles interact and resonate with the incident TE- and TM-polarized light. These factors collectively optimize the grating’s polarization sensitivity and its filtering efficiency for applications like green light filtering [23–26]. Fig. 3The electric and magnetic responses of a TM mode b TE mode The electric and magnetic responses of a TM mode b TE mode We further investigated the impact of structural parameters on the green light filtering performance of the GMR gratings. The results, illustrated in Fig. 4, highlight the behavior of an effective green light filter, particularly around the 550 nm wavelength, corresponding to green light. The effects of varying key geometrical configurations, including the period (P), filling factor (F), incident angle, and polarization angle, were studied while keeping other parameters constant to optimize optical performance. As shown in Fig. 4a, altering the grating period (P) from 300 nm to 400 nm causes the resonance wavelength to redshift from 470 nm to 600 nm, with transmission intensities consistently exceeding 90%. Figure 4b demonstrates that adjusting the filling factor (F) from 0.6 to 0.78 results in a slight shift of the resonant peak from 545 nm to 552 nm and a reduction in transmission intensity. The increase in the filling factor enhances optical alignment and material coverage, improving the filter’s ability to isolate the green wavelength by efficiently absorbing or reflecting non-green wavelengths. These findings indicate that careful modification of geometrical parameters significantly influences the resonance conditions for 1D GMR gratings. This optimization promotes the overlapping interaction of electric and magnetic responses, yielding high resonance intensities within the visible spectrum, particularly in the green light range. Varying the thickness of the grating layer from 5 to 55 nm, as shown in Fig. 4c, significantly impacts the interference patterns, providing precise control over the intensity and spectral behavior of transmitted and reflected light. This control is essential for designing filters that can selectively enhance or suppress specific wavelengths, such as green light. By fine-tuning the constructive and destructive interference mechanisms, the thickness variation enables the optimization of spectral performance, ensuring the desired wavelength is effectively filtered while minimizing undesired spectral components. Figure 4e illustrates the effect of varying the polarization angle from 0 to 90 degrees. The resonance wavelength shifts and strengthens with increasing polarization angle, exhibiting a more pronounced resonance effect at larger angles. This behavior is attributed to the enhanced alignment of the electric field (E) with the grating structure as the divergence angle increases, resulting in a more efficient coupling of light into the GMR structure and amplifying the resonance effect. Consequently, the resonance wavelength undergoes both a spectral shift and an increase in resonance intensity, demonstrating its dependence on the polarization angle [27]. Similarly, Fig. 4d highlights the strong resonance effect observed at low incident angles, which is well-suited for green light filtering applications. By maximizing energy coupling and constructive interference at specific wavelengths, the filter achieves high selectivity and performance for green light. This makes it particularly effective for applications requiring precise spectral filtering or enhancement, ensuring optimal functionality within the target wavelength range. Fig. 4The characteristic behaviors with diverse geometrical configurations a varying period b changing the filling factor c changing the thickness d incident angle and e polarization angle The characteristic behaviors with diverse geometrical configurations a varying period b changing the filling factor c changing the thickness d incident angle and e polarization angle We selected a grating period of P = 300 nm for blue wavelengths and P = 430 nm. for red wavelengths, keeping other parameters constant. To evaluate their performance, we examined the geometrical effects on these two filters. The simulated transmittance spectra of the GMR gratings under various polarization states are shown in Figure S1a (SI). As the filling factor increases from 0.60 to 0.78 (Figure S1b, SI), the peak intensity decreases. Additionally, changing the grating thickness from 5 to 50 nm (Figure S1c, SI) in the 627–680 nm wavelength range significantly impacts the performance of the red-light filter. As the thickness increases, the interference patterns shift, altering the intensity and spectral behavior of the transmitted light. Interestingly, the polarization and incident angles (Figures S2e and S2d, SI) do not show any resonance effect in this case. At these angles, the interaction between the light and the filter is not