Optical detection system for identifying gallium nitride lattice defects

Abstract

The present application relates to the field of semiconductor wafer optical detection, specifically to an optical detection system for identifying gallium nitride lattice defects; comprising an illumination module, a detection module, a scanning control module, a scanning execution mechanism, a data processing module and a display; the system outputs polarized state periodically switched ultraviolet light to irradiate the wafer, obtains the harmonic amplitude by harmonic demodulation of the received light to determine the polarization rotation amount, and compares with the background threshold to determine abnormal pixels; the core is to accurately identify lattice defects and dislocation types by combining polarization modulation, harmonic demodulation and hierarchical scanning; the present application effectively overcomes the problem of weak signal identification, balances the detection accuracy and full wafer scanning efficiency, and improves the stability of the dense overlapping area determination.

Description

Technical Field

[0001] This invention relates to the field of optical inspection of semiconductor wafers, specifically to an optical inspection system for identifying gallium nitride lattice defects. Background Technology

[0002] As the application of gallium nitride (GaN) wafers in power devices and optoelectronic devices continues to expand, the impact of lattice defects, especially dislocations, on device performance, yield, and consistency is becoming increasingly prominent. How to quickly identify defects during wafer inspection while balancing inspection accuracy and large-area scanning efficiency has become an important problem that needs to be solved in the current wafer inspection field. Traditional GaN lattice defect detection currently mainly relies on the following methods: surface imaging detection based on differences in luminous intensity, microscopic observation detection based on morphological changes, and detection by high-resolution scanning of local areas point by point. However, methods based on differences in luminescence intensity, morphological changes, and point-by-point high-resolution scanning all have certain shortcomings. For example, detection methods based on differences in luminescence intensity are easily affected by background fluctuations, light intensity drift, and the indistinctness of weak defect signals, resulting in insufficient ability to identify weak disturbances caused by single or a small number of dislocations. Detection methods based on morphological changes are difficult to directly reflect the differences in polarization response caused by local lattice stress fields, and have limited ability to distinguish defect types. Although point-by-point high-resolution scanning can improve local detection accuracy, its scanning efficiency is low when detecting the entire wafer, and it is prone to problems such as excessive data processing volume in areas with dense or overlapping defects, leading to insufficient judgment stability. Summary of the Invention

[0003] The purpose of this invention is to provide an optical detection system for identifying gallium nitride (GaN) lattice defects. Existing GaN lattice defect detection technologies have significant shortcomings in identifying weak disturbances caused by single or small numbers of dislocations, distinguishing defect types, and achieving high efficiency in large-area scanning and stable determination of densely overlapping regions. There is an urgent need to propose an optical detection system that can accurately identify lattice defects and dislocation types by combining polarization modulation, harmonic demodulation, and hierarchical scanning, while balancing local detection accuracy and overall scanning efficiency. The technical solution of this invention includes: An optical inspection system for identifying gallium nitride lattice defects includes an illumination module, an inspection module, a scan control module, a scan execution mechanism, a data processing module, and a display. The illumination module is used to output ultraviolet detection light with periodically switched polarization states and irradiate the gallium nitride wafer. The detection module is used to receive the reflected or scattered light from the gallium nitride wafer at the current scanning position, and to perform harmonic demodulation on the detection signal to obtain the first harmonic amplitude and the second harmonic amplitude. The data processing module is used to determine the polarization rotation amount of the pixel corresponding to the current scanning position based on the first harmonic amplitude and the second harmonic amplitude, and compare the polarization rotation amount with the background threshold to determine whether the current pixel is an abnormal pixel or a background pixel. The background threshold is determined based on the mean and standard deviation of the polarization rotation amount of background pixels in the defect-free area obtained in advance. The scanning control module is used to control the scanning execution mechanism to perform full-film coarse scanning, partial fine scanning, or regional integral scanning. The data processing module is used to record defect identification results and scanning trajectories, and outputs the detection results through the display.

[0004] Preferably, the illumination module includes a pulsed laser, a linear polarizer, and a photoelastic modulator; the pulsed laser is used to output the ultraviolet detection light; the linear polarizer is used to convert the ultraviolet detection light into linearly polarized light; and the photoelastic modulator is used to periodically polarize the linearly polarized light. The detection module includes a polarizer, a photodetector, a reference detector, a dual-channel lock-in amplifier, a dark field detection channel, and a position detection unit. The reference detector is used to monitor the laser power. The dual-channel lock-in amplifier is used to perform phase-locked demodulation on the electrical signal converted and output by the photodetector to obtain the first harmonic amplitude and the second harmonic amplitude. The dark field detection channel is used to output a pixel shielding signal characterizing whether the pixel corresponding to the current scanning position is effective. The position detection unit is used to receive the reflected light and output the spot centroid offset. The scanning execution mechanism includes a dual-axis scanning galvanometer and a piezoelectric platform; the dual-axis scanning galvanometer is used to perform a full-wafer coarse scan; the piezoelectric platform is used to perform a local fine scan and to adjust the tilt angle of the gallium nitride wafer.

[0005] Preferably, the data processing module is further configured to normalize the detection signal corresponding to the current pixel based on the laser power monitored by the reference detector, and obtain the first harmonic amplitude and the second harmonic amplitude output by the dual-channel lock-in amplifier; When the pixel masking signal indicates that the current pixel is a valid pixel, the polarization rotation amount of the current pixel is determined based on the first harmonic amplitude and the second harmonic amplitude, and the polarization rotation amount is compared with the background threshold. When the polarization rotation amount is greater than the background threshold, the current pixel is determined to be an abnormal pixel; When the polarization rotation amount is less than or equal to the background threshold, the current pixel is determined to be a background pixel.

[0006] Preferably, the data processing module is further configured to identify the dislocation type based on the ratio of the first harmonic amplitude to the second harmonic amplitude, wherein the dislocation type includes edge dislocations, screw dislocations, and mixed dislocations; The data processing module is used to determine whether the dislocation type of the current pixel is an edge dislocation, a screw dislocation, or a mixed dislocation by comparing the ratio with preset upper and lower threshold values.

[0007] Preferably, the scanning control module is further configured to determine candidate regions based on the distribution of connected abnormal pixels in the two-dimensional pixel array generated by the full-slice coarse scan that satisfy a pre-set pixel adjacency rule; Calculate the number of valid abnormal pixels contained in the candidate region, and remove invalid pixels based on the pixel masking signal; The number of valid abnormal pixels is compared with a pixel number threshold, which is set according to a preset coarse scan resolution and a calibration defect size. When the number of valid abnormal pixels is less than or equal to the pixel number threshold, the piezoelectric platform is controlled to perform a local fine scan on the candidate region. When the number of valid abnormal pixels is greater than the number of pixels threshold, the candidate region is determined to be a defect overlap region, and the scanning execution mechanism is controlled to perform regional integral scanning on the defect overlap region.

[0008] Preferably, the data processing module is further configured to update the background threshold; When the current pixel is determined to be a background pixel, the mean and standard deviation of the polarization rotation of the background pixels in the defect-free region are updated using the polarization rotation of the background pixel. When the current pixel is determined to be an abnormal pixel, or when the pixel masking signal indicates that the current pixel is an invalid pixel, the current pixel will not participate in the update of the mean and standard deviation of the polarization rotation.

