A flexible programmable light field pixelated modulation device

Abstract

The application discloses a kind of flexible programmable light field pixelization regulation and control devices.The device includes flexible substrate, light emitting element array, matrix addressing driving circuit and controller.Controller is configured to receive target light field distribution data and calculate the driving parameter of each light emitting element unit, support on-demand switching between mode one and mode two;Mode one is based on spatial interpolation method and is decomposed into the driving parameter of each light emitting element unit;Mode two is based on machine learning model end-to-end calculation of the driving parameter of each light emitting element unit.The device can automatically select the optimal control mode for the two-dimensional Fourier transform feature evaluation of target light field distribution, and has built-in PID and AI double time scale cascade control and safety limit, to realize light biological safety protection.AI supports online self-learning to continuously improve the synthesis accuracy.The device is closely attached to various shapes of target surface through flexible substrate, to realize pixel-level accurate programming distribution of light intensity, wavelength and light color.The device is suitable for skin partition phototherapy, optogenetic neural regulation, light-cured 3D printing, display backlight and digital PCR detection and other fields.

Description

Technical Field

[0001] This invention belongs to the field of light field modulation and semiconductor optoelectronics technology, specifically relating to a flexible programmable light field pixelation modulation device that can be attached to the surface of various shaped targets. This device achieves pixel-level precise programming distribution of light intensity, wavelength, and color within the target space through an independently addressable array of light-emitting elements and a dual-mode light field synthesis algorithm. It is particularly suitable for applications requiring precise spatial light field control, such as skin zone phototherapy, optogenetic neuromodulation, industrial photocuring, and display backlighting. Background Technology

[0002] Light field modulation refers to the technology of precisely controlling parameters such as light intensity, wavelength, color, phase, and polarization at various locations within a target space. With the rapid development of solid-state light source technology, light field modulation has evolved from traditional single uniform illumination to the intelligent requirement of spatially programming the light field distribution within a target space. In optogenetics, there is a need to precisely activate specific brain regions without interfering with adjacent areas; in phototherapy for skin, different regions require differentiated illumination schemes with different wavelengths and intensities; and in digital PCR fluorescence detection, multi-wavelength excitation light is needed to precisely cover each reaction unit of the micropore array.

[0003] Existing light field manipulation devices generally suffer from the following substantial defects.

[0004] First, the layout of the light-emitting elements is fixed, and the light field distribution is determined by the hardware. Traditional LED arrays use rigid PCBs, and the position and orientation of the light-emitting elements cannot be changed once manufactured, making it impossible to dynamically adjust the light field distribution according to the shape and size of the target area. For curved or irregular target surfaces, rigid PCBs cannot be fitted, resulting in reduced light field transmission efficiency and severe edge attenuation.

[0005] Second, the light field distribution is not programmable. Existing systems can typically only adjust the overall brightness or switch between simple zones, and cannot achieve precise pixel-level programming of light intensity, color, and wavelength in space. For example, the light intensity uniformity of existing LED surface light sources is typically between 20% and 30%, which is far from meeting the requirements of precision detection.

[0006] Third, there is a lack of spatial light field synthesis algorithms. Existing systems cannot automatically solve for the driving parameters of each emitting element based on the target light field distribution data. For an array of N emitting elements, the target light field distribution is a continuous function. How to inversely decompose this continuous function into N discrete driving parameters is a typical underdetermined inverse problem. The lack of an effective inverse solution algorithm leads to low light field synthesis accuracy and uneven edge transitions.

[0007] Fourth, the multi-wavelength coordinated control capability is insufficient. Existing devices rely on filter wheels or mechanical switching when switching between different wavelength excitation light, with switching times exceeding hundreds of milliseconds, making it impossible to achieve real-time multi-wavelength parallel excitation and spatial mixing.

[0008] Fifth, there is a lack of AI-driven parameter self-optimization capabilities. Existing devices rely on repeated manual adjustments to the driving parameters of the light-emitting elements, lacking the ability to automatically and adaptively optimize the light field distribution based on the optical properties of the target surface and the application scenario.

[0009] Therefore, there is an urgent need for a flexible programmable light field pixelation control device that can closely fit the target surface, has independently controllable light-emitting elements, supports pixel-level programming of light field distribution, has dual-mode driving capability, and can automatically select the optimal control mode according to the target characteristics.

[0010] It should be noted that the information disclosed in the background section above is only for understanding the background of this application, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0011] This invention aims to overcome the shortcomings of existing light field control devices, such as fixed layout of light-emitting elements, non-programmable light field distribution, and lack of AI adaptation, and provides a pixelated light field control device based on a flexible programmable light-emitting element array. This device achieves close adhesion to the target surface through a flexible substrate, employs an independently addressable multi-wavelength light-emitting element array and a dual-mode light field synthesis algorithm to achieve pixel-level precise programming distribution of light intensity, wavelength, and color within the target space, and supports automatic selection of the optimal control mode based on the target light field characteristics.

[0012] To achieve the above objectives, the present invention adopts the following technical solution.

[0013] In a first aspect of the present invention, a flexible programmable light field pixelation control device is characterized by comprising: A flexible substrate (1) has a lower surface with a shape that matches the contour of the target surface and is fixed relative to the target surface by an adhesive layer or a mechanical fixing device, and remains fixed after being attached; The light-emitting element array (2) is disposed on the flexible substrate (1). The light-emitting element array (2) contains multiple light-emitting element units arranged in a matrix manner. Each light-emitting element unit is independently controllable and is used to output a controllable light field to the target space. A matrix addressing drive circuit (3) is electrically connected to the light-emitting element array (2) to provide an independent and controllable current or pulse width modulation drive signal for each light-emitting element unit; A controller (5) is communicatively connected to the driving circuit (3) and is configured to: receive target light field distribution data, calculate the driving parameters of each light-emitting element unit based on the target light field distribution data, and output the driving parameters to the light-emitting element array (2) through the driving circuit (3), so that each light-emitting element unit can work together to synthesize a light field that approximates the target light field distribution in the target space.

[0014] Furthermore, the controller (5) is programmed to select or automatically switch between mode one and mode two as needed: (a) Mode 1 is a target field inverse decomposition and virtual center synthesis algorithm, which decomposes the continuous target light field into driving parameters of each light-emitting element unit based on the spatial interpolation method. (b) Mode 2 is an AI end-to-end driven synthesis algorithm, which calculates the driving parameters of each light-emitting element unit from the target light field distribution data end-to-end based on a machine learning model.

[0015] Furthermore, the spatial interpolation method is selected from at least one of inverse distance weighted interpolation, Kriging interpolation, radial basis function interpolation, or spline interpolation.

[0016] Furthermore, the automatic switching is triggered in the following manner: a two-dimensional Fourier transform is performed on the target light field distribution to extract the spatial spectrum mean, and the number of wavelength switching times in the target area is counted; when the spatial spectrum mean is lower than a preset threshold and the number of wavelength switching times is lower than a preset threshold, mode one is automatically selected; otherwise, mode two is automatically selected.

[0017] Furthermore, in the second mode, the machine learning model deployed in the controller (5) is used to calculate the driving parameters of each light-emitting element unit from the target light field distribution data end-to-end. The machine learning model is selected from at least one of fully connected neural networks, convolutional neural networks, Transformers, graph neural networks, recurrent neural networks, long short-term memory networks, generative adversarial networks, diffusion models, or other end-to-end generative models based on deep learning.

[0018] Furthermore, the controller (5) is also programmed to: fine-tune the machine learning model in mode two online based on the deviation data between the actual light field distribution and the target light field distribution collected by the light field feedback module (8) during the operation of the device after deployment, so that the synthesis accuracy gradually improves with the increase of usage times; the online fine-tuning adopts at least one of the following: low learning rate gradient descent method, elastic weight consolidation, and incremental learning method based on playback cache; when the synthesis error detected by the light field feedback module (8) exceeds the preset threshold N times consecutively, fine-tuning is triggered, where N is an integer in the range of 3 to 50.

[0019] Furthermore, when inverse distance weighted interpolation is used in Mode 1, the target field inverse decomposition and virtual center synthesis algorithm includes the following steps: (a) receiving target light field distribution data; (b) setting one or more virtual light field centers (7) in the target space; (c) using inverse distance weighting to calculate the light intensity weight of each pixel in the target space, the weight being determined by the spatial distance and exponent between each virtual light field center (7) and each light-emitting element unit, the closer the distance, the greater the weight; the farther the distance, the smaller the weight; (d) mapping the weight values ​​to the driving parameters of each light-emitting element unit.

[0020] Furthermore, the distance is at least one of Euclidean distance, Manhattan distance, Chebyshev distance, Minkowski distance, Mahalanobis distance, or geodesic distance; the power exponent is in the range of 0.5 to 10.

[0021] In some embodiments, the light-emitting element unit is selected from at least one of micro light-emitting diodes, micro-LEDs, organic light-emitting diodes, quantum dot light-emitting diodes, quantum dot color conversion Mini-LEDs, laser diodes, vertical-cavity surface-emitting lasers, perovskite light-emitting diodes, carbon quantum dot light-emitting diodes, graphene quantum dot light-emitting diodes, two-dimensional material exciton light-emitting diodes, and electroluminescent devices; each light-emitting element unit contains at least two light-emitting chips with different center wavelengths, the wavelengths being selected from any wavelength in the ultraviolet UV-C to near-infrared NIR-II range; the light-emitting chips in the same unit can be controlled independently.

