Optical sensor for biomolecule detection and reverse design method thereof

Abstract

The invention relates to the technical field of biomolecule detection, and discloses a reverse design method of an optical sensor for biomolecule detection, which comprises the following specific steps: S1, defining target spectral response, constructing a parameterized model and generating an initial structure library; s2, performing batch simulation by adopting Lumeral FDTD to generate an FDTD simulation data set, training a GAN network model and outputting candidate structure parameters; and S3, optimizing the GAN network model through a Pareto frontier, and outputting an optimal structure parameter. According to the optical sensor for biomolecule detection and the reverse design method thereof, edge hole displacement optimization is realized through the L3-type photonic crystal microcavity, electric field locality is effectively enhanced, near-infrared band optimization is realized, an optical communication window is matched, autofluorescence interference of a biological sample is reduced, specificity is enhanced by designing a multi-layer biological interface, and the detection accuracy is improved. And optimal selection of a parameter structure is realized through reverse design, and a deep learning mechanism is introduced to effectively improve the detection precision of biomolecules.

Description

Technical Field

[0001] The present invention relates to the technical field of biomolecule detection, and in particular to an optical sensor for biomolecule detection and a reverse design method thereof. Background Art

[0002] Silicon-based optical sensors are miniaturized devices that utilize the optical properties of semiconductor silicon materials to achieve high-sensitivity detection by detecting signal changes caused by the interaction between light and biological molecules. They have the advantages of high-sensitivity detection and specific molecular recognition performance in the biomolecule detection industry, and therefore have important application value and application prospects.

[0003] After searching, it was found that according to the invention patent with Chinese patent publication number CN118150523A, a silicon photonic sensor chip, a biomolecule sensing method and a biodetection device are disclosed. The silicon photonic sensor chip and biomolecule sensing method in the invention patent are achieved by opening interconnected microchannels and mounting grooves in the cladding. The mounting grooves and the silicon waveguide are separated by a first distance. While satisfying the evanescent field coupling between the silicon waveguide and the microsphere cavity placed in the mounting groove, the silicon waveguide is still fully covered by the cladding, thus avoiding exposure of the silicon waveguide. Through the setting of the microsphere cavity, the microsphere cavity is utilized to contact the biological sample flowing in the microchannel, and then the silicon waveguide is coupled with the microsphere cavity to detect the movement of the resonant wavelength output by the silicon waveguide, thereby realizing refractive index sensing and biomolecule sensing. By directly contacting the biomolecules with the microsphere cavity, the sensing sensitivity is enhanced while also avoiding contamination of the silicon waveguide, thus enabling detection of different biomolecules. However, this silicon optical sensor chip, biomolecule sensing method, and biodetection device utilizes evanescent field coupling between a microsphere cavity and a silicon waveguide via a 50-200nm gap. The energy exchange efficiency is significantly affected by distance. When the gap is too large, the coupling efficiency significantly decreases, resulting in a reduced signal-to-noise ratio. Furthermore, the Q value of the microsphere cavity is often limited by surface roughness and mode leakage, making it difficult to accurately detect wavelength shifts caused by biomolecule binding. To address these issues, an optical sensor for biomolecule detection and its inverse design method are proposed. Summary of the Invention

[0004] (1) Technical problems solved In response to the shortcomings of the existing technology, the present invention provides an optical sensor for biomolecule detection and a reverse design method thereof, which has the advantages of improving optical performance based on photonic crystal microcavities and achieving structural design optimization through reverse design algorithms, solving the problem in the above-mentioned background technology that the detection limit and sensitivity of existing silicon optical sensor chips need to be further optimized and improved.

[0005] (2) Technical solution

[0006] To achieve the above-mentioned purpose of improving optical performance based on photonic crystal microcavities and realizing structural design optimization through an inverse design algorithm, the present invention provides the following technical solution: an optical sensor for biomolecule detection, comprising a silicon wafer substrate on an insulator, the silicon wafer substrate comprising a substrate layer, a buried oxide layer, and a device layer stacked in sequence from bottom to top, a photonic crystal microcavity structure layer etched inside the device layer, a waveguide layer designed on top of the device layer, the surface of the silicon wafer substrate modified with a functionalized layer, and an encapsulation layer designed on the outside of the functionalized layer.