[0009] Preferably, the data processing module is further configured to adjust the driving voltage amplitude of the photoelastic modulator according to the deviation between the output signal of the reference detector and the pre-calibrated reference value, and adjust the phase delay of the photoelastic modulator to a pre-calibrated quarter-wave phase delay. Based on the centroid offset of the light spot output after the reflected light is received by the position detection unit in the detection module, the piezoelectric platform is controlled to adjust the tilt angle of the gallium nitride wafer.

[0010] Preferably, the scanning control module is further configured to, during local fine scanning, control the piezoelectric platform to scan the candidate region and its surrounding neighborhood within a pre-defined range, wherein the neighborhood range is pre-defined according to the local fine scanning resolution; the scanning control module is further configured to, during regional integral scanning, use a moving sliding window to perform integral scanning of the defect overlap region and integrate the detection signal within the window, wherein the window size and step size of the sliding window are determined according to a pre-defined coarse scanning resolution; the data processing module is further configured to, for pixels marked as invalid by the pixel masking signal, use a spatial interpolation algorithm to complete the pixel based on the detection value of the valid pixels in its neighborhood.

[0011] Preferably, the data processing module is further configured to record detection process data, which includes the first harmonic amplitude, the second harmonic amplitude, the pixel shielding signal, and the spot centroid offset. Record the threshold update operation performed by the data processing module, the drive voltage adjustment operation of the photoelastic modulator, and the tilt angle adjustment operation of the piezoelectric platform; It also records the coordinates of the defect, the type of dislocation, the scan path, and the corresponding time series.

[0012] Preferably, the data processing module is used to display the detection results through the display, and the detection results include defect distribution, dislocation type, scan path and detection process record.

[0013] Compared with the prior art, the present invention has the following beneficial effects: In the above scheme, this invention combines polarization modulation, harmonic demodulation, and hierarchical scanning mechanisms. Through a data processing module, the influence of dislocation strain field on the polarization state is transformed into a quantifiable polarization rotation quantity, and the background threshold is dynamically updated. This scheme effectively overcomes the challenge of weak signal identification. While ensuring the accuracy of single-defect identification, it improves the efficiency of full-film scanning through regional integral scanning, significantly enhancing the stability of defect determination in densely overlapping areas. Attached Figure Description

