Wafer intrinsic defect classification method and system based on infrared penetration characteristics

Abstract

The present application relates to a wafer intrinsic defect classification method and system based on infrared penetration characteristics, comprising: driving a scanning module to scan the wafer globally to generate a suspected defect positioning signal with absolute physical coordinates; detecting the corresponding wafer warping bias to generate an analog modulation voltage signal; dynamically adjusting the infrared exposure time and longitudinal down-step pitch according to the signal to construct a dynamic non-diffraction scanning trajectory; performing a walking fault scan along the trajectory to capture an anti-deformation infrared transmission image sequence with depth labels; processing the sequence and extracting the Z-domain phase inversion feature evolving with depth; if the light intensity cross-over turning point is analyzed, it is classified as the first intrinsic defect of physical refraction effect; if it presents a monotonic linear decay without abrupt change, it is classified as the second extrinsic defect of pure physical obstruction and the data is output. The present application solves the problems of depth measurement deviation caused by wafer warping and the difficulty in effectively distinguishing defects of different physical causes.

Description

Technical Field

[0001] This invention belongs to the technical field of semiconductor testing, and relates to a method and system for classifying intrinsic defects in wafers based on infrared penetration characteristics. Background Technology

[0002] In semiconductor device manufacturing, wafer defect detection is a crucial step in ensuring product yield and reliability. With increasing chip integration and the application of three-dimensional stacked structures, the technical challenge of detection lies not only in detecting minute defects but also in accurately classifying them. This is particularly important in distinguishing between intrinsic wafer defects with significant performance impacts, such as bubbles and dislocations, and surface-level defects like particles and contaminants. Intrinsic wafer defects are hidden within the material; their type, three-dimensional location, and size affect subsequent processes and the electrical performance of the final device. Therefore, accurate physical characterization and localization of these defects is a major technical requirement currently facing the industry.

[0003] To address these needs, the industry currently employs various optical and non-optical inspection solutions. Surface scanning systems based on the visible light scattering principle can quickly detect surface defects on wafers, but they struggle to effectively acquire information about the wafer's interior. To detect internal defects, methods capable of penetrating silicon, such as infrared optical microscopy or X-ray computed tomography, are commonly used. Taking infrared transmission imaging as an example, this technology utilizes the high transmittance of silicon in specific wavelengths. By acquiring image sequences at different focal planes with a fixed step size, it reconstructs the three-dimensional morphology of defects. Then, it extracts morphological features such as size, shape, and contrast from the images and combines these features with relevant algorithmic models for statistical classification.

[0004] However, the aforementioned existing technologies have certain limitations in addressing complex inspection needs. First, wafers often undergo macroscopic warping deformation after thermal processing. When faced with warping, the aforementioned infrared microscopy system using fixed-step scanning is prone to causing the focal plane to become non-parallel to the wafer surface, resulting in out-of-focus images in some areas. This adversely affects the accuracy of defect depth localization and morphology analysis. Second, the aforementioned classification methods that rely solely on morphological features and statistical models have insufficient physical interpretability. Defects with different physical origins, such as internal microbubbles and deeply buried metal particles, may present similar morphologies in two-dimensional projections or blurred three-dimensional reconstruction images. This can easily lead to confusion in the classification model, making it difficult to provide accurate physical qualitative conclusions. For example, after identifying anomalies in light intensity, existing methods struggle to effectively distinguish whether the anomaly is caused by internal refraction effects or purely by physical occlusion. Summary of the Invention

[0005] In a first aspect, the present invention provides a method for classifying intrinsic defects in wafers based on infrared penetration characteristics, comprising the following steps: S1. Drive the infrared scanning module to perform a global scan of the wafer, extract the transmission abnormal area and generate a suspected defect location signal containing absolute physical coordinates; S2. Detect the wafer warpage bias at the physical location corresponding to the suspected defect location signal, and generate an analog modulated voltage signal for underlying hardware interlocking based on the wafer warpage bias. S3. Based on the analog modulation voltage signal, dynamically adjust the exposure time of the infrared detection component and the downward stepping distance of the longitudinal stepper to construct a dynamic aberration-free scanning trajectory that eliminates wavefront distortion. S4. Drive the infrared array acquisition module to perform a wandering tomography along a dynamic aberration-free scanning trajectory to capture a deformation-resistant infrared transmission image sequence with absolute depth labels. S5. Process the deformation-resistant infrared transmission image sequence to aggregate pixel blocks with overlapping spatial coordinates and extract the Z-domain phase reversal features that evolve layer by layer with the detection depth. S6. Analyze the nonlinear photoelectric abrupt change process within the phase reversal feature of the Z domain. In response to the capture of the light intensity crossover reversal point, classify and grade the defects in the corresponding region as first intrinsic defects with physical refraction effects. S7. In response to the monotonically linear decay without abrupt changes in the phase reversal characteristics of the Z-domain, the defects in the corresponding region are classified and graded as second intriguing defects with pure physical occlusion, and the closed-loop data of physical grading of wafer defects is output.

[0006] A further aspect of the present invention involves driving an infrared scanning module to perform a global scan of the wafer, comprising the following steps: The initial two-dimensional infrared grayscale image penetrating the wafer is captured using a large field-of-view time-delay integral infrared scanning module. Thresholding segmentation is performed on the initial infrared two-dimensional grayscale image to extract continuous dark spot pixel regions; The geometric center of the continuous dark spot pixel region is mapped to the physical position in the wafer absolute coordinate system through the coordinate calibration matrix, and the suspected defect location signal containing horizontal and vertical two-dimensional spatial coordinates is output.

