On-chip integrated light field encoding device and design method and application thereof

Abstract

The invention discloses an on-chip integrated light field encoding device and a design method and application thereof. The light field encoding device comprises a substrate layer, an induction layer and an encoding layer. The substrate layer has a first surface and a second surface opposite to each other, the sensing layer is arranged on the first surface, and the coding layer is arranged on the second surface; the coding layer comprises a plurality of coding areas, and when the target light penetrates through the coding layer, modulation responses of the coding areas to the target light are different; the sensing layer has a patterning sensing function and can sense the modulation state of transmission light which transmits the coding layer and the substrate layer in different areas. Displacement errors caused by system-level integration alignment can be reduced, the preparation process difficulty and the preparation cost are reduced, a large amount of optical loss in a traditional optical system can be avoided, the light utilization efficiency of an imaging chip is greatly improved, particularly, the working stability of the imaging chip is improved, and the optical system is suitable for large-scale popularization and application. And the imaging influence caused by insufficient quantum efficiency of a partial wave band detector can be reduced.

Description

Technical Field

[0001] The present invention relates to the field of spectral imaging technology, and in particular to an on-chip integrated light field coding device, a design method and an application thereof. Background Art

[0002] Today, the Fourth Industrial Revolution is sweeping the globe. With the rapid development of information technologies such as big data and artificial intelligence, the importance of a new generation of intelligent photoelectric detection technology is becoming increasingly prominent. Advanced photoelectric detection and recognition systems play a key role in both civilian intelligent systems and future warfare, becoming the indispensable "eyes" and "brains" of dynamic perception and intelligent recognition. In this context, spectral imaging technology, capable of simultaneously acquiring both spatial information of a target and spectral information representing material properties, significantly increases the dimensionality and magnitude of imaging data and has become a core technology in the field of intelligent photoelectric detection.

[0003] In the field of spectral imaging technology, traditional spectrometers are mainly divided into three categories: dispersion type, filter type and Fourier transform type, but these technical paths all have significant defects. For example:

[0004] (1) Dispersive spectrometers rely on the spatial dispersion principle of gratings or prisms and need to obtain spectral information point by point through mechanical scanning, resulting in a bulky system and low acquisition efficiency.

[0005] (2) Filter-type spectrometers use tunable filters to achieve wavelength selection, but they have problems such as low light flux and limited spectral resolution. In addition, filters need to be frequently switched during multi-band detection, which restricts real-time performance.

[0006] (3) The Fourier transform spectrometer is based on the principle of interference. Although it can obtain a high signal-to-noise ratio spectrum, it requires a precise moving mirror to scan the time dimension, which has extremely high requirements for mechanical stability and is difficult to miniaturize.

[0007] The common defects of these three types of traditional technologies are concentrated in three aspects: first, they rely on mechanical scanning mechanisms in spatial or temporal dimensions, resulting in high system complexity and limited response speed; second, the core optical components (such as gratings, filter arrays, interferometers, etc.) are bulky, and the weight of the whole machine often reaches tens of kilograms, which seriously limits their portable applications; third, the coupling design of precision optical components and complex mechanical structures significantly increases manufacturing costs. The price of a single device can easily reach hundreds of thousands of yuan, which is difficult to meet the needs of large-scale popularization.

[0008] Computational spectral imaging technology has emerged to overcome fundamental bottlenecks such as reliance on mechanical scanning and bulky size in traditional spectral imaging technology. The core innovation of this technology lies in transferring the burden of physical spectrometry to the digital domain: on the optical side, the incident light field is encoded in a hybrid spatial-spectral dimension through a micro-nanostructure modulation layer, so that the two-dimensional image acquired in a single snapshot contains three-dimensional spectral information; on the algorithm side, advanced computing methods such as compressed sensing and neural networks are used to decode and reconstruct a high-precision spectral cube from the encoded image. This "physical encoding + digital decoding" synergistic mechanism essentially reduces hardware complexity through computing power. When hardware miniaturization leads to limited spectroscopic capabilities of physical devices, computational reconstruction technology can reversely compensate for key indicators such as spectral resolution, successfully breaking the industry curse of "device shrinkage - performance degradation."

[0009] The computational spectral imaging system is mainly divided into two key steps:

[0010] ① Light field coding: Use microstructures such as metasurfaces and photonic crystals to perform multi-dimensional modulation of the incident light wavefront. By designing the low correlation of the coding matrix, it ensures that light signals of different wavelengths form a unique and identifiable imprint on the detector.

[0011] ② Algorithm decoding: Construct a mathematical model of the spectral response matrix and coding characteristics, and use algorithms such as sparse optimization to decouple each wavelength component from the aliased signal.

[0012] However, traditional computational spectral imaging systems are still constrained by a discrete architecture—the modulation layer and detector are stacked using separate devices.

[0013] Currently, there are three main types of traditional computational spectral imaging systems at home and abroad: image plane coding, point spread function coding, and spectral response coding.