[0014] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. The drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0015] Figure 1 This is a schematic diagram of the modules of the optical detection system for identifying gallium nitride lattice defects provided in an embodiment of the present invention. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0017] Please see the appendix Figure 1 An optical detection system for identifying gallium nitride lattice defects includes an illumination module, a detection module, a scan control module, a scan execution mechanism, a data processing module, and a display. The illumination module is used to output ultraviolet detection light with periodically switching polarization states and illuminate the gallium nitride wafer. The detection module is used to receive reflected or scattered light from the gallium nitride wafer at the current scan position and perform harmonic demodulation on the detection signal to obtain the first harmonic amplitude and the second harmonic amplitude. The data processing module determines the polarization rotation amount of the pixel corresponding to the current scanning position based on the first harmonic amplitude and the second harmonic amplitude, and compares the polarization rotation amount with the background threshold to determine whether the current pixel is an abnormal pixel or a background pixel. The background threshold is determined based on the mean and standard deviation of the polarization rotation amount of background pixels in the defect-free area obtained in advance. The scanning control module controls the scanning execution mechanism to perform full-area coarse scanning, partial fine scanning, or area integration scanning. The data processing module records the defect identification results and scanning trajectory, and outputs the detection results to the display. An optical detection system for identifying gallium nitride lattice defects includes an illumination module comprising a pulsed laser, a linear polarizer, and a photoelastic modulator. The pulsed laser outputs ultraviolet detection light; the linear polarizer converts the ultraviolet detection light into linearly polarized light; and the photoelastic modulator periodically modulates the polarization of the linearly polarized light. The detection module includes a polarizer, a photodetector, a reference detector, a dual-channel lock-in amplifier, a dark field detection channel, and a position detection unit. The reference detector is used to monitor the laser power. The dual-channel lock-in amplifier is used to perform phase-locked demodulation on the electrical signal converted by the photodetector to obtain the first harmonic amplitude and the second harmonic amplitude. The dark field detection channel is used to output a pixel shielding signal that characterizes whether the pixel corresponding to the current scanning position is effective. The position detection unit is used to receive reflected light and output the centroid offset of the light spot; the scanning execution mechanism includes a dual-axis scanning galvanometer and a piezoelectric platform; the dual-axis scanning galvanometer is used to perform a full-wafer coarse scan; the piezoelectric platform is used to perform a local fine scan and to adjust the tilt angle of the gallium nitride wafer; An optical detection system for identifying gallium nitride lattice defects includes a data processing module that normalizes the detection signal corresponding to the current pixel based on the laser power monitored by the reference detector, and obtains the first harmonic amplitude and the second harmonic amplitude output by the dual-channel lock-in amplifier. When the pixel masking signal indicates that the current pixel is a valid pixel, the polarization rotation amount of the current pixel is determined based on the first harmonic amplitude and the second harmonic amplitude, and the polarization rotation amount is compared with the background threshold; when the polarization rotation amount is greater than the background threshold, the current pixel is determined to be an abnormal pixel; when the polarization rotation amount is less than or equal to the background threshold, the current pixel is determined to be a background pixel. This embodiment provides a basic inspection process for gallium nitride wafer inspection; the illumination module, inspection module, scan control module, scan execution mechanism, data processing module and display are connected by electrical and signal connections to form the whole machine; The polarization state periodically switching ultraviolet detection light refers to the ultraviolet detection light output by the illumination module changing back and forth between different polarization states according to a preset rule within one modulation cycle; the current pixel refers to the single sampling unit processed by the detection module and the data processing module when the scanning execution mechanism stops or sweeps the ultraviolet detection beam across a certain sampling position on the wafer surface; the background threshold refers to the judgment value used to distinguish abnormal pixels from background pixels, which is obtained by statistically analyzing the effective background pixels that have been confirmed to belong to the defect-free area. In this embodiment, the illumination module incident ultraviolet detection light onto the gallium nitride wafer; the ultraviolet detection light is preferably selected from a band higher than the gallium nitride bandgap, so that the detection response mainly comes from the epitaxial layer region; in order to enable the subsequent detection module to separate the polarization information caused by lattice stress from the intensity change, the illumination module first outputs linearly polarized ultraviolet detection light, and then applies periodic polarization modulation to the linearly polarized ultraviolet detection light. Therefore, when the ultraviolet detection light shines on a region without obvious defects, the reflected or scattered light received by the detection module mainly represents the system's background response; when the ultraviolet detection light shines on a dislocation region, the local birefringence caused by the dislocation strain field will change the polarization state of the returning beam, thereby causing changes in the harmonic components in the detection signal. After receiving reflected or scattered light from the gallium nitride wafer, the detection module converts the optical signal into an electrical signal and performs harmonic demodulation according to the modulation frequency and its harmonic components to obtain the first harmonic amplitude and the second harmonic amplitude. During harmonic demodulation, in order to avoid timing and control link conflicts caused by physical hysteresis, there is a physical timing coordination rule between the scanning actuator driving pixel position update and the internal integral filtering of the lock-in amplifier. Specifically, in the full-film coarse scan mode, the dwell time of the ultraviolet detection beam at each current pixel is strictly configured by the hardware clock to be greater than a preset multiple of the integration time constant of the low-pass filter of the dual-channel lock-in amplifier, preferably 3 to 5 times. If the scan dwell time is shorter than this matching relationship, the polarization residual signal of the previous pixel will interfere with the detection of the current pixel through the discharge hysteresis effect of the low-pass filter capacitor, causing the boundary of the microscale stress field to have a trailing artifact in the spatial domain. If the dwell time is too long, it will cause the detection throughput of the entire film to decrease. By locking the above-mentioned multiple relationship between the dwell time and the integration time constant, it is ensured that the harmonic amplitude output at each pixel position tends to and stabilizes at the steady-state value of the physical response in time sequence; the first harmonic amplitude and the second harmonic amplitude referred to here refer to the demodulated output corresponding to the polarization modulation fundamental frequency component and the second harmonic component, respectively. After reading the two amplitude values, the data processing module does not directly use the total light intensity as the defect criterion. Instead, based on the pre-established calibration relationship, it converts the first harmonic amplitude and the second harmonic amplitude into the polarization rotation amount of the current pixel. This polarization rotation amount represents the degree of change of the polarization state of the ultraviolet detection light at the current sampling position relative to the reference state. Its physical source is the modulation of the polarization mode by the local lattice stress field. During the pixel determination process, the data processing module first receives the detection signal corresponding to the current pixel, and then determines whether the pixel meets the conditions for participating in the statistics and determination. If the current pixel corresponds to a valid background area, the polarization rotation amount of the pixel is included in the background statistics set. If the current pixel corresponds to a region with abnormal polarization disturbance, it is not used as a background sample. For the effective set of background pixels, the data processing module calculates the mean and standard deviation of the polarization rotation, and forms a background threshold accordingly. The background threshold is preferably in the form of the mean plus a preset multiple of the standard deviation, where the preset multiple can be a range of values ​​that make the background false judgment rate meet the detection requirements, and preferably a multiple that can distinguish between normal background fluctuations and defect disturbances. Since the polarization rotation amount in the defect-free region is mainly affected by the system background and shot noise, the dislocation region will introduce additional polarization changes that are higher than the statistical fluctuations of the background. Therefore, a threshold is constructed by means and standard deviation to separate background random fluctuations from defect signals. Before the formal scan, the system performs a pre-scan on a known standard defect-free sample or a specified edge defect-free reference region of the current wafer to obtain the initial polarization rotation mean and standard deviation to initialize the background threshold. In the current pixel determination sequence, the detection module first outputs the first harmonic amplitude and the second harmonic amplitude of the current pixel. Then, the data