[0007] Preferably, the substrate layer is a high-resistance silicon structure with a thickness of 500 μm, and the buried oxide layer is a silicon dioxide structure with a thickness of 2 μm; The photonic crystal microcavity structure layer includes an air hole array arranged in a hexagonal lattice, an L3 type microcavity is formed in the air hole array, and the waveguide layer is a ridge silicon waveguide structure with a width of 450nm and a height of 220nm.

[0008] Preferably, the functionalized layer comprises a passivation layer, a silanization layer, a crosslinking layer, a probe layer and a sealing layer stacked sequentially from bottom to top, wherein the passivation layer is a thermally grown silica structure, the silanization layer is an APTMS structure, the crosslinking layer is a glutaraldehyde structure, the probe layer is a biomolecule structure, and the sealing layer is bovine serum albumin; The encapsulation layer includes a microfluidic channel and a sealing layer. The microfluidic channel is a PDMS structure, and the sealing layer is an oxygen plasma activated PDMS structure.

[0009] A method for reverse engineering an optical sensor for biomolecule detection comprises the following specific steps: S1. Define the target spectral response, build a parameterized model and generate an initial structure library; S2. Use Lumerical FDTD batch simulation to generate FDTD simulation datasets, train the GAN network model, and output candidate structural parameters; S3. Optimize the GAN network model through the Pareto frontier and output the optimal structural parameters.

[0010] Preferably, the step of defining the target spectral response for parameterized modeling includes: 1) The hole radius r, lattice constant a and displacement factor δ are used as design variables and constraints, which can be expressed as: ; 2) Construct an objective function to maximize the quality factor Q, sensitivity S and mode volume V, which can be expressed as:

[0011] where Q(p) represents the quality factor and , is the resonant wavelength, is the full width at half maximum; S(p) is the sensitivity and , represents the wavelength drift caused by unit refractive index change; V m (p) is the mode volume and , ϵ is the dielectric constant, and E is the electric field distribution.

[0012] Preferably, the electromagnetic field simulation decomposition is performed by the finite difference time domain method, which is expressed as:

[0013] Use Lumerical API to batch scan the parameter space, and the number of uniform sampling in the constraint space is N=10 4 Group , generate and output FDTD simulation training data, including transmission spectrum With optical indicators ; The steps of building and training the GAN network model include: 1) Construct the generator G through the U-Net structure, expressed as:

[0014] Input noise Target transmission spectrum , the output structure parameters are , the loss function of constructing the generator G is expressed as:

[0015] Which sets =0.5, L2 is the regularization weight; 2) Construct the discriminator D through the CNN network structure, expressed as:

[0016] in is the feature extractor, σ is the Sigmoid activation function, and the discriminator D is constructed with a Wasserstein loss function with gradient penalty, which is expressed as:

[0017] in represents the random interpolation of real and generated samples, β is the gradient penalty coefficient and its value is 10; 3) Combined with the generator G loss function and the discriminator D loss function The GAN network model is jointly trained and expressed as , , build the Adam optimizer and set the batch size and number of iterations.

[0018] Preferably, the GAN network model is optimized according to the Pareto frontier, and the steps include: 1) The number of generated samples according to the trained GAN network model is M=10 4 Group candidate structure parameters ; 2) Use non-dominated sorting to solve Dominate , if and only if at least one of the following inequality conditions is satisfied:

[0019] 3) Optimize the super volume index, expressed as:

[0020] in represents the target vector and , The governing space defined for the reference point.

