Quartz wafer defect detection equipment with multi-spectral imaging detection function

Abstract

The application discloses a quartz wafer defect detection equipment with a multispectral imaging detection function, relates to the technical field of quartz wafer defect detection, and comprises a rack serving as a mounting and positioning carrier of various components of the equipment; a wafer conveying mechanism is arranged on the rack and comprises a vacuum adsorption platform for fixing a quartz wafer and a linear motor driving guide rail for driving the vacuum adsorption platform to move linearly. In the application, the photonic crystal superstructure surface light splitting module in the multispectral imaging module eliminates the multi-waveband optical path difference and coaxial deviation, the multispectral coplanar integrated detector array solves the response and transmission delay problems through a shared trigger circuit and a parallel reading channel, the wide-spectrum light source is used to ensure high-quality acquisition of multi-waveband signals, and the time and space synchronization errors are eliminated from the optical and hardware levels; the closed-loop control system relies on the high-frequency acquisition of motion parameters of the laser Doppler interferometer to realize time sequence accurate matching under dynamic scanning and takes into account the detection accuracy and efficiency.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of quartz wafer defect detection, in particular to a quartz wafer defect detection device with multi-spectral imaging detection function. BACKGROUND

[0002] As a key basic material for semiconductor lithography, etching and other core processes, the surface scratches, particle contamination and subsurface bubbles, lattice defects of quartz wafers directly affect the yield of chips. Multi-spectral imaging detection technology has become the core technology path for quartz wafer defect detection because it can capture the unique optical response of different defects.

[0003] For example, a wafer detection scattering light collection objective lens and wafer detection equipment with publication number CN119620341A is applied to wafer defect detection. The scattering light collection lens includes a first lens group, a second lens group, a third lens group, and a fourth lens group along the optical axis direction from the object plane to the image plane, a total of twelve lenses and an aperture stop, and the twelve lenses and the aperture stop are coaxially arranged. The first lens group is a catadioptric lens group with positive focal power, which is used to image the object plane to the intermediate image plane by twice reflection.

[0004] However, in the prior art, the quartz wafer defect detection device with multi-spectral imaging detection function has inherent differences in the response speed of the detector, the circuit trigger delay of different spectral channels at the hardware level, the switching delay of the mechanical light splitting structure such as the filter wheel at the optical level, and the different waveband light path propagation times caused by the prism and other light splitting elements at the control level. The transmission deviation of the trigger signal and the mismatch of the motion speed further amplify the error during dynamic scanning. In order to avoid this error, the existing equipment mostly adopts an intermittent process of "wafer static collection - moving positioning", which prolongs the detection time of a single wafer to tens of seconds, making it difficult to adapt to the production capacity demand of the production line. SUMMARY

[0005] The purpose of the present application is to provide a quartz wafer defect detection device with multi-spectral imaging detection function to solve the problems raised in the background art.

[0006] To achieve the above purpose, the present application provides the following technical scheme: a quartz wafer defect detection device with multi-spectral imaging detection function, comprising a rack as the mounting and positioning carrier of various components of the device; a wafer conveying mechanism provided on the rack, comprising a vacuum adsorption platform for fixing the quartz wafer and a linear motor driven guide rail for driving the vacuum adsorption platform to move linearly, for realizing the feeding, accurate positioning, continuous conveying and discharging of the quartz wafer;

[0007] A multi-spectral imaging module is arranged on the rack and located directly above the detection path of the wafer conveying mechanism, used for synchronously collecting ultraviolet, visible light, near-infrared multi-band image data of the quartz wafer, and includes a light source assembly, a photonic crystal superstructure surface light splitting module and a multi-spectral coplanar integrated detector array connected in sequence in the light path; the light source assembly is a wide-spectrum area light source, covering the ultraviolet to near-infrared wave band;

[0008] A closed-loop control system is electrically connected with the wafer conveying mechanism and the multi-spectral imaging module, used for realizing the space-time synchronization of multi-spectral data by collecting motion parameters and dynamically adjusting the acquisition time sequence in real time, and includes a laser Doppler interferometer, an FPGA main controller and a driving module;

[0009] A data processing module is electrically connected with the multi-spectral imaging module and the closed-loop control system, used for real-time registration and defect analysis of multi-spectral image data, and outputs the defect position, type and size parameters, and includes an ASIC on-chip computing unit and a back-end analysis unit.