[0022] Furthermore, the device also includes an optical field feedback module (8), which includes at least one photoelectric sensor selected from at least one of photodiode, avalanche photodiode, complementary metal-oxide-semiconductor image sensor, charge-coupled device image sensor, multispectral sensor, and imaging spectrometer; the optical field feedback module (8) collects actual optical field distribution data and feeds it back to the controller (5) for closed-loop control and model fine-tuning.

[0023] In some embodiments, the matrix addressing drive circuit (3) adopts a passive matrix architecture or an active matrix architecture. The thin-film transistors of the active matrix are selected from at least one of amorphous silicon thin-film transistors, low-temperature polycrystalline silicon thin-film transistors, indium gallium zinc oxide thin-film transistors, or organic thin-film transistors. The size of the light-emitting element array (2) is M rows × N columns, where M and N are both integers not less than 1, and M × N is not less than 2. The center spacing of the light-emitting element units is 0.05 mm to 50 mm.

[0024] Furthermore, the device also includes an optical control element selected from at least one of a microlens array, a Fresnel lens array, a diffractive optical element, a metasurface lens, a photonic crystal, or an optical waveguide.

[0025] Furthermore, the device also includes a thermally conductive layer (10a) with a thermal conductivity of not less than 1 W / (m·K) and a light transmittance of not less than 70% when placed in an optical path.

[0026] Furthermore, the device also includes a multispectral sensor or imaging spectrometer for online calibration of the light-emitting element array (2), and the calibration data is stored in the non-volatile memory of the controller (5).

[0027] In some embodiments, the flexible substrate (1) is made of at least one of polyimide, polyethylene terephthalate, polydimethylsiloxane, thermoplastic polyurethane, polyethylene naphthalate, liquid crystal polymer, polyetheretherketone, and polycarbonate, and has a thickness of 5 μm to 2 mm; the controller (5) is made of at least one of microcontroller, digital signal processor, field programmable gate array, application-specific integrated circuit, system-on-a-chip, or neural network processor.

[0028] Furthermore, the training data for the machine learning model deployed in the controller (5) is generated by the following method: in the optical simulation environment, for a large number of randomly generated target light field distributions, the optimal driving parameters of each light-emitting element unit are iteratively searched by a global optimization solver, and the target light field distribution-optimal driving parameter pair is constructed as a training dataset; the optical simulation environment adopts at least one of the following: an optical simulation environment based on Monte Carlo ray tracing, a simulation model based on physical optics, or a wave optics simulation model; the global optimization solver is selected from at least one of the following: genetic algorithm, Bayesian optimization, simulated annealing, particle swarm optimization, or differential evolution.

[0029] Furthermore, the controller (5) is also programmed to execute a coarse-fine dual-layer driving parameter calculation strategy: the first layer generates a set of coarse-adjustment driving parameters based on the target light field distribution using either mode one or mode two; the second layer receives the actual light field distribution collected by the light field feedback module (8) through a proportional-integral-derivative closed-loop controller, generates a fine-adjustment correction amount and superimposes it onto the coarse-adjustment driving parameters, and the final driving parameters after superposition are output to the light-emitting element array (2) via the driving circuit (3).

[0030] Furthermore, the controller (5) is also programmed to perform PID and AI dual-time-scale cascade control: the first loop PID fast closed loop is updated at a frequency of not less than 100Hz according to the light field feedback module (8) to fine-tune the driving parameters in real time to ensure photobiological safety limits; the second loop AI slow online self-learning updates the model weights according to the long-term deviation statistics data at an update frequency of 0.1Hz to 10Hz; wherein the hard-coded safety limits of the PID fast loop take precedence over the output of the AI ​​slow loop, and when a conflict occurs, the PID limiter covers the AI ​​output.

[0031] Furthermore, the light-emitting element array (2) and controller (5) support three-dimensional stereoscopic light field distribution control; in the three-dimensional mode, the target light field distribution is a three-dimensional voxel grid, and the distance calculation in the spatial interpolation method and the input of the machine learning model are both extended to three dimensions.

[0032] In some embodiments, the light field feedback module (8) includes at least one photoelectric sensor, which is selected from at least one of photodiode, avalanche photodiode, silicon photomultiplier tube, complementary metal oxide semiconductor image sensor, charge-coupled device image sensor, photoresistor or multispectral camera, for real-time acquisition of the actual light field distribution in the target space and feedback to the controller (5) to form closed-loop regulation; the dynamic range of the photoelectric sensor is 40dB to 140dB, and the spectral response range covers 200nm to 2000nm.

[0033] In some embodiments, the device further includes an optical isolation element and / or a heat insulation layer (10b), wherein the optical isolation element is selected from at least one of a microlens array, a Fresnel lens array, a diffractive optical element, a holographic diffuser, or an optical waveguide, and is used to shape and guide the light beam output by the light-emitting unit array (2); the heat insulation layer (10b) has a thermal conductivity of less than 0.5 W / (m·K) and is used to provide effective thermal isolation in the optical path.

[0034] Furthermore, the controller (5) is also programmed to execute a photobiological safety strategy: calculate the light intensity attenuation within the tissue according to the Beer-Lambert law, and constrain the upper limit of the drive parameters according to the maximum permissible exposure safety limit; when the light field feedback module (8) detects that the output light intensity is about to exceed the MPE, the PID fast loop reduces the drive parameters to a safe range within a single control cycle.

[0035] In some embodiments, the light-emitting unit is selected from at least one of micro light-emitting diodes, micro-LEDs, organic light-emitting diodes, quantum dot light-emitting diodes, quantum dot color conversion Mini-LEDs, laser diodes, vertical cavity surface-emitting lasers, or micro plasma light sources; the center wavelength range of the light-emitting unit is 280nm to 1550nm, covering the ultraviolet, visible, and near-infrared bands; the external dimensions of a single light-emitting unit are 1μm×1μm to 1mm×1mm.

[0036] In some embodiments, the matrix addressing drive circuit (3) adopts a thin-film transistor backplane architecture. The thin-film transistor is selected from at least one of amorphous silicon thin-film transistors, low-temperature polycrystalline silicon thin-film transistors, indium gallium zinc oxide thin-film transistors, or organic thin-film transistors. It supports pixel-level light emission control with independent addressing, and the driving current ranges from 0.1μA to 100mA. The grayscale resolution is not less than 12 bits.

[0037] In some embodiments, the flexible substrate (1) is made of at least one of polyimide, polyethylene terephthalate, polydimethylsiloxane, thermoplastic polyurethane, polyethylene naphthalate, liquid crystal polymer, polyetheretherketone, or polycarbonate, and has a thickness of 5 μm to 2 mm.

[0038] In some embodiments, the quantum dot emitting layer or carbon quantum dot emitting layer in the above-mentioned light-emitting element unit is prepared using a solution method, which includes, but is not limited to, spin coating, inkjet printing, and slot coating. The quantum dot emitting layer or carbon quantum dot emitting layer generates exciton light through electron-hole recombination after an electric field is applied. Its device structure includes an anode, a hole injection layer, an emitting layer, an electron transport layer, and a cathode. The carbon quantum dot emitting layer is synthesized using a hydrothermal method or a microwave-assisted method.

[0039] In a second aspect of the invention, a computer-readable storage medium having a computer program stored thereon, characterized in that, when executed by a processor, the program implements a control method for the apparatus as claimed in any one of claims 1 to 23.

[0040] The present invention has the following beneficial effects.

[0041] (1) Flexible and adaptable with strong versatility. Using a flexible substrate, it can be applied to target surfaces of various shapes in one go to achieve gapless optical coupling. The flexible array can be rolled up and bent to adapt to different curved surfaces. Those skilled in the art can adjust the array size and control parameters according to the shape of the target surface without the need for additional creative design.

[0042] (2) Pixel-level programmable light field. Through independently addressable light-emitting element arrays and dual-mode light field synthesis algorithm, the pixel-level precise distribution of light intensity, wavelength and color in space can be achieved, and the light field non-uniformity can be controlled to within 5%.

[0043] (3) Dual-mode intelligent switching. For simple, low-frequency light field distributions, the system automatically selects a high-efficiency spatial interpolation mode; for complex, high-frequency light field distributions, the system automatically selects a high-precision AI mode, balancing efficiency and accuracy.

[0044] (4) Artificial intelligence adaptive optimization. AI algorithms are used to automatically optimize the driving parameters of the light-emitting element based on the optical properties of the target surface and the application task. Online self-learning continuously improves the synthesis accuracy, significantly improving the adaptability to different scenarios and the consistency of light field quality.

[0045] (5) Real-time collaborative output of multiple wavelengths. Each light-emitting element unit can integrate multiple light-emitting chips of different wavelengths. Through independent PWM control, multiple wavelengths can be mixed in any proportion in the same spatial position. The wavelength switching time is reduced from the hundreds of milliseconds level of traditional filter wheels to the microsecond level.