[0014] The main idea of ​​image plane encoding is based on the principle of spectral information compression on the intermediate image plane. After the target is imaged on the modulation plane through the objective lens, the spatial light modulator is used to implement pixel-level light field cropping. Typical solutions include:

[0015] Transmissive encoding: A chrome-plated quartz mask is used to achieve fixed-mode spectral filtering.

[0016] Dynamic encoding: Application of digital micromirror arrays or liquid crystal on silicon spatial modulators for programmable light intensity modulation.

[0017] By selectively filtering out redundant spectral data and reusing effective information in the spatial dimension, coding constraints are provided for subsequent spectral reconstruction.

[0018] Point spread function encoding refers to the use of the spatial response characteristics of the optical system to implement spectral dimension encoding. The core lies in designing a wavelength-sensitive point spread function. Implementation paths include:

[0019] Diffraction control method: Using diffraction optical elements, the amplitude / phase distribution of light waves is controlled to produce a wavelength-dependent point spread function.

[0020] Scattering medium method: Use irregular surfaces or composite media to induce random scattering and construct the nonlinear variation of PSF with wavelength.

[0021] Dispersion coupling method: introduce dispersive elements such as prisms and gratings to form a mapping relationship between the point spread function and the incident wavelength.

[0022] Spectral response coding technology refers to the information encoding achieved through the differentiation of the spectral sensitivity of the detector pixels. The main technical routes are:

[0023] Spectral filter array: A specific bandpass filter array is integrated on the surface of the CMOS pixel to reconstruct the spectral image based on the complementary information provided by different channels.

[0024] Multi-aperture spectrometry: Spatial replication achieved by simultaneous imaging using multiple apertures.

[0025] Sequential switching acquisition: Time-sharing multispectral imaging is achieved through tunable filter rotation, and the static characteristics of the scene must be ensured.

[0026] Many existing devices for implementing light field encoding require complex processing techniques, are expensive, or make the system bulky and heavy. System calibration and robustness also need to be improved. These include the following:

[0027] (1) Cost issue: The cost of digital micromirror arrays and liquid crystal on silicon light modulators is very high, so the manufacturing cost of spectral imaging systems using image plane coding is relatively high.

[0028] (2) Limited usage scenarios: Both image plane coding and point spread function coding computational spectrometers are very susceptible to environmental influences, and time-series switching acquisition can only be applied to static scenes.

[0029] (3) Volume problem: These traditional computational spectrometers all have a huge optical system volume and cannot be used in miniaturized scenarios.

[0030] (4) Low light utilization: The optical system leads to reduced light flux, and when the quantum efficiency of the infrared detector is low, long exposure time is required and the image quality is reduced.

[0031] (5) Usage conditions: Due to the complexity of the mechanical structure and optical system, it is sensitive to vibration. Summary of the Invention

[0032] In view of the deficiencies in the prior art, the present invention aims to provide an on-chip integrated optical field coding device, a design method and an application thereof.

[0033] To achieve the aforementioned object of the invention, the technical solutions adopted by the present invention include:

[0034] In a first aspect, the present invention provides an on-chip integrated optical field coding device, which includes a substrate layer, a sensing layer, and a coding layer;

[0035] The base layer has a first surface and a second surface facing each other, the sensing layer is arranged on the first surface, and the coding layer is arranged on the second surface;

[0036] The coding layer includes a plurality of coding regions, and when the target light passes through the coding layer, each coding region has a different modulation response to the target light;

[0037] The sensing layer has a patterned sensing function and can sense the modulation state of the transmitted light that passes through the coding layer and the base layer in different areas.

[0038] In a second aspect, the present invention further provides a method for designing a light field coding device, comprising:

[0039] Providing a variety of different metasurface micro-nanostructure units to form a selection set;

[0040] Simulating all the metasurface micro-nanostructure units to obtain corresponding wide-spectrum response characteristics;

[0041] A greedy algorithm is used to select a preset number of target sets from the candidate set, wherein the metasurface micro-nanostructure units of the target sets have the lowest correlation with each other;

[0042] Each super-surface micro-nanostructure unit in the target set is used as a coding region to form a coding layer.

[0043] In a third aspect, the present invention further provides a spectral imaging observation system, which includes: the above-mentioned light field encoding device, an incident module, and an analysis module;

[0044] The incident module is used to transmit the target light to the light field encoding device, and the parsing module stores a deep neural network model for parsing the patterned light signal emitted by the sensing layer and restoring the reconstructed spectrum of the target light.

[0045] In a fourth aspect, the present invention further provides a model training method in the above-mentioned spectral imaging observation system, which comprises:

[0046] Provide a variety of training incident lights with different reconstructed spectra, and use the reconstructed spectra as training labels;

[0047] Allowing the training incident light to irradiate the light field encoding device through the incident module to obtain a patterned light signal as training data;

[0048] The deep neural network model is trained using the training data and the training labels so that the reconstructed spectrum output by the deep neural network model is close to the training labels.