processing module converts these two amplitudes into the polarization rotation amount of the current pixel and compares the polarization rotation amount with the background threshold. When the polarization rotation amount is greater than the background threshold, the data processing module outputs an abnormal pixel mark. When the polarization rotation amount is less than or equal to the background threshold, the data processing module outputs a background pixel mark. This output result is written into the defect identification result cache and sent to the scanning control module as the basis for subsequent scanning path selection. After receiving the abnormal pixel distribution results, the scanning control module controls the scanning execution mechanism to switch between different scanning modes. Full-wafer coarse scanning refers to using a first preset step size to perform a global scan mapping on the entire wafer to obtain the approximate distribution of abnormal pixels. Local fine scanning refers to using a second preset step size smaller than the first preset step size to perform local scanning on the identified candidate abnormal regions in order to improve spatial resolution. Regional integration scanning refers to the planar integration detection of regions with dense anomalous pixels and where individual dislocations are difficult to separate, in order to output stress anomaly results at the regional level. The scanning control module first controls the scanning execution mechanism to complete a coarse scan of the entire film. After identifying anomalous pixel clusters in the coarse scan results, it then decides whether to proceed to a local fine scan or regional integration scan based on the degree of clustering and the size of the region. In this way, the detection throughput and spatial resolution are balanced in the full-film detection, and the repeated high-resolution sampling of non-annomous regions is reduced through a hierarchical scanning mechanism. In a typical detection scenario, the scanning actuator first drives the ultraviolet detection beam to scan the entire gallium nitride wafer in a preset row and column order. The detection module outputs the harmonic demodulation result for each pixel position, and the data processing module performs polarization rotation conversion and background comparison pixel by pixel. If multiple abnormal pixels appear consecutively in a certain local area, the scanning control module registers that area as an area to be further processed. For areas with small size and few abnormal pixels, the scanning control module outputs a local fine scan command; for areas with continuous distribution of abnormal pixels and showing an overlapping trend, the scanning control module outputs an area integration scan command; finally, the data processing module associates and stores the defect identification results, the corresponding scanning trajectory and the judgment category of each area, and displays the defect distribution map, the location of abnormal areas and the scanning path on the display. In this embodiment, the structure further defines the hardware configuration of each module; the pulsed laser in the illumination module is used to provide ultraviolet detection light, the linear polarizer organizes the laser output into linearly polarized light with a defined polarization direction, and the photoelastic modulator applies a periodic phase delay to the linearly polarized light, so that the emitted ultraviolet detection light undergoes polarization state switching within one system cycle. Preferably, the photoelastic modulator operates in its mechanical resonant modulation state, and the modulation peak phase delay is selected as a set value that weakens the DC background term, thereby reducing the background polarization error caused by the modulator itself; the analyzer is set in the return optical path to convert polarization changes into intensity changes; the photodetector converts this intensity change into an electrical signal; the dual-channel lock-in amplifier locks the fundamental frequency and second harmonic reference signals respectively to output the first harmonic amplitude and the second harmonic amplitude; the reference detector detects changes in illumination power; the dark field detection channel is used to detect surface particle scattering and output a pixel shielding signal; the dual-axis scanning galvanometer undertakes the coarse scanning of the entire image, and the piezoelectric platform undertakes the fine scanning of local areas and tilt adjustment; In this embodiment, the normalization process is performed in the following order: the reference detector first samples the reference light synchronized with the main ultraviolet detection light and outputs a reference signal characterizing the current illumination power; in order to prevent the high-frequency carrier from generating parasitic phase delay and disrupting the phase-locked demodulation law due to direct division operation at the radio frequency front end, the dual-channel lock-in amplifier receives the main detection electrical signal and performs real-time phase-locked demodulation, outputting the original first harmonic amplitude value and second harmonic amplitude value. After reading the reference signal, the data processing module demodulates the output of the main detection electrical signal corresponding to the current pixel and performs power normalization to eliminate the direct influence of laser pulse fluctuations on harmonic amplitude. The specific processing logic of this power normalization is as follows: a nominal laser power reference value is preset, and the currently measured reference signal is divided by the nominal laser power reference value to obtain a dimensionless light intensity fluctuation ratio factor; it is then determined whether the light intensity fluctuation ratio factor is within a preset effective power range, for example, between 0.8 and 1.2. If the intensity fluctuation ratio factor is within the effective power range, the data processing module divides the original first harmonic amplitude and the original second harmonic amplitude by the intensity fluctuation ratio factor in the digital domain to obtain the final normalized first harmonic amplitude and second harmonic amplitude. If the intensity fluctuation ratio factor is lower than the lower limit of the effective power range, it indicates that the instantaneous laser power is too weak, which may lead to a low signal-to-noise ratio. The data processing module will directly trigger the pixel shielding signal and mark the current pixel as an invalid pixel. If the intensity fluctuation ratio factor is higher than the upper limit of the effective power range, it indicates that the laser has pulse spike overload, and the current pixel is marked as an invalid pixel. Through digital domain division and normalization with boundary constraints, the overload saturation of the divider caused by the incident light power fluctuation of different pixels is avoided, thereby ensuring the demodulation integrity in the timing, and making the normalized harmonic amplitude stable to characterize the polarization change caused by the local stress field of lattice dislocations. The pixel masking signal is used to indicate whether the current pixel is valid. The valid pixel here refers to a pixel that is not masked by surface particles, abnormal scattering or other artifacts and can be used for polarization calculation and statistical update. The invalid pixel is a pixel that is identified by the dark field detection channel and does not participate in the polarization rotation determination. When the pixel masking signal indicates that the current pixel is a valid pixel, the data processing module converts the first harmonic amplitude and the second harmonic amplitude into the polarization rotation of the current pixel according to the calibration relationship. This calibration relationship can be established using standard samples. Essentially, it maps the phase-locked loop output amplitude to a known polarization rotation response. The specific steps are as follows: Select a series of standard samples with different known polarization rotation values, place them sequentially in the optical path for testing, and record the first harmonic amplitude and second harmonic amplitude corresponding to each known polarization rotation value. Use the first harmonic amplitude and second harmonic amplitude as input coordinates and the polarization rotation value as output value to construct a two-dimensional lookup table. During actual detection and calculation, the data processing module reads the first harmonic amplitude and the second harmonic amplitude of the current pixel and performs nearest neighbor indexing on them in a two-dimensional lookup table. If the coordinate is not on a node in the table, the output values ​​of the four adjacent nodes are extracted, and the polarization rotation of the current pixel is calculated using bilinear interpolation. For example, if the measured first harmonic amplitude is V1 and the second harmonic amplitude is V2, the values ​​of the four rectangular nodes surrounding the coordinate position in the two-dimensional lookup table are obtained, and the horizontal and vertical weighted summation is performed according to the inverse weight of the area from the coordinate point to the row and column of each node to determine the polarization rotation. The data processing module compares the polarization rotation amount of the current pixel with the background threshold. If the result is greater than the background threshold, it is determined to be an abnormal pixel; if the result is less than or equal to the background threshold, it is determined to be a background pixel. If the pixel masking signal indicates that the current pixel is an invalid pixel, then the pixel will not enter the polarization rotation amount comparison process, nor will it participate in the background statistics. Through this processing, pseudo signals caused by surface dust or abnormal scattering will not be written into the result image as real defects. Through the above structure and processing flow, this embodiment uses a combination of polarization modulation, harmonic demodulation and hierarchical scanning to transform the influence of dislocation strain field on polarization state into quantifiable pixel determination results. Compared with detection methods that rely solely on differences in luminous intensity, this embodiment is based on polarization rotation, a physical quantity related to local stress. Therefore, it is more suitable for identifying weak disturbances caused by single or a small number of dislocations, and provides a unified signal input for subsequent dislocation type differentiation and region scanning mode switching.