[0046] (6) Closed-loop feedback ensures long-term stability. The actual light field distribution is monitored in real time and automatically corrected by the light field feedback module. This can compensate for deviations caused by factors such as aging of light-emitting elements, temperature drift, and ambient light interference, ensuring the long-term stability and repeatability of the light field distribution. Closed-loop control can compensate for long-term optical power drift to within 3%.

[0047] (7) Comprehensive safety protection. Built-in PID and AI dual time-scale cascaded control and MPE limit constraints ensure that the output light field is within the photobiological safety range in various operating modes, making it suitable for medical and cosmetic applications for human use.

[0048] (8) Costs are controllable and mass production is easy. The light-emitting element array can be mass-produced using standard Mini-LED or Micro-LED mass transfer technology, and the flexible substrate and driving circuit can be manufactured using mature flexible printed circuit technology, so the overall material cost is controllable.

[0049] (9) Nanomaterial enhancement. Carbon quantum dots and graphene quantum dots are introduced as novel electroluminescent materials. They are combined with silver nanowires and graphene transparent flexible electrodes to replace traditional ITO electrodes, which greatly improves the bending reliability and optical transmittance of the flexible array. Plasmon nanoantenna arrays are used to enhance the local light field intensity and improve the fluorescence detection sensitivity. An upconversion nanoparticle wavelength conversion layer is integrated to realize a non-traditional wavelength conversion path of near-infrared excitation to visible light emission.

[0050] Other beneficial effects of the embodiments of the present invention will be further described below. Attached Figure Description

[0051] Figure 1This is a schematic diagram of the overall structure of the device of the present invention. The diagram shows the spatial relationship of the flexible substrate (1), the light-emitting element array (2), the matrix addressing driving circuit (3), the controller (5), the virtual light field center (7), the light field feedback module (8), the microlens array (9), and the transparent heat-conducting layer (10a).

[0052] Figure 2 This is a top view of the structure of the light-emitting element array. The figure shows the arrangement of the 16×16 light-emitting element array (2) on the flexible substrate (1) and the wiring method of the row and column addressing wires.

[0053] Figure 3 This is a schematic diagram of a single multi-wavelength light-emitting element unit. The diagram distinguishes the layout of three chips by different shades of gray, where light gray represents the red light chip, medium gray represents the blue light chip, and dark gray represents the near-infrared chip.

[0054] Figure 4 This is a schematic diagram of the virtual light field center synthesis algorithm, showing the spatial distance relationship between the virtual light field center (7) and each light-emitting element unit.

[0055] Figure 5 The flowchart shows the dual-mode switching process, where the controller (5) automatically selects either mode one or mode two based on the spatial spectrum and wavelength switching number of the target light field distribution.

[0056] Figure 6 The online self-learning flowchart shows that the light field feedback module (8) collects the deviation data between the actual light field and the target light field. When the deviation exceeds the preset threshold for N consecutive times, the controller (5) triggers the online fine-tuning of the machine learning model.

[0057] Figure 7 The diagram shows a cascaded control architecture with PID and AI, illustrating the timing relationship and safety limit priority of the first loop PID fast closed loop and the second loop AI slow self-learning. Detailed Implementation

[0058] The embodiments of the present invention will be described in detail below. It should be emphasized that the following description is merely exemplary and not intended to limit the scope and application of the present invention.

[0059] It should be noted that when a component is referred to as "fixed to" or "set on" another component, it can be directly on or indirectly on that other component. When a component is referred to as "connected to" another component, it can be directly connected to or indirectly connected to that other component. Furthermore, a connection can be used for fixing, coupling, or communication.

[0060] It should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", and "outer" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing the embodiments of the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the present invention.

[0061] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined with "first" or "second" may explicitly or implicitly include one or more of that feature.

[0062] General Description. Target Surface. The "target surface" as used in this invention refers to any surface on which a programmable light field distribution needs to be generated in the space above it, including but not limited to planes, curved surfaces, spheres, cylinders, cones, freeform surfaces, human skin surfaces (face, torso, limbs), animal skull or brain tissue surfaces, the bottom surface of the reaction chamber of a microfluidic chip, tissue phantom surfaces, curved surfaces of building interior walls, surfaces of industrially cured workpieces, display panel backplates, etc. Those skilled in the art should understand that the size, shape, and number of light-emitting element units of the light-emitting element array (2) can be adjusted according to the specific shape and size of the target surface. For example, a rolled-up bonding method can be used for cylindrical surfaces, and a planar bonding method can be used for planar surfaces. By adjusting the array layout and control parameters, this invention can be adapted to different target surfaces.

[0063] General Description. Types of Light-Emitting Elements. The "light-emitting element unit" described in this invention refers to any device or apparatus capable of converting an input electrical driving signal into an output light field energy within a target space. The micro-LEDs, Micro-LEDs, OLEDs, QLEDs, laser diodes, and VCSELs listed in the above embodiments are merely exemplary light-emitting element types and are not intended to limit the scope of protection of this invention. Any other type of light-emitting element with similar light field output functionality, including but not limited to perovskite light-emitting diodes, carbon quantum dot light-emitting diodes, graphene quantum dot light-emitting diodes, two-dimensional material exciton light-emitting diodes, and electroluminescent devices, as long as they can generate a light field energy output positively correlated with their driving parameters within the target space, can be used as a constituent unit of the light-emitting element array of this device and all fall within the scope of protection of this invention.

[0064] General Description. Array Specifications and Arrangement. The scale of the light-emitting element array (2) described in this invention is not limited by the specific implementation. The number of rows M can be 1 to 10000, the number of columns N can be 1 to 10000, and M×N≥2. The array arrangement includes, but is not limited to, rectangular matrix arrangement, staggered arrangement, ring arrangement, radial arrangement, and non-uniform density arrangement. The center-to-center spacing of the light-emitting element units can be 0.05mm to 50mm, and in a preferred embodiment, it is 0.1mm to 10mm. The shape of the light-emitting element units in the drawings is only schematic. The actual external dimensions of a single light-emitting element unit can be any geometric shape, including but not limited to square, rectangle, circle, triangle, hexagon, etc. Those skilled in the art can select a suitable shape according to the process conditions and light field resolution requirements, all of which fall within the protection scope of this invention.

[0065] Application Area Description. The flexible programmable light field pixelation control device provided by this invention is a general platform. The above embodiments are only used to exemplify some applications of this invention and are not intended to limit the scope of application of this invention. Based on the principles of this invention, this device can be widely used in any field requiring precise spatial light field control, including but not limited to skin zone phototherapy, optogenetic neural modulation, biological detection, digital PCR parallel fluorescence excitation and readout, intelligent lighting and architectural lighting design, ultraviolet curing 3D printing, display backlighting and local dimming, visible light communication, optical tweezers array manipulation, plant factory supplemental lighting, fluorescence microscope excitation light source, surface plasmon resonance sensing excitation, etc. These applications all fall within the protection scope of this invention. For planar applications requiring precise positioning and thermal management, such as chip heat dissipation and semiconductor wafer inspection, the substrate can be a rigid substrate (such as a ceramic substrate, aluminum substrate, silicon substrate, glass substrate, or printed circuit board), providing good thermal conduction path and mechanical stability while maintaining planar rigidity. The above-mentioned rigid substrate implementation scheme also falls within the protection scope of this invention.

[0066] General Specifications Regarding Size and Materials. The size and materials of the light-emitting element unit described in this invention are not limited to specific embodiments. The external dimensions of the light-emitting element unit can be selected from 1μm × 1μm to 10mm × 10mm, depending on the spatial resolution requirements of the application scenario. The light-emitting element can be made of III-V group semiconductor materials such as GaN-based, AlGaInP-based, and InGaAs-based materials, or it can be made of organic light-emitting materials, quantum dot materials, or perovskite materials. The wavelength covers the ultraviolet to near-infrared range. Those skilled in the art can select appropriate sizes and materials based on the target wavelength, manufacturing cost, and process conditions, all of which fall within the protection scope of this invention.

[0067] General Description of Substrate Material. The substrate (1) described in this invention can be flexible or rigid. Flexible substrates can be made of materials selected from polyimide, polyethylene terephthalate, polydimethylsiloxane, thermoplastic polyurethane, polyethylene naphthalate, liquid crystal polymers, polyetheretherketone, polycarbonate, Parylene-C, etc., for bonding to curved target surfaces. Rigid substrates can be made of materials selected from at least one of FR-4 epoxy glass cloth laminate, ceramic substrate, aluminum substrate, copper substrate, glass substrate, silicon substrate, or printed circuit board, for planar target surfaces. Those skilled in the art should understand that the core of this invention is the collaborative architecture of the programmable light-emitting element array and the matrix addressing driving circuit. The choice of substrate material depends on the specific application scenario. Both flexible and rigid substrates can realize the programmable pixelated light field modulation function of the light-emitting element array, and both fall within the protection scope of this invention.

