Nanophotonic biosensors have emerged as transformative tools in biomedical diagnostics, environmental monitoring and food safety, offering rapid, label-free and highly sensitive detection of a wide range of biomolecules. By manipulating light at the nanoscale, these devices can detect subtle changes in the local refractive index (RI) or the presence of specific analytes, enabling early disease diagnosis, real-time monitoring and point-of-care testing [1–3]. Traditionally, plasmonic biosensors based on noble metals such as gold and silver have been the cornerstone of nanophotonic sensing technologies [1]. Their strong localized surface plasmon resonances (LSPRs) provide significant electromagnetic field enhancement, which is valuable for sensitive detection of biomolecular interactions [4]. However, plasmonic sensors face fundamental limitations, including broad resonance linewidths and significant energy dissipation due to intrinsic metal losses, photothermal heating [5] and limited compatibility with standard semiconductor processing [6]. These factors restrict their sensitivity, spectral resolution and broader adoption in clinical and industrial settings [7].

To address these limitations, research has increasingly focused on high-refractive-index dielectric materials, particularly silicon (Si), as promising alternatives for next-generation refractometric biosensors. Silicon nanostructures support sharp and low-loss Mie-type resonances, which correspond to electric and magnetic multipolar modes generated by the resonant excitation of displacement currents within subwavelength dielectric particles. These resonances provide strong field confinement, high quality factors and low absorption losses, enabling improved spectral resolution and reduced background noise, significantly enhancing sensing performance that matches or exceeds conventional nanoplasmonic platforms [8–12]. For example, Bontempi et al demonstrated highly sensitive biosensors based on periodic arrays of Si nanodisks (SiNDs), fabricated using electron beam lithography (EBL) and reactive ion etching (RIE). Their devices achieved streptavidin detection limit around 100 pM in the near-infrared (NIR) range, outperforming plasmonic sensors by two orders of magnitude [13]. Similarly, Yavas et al developed on-chip biosensing platforms using SiND arrays, fabricated via EBL and integrated into microfluidic chips. These sensors, interrogated by transmission microscopy in the visible-NIR (VIS-NIR), enabled real-time resonance tracking and demonstrated clinically relevant sensitivity for prostate-specific antigen detection in human serum [14]. In a subsequent work, they provided a deeper understanding of the sensing mechanisms by disentangling the roles of the electric and magnetic dipole modes in the optical response. They further demonstrated that semi-random arrays of Si nanocylinders, fabricated through colloidal lithography, also support both resonances with high biomolecular sensitivity [15]. More recently, Jahani et al developed Si metasurfaces based on nanodisk arrays, also fabricated by EBL, operating in the visible to NIR range and enabling label-free biosensing with femtomolar detection limits and multiplexed analysis. Integration with microfluidic systems allowed real-time, high-sensitivity biomarker detection [16]. Taken together, these studies highlight that Si-based dielectric nanostructures, ranging from ordered to semi-random arrays, offer a versatile and highly sensitive platform for refractometric biosensing, potentially surpassing conventional plasmonic sensors [17].

However, despite these advances, the field still requires further developments to improve large-area cost-effective fabrication and fully exploit these low-loss, high-Q Mie resonances for enhanced sensing performance. Most reported devices are optimized for operation in the NIR or infrared spectral ranges, which may introduce absorption interferences in biological analysis [18]. Additionally, the fabrication of high-quality Si nanostructures often relies on expensive, low-throughput techniques such as EBL, limiting scalability and commercial viability [19]. Many of these systems also require bulky, laboratory-grade instrumentation such as high-end microscopes and NIR-IR spectrometers, making them impractical for integration into compact, portable and a user-friendly affordable device [20].

In this work, we report the design, optimization and fabrication of cost-effective all-dielectric nanoresonators tailored for high-resolution biosensing in the visible spectral range. The sensor consists of a semi-ordered array of SiNDs patterned on a glass substrate, exhibiting narrow-band Mie resonances with refractometric sensitivity. Guided by computational modelling, we employed colloidal lithography methods to optimize geometric parameters governing resonance quality and sensor sensitivity. These techniques use self-assembled colloidal particles as nanoscale masks or templates, enabling large-area, cost-effective fabrication of periodic nanostructures with controlled geometries. The sensors are integrated into a compact prism-coupling platform with user-friendly operation, incorporating a semi-automated microfluidic module and software-controlled readout for real-time monitoring of wavelength shifts (figure 1). As a proof of concept, we further demonstrate the stable and efficient biofunctionalization of the SiNDs with supported lipid bilayers (SLBs), creating biomimetic interfaces suitable for on-chip investigations of cell communication and dynamic biological processes. SLB coatings have been recently introduced on gold-based plasmonic sensors to mimic tumor cell membranes and study immune-tumor cell interactions, showing great potential for evaluating advanced cell-directed cancer therapies [21, 22]. However, photothermal heating inherent to plasmonic substrates has been shown to compromise cell viability during prolonged measurements. In contrast, all-dielectric nanostructures, characterized by negligible photothermal effects, provide a more biocompatible environment that preserves the structural integrity of lipid membranes and living cells during optical interrogation. Our SLB-functionalized dielectric sensors could thus enable label-free, non-invasive and real-time monitoring of biological activity, offering a promising route for exploring complex cellular processes under physiologically relevant conditions. Overall, this work outlines a practical pathway toward scalable, visible-range dielectric biosensors, bridging the gap between fundamental nanophotonic research and real-world bioanalytical applications.