[0018] An optical detection system for identifying gallium nitride lattice defects includes a data processing module that is also used to identify dislocation types based on the ratio of the first harmonic amplitude to the second harmonic amplitude. The dislocation types include edge dislocations, screw dislocations, and mixed dislocations. The data processing module is used to determine the dislocation type of the current pixel as an edge dislocation, screw dislocation, or mixed dislocation by comparing the ratio with preset upper and lower threshold values. The optical inspection system for identifying gallium nitride lattice defects also includes a scanning control module that determines candidate regions based on the distribution of connected aberrant pixels that meet pre-defined pixel adjacency rules in a two-dimensional pixel array formed by full-wave coarse scanning; calculates the number of valid aberrant pixels contained in the candidate regions; and removes invalid pixels by combining pixel masking signals. The number of valid abnormal pixels is compared with a pixel number threshold, which is set according to a pre-set coarse scan resolution and a calibrated defect size. When the number of valid abnormal pixels is less than or equal to the pixel number threshold, the piezoelectric platform is controlled to perform a local fine scan on the candidate area. When the number of valid abnormal pixels is greater than the pixel number threshold, the candidate area is determined to be a defect overlap area, and the scanning execution mechanism is controlled to perform regional integration scan on the defect overlap area. In an optical inspection system for identifying gallium nitride lattice defects, the data processing module is also used to update the background threshold. When the current pixel is determined to be a background pixel, the mean and standard deviation of the polarization rotation of the background pixels in the defect-free region are updated using the polarization rotation of the background pixel. When the current pixel is determined to be an abnormal pixel, or when the pixel masking signal indicates that the current pixel is an invalid pixel, the current pixel does not participate in the update of the mean and standard deviation of the polarization rotation. Based on the aforementioned pixel anomaly determination, this implementation further provides the specific process of dislocation type identification, candidate region screening, and dynamic background threshold updating; the dislocation type referred to here is the classification of defects based on the polarization response characteristics at the current pixel, including edge dislocations, screw dislocations, and mixed dislocations; The candidate region refers to a local region in the whole-image coarse scan result that is composed of connected abnormal pixels that meet the preset pixel adjacency rules and is to be further determined whether to perform fine scanning or integral scanning; the pixel number threshold refers to the counting threshold that distinguishes the candidate region into distinguishable single defect regions and defect overlapping regions. In the dislocation type identification stage, the data processing module takes the current pixel that has passed the validity judgment and been identified as an abnormal pixel as input; for the current pixel, the data processing module reads the first harmonic amplitude and the second harmonic amplitude and calculates the ratio between the two; before calculating the ratio, the data processing module determines whether the second harmonic amplitude is less than or equal to the preset system background noise tolerance value. If the second harmonic amplitude is less than or equal to the tolerance value, it indicates that the second harmonic signal is too weak to support an effective division operation. The pixel is then directly marked as a weak signal defect that cannot be accurately classified, and no further ratio calculations are performed. If the second harmonic amplitude is greater than the tolerance value, a preset non-zero protection constant is added to the second harmonic amplitude as the denominator, and then the first harmonic amplitude is divided by the denominator. The system calculation logic for this ratio operation uses the formula... By first harmonic amplitude Divide by the second harmonic amplitude With the preset non-zero protection constant The sum of these values ​​is used to calculate the ratio result for type determination. For example, if the protection constant is set to 1e-6, and the measured first harmonic amplitude of an input pixel is 0.05V and the second harmonic amplitude is 0.02V, since it is greater than the background tolerance, the system performs the division operation with a non-zero bias constant to obtain a ratio of approximately 2.500. By introducing a non-zero protection constant, division-to-zero errors can be avoided and ratio abrupt changes caused by small noise fluctuations can be suppressed, ensuring the numerical stability of dislocation type discrimination. This ratio reflects the relative strength of different harmonic components in the current pixel polarization response. Since the local stress distribution corresponding to different dislocation types is different, the response ratio of the returned beam in the fundamental frequency and second harmonic channels is also different. Therefore, this ratio can be used as a dislocation type discrimination quantity. To ensure a consistent basis for judgment, the data processing module pre-stores the upper and lower threshold values ​​obtained from the calibration of standard samples. The standard samples are preferably gallium nitride samples with known dislocation types. By repeatedly detecting the corresponding regions, the distribution range of each type of dislocation in the ratio of the first harmonic amplitude to the second harmonic amplitude can be obtained. Then, the boundary value between adjacent categories is selected as the upper and lower threshold values. In the specific determination, the data processing module first calculates the ratio of the current pixel, and then compares them sequentially according to the conditional branches; if the ratio is greater than the upper threshold, an edge dislocation mark is output; if the ratio is less than the lower threshold, a screw dislocation mark is output; if the ratio is greater than or equal to the lower threshold and less than or equal to the upper threshold, a mixed dislocation mark is output. Preferably, the selection of the upper and lower threshold values ​​satisfies the principle that the ratio distribution ranges of different types of standard samples should be separated as much as possible to reduce misjudgment in the boundary area. In this embodiment, the ratio determination is performed under the condition that abnormal pixels have been screened out. Therefore, the ratio does not play the role of initial screening for defects, but rather the role of further subdividing the defect into different types. This arrangement can prevent background noise pixels from directly entering the dislocation classification process. During the scanning decision-making stage, the scanning control module receives the two-dimensional pixel array obtained from the full-area coarse scan. The pixel adjacency rule referred to here is the adjacency relationship rule used to determine whether two abnormal pixels belong to the same connected region. In response to the problem of data matrix structure breakage caused by unforeseen surface dust and other obstructions during scanning, if the strict four-neighbor or eight-neighbor contact determination is directly applied, the defect response that originally belonged to the same physical strain field will be physically broken by invalid pixels, resulting in topological splitting of the connected graph. Therefore, it is preferable to use an adjacency determination with a through-hole tolerance mechanism: if the spatial Euclidean distance between two abnormal pixels in the two-dimensional scanning array is less than or equal to the preset tolerance span threshold, such as 2 pixel steps, and the matrix elements in the middle are only pixels marked as invalid by the pixel masking signal, then these two groups of abnormal pixels are still merged into the same region by the graph theory algorithm in terms of data structure. Based on the above rules, the scanning control module first traverses each abnormal pixel in the coarse scan result, merges abnormal pixels that satisfy the adjacency rule, and forms one or more connected abnormal pixel sets with topological integrity; then, a candidate region is defined for each connected abnormal pixel set; during the statistical stage, the scanning control module combines the pixel masking signal to not count invalid pixels in the set itself, but only accumulates and sums the valid abnormal pixels, thereby obtaining the exact number of valid abnormal pixels contained in each candidate region, ensuring that the count value will not produce statistical deviation due to the presence of invalid pixels; The pixel number threshold is preset based on the scanning resolution and the calibration defect size during coarse scanning; the scanning resolution refers to the spatial step distance between adjacent pixels during coarse scanning, and the calibration defect size refers to the physical diameter affected by the stress field of a single dislocation measured through a standard sample. To provide a quantitative basis for setting the threshold, the logic for determining the pixel number threshold is as follows: The data processing module divides the pre-acquired calibration defect size by the scanning resolution to calculate the theoretical number of pixels spanned in a single dimension; the theoretical number of pixels spanned in a single dimension is squared to obtain the corresponding theoretical coverage pixel area; considering the spot diffusion and signal spread in the actual scan, the theoretical coverage pixel area is multiplied by a preset tolerance amplification factor, the preferred range of which is 1.2 to 1.5, and the resulting product is rounded up and set as the pixel number threshold. As an example setting: if the size of a single calibrated defect measured in the standard sample is 30 micrometers, and the current coarse scanning system is set to a scanning resolution of 10 micrometers / pixel, then the theoretical span of a single dimension is 3, and its square gives a theoretical coverage pixel area of ​​9; if the system uses a preset tolerance amplification factor of 1.2, the two are multiplied together to get 10.8, and after rounding up, the pixel number threshold is 11; since the projected area of ​​a single dislocation in the coarse scan is limited by the theoretical coverage area, if the number of effective abnormal pixels in a candidate region does not exceed the pixel number threshold, such as 11 in the aforementioned example, it means that the region is more likely to correspond to a single defect or a small number of defects that are separated from each other; If the threshold is exceeded, it indicates that the defect distribution in the region is denser, and multiple dislocation stress fields have interfered or overlapped with each other under the currently set coarse scan resolution. Based on this, the scan control module compares the number of effective abnormal pixels in each candidate region with the pixel number threshold: when the number of effective abnormal pixels is less than or equal to the pixel number threshold, a local fine scan command is output, and the piezoelectric platform performs a local fine scan on the candidate region; when the number of effective abnormal pixels is greater than the pixel number threshold, the candidate region is determined to be a defect overlap region, and a region integration scan command is output, and the scan execution mechanism performs an integration scan on the region. In a continuous processing flow, the dual-axis scanning galvanometer first completes a coarse scan of the entire film. The data processing module forms anomaly markers pixel by pixel and outputs the preliminary judgment result of the dislocation type simultaneously. After the coarse scan is completed, the scanning control module forms candidate regions based on the connectivity distribution of the abnormal pixels in the two-dimensional pixel array. It counts the number of valid abnormal pixels in each candidate region and then decides whether the candidate region enters a local fine scan or a regional integral scan. If a local fine scan is entered, the piezoelectric platform re-covers the candidate region with a smaller step size, improving the resolution of single defect contours. If a regional integration scan is entered, the goal is no longer to separate individual dislocations, but to characterize the overall anomaly degree of the region by the signal integration result within the region. Through this processing method, the scanning time is mainly used for candidate regions with obvious single defect features, while a more suitable regional characterization method is used for densely anomaly regions, avoiding repeated sampling in overlapping defect regions during local fine scans that makes it difficult to obtain single defect results. In this embodiment, the background threshold is not a fixed constant, but is updated as effective background pixels accumulate; during the pixel-by-pixel processing, the data processing module first performs validity and anomaly checks on each current pixel, and then decides whether to update the background statistics. When the current pixel is determined to be a background pixel, it means that the polarization rotation amount of the pixel mainly represents the background level under the current detection conditions. Therefore, the data processing module incorporates the polarization rotation amount of the background pixel into the set of valid background pixels to update the mean and standard deviation of the polarization rotation amount. When the current pixel is determined to be an abnormal pixel, it means that the pixel contains additional polarization changes caused by defects and should not be included in the background statistics. When the pixel masking signal indicates that the current pixel is an invalid pixel, the pixel will not participate in the update of the mean and standard deviation. This configuration is intended to prevent the background threshold from being raised and the sensitivity to real defects from being increased if abnormal or invalid pixels are written into the background statistics. Background threshold updates preferably use a pixel-by-pixel recursive method; each time the data processing module receives a new background pixel, it updates the mean and standard deviation based on the existing background statistics and the polarization rotation of the background pixel, and then recalculates the background threshold based on the updated mean and standard deviation. Specifically, to avoid repeatedly traversing the entire historical data, the update process adopts an online recursive logic: the polarization rotation amount of the current background pixel is subtracted from the old mean at the previous moment, the difference is divided by the current cumulative total number of background pixels, and the resulting correction is added to the old mean to obtain a new mean; at the same time, the product of the polarization rotation amount and the difference between the old and new means is used to accumulate the sum of squared differences, thereby calculating the new standard deviation in real time; the recursive update enables the system to dynamically adjust the judgment benchmark in response to the low-frequency fluctuations of the background. For example, when there is a physical gradient in reflectivity or a steady-state drift in the system background in different regions of the wafer, the dynamically updated background threshold can be closer to the true background level of the current scanned area, thereby reducing the possibility of misjudging normal areas as abnormal areas or submerging weak defects in excessively high thresholds. In the detection scenario corresponding to this embodiment, when the coarse scan enters a certain area, if the surface of the area is clean and without obvious defects, multiple consecutive valid pixels are determined to be background pixels, and the data processing module continuously updates the background mean and standard deviation; if a local dislocation area appears, the polarization rotation amount of the pixel exceeds the current background threshold, the data processing module immediately marks it as an abnormal pixel, and at the same time suspends the entry of the pixel into the background statistics. As the coarse scan continues to move to the subsequent defect-free area, the background statistics are updated again. With this update method, the background statistics are always composed of effective background pixels, and dislocation signals and particle artifacts will not change the background judgment benchmark. Thus, dislocation type identification, candidate region formation and scanning mode switching are all based on a more stable background boundary. Through the above processing, this embodiment continues to complete defect type discrimination after anomaly detection and transforms pixel-level judgment results into region-level scanning decisions; at the same time, by selectively updating the background threshold, the judgment benchmark in the whole-area coarse scan is adjusted according to the change of effective background pixels; thus, the system can distinguish between anomalies and background at the single pixel level, distinguish between finely scannable areas and overlapping areas at the region level, and distinguish between edge dislocations, screw dislocations and mixed dislocations at the defect level.