[0010] Preferably, the photonic crystal superstructure surface light splitting module includes a substrate of quartz glass and a silicon nitride nanocolumn array arranged on the substrate; the nanocolumn array is composed of a plurality of nanocolumns arranged in a predetermined periodical pattern, and the height and spacing of the nanocolumns are designed according to the optical path difference of the target spectral wave band, so that the optical path difference of the incident light of the ultraviolet, visible light and near-infrared wave band is compensated to 0±5ps after the modulation of the nanocolumn array, and the propagation directions of the light of each wave band are coaxial, with a coaxiality deviation of less than 0.1°.

[0011] Preferably, the target spectral wave band includes a 260nm ultraviolet wave band, a 550nm visible light wave band and a 940nm near-infrared wave band; the period of the nanocolumn array is 200-400nm, and the height-diameter ratio of the nanocolumns is 2-5:1.

[0012] Preferably, the multi-spectral coplanar integrated detector array is prepared by a CMOS substrate epitaxial heterojunction process, and AlGaN-based ultraviolet light sensing units, Si-based visible light sensing units and InGaAs-based near-infrared light sensing units are arranged on the same chip substrate in a pixel-level staggered manner; the detector array also integrates a shared trigger circuit and a parallel charge reading channel; the shared trigger circuit receives a synchronization trigger signal from the FPGA main controller to drive the synchronization exposure of each light sensing unit, with an exposure delay deviation of less than 10ps; the parallel charge reading channel synchronously outputs image data of each wave band, with a data transmission delay difference of less than 50ns.

[0013] Preferably, the laser Doppler interferometer is arranged on the frame and parallel to the movement track of the linear motor drive rail, with a sampling frequency of ≥1 MHz, for real-time acquisition of the moving speed, acceleration and instantaneous position parameters of the vacuum adsorption platform, with a position detection accuracy of ±0.01 μm.

[0014] Preferably, the FPGA main controller is internally provided with a motion-trigger timing mapping model, which dynamically calculates and adjusts the exposure trigger time of the multi-spectral coplanar integrated detector array according to the real-time motion parameters acquired by the laser Doppler interferometer, with an adjustment accuracy of 10 ps and a trigger parameter update frequency of 1 ms.

[0015] Preferably, the driving module is electrically connected with the linear motor drive rail, and can control the motion speed of the vacuum adsorption platform to adaptively change in the range of 50-300 mm / s, with a speed fluctuation error of <0.5%, and can realize millisecond-level speed correction in response to the instructions of the FPGA main controller.

[0016] Preferably, the ASIC on-chip computing unit is directly connected with the output end of the multi-spectral coplanar integrated detector array, and is internally provided with a preset multi-spectral pixel coordinate mapping table and a dynamic feature anchor point matching algorithm; the dynamic feature anchor point matching algorithm takes the lattice periodic texture on the surface of the quartz wafer as an anchor point, and real-time fine-tunes the pixel positions of different spectral images, with a registration error of <0.05 pixels and a single-frame image registration time consumption of <100 μs.

[0017] Preferably, the light source assembly has a light intensity uniformity of ≥90%, an ultraviolet waveband radiation intensity of ≥10 mW / cm² and a near-infrared waveband radiation intensity of ≥20 mW / cm²; the adsorption force of the vacuum adsorption platform can be adjusted in the range of 0.1-0.5 MPa, and is suitable for quartz wafers with a diameter of 150-300 mm.

[0018] Preferably, the rear-end analysis unit adopts a deep learning model based on an attention mechanism for defect identification and classification of the registered multi-spectral images.

[0019] Compared with the prior art, the present application has the following beneficial effects:

[0020] 1. In this invention, the wafer transport mechanism utilizes a porous ceramic adsorption platform and a high-precision linear motor guide rail to achieve stable wafer fixation and high-speed, stable movement, meeting the continuous transport requirements of mass production lines. In the multispectral imaging module, the photonic crystal metasurface beam-splitting module eliminates multi-band optical path difference and coaxial deviation. The multispectral coplanar integrated detector array solves response and transmission delay problems through a shared trigger circuit and parallel reading channels. Combined with a broadband light source, it ensures high-quality acquisition of multi-band signals, eliminating spatiotemporal synchronization errors from both optical and hardware perspectives. The closed-loop control system relies on a laser Doppler interferometer to collect motion parameters at high frequency, forming a "sensing-computation-drive" closed loop to achieve precise timing matching under dynamic scanning, balancing detection accuracy and efficiency. It boasts high defect location accuracy and a high detection rate for common defects, perfectly meeting the "high efficiency + high precision" quality control requirements of mass production lines while reducing maintenance and calibration frequency and costs.