[0021] (3) Beneficial effects Compared with the prior art, the present invention provides an optical sensor for biomolecule detection and a reverse design method thereof, which has the following beneficial effects: 1. This optical sensor for biomolecule detection and its reverse design method achieve edge hole displacement optimization through an L3-type photonic crystal microcavity, effectively enhancing the localization of the electric field, optimizing and matching the optical communication window in the near-infrared band, and reducing the interference of spontaneous fluorescence of biological samples.

[0022] 2. This optical sensor for biomolecule detection and its reverse design method enhance specificity by designing a multi-layer biointerface, achieve optimal parameter structure selection through reverse design, and introduce a deep learning mechanism to effectively improve the accuracy of biomolecule detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 Schematic diagram of the optical sensor structure of the present invention; Figure 2 This is a schematic diagram of the structure of the photonic crystal microcavity structure layer of the present invention; Figure 3 Schematic diagram of the waveguide layer structure of the present invention; Figure 4 This is a flow chart of the reverse design method of an optical sensor according to the present invention; Figure 5 This is a flow chart of the method for studying the optical wavelength range of the present invention in the near-infrared optical communication wavelength.

[0024] In the figure: 1. Silicon wafer substrate; 101. Substrate layer; 102. Buried oxide layer; 103. Device layer; 2. Photonic crystal microcavity structure layer; 201. Air hole array; 202. L3 type microcavity; 3. Waveguide layer; 4. Functionalization layer; 401. Passivation layer; 402. Silanization layer; 403. Cross-linking layer; 404. Probe layer; 405. Sealing layer; 5. Encapsulation layer; 501. Microfluidic channel; 502. Sealing layer. DETAILED DESCRIPTION

[0025] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0026] Example 1 In this embodiment, the specific structure of the silicon wafer substrate 1 includes: 1) A substrate layer 101 having a thickness of 500 μm and made of a high-resistance silicon structure, which is mainly used for mechanical support to reduce optical loss; 2) A buried oxide layer 102 made of silicon dioxide SiO2 with a thickness of 2 μm, serving as an optical isolation layer to control the refractive index contrast; 3) a device layer 103 having a thickness of 220 nm and made of single-crystal silicon, serving as the core layer of the photonic crystal microcavity; 4) The waveguide layer 3 of the ridge silicon waveguide structure has a width of 450 nm and a height of 220 nm, serving as a light transmission channel; The photonic crystal microcavity structure layer 2 is etched in the device layer 103. Its structure includes an air hole array 201 and an L3-type microcavity 202. The air hole array 201 is arranged in a hexagonal lattice with a lattice constant of a=415 nm and forms a photonic band gap. The size of a single hole in the air hole array 201 can enhance the light-matter interaction volume. The L3-type microcavity 202 with a line defect formed by removing three holes on the air hole array 201 is removed.

[0027] Example 2 In this embodiment, the functionalized layer 4 serves as a surface modification system and mainly includes: 1) a passivation layer 401 of thermally grown silicon dioxide structure having a thickness of 50 nm, which is mainly used to protect the surface of the silicon wafer substrate 1 and provide a hydroxylation base; 2) APTMS [(CH3O)3Si(CH2)3NH2] structure silanization layer 402, with a thickness of 0.8~1.2 nm, is mainly used to form amino terminals, with a surface density of about 3×10 14 sites / cm2 ; 3) Glutaraldehyde (OHC-(CH2)3-CHO) structure cross-linking layer 403, with a thickness of 1.5 nm, mainly used for the reaction between aldehyde groups and amino groups; 4) a biomolecular structure probe layer 404, having a thickness of 5 to 10 nm and mainly used for specifically capturing the target; 5) Bovine serum albumin (BSA) structured blocking layer 405, with a thickness of 4-6 nm, is mainly used to cover nonspecific sites and reduce false positives; The surface modification steps include: Oxygen plasma treatment was used to generate Si-OH, which was self-assembled into a monolayer using 3-aminopropyltrimethoxysilane (APTMS), and glutaraldehyde was used to bridge the amino groups with the amino groups of biomolecules. 2) by covalently immobilizing anti-EGFR antibodies and then blocking nonspecific sites with bovine serum albumin (BSA); The molecular capture mechanism is as follows: target biomolecules bind to antibodies, the local refractive index increases, and the microcavity resonance wavelength redshifts. The theoretical detection limit is set at 10 pg / mm², and the refractive index sensitivity is 200 nm / RIU. The biomolecule attachment mechanism is expressed as follows: Si-OH —— APTMS —— -NH2 │ + OHC-(CH2)3-CHO —— -N=CH-(CH2)3-CH=N- │ + Antibody-NH2——covalent bond fixation.