[0019] In the optical detection system for identifying gallium nitride lattice defects, the data processing module is also used to adjust the driving voltage amplitude of the photoelastic modulator according to the deviation between the output signal of the reference detector and the pre-calibrated reference value, adjust the phase delay of the photoelastic modulator to the pre-calibrated quarter-wave phase delay, and control the piezoelectric platform to adjust the tilt angle of the gallium nitride wafer according to the centroid offset of the light spot output by the position detection unit in the detection module after receiving the reflected light. In the optical detection system for identifying gallium nitride lattice defects, the scanning control module is also used to control the piezoelectric platform to scan the candidate region and its surrounding neighborhood within a pre-set range during local fine scanning. The neighborhood range is pre-set according to the local fine scanning resolution. The scanning control module is also used to perform integral scanning of the defect overlap area by moving a sliding window during the execution of regional integral scanning, and to integrate the detection signal within the window. The window size and step size of the sliding window are determined according to the preset coarse scan resolution. The data processing module is also used to complete the pixels marked as invalid by the pixel masking signal by using a spatial interpolation algorithm based on the detection values ​​of the valid pixels in its neighborhood. An optical inspection system for identifying gallium nitride lattice defects includes a data processing module that records inspection process data, including the amplitude of the first harmonic, the amplitude of the second harmonic, the pixel shielding signal, and the centroid offset of the light spot; records threshold update operations, photoelastic modulator drive voltage adjustment operations, and piezoelectric platform tilt angle adjustment operations performed by the data processing module; and records the coordinate position of the defect, dislocation type, scan path, and corresponding time series. An optical inspection system for identifying gallium nitride lattice defects includes a data processing module that displays the inspection results on a monitor. The results include defect distribution, dislocation type, scan path, and inspection process record. This embodiment provides a detailed implementation of the detection system in terms of modulation stabilization, tilt compensation, fine scanning and integral scanning execution, process recording, and result display. The pre-calibrated quarter-wave phase delay refers to the target phase delay of the photoelastic modulator in the calibration state. The centroid offset of the reflected spot refers to the deviation of the energy center of the reflected beam in the position detection channel from the reference position. The sliding window refers to a scanning calculation unit that moves in a preset step size within the defect overlap area to collect the detection signals of multiple pixels in a local area. In terms of modulation stabilization control, the reference detector continuously outputs a reference signal characterizing the change in reference light intensity. The data processing module compares this reference signal with a pre-calibrated reference value to obtain the current deviation. This reference value corresponds to the signal level that the reference detector should output when the photoelastic modulator is at the target modulation depth. When the deviation is within the allowable range, the data processing module maintains the current drive voltage amplitude of the photoelastic modulator unchanged; when the deviation exceeds the allowable range, the data processing module calculates and outputs the drive adjustment amount based on the proportional-integral control logic to adjust the drive voltage amplitude of the photoelastic modulator; the specific calculation rule is as follows: the data processing module multiplies the current deviation value by a preset proportional coefficient to obtain the proportional term, multiplies the sum of the current deviation and the historical deviation values ​​by a preset integral coefficient to obtain the integral term, adds the proportional term and the integral term to obtain the compensation adjustment amount of the drive voltage, and then adds the compensation adjustment amount to the current drive voltage amplitude; Through this determined feedback adjustment calculation, the corresponding phase delay is brought back to near the pre-calibrated quarter-wave phase delay. This configuration aims to address the fact that the actual phase delay of the photoelastic modulator is affected by changes in the operating state. If the phase delay deviates from the calibration value, the physical correspondence between the first and second harmonic amplitudes will change, thereby affecting the polarization rotation conversion and dislocation type determination. By continuously comparing and quantitatively compensating through a reference detector, the harmonic demodulation results between subsequent pixels can be kept comparable. Regarding tilt adjustment, the detection system determines the tilt state of the current local surface of the wafer relative to the incident light based on the centroid offset of the reflected light spot. The centroid offset of the reflected light spot can be output by the position detection unit in the detection module. After the data processing module reads the offset, it compares it with the reference position to obtain the offset components along the two adjustment directions, such as the X-axis and Y-axis. If the offset components are within the allowable range, the piezoelectric platform maintains the current posture. If the offset component exceeds the allowable range, the data processing module outputs a tilt adjustment command to the piezoelectric platform based on the coordinate mapping relationship. The specific command mapping rule is as follows: the data processing module divides the X-axis offset component and the Y-axis offset component by the angular amplification factor pre-calibrated by the optical path system to calculate the actual angular deviation of the current gallium nitride wafer in the corresponding axis. The negative of the angle deviation is taken as the target compensation angle. Based on the voltage-angle response function of the piezoelectric platform, the required driving voltage increment for each actuator is calculated as the tilt adjustment command output, so that the piezoelectric platform changes the local tilt angle of the wafer until the centroid of the reflected light spot returns to the allowable range. The reason for adopting this method is that the tilt caused by wafer warping or clamping will change the reflection geometry and mix geometric factors into the polarization detection results. By performing inverse quantization calculation based on the centroid offset of the reflected light spot and adjusting the piezoelectric platform, the incident state of the ultraviolet detection light relative to the wafer surface can be kept stable, so that the harmonic variation mainly reflects the polarization variation caused by lattice defects. During the local fine scanning, the scanning control module has already determined the candidate region based on the coarse scanning results. In order to avoid the defect boundary being truncated due to the coarse scanning positioning error, the piezoelectric platform not only scans the candidate region itself, but also performs a coverage scan on the neighborhood within a pre-set range around it. The neighborhood range is pre-set according to the local fine scanning resolution, and its selection principle is: under the condition that the local fine scanning step distance is known, the associated abnormal responses that may be included around the candidate region are also included in the fine scanning range. In terms of execution sequence, the piezoelectric platform first moves to the starting position of the candidate region, and then samples point by point according to the local fine scanning path. The detection module synchronously outputs the first harmonic amplitude, the second harmonic amplitude, and the pixel shielding signal. The data processing module performs the same validity judgment, polarization rotation calculation, and dislocation type identification for each fine-scanned pixel as it does for the coarse scan. Since the resolution of the local fine scan is higher than that of the coarse scan, the area that was originally a few abnormal pixels in the coarse scan can be further separated into the spatial contour and type distribution of individual dislocations in the fine scan image. When performing regional integral scanning, the prerequisite is that the scanning control module has determined that the candidate region is a defect overlap region. At this time, the fine scanning method aimed at identifying a single dislocation is stopped, and the integral scanning of the defect overlap region is performed by moving the sliding window. Specifically, the scanning execution mechanism sets the window size and step size of the sliding window according to the preset scanning resolution. For example, a 3×3 pixel array is set as a single sliding window, and the movement step size is set to 1 pixel. When the sliding window covers the current local area, the data processing module collects the polarization rotation amount of all effective pixels in the window and performs regional integration processing on these polarization rotation amounts: the polarization rotation amounts of all effective pixels in the window are accumulated and summed to obtain the initial total integration amount of the region. Since the number of effective pixels may change due to pixel masking during window movement, the data processing module divides the initial total integration by the total number of effective pixels actually contained in the current sliding window to calculate the average response intensity of a single pixel. Finally, the average response intensity is output as the regional response value representing the overall defect density of the local area corresponding to the sliding window. The sliding window moves to the next position by step size, and the above process is repeated until the entire defect overlap area is covered. The reason for using a moving sliding window is that when multiple dislocation strain fields overlap in space, the classification conditions at the single pixel or single defect level are weakened. Continuing to pursue the separation of a single dislocation will increase the computational complexity of the system and make the results unstable. At this time, it is more appropriate to characterize the overall anomaly of the region by the integral result within the window. For pixels marked as invalid by the pixel masking signal, this implementation does not directly retain them as null values, but instead performs interpolation to complete them based on the detection values ​​of adjacent valid pixels. The specific process is as follows: the data processing module first determines whether there are adjacent valid pixels around the current invalid pixel; if there are, it reads the detection values ​​of the adjacent valid pixels, namely the first harmonic amplitude value and the second harmonic amplitude value, and forms the complete value corresponding to the invalid pixel according to the neighborhood relationship. The neighborhood relationship uses a distance-inverse weighted average algorithm. The data processing module extracts the detection values ​​of all valid pixels within the eight-neighborhood centered on the invalid pixel, calculates the spatial Euclidean distance between each valid pixel and the invalid pixel, and uses the reciprocal of the Euclidean distance as the weight of the corresponding detection value. The first harmonic amplitude and the second harmonic amplitude of each valid pixel are multiplied by their respective weights and summed. Dividing by the sum of all weights, the two results obtained are the supplementary values ​​of the first harmonic amplitude and the second harmonic amplitude of the invalid pixel. Based on spatial interpolation logic, the formula for calculating the padding value of the invalid pixel is as follows: This formula will extract the total number of valid pixels. The Middle The corresponding harmonic amplitude of each effective pixel Multiply by its spatial Euclidean distance to the invalid pixel The inverse of the distance is summed and divided by the sum of the inverses of all spatial Euclidean distances to obtain the invalid pixel padding value to be output. ; For example, if two valid pixels with amplitudes of 0.10V and 0.30V are detected around an invalid pixel at a distance of 1 pixel step, and another valid pixel with an amplitude of 0.20V at a distance of 1.414 pixel step, the data processing module calculates the completion value to be approximately 0.18V based on the aforementioned inverse weighted logic. If there are insufficient adjacent valid pixels, the masking mark of the invalid pixel is retained, and completion is performed after sufficient valid pixels are obtained in a larger neighborhood. This interpolation completion is only used for continuity processing during the image formation stage and region integration. It does not change the rule that invalid pixels do not participate in defect judgment, background statistics, and dislocation classification. The principle of the above processing is that if individual invalid pixels caused by surface dust are left blank, it will cause broken areas in the local result image. Interpolation completion can improve the continuity of display and the stability of the integration area. Regarding the recording of the detection process, the data processing module saves the corresponding detection process data for each pixel or each scanning window; the recorded data includes at least the first harmonic amplitude, the second harmonic amplitude, the pixel shielding signal, and the centroid offset of the reflected light spot; in addition to the original detection data, the data processing module also records the threshold update operation, the photoelastic modulator drive voltage adjustment operation, and the piezoelectric platform tilt angle adjustment operation performed by itself. The recording operation referred to here is not simply about retaining the final result, but rather about storing the parameter changes corresponding to each update or adjustment in association with the scanning position. For example, when the background threshold is updated during the scanning of a certain area, the data processing module associates the statistical values ​​before and after the update with the corresponding pixel segment; when the reference signal deviation triggers the adjustment of the photoelastic modulator drive voltage, the adjustment amount is associated with the corresponding scanning stage; when the reflected light spot shift causes a change in the piezoelectric platform's attitude, the adjustment amount is associated with the corresponding position. After recording in this way, when reviewing the results later, it is not only possible to know what kind of defect exists at a certain location, but also to trace back to the detection state under which the defect was obtained. In addition to process quantities, the data processing module also records the coordinates of the defect, the type of dislocation, the scan path, and the corresponding time series. The time series referred to here is a sequence formed by arranging various detection data and control data according to the system sampling order, which is used to represent the continuous change relationship of each parameter during the scanning process. By mapping the defect coordinates, dislocation type, and scan path, the operator can view on the monitor whether a certain defect was discovered by coarse scanning, confirmed by local fine scanning, or characterized by regional integral scanning. By correlating the first harmonic amplitude, the second harmonic amplitude, the pixel shielding signal, the centroid offset of the reflected light spot with the relevant adjustment operations, it is also possible to determine whether the detection results are affected by local contamination, changes in the incident tilt angle, or modulation depth drift. Regarding the result display, the data processing module organizes the recorded data into detection results and outputs them through the display; the display content includes at least the defect distribution, dislocation type, scan path, and detection process record; preferably, the display shows the defect location in the form of a two-dimensional distribution map of the wafer, and uses different display marks for edge dislocations, screw dislocations, and mixed dislocations; Different paths or boundary markers are used for areas confirmed by local fine scanning and areas processed by regional integral scanning; for the detection process record, the corresponding harmonic amplitude changes, pixel shielding status, centroid shift of reflected light spot, and threshold update and adjustment operation records can be retrieved by selecting the area; thus, when users view the final results, they can not only obtain the spatial distribution and type distribution of defects, but also obtain the detection process information corresponding to the results, which is convenient for judging the reliability of the results of a certain area and the source of the scanning method; In one application scenario, the system first performs a coarse scan of the entire wafer to form candidate regions. A local fine scan is then performed on one type of candidate region, while an integral scan is performed on another type of defect-overlapping region. During the scan, if the reference detector output deviates from the baseline value, the data processing module adjusts the driving voltage amplitude of the photoelastic modulator. If the centroid of the reflected light spot shifts, the piezoelectric platform is controlled to adjust the wafer tilt angle. If the dark field detection channel marks an invalid pixel, that pixel is not included in defect judgment and statistics, and is filled in during the image formation stage using adjacent valid pixels. Finally, the display outputs a defect distribution map of the entire wafer, which includes markers for different dislocation types, coarse and fine scan paths, integral scan areas, and corresponding detection process records for each area. Through the above implementation method, the detection results not only reflect the defects themselves, but also retain the modulation state, posture adjustment state, and pixel validity information during the detection process, which is beneficial for reviewing the detection results of the entire wafer.