[0021] 2. In this invention, the stable fixation and ultra-low fluctuation motion of the wafer prevent wafer attitude deviation from affecting defect location. In the multispectral imaging module, a broadband light source combined with an optical fiber coupling system ensures stable light intensity. The four-band photonic crystal metasurface eliminates beam splitting delay and coaxial deviation. The high-resolution detector array accurately captures near-infrared weak scattering signals with low trigger delay and high quantum efficiency, achieving synchronous and high-quality acquisition of subsurface defect signals from both optical and hardware perspectives. The closed-loop control system relies on a 2MHz high-frequency laser interferometer and an FPGA model optimized by Kalman filtering, combined with an EtherCAT real-time communication drive module, to form a "motion prediction-timing control-speed correction" closed loop, ensuring the accuracy of defect location under dynamic scanning. The data processing module, through a deep learning-registered ASIC unit and a Spectral-Transformer model, integrates multi-band features to accurately identify defects, perfectly adapting to the stringent control requirements of high-end wafer fabs for "high-precision identification, deep detection, and full-process traceability" of latent defects. Attached Figure Description

[0022] Figure 1 This is a flowchart of the quartz wafer defect detection device with multispectral imaging detection function of the present invention;

[0023] Figure 2 This is a first three-dimensional structural schematic diagram of the quartz wafer defect detection device with multispectral imaging detection function according to the present invention.

[0024] Figure 3 This is a second three-dimensional structural schematic diagram of the quartz wafer defect detection device with multispectral imaging detection function of the present invention.

[0025] Figure 4 This is a top view schematic diagram of the quartz wafer defect detection device with multispectral imaging detection function according to the present invention.

[0026] Figure 5 This is a three-dimensional structural diagram of the wafer transport mechanism in the quartz wafer defect detection equipment with multispectral imaging detection function of the present invention;

[0027] Figure 6 This is a three-dimensional structural diagram of the multispectral imaging module in the quartz wafer defect detection equipment with multispectral imaging detection function of the present invention.

[0028] In the diagram: 1. Rack; 2. Wafer transport mechanism; 21. Vacuum adsorption platform; 22. Linear motor drive rail; 3. Multispectral imaging module; 4. Closed-loop control system. Detailed Implementation

[0029] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] Example 1: Refer to Figures 1-6 As shown: Quartz wafer defect inspection equipment with multispectral imaging detection function is designed to meet the routine defect inspection needs of quartz wafers in semiconductor mass production lines. It needs to balance inspection efficiency and basic accuracy. The equipment needs to meet the continuous inspection rhythm of "second-level / wafer" in mass production lines, while covering common defects such as surface scratches and particle contamination.

[0031] The frame 1 adopts an integral cast aluminum alloy structure, and is equipped with vibration damping feet at the bottom. The table levelness error is <0.02mm / m, providing a rigid mounting benchmark for each component.

[0032] The wafer transport mechanism 2 includes a vacuum adsorption platform 21 and a linear motor drive rail 22. The vacuum adsorption platform adopts a porous ceramic panel, and the adsorption force can be adjusted within the range of 0.2-0.5MPa. It is suitable for quartz wafers with a diameter of 300mm, and the wafer flatness deviation after adsorption is <0.01mm. The linear motor drive rail is selected from the Yaskawa SGLF series, with a stroke of 600mm, a positioning accuracy of ±0.1μm, and a maximum movement speed of 300mm / s.

[0033] The light source component in the multispectral imaging module 3 has a spectral range of 200-1100nm, a light intensity uniformity of 92%, an ultraviolet band (260nm) radiation intensity of 12mW / cm², a near-infrared band (940nm) radiation intensity of 25mW / cm², and the distance between the light source outlet and the wafer surface is 50mm.