[0068] Working principle of a single light-emitting element unit. Each light-emitting element unit contains at least one light-emitting chip. When the driving circuit (3) selects the unit through row and column addressing and injects current into it, the light-emitting chip emits light of corresponding intensity according to the magnitude of the injected current. In this invention, a single light-emitting element unit does not work in isolation. Instead, the controller (5) allocates the driving parameters of each unit according to the virtual light field center algorithm or AI-driven synthesis mode, so that the emitted light of all light-emitting element units is superimposed in intensity within the target space. Since light is incoherently superimposed, the light of each light-emitting element unit propagates and superimposes independently in space. Each light-emitting element unit can be regarded as an "optical pixel" constituting a programmable light field. The cooperative output of all pixels forms the desired light field distribution within the target space. The luminous intensity of each light-emitting element unit is independently controllable, and the luminous color can be continuously adjusted by the PWM ratio of multiple wavelength channels.

[0069] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0070] Example 1: Flexible programmable light field pixelation control device for skin zone phototherapy This embodiment provides a specific implementation scheme for a skin phototherapy device.

[0071] 1. Hardware configuration (e.g.) Figure 1 (As shown) The flexible substrate (1) is a 50μm thick polyimide film with a size of 50mm×50mm. Its bottom surface is statically bonded to the target skin area in one go using medical-grade double-sided adhesive.

[0072] The light-emitting element array (2) has a size of 32×32, with a total of 1024 light-emitting element units (e.g. Figure 2As shown), a flip-chip Mini-LED, measuring 150μm × 150μm, is bonded to a flexible substrate (1) using anisotropic conductive adhesive. Each light-emitting element unit integrates three light-emitting chips: a blue light chip with a center wavelength of 450nm for acne treatment; a red light chip with a center wavelength of 633nm for promoting collagen regeneration; and a near-infrared chip with a center wavelength of 830nm for deep tissue penetration. The three chips are arranged in a triangular pattern within the unit and are each driven independently by PWM.

[0073] The matrix addressing drive circuit (3) adopts an active matrix architecture. Each light-emitting element unit is equipped with an IGZO thin-film transistor switch and a storage capacitor, supporting 16-bit PWM grayscale control. The driver IC uses Texas Instruments TLC59582, which supports 48 channels of constant current output, with each channel's current adjustable from 0.1mA to 30mA.

[0074] The controller (5) adopts a dual-core heterogeneous system-on-a-chip. The first core, ARM Cortex-M7, runs real-time driver parameter calculation and virtual center update with an update frequency of 100Hz; the second core, ARM Cortex-M4, runs AI inference and communication.

[0075] The light field feedback module (8) integrates eight silicon photodiodes at the edge of the flexible substrate (1), which are evenly distributed around the treatment area to monitor the light intensity distribution on the target surface in real time.

[0076] The optical control element integrates a microlens array (9) above the light-emitting element array (2). Each microlens covers 2×2 light-emitting element units, collimates the emitted beam, compresses the beam divergence angle from 120° to about 30°, and improves the optical power density of the skin surface.

[0077] AI models employ, for example Figure 6 The supervised learning training paradigm shown uses a fully connected neural network as the policy network. The input is the optical characteristic vector of the target skin region and the lesion type encoding. The output is the predicted PWM duty cycle value of each light-emitting element unit and the position and exponent of the virtual light field center (7). The training data is constructed by randomly generating a large number of target light field distribution samples in the optical simulation environment and using the optimal driving parameters corresponding to each light field distribution to calculate the labels using full-wave optical simulation. After training, the model is quantized into 8-bit integer weights and deployed on the Cortex-M4 core of the controller. The inference time is less than 10ms.

[0078] 2. Workflow Step S1: Apply the flexible array to the patient's target skin area and press it firmly with a medical elastic bandage.

[0079] Step S2: The doctor sets the treatment plan through the human-computer interaction interface, setting blue light (450nm, power density 30mW / cm²) on both sides of the nose and red light (633nm, power density 15mW / cm²) on the cheek area, and uploads the target light field distribution map.

[0080] Step S3: The controller (5) performs a two-dimensional Fourier transform on the target light field distribution and extracts the spatial spectrum mean. Since the lesion area of ​​the skin zone phototherapy has a clear boundary and a simple wavelength distribution, the spectrum mean is lower than the preset threshold. The controller (5) automatically selects mode one (e.g., ...). Figure 5 (As shown).

[0081] Step S4: The controller (5) executes the target field inverse decomposition algorithm and sets two virtual light field centers (7) according to the target light field distribution. Virtual center A is located in the nasal alar region, with a target wavelength of 450nm and a target light intensity of 30mW / cm²; virtual center B is located in the cheek region, with a target wavelength of 633nm and a target light intensity of 15mW / cm². Using the following... Figure 4 The inverse distance weighting method (power exponent p=2) shown calculates the weight contribution of 1024 light-emitting element units to each virtual center, and the superposition is mapped to the driving parameters.

[0082] Step S5: The driving circuit (3) outputs a PWM signal according to the calculated driving parameters, and 1024 light-emitting element units emit light in tandem, synthesizing the target light field distribution on the skin surface. The light field feedback module (8) collects the light intensity of each area in real time and feeds it back to the controller (5) to execute the following... Figure 7 The PID fast closed-loop loop shown is fine-tuned and corrected.

[0083] Step S6: During the treatment process, the controller (5) continuously calculates the deviation between the synthesized light field and the target light field (e.g., Figure 6 As shown in the diagram, if the deviation is less than 5%, the parameters remain unchanged; if the deviation exceeds the threshold, the drive parameters are adjusted via the PID loop. The device complies with ANSI Z136.1 and IEC 60825-1 photobiological safety standards, and the PID fast loop ensures that the light power density on the skin surface is always within the MPE safety limit. The output is automatically cut off at the end of treatment.

[0084] 3. Effect Verification Based on simulation analysis and literature reports on the performance of similar devices, the expected results when using the device of this embodiment to treat 10 patients for 8 weeks are as follows: Compared with conventional LED masks, the light field non-uniformity is reduced from 25% to 4.3%; the spatial distribution accuracy of light dose in the treatment area (light power density deviation in 90% of the treatment area) reaches ±2mm spatial resolution; and the PID and MPE safety limits ensure zero thermal injury events. AI online self-learning can improve the synthesis accuracy by 12% after week 4 compared to week 1. The stability of light power density in the treatment area (standard deviation / mean ratio) is reduced from 18.2% with conventional LED masks to 3.8%.

[0085] Example 2: High-resolution light field pixelation control device for optogenetic neural modulation This embodiment provides a specific implementation scheme for optogenetic experiments, and its device architecture is as follows: Figure 1 As shown.

[0086] 1. Hardware Components The flexible substrate (1) is a Parylene-C film with a thickness of 25 μm and a size of 20 mm × 30 mm. It is transparent and biocompatible. The bottom contour is molded to match the curvature of the mouse skull and is bonded by a biocompatible photocurable adhesive.

[0087] The light-emitting element array (2) has a size of 32×32, with a total of 1024 Micro-LED units (e.g. Figure 2 As shown, the unit has dimensions of 80μm×80μm, a center-to-center spacing of 150μm, and an emission wavelength of 470nm per unit. It is integrated onto a flexible substrate using a mass transfer process.

[0088] The matrix addressing drive circuit (3) adopts a passive matrix architecture with 32 row lines and 32 column lines. Each Micro-LED is connected in series with a Schottky diode. The row drive uses a constant current source, and the column scanning frequency is 16kHz.

[0089] The controller (5) uses the Xilinx Zynq-7020 SoC. The programmable logic part implements the generation of 32×32 row and column scan timing, and the processor part runs AI inference.

[0090] The light field feedback module (8) integrates four miniature multispectral sensors at the edge of the array, covering the visible to near-infrared bands, with a sampling rate of 50Hz.

[0091] 2. Workflow Step S1: During the surgery, the flexible array is attached to the surface of the mouse skull, covering the primary visual cortex V1 of the target brain region, and fixed with dental cement.

[0092] Step S2: The operator defines the target light field distribution, sets the light intensity to 100% in the (15,18) to (17,20) unit region, and sets it to 0% in the remaining regions.

[0093] Step S3: The controller (5) analyzes the target distribution, which is a simple rectangular region with a single wavelength and a spatial spectrum concentrated in the low frequency range, and automatically selects mode one. A virtual light field center (7) is set in the center of the target region, and the weights are calculated using IDW (p=3, sharp transition) (e.g. Figure 3 (As shown).

[0094] Step S4: Due to the extremely high spatial resolution requirements of optogenetic experiments, Mode 1 failed to adequately suppress light leakage outside the target region. The device detected that the light leakage detected by the light field feedback module (8) in the adjacent region exceeded 5%, and the controller (5) pressed the button as follows. Figure 5 The dual-mode switching logic shown automatically switches to mode two for fine-tuning compensation. The AI ​​model outputs end-to-end driving parameters, suppressing light leakage to 2.3%.

[0095] Step S5: During operation, such as Figure 7 The first PID fast closed-loop circuit (200Hz) shown fine-tunes the drive current in real time based on light field feedback to compensate for the temperature drift of Micro-LEDs. The AI ​​slow loop evaluates the synthesis accuracy every 30 seconds (e.g., Figure 6 As shown in the figure, online fine-tuning with a low learning rate may be necessary.