[0007] A further aspect of the present invention involves detecting wafer warpage bias at the physical location corresponding to a suspected defect location signal, comprising the following steps: The suspected defect location signal is used as a hardwired trigger switch to drive the surface manifold coherent encoder to release coherent laser to the horizontal and vertical two-dimensional spatial coordinates. Receive the interference echo reflected back from the physical surface of the wafer by coaxial coherent laser, and analyze the radius of curvature of the current tiny region in the horizontal and vertical two-dimensional spatial coordinates and the optical astigmatism bias vector as the wafer warping bias; The radius of curvature and the optical astigmatism bias vector are transformed into a continuous topological waveform that characterizes the local thickness distortion of the wafer, and the continuous topological waveform is amplified by an operational amplifier and output as an analog modulated voltage signal.

[0008] A further aspect of the present invention involves dynamically adjusting the exposure time of the infrared detection component and the downward stepping distance of the longitudinal stepping actuator, comprising the following steps: The analog modulation voltage signal is loaded into the exposure timing generator of the infrared phase gate array driver, which is an infrared detection component, and the photoelectric integration exposure time of a single local infrared pixel in the infrared phase gate array driver is dynamically adjusted by using the voltage amplitude of the analog modulation voltage signal. The inverted component of the analog modulated voltage signal is used to reshape the longitudinal micron-level downward stepping distance of the piezoelectric stepper, which is used as a longitudinal stepper, in real time when it cuts into the wafer thickness. By substituting the photoelectric integration exposure time and the longitudinal micron-level downward step spacing into the displacement compensation, a dynamic aberration-free scanning trajectory that eliminates wavefront distortion is generated.

[0009] A further aspect of the present invention involves driving an infrared array acquisition module to perform a wandering tomography scan along a dynamic aberration-free scanning trajectory, comprising the following steps: Using the high-light reflection point established by the pre-set coaxial coherent laser on the physical surface of the wafer as the absolute zero longitudinal starting point, the light-emitting and pixel photosensitive elements of the short-wave infrared array acquisition module, which serves as the infrared array acquisition module, are activated. The constrained shortwave infrared array acquisition module performs a non-equidistant downward cutting action on the vertical focusing plane along the dynamic aberration-free scanning trajectory, according to the timing sequence of the preset longitudinal micron-level downward stepping interval and the preset photoelectric integration exposure time. The transmitted photon stream that completely penetrates the corresponding regions of the horizontal and vertical two-dimensional spatial coordinates is continuously captured, and the aggregated output is a sequence of deformation-resistant infrared transmission images with sequentially increasing depth labels.

[0010] A further aspect of this invention involves extracting the Z-domain phase reversal features that evolve layer by layer with the detection depth, including the following steps: By stacking the anti-deformation infrared transmission image sequence according to the sequentially increasing depth labels, a virtual three-dimensional photon transmission energy field characterizing the transmission attenuation probability inside the wafer is constructed. Lock the pixel block formed by the central pixel that vertically penetrates the horizontal and vertical two-dimensional spatial coordinates and the surrounding ring of adjacent pixels, and track the light intensity gradient trend of the pixel block from shallow to deep in the virtual three-dimensional photon transmission energy field. The photon spatial phase transition curves generated by the transmitted light intensity of the central pixel and the diffracted light intensity of the surrounding ring of pixels when crossing medium planes of different depths are extracted to generate Z-domain phase reversal features.

[0011] A further aspect of the present invention, in response to the capture of a light intensity crossover point, classifies and categorizes it as a first intrinsic defect with a physical refraction effect, comprising the following steps: Along the depth coordinates of the Z-domain phase reversal feature, we identify divergent phase nodes where the central pixel exhibits a local minimum and a ring of adjacent pixels derives a bright diffraction ring and exhibits a local maximum. By comparing the Z-domain phase reversal features under deeper dielectric layers, the refocusing phase nodes that detect the light intensity playback of the central pixel and show local maxima, and the energy extraction, collapse and darkening of the bright diffraction ring and show local minima are detected. When diverging phase nodes and reconverging phase nodes appear sequentially in a continuous longitudinal depth to form an intensity crossover point, the current region is classified as a first intrinsic defect. First intrinsic defects include bubbles and pores hidden inside the wafer.

[0012] A further aspect of the present invention, in response to the Z-domain phase reversal feature exhibiting a monotonically linear decay without abrupt changes, classifies and categorizes it as a second eigenvalue defect of pure physical occlusion, comprising the following steps: Determine whether the representative light intensity of the center pixel in the Z-domain phase reversal feature conforms to the monotonic photon occlusion absorption property as the absolute depth label increases; If the entire jump curve fails to capture the preset light intensity crossover point and confirms that it only exhibits monotonous linear decay, it is determined that the refractive index rearrangement has not occurred at the interface of the probe medium, and it is forcibly classified and graded as a second intriguing defect. The second intriguing defect includes surface particles or surface protrusions. Summarize the information of the first intrinsic defect and the second non-intrinsic defect, and output the closed-loop data of wafer defect physical classification.

[0013] A further aspect of the present invention involves summarizing information on the first intrinsic defect and the second non-intrinsic defect to output closed-loop data for wafer defect physical classification, including the following steps: Obtain the three-dimensional depth coordinates and equivalent astigmatic volume of the first intrinsic defect; Extract the two-dimensional coordinates of the surface of the second intrinsic defect; The three-dimensional depth coordinates, equivalent astigmatic volume, and surface two-dimensional coordinates are structured and encapsulated to generate and output closed-loop data for wafer defect physical classification based on pure physical underlying optical features.