Figure 1. Schematic representation of the silicon nanodisk (SiND) sensor device. Polarized broadband light is coupled into the sensor through a prism at a 66° incidence angle and the transmitted signal is collected by a mini-spectrometer. The left panel shows a representative resonance spectrum, where a change in the surrounding refractive index induces a shift in the wavelength position. The right panel displays a real-time sensorgram corresponding to the dynamic wavelength shift during refractive index variations.

Download figure:

Standard image High-resolution image 2.1. Chemicals

For hole-mask lithography (HCL) nanofabrication, poly(diallyldimethylammonium chloride) (PDDA) was acquired from Merck (Steinheim, Germany) and sulfate latex beads (8% w/v, 0.3 μm) from Thermo Fisher Scientific (Paisley, United Kingdom). Polycrystalline silicon (poly-Si)-coated fused silica wafers (6-inch diameter, 500 ± 20 μm thickness, double-side polished, Ra < 1 nm) were acquired from Siegert Wafer GmbH (Aachen, Germany). The wafers were diced into 25 × 25 mm2 squares, followed by sputter deposition of a 110 nm ± 5% n-type poly-Si film on the front surface of each substrate. Common solvents, like isopropanol, acetone, ethanol and hydrochloric acid (HCl, 37%) were purchased from Panreac AppliChem (Barcelona, Spain). Standard chemicals including gold etchant (potassium monoiodide, KI) and sodium hydroxide (NaOH), reagents for buffer preparation (10 mM PBS (phosphate buffer saline, pH 7.4)) and MES (2-(N-morpholino) ethane sulfonic acid, pH 5.0)), reagents for carboxylic groups activation (1-ethyl-3(3-dimethylaminopropyl) carbodiimide hydrochloride (EDC)) and sulfo-N-hydroxysuccinimide (s-NHS), and for deactivation ethanolamine (EA) were purchased to Sigma-Aldrich/Merck (Steinhem, Germany). Lipids (POPC, 1-palmitoyl-2-oleoyl-snglycero-3-phosphocholine) and DOPS (1,2-dioleoyl-sn-glycero-3-phospho-L-serine) and Biot-PE (1,2-dioleoyl-sn-glycero-3-phosphoethanolamine-N-biotinyl) (sodium salt) were provided by Avanti Polar Lipids (Alabaster, Alabama, USA) through Merck.

2.1.1. Computational modelling

Computational modelling of SiND arrays on a glass substrate were performed using finite element method (FEM) in COMSOL Multiphysics. The analysis concentrated on near-field electromagnetic distributions and resonance wavelengths. The computational model consisted of a single unit cell of the SiNDs array, constructed in three dimensions and subjected to periodic boundary conditions to emulate an infinite array. The SiND was positioned on a glass substrate, with water (RI = 1.33) serving as the cladding medium. Geometric parameters, including nanodisk diameter (d) ranging from 100 to 300 nm, height (h) from 90 to 150 nm, gap (g) from 150 nm to 600 nm and angle of incidence between 64° and 70°, were systematically varied. The glass substrate RI was determined using the Sellmeier dispersion relation [23] accounting for wavelength-dependent variations. Silicon RI values were interpolated from literature [24] ensuring fidelity across the investigated wavelength range. A polarized light source ranging from 400 to 900 nm was introduced from below the substrate at an incident angle of 66°, incorporating both transverse magnetic (TM) and transverse electric (TE) polarizations to comprehensively characterize the sensor optical response. The optical response was collected from the top of the cladding layer. To accurately resolve fine-scale light-matter interactions, a non-uniform mesh with element sizes ranging from 3 to 100 nm was employed. Electromagnetic field distributions were analyzed to elucidate multi-slice polarization effects, as illustrated in supplementary information (SI) figure S1.