[0096] 3. Effect Verification Simulation analysis shows that when the device in this embodiment is used for optogenetic neuromodulation, the activation rate of the target neurons based on calcium imaging reaches 95%, and the false activation rate of adjacent neurons is less than 3%. The spatial resolution of the light field (full width at half maximum) can reach 85μm (consistent with the physical size of the light-emitting element unit of 80μm), and the PID 200Hz closed loop ensures that the light pulse response delay is less than 5ms. Online self-learning ensures that the light leakage rate increases by only 0.8% after 8 hours of continuous operation. The MPE limit (470nm continuous wave irradiation of brain tissue: approximately 2.5mW / mm², according to ANSI Z136.1) ensures that the temperature rise of brain tissue is less than 1°C, with no thermal damage.

[0097] Example 3: Flexible programmable light field device for parallel fluorescence excitation and readout in digital PCR This embodiment provides a specific implementation scheme for multi-channel fluorescence parallel detection in digital PCR (dPCR), and its device architecture is as follows: Figure 1 As shown.

[0098] 1. Hardware Components The flexible substrate (1) is a 50 μm thick polyimide film with a size of 75 mm × 25 mm, which is consistent with the size of a standard microscope slide. The bottom surface is statically bonded to the bottom surface of the microfluidic dPCR chip by optical grade double-sided adhesive.

[0099] The light-emitting element array (2) has a size of 48×48, with a total of 2304 light-emitting element units (e.g. Figure 2 As shown in the diagram, the arrangement corresponds one-to-one with the physical layout of the dPCR chip microarray. Each light-emitting element unit integrates four light-emitting chips of different wavelengths: a 470nm blue chip for FAM dye excitation; a 535nm green chip for HEX / VIC dye excitation; a 585nm yellow chip for ROX dye excitation; and a 640nm red chip for Cy5 dye excitation. The four chips are arranged within the unit as follows: Figure 3 The multi-wavelength triangular / square layout shown (four-chip square layout in this embodiment) is driven independently by PWM, supporting simultaneous or time-division sequential excitation of four channels of fluorescence. The external dimensions of a single unit are 100μm×100μm, matching the diameter of the dPCR micropores (approximately 80-120μm).

[0100] The matrix-addressable driver circuit (3) adopts an active matrix architecture. Each light-emitting element unit is equipped with an IGZO thin-film transistor switch and a storage capacitor, supporting 12-bit PWM grayscale control. The driver IC supports 48 channels of constant current output, with each channel's current adjustable from 0.05mA to 20mA. The full refresh rate is 250Hz, meeting the timing requirements for real-time fluorescence acquisition.

[0101] The controller (5) adopts an FPGA+ARM heterogeneous architecture and a Xilinx Zynq-7020 SoC. The FPGA programmable logic part realizes the parallel row and column scanning timing generation of the 48×48 array and the synchronous acquisition and triggering of four-channel fluorescence signals. The ARM processor part runs the AI ​​multicolor fluorescence compensation model.

[0102] The light field feedback module (8) is equipped with a CMOS image sensor above the dPCR chip, with a pixel resolution of 1920×1080, a dynamic range of 72dB, and a frame rate of 30fps (1 to 2 frames of fluorescence images can be acquired within a 30ms acquisition window for signal averaging). In each fluorescence reading cycle, the CMOS acquires multi-channel fluorescence images and feeds them back to the controller (5) for closed-loop adjustment of the excitation light intensity of each micropore and AI multicolor compensation.

[0103] 2. Workflow Step S1: Attach the flexible light field array to the bottom surface of the dPCR chip, aligning the center of each light-emitting element unit with the center of the corresponding micropore. Visual alignment is assisted by a low-power LED illumination mode, achieving an alignment accuracy of ±15μm.

[0104] Step S2: The controller (5) reads the microwell partition information and PCR amplification protocol of the dPCR chip. For singleton PCR reactions, all microwells are set to use a single excitation wavelength (e.g., 470nm excitation for FAM detection). For multiplex PCR reactions (e.g., quadruple respiratory pathogen panel), the microwells are grouped by partition: A zone FAM, B zone HEX, C zone ROX, and D zone Cy5, with each zone set with a corresponding excitation wavelength and power density (typical value 0.5-5mW / mm²).

[0105] Step S3: The controller (5) performs a two-dimensional FFT analysis on the target light field distribution. For single detection (single wavelength, large-area uniform excitation), the spatial spectrum energy is concentrated in the low frequency, and the average spectrum value is lower than the preset threshold. Mode 1 is automatically selected, and the driving parameters are calculated using IDW (p=2). For multiple detection (multiple regions and multiple wavelengths, rich high-frequency components in the spatial spectrum), the controller (5) performs a two-dimensional FFT analysis on the target light field distribution. Figure 5 The dual-mode switching logic shown automatically selects mode two, and the AI ​​model outputs the four-channel drive parameters of each unit end-to-end.

[0106] Step S4: Perform fluorescence excitation and acquisition cycle. Within each cycle (50ms): 0-10ms is the excitation phase, the light-emitting element array outputs according to the driving parameters; 10-15ms is the delay phase, waiting for the fluorescence signal to stabilize after the excitation light is turned off; 15-45ms is the acquisition phase, the CMOS sensor acquires fluorescence images at a frame rate of 30fps; 45-50ms is the data processing phase, the controller (5) performs fluorescence intensity extraction and multicolor compensation calculation.

[0107] Step S5: AI Multicolor Fluorescence Compensation. In multiplex PCR scenarios, the emission spectra of different fluorescent dyes overlap (e.g., the FAM emission tail extends into the HEX channel, and the ROX emission tail extends into the Cy5 channel). The AI ​​model takes the original four-channel CMOS fluorescence image as input and outputs the compensated fluorescence intensity values ​​for each channel of each microwell end-to-end. The model uses... Figure 6 The supervised learning training paradigm shown uses a 3-layer CNN architecture and was trained in a simulation environment with 10,000 sets of simulated multicolor fluorescence data (including different crosstalk ratios, different concentration gradients, and different background noise levels). Deployed on the ARM core of a Zynq-7020, the single inference time is <5ms. Compared with the traditional matrix compensation method, the AI ​​model's compensation error in the low concentration range (1-10 copies / μL) is reduced from an average of 18.3% to 4.7%.

[0108] Step S6: Online self-learning drift compensation. During long-term continuous operation (≥4 hours), the LED output light power slowly decreases due to temperature rise (typical temperature drift coefficient -0.3% / °C), resulting in a decrease in fluorescence excitation efficiency. The CMOS sensor in the light field feedback module (8) collects the fluorescence intensity of each micropore every 30 cycles, and the controller (5) calculates the fluorescence decay trend. When the average fluorescence decay of 50 consecutive samples exceeds 8%, the EWC online fine-tuning is triggered to update the excitation-fluorescence mapping weights in the AI ​​model and compensate the fluorescence decay to ≤3%.

[0109] Step S7: The device complies with IEC 60825-1 laser product safety standard and ANSI Z136.1 photobiological safety standard. For example... Figure 7 The first PID fast loop (200Hz) shown monitors the output optical power of each light-emitting element unit in real time based on CMOS feedback. When the output of any unit exceeds the preset safety limit (5mW / mm²), the PWM duty cycle is reduced to the safe range within a single control cycle.

[0110] 3. Effect Verification Based on simulation analysis, when using the device of this embodiment to detect dPCR quadrivalent respiratory pathogens (influenza A, influenza B, respiratory syncytial virus, and adenovirus), the expected settings are as follows: five concentration gradients for each target (1, 5, 25, 125, and 625 copies / μL), three replicates for each concentration gradient, for a total of 60 detection reactions. Compared with the traditional LED surface light source + mechanical filter wheel solution: (a) Fluorescence detection sensitivity: After AI multicolor compensation, the signal-to-noise ratio of the FAM channel at a concentration of 1 copy / μL increased from 2.1 to 5.8 in the traditional scheme; the HEX channel from 1.8 to 5.2; the ROX channel from 2.3 to 6.1; and the Cy5 channel from 1.5 to 4.9. The average limit of detection (LOD) of the four channels decreased from 3.2 copies / μL to 0.8 copies / μL.

[0111] (b) Crosstalk between channels: The crosstalk rate of the FAM→HEX channel using the traditional compensation matrix method is 12.7%, which is expected to be reduced to 3.1% after AI compensation; the crosstalk rate of the ROX→Cy5 channel can be reduced from 14.2% to 2.8%.

[0112] (c) Detection throughput and speed: This device can analyze 2304 microwells in parallel in a single detection. The total cycle time for four-channel fluorescence excitation and acquisition is 50ms. The time for a single complete four-channel scan is reduced from 4.2 seconds with the mechanical filter wheel to 50ms, and the throughput can be increased by 84 times.

[0113] (d) Long-term stability: After 8 hours of continuous operation, the attenuation rate of fluorescence signals in each channel is ≤2.5% after online fine-tuning compensation by EWC, while the attenuation rate of the uncompensated traditional scheme is about 18%.

[0114] Meanwhile, this embodiment demonstrates the application prospects of this device in large-scale discrete microarray scenarios. For three-dimensional micropore culture systems (such as 3D cell spheres or organoids in micropores), the controller (5) supports extending the above-mentioned two-dimensional micropore array light field scheme to a three-dimensional voxel grid, and independently compensates the light intensity of each depth layer according to the Beer-Lambert law, so as to realize differentiated light dose delivery at different depths in the micropores, covering the three-dimensional light field distribution control described in claim 19.