[0014] Secondly, the present invention provides a wafer intrinsic defect classification system based on infrared penetration characteristics, comprising the following modules: The defect location module is used to drive the infrared scanning module to perform a global scan of the wafer, extract the transmission abnormal area, and generate a suspected defect location signal containing absolute physical coordinates. The warpage detection module is used to detect the wafer warpage bias at the physical location corresponding to the suspected defect location signal, and to generate an analog modulated voltage signal for underlying hardware interlocking based on the wafer warpage bias. The scanning trajectory generation module is used to dynamically adjust the exposure time of the infrared detection component and the downward stepping distance of the longitudinal stepper actuator according to the analog modulation voltage signal, so as to construct a dynamic aberration-free scanning trajectory that eliminates wavefront distortion. The tomographic image acquisition module is used to drive the infrared array acquisition module to perform a wandering tomographic scan along a dynamic aberration-free scanning trajectory and capture a deformation-resistant infrared transmission image sequence with absolute depth labels. The optical feature extraction module is used to process the deformation-resistant infrared transmission image sequence to aggregate pixel blocks with overlapping spatial coordinates and extract the Z-domain phase reversal features that evolve layer by layer with the detection depth. The intrinsic defect classification module is used to analyze the nonlinear photoelectric abrupt change process within the phase reversal feature of the Z domain. In response to the capture of the light intensity crossover reversal point, the defects in the corresponding region are classified and graded as first intrinsic defects with physical refraction effects. The defect classification and output module is used to classify and classify the defects in the corresponding region as second intriguing defects with pure physical occlusion in response to the monotonous linear decay without abrupt changes in the phase reversal characteristics of the Z-domain, and output closed-loop data of wafer defect physical classification.

[0015] In summary, the present invention has the following beneficial technical effects: 1. This invention, after locating suspected defects, utilizes a surface manifold coherent encoder to detect localized wafer warpage bias and generate an analog modulated voltage signal. This signal is then used to directly modulate the longitudinal stepping spacing of subsequent piezoelectric stepper actuators. This mechanism converts the physical deformation parameters of the wafer surface into displacement compensation for Z-axis scanning in real time, enabling the downward stepping of tomographic scanning to dynamically follow the wafer's surface contour. Through this hardware-closed-loop real-time compensation method, this invention constructs a dynamic, aberration-free scanning trajectory during tomographic scanning, reducing focal plane drift caused by wafer warpage, thereby improving the depth positioning accuracy of subsequent 3D data acquisition.

[0016] 2. The analog modulation voltage signal used in this invention drives the longitudinal stepper actuator and simultaneously modulates the photoelectric integration exposure time of the infrared detection component. This mechanism can dynamically adjust the detector's exposure time for adaptive control based on the optical astigmatism or focusing effect caused by wafer warping. For example, when a local protrusion is detected causing transmitted light divergence and signal weakening, the system automatically extends the exposure time to supplement the photon collection. This adjustment method ensures that the intensity of the infrared transmitted signal collected on a specific warped surface remains within a relatively stable range, improving the signal-to-noise ratio of the original image sequence and providing a better data foundation for subsequent accurate physical feature extraction.

[0017] 3. This invention processes infrared transmission image sequences with deformation resistance to extract Z-domain phase reversal features that evolve layer by layer with detection depth, and uses these features to physically classify defects. This feature tracks the evolution of the intensity contrast between the defect center point and its surrounding diffraction rings with depth. The classification logic is that low-refractive-index inclusions such as bubbles and pores inside the wafer cause the transmitted infrared light to first diverge and then refocus, thus forming characteristic crossover points on the intensity curve; while surface particles and other obstructive defects exhibit monotonic intensity attenuation. By identifying this optical behavior difference with specific physical significance, this invention can effectively distinguish between intrinsic internal defects and extrinsic surface defects, thereby improving the accuracy and physical interpretability of defect classification. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. The drawings are used to provide a further understanding of the present invention.

[0019] Figure 1 A flowchart illustrating an embodiment of this application is disclosed.

[0020] Figure 2 Structural schematic diagrams of embodiments of this application are disclosed. Detailed Implementation

[0021] The following is in conjunction with the appendix Figures 1-2 A preferred description of the present invention is provided below.

[0022] See attached document Figure 1 This invention proposes a wafer intrinsic defect classification method based on infrared penetration characteristics, comprising the following steps: S1. Drive the infrared scanning module to perform a global scan of the wafer, extract the transmission abnormal area and generate a suspected defect location signal containing absolute physical coordinates; S2. Detect the wafer warpage bias at the physical location corresponding to the suspected defect location signal, and generate an analog modulated voltage signal for underlying hardware interlocking based on the wafer warpage bias. S3. Based on the analog modulation voltage signal, dynamically adjust the exposure time of the infrared detection component and the downward stepping distance of the longitudinal stepper to construct a dynamic aberration-free scanning trajectory that eliminates wavefront distortion. S4. Drive the infrared array acquisition module to perform a wandering tomography along a dynamic aberration-free scanning trajectory to capture a deformation-resistant infrared transmission image sequence with absolute depth labels. S5. Process the deformation-resistant infrared transmission image sequence to aggregate pixel blocks with overlapping spatial coordinates and extract the Z-domain phase reversal features that evolve layer by layer with the detection depth. S6. Analyze the nonlinear photoelectric abrupt change process within the phase reversal feature of the Z domain. In response to the capture of the light intensity crossover reversal point, classify and grade the defects in the corresponding region as first intrinsic defects with physical refraction effects. S7. In response to the monotonically linear decay without abrupt changes in the phase reversal characteristics of the Z-domain, the defects in the corresponding region are classified and graded as second intriguing defects with pure physical occlusion, and the closed-loop data of physical grading of wafer defects is output.