[0049] Based on the above technical solution, compared with the prior art, the beneficial effects of the present invention include at least:

[0050] The technical solution provided by the present invention integrates the sensing layer and the coding layer on both the front and back sides of the substrate layer, which can reduce the displacement error caused by the alignment of traditional filters and detector pixels, and also reduces the difficulty of the manufacturing process and the cost required for manufacturing. It can also avoid a large amount of optical loss in traditional optical systems, greatly improve the light utilization efficiency of the imaging chip, especially improve the working stability of the imaging chip in special environments such as low-light environments, and can also reduce the imaging impact caused by insufficient quantum efficiency of detectors in some bands.

[0051] The on-chip integrated process is compatible with traditional semiconductor manufacturing processes, so there is no need to further explore the process, which reduces the cost of the manufacturing process and lays a solid foundation for rapid commercialization.

[0052] Light field coding is the modulation and encoding of light fields over a wide band. Combined with supporting deep learning algorithms, it can achieve real-time spectral detection. As computing power increases, ultrafast spectral imaging can be further realized, solving the problem of slow traditional spectral imaging rates, expanding application scenarios, and accelerating the pace of commercialization.

[0053] Light field coding devices are on the same order of magnitude as detectors, radically miniaturizing spectral imaging systems to the same scale as chips, resolving the issues of bulk and heaviness. Light field coding devices can be directly integrated into existing electronic products, enabling rapid commercialization.

[0054] By selecting the material system and the detector band, it can also be expanded to any band.

[0055] The above description is only an overview of the technical solution of the present invention. In order to enable those skilled in the art to more clearly understand the technical means of this application and implement them according to the contents of the specification, the following is an explanation of the preferred embodiments of the present invention with detailed drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 This is a schematic structural diagram of an optical field coding device provided by a typical embodiment of the present invention;

[0057] Figure 2 This is a schematic diagram of a portion of the structure of an optical field coding device provided by a typical embodiment of the present invention;

[0058] Figure 3This is a schematic diagram of a spectrum reconstruction process provided by a typical embodiment of the present invention;

[0059] Figure 4 1 is a schematic top view of the structure of different metasurface micro-nanostructure units of a light field coding device provided by a typical embodiment of the present invention;

[0060] Figure 5 This is a schematic diagram of a screening process based on a greedy algorithm provided in a typical implementation case of the present invention;

[0061] Figure 6 This is a graph showing correlation coefficients and transmission efficiency of different metasurface micro-nanostructures in a light field coding device provided by a typical embodiment of the present invention;

[0062] Figure 7 This is a schematic diagram of the preparation process of an optical field coding device provided by a typical embodiment of the present invention;

[0063] Figure 8 This is a schematic diagram of the structure of a spectral imaging observation system provided by a typical embodiment of the present invention.

[0064] Figure 9 These are the training set and test set spectra of the spectral imaging observation system provided by a typical embodiment of the present invention;

[0065] Figure 10 This is a spectrum reconstruction diagram of a spectrum imaging observation system incident with a continuous light source, provided by a typical embodiment of the present invention;

[0066] Figure 11 This is an example diagram of the spectral imaging restoration effect under a single channel of a spectral imaging observation system provided in a typical implementation case of the present invention. DETAILED DESCRIPTION

[0067] In view of the shortcomings of the prior art, the inventors of this case, after long-term research and extensive practice, have proposed the technical solution of the present invention. The following will further explain this technical solution, its implementation process and principles.

[0068] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the scope of protection of the present invention is not limited to the specific embodiments disclosed below.

[0069] An embodiment of the present invention provides an on-chip integrated optical field coding device, which includes a base layer, a sensing layer and a coding layer; the base layer has a first surface and a second surface facing each other, the sensing layer is arranged on the first surface, and the coding layer is arranged on the second surface; the coding layer includes multiple coding regions, and when target light passes through the coding layer, each coding region has a different modulation response to the target light; the sensing layer has a patterned sensing function and can sense the modulation state of the transmitted light passing through the coding layer and the base layer in different areas.

[0070] Based on the above technical solution, the preferred embodiment of the present invention designs a computational mid-infrared micro-spectral imaging chip with integrated on-chip integration of metasurface detectors based on the combination of four technologies: metasurface technology, light field coding technology, algorithm decoding technology, and encoder-detector integration, and proposes a corresponding preparation process and supporting algorithm.

[0071] Compared with traditional computational mid-infrared spectral imaging chips, not only does it use metasurface technology to miniaturize the light field encoding device, but it also solves the nanoscale alignment accuracy requirements (<100nm) and heterogeneous integration process problems by integrating the encoder and the detector on-chip. In terms of metasurface material selection, a material system compatible with mainstream semiconductor processes was selected, which greatly reduced the preparation cost and also reduced the complexity and cost of the on-chip integrated integration process. Through the carefully designed metasurface micro-nano structure, not only the light utilization rate is greatly improved, but also the spectral encoding correlation between metasurfaces is greatly reduced. Through the careful design of the supporting algorithm, the prior knowledge of the encoding matrix is ​​embedded into the algorithm, thereby realizing a mid-infrared micro-spectral imaging chip with high spectral resolution and high imaging resolution.