[0020] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. An optical detection system for identifying gallium nitride lattice defects, characterized in that, It includes a lighting module, a detection module, a scanning control module, a scanning actuator, a data processing module, and a display. The illumination module is used to output ultraviolet detection light with periodically switched polarization states and irradiate the gallium nitride wafer. The detection module is used to receive the reflected or scattered light from the gallium nitride wafer at the current scanning position, and to perform harmonic demodulation on the detection signal to obtain the first harmonic amplitude and the second harmonic amplitude. The data processing module is used to determine the polarization rotation amount of the pixel corresponding to the current scanning position based on the first harmonic amplitude and the second harmonic amplitude, and compare the polarization rotation amount with the background threshold to determine whether the current pixel is an abnormal pixel or a background pixel. The background threshold is determined based on the mean and standard deviation of the polarization rotation amount of background pixels in the defect-free area obtained in advance. The scanning control module is used to control the scanning execution mechanism to perform full-film coarse scanning, partial fine scanning, or regional integral scanning. The data processing module is used to record defect identification results and scanning trajectories, and outputs the detection results through the display.

2. The optical detection system for identifying gallium nitride lattice defects according to claim 1, characterized in that: The illumination module includes a pulsed laser, a linear polarizer, and a photoelastic modulator; the pulsed laser is used to output the ultraviolet detection light; the linear polarizer is used to convert the ultraviolet detection light into linearly polarized light; and the photoelastic modulator is used to periodically polarize the linearly polarized light. The detection module includes a polarizer, a photodetector, a reference detector, a dual-channel lock-in amplifier, a dark field detection channel, and a position detection unit. The reference detector is used to monitor the laser power. The dual-channel lock-in amplifier is used to perform phase-locked demodulation on the electrical signal converted and output by the photodetector to obtain the first harmonic amplitude and the second harmonic amplitude. The dark field detection channel is used to output a pixel shielding signal characterizing whether the pixel corresponding to the current scanning position is effective. The position detection unit is used to receive the reflected light and output the spot centroid offset. The scanning execution mechanism includes a dual-axis scanning galvanometer and a piezoelectric platform; the dual-axis scanning galvanometer is used to perform a full-wafer coarse scan; the piezoelectric platform is used to perform a local fine scan and to adjust the tilt angle of the gallium nitride wafer.