optimized, leading to weaker resonant modes. This highlights the importance of careful tuning of these parameters to maximize the filter’s performance and resonance strength. The simulated transmittance spectra of the GMR gratings with various polarization states are shown in Figure S2a (SI). When the filling factor increases from 0.60 to 0.78, a red shift is observed in the wavelength range of 447–462 nm (Figure S2b, SI), accompanied by higher full-width at half-maximum (FWHM) values. This indicates that increasing the filling factor broadens the resonance peak. In Figure S2c (SI), the change in grating thickness from 5 to 50 nm in the 447–490 nm wavelength range significantly impacts the performance of the blue light filter. As the thickness increases, the interference patterns alter, affecting the filter’s ability to block or transmit specific wavelengths. This highlights the importance of precisely controlling the thickness to achieve the desired filtering effect. Figure S2d (SI) shows that a change in polarization angle results in a green shift in the resonance wavelength. The resonance effect becomes stronger as the polarization angle increases up to 90 degrees. Finally, in Figure S2e (SI), a strong resonance effect is observed when light strikes the filter at incident angles below 20 degrees. This emphasizes the importance of optimizing both polarization and incident angles for enhancing the filter’s performance [28–30]. We further explore the influence of varying the refractive index (RI), which has numerous sensing applications. The simulated devices were exposed to a normal background with the RI changing from 1.0 to 1.10. As shown in Fig. 5a, the resonance wavelength (green) exhibited a redshift as the RI increased from 1.0 to 1.10. This behavior indicates that the resonance wavelength is linearly dependent on the surrounding RI. To further analyze the filtering behavior, we calculated the Q-factor, sensitivity, and figure of merit (FOM). The Q-factor is defined as the ratio between the resonant wavelength shift (Δλ) and the full-width at half-maximum (FWHM) [31, 32]. It is directly related to the precision of the color filter. The shift in Δλ due to changes in the RI of the ambient medium is known as sensitivity. Higher Q-factors indicate greater precision of the green light filter in response to variations in the surrounding environment. As presented in Fig. 5b, the Q-factors for green, blue, and red-light filters vary within the ranges of 142.75 to 451.64, 93.87 to 187.01, and 131.83 to 331.46, respectively, as the RI changes from 1.0 to 1.10. Notably, the Q-factor for the green light filter shows a linear increase, which is larger than that for the red and blue light filters. This suggests that energy loss is reduced in the green light filter, enhancing its filtering efficiency. The FOM values also exhibit a linear dependence, as shown in Fig. 5c. For the red and blue wavelengths, the FOM remains relatively constant at 107.98, while the green light filter demonstrates a standard FOM value of 301.11, indicating excellent performance as the ambient RI varies from 1.0 to 1.10. Furthermore, the refractive index variation significantly affects the sensitivity of the blue light filter, as illustrated in Fig. 5d, where sensitivity changes from 1.79 nm/RIU to 266.19 nm/RIU. However, the sensitivity of the red and green light filters is only slightly influenced by changes in the RI. Specifically, the sensitivity of the red filter varies from 1.79 to 208.60 nm/RIU, while the green filter’s sensitivity ranges from 1.79 to 454.70 nm/RIU. Fig. 5a Resonances, b Q-factors, c FOMs e and sensitivities of green, blue, and red light to different n values, respectively a Resonances, b Q-factors, c FOMs e and sensitivities of green, blue, and red light to different n values, respectively Fabrication of the GMR gratings We utilized the magnetron sputtering technique to engineer the GMR grating. The fabrication process is schematically illustrated in Figure S3. The typical sequence of steps for the photolithography process using holographic lithography is as follows. First, a glass substrate was cleaned by washing with acetone and ethanol for 15 min under sonication to remove impurities from the surface. Afterward, the substrate was rinsed with deionized (DI) water in large amounts, and the residual moisture was blown away using an N2 gas gun. The cleaned substrate was then placed on a hot plate and heated at 100 °C for 30 min to activate the surface. Next, a photoresist (RZJ304-10) was diluted with RZR-3100 in a ratio of 1:1.6 and spin-coated onto the substrate at 3000 rpm for 45 s. This process created a uniform photoresist layer with a thickness of approximately 200 nm. The substrate with the photoresist was then heated at 100 °C for 1 min on a hot plate. Once the substrate cooled to ambient temperature, it was mounted