[0028] Example 3 In this embodiment, the specific steps of combining deep learning with biomolecule detection to study the optical wavelength range within the near-infrared optical communication wavelength include: 1) Input the original transmission spectrum , N = 500 wavelength points, the spectrum is preprocessed by wavelet denoising and the denoised spectrum is output , its wavelet threshold denoising formula is expressed as:

[0029] in is the scaling function, is the wavelet function, c j,k , d j,k are the scaling coefficient and the wavelet coefficient respectively, represents the soft threshold function, represents the layer dependency threshold, σ is the noise standard deviation; 2) Based on the Lorentz fitting, the resonance peak characteristics are extracted and the physical model of the photonic crystal microcavity transmission spectrum is constructed as follows:

[0030] Where A is the baseline intensity, B is the resonance depth and 0 < B < 1, is the resonant wavelength, is the full width at half maximum, C, All are linear drift compensation items; Construct the target optimization function and solve it according to the Levenberg-Marquardt algorithm, which is expressed as:

[0031] 3) Construct time-shift features and obtain dynamic feature vectors, which are expressed as: ,

[0032] Where t is the sampling time point, Q(t) is the real-time quality factor, and R(t) is the resonance contrast; Feature extraction is extracted through the 1D-CNN network architecture, which is expressed as:

[0033] Where K is the convolution kernel size, D is the number of input channels, and c is the output channel index; 4) Combine LSTM with dynamic modeling and input CNN feature sequence , the final hidden state output is , its LSTM unit equation includes:

[0034] Where σ is the sigmoid function, ⊙ represents element-wise multiplication, and W and b represent trainable weights; 5) Transformer concentration prediction, input Add position coding, and its position coding formula is expressed as: ,

[0035] where d model =128 indicates the feature dimension, pos is the time step index; Introduce the multi-head attention mechanism, expressed as:

[0036] The concentration decoding process is expressed as:

[0037] where c max Indicates the maximum concentration, then its concentration output range is [0, c max ]; 6) Construct a composite loss function to train the LSTM dynamics model, expressed as:

[0038] Where β is the balance weight, BCE is the binary cross entropy, a Nadam optimizer is constructed and the batch size is set, and an algorithm test index is constructed based on wavelength detection and concentration detection, which is expressed as: Wavelength detection indicators:

[0039] Concentration detection indicators:

[0040] In summary, the optical sensor for biomolecule detection and its inverse design method achieve edge hole displacement optimization through L3-type photonic crystal microcavity, effectively enhance the electric field localization, optimize and match the optical communication window for the near-infrared band, reduce the spontaneous fluorescence interference of biological samples, enhance specificity by designing a multi-layer biological interface, achieve optimal parameter structure selection through inverse design, and introduce a deep learning mechanism to effectively improve the accuracy of biomolecule detection.

[0041] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply the existence of any such actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or device comprising the element.

[0042] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. An optical sensor for biomolecule detection, characterized in that: The invention comprises a silicon wafer substrate (1) on an insulator, wherein the silicon wafer substrate (1) comprises a substrate layer (101), a buried oxide layer (102) and a device layer (103) stacked in sequence from bottom to top, a photonic crystal microcavity structure layer (2) is etched inside the device layer (103), a waveguide layer (3) is designed on top of the device layer (103), the surface of the silicon wafer substrate (1) is modified with a functionalized layer (4), and an encapsulation layer (5) is designed on the outside of the functionalized layer (4).