[0034] Photonic crystal metasurface beam splitter module: The substrate is quartz glass (2mm thick), and a silicon nitride nanopillar array is grown on the surface with a period of 300nm. The nanopillars have a diameter of 100nm and a height of 300nm (height-to-diameter ratio 3:1). It is designed for the 260nm ultraviolet, 550nm visible light, and 940nm near-infrared bands. After modulation, the optical path difference of the three bands is compensated to 0±4ps, and the coaxiality deviation of the emitted light is 0.08°.

[0035] Multispectral coplanar integrated detector array: fabricated using CMOS substrate epitaxial heterojunction process, with a pixel size of 5μm and a resolution of 2048×2048; AlGaN-based ultraviolet photosensitive unit with a quantum efficiency of 65%, Si-based visible light photosensitive unit with a quantum efficiency of 75%, and InGaAs-based near-infrared photosensitive unit with a response range of 900-1700nm; shared trigger circuit with a delay deviation of 8ps, and parallel charge readout channel with a data transmission delay difference of 40ns.

[0036] The laser Doppler interferometer in the closed-loop control system 4 has a sampling frequency of 1MHz and a position detection accuracy of ±0.01μm. It is mounted on the side of the frame and parallel to the motion trajectory of the linear motor drive rail. The laser beam is incident on the reflective target on the side of the vacuum adsorption platform. The FPGA main controller has a built-in motion-trigger timing mapping model and algorithm logic written in C language. It can update the trigger time every 1ms according to the motion parameters and adjust the accuracy by 10ps.

[0037] The drive module is electrically connected to the linear motor drive rail, with a speed control range of 50-300mm / s, a speed fluctuation error of 0.4%, and a speed correction delay of <2ms in response to FPGA commands.

[0038] The ASIC on-chip computing unit in the data processing module adopts an integrated multispectral pixel coordinate mapping table. The dynamic feature anchor point matching algorithm uses the 1μm periodic texture of the quartz lattice as the anchor point. The registration time for a single frame image is 80μs, and the registration error is 0.04 pixels.

[0039] The backend analysis unit combines a deep learning model with an attention mechanism. The training dataset contains 100,000 labeled defect images and can identify six common defects such as surface scratches and particulate contamination.

[0040] The quartz wafer is fed and pre-positioned by a robotic arm to the vacuum adsorption platform 21. The platform starts to fix the wafer with an adsorption force of 0.3MPa. The laser positioning system (Keyence LK-H020) completes the alignment of the wafer center with the detection reference, with an alignment error of <0.02mm.

[0041] Dynamic scanning and synchronous acquisition: The drive module controls the linear motor to drive the guide rail to move at a uniform speed of 200mm / s. The laser Doppler interferometer collects the platform's moving speed, acceleration, and instantaneous position data in real time and transmits them to the FPGA main controller. The FPGA generates a synchronous trigger signal based on the motion parameters to drive the multispectral coplanar integrated detector array to expose. The light source component illuminates the wafer surface, and the reflected light is split by the photonic crystal metasurface and synchronously incident on the three types of photosensitive units. The raw data is output through parallel channels.

[0042] Real-time registration and defect analysis: The on-chip computing unit of the ASIC receives the raw data, completes the initial registration through the coordinate mapping table, and then fine-tunes the deviation through the dynamic feature anchoring algorithm; the registered data is transmitted to the back-end analysis unit, and the model completes defect identification and classification within 50ms, and calculates parameters such as defect location (accuracy ±0.05μm) and size.

[0043] Unloading and sorting: The guide rail drives the platform to the unloading position. Based on the test results, the robot moves the qualified wafers to the finished product box and the defective wafers to the rework area. The vacuum adsorption force is released, and the single test is completed.