onto the exposure stage. Before exposing the substrate, the energy of the exposed area was calculated using a power meter. A Lloyd’s mirror configuration was employed for fast exposure using a 50-mW helium–cadmium (He–Cd) continuous wave laser (λ = 325 nm) as the light source. Based on experimental experience, the optimal exposure dose for patterning the gratings was determined to be approximately 65 mJ/cm2. A shutter was used to accurately control the exposure time. The next critical step in the process was to adjust the grating profile by controlling the development time. The substrate was immersed in the development solution (NMD-3) for 35 s to achieve a trapezoidal grating profile, after which it was rinsed with DI water for 60 s to stop the development process. Finally, a 36 nm dielectric TiO2 layer was coated on the photoresist grating using magnetron sputtering technology. Afterward, the substrate was washed with DI water and dried under an N2 gas stream. The resulting GMR grating, shown in the scanning electron microscope (SEM) image in Fig. 6, was successfully fabricated. Fig. 6SEM micrographs of the synthesized TiO2 GMR grating on a SiO2 substrate SEM micrographs of the synthesized TiO2 GMR grating on a SiO2 substrate It is evident from Figure S3 that the synthesized GMR gratings exhibit trapezoidal shapes, which are advantageous in the grating fabrication process. Throughout the fabrication, careful consideration is given to the uncertainties associated with the refractive indices of the materials, particularly TiO2, as these factors significantly influence the resonance characteristics of the GMR gratings. To address this, our fabrication method is primarily guided by a theoretical framework to determine the optimized structural parameters that effectively enhance the GMR effect. This approach ensures that the fabricated gratings align with the desired performance characteristics, even in the presence of practical material uncertainties. By integrating theoretical predictions with careful fabrication processes, we can achieve reliable and precise results for the GMR gratings. Resonance properties of the synthesized GMR gratings The optical properties of the fabricated GMR gratings were examined using an adjustable angle spectroscopic ellipsometer (VASE) from J.A. Woollam Co. The ellipsometer was configured to analyze the spectral transmittance of the gratings at normal incidence for both TE and TM polarized light. The collimated beam had a spot diameter of 3 mm, which is sufficiently smaller than the grating area of 7 × 7 mm2. Measurements were performed with a scanning wavelength step of 0.2 nm over the range of 400 to 700 nm. The wavelength-dependent refractive index (n) and extinction coefficient (k) of the sputtered TiO2 layer were measured using spectroscopic ellipsometry across 450–700 nm. The data, summarized in (Table S1, SI), show a high refractive index and negligible absorption in the visible range, confirming the suitability of the TiO2 film for supporting high-Q guided-mode resonances. These measured nk values were directly used in the simulations to accurately model the optical response of the fabricated GMR gratings. The calculated zeroth-order transmittance spectra of the grating, shown in Fig. 2, highlight the polarizing GMR effect around a wavelength of 500 nm. A small deviation of 10 nm from the expected 500 nm resonance wavelength is observed, primarily due to slight variations in the applied grating shapes compared to the designed specifications. Although the resonance wavelength can be further fine-tuned by adjusting fabrication parameters, such as increasing the grating period, the current synthesis results already demonstrate a GMR effect close to the targeted wavelength, confirming the effectiveness of the proposed design [33]. Additionally, there is a notable agreement between the simulated and experimental results. Key resonance features, such as the line profiles, resonance linewidths, and diffraction characteristics, are successfully replicated in the experimental data. This validates the consistency of both the theoretical model and the synthesis method. The obtained outcomes provide an experimental confirmation of the polarization-sensitive behavior of the 1D GMR gratings under normal incidence, demonstrating the accuracy of the theoretical understanding and the effectiveness of the fabrication process [34]. Fig. 7Transmittance spectra of the grating filter under different polarization at normal incidence Transmittance spectra of the grating filter under different polarization at normal incidence Using an ellipsometer, we measured the transmission spectrum of the grating filter under different polarizations at normal incidence, as shown in Fig. 7. The results demonstrate that the fabricated grating is polarization-sensitive, with bare glass showing no resonance effects in either polarization mode. Since the glass substrate itself has a transparency of around 80%, the experimental results align well with the simulation data in terms of transmittance amplitude and resonant wavelength. However, surface roughness and coating unevenness lead to light scattering, which results in optical losses and broadens the FWHM of the resonance dips. Despite these challenges, the grating holds significant promise for practical applications, thanks to its simpler fabrication technique and lower manufacturing costs [35, 36]. Our ongoing efforts aim to improve resonance performance while maintaining the relatively simple geometry of the structure. 