3. The optical detection system for identifying gallium nitride lattice defects according to claim 2, characterized in that: The data processing module is also used to normalize the detection signal corresponding to the current pixel according to the laser power monitored by the reference detector, and obtain the first harmonic amplitude and the second harmonic amplitude output by the dual-channel lock-in amplifier. When the pixel masking signal indicates that the current pixel is a valid pixel, the polarization rotation amount of the current pixel is determined based on the first harmonic amplitude and the second harmonic amplitude, and the polarization rotation amount is compared with the background threshold. When the polarization rotation amount is greater than the background threshold, the current pixel is determined to be an abnormal pixel; When the polarization rotation amount is less than or equal to the background threshold, the current pixel is determined to be a background pixel.

4. The optical detection system for identifying gallium nitride lattice defects according to claim 3, characterized in that: The data processing module is also used to identify the dislocation type based on the ratio of the first harmonic amplitude to the second harmonic amplitude, wherein the dislocation type includes edge dislocation, screw dislocation and mixed dislocation; The data processing module is used to determine whether the dislocation type of the current pixel is an edge dislocation, a screw dislocation, or a mixed dislocation by comparing the ratio with preset upper and lower threshold values.

5. The optical detection system for identifying gallium nitride lattice defects according to claim 4, characterized in that: The scanning control module is also used to determine candidate regions based on the distribution of connected abnormal pixels in the two-dimensional pixel array generated by the full-slice coarse scan that meet the preset pixel adjacency rules. Calculate the number of valid abnormal pixels contained in the candidate region, and remove invalid pixels based on the pixel masking signal; The number of valid abnormal pixels is compared with a pixel number threshold, which is set according to a preset coarse scan resolution and a calibration defect size. When the number of valid abnormal pixels is less than or equal to the pixel number threshold, the piezoelectric platform is controlled to perform a local fine scan on the candidate region. When the number of valid abnormal pixels is greater than the number of pixels threshold, the candidate region is determined to be a defect overlap region, and the scanning execution mechanism is controlled to perform regional integral scanning on the defect overlap region.

6. The optical detection system for identifying gallium nitride lattice defects according to claim 5, characterized in that: The data processing module is also used to update the background threshold; When the current pixel is determined to be a background pixel, the mean and standard deviation of the polarization rotation of the background pixels in the defect-free region are updated using the polarization rotation of the background pixel. When the current pixel is determined to be an abnormal pixel, or when the pixel masking signal indicates that the current pixel is an invalid pixel, the current pixel will not participate in the update of the mean and standard deviation of the polarization rotation.

7. The optical detection system for identifying gallium nitride lattice defects according to claim 6, characterized in that: The data processing module is also used to adjust the driving voltage amplitude of the photoelastic modulator according to the deviation between the output signal of the reference detector and the pre-calibrated reference value, and adjust the phase delay of the photoelastic modulator to the pre-calibrated quarter-wave phase delay. Based on the centroid offset of the light spot output after the reflected light is received by the position detection unit in the detection module, the piezoelectric platform is controlled to adjust the tilt angle of the gallium nitride wafer.

8. The optical detection system for identifying gallium nitride lattice defects according to claim 7, characterized in that: The scanning control module is further configured to, during local fine scanning, control the piezoelectric platform to scan the candidate region and its surrounding neighborhood within a pre-defined range, wherein the neighborhood range is pre-defined according to the local fine scanning resolution; the scanning control module is further configured to, during regional integration scanning, use a moving sliding window to perform integration scanning on the defect overlap region and integrate the detection signal within the window, wherein the window size and step size of the sliding window are determined according to a pre-defined coarse scanning resolution; the data processing module is further configured to, for pixels marked as invalid by the pixel masking signal, use a spatial interpolation algorithm to complete the pixel based on the detection value of the valid pixels in its neighborhood.

9. The optical detection system for identifying gallium nitride lattice defects according to claim 8, characterized in that: The data processing module is also used to record detection process data, which includes the first harmonic amplitude, the second harmonic amplitude, the pixel shielding signal, and the spot centroid offset. Record the threshold update operation performed by the data processing module, the drive voltage adjustment operation of the photoelastic modulator, and the tilt angle adjustment operation of the piezoelectric platform; It also records the coordinates of the defect, the type of dislocation, the scan path, and the corresponding time series.

10. The optical detection system for identifying gallium nitride lattice defects according to claim 9, characterized in that: The data processing module is used to display the detection results on the display, and the detection results include defect distribution, dislocation type, scan path and detection process record.

Images