2. The optical sensor for biomolecule detection and the reverse design method thereof according to claim 1, characterized in that: The substrate layer (101) is a high-resistance silicon structure with a thickness of 500 μm, and the buried oxide layer (102) is a silicon dioxide structure with a thickness of 2 μm; The photonic crystal microcavity structure layer (2) comprises an air hole array (201) arranged in a hexagonal lattice, an L3 type microcavity (202) is formed in the air hole array (201), and the waveguide layer (3) is a ridge silicon waveguide structure with a width of 450 nm and a height of 220 nm.

3. The optical sensor for biomolecule detection and the reverse design method thereof according to claim 1, characterized in that: The functionalized layer (4) comprises a passivation layer (401), a silanization layer (402), a cross-linking layer (403), a probe layer (404) and a sealing layer (405) stacked in sequence from bottom to top, wherein the passivation layer (401) is a thermally grown silicon dioxide structure, the silanization layer (402) is an APTMS structure, the cross-linking layer (403) is a glutaraldehyde structure, the probe layer (404) is a biomolecule structure, and the sealing layer (405) is bovine serum albumin; The encapsulation layer (5) comprises a microfluidic channel (501) and a sealing layer (502); the microfluidic channel (501) is a PDMS structure, and the sealing layer (502) is an oxygen plasma activated PDMS structure.

4. A method for reverse engineering an optical sensor for biomolecule detection, characterized in that: The specific steps include: S1. Define the target spectral response, build a parameterized model and generate an initial structure library; S2. Use Lumerical FDTD batch simulation to generate FDTD simulation datasets, train the GAN network model, and output candidate structural parameters; S3. Optimize the GAN network model through the Pareto frontier and output the optimal structural parameters.

5. The method for reverse design of an optical sensor for biomolecule detection according to claim 4, characterized in that: The steps to define the target spectral response for parametric modeling include: 1) Based on the hole radius r, lattice constant a and displacement factor As design variables and constraints, they are expressed as: ; 2) Construct an objective function to maximize the quality factor Q, sensitivity S and mode volume V, which can be expressed as: ； where Q(p) represents the quality factor and , is the resonant wavelength, is the full width at half maximum; S(p) is the sensitivity and , represents the wavelength drift caused by unit refractive index change; V m (p) is the mode volume and , ϵ is the dielectric constant, and E is the electric field distribution.

6. The method for reverse design of an optical sensor for biomolecule detection according to claim 5, characterized in that: The electromagnetic field simulation is decomposed by the finite difference time domain method and expressed as: ； Use Lumerical API to batch scan the parameter space, and the number of uniform sampling in the constraint space is N=10 4 Group , generate and output FDTD simulation training data, including transmission spectrum With optical indicators ; The steps of building and training the GAN network model include: 1) Construct the generator G through the U-Net structure, expressed as: ； Input noise Target transmission spectrum , the output structure parameters are , the loss function of constructing the generator G is expressed as: ； Which sets =0.5, L2 is the regularization weight; 2) Construct the discriminator D through the CNN network structure, expressed as: ； in is the feature extractor, σ is the Sigmoid activation function, and the discriminator D is constructed with a Wasserstein loss function with gradient penalty, which is expressed as: ； in represents the random interpolation of real and generated samples, β is the gradient penalty coefficient and its value is 10; 3) Combined with the generator G loss function and the discriminator D loss function The GAN network model is jointly trained and expressed as , , build the Adam optimizer and set the batch size and number of iterations.

7. The optical sensor for biomolecule detection and the reverse design method thereof according to claim 1, characterized in that: The steps of optimizing the GAN network model according to the Pareto frontier include: 1) The number of generated samples according to the trained GAN network model is M=10 4 Group candidate structure parameters ; 2) Use non-dominated sorting to solve Dominate , if and only if at least one of the following inequality conditions is satisfied: ; 3) Optimize the super volume index, expressed as: ； in represents the target vector and , The governing space defined for the reference point.

Images