[0044] In this embodiment, the rigid structure and vibration reduction design of the frame 1 provide a stable installation benchmark for the equipment, ensuring the operational accuracy of each component; the wafer transport mechanism 2 achieves stable fixing and high-speed, stable movement of the wafer through a porous ceramic adsorption platform and a high-precision linear motor guide rail, adapting to the continuous transport requirements of mass production lines; in the multispectral imaging module 3, the photonic crystal metasurface beam splitting module eliminates multi-band optical path difference and coaxial deviation, and the multispectral coplanar integrated detector array solves response and transmission delay problems through a shared trigger circuit and parallel reading channels, and works with a broadband light source to ensure high-quality acquisition of multi-band signals, eliminating spatiotemporal synchronization errors from both optical and hardware perspectives; the laser Doppler interferometer in the closed-loop control system 4 acquires motion parameters at high frequency. The FPGA main controller updates the trigger time at the 1ms level (10ps accuracy), and the drive module precisely controls the speed (fluctuation error of 0.4%), forming a closed loop of "sensing-computing-driving" to achieve precise timing matching under dynamic scanning. The ASIC on-chip computing unit of the data processing module completes high-precision registration at a speed of 80μs / frame (error of 0.04 pixels), and the back-end model quickly identifies 6 types of defects (within 50ms), ensuring detection accuracy and efficiency. The entire process achieves "second-level / wafer" continuous inspection of 300mm wafers, with defect positioning accuracy of ±0.05μm and a high detection rate of common defects, perfectly adapting to the "high efficiency + high precision" quality control requirements of mass production lines, while reducing the frequency and cost of operation and maintenance calibration.

[0045] Example 2: According to Figures 1-6As shown, this implementation focuses on detecting latent defects such as subsurface bubbles and micro-lattice defects, requiring defect detection accuracy to reach the 0.05μm level and subsurface defect detection depth ≥50μm, to meet the stringent quality control standards of high-end wafer fabs.

[0046] The frame 1 adopts a granite base + aluminum alloy frame structure, with a table levelness error of <0.01mm / m and a built-in constant temperature control system (temperature control accuracy ±0.5℃) to reduce the impact of temperature on the optical system.

[0047] Wafer transport mechanism 2: The vacuum adsorption platform adopts a porous sapphire panel, with an adsorption force adjustment range of 0.3-0.5MPa, suitable for 300mm wafers, and the flatness deviation after adsorption is <0.005mm; the linear motor drive guide rail is selected from THKSSR series, with a positioning accuracy of ±0.05μm, a movement speed of 50-250mm / s, and is equipped with an air bearing vibration reduction structure.

[0048] The light source component of the multispectral imaging module 3 adopts a broadband light source with a spectral range of 200-1700nm, a light intensity uniformity of 95%, an ultraviolet radiation intensity of 15mW / cm² at 260nm, a near-infrared radiation intensity of 30mW / cm² at 940nm, and is equipped with an optical fiber coupled light output system with a light intensity stability of ±1% / h.

[0049] Photonic crystal metasurface beam splitter module: quartz glass substrate thickness 3mm, silicon nitride nanopillar array period 250nm, nanopillar diameter 80nm, height 240nm (height-to-diameter ratio 3:1); designed for four wavelength bands of 260nm, 550nm, 940nm, and 1310nm, optical path difference compensation to 0±3ps, and output light coaxiality deviation 0.05°.

[0050] Multispectral coplanar integrated detector array: pixel size 3.5μm, resolution 4096×4096; AlGaN-based ultraviolet photosensitive unit quantum efficiency 70%, InGaAs-based near-infrared photosensitive unit response range 900-1700nm, quantum efficiency 72%; shared trigger circuit delay deviation 5ps, parallel readout channel transmission delay difference 30ns.

[0051] The laser Doppler interferometer in the closed-loop control system 4 has a sampling frequency of 2MHz, a position detection accuracy of ±0.005μm, and a laser beam incident on a high-precision reflector of the vacuum adsorption platform with a signal-to-noise ratio of ≥50dB.

[0052] The FPGA main controller has a built-in optimized motion-trigger timing mapping model, and introduces a Kalman filter algorithm to predict motion trajectories, with an adjustment accuracy of 5ps and a parameter update frequency of 2ms.

[0053] The drive module uses a servo driver with a speed fluctuation error of 0.2% and a response latency of <1ms. It supports real-time communication with the FPGA via EtherCAT.

[0054] The ASIC on-chip computing unit registration algorithm in data processing module 5 introduces a lightweight deep learning model, with a single frame registration time of 60μs and a registration error of 0.03 pixels; the dynamic feature anchor point matching algorithm supports dual anchor point calibration of lattice texture and manual marking.

[0055] Backend analysis unit: The Spectral-Transformer deep learning model is used to integrate multi-band spectral-spatial features. The training dataset contains 200,000 labeled images, covering 8 types of defects such as surface scratches and subsurface bubbles, including 50,000 labeled samples of subsurface defects.