[0072] Breakthroughs in optical metasurface technology offer a revolutionary solution for the miniaturization of spectral imaging systems. Metasurfaces not only compress spectral modulation capabilities to subwavelength scales but can also be monolithically integrated with detectors using CMOS-compatible processes, fundamentally eliminating inter-layer misalignment errors and ushering in a new "device-as-system" paradigm. Metasurfaces, composed of subwavelength artificial microstructures, enable multi-dimensional manipulation of lightwave phase, amplitude, and polarization within submicron thicknesses. Compared to traditional optical components, their key advantages lie in: 1) Deep integration—by fabricating spectral modulation microstructures directly on the detector pixel surface, they achieve integrated "modulation-detection" monolithic integration, completely eliminating misalignment errors; 2) Extreme miniaturization—with a single metasurface device less than a micron thick, combined with CMOS processes, they enable chip-scale spectral imaging systems, three orders of magnitude smaller than traditional devices; 3) Functional programmability—by designing microstructure geometry, spectral response characteristics can be flexibly customized, supporting dynamic reconfiguration of detection modes. Metasurface-based spectral chips not only reduce system size to the size of a coin, but also reduce manufacturing costs by over 90% through algorithm-hardware co-optimization.

[0073] The computational mid-infrared micro-spectral imaging chip with on-chip metasurface detector integration is a spectral imaging chip technology that integrates metasurface micro-nanostructures directly onto mid-infrared detector pixels. It encodes the mid-infrared light incident on the chip into a light field. The detector reads the data and then reconstructs the spectral information and spectral imaging using a specially designed spectral reconstruction algorithm. Furthermore, the on-chip integration of the detector and metasurface micro-nanostructure overcomes the challenges of nanometer-scale alignment accuracy (<100nm) and heterogeneous integration, thereby reducing processing costs and realizing a low-cost spectral imaging chip with mass production potential and strong application prospects. Leveraging the advantages of on-chip integration, spectral encoding can be performed across all detector pixels. Leveraging the properties of this global encoding, targeted algorithms can be developed to achieve spectral imaging with both high spectral resolution and high imaging resolution. This not only meets the demand for the lowest possible processing cost, but also addresses the challenges of high spectral resolution and high imaging resolution.

[0074] Spectral imaging chip technology is already entering the application landscape. By deeply integrating fundamental research, process breakthroughs, and practical applications, this technology is driving a revolutionary shift in spectral imaging from high-end laboratory instruments to ubiquitous intelligent sensing terminals. With continued technological advancement, its maturity is increasing, and its cost is decreasing, providing spectral imaging data with higher spectral and imaging resolutions for applications across various fields.

[0075] Of course, in alternative embodiments, metasurface technology may not be used, and quantum dot films, two-dimensional material layers, chemical coating layers, etc. may be used, as long as different spectral responses can produce large differences.

[0076] In some embodiments, the coding region includes any one or a combination of two or more of a super surface, a quantum dot film, a two-dimensional material layer, and a chemical coating layer.

[0077] In some embodiments, the coding regions are selected from supersurfaces, and the height of the supersurfaces in all the coding regions is equal.

[0078] In some embodiments, the coding regions are arranged in an array on the coding layer.

[0079] An embodiment of the present invention further provides a method for designing a light field coding device provided in any of the above embodiments, comprising the following steps:

[0080] Providing a variety of different metasurface micro-nanostructure units to form a selection set;

[0081] Simulating all the metasurface micro-nanostructure units to obtain corresponding wide-spectrum response characteristics;

[0082] A greedy algorithm is used to select a preset number of target sets from the candidate set, wherein the metasurface micro-nanostructure units of the target sets have the lowest correlation with each other;

[0083] Each super-surface micro-nanostructure unit in the target set is used as a coding region to form a coding layer.

[0084] In some embodiments, the target set screening process specifically includes:

[0085] Select a metasurface micro-nanostructure unit and use the corresponding wide spectrum response characteristics as a primitive;

[0086] Calculating the correlation coefficient between the primitive and the wide spectrum response characteristics of the remaining metasurface microstructure units, and selecting a metasurface micro-nanostructure unit with the smallest correlation coefficient to be included in the observation matrix;

[0087] Traversing all the metasurface microstructures as the primitives, and incorporating the corresponding metasurface microstructure units into the observation matrix;

[0088] A group with the smallest average correlation coefficient is selected from the observation matrix as the target set.

[0089] In some embodiments, the structural parameters of the metasurface micro-nanostructure units are different, and the structural parameters include period P, duty cycle a, line width w, and rotation angle θ.