[0023] In one embodiment of the present invention, step S1 includes the following steps: An initial infrared two-dimensional grayscale image penetrating the wafer is captured using a large field-of-view time-delay integral infrared scanning module; threshold segmentation is performed on the initial infrared two-dimensional grayscale image to extract continuous dark spot pixel regions; the geometric center position of the continuous dark spot pixel regions is mapped to the physical position in the wafer absolute coordinate system through a coordinate calibration matrix, and a suspected defect location signal containing horizontal and vertical two-dimensional spatial coordinates is output.

[0024] In this embodiment, the infrared scanning module can specifically be a large field-of-view time-delay integral infrared scanning module. Specifically, the large field-of-view time-delay integral infrared scanning module is driven to perform a global scan of the wafer, extracting areas of transmission anomalies and generating suspected defect location signals containing absolute physical coordinates. This step is coordinated and executed by the central control unit of an integrated wafer defect classification system.

[0025] The central control unit sends a digital activation command to a programmable current source coupled to a short-wave infrared linear array light source. The short-wave infrared linear array light source comprises an array of laser diodes arranged linearly perpendicular to the scanning direction, emitting spectral energy concentrated at a central wavelength of [missing information]. Within a narrow band, uniform linear transmission illumination is provided to the back of the target wafer fixed on a vacuum adsorption wafer chuck through beam shaping elements such as Powell lenses or microlens arrays.

[0026] The central control unit simultaneously activates the linear translation stage and image acquisition card of the large field-of-view time-delay integrated infrared scanning module. The linear translation stage drive module operates at a preset constant speed. Scanning is performed along the wafer diameter. The photosensitive chip inside the time-delay integration infrared scanning module, such as a multi-level indium gallium arsenide focal plane array, continuously receives infrared photons penetrating the target wafer during the scanning process. Its time-delay integration mode synchronously accumulates the charges generated at the same point at different times in the column transfer register of the photosensitive chip, thereby obtaining a high signal-to-noise ratio initial infrared two-dimensional grayscale image under high-speed scanning. This image... Based on pixel coordinates The grayscale matrix is ​​indexed.

[0027] The image processing module performs threshold segmentation on the initial infrared two-dimensional grayscale image, automatically calculating the grayscale threshold using the Otsu method or pre-setting it based on historical statistical data. Pixels with gray values ​​below a gray threshold in the image are identified as transmission anomalous points. Then, using an eight-neighbor or four-neighbor connectivity component labeling algorithm, spatially adjacent transmission anomalous points are aggregated into several independent, continuous dark spot pixel regions. For each extracted continuous dark spot pixel region Calculate its geometric center pixel coordinates .

[0028] By calling the coordinate calibration matrix pre-calibrated and stored by the system. Set the coordinates of each geometric center pixel. Linear mapping from the pixel coordinate system of a two-dimensional image to the physical location in the absolute coordinate system of the target wafer. The completion of this process signifies the generation of a structured dataset containing the horizontal and vertical two-dimensional spatial coordinates of all abnormal areas, i.e., the suspected defect location signals, which is then cached on the high-speed data bus for subsequent steps.

[0029] In this implementation, the mapping from pixel coordinates to physical coordinates can be achieved through affine transformation, the mathematical expression of which is as follows: In the formula, and These are the physical x and y coordinates of the target wafer in the absolute coordinate system, obtained through calculation. and These are the x and y coordinates of the geometric center pixel of the continuous dark spot pixel region in the initial infrared two-dimensional grayscale image. It is a coordinate calibration matrix, which includes rotation and scaling factors. and translation coefficient It is determined by taking pictures of a standard calibration board with known physical coordinates and performing least squares fitting during equipment manufacturing or maintenance.

[0030] In one embodiment of the invention, the center wavelength can be set to 1050 nm. This value is an optimized selection based on the spectral transmittance curve analysis of standard doped silicon materials. In this band, silicon has a low absorption coefficient, allowing infrared light to effectively penetrate the entire wafer thickness, while this wavelength remains within the response range of high-performance indium gallium arsenide detectors. Grayscale threshold The setting is based on the following: assuming the application scenario is standard silicon wafer inspection with uniform background transmittance, a fixed grayscale threshold can be set, for example, three standard deviations below the average background grayscale; or, in the case of uneven background light intensity, the Otsu method can be used to dynamically calculate the global threshold to improve the robustness of segmentation.

[0031] It should be noted that the coordinate calibration matrix The accuracy is established in the offline calibration process. This process uses a quartz calibration plate engraved with a precise grid or feature point array. A high-precision motion platform moves the calibration plate to multiple known physical locations, and images are captured by a large field-of-view time-delay integral infrared scanning module. Finally, the values ​​of each element of the matrix are obtained by solving an overdetermined linear equation system, ensuring that the mapping accuracy is on the micrometer scale. In a specific implementation, the suspected defect location signal can be a first-in-first-out (FIFO) queue. Each element in the queue is a tuple containing two floating-point numbers, each corresponding to a suspected defect point. and Physical coordinates.