[0115] Example 4: Nanomaterial-enhanced flexible programmable optical field device for high-throughput drug screening This embodiment provides a specific implementation scheme for high-throughput drug screening in 96-well plates, integrating plasmonic nanoantenna array fluorescence enhancement, upconversion nanoparticle wavelength conversion, and flexible transparent conductive thin-film electrodes, covering the nanomaterial enhancement technology features described in claims 21, 22, and 23. Its device architecture is as follows: Figure 1 As shown.

[0116] 1. Hardware Components The flexible substrate (1) is a 75 μm thick polyethylene terephthalate (PET) film with dimensions of 128 mm × 86 mm, matching the bottom dimensions of a standard 96-well plate. The bottom surface is bonded to the bottom surface of the plate using an optically grade UV-curable adhesive. Furthermore, a transparent conductive silver nanowire film is fabricated on the flexible substrate as a flexible transparent electrode layer (claim 21), replacing the traditional indium tin oxide (ITO) electrode. The silver nanowires have a diameter of 40 nm and a length of 25 μm, and are deposited using the Meyer rod coating method, exhibiting a sheet resistance of 22 Ω / sq and a visible light transmittance of 87%. Compared to the ITO electrode (sheet resistance 45 Ω / sq, resistance increases by 300% when bending radius ≥ 30 mm), the silver nanowire electrode shows a sheet resistance change of <5% at a bending radius of 5 mm, significantly improving the electrical reliability and optical transmittance of the flexible array during repeated bonding and peeling of the plate.

[0117] The light-emitting element array (2) is customized into an 8-row × 12-column array according to the 96-hole plate layout (e.g. Figure 2As shown), each hole corresponds to a light-emitting element unit group. Each unit group contains four types of sub-pixels: the first sub-pixel is a standard blue Micro-LED chip (470nm) for conventional fluorescence excitation; the second sub-pixel is a standard green Micro-LED chip (535nm); the third sub-pixel is a plasmon-enhanced pixel (claim 22), which is a gold nanoparticle dimer array fabricated on the surface of the green Micro-LED chip (535nm) using electron beam lithography and lift-off processes. The nanoparticles have a diameter of 80nm, a gap of 8nm, and an array period of 300nm. The local surface plasmon resonance peak of the gold dimer is located at approximately 530nm, which precisely matches the 535nm excitation wavelength. The local electromagnetic field enhancement factor in the gap region is approximately 8000 times (falling into the 10-100000× range described in claim 22). The fourth sub-pixel is an upconversion emission auxiliary pixel (claim 23), consisting of a 5μm thick layer of NaYF4:Yb(20%),Er(2%) upconversion nanoparticles spin-coated above a near-infrared VCSEL (980nm, vertical-cavity surface-emitting laser). The 980nm excitation light, after anti-Stokes conversion by the upconversion layer, emits 540nm green light and 660nm red light, used for excitating fluorescent proteins within deep cell spheres (≥200μm diameter). Compared to direct visible light (penetration depth approximately 0.3mm), the 980nm near-infrared light can penetrate to a depth of over 1.5mm, overcoming the bottleneck of photoactivation within 3D cell spheres.

[0118] The matrix addressing drive circuit (3) adopts a low-temperature polycrystalline silicon (LTPS) thin-film transistor active matrix architecture. Each light-emitting element unit group is configured with four independent 12-bit PWM channels to drive four types of sub-pixels respectively. The drive current range is adjustable from 0.1μA to 50mA.

[0119] The controller (5) uses NVIDIA Jetson Orin Nano with 8GB of video memory and runs an AI drug response prediction model and a light field synthesis model.

[0120] The light field feedback module (8) is equipped with a multispectral camera above the 96-well plate, covering the spectral range of 350nm to 1700nm, and includes 8 spectral channels (bandwidth 20nm) for simultaneously acquiring conventional fluorescence (FITC, PE, DAPI, etc.), plasmon-enhanced fluorescence and upconversion luminescence signals.

[0121] 2. Workflow Step S1: Attach the flexible light field array to the bottom surface of the 96-well plate, and fix it after the optical-grade UV curing adhesive has cured for 30 seconds.

[0122] Step S2: The operator sets an independent light stimulation and detection scheme for each well using the control software. Taking antitumor drug dose-response screening as an example: the 96-well plate is divided into 12 columns (drug concentration gradient: 0.01, 0.03, 0.1, 0.3, 1, 3, 10, 30, 100, 300, 1000, 3000 nM) and 8 rows (8 candidate compounds + 2 rows of negative / positive controls). For conventional 2D adherent culture wells (rows 1-4), standard blue / green LED pixels are assigned for FITC and PE dual-color fluorescence detection. For 3D tumor spheroid culture wells (rows 5-8), plasmon-enhanced pixels are assigned for fluorescence enhancement detection of low-abundance apoptosis marker (caspase-3), and upconversion pixels are assigned for near-infrared excitation inside the deep spheroids.

[0123] Step S3: The controller (5) generates differentiated target light field distributions according to the detection requirements of each aperture. For standard apertures, mode one is used, with IDW (p=2) uniformly distributing excitation light (e.g., Figure 4 (As shown). For holes requiring fluorescence enhancement, plasmon-enhanced holes are constructed using methods such as... Figure 5 Mode 2 (AI light field synthesis) shown in the figure enhances the pixel coverage area through methods such as... Figure 7 The PID fast loop shown (200Hz) keeps the excitation power density stable within ±2%.

[0124] Step S4: Perform full-plate fluorescence readings every 2 hours. Acquire fluorescence images of the FITC channel (excitation 470nm / emission 520nm) and PE channel (excitation 535nm / emission 575nm) in the regular wells. For the plasmon-enhanced wells, excitation is performed at 535nm and the emission channel at 553nm is acquired using a multispectral camera. The fluorescence signal of caspase-3 labeled with AlexaFluor 532 is amplified to approximately 8000 times that of the unenhanced wells (approximately 3500 times in actual biological samples due to molecular diffusion effects) using a gold nanoantenna local field enhancement (excitation 535nm / emission 553nm). This allows for the detection of apoptosis rates as low as 0.5% in the early stages of drug action (6 hours), while the detection limit for the unenhanced control group is approximately 18% apoptosis rate. The upconversion auxiliary well was excited by a 980nm VCSEL and the 540nm channel was acquired by a multispectral camera. It successfully achieved fluorescence imaging of tdTomato-labeled cells inside a tumor spheroid with a diameter of about 300μm (540nm upconversion emission excites the 554nm absorption band of tdTomato). The penetration depth reached 280μm, while the control group without upconversion (direct excitation at 554nm) could only clearly image the surface layer of the spheroid, about 80μm.

[0125] Step S5: As Figure 7The second AI slow self-learning loop (0.2Hz) shown updates the drug response prediction model based on multispectral data collected every 2 hours, adaptively optimizing the light dose and acquisition parameters for subsequent readings as screening progresses. EWC incremental learning (e.g.) Figure 6 (As shown) Ensure that the model does not suffer from catastrophic forgetting across different drug target categories.

[0126] 3. Effect Verification Based on simulation analysis and literature reports, when using the device in this embodiment to test the antitumor activity of a kinase inhibitor library (128 compounds) in a 96-well plate, the tumor cell line was HCT116 (human colon cancer), and the detection endpoints were caspase-3 activation (apoptosis) and ATP content (activity). Compared with a traditional ELISA reader (Tecan Spark): (a) Detection sensitivity: Plasmon-enhanced fluorescence reduced the detection limit of caspase-3 from 12.5 pg / mL to 0.8 pg / mL (an improvement of about 16 times), consistent with the range of 10-100000× enhancement factor in claim 22.

[0127] (b) 3D penetration depth: Upconversion-assisted excitation can achieve fluorescence imaging at a depth of 280 μm in 3D tumor spheres, which is 3.5 times that of traditional visible light excitation (80 μm), covering the main internal regions of typical 3D spheres (200-500 μm in diameter).

[0128] (c) Reliability of flexible electrodes: The sheet resistance of the silver nanowire transparent electrode only increases by 8% after 1000 bending cycles (bending radius 5mm), while the ITO electrode shows obvious cracks and a sheet resistance increase of more than 500% after 50 cycles, consistent with the reliability advantage of the flexible transparent electrode in wearable / fittable applications as described in claim 21.

[0129] (d) AI self-learning value: During 48 hours of continuous screening, the AI ​​slow loop adaptively adjusts the light dose and acquisition parameters of the later holes based on the fluorescence readings of the earlier holes, which can reduce the inter-hole variation coefficient (CV) of the fluorescence signal of the whole plate from the initial 18% to 4.3% after 6 hours.

[0130] Example 5: Flexible programmable light field device for light modulation in three-dimensional organoid culture This embodiment provides a specific implementation scheme for spatially programmable photostimulation in three-dimensional culture of intestinal organoids, demonstrating the three-dimensional light field distribution control capability as described in claim 19. Its device architecture is as follows: Figure 1 As shown.