[0090] In some embodiments, the structural parameters range from period P=μm-5μm, a=0.3-0.8, w=1 / 3a-2 / 3a, θ=0-80°.

[0091] An embodiment of the present invention also provides a spectral imaging observation system, which includes: a light field encoding device, an incident module, and an analysis module provided in any of the above embodiments; the incident module is used to transmit the target light to the light field encoding device, and the analysis module stores a deep neural network model for analyzing the patterned light signal emitted by the sensing layer and restoring the reconstructed spectrum of the target light.

[0092] The embodiment of the present invention further provides a model training method in the above-mentioned spectral imaging observation system, which includes the following steps:

[0093] Provide a variety of training incident lights with different reconstructed spectra, and use the reconstructed spectra as training labels;

[0094] Allowing the training incident light to irradiate the light field encoding device through the incident module to obtain a patterned light signal as training data;

[0095] The deep neural network model is trained using the training data and the training labels so that the reconstructed spectrum output by the deep neural network model is close to the training labels.

[0096] The technical solution provided by the present invention is divided into three parts: (1) the design of a light field coding device based on a metasurface, (2) the preparation process of the light field coding device and its on-chip integration with the detector, and (3) calibration testing and supporting algorithm design. These three aspects will be introduced in detail below.

[0097] (1) Design of light field coding device based on metasurface: The light field coding device of the present invention is composed of detector pixels, SiO2 buffer layer and a-Si metasurface micro-nanostructure layer (note that the buffer layer here can be any material, such as silicon dioxide, aluminum oxide, etc., and the material of the micro-nanostructure layer can also be any material with high transmission efficiency at mid-infrared wavelengths, such as a-Si, Ge, two-dimensional materials, quantum dot materials, etc.).

[0098] The light field encoding device is a buffer layer (thickness can be selected from 0.1μm-2μm) superimposed on a metasurface micro-nanostructure layer (note that the structure layer here can be a stack of multiple materials and then the corresponding metasurface structure is designed, and the thickness can be from 1μm to 5μm). The schematic diagram of a single device is as follows Figure 1 This simple stacking and carefully designed micro-nanostructure (the design principle is mainly based on the near-field coupling of the internal structure of the structural unit) is easy to process and can encode the incident infrared light, thereby achieving a high-efficiency and highly differentiated light field encoding effect.

[0099] For the design of metasurface micro-nanostructures, we calculated the transmittance and diffraction efficiency of various metasurface structures (circular, rectangular, cross, H-shaped, etc.) and various structural parameters (length, width, and height) by scanning a large number of structural units with different parameters. Using a greedy algorithm, we selected transmission responses with low similarity, high efficiency, and high diffraction efficiency, and obtained the corresponding metasurface structural parameter results. (All simulations were performed using simulation software such as COMSOL.)

[0100] Of course, the choice of buffer layer and metasurface micro-nanostructure materials can be changed according to the required application band. For example, when choosing a material that is transparent in the visible light band, it can be used as a light field coding device in the visible light band, and is expected to realize applications such as cross-band spectral coding devices and even cross-band spectral imaging.

[0101] (2) Preparation process of light field coding device and its integration process with detector chip: First, the processed mid-infrared detector device (the mid-infrared detector can be a mid-infrared detector of any material system, such as a superlattice system, a two-dimensional material system, etc.) is simply cleaned, and then a 0.1μm thick SiO2 is grown on the substrate of the device as a buffer layer (the thickness can be selected between 0.1μm and 2μm), and then a 1μm thick a-Si material is deposited on the buffer layer as a super-surface micro-nano structure layer. Then, positive photoresist AZ1500 is sprayed and the device is patterned by photolithography and development technology (any photolithography machine model can be used, and photolithography machines with different resolutions can be selected according to the actual application requirements for the unit device size). Next, Ti / Au / Cr is deposited using an electron beam evaporation coating machine, where Cr is used as a hard mask material for etching a-Si materials. An NLD etcher is used with Cr as a mask to etch away a-Si materials other than those without patterning. After etching, the sample is immersed in an acetone solution for 12 hours. A rubber-tipped dropper is used to gently blow on the surface in the solution, and ultrasonically applied for 30 seconds × 4 times to completely remove the photoresist, thereby obtaining the desired super-surface micro-nano structure. Finally, the sample is rinsed several times with isopropyl alcohol and deionized water, and dried with nitrogen to complete the preparation (as shown in FIG. Figure 2 shown).

[0102] 3) Calibration test and supporting algorithm design

[0103] Prototype calibration: A medium-wave infrared monochromatic light source is used to illuminate the prototype, and the detector circuit directly reads the response data of each pixel to obtain the experimental frequency response curve of the metasurface in each channel within 1.8 to 5.4 μm, and further calibrate the prototype performance.