[0032] In one example, the central control unit drives a large field-of-view time-delay integrated infrared scanning module to complete a global scan, generating an initial 2D infrared grayscale image of 20480px × 20480px, with a grayscale range of 0 to 4095 (12-bit). The image processing module then calculates and sets a grayscale threshold. The value is 850. During image analysis, a continuous dark spot pixel region was identified, whose pixel coordinate set contains three points: {(5120, 10240), (5121, 10240), (5120, 10241)}. The geometric center pixel coordinate of this region is... Calculated The coordinate calibration matrix of the system call. It was obtained through prior calibration; let's assume its value is: This matrix represents a 0.005mm dimension for each pixel, with a physical offset of (10.5, 12.0)mm from the origin of the coordinate system. Substituting the coordinates of the geometric center pixel into the formula yields the physical coordinates. ,as well as Therefore, the suspected defect location signal ultimately generated by the system and transmitted to downstream steps contains specific coordinate points. The coordinates precisely point to a transmission anomaly location on the physical surface of the target wafer.

[0033] In one embodiment of the present invention, step S2 includes the following steps: The suspected defect location signal is used as a hardwired trigger switch to drive the surface manifold coherent encoder to release coaxial coherent laser to the horizontal and vertical two-dimensional spatial coordinates; the interference echo reflected back from the physical surface of the wafer by the coaxial coherent laser is received, and the radius of curvature and optical astigmatism bias vector of the current small region in the horizontal and vertical two-dimensional spatial coordinates are analyzed as the wafer warpage bias; the radius of curvature and optical astigmatism bias vector are converted into a continuous topological waveform characterizing the local thickness distortion of the wafer, and the continuous topological waveform is amplified by an operational amplifier and output as an analog modulated voltage signal.

[0034] In this embodiment, the specific method for detecting wafer warpage bias is as follows: A surface topography detection device, such as a surface manifold coherent encoder, is activated to detect the wafer warpage bias at the physical location corresponding to the suspected defect location signal, generating an analog modulation voltage signal for underlying hardware interlocking. This signal is triggered immediately after the suspected defect location signal is generated.

[0035] Specifically, the central control unit will use the horizontal and vertical two-dimensional spatial coordinates contained in the suspected defect location signal. As a hardware address, it is directly loaded into the motion controller of the two-dimensional precision electronically controlled translation stage of the surface manifold coherent encoder. The motion controller then drives the optical probe of the surface manifold coherent encoder to move above the target wafer surface, so that its optical axis is precisely aligned with the horizontal and vertical two-dimensional spatial coordinates.

[0036] After positioning is complete, the system immediately triggers a coaxial coherent laser source within the surface manifold coherent encoder via hardwired connection. This source, such as a frequency-stabilized helium-neon laser or a distributed feedback laser diode, emits a collimated Gaussian laser beam towards a tiny region on the wafer's physical surface corresponding to the two-dimensional spatial coordinates. This laser beam is incident perpendicularly to the target wafer surface after passing through a beam splitter and is reflected. The receiving unit of the surface manifold coherent encoder, such as a Shaker-Hartmann wavefront sensor equipped with a microlens array, captures the interference echo reflected from the wafer's physical surface and re-passing through the beam splitter. By analyzing the displacement of the focal spot array formed on the Shaker-Hartmann wavefront sensor relative to the ideal grid position under the reference plane wave, the wavefront phase distortion of the interference echo can be reconstructed. The signal processor fits the wavefront phase distortion using Zernike polynomials and extracts the low-order aberration components that dominate the wavefront shape, i.e., the radius of curvature. With optical astigmatism bias vector .

[0037] Subsequently, a dedicated digital-to-analog converter and signal conditioning circuit converts the resolved radius of curvature and optical astigmatism bias vector magnitudes into instantaneous levels characterizing local thickness distortion of the wafer through a preset weighted mapping function. The continuous variation of this level constitutes the continuous topological waveform. Finally, this instantaneous level is fed into the non-inverting input of an operational amplifier for voltage amplification. The operational amplifier's gain is set according to its feedback network. At the output, an analog modulated voltage signal with sufficient driving capability and a voltage amplitude proportional to the wafer warpage bias is generated. It then outputs the data to subsequent hardware modules.

[0038] In this embodiment, the generation process of the analog modulated voltage signal can be described by the following formula: In the formula, It is the final output analog modulated voltage signal. It is the closed-loop voltage gain of the operational amplifier. It is the radius of curvature obtained from analysis. It is the optical astigmatism bias vector obtained by analysis. It is the proportionality coefficient that converts curvature into voltage. It is the proportionality coefficient that converts the magnitude of astigmatism into voltage. It is a DC bias voltage used to ensure the output reference level when there is no warping bias.

[0039] It should be noted that a surface manifold coherent encoder is a non-contact optical profile measurement device. Essentially, it utilizes the principle of light interference to accurately reproduce the three-dimensional shape of an object's surface. Coaxial coherent laser refers to a laser whose emission and reception paths share most of the optical axis, ensuring the perpendicularity and accuracy of the measurement. Optical astigmatism bias vector. It is a two-dimensional vector whose magnitude represents the size of the astigmatism and whose direction represents the axis of the astigmatism. Together with the radius of curvature, it describes the local quadratic surface shape of the wafer at that point.

[0040] The continuous topology waveform is a time-continuous voltage signal. Its amplitude reflects the warping changes along the path traversed by the optical probe as it scans the wafer surface in real time; it is not a static value. (Scale factor) and These are the system calibration parameters, whose values ​​are set based on the response sensitivity of the downstream hardware, namely the infrared phase-gated array driver, to the modulation voltage, to ensure that the warpage bias can be accurately translated into effective hardware control commands. The operational amplifier gain... It is configured to amplify the converted weak voltage signal to an industry-standard control voltage range, such as 0V to 5V or -10V to +10V.