[0056] Material loading and pre-positioning: Vacuum suction cup robotic arms transport wafers, vacuum adsorption platforms fix them with an adsorption force of 0.4MPa, and laser positioning systems complete the alignment with an alignment error of <0.01mm; the constant temperature system is activated to stabilize the internal temperature of the equipment at 23±0.5℃.

[0057] Dynamic scanning and synchronous acquisition: The linear motor drives the guide rail to move at a constant speed of 150 mm / s. The laser Doppler interferometer acquires motion parameters every 0.5 ms and transmits them to the FPGA main controller. The FPGA combines Kalman filtering to predict the trajectory and generate a synchronous trigger signal. The exposure time of the detector array is dynamically adjusted according to the band characteristics. After the reflected light is split by the metasurface, the four bands of light are synchronously incident on the detector and the data is output in parallel.

[0058] Real-time registration and defect analysis: The ASIC unit first registers the data using a coordinate mapping table, and then fine-tunes it using lattice texture as an anchor point. Near-infrared feature extraction is enhanced for subsurface defects. The back-end analysis unit completes defect identification within 80ms and calculates the subsurface defect depth using a multi-band light intensity attenuation model. The depth detection range is 1-50μm.

[0059] Material unloading and sorting: After inspection, the wafers are transferred to the sorting machine by a robotic arm. They are sorted into the rework area or scrap area according to the defect type (such as repairable scratches / non-repairable bubbles). The inspection data is automatically uploaded to the MES system.

[0060] In this embodiment, the frame 1 adopts a granite base and a constant temperature control system, with a platform levelness of <0.01mm / m and a temperature control accuracy of ±0.5℃, reducing the interference of temperature and vibration on the optical system from the environmental source, and providing a stable benchmark for high-precision detection; the wafer transport mechanism 2 uses a porous sapphire adsorption platform (flatness deviation <0.005mm) and a high-precision guide rail with air bearings (positioning accuracy ±0.05μm) to achieve stable fixation and ultra-low fluctuation movement of 300mm wafers, avoiding wafer posture deviation from affecting defect positioning; in the multispectral imaging module 3, a broadband light source (200-1700nm) combined with an optical fiber coupling system ensures stable light intensity, a four-band photonic crystal metasurface (optical path difference compensation 0±3ps) eliminates beam splitting delay and coaxial deviation, and a high-resolution detector array (3.5μm pixels) accurately captures near-infrared weak scattering signals with low trigger delay (5ps) and high quantum efficiency. The system achieves synchronous, high-quality acquisition of subsurface defect signals from both optical and hardware perspectives. The closed-loop control system 4 relies on a 2MHz high-frequency laser interferometer (accuracy ±0.005μm) and an FPGA model optimized by Kalman filtering (trigger accuracy 5ps), combined with an EtherCAT real-time communication drive module (response latency <1ms), forming a closed loop of "motion prediction - timing control - speed correction" to ensure defect positioning accuracy of 0.05μm under dynamic scanning. The data processing module uses ASIC units registered with deep learning (error 0.03 pixels) and a Spectral-Transformer model to accurately identify 8 types of defects by integrating multi-band features. Based on a model trained with 50,000 subsurface samples, it achieves depth detection from 1 to 50μm, and the data is automatically uploaded to the MES system, perfectly meeting the stringent control requirements of high-end wafer fabs for "high-precision identification, depth detection, and full-process traceability" of latent defects.