[0104] Data Acquisition: Continuous light sources with varying frequency response characteristics, both domestic and foreign, were used to illuminate the prototype and standard spectrometer systems, generating the training sets required for the spectral reconstruction algorithm. Mid-infrared spectral imaging data was first acquired using a pre-assembled acquisition system (using a homologous mid-infrared detector to filter and capture spectral imaging data in the corresponding wavelength band).

[0105] Performance test: Continuous light source is incident on the prototype to obtain its spectrum reconstruction and compare it with the test results of the standard spectrometer system to evaluate the spectrum detection performance of the verification prototype. The built acquisition system is used to image on the prepared spectrum imaging chip, and the data is put into the spectrum imaging algorithm for reconstruction and compared with the previously collected imaging data to evaluate the spectrum imaging performance of the verification prototype (the reconstruction flow chart is as follows Figure 3 shown).

[0106] The proposed computational mid-infrared micro-spectral imaging chip, which integrates a metasurface detector on-chip, is of great significance for the realization of on-chip micro-spectral imaging chips. In theory, the chip can be designed to capture any wavelength by simply changing the shape and structural parameters of the metasurface's micro-nanostructure units and selecting the appropriate material system and detectors for the corresponding wavelengths.

[0107] Furthermore, the fabrication process requires only a single lithography and etching step to produce the desired wavelength band of light field coding devices. All metasurface micro-nanostructure units maintain identical heights, enabling the fabrication of large-scale light field coding arrays in one go. This fabrication process not only greatly simplifies the overall process flow but also significantly reduces fabrication costs due to on-chip integration, significantly advancing the commercialization of micro-spectral imaging chip technology.

[0108] Combined with deep learning algorithms, the imaging speed can be greatly improved to achieve fast spectral imaging. The introduction of physical prior knowledge in the algorithm perfectly solves the imaging resolution limitations brought by traditional filters, and truly achieves spectral detection with high spectral resolution in the mid-infrared band and hyperspectral imaging with high imaging resolution.

[0109] The present invention is perfectly compatible with existing mature semiconductor processes and CMOS integration in terms of preparation process, and the preparation process is simple, which truly solves various shortcomings of existing spectral imaging systems (imaging resolution, spectral resolution, imaging rate).

[0110] The technical solution of the present invention is further described in detail below through several embodiments and in conjunction with the accompanying drawings. However, the selected embodiments are only used to illustrate the present invention and are not intended to limit the scope of the present invention.

[0111] Example

[0112] The present invention discloses a design scheme, corresponding preparation process and supporting algorithm for a computational mid-infrared micro-spectral imaging chip integrated on a metasurface detector chip. By changing the shape and structural parameters of the metasurface micro-nano structural unit, a light field coding device with different spectral responses in a wide band can be realized. By scanning a large number of structural parameters, a greedy algorithm is combined to calculate the correlation of spectral responses between different structural units. Then, a combination of structural units with low correlation and the highest efficiency is obtained as a light field coding device. Then, a plasma enhanced chemical vapor deposition device (PECVD), an electron beam lithography machine, a magnetic neutral loop discharge plasma source etching device (NLD) and other equipment are used to prepare the designed spectral coding device on the detector pixel. Finally, a large number of spectral responses passing through the light field coding device are used as training data and as input data for deep learning for training, and then the trained neural network is used for spectral reconstruction, and an image deep learning algorithm is used in combination with physical priors to realize hyperspectral imaging. The following will introduce the specific embodiments of design and preparation, spectral reconstruction, and spectral imaging:

[0113] (1) Design of optical field coding device: The design principle of the metasurface micro-nanostructure unit in the present invention is mainly to utilize the change of wide spectral response caused by the near-field coupling of the internal structure of the micro-nanounit. This is done by simply stacking materials on the detector pixel and then selectively etching the specific functional layer. The side view of the structure is shown in Figure 2. Figure 1 , Figure 2 As shown, its simple structure top view is shown in Figure 4 . (P is the period, a is the duty cycle, θ is the rotation angle, and w is a parameter related to a) The parameter settings here are: period P is 1μm-5μm, duty cycle a is 0.3-0.8, w is 1 / 3a-2 / 3a, and θ is 0-80 degrees. (The parameter selection here is mainly to match the wavelength, because the essence of the metasurface is to control the light field through micro-nanostructure units of the half-wavelength order. Considering the processing difficulty and the actual application band, the above parameters are chosen for scanning here).

[0114] The micro-nanostructure units were simulated using COMSOL multiphysics software. The incident light source was set to a superposition of TE and TM polarized light, and periodic boundary conditions were added on both sides of the units to simulate the electromagnetic properties of the micro-nanostructure units within an infinite periodic structure. By scanning the structural parameters, a large number of broad spectral response results were obtained. A greedy algorithm was then used to select 16 micro-nanostructure units with low correlation. These were then integrated on-chip with the detector as optical field encoding devices.

[0115] The screening strategy based on the greedy algorithm is as follows Figure 5 shown.