2.2. Fabrication of SiNDs sensors

SiND sensor arrays were fabricated on 25 × 25 mm2 glass substrates, pre-coated with a 110 nm layer of poly-Si. The substrates were sequentially cleaned by rinsing with 0.5 wt% SDS, 0.1 M HCl and Milli-Q water, then dried under a nitrogen stream. For further decontamination, the substrates were sonicated for 1 min each in acetone, isopropanol and Milli-Q water at 65 °C, followed by a final rinse and nitrogen drying. A bi-layer metal mask (2 nm titanium and 50 nm gold) was then deposited via electron beam evaporation. The metal-coated substrates were treated with oxygen plasma (18 s, 75 W, 75 mTorr, 40% O₂) to increase surface hydrophilicity. A 0.2 wt% solution of PDDA was applied for 1 min to promote surface charge, then thoroughly rinsed and dried. Colloidal polystyrene beads (60–300 nm diameter, 0.2%–0.6% concentration) were dispersed on the surface for 1 min as an etch mask. RIE was carried out for 2 min and 30 s in argon (40 sccm; ICP power: 250 W, HF: 40 W; 6 mTorr) to transfer the pattern to the metal layers. Beads were removed by tape stripping and the exposed silicon was etched with a C₄F₈/SF₆ (30/10 sccm) mixture for 75–250 s, adjusting rates (60–195 nm min−1) for target nanodisks geometry. Gold lift-off was performed by immersion in either piranha solution or potassium iodide (KI)-solution for 2–20 min. Final sonication at 25 °C–65 °C for 2–20 min ensured complete removal of residual gold.

2.3. Scanning electron microscopy (SEM) analysis

The morphology and spatial distribution of the fabricated silicon nanodisk arrays were examined using field emission scanning electron microscopy (FEI-SEM Magellan 400L, Oxford Instruments). SEM images were captured at various magnifications (1 µm–500 nm) to assess the uniformity, surface coverage and structural reliability of the nanodisks across the entire substrate. Quantitative analysis of nanodisk dimensions and gap distances was conducted using ImageJ software (National Institutes of Health, USA). Nanodisk heights were extracted from SEM images acquired at a fixed tilt angle by measuring the projected vertical dimension, with the reported uncertainty corresponding to the repeatability obtained from multiple measurements under identical imaging conditions. The inter-disk gap (g) was defined was defined as the minimum edge-to-edge distance between a nanodisk and its nearest neighbor, measured along the line connecting their centers, which represents the physically relevant spacing for near-field and optical coupling. Measurements were carried out on at least three SEM images acquired at different locations across the chip to ensure statistical representativeness.

2.4. Optical platform

The optical platform employed in this study is a self-designed, compact instrument (20 × 20 cm), previously employed as SPR and LSPR biosensor system [22, 25–28]. A collimated, TM-polarized polychromatic light beam is directed onto the sensor surface, with an adjustable incident angle ranging from 45° to 90° used for optimizing resonance conditions during sensor device characterization. The sensor chips were mounted between a trapezoidal prism (RI = 1.52) and immersion oil (RI = 1.51). Sample introduction was managed via a microfluidic chamber featuring a fluidic channel approximately 10 mm in length (15 mm including the o-ring seal), 2–3 mm in width, and 0.2 mm in depth, controlled by a syringe pump and injection loop to ensure steady and reproducible analyte delivery to the sensor surface. The reflected light from the sensor was collected using a fiber-coupled mini-spectrometer (Flame S 350–1050 nm, Ocean Optics), configured with a 25 μm entrance slit and grating #4 providing a spectral resolution of approximately 1.5 nm (full width half maximum, FWHM) over the 350–1050 nm range, while real-time monitoring of resonance shifts was performed using custom Python-based software with polynomial fitting algorithms. The central wavelength of the resonance dip was extracted and tracked by applying a high-order polynomial filt to the reflectance spectrum around the dip, enabling accurate determination of the resonance position and its spectral shifts. For each measurement, 300 reflectivity spectra were acquired with an integration time of 3 ms per spectrum, optimizing both the signal-to-noise ratio and temporal resolution. The intensity values shown in the spectra correspond to the raw photon counts recorded by the detector, labeled as Intensity (a.u.) because they represent arbitrary, spectrometer-dependent units rather than absolute radiometric values (see figure 1, left panel). No normalization to unity is performed; the only preprocessing consists of dark-spectra subtraction, where a spectrum acquired with the light source turned off is subtracted from each measurement to remove electronic noise and background signal. Beyond this correction, all spectra are presented in their unnormalized form to preserve the absolute resonance depth and contrast, which facilitates accurate tracking of the resonance position. Resonance shifts, defined as the displacement of the spectral dip (Δλ, nm), are used as the primary metric for evaluating the sensor performance. Ultimately, the sensor platform is designed to operate in a fixed-angle configuration, offering a compact, robust and affordable solution suitable for practical real-world applications.

2.5. Bulk sensitivity assessment

To assess the bulk sensitivity of the silicon nanophotonic sensors, HCl solutions at concentrations ranging from 25 mM to 500 mM were injected onto the sensor surface. Each injection was performed at a controlled flow rate of 30 μl min−1, with a total injection volume of 150 μl per concentration.