[0061] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A quartz wafer defect detection device with multispectral imaging detection function, characterized in that: include: The frame (1) serves as the mounting and positioning carrier for the various components of the equipment; The wafer conveying mechanism (2) is located on the frame (1) and includes a vacuum adsorption platform (21) for fixing the quartz wafer and a linear motor drive rail (22) for driving the vacuum adsorption platform (21) to move linearly, for realizing the loading, precise positioning, continuous conveying and unloading of the quartz wafer; The multispectral imaging module (3) is mounted on the rack (1) and located directly above the detection path of the wafer transport mechanism (2). It is used to simultaneously acquire ultraviolet, visible, and near-infrared multi-band image data of the quartz wafer. It includes a light source component, a photonic crystal metasurface beam splitting module, and a multispectral coplanar integrated detector array connected in sequence by optical paths. The light source component is a broadband surface light source with a spectrum covering the ultraviolet to near-infrared bands. The closed-loop control system (4) is electrically connected to the wafer conveying mechanism (2) and the multispectral imaging module (3) respectively. It is used to realize the spatiotemporal synchronization of multispectral data by real-time acquisition of motion parameters and dynamic adjustment of acquisition timing. It includes a laser Doppler interferometer, an FPGA main controller and a drive module. The data processing module is electrically connected to the multispectral imaging module (3) and the closed-loop control system (4) for real-time registration and defect analysis of multispectral image data, and outputs defect location, type and size parameters. It includes an ASIC on-chip computing unit and a back-end analysis unit. The photonic crystal metasurface beam-splitting module includes a quartz glass substrate and a silicon nitride nanopillar array disposed on the substrate. The nanopillar array is composed of several nanopillars arranged according to a preset periodic rule. The height and spacing of the nanopillars are customized according to the optical path difference of the target spectral band, so that the optical path difference of the incident light in the ultraviolet, visible and near-infrared bands is compensated to 0±5ps after being modulated by the nanopillar array, and the propagation direction of the outgoing light in each band is coaxial with a coaxiality deviation of <0.1°. The multispectral coplanar integrated detector array is fabricated using a CMOS substrate epitaxial heterojunction process. AlGaN-based ultraviolet photosensitive units, Si-based visible light photosensitive units, and InGaAs-based near-infrared photosensitive units are arranged pixel-level alternately on the same chip substrate. The detector array also integrates a shared trigger circuit and a parallel charge readout channel. The shared trigger circuit receives the synchronous trigger signal from the FPGA main controller and drives each photosensitive unit to be exposed synchronously with an exposure delay deviation of <10ps. The parallel charge readout channel synchronously outputs image data for each band with a data transmission delay difference of <50ns. The laser Doppler interferometer is mounted on the frame (1) and parallel to the motion trajectory of the linear motor drive rail (22). The sampling frequency is ≥1MHz. It is used to collect the moving speed, acceleration and instantaneous position parameters of the vacuum adsorption platform (21) in real time. The position detection accuracy is ±0.01μm. The FPGA main controller has a built-in motion-trigger timing mapping model. The model dynamically calculates and adjusts the exposure triggering time of the multispectral coplanar integrated detector array based on the real-time motion parameters collected by the laser Doppler interferometer. The adjustment accuracy is up to 10 ps, ​​and the triggering parameters are updated every 1 ms.

2. The quartz wafer defect detection equipment with multispectral imaging detection function according to claim 1, characterized in that: The target spectral bands include the 260nm ultraviolet band, the 550nm visible light band, and the 940nm near-infrared band; the period of the nanopillar array is 200-400nm, and the height-to-diameter ratio of the nanopillars is 2-5:

1.

3. The quartz wafer defect detection equipment with multispectral imaging detection function according to claim 1, characterized in that: The drive module is electrically connected to the linear motor drive rail (22), and can adjust the movement speed of the vacuum adsorption platform (21) to adapt within the range of 50-300mm / s with a speed fluctuation error of <0.5%, and respond to the instructions of the FPGA main controller to achieve millisecond-level speed correction.

4. The quartz wafer defect detection equipment with multispectral imaging detection function according to claim 1, characterized in that: The on-chip computing unit of the ASIC is directly connected to the output of the multispectral coplanar integrated detector array, and has a built-in preset multispectral pixel coordinate mapping table and dynamic feature anchor point matching algorithm. The dynamic feature anchor point matching algorithm uses the lattice periodic texture of the quartz wafer surface as anchor points to finely adjust the pixel position of different spectral images in real time, with a registration error of <0.05 pixels and a single frame image registration time of <100μs.

5. The quartz wafer defect detection equipment with multispectral imaging detection function according to claim 1, characterized in that: The light intensity uniformity of the light source component is ≥90%, the ultraviolet band radiation intensity is ≥10mW / cm², and the near-infrared band radiation intensity is ≥20mW / cm²; the adsorption force of the vacuum adsorption platform (21) can be adjusted within the range of 0.1-0.5MPa, and it is suitable for quartz wafers with a diameter of 150-300mm.

6. The quartz wafer defect detection equipment with multispectral imaging detection function according to claim 1, characterized in that: The back-end analysis unit uses a deep learning model based on an attention mechanism to identify and classify defects in the registered multispectral images.

Images