[0041] For example, the central control unit receives the coordinates from the suspected defect location signal. The optical probe of the surface manifold coherent encoder is precisely moved to this position. Suppose that at this moment, the Shackleton wavefront sensor determines that there is a slight bulge warp on the wafer surface at this point, with an equivalent radius of curvature... for Simultaneously, a small optical astigmatism bias vector exists, whose magnitude... The value is 0.05 diopter. The system's preset calibration parameter is: scaling factor. The proportionality coefficient is 200 V·m. 10V / diopter, DC bias voltage The voltage is 0.1V. First, the signal conditioning circuit calculates the unamplified voltage value, which is calculated as follows: The 3.1V signal is then fed into an operational amplifier configured as a non-inverting amplifier, whose feedback resistor-to-input resistor ratio is set to 1, thereby increasing its voltage gain. The value is 2. Therefore, the final analog modulated voltage signal output from the operational amplifier is 2. The amplitude is This 6.2V voltage signal, as an analog quantity that accurately encodes the physical warping information at that point, is transmitted in real time and without delay to the infrared phase-gated array driver in subsequent steps to construct dynamic compensation.

[0042] In one embodiment of the present invention, step S3 includes the following steps: An analog modulated voltage signal is loaded into the exposure timing generator of the infrared phase-gated array driver, which serves as the infrared detection component. The photoelectric integration exposure time of a single local infrared pixel in the infrared phase-gated array driver is dynamically adjusted using the voltage amplitude of the analog modulated voltage signal. The analog modulated voltage signal is then sent to a precision inverting operational amplifier circuit to generate an inverted component with the opposite polarity to the original signal. The inverted component is then sent to the power drive module that drives the piezoelectric stepper actuator. The power drive module adjusts the voltage of the next drive pulse applied to the piezoelectric stepper actuator in real time according to the voltage amplitude of the inverted component, thus reshaping the longitudinal micron-level downward step spacing of a single downward probe. The photoelectric integration exposure time and the longitudinal micron-level downward step spacing are substituted into displacement compensation to generate a dynamic aberration-free scanning trajectory that eliminates wavefront distortion.

[0043] In this embodiment, the infrared detection component is specifically embodied as an infrared phase-gated array driver, and the longitudinal stepper is specifically embodied as a piezoelectric stepper actuator. Specifically, after generating an analog modulation voltage signal, the photoelectric integration exposure gate of the infrared phase-gated array driver and the piezoelectric stepper actuator are modulated, using the analog modulation voltage signal to construct a dynamic, aberration-free scanning trajectory that eliminates wavefront distortion. This process is the core of the opto-mechatronics closed-loop control. The analog modulation voltage signal... The signals are fed into two independent signal processing paths in parallel: In the first path, an analog modulated voltage signal is directly applied to the voltage control input of the exposure timing generator within the infrared phase-gated array driver. This timing generator contains a high-frequency relaxation oscillator circuit whose oscillation period determines the integration time of the infrared pixels. The analog modulated voltage signal modulates the width of the output pulse by altering the charging or discharging rate of the key timing capacitor in this oscillator circuit, thus physically replacing the static clock reference provided by the internal fixed-frequency crystal oscillator with an electrical signal overlay. Specifically, an increase or decrease in the voltage amplitude will linearly lengthen or shorten the photoelectric integration exposure time.

[0044] In the second path, the analog modulated voltage signal is fed into an inverting operational amplifier circuit to generate an inverted component with the opposite polarity to the original signal. The inverted component is then used as a control signal and sent to the power drive module that drives the piezoelectric stepper actuator. The piezoelectric stepper actuator is responsible for driving the piezoelectric stepper actuator perpendicular to the wafer surface. The piezoelectric stepper performs micrometer-level displacement along the axial direction. The power drive module adjusts the voltage of the next drive pulse applied to the piezoelectric stepper in real time based on the voltage amplitude of the inverting component, thereby reshaping the step spacing of its single downward movement.

[0045] By converting the voltage signal corresponding to wafer surface warping into synchronous reverse compensation for exposure time and step spacing, the scanning path executed by the system constitutes a dynamic aberration-free scanning trajectory that can actively counteract the influence of wafer macroscopic deformation.

[0046] In this embodiment, the photoelectric integration exposure time is dynamically adjusted. and longitudinal micron-level downward step spacing It can be described by the following formula: In the formula, In the first Photoelectric integration exposure time at each detection point. The baseline exposure time corresponds to a perfectly flat wafer surface. It is the exposure time modulation coefficient, and its unit is s / V. In the The amplitude of the analog modulated voltage signal generated in step S2 at each detection point. In the first The next longitudinal micrometer-level downward probe step interval after each detection point, in the formula... This is the function for finding the maximum value. This is a baseline step spacing. It is a step pitch modulation coefficient, and its unit is m / V. With a preset minimum safe step-down threshold, such as 0.1μm, this clamping constraint logic can prevent damage caused by detecting extreme wafer warpage. If the calculated value is negative or zero, ensure that the probe is in a safe cutting state without disrupting the scan queue.

[0047] It should be noted that the infrared phase-gated array driver is a highly integrated control unit that not only manages the pixel readout of the infrared focal plane array but also synchronously controls the related mechanical movements. The electrical signal coverage method of the exposure timing generator means that the system's dynamic response speed is limited only by the response bandwidth of the electronic components, which is much faster than traditional control methods based on software calculation and feedback, achieving real-time compensation at the hardware level.