2.6. Sensor biofunctionalization

SLBs were formed on SiNDs sensor surfaces via rupture and fusion of small unilamellar vesicles (SUVs). Protocols for lipid preparation followed manufacturer guidelines, and protocols for sensor functionalization were carried out according to previously optimized procedures [21, 25]. Before mounting the sensors onto the optical platform, sensor chips were thoroughly cleaned using sequential rinses with 0.5% (w/v) SDS solution, 0.1 M HCl and Milli-Q water. Subsequently, the sensor chips underwent sonication in acetone, isopropanol and Milli-Q water for 1 min at 65 °C to remove any residual organic contaminants. Two distinct lipid compositions were employed to form (i) biotinylated SLBs and (ii) multifunctional SLBs incorporating both biotinylated and carboxylic (–COOH) functional groups. For the biotinylated-SLB, a lipid mixture of POPC and Biot-PE at a 10:1 molar ratio was used. To obtain the multifunctional SLB, a ternary lipid mixture of POPC, DOPS and Biot-PE at a 10:1:1 molar ratio was prepared. In each case, lipids were first dissolved in chloroform to yield a total lipid mass of 1 mg. The solvent was evaporated under a gentle nitrogen stream followed by vacuum drying for at least 2 h to remove residual solvent, forming a uniform lipid film. The dried films were rehydrated with 1 ml of PBS (10 mM, pH 7.4) to a final concentration of 1 mg ml−1. The resulting lipid suspensions were vortexed and sonicated for 1 h at room temperature to produce heterogeneous vesicle dispersions. SUVs of uniform size were then obtained by extruding the suspensions 100 times through a 0.1 μm polycarbonate membrane using a mini-extruder maintained at 60 °C. The freshly prepared SUV solutions were immediately introduced onto the sensor surface to induce vesicle rupture and lipid bilayer formation. This process yielded stable and homogeneous carboxylic and multifunctional SLBs directly on the sensor surface. To improve bilayer uniformity and remove weakly adsorbed vesicles, the surface was rinsed with 10 mM HCl. Bilayer integrity and reproducibility were subsequently assessed by treating the functionalized surface with 0.1% SDS to disrupt the lipid bilayer and regenerate the sensor surface for subsequent functionalization cycles.

After stabilization of the functional SLBs, receptor molecules were immobilized onto the respective bilayer surfaces. Biotinylated SLB was immobilized with Biotin-cell receptors (30 μg ml−1 in 50 Mm PBS buffer, pH 6.5), while multifunctional SLBs were co-immobilized with biotin-cell receptors (30 μg ml−1 in 50 mM PBS buffer, pH 6.5) and COOH-cell receptors (50 μg ml−1 in 50 Mm MES buffer, pH 5.5). Receptor immobilization was achieved using EDC/NHS (0.4 M/0.1 M) chemistry in MES buffer (50 mM, pH 5.5). To quench residual active esters and block unreacted carboxyl groups, 1 M ethanolamine (pH 8.0) was injected for 1 min. All sensing measurements were performed using PBS (10 mM, pH 7.4) as the running buffer at a constant flow rate of 20 μl min−1.

2.7. Data analysis

All nanophotonic sensor data were processed and analyzed using Origin 8.0 (OriginLab, MA, USA) and GraphPad Prism (GraphPad Software, CA, USA). Calibration curve was generated by plotting the mean resonance wavelength shift (Δλ) and corresponding standard deviation (SD) against changes in RI (Δn), with linear regression applied for curve fitting. The experimental limit of detection (LOD) was determined as three times the SD of the background signal, using water as the running buffer. Specifically, the sensor baseline was measured under water flow to calculate the mean and SD of the background, yielding a noise value around 10−4 nm. This noise value was then interpolated into the linear regression equation of the calibration curve to obtain the LOD in terms of RI change. Measurement precision was assessed by calculating the coefficient of variation (CV%), defined as the ratio of the SD to the mean value.

3.1. Computational design of SiNDs sensor

Systematic electromagnetic simulations were conducted using the FEM in COMSOL Multiphysics, considering a single SiND unit cell under periodic boundary conditions. The optimization focused on three key geometric parameters: nanodisk diameter, height and gap, each affecting the optical response of the nanophotonic sensor. Detailed results of parameter sweep and spectral analyses are provided in the supplementary information (section S1). The simulations revealed that under TM-polarized excitation, the SiND array supports two distinct resonance peaks within the visible range, with a prominent one, centered at 685 nm, exhibiting narrow linewidth (FWHM = 8 nm). In contrast, TE polarization produced a single, broader resonance at 665 nm (FWHM = 22 nm). This demonstrates that, according to recent reports, TM polarization enables sharper and more confined resonant modes in the SiND configuration (figure S1(c)). Further parametric tuning identified a nanodisk diameter of 300 nm as optimal, producing the strongest and narrowest resonance features due to enhanced electromagnetic confinement and well-separated higher-order modes (figure S1(d)) [29]. Variation of the nanodisk height in the range 90–150 nm indicated that 110 nm yields the most intense and sharp resonance peak (FWHM = 14 nm) around 622 nm, making it particularly suitable for biosensing applications (figure S1(e)). The interparticle gap was found to play a critical role: a gap of 150 nm (g = d/2) facilitated strong near-field coupling, resulting in the narrowest resonance peak (FWHM = 5 nm) near 720 nm, whereas larger gaps reduced coupling and broadened resonances (figure S1(f)). The angle of incidence was also varied from 64° to 77°, showing that θ =  66° resulted in a sharper resonance (figure S1(g)). With these optimized geometries (d = 300 nm, h = 110 nm, g = 150 nm and θ = 66°), we evaluated the refractometric sensitivity of the resonance in response to subtle RI changes of the aqueous cladding from 1.331 to 1.334 (figures S1(h) and (i)). Results predicted a bulk sensitivity of approximately 200 nm RIU−1 (figure S1(j)), which would be in the range of conventional nanoplasmonic sensors (100–300 nm RIU−1) [30] and other Si-based nanostructured sensors (70–263 nm RIU)−1 [31].