[0131] 1. Hardware Components The flexible substrate (1) is a 100 μm thick transparent polydimethylsiloxane (PDMS) film with dimensions of 35 mm × 35 mm, which matches the bottom of a 35 mm cell culture dish and is irreversibly bonded to the bottom surface of the culture dish after treatment with oxygen plasma. PDMS has excellent optical transparency (400-800 nm transmittance > 92%) and biocompatibility.

[0132] The light-emitting element array (2) has a size of 32×32, with a total of 1024 Micro-LED units (e.g. Figure 2 As shown, each unit measures 100μm × 100μm with a center-to-center spacing of 200μm and is integrated onto a PDMS substrate using mass transfer technology. Each light-emitting element unit integrates three light-emitting chips: a 450nm blue chip for Channelrhodopsin-2 (ChR2) optogenetic activation; a 590nm yellow chip for Hallorhodopsin (NpHR) optogenetic suppression; and a 630nm red chip for ReaChR redshift optogenetic activation. The three chips are arranged in a triangular pattern within the unit, each driven independently by a 12-bit PWM.

[0133] The matrix addressing drive circuit (3) adopts an active matrix architecture, IGZO thin film transistors, and three independent PWM channels in each unit.

[0134] The controller (5) uses a Xilinx Zynq-7020 SoC. The FPGA programmable logic section implements parallel row and column scanning of a 32×32 array and three-dimensional voxel light field calculation. The ARM processor section runs a real-time three-dimensional light attenuation model and an AI three-dimensional light field optimization model.

[0135] The light field feedback module (8) is equipped with a micro multispectral sensor and a CMOS image sensor above the culture dish. The multispectral sensor covers the visible light band and is used to monitor the intensity distribution of scattered light at different depths of the organoid in real time.

[0136] 2. Workflow Step S1: Place a Matrigel droplet containing intestinal organoids (approximately 4 mm in diameter, 1.5 mm in height, and containing approximately 50 organoids) in the center of the culture dish. A flexible light field array is then attached to the bottom of the culture dish.

[0137] Step S2: The operator defines the three-dimensional target light field distribution. The organoid is scanned along the Z-axis using a confocal microscope to obtain its three-dimensional spatial coordinates (voxel resolution 20μm×20μm×20μm, totaling approximately 75×75×75 voxels). The target light dose is labeled for each voxel: the Lgr5+ stem cell region (located at the bottom of the crypts) requires 590nm suppression light; the villous region (located at the apex) requires 450nm activation light; and free single cells in the matrix gel require 630nm redshift activation light.

[0138] Step S3: The controller (5) calculates the three-dimensional light attenuation compensation. For each target voxel (x, y, z), the light attenuation from the bottom surface emitting element unit to the target voxel is calculated according to the Beer-Lambert law: I(z) = I0 × exp(-μz), where z is the depth (vertical distance from the bottom surface to the target voxel), and μ is the tissue attenuation coefficient (μ≈28cm⁻¹ at 450nm, μ≈22cm⁻¹ at 590nm, and μ≈18cm⁻¹ at 630nm). Based on the above model, the initial driving parameters required for each target position of each emitting element unit in the three-dimensional voxel grid are calculated.

[0139] Step S4: For a standard culture cycle (light stimulation every 24 hours), the device is operated as follows: Figure 5 The dual-mode switching logic shown adopts mode one, as follows: Figure 4 The IDW principle shown is extended to a three-dimensional distance (p=2) weighted calculation of driving parameters. The distance is a three-dimensional Euclidean distance, taking into account the straight-line distance from each emitting element unit to each target voxel and light attenuation compensation.

[0140] Step S5: For fine-grained optogenetic manipulation (single organoid resolution), the device employs Mode 2. The AI ​​3D optical field optimization model takes the 3D target voxel distribution and attenuation coefficient as input, and outputs an end-to-end 32×32 three-channel (450 / 590 / 630nm) driving parameter matrix. The model is trained in an optical simulation environment (based on Monte Carlo ray tracing, simulating heterogeneous scattering and absorption of organoid tissues) with 5000 sets of random 3D target distributions.

[0141] Step S6: As Figure 7 The first PID fast closed loop (200Hz) shown finely adjusts the PWM duty cycle of each unit based on the real-time reading of the multispectral sensor in the light field feedback module (8). The multispectral sensor collects the emitted light scattered by the organoid from the side of the culture dish, and utilizes the differential scattering characteristics of different wavelengths (in tissue optics, the scattering coefficient is approximately inversely proportional to the wavelength, and the qualitative direction is related to the Rayleigh scattering intensity ∝ 1 / λ). 4 (Consistent) The depth-resolution light field distribution is calculated and fed back to the PID loop for independent compensation of the depth layer.

[0142] Step S7: AI slow self-learning loop (e.g.) Figure 6 (As shown). During the two-week long-term culture, the organoids continued to grow (diameter increased from approximately 100 μm to approximately 400 μm), resulting in gradual changes in light attenuation characteristics. The AI ​​loop updated the weights of the three-dimensional attenuation model and the light field synthesis model every 24 hours based on the multispectral data from the light field feedback module. EWC incremental learning ensured that the training distribution of the original organoid morphology was not forgotten when new knowledge was incorporated into the model.

[0143] 3. Effect Verification Based on simulation analysis, the expected results of using the device of this embodiment to perform three-dimensional optogenetic regulation of intestinal organoids for 14 days are as follows: (a) Depth-resolved activation specificity: The Lgr5-GFP reporter organoid achieved an optogenetic inhibition efficiency of 82% at 590 nm in the bottom of the crypt (approximately 280 μm deep) (percentage of cells with ≥50% reduction in GFP fluorescence intensity), while the false inhibition rate in the non-target region at the top of the villi (approximately 50 μm deep) was only 4.3%. Three-dimensional light attenuation compensation reduced the actual light dose deviation at each depth from -65% to +25% before compensation (overexposure in shallow layers and underexposure in deep layers) to -8% to +6% after compensation.

[0144] (b) Single organoid resolution: The 200μm spacing of the 32×32 array ensures that each organoid with a diameter of approximately 150-300μm is covered by an average of approximately 6-28 luminescent element units, enabling independent photomodulation at the single organoid level. The false activation rate of adjacent organoids (spacing ≥400μm) is <5%.

[0145] (c) Long-term culture stability: During the 14-day culture period, AI self-learning enabled the accuracy of 3D light field synthesis to adaptively optimize with organoid growth. The root mean square error (RMSE) of the target light field versus the actual light field between day 1 and day 14 was expected to decrease from an initial 18.2% to 7.8% by day 7 and stabilize at 6.1% by day 14. In contrast, the deviation of the fixed parameter scheme (without AI self-learning) continued to deteriorate from 18.2% on day 1 to 42.5% on day 14.

[0146] (d) Organoid viability: Compared with the control group without light stimulation, the cell viability (Calcein-AM / PI staining) of the organoids precisely regulated by this device after 14 days of culture was not expected to be statistically significant (p>0.05), indicating that the photobiological safety of the PID safety limit and the Beer-Lambert attenuation compensation met expectations.

[0147] It should be noted that the simulation data and expected performance data in the above embodiments are only used to illustrate and verify the technical solution and expected effects of the present invention. Those skilled in the art should understand that the specific parameters in actual applications can be adjusted according to experimental conditions, application scenarios and device selection, and these adjustments all fall within the protection scope of the present invention.

[0148] The above description, in conjunction with specific and preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various substitutions or modifications can be made to these described embodiments without departing from the inventive concept, and all such substitutions or modifications should be considered within the scope of protection of the present invention.

Claims

1. A flexible programmable light field pixelation control device, characterized in that, include: A flexible substrate (1) has a lower surface with a shape that matches the contour of the target surface and is fixed relative to the target surface by an adhesive layer or a mechanical fixing device; The light-emitting element array (2) is disposed on the flexible substrate (1). The light-emitting element array (2) contains multiple light-emitting element units arranged in a matrix manner. Each light-emitting element unit is independently controllable and is used to output a controllable light field to the target space. A matrix addressing drive circuit (3) is electrically connected to the light-emitting element array (2) to provide an independent and controllable current or pulse width modulation drive signal for each light-emitting element unit; A controller (5) is communicatively connected to the driving circuit (3) and is configured to: receive target light field distribution data, calculate the driving parameters of each light-emitting element unit based on the target light field distribution data, and output the driving parameters to the light-emitting element array (2) through the driving circuit (3), so that each light-emitting element unit can work together to synthesize a light field that approximates the target light field distribution in the target space.

2. The apparatus according to claim 1, characterized in that, The controller (5) is programmed to select or automatically switch between Mode 1 and Mode 2 as needed: (a) Mode 1 is a target field inverse decomposition and virtual center synthesis algorithm, which decomposes the continuous target light field into driving parameters of each light-emitting element unit based on the spatial interpolation method. (b) Mode 2 is an AI end-to-end driven synthesis algorithm, which calculates the driving parameters of each light-emitting element unit from the target light field distribution data end-to-end based on a machine learning model.

3. The apparatus according to claim 2, characterized in that, The spatial interpolation method is selected from at least one of inverse distance weighted interpolation, Kriging interpolation, radial basis function interpolation, or spline interpolation.