[0048] A piezoelectric stepper actuator is a device that uses the inverse piezoelectric effect to achieve precise displacement. It has a fast response speed and displacement resolution down to the nanometer level, making it suitable for achieving micrometer-level stepping. (Reference exposure time) and reference step spacing These are system parameters that are pre-set based on the optical properties of the wafer material being tested, the expected size of the defects, and the desired scanning speed.

[0049] Exposure time modulation coefficient and step pitch modulation coefficient This is a key calibration parameter of the system. Its value is obtained by scanning a standard part with a known warp morphology, analyzing the changes in signal intensity and aberration, and then calculating it using an optimization algorithm to ensure that the compensation effect is optimal.

[0050] For example, continuing from the previous embodiment, the surface manifold coherent encoder detected a protrusion warp at the coordinate point (36.10165, 63.20165) mm and output an analog modulated voltage signal with an amplitude of 6.2V, i.e. Assume the system's preset baseline parameter is: baseline exposure time. 100μs, exposure time modulation coefficient 5 μs / V; reference step spacing The step spacing modulation coefficient is 2.0 μm. The value is 0.1 μm / V. First, calculate the photoelectric integration exposure time at this point: This means that, since the detected warping of the protrusion may cause signal divergence, the system automatically extends the exposure time from the baseline of 100 μs to 131 μs to ensure sufficient photon collection. Simultaneously, the next vertical micrometer-level downward step spacing is calculated: This means that, in order to follow the protruding wafer surface contour, the system automatically reduces the next vertical penetration distance from the standard 2.0 μm to 1.38 μm. Therefore, on the dynamic aberration-free scanning trajectory, the scanning action for this specific suspected defect point is precisely defined as: acquiring data with an exposure time of 131 μs, and then penetrating 1.38 μm into the wafer interior, thereby creating aberration-free acquisition condition with real-time physical compensation for subsequent tomographic scanning.

[0051] In one embodiment of the present invention, step S4 includes the following steps: Using the high-reflection point established by the pre-set coaxial coherent laser on the physical surface of the wafer as the absolute zero longitudinal starting point, the light-emitting and pixel photosensitive elements of the short-wavelength infrared array acquisition module, which serves as the infrared array acquisition module, are activated. The short-wavelength infrared array acquisition module is constrained to perform non-equidistant downward cutting along the dynamic aberration-free scanning trajectory according to the time sequence of the pre-set longitudinal micron-level downward stepping spacing and the pre-set photoelectric integration exposure time, and continuously captures the transmitted photon flow that completely penetrates the corresponding regions of the horizontal and vertical two-dimensional spatial coordinates, and aggregates and outputs a deformation-resistant infrared transmission image sequence with sequentially increasing depth labels.

[0052] In this embodiment, the infrared array acquisition module specifically employs a short-wave infrared array acquisition module. Specifically, after constructing a dynamic aberration-free scanning trajectory, the short-wave infrared array acquisition module is driven to perform a walking tomographic scan along the dynamic aberration-free scanning trajectory, capturing a deformation-resistant infrared transmission image sequence with absolute depth labels. This step is crucial for acquiring the three-dimensional structural information inside the wafer.

[0053] First, the central control unit confirms that the coherent laser has established a stable specular reflection point on the physical surface of the wafer. The location of this reflection point within the optical system is defined as... The absolute zero longitudinal starting point of axis detection, i.e., depth. Subsequently, the central control unit sends a synchronization trigger signal, simultaneously activating the light-emitting units and pixel photosensitive elements inside the short-wave infrared array acquisition module. The light-emitting unit, such as a high-brightness short-wave infrared LED array light source, focuses illumination on the corresponding area in the horizontal and vertical two-dimensional spatial coordinates. At the same time, the pixel photosensitive elements of the short-wave infrared array acquisition module, namely the high-resolution InGaAs focal plane array, begin preparing for the first photon integration.

[0054] Based on the dynamically generated timing commands in S3, the motion system of the shortwave infrared array acquisition module strictly constrains its optical lens to ensure that it follows the first modulated photoelectric integration exposure time. and the first modulated longitudinal micrometer-level downward step spacing The system operates on sequences that interact. Specifically, the system in The position, with The exposure time is used to capture the first frame of the image, which records the transmitted photon stream that completely penetrates the corresponding regions in the horizontal and vertical two-dimensional spatial coordinates, and assigns it an absolute depth label. .

[0055] After the data acquisition is completed, the piezoelectric stepper actuator immediately performs a non-equidistant vertical downward cutting motion, the displacement of which is precisely equal to... In the new location The system then uses a new exposure time. Capture the second frame image and assign a depth label. This process repeats continuously, capturing a series of images until the total depth reaches a preset value, such as penetrating the entire wafer thickness. In this way, the system aggregates and outputs a sequence of deformation-resistant infrared transmission images, eliminating astigmatic interference caused by macroscopic deformation, and each frame contains an in-order depth label.

[0056] In this embodiment, the deformation-resistant infrared transmission image sequence can be formally represented as a set. Each frame of the image Capture and depth tagging The calculation follows the following recursive relationship: in, It is the first The absolute depth label of the frame image. . In capturing the first After the frame image, the longitudinal micron-level downward step spacing is calculated based on the warping information at that point. It is the first A frame of two-dimensional infrared transmission image. It is based on the first The photoelectric integral exposure time is calculated from the warp information of each detection point. The function represents performing an image capture action at a specified depth and exposure time. It is the total number of slices in the computed tomography scan.