3.2. Nanofabrication and characterization of SiNDs sensors

To fabricate the SiND sensor arrays, we adapted and further optimized a previously developed protocol based on hole-mask colloidal lithography (HCL) (figure 2(a), details in SI–S2). HCL is a versatile and cost-effective nanofabrication method that enables large-area patterning of short-range ordered nanostructures without the need for complex lithographic equipment. In our approach, poly-Si substrates were coated with a 2 nm Ti/50 nm Au bilayer and treated with oxygen plasma to enhance surface hydrophilicity. PDDA and sulfate latex beads (60–300 nm, 0.2%–0.6%) were then deposited to form a short-range ordered colloidal mask. The Au layer was etched by Ar-based RIE, and the beads were subsequently removed by gentle tape stripping, leaving a nanohole pattern that served as an etch mask for the Si layer. A thorough and systematic optimization of the underlying poly-Si etching process parameters was carried out to precisely control the final nanodisk geometry. In particular, the C₄H₈/SF₆ plasma etching rate (60–190 nm min−1) and etching duration (75–250 s) were iteratively tuned to define the poly-SiND height and sidewall profile with high morphological fidelity (figures S2 and S3). Finally, a KI-based lift-off assisted by mild sonication ensured complete removal of the residual Au without damaging the nanostructures.

Figure 2. (a) Schematic illustration of SiNDs arrays fabrication via hole-mask-colloidal lithography (HCL) process. Sulfate latex beads are self-assembled on a polycrystalline silicon (poly-Si)-coated glass substrate, followed by titanium/gold deposition to define the mask. Subsequent bead removal exposes the gold mask openings, which are transferred into the silicon layer by reactive ion etching. A final KI-based lift-off step removes the residual gold, yielding clean SiND arrays; (b) and (c) SEM images of a short-range ordered nanodisks array obtained under optimized conditions (d = 313 ± 15 nm, h = 108.5 ± 2.5 nm, g = 156.2 ± 8.5 nm), confirming well-defined geometry and random distribution across the substrate.

Download figure:

Standard image High-resolution image

Target geometries (d = 300 nm, h = 110 nm, g = 150 nm) were achieved using a 0.4% colloid concentration and optimized plasma conditions corresponding to an etch rate of 150 nm min−1 for 200 s. Morphological analysis by SEM confirmed the formation of well-defined, short-range ordered SiND arrays (figures 2(b) and (c)). Across multiple fabrication batches (⩾4 batches, 4 chips each), the resulting nanodisks exhibited a mean diameter of 313 ± 15 nm, a height of 108.5 ± 2.5 nm, and an edge-to-edge gap of 156 ± 8.5 nm, closely matching the target values. The measured diameter uniformity (CV = 8.5%) and consistent height demonstrate the high reproducibility achieved through process refinement. Over-etching tests revealed edge degradation and loss of definition, underscoring the importance of precise temporal control and parameter stability to preserve nanodisk morphology (see S3). Overall, this extensive optimization effort established a refined HCL approach that enables accurate, reproducible, and scalable fabrication of SiND arrays with tunable geometries. The process allows independent adjustment of key parameters while maintaining high uniformity across large areas, offering a robust and cost-effective alternative to conventional e-beam lithography [21]. The resulting residue-free SiND structures provide a reliable platform for next-generation label-free nanophotonic biosensors.

3.3. Optical spectroscopy characterization and refractometric sensing evaluation

Following the comprehensive optimization and structural evaluation of the SiNDs arrays, we systematically investigated their optical properties to elucidate resonance and sensing behavior. We employed a custom-built, compact optical setup (20 × 20 cm footprint) for prism-coupling excitation of electromagnetic resonances. The system, originally developed for SPR and LSPR sensing applications, integrates readily available optomechanical components and low-cost instrumentation, offering an affordable and portable alternative to conventional benchtop and microscopy setups. A collimated, polarized white light beam is directed onto the sensor surface through a prism coupling configuration, with the incident angle adjustable between 45° and 90° to optimize the resonance condition. Reflected light is collected by an optical fiber and analyzed with a compact customized spectrometer, while data acquisition and real-time resonance tracking are performed using custom Python-based software. The platform also supports flow-cell operation through a small-volume microfluidic chamber enabling stable and reproducible sample delivery.