[0116] Through simple screening, light field coding devices with low correlation coefficients can be selected. Figure 6 This is the correlation coefficient and transmission efficiency diagram of 16 micro-nano structure units. The average correlation coefficient here is 0.28 (usually a correlation coefficient below 0.3 is acceptable).

[0117] (2) Fabrication process of optical field encoding devices: In this embodiment, unlike the traditional spectral imaging system where the encoding device and detection system are separated, the encoding device and the detection device are integrated into one structure. By integrating the encoder and the detector on-chip, the nanometer-level alignment accuracy requirement (<100nm) and the heterogeneous integration process challenges are solved. In terms of the selection of metasurface materials, a material system compatible with mainstream semiconductor processes is selected, which greatly reduces the preparation cost and also reduces the complexity and cost of the on-chip integration process.

[0118] The preparation mainly uses plasma enhanced chemical vapor deposition coating equipment (PECVD), electron beam lithography equipment, magnetic neutral loop discharge plasma source etching equipment (NLD) and other instruments. The main process steps are: detector substrate preparation, PECVD coating of SiO2 film, PECVD coating of a-Si film, photoresist, electron beam lithography, electron beam evaporation of top layer Cr, lift-off, NLD etching, and Cr removal. The specific process flow chart is shown in Figure 7 As shown, the following steps are included.

[0119] 1) Substrate preparation: The present invention integrates the light field encoding device and the detector chip, so the substrate is the original detector chip. Since it is an on-chip integration, it can be directly processed on the prepared detector, and a simple substrate cleaning step is sufficient. The substrate cleaning steps usually taken are as follows: first, take a glass beaker and a polytetrafluoroethylene flower basket of appropriate capacity, rinse them repeatedly with deionized water, and use a nitrogen gun to blow dry the excess water on the inner wall of the beaker and the flower basket; then take out the sample and put it into the flower basket, and add acetone solution to the beaker, immerse the flower basket and the sample into the solution, and then put it into an ultrasonic cleaner for cleaning for 5 minutes, and repeat this operation twice; then replace the isopropyl alcohol solution and clean the substrate for 5 minutes, repeat twice, and the other operations are the same as the previous step; finally, rinse the substrate repeatedly with deionized water, blow dry it with nitrogen, and put it into the sample box.

[0120] 2) PECVD coating of SiO2 film: A 0.24 μm thick SiO2 film (the thickness can be selected from 0.1 μm to 2 μm) is prepared using a PECVD system with a growth rate of 120 nm / min, a growth time of 2 min, and an operating temperature of 100°C.

[0121] 3) PECVD coating of a-Si film: Use the PECVD system to prepare a-Si film with a thickness of 3μm (the thickness can be selected from 1μm to 3μm, mainly based on the parameters of your design structure), with a growth rate of 300nm / min, a growth time of 10min, and an operating temperature of 250℃.

[0122] 4) Spreading: Use A4 photoresist for electron beam lithography, speed 600-6s, speed 4000-30s. Then bake the plate at 180℃ for 10 minutes.

[0123] 5) Electron beam lithography: Any type of electron beam lithography machine can be used, using specific dose process parameters for lithography at a speed of 1200um / min.

[0124] 6) Development: After exposure, the film was developed using a mixture of tetramethylolpropane and isopropyl alcohol in a ratio of 1:3. The development time was 100 seconds and the fixing time was 30 seconds.

[0125] 7) Gluing: Use a glue machine with 200W power for one minute.

[0126] 8) Electron beam evaporation of top layer Cr: Cr is deposited using an electron beam evaporation coating machine with a thickness of 50 nm.

[0127] 9) Metal Stripping: After coating, soak the sample in NMP solution for 12 hours. Then, place the sample in an ultrasonic cleaner for 15 minutes, then place it in an 85°C water bath for 15 minutes. After the water bath, place the sample in an ultrasonic cleaner for another 15 minutes.

[0128] 10) NLD Etching: Next, the structure of the light field encoding device is fabricated by etching downwards, using Cr as a mask. An NLD etcher is used to remove the unwanted a-Si in one go, with an etching time set to 300 seconds. Even if some removal of the underlying SiO2 is minimal, the impact on the final result is negligible.

[0129] 11) Removal of residual Cr: Immerse the prepared sample in Cr etching solution for 100 seconds to remove the Cr mask residue on the surface of the structure.

[0130] Of course, the above process steps are exemplary. Based on the device structure and application principle proposed in the present invention, those skilled in the art may replace them with other equivalent preparation processes to obtain devices with equivalent functions, which still fall within the scope of protection of the present invention.

[0131] (3) Spectral reconstruction implementation: Figure 8As shown, a mid-wave infrared monochromatic light source was used to illuminate the prototype. The detector circuit directly read out the response data of each pixel, obtaining the experimental frequency response curves of the metasurface in each channel within the 1.8-5.4μm range and further calibrating the prototype performance. Then, continuous light sources from China and other countries with different frequency response characteristics were respectively incident on the prototype and a standard spectrometer system to obtain the training set required for the spectral reconstruction algorithm. Mid-infrared spectral imaging data was first collected using the established acquisition system (a homologous mid-infrared detector was required to filter and collect spectral imaging data in the corresponding band).

