Digital nanophotonic biosensing empowered by silicon Mie voids

Abstract

Optical biosensors are indispensable in medical and environmental diagnostics, yet existing approaches are fundamentally limited in their sensitivity due to ensemble-averaged measurements. Digital biosensing has emerged as a promising solution for resolving individual binding events, thereby providing signals at very low analyte concentrations down to the single-molecule level. Here, we present a novel concept for digital optical biosensing empowered by dielectric Mie voids, combining nanoparticle-based contrast enhancement and deep learning for ultrasensitive biomarker detection. The resonantly trapped light in the air cavities of the periodic Mie void arrays ensures strong overlap between the near-fields and the single gold nanoparticles that are captured on the surface in the presence of the protein biomarker. Remarkably, this strong interaction creates high-contrast digital signals for the precise counting of single nanoparticles located both within and outside the voids, yielding efficient use of the entire sensor area for high sensitivity. We employ deep-ultraviolet (DUV) lithography for the scalable and low-cost production of Mie voids in silicon wafers and automated image analysis with a convolutional neural network for robust nanoparticle counting. As a proof of our concept, we demonstrate the detection of an important disease biomarker, interleukin-6 (IL-6), from small sample volumes at concentrations as low as 1.84 pg/ml, within the physiological range of healthy individuals. Owing to its scalability, precision, and adaptability, our digital nanophotonic biosensing approach based on silicon Mie voids establishes a versatile route for applications ranging from bioanalytics to health and environmental monitoring.

Figures (5)

• Figure 1: CNN-enhanced digital biosensing with dielectric Mie voids. a) Artistic view of a silicon Mie void sensor with bound gold nanoparticles (NPs). The biomarker molecule (blue) is 'sandwiched' between the detection (attached to the NP) and capture (attached to the sensor surface) antibodies, ensuring selective immobilization of the nanoparticle on the sensor. b) Simulation of the change in the scattering properties of a silicon Mie void upon binding of a gold NP. A decrease in the scattering cross section of the void leads to a detectable signal change in reflection for detecting and counting individual NPs. c) Top image: Photograph of a 4-inch silicon wafer containing 24 Mie void chips fabricated using DUV lithography; bottom left image: photograph of a single chip with Mie void arrays obtained by dicing the wafer; bottom right image: SEM image of Mie voids with bound gold NPs. d) Illustration of the CNN aided digital NP counting scheme for biomarker detection: incident light reflects from a Mie void array with Au NPs, forming an optical image on the CCD camera. Nanoparticles induce local perturbations in reflectivity, creating brighter (for those located inside ("In") the voids) or darker (for those located outside ("Out") of the voids) spots compared to the background. The resulting image is post-processed using a pre-trained CNN model to count the total number of binding events. e) Calibration curve showing the total number of detected NPs for different biomarker concentrations.
• Figure 2: Simulated reflectance spectra and fabricated structures. a) Linearly polarized light reflectance spectra of the silicon Mie void array normalized to the reflectance from a bare silicon substrate. The reflectance signal increases when a gold NP is located inside ("In") of the void and depends on the NP position (at the centre or near the void edge). The reflectance signal decreases when a gold NP is outside ("Out") of the voids. The insets show the corresponding electric field distributions of the void, with the polarization direction shown by an arrow in each inset. b) False color SEM image of a DUV-fabricated Mie void array with bound gold NPs.
• Figure 3: Optical imaging. a) Schematic view of the hyperspectral imaging configuration used to analyze the optical response of Mie voids with individual NPs. The Mie voids array is illuminated with a narrow-band tunable source. The reflected light is collected by a CCD camera matrix, forming a hyperspectral cube -- a stack of images, each corresponding to a certain illumination wavelength. Areas of the Mie void array with Au NPs projected onto the camera matrix generate either brighter (NP "In") or darker (NP "Out") pixels over a broad wavelength span. b) Experimental reflectance spectra extracted from the acquired hyperspectral cube. Each point in this graph takes the intensity value of a certain pixel (corresponding to the areas with NP either "In" or "Out") from a given wavelength $\lambda_i$ image. The reflectance $R_0$ from empty pixels (averaged) is shown with a thinner gray line. The presence of a nanoparticle induces reflectance change, $\Delta R$. The semitransparent region denotes the integration range used for condensing hyperspectral data into final NP maps. The insets show SEM images of the areas corresponding to the extracted pixels. The squares depict the size of a camera pixel relative to the magnified image of the voids. c) Average differential reflectivity $(\Delta R/R_0)^{\mathrm{avg}}$ maps. The NP map is obtained by subtracting the 'Before adding NPs' image from the 'After adding NPs' one. The final NP map is overlapped with a SEM image of the corresponding area to confirm the origin of bright and dark spots. Gray circles denote positions of NPs in the SEM image.
• Figure 4: CNN-Aided digital detection and counting of NPs on Mie voids. a) Labeling of the NP maps using SEM images for training and validation of the model. The maps are marked with 'ground truth' pixel locations acquired from SEM images for "In" and "Out" NPs. b) Illustration of the U-Net architecture of the CNN. The framework utilizes three hidden layers. The numbers above each node indicate the number of applied convolutional filters at this step. The NP map from the validation dataset (d) is processed with the trained CNN model to generate two probability maps for "In" (c) and "Out" (e) NPs. 'Ground truth' locations of the respective NPs are indicated with cyan circles in the "In" image and purple triangles in the "Out" image. f, h) Comparison of precision-recall curves between the CNN-based algorithm applied to "In" (f) and "Out" (h) probability maps and simple thresholding of the original image. Markers on the curves indicate the highest $F_1$-score values for both methods. g) Combined statistics of the performance of both methods on the validation dataset.
• Figure 5: IL-6 biomarker detection. a) Sample preparation procedure. I. Pre-functionalized NPs are mixed with IL-6 molecules in a buffer. IL-6 molecules bind to NPs via antibody-analyte recognition. II. The resulting solution is deposited onto the Mie void chip and left for incubation for one hour. NPs with IL-6 molecules get immobilized on the sensor surface through capture antibody-analyte recognition. III. The chip is washed to remove unbound NPs and dried before measurement. b) Calibration curve showing the detected number of NPs (both "In" and "Out") versus the concentration of IL-6. The output signal of the digital biosensing platform is presented with the mean values at each concentration point, with the error bars showing $\pm$s.d. (standard deviation). The limit of detection is defined as three times the standard deviation of the blank signal (0 pg/ml) counted from the mean value, and is depicted with a gray dashed line. c) Snippets from NP maps corresponding to different concentrations of IL-6. The number of immobilized NPs increases with increasing concentration.