In this platform, SiND sensor chips were mounted for direct angle-resolved spectroscopy, enabling systematic characterization under both TM and TE polarized light at incidence angles between 64° and 70°, according to simulation predictions (figure S1(g)). In both polarizations, a pronounced resonance peak was observed in the 700–730 nm wavelength range, with peak intensity and spectra definition improving at higher incidence angles (figure S4). The optimal response was obtained at 66° under TM-polarized illumination, with a resonance centered around 720 nm and a FWHM of 93 nm (figure 3(a)). The experimentally observed resonance linewidth is broader than that predicted by simulations (FWHM 8 nm), which assume perfectly periodic and uniform arrays. Such differences are consistent with observations in plasmonic and photonic nanostructures where experimental linewidths often exceed simulation prediction due to additional loss channels and structural variations [7, 32]. The broadening can be therefore attributed to the intrinsic random spatial distribution introduced during the HCL fabrication process. Nevertheless, the measured resonance position and polarization dependence are in good agreement with FEM simulations, which identified TM polarization as optimal for sharp and well-confined resonance modes (figure S1(c)). The experimental resonance at 720 nm closely corresponds to the simulated mode arising from strong near-field coupling at an interparticle gap of 150 nm, confirming that the optimized geometrical parameters effectively translate into the expected optical behavior despite short-range order and fabrication-induced deviations. Furthermore, despite these irregularity effects, the measured FWHM remains among the narrowest reported for colloidal lithography-fabricated SiNDs arrays and is consistent with previous demonstrations of high quality resonances with linewidths 100 nm [33].

Figure 3. (a) Representative resonance spectra acquired under optimized TM configuration at a 66° incident angle, displaying resonance wavelength position and full width at half maximum (FWHM) values obtained directly from the device. (b) Bulk sensitivity calibration curve showing the linear correlation between resonance wavelength shift (Δλ) and refractive index change (Δn). Each data point represents the mean ± standard deviation from triplicate measurements, highlighting the excellent reproducibility and precision. The calculated noise level (0.0023) was determined as the standard deviation from measurements under constant refractive index conditions. (c) Real-time sensograms showing resonance intensity changes of the silicon nanophotonic sensor during sequential injections of different HCl concentrations (25 mM to 500 mM) over a Milli-Q water running flow. Arrows indicate sample entry into the sensing chamber and sample exit through running water washing.

Download figure:

Standard image High-resolution image

To evaluate the bulk refractometric performance of the SiNDs sensors, we monitored the resonance wavelength shift (Δλ, nm) in response to controlled changes in the RI (Δn, RIU) of the surrounding medium (figure 3(b)). Sequential injections of HCl solutions at different concentrations were introduced into the microfluidic system, and the sensor response was recorded in real time, producing dynamic sensorgrams that reflect the evolution of the resonance peak with each injection (figure 3(c)). From these measurements, a calibration curve was obtained by plotting the steady-state resonance wavelength shift against the corresponding RI change (figure 3(b)). Linear regression of this curve yielded a bulk sensitivity (SB) of 543 nm RIU−1. Using the noise level of the system, we further determined a high-resolution LOD of 4.04 × 10−6 RIU. This experimental sensitivity significantly exceeds the 200 nm RIU−1 predicted by FEM simulations for the optimized unit cell, highlighting the contribution of collective effects and the large-area short-range ordering in fabricated arrays. The higher-than-predicted sensitivity may be attributed to enhanced near-field interactions between neighboring nanostructures, including intra-array coupling and the possible formation of small clusters, which are not fully captured in simulations of an isolated unit cell. Such interactions can produce slightly detuned local resonances and hybridized modes, effectively increasing the overlap of the electromagnetic field with the surrounding medium and, consequently, the bulk refractive-index response. Similar effects have been observed in plasmonic and photonic systems, where short-range order, semi-random arrangements, or disordered aggregates have been reported to increase sensitivity compared to perfectly periodic arrays [34–36]. These observations suggest that the spatially heterogeneous nature of the fabricated SiND arrays contributes to the enhanced refractometric performance measured experimentally, beyond what is predicted for an idealized, perfectly periodic structure. We note, however, that these explanations remain hypotheses that would require detailed simulations or further modeling to be fully confirmed.