[0132] Here, a deep neural network is used as the training network for spectral reconstruction. For the specific process, refer to Figure 3 shown.

[0133] Figure 9 The specific effect of spectral reconstruction is shown in the figure. For spectral imaging, U-Net and residual connection are used as the basic model, and the response of the light field encoding device to mid-infrared light is embedded as the prior physical knowledge of the network. The test set effect is as follows: Figure 10 and Figure 11 shown.

[0134] From the test results, it can be concluded that the spectral imaging of this embodiment can achieve a peak signal-to-noise ratio of up to 43.4 dB and an average structural similarity of 0.94 (maximum 0.99) (the maximum value is 1) in prediction, and a minimum peak signal-to-noise ratio of 33.9 dB and an average structural similarity of 0.89.

[0135] Unlike traditional spectral imaging systems, the present invention proposes a computational mid-infrared micro-spectral imaging chip designed by combining four technologies: on-chip integration with detectors, metasurface technology (in actual applications, quantum dot films, two-dimensional material layers, chemical coating layers, etc. can also be used as equivalent substitutes, as long as they can produce large differences in different spectral responses), light field coding technology, and algorithm decoding technology. This chip can achieve high-resolution spectral imaging and high-resolution detection effects.

[0136] It should be understood that the above embodiments are merely illustrative of the technical concepts and features of the present invention. Their purpose is to enable those skilled in the art to understand the contents of the present invention and implement them accordingly. They are not intended to limit the scope of protection of the present invention. Any equivalent variations or modifications made in accordance with the spirit and substance of the present invention are intended to be encompassed within the scope of protection of the present invention.

Claims

1. An on-chip integrated optical field coding device, characterized in that: Includes base layer, sensing layer and coding layer; The base layer has a first surface and a second surface facing each other, the sensing layer is arranged on the first surface, and the encoding layer is arranged on the second surface; The coding layer includes a plurality of coding regions, and when the target light passes through the coding layer, each coding region has a different modulation response to the target light; The sensing layer has a patterned sensing function and can sense the modulation state of the transmitted light that passes through the coding layer and the base layer in different areas.

2. The optical field encoding device according to claim 1, wherein: The coding region includes any one or a combination of two or more of a super surface, a quantum dot film, a two-dimensional material layer, and a chemical coating layer.

3. The optical field coding device according to claim 2, wherein: The coding regions are selected from supersurfaces, and the heights of the supersurfaces in all the coding regions are equal.

4. The optical field coding device according to claim 1, wherein: The coding regions are arranged in an array on the coding layer.

5. A method for designing a light field coding device, characterized in that: include: Providing a variety of different metasurface micro-nanostructure units to form a selection set; Simulating all the metasurface micro-nanostructure units to obtain corresponding wide-spectrum response characteristics; A greedy algorithm is used to select a preset number of target sets from the candidate set, wherein the metasurface micro-nanostructure units of the target sets have the lowest correlation with each other; Each super-surface micro-nanostructure unit in the target set is used as a coding region to form a coding layer.

6. The design method according to claim 5, characterized in that: The target set screening process specifically includes: Select a metasurface micro-nanostructure unit and use the corresponding wide spectrum response characteristics as a primitive; Calculating the correlation coefficient between the primitive and the wide spectrum response characteristics of the remaining metasurface microstructure units, and selecting a metasurface micro-nanostructure unit with the smallest correlation coefficient to be included in the observation matrix; Traversing all the metasurface microstructures as the primitives, and incorporating the corresponding metasurface microstructure units into the observation matrix; A group with the smallest average correlation coefficient is selected from the observation matrix as the target set.

7. The design method according to claim 5, characterized in that: The structural parameters of the metasurface micro-nanostructure units are different, and the structural parameters include period P, duty cycle a, line width w, and rotation angle θ.

8. The design method according to claim 7, characterized in that: The value ranges of the structural parameters are period P=1μm-5μm, a=0.3-0.8, w=1 / 3a-2 / 3a, θ=0-80°.

9. A spectral imaging observation system, characterized in that: include: The light field coding device, incident module, and analysis module according to any one of claims 1 to 4; The incident module is used to transmit the target light to the light field encoding device, and the parsing module stores a deep neural network model for parsing the patterned light signal emitted by the sensing layer and restoring the reconstructed spectrum of the target light.

10. The model training method in the spectral imaging observation system according to claim 9, characterized in that: include: Provide a variety of training incident lights with different reconstructed spectra, and use the reconstructed spectra as training labels; Allowing the training incident light to irradiate the light field encoding device through the incident module to obtain a patterned light signal as training data; The deep neural network model is trained using the training data and the training labels so that the reconstructed spectrum output by the deep neural network model is close to the training labels.

Images