4. The apparatus according to claim 2, characterized in that, The automatic switching is triggered as follows: a two-dimensional Fourier transform is performed on the target light field distribution to extract the spatial spectrum mean, and the number of wavelength switching times within the target area is counted; when the spatial spectrum mean is lower than a preset threshold and the number of wavelength switching times is lower than a preset threshold, mode one is automatically selected; otherwise, mode two is automatically selected.

5. The apparatus according to claim 2, characterized in that, In Mode 2, the machine learning model deployed in the controller (5) is used to calculate the driving parameters of each light-emitting element unit from the target light field distribution data end-to-end. The machine learning model is selected from at least one of fully connected neural networks, convolutional neural networks, Transformers, graph neural networks, recurrent neural networks, long short-term memory networks, generative adversarial networks, diffusion models, or other end-to-end generative models based on deep learning.

6. The apparatus according to claim 5, characterized in that, The controller (5) is also programmed to: fine-tune the machine learning model online based on the deviation data between the actual light field distribution and the target light field distribution collected by the light field feedback module (8) during the operation of the device after deployment, so that the synthesis accuracy gradually improves with the increase of the number of uses; The online fine-tuning employs at least one of the following: low learning rate gradient descent, elastic weight consolidation, and incremental learning based on playback cache. Fine-tuning is triggered when the synthesis error detected by the light field feedback module (8) exceeds a preset threshold N times consecutively, where N is an integer in the range of 3 to 50.

7. The apparatus according to claim 3, characterized in that, The inverse distance-weighted interpolation includes the following steps: (a) Receive target light field distribution data; (b) Set one or more virtual light field centers (7) in the target space; (c) Calculate the light intensity weight of each pixel in the target space using the inverse distance weighting method. The weight is determined by the spatial distance and exponent between each virtual light field center (7) and each light-emitting element unit; (d) Map the weight values ​​to the driving parameters of each light-emitting element unit.

8. The apparatus according to claim 7, characterized in that, The distance is at least one of Euclidean distance, Manhattan distance, Chebyshev distance, Minkowski distance, Mahalanobis distance, or geodesic distance; the power exponent is in the range of 0.5 to 10.

9. The apparatus according to claim 1, characterized in that, The light-emitting element unit is selected from at least one of the following: micro light-emitting diode, micro-LED, organic light-emitting diode, quantum dot light-emitting diode, quantum dot color conversion Mini-LED, laser diode, vertical cavity surface emission laser, perovskite light-emitting diode, carbon quantum dot light-emitting diode, graphene quantum dot light-emitting diode, two-dimensional material exciton light-emitting diode, and electroluminescent device; each light-emitting element unit contains at least two light-emitting chips with different center wavelengths, the wavelengths being selected from any wavelength in the ultraviolet UV-C to near-infrared NIR-II range; the light-emitting chips in the same unit can be controlled independently.

10. The apparatus according to claim 1, characterized in that, The device further includes an optical field feedback module (8), which includes at least one photoelectric sensor selected from at least one of photodiode, avalanche photodiode, complementary metal oxide semiconductor image sensor, charge-coupled device image sensor, multispectral sensor, and imaging spectrometer; the optical field feedback module (8) collects actual optical field distribution data and feeds it back to the controller (5) for closed-loop control and model fine-tuning.

11. The apparatus according to claim 1, characterized in that, The matrix addressing drive circuit (3) adopts a passive matrix architecture or an active matrix architecture. The thin film transistors of the active matrix are selected from at least one of amorphous silicon thin film transistors, low-temperature polycrystalline silicon thin film transistors, indium gallium zinc oxide thin film transistors, or organic thin film transistors. The size of the light-emitting element array (2) is M rows × N columns, where M and N are both integers not less than 1, and M × N is not less than 2. The center spacing of the light-emitting element units is 0.05 mm to 50 mm.

12. The apparatus according to claim 1, characterized in that, The device further includes an optical control element, which is selected from at least one of a microlens array, a Fresnel lens array, a diffractive optical element, a metasurface lens, a photonic crystal, or an optical waveguide.

13. The apparatus according to claim 1, characterized in that, The device also includes a thermally conductive layer (10a) with a thermal conductivity of not less than 1 W / (m·K) and a light transmittance of not less than 70% when placed in an optical path.

14. The apparatus according to claim 1, characterized in that, The device also includes a multispectral sensor or imaging spectrometer for online calibration of the light-emitting element array (2), and the calibration data is stored in the non-volatile memory of the controller (5).

15. The apparatus according to claim 1, characterized in that, The flexible substrate (1) is made of at least one of polyimide, polyethylene terephthalate, polydimethylsiloxane, thermoplastic polyurethane, polyethylene naphthalate, liquid crystal polymer, polyetheretherketone, and polycarbonate, and has a thickness of 5 μm to 2 mm; the controller (5) is made of at least one of microcontroller, digital signal processor, field programmable gate array, application-specific integrated circuit, system-on-a-chip, or neural network processor.

16. The apparatus according to claim 1, characterized in that, The training data of the machine learning model deployed in the controller (5) is generated by the following method: In the optical simulation environment, for a large number of randomly generated target light field distributions, the optimal driving parameters of each light-emitting element unit are iteratively searched by the global optimization solver, and the target light field distribution-optimal driving parameter pair is constructed as a training dataset; The optical simulation environment adopts at least one of the following: Monte Carlo ray tracing-based optical simulation environment, physical optics-based simulation model, or wave optics simulation model; the global optimization solver is selected from at least one of the following: genetic algorithm, Bayesian optimization, simulated annealing, particle swarm optimization, or differential evolution.

17. The apparatus according to claim 2, 3 or 4, characterized in that, The controller (5) is also programmed to execute a coarse-fine dual-layer driving parameter calculation strategy: the first layer generates a set of coarse-tuned driving parameters based on the target light field distribution using either mode one or mode two; The second layer receives the actual light field distribution collected by the light field feedback module (8) through the proportional-integral-derivative closed-loop controller, generates fine-tuning correction values ​​and superimposes them onto the coarse-tuning driving parameters, and the final driving parameters after superposition are output to the light-emitting element array (2) via the driving circuit (3).

18. The apparatus according to claim 2 or 4, characterized in that, The controller (5) is also programmed to perform PID and AI dual time-scale cascade control: the first loop PID fast closed loop is updated at a frequency of not less than 100Hz according to the light field feedback module (8) to fine-tune the driving parameters in real time to ensure photobiological safety limits; the second loop AI slow online self-learning is updated at a frequency of 0.1Hz to 10Hz according to the long-term deviation statistics to update the model weights; wherein the hard-coded safety limit of the PID fast loop takes precedence over the output of the AI ​​slow loop, and when a conflict occurs, the PID limiter covers the AI ​​output.

19. The apparatus according to claim 3 or 4, characterized in that, The light-emitting element array (2) and controller (5) support three-dimensional stereo light field distribution control; in the three-dimensional mode, the target light field distribution is a three-dimensional voxel grid, and the distance calculation in the inverse distance weighting method and the input of the machine learning model are both extended to three dimensions.

20. The apparatus according to claim 1, characterized in that, The controller (5) is also programmed to execute a photobiological safety strategy: calculate the light intensity attenuation within the tissue according to the Beer-Lambert law, and constrain the upper limit of the drive parameters according to the maximum permissible exposure safety limit; When the light field feedback module (8) detects that the output light intensity is about to exceed the MPE, the PID fast loop reduces the driving parameters to a safe range within a single control cycle.

21. The apparatus according to claim 1, characterized in that, A flexible transparent conductive film is also provided on the flexible substrate (1) as an electrode layer to carry the driving current from the matrix addressing driving circuit (3) to each light-emitting element unit; the material of the flexible transparent conductive film is selected from at least one of silver nanowire transparent conductive network, graphene transparent conductive film, copper nanowire transparent conductive network or carbon nanotube transparent conductive film, with a light transmittance of not less than 80% and a sheet resistance of not more than 50Ω / sq.

22. The apparatus according to claim 12, characterized in that, The optical control element includes a plasmonic nanoantenna array, which is composed of periodically arranged dimers or polymers of metal nanoparticles. The metal nanoparticles are selected from at least one of gold, silver, aluminum, and copper. The spacing between the nanoparticles is 1 nm to 50 nm, and the array period is 10 nm to 500 nm. The plasmonic nanoantenna array generates near-field enhancement under local surface plasmonic resonance conditions, with an enhancement factor of 10 to 100,000 times and a resonance wavelength covering the range of 350 nm to 900 nm.

23. The apparatus according to claim 12, characterized in that, The optical control element includes an upconversion nanoparticle wavelength conversion layer, which is composed of fluoride matrix nanoparticles co-doped with sensitizer and activator. The matrix is ​​selected from at least one of NaYF4, NaGdF4, LiYF4, and BaYF5. The sensitizer is selected from at least one of Yb³⁺ and Nd³⁺. The activator is selected from at least one of Er³⁺, Tm³⁺, and Ho³⁺. The upconversion nanoparticle wavelength conversion layer converts near-infrared excitation light into visible light or ultraviolet emission light through an anti-Stokes process.

24. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the control method of the apparatus as described in any one of claims 1 to 23.

Images