Finally, we compared the SiND sensor performance with our previously optimized plasmonic gold nanodisk (AuND) sensors (d = 100 nm, h = 20 nm), fabricated using a similar HCL protocol and tested under identical conditions in the same optical platform (λ = 680 nm, FWHM = 67 nm, SB = 183 nm RIU−1, LOD = 1.3 × 10−⁵ RIU) [21], (figure S5). Despite exhibiting a slightly broader resonance bandwidth, the SiND sensor developed in this study exhibits substantially higher bulk sensitivity and a better detection limit. This improvement may arise from silicon high RI and low optical losses, which enhance light confinement within the nanodisks and increase the interaction of the resonant modes with the surrounding medium, producing larger resonance shifts for a given RI change. Relative to previously reported dielectric nanoresonator sensors, our SiND platform demonstrates superior or comparable sensing performance. For example, random arrays of Si nanocylinders with a height of 130 nm and diameter of 280 nm have achieved a bulk sensitivity of only 86 nm RIU−1 [15], while periodic arrays of 50 nm-tall SiNDs have reached 227 nm RIU−1 [14]. Notably, the sensitivity of our sensor (543 nm RIU−1) lies at the upper end of the range reported for Si-based nanophotonic sensors, which typically spans from 86 to 657 nm RIU−1 [37]. A summary of the sensitivity, fabrication methods and operational ranges of representative silicon-based sensors is provided in table 1. Furthermore, most prior systems operate in the NIR region, are fabricated using EBL and require complex instrumentation such as supercontinuum lasers and NIR spectrometers for characterization. In contrast, the SiNDs sensor presented here are specifically optimized for visible range operation and are compatible with compact, portable laboratory device that integrate microfluidics and real-time monitoring software. This design eliminates the complexity and high costs associated with NIR-based systems and EBL fabrication, offering a more cost-effective and scalable solution. The combination of high sensitivity, operational simplicity and scalability makes our high-RI nanophotonic sensor a promising platform and a potential candidate for biological sensing applications.

Table 1. State-of-the-art comparison of Si-based nanoresonator sensors.

Sensor structuresDimensions (nm)Fabrication methodOperation rangeSensitivity (nm RIU−1)ReferencesNanodisksh = 110 nm d = 300 nmHCLVIS543This workPillarsh = 320 nmd = 100–140 nmEBLNIR413[38]Nanodisksh = 50 nm d = 280 nmEBLVIS-NIR227[14]Nanocylindersh = 130 nm d = 280 nmHCLIR86[15]Nanoantennah = 190 nm d = 330 nmEBLIR—[30]3.4. Biomimetic functionalization for on-chip cell studies

To validate the proposed sensor device for biological analysis, we carried out a series of experiments aimed at establishing the formation, stability, and functionality of artificial cell membranes on the SiND sensors as a foundation for on-chip cell studies. These membranes, formed by SLBs decorated with specific cell receptors, provide fluid, biomimetic interfaces to better reproduce native cell membrane environments, enabling the study of dynamic and collective biological processes directly on the chip. We anticipate that silicon sustained hydrophilicity and low optical losses could enable more stable SLB assembly and improved biocompatibility relative to metallic plasmonic substrates, where localized heating may compromise the integrity of lipid membranes and the viability of biological entities.

To evaluate the suitability of the SiND sensors for biomimetic surface functionalization, we first investigated the formation of SLBs via vesicle fusion. An illustrative diagram of the biofunctionalization is provided in figure S6. SUVs composed of a lipid mixture (POPC:DOPS (10:1)) were prepared by extrusion and introduced on the SiND sensor surface. These vesicles spontaneously ruptured and fused, yielding stable carboxyl-functionalized SLB, evidenced by a significant resonance wavelength shift (Δλ ≈ 12 nm) and subsequent introduction of mild HCl solution (figure 4). This cleaning step indicates that the lipid membranes are resilient to pH changes and confirm the formation of a continuous bilayer rather than loosely adsorbed vesicles. To further evaluate surface robustness and reusability, the bilayers were disrupted with 0.1% SDS and subsequently reassembled in flow, without removing the sensor chip from the platform. This cycle could be repeated multiple times with highly reproducible SLB formation, emphasizing the persistent hydrophilicity of Si surfaces as a key advantage of dielectric sensors and enabling long-term, cyclic biofunctionalization.

Figure 4. Schematic and real-time SPR sensogram illustrating the cyclic formation and regeneration of a supported lipid bilayer (SLB) on silicon nanodisk (SiND) sensor. SLB membranes were assembled via disruption of small unilamellar vesicles (SUVs), followed by repeated sensor regeneration steps using 10 mM HCl cleaning and 0.1% SDS washing protocols, demonstrating reproducible SLB formation and robust sensor surface recovery.