1D GMR gratings with polarization-sensitive resonance at normal incidence offer significant advantages for applications that require precise polarization control and spectral filtering. In telecommunications, these gratings enhance signal clarity by selectively isolating polarization modes, reducing interference and improving overall communication quality. Their fine spectral resolution makes them valuable in spectroscopy, enabling accurate measurements and analyses of materials. In imaging systems, these gratings can control polarization to enhance contrast and reduce glare, improving image quality. The TiO2-coated trapezoidal design allows for cost-effective and scalable production through nanoimprinting and photolithography, making them a practical solution for compact optical systems that demand high-performance filtering. With their high refractive index and stable resonance characteristics, these gratings are poised to serve as versatile, modern solutions in advanced filtering and sensing technologies. Table 1 shows the experimental and theoretical comparison of different matrices [37–40]. Table 1Experimental and theoretical comparison of different matricesMetricsExperimentSimulation% DifferenceResonant wavelength (λ_res (nm))545 ± 2538 ± 0.5+ 1.3%FWHM (nm)7.8 ± 0.56.9 ± 0.2+ 13.0%Peak transmission (Tpeak)0.62 ± 0.030.68 ± 0.02− 8.8%Q = λ_res / FWHM69.9 ± 4.578.0 ± 2.6− 10.4% Table 1Experimental and theoretical comparison of different matricesMetricsExperimentSimulation% DifferenceResonant wavelength (λ_res (nm))545 ± 2538 ± 0.5+ 1.3%FWHM (nm)7.8 ± 0.56.9 ± 0.2+ 13.0%Peak transmission (Tpeak)0.62 ± 0.030.68 ± 0.02− 8.8%Q = λ_res / FWHM69.9 ± 4.578.0 ± 2.6− 10.4% Experimental and theoretical comparison of different matrices The polarization performance of the TiO2-coated trapezoidal GMR grating was experimentally characterized using an ellipsometer by measuring the transmission spectra under transverse electric (TE) and transverse magnetic (TM) polarizations at normal incidence. The results confirm a clear polarization-dependent resonance, with the grating exhibiting a strong dip in one polarization while the orthogonal polarization remains largely unaffected, consistent with the simulated behavior. This demonstrates that the device functions as a polarization-sensitive green-light filter. The experimental transmission amplitudes, resonance wavelengths, and line shapes closely match the simulation results, indicating good agreement between the design and fabrication. From these data, the polarization-dependent performance metrics, such as extinction ratio (ER) and polarization-dependent loss (PDL), can be directly calculated for each resonance wavelength. The observed polarization selectivity arises from the asymmetric trapezoidal geometry of the grating, which induces anisotropic coupling of guided modes and confirms the effectiveness of the proposed design for integrated photonic filtering applications. Conclusion In this study, we successfully fabricated a one-dimensional (1D) polarization-sensitive GMR grating using a TiO2-coated trapezoidal photoresist structure on a SiO2 substrate. The grating was produced using a low-cost magnetron sputtering technique, eliminating the need for etching processes. This marks the first experimental realization of a 1D polarizing GMR grating, showcasing the potential for large-scale, cost-effective production of practical devices. The grating’s polarization-sensitive resonance enables precise spectral filtering and light manipulation, making it ideal for applications in telecommunications, imaging, and spectroscopy. Its simple fabrication method and scalability suggest that these devices can be efficiently produced for use in fields such as biosensing and fluorescence enhancement. The ability to adjust the resonance wavelength by modifying fabrication parameters further enhances the device’s versatility. With improved manufacturing consistency, these gratings have the potential to revolutionize optical technologies, offering high-performance and cost-effective solutions for various scientific and commercial applications. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1. Supplementary Material 1.