[0057] In this embodiment, the short-wave infrared array acquisition module is used for high-magnification microscopic imaging. It and the large field-of-view scanning module in S1 are two independent physical components; the former prioritizes resolution, while the latter prioritizes scanning speed. The establishment of the speckle center, a bright speckle pattern formed on the wafer surface by a coherent coherent laser, is used as a physical reference. The relative displacement of the axes provides a zero point unaffected by optical system aberrations. Non-equidistant downcutting is one of the core features of this invention, meaning that the scanning step size is not fixed but dynamically adjusted according to the local topography of the wafer surface, which is the basis for achieving deformation-resistant scanning.

[0058] The sequentially increasing depth labels are not simply image sequence numbers, but floating-point values ​​strictly correlated with the physical depth distance. Their precision directly determines the accuracy of subsequent 3D reconstruction and defect localization. The "deformation-resistant" characteristic of the deformation-resistant infrared transmission image sequence stems from the synergistic effect of steps S3 and S4. Through real-time compensation at the hardware level, the acquired image sequence can reflect the internal structure of an object in an ideally flat state, even on a warped wafer, providing high-quality raw data for subsequent defect analysis.

[0059] For example, following the foregoing embodiment, the shortwave infrared array acquisition module initiates a wandering tomographic scan at the coordinate point (36.10165, 63.20165) mm. The scan start point is defined as the absolute zero longitudinal starting point. According to S3's calculations, the photoelectric integration exposure time for the first acquisition... The duration is 131 μs. After completing the acquisition, the system performs its first vertical micrometer-level downward probe step. The value is 1.38 μm. Therefore, the system captures the first frame image. And assign it a depth label Subsequently, the piezoelectric stepper actuator moved the optical lens down 1.38 μm. At the new depth At this point, the system reactivates the surface manifold coherent encoder. Assuming the detected warpage decreases, a new exposure time is calculated. The new step spacing is 115 μs. The depth was 1.85 μm. The system then captured a second frame at this depth using an exposure time of 115 μs. and assign it a deep label. Subsequently, the piezoelectric stepper actuator moves down another 1.85 μm. This process continues, for example, acquiring a total of 100 layers. The system ultimately outputs a deformation-resistant infrared transmission image sequence containing 100 pairs (image, depth label): Each depth label is the sum of the previous depth and the corresponding step spacing. This sequence faithfully records the transmission characteristics inside the wafer at different depths, and depth measurement errors caused by wafer warping have been eliminated through hardware means.

[0060] In one embodiment of the present invention, step S5 includes the following steps: Stack the anti-deformation infrared transmission image sequence according to the sequentially increasing depth labels to construct a virtual three-dimensional photon transmission energy field characterizing the transmission attenuation probability inside the wafer; lock the pixel block formed by the central pixel and the surrounding ring of pixels that vertically penetrate the horizontal and vertical two-dimensional spatial coordinates; track the light intensity gradient trend of the pixel block from shallow to deep in the virtual three-dimensional photon transmission energy field; extract the photon spatial phase jump curves generated by the transmitted light intensity of the central pixel and the diffracted light intensity of the surrounding ring of pixels when crossing medium planes of different depths, and generate Z-domain phase reversal features.

[0061] The image analysis processor processes the deformation-resistant infrared transmission image sequence to aggregate pixel blocks with overlapping spatial coordinates, and extracts the layer-by-layer evolution with detection depth. Domain phase reversal features. The aim is to extract optical-physical features in three-dimensional space from a series of two-dimensional images.

[0062] The processor increments the depth label sequentially. Deformation-resistant infrared transmission image sequence Virtual 3D spatial alignment is performed in computer memory. Specifically, it aligns each frame of the image... As a two-dimensional slice, placed in a three-dimensional coordinate system On the plane, a voxel data structure is constructed, namely a virtual three-dimensional photon transmission energy field. Each voxel in this energy field... The value represents the physical coordinates. ,depth The intensity of infrared transmitted photons at that location. Next, the system uses the horizontal and vertical two-dimensional spatial coordinates determined in S1. A vertically oriented voxel column is locked within a virtual three-dimensional photon transmission energy field. Centered on this voxel column, the system defines a two-dimensional pixel template, which represents the horizontal and vertical spatial coordinates. The corresponding voxel is defined as the center pixel, and the eight or more pixels immediately surrounding the center pixel in a ring shape are defined as the outer ring of adjacent pixels. These two parts together constitute a pixel block.

[0063] The processor tracks this pixel block in the virtual three-dimensional photon transmission energy field, along... The gradient trend of light intensity along the axis from light to dark, that is, from... arrive The specific approach is to, for each depth layer Calculate the average gray value of the center pixel. The average gray value of the ring of pixels adjacent to the outer edge .

[0064] The function curves showing how these two light intensity values ​​change with depth, i.e. and These two curves, exhibiting relative intensity changes and spatial phase relationships when traversing medium planes of different depths, constitute the photon spatial phase jump curve characterizing photon propagation behavior. This curve is the one ultimately generated in this step. Domain phase reversal feature.

[0065] In this embodiment, The core of the domain phase reversal feature is two light intensity variation curves with depth, which are calculated as follows: In the formula, For depth The intensity of transmitted light at the center pixel. For depth The average diffraction intensity of adjacent pixels on the periphery. It is a function used to obtain an image. Corresponding physical coordinates The pixel grayscale value. Indicates revolving around the center point A circular set of pixel coordinates. It represents the total number of pixels within the annular region.

[0066] It should be noted that the virtual 3D photon transmission energy field is a digitized 3D array, which provides the data foundation for subsequent nonlocal feature analysis. The definition of the central pixel and the surrounding ring of pixels is based on the point spread function (PSF) or diffraction pattern analysis method in simulating optics. The light intensity of the central pixel mainly reflects the energy of the directly transmitted photons