Download figure:

Standard image High-resolution image

Following the demonstration of stable and regenerable SLBs, we next evaluated their functionalization with cell receptors using two lipid formulations, highlighting both selective and multi-targeted immobilization capabilities (figure 5). Biotin-functional lipid mixtures (POPC:Biot-PE, 10:1) formed stable biotinylated SLBs, producing an equilibrium resonance shift of Δλ ≈ 14 nm. The bilayer remained stable during NaOH cleaning, confirming robust membrane integrity (figure 5(a)). Neutravidin (50 μg ml−1) was then introduced, yielding a Δλ ≈ 7.6 nm shift, followed by NaOH cleaning to confirm stability (figure 5(b)). Subsequent injection of a model biotinylated cell receptor (angiotensin-converting enzyme 2, ACE2, 30 μg ml−1) induced an additional Δλ ≈ 3.4 nm shift, with only a small Δλ decrease (∼−0.9 nm) after final washing, indicating slight removal of loosely adsorbed material. These results prove that SLB-functionalized SiND sensors can reliably support selective and stable receptor immobilization, providing a biomimetic sensor for on-chip cell studies.

Figure 5. (a) Schematic illustration and corresponding real-time sensograms showing the formation of a supported lipid bilayer (SLB) with biotin functionalities (POPC:Biot-PE) on a SiND surface via vesicle disruption (middle panel), followed by sequential immobilization of neutravidin and biotinylated receptor, with NaOH cleaning confirming stability (right panel). (b) Schematic illustration and sensograms showing the assembly of a complex SLB (POPC:DOPS:Biot-PE, 10:1:1) on the SiND surface (middle panel), followed by the immobilization of two different receptors: biotinylated and non-conjugated cell receptors, using neutravidin capture and EDC/NHS chemistry, respectively (right panel).

Download figure:

Standard image High-resolution image

Nonetheless, to closely mimic native cell membranes, the incorporation of a single receptors is often insufficient. We therefore evaluated the platform for multi-targeted functionalization through site-specific co-immobilization of multiple ligands. Multifunctional SLBs were formed from POPC:DOPS:Biot-PE (10:1:1) lipid mixtures, combining carboxyl and biotin moieties. SLB formation generated resonance shifts comparable to single-component bilayers (Δλ ≈ 14 nm, figure 5(c)). Sequential receptor immobilization was then performed: biotinylated cell receptors were first attached via strong biotin–neutravidin interactions (Δλ ≈ 3.4 nm), followed by covalent coupling of unconjugated receptors via carbodiimide chemistry between COOH functional groups and primary amines (NH2) of the proteins (Δλ ≈ 8 nm), as shown in the sensorgram (figure 5(d)). Cleaning and inactivation steps using NaOH and ethanolamine effectively stabilized the bilayer architecture. These proof-of-concept experiments demonstrate that multifunctional SLB-coated SIND sensors can support site-selective integration of multiple receptors, establishing a versatile platform for advanced biosensing and for investigating complex biological interactions, such as cell–cell communication.

We have developed a cost-effective all-dielectric nanoresonator sensor platform optimized for visible-light operation, combining advantages over both dielectric and plasmonic biosensors. Using scalable HCL, we reproducibly fabricated short-range ordered SiND arrays and integrated them into a compact optical platform for user-friendly operation. The sensors exhibit sharp Mie-type resonances centered around 720 nm, with high refractometric sensitivity (543 nm RIU−1) and excellent detection limits in the 10−6 RIU range. These features position our SiND sensors as promising platform for sensitive, label-free detection in biomedical and environmental applications.

The intrinsic properties of SiNDs, including superior spectral resolution and minimal plasmonic losses compared to gold nanostructures, make them particularly suitable for complex biological studies where sensitivity and efficient heat dissipation are critical. Their hydrophilic surfaces further facilitate advanced biomimetic functionalization, allowing the formation of stable lipid membranes that serve as robust cell-mimetic interfaces. We demonstrated that these lipid bilayers can be repeatedly assembled, disrupted, and reassembled in flow, highlighting the reusability and operational stability of the SiND sensors. Furthermore, multifunctional SLBs allow site-selective co-immobilization of multiple ligands, enabling multi-targeted sensing on a single, biomimetic device.

Future work will focus on systematic evaluation of its biosensing performance and integration with advanced microfluidics to enable real-time, on-chip live cell analysis. Altogether, these results demonstrate that dielectric nanophotonic sensors provide a versatile and robust platform for advanced studies of complex biological processes, with potential applications ranging from fundamental investigations of cell communication to the design and development of personalized cell-based therapies.

The research work has been funded through the PRIME project (PID2023-151524OB-I00), funded by MICIU/AEI/ 10.13039/501100011033 and by ERDF/EU. MS acknowledges the economic support of the Ramon y Cajal Grant (RYC2020-029015-I) funded by the AEI/MCIN and the European Union NextGenerationEU/PRTR. The ICN2 is supported by the Severo Ochoa Centres of Excellence Programme, Grant CEX2021-001214-S, funded by MCIN/AEI/10.13039.501100011033. ICN2 is funded by the CERCA programme (Generalitat de Catalunya).

The datasets generated and/or analyzed during the current study are available from the corresponding author on reasonable request.

All data that support the findings of this study are included within the article (and any supplementary files).

The authors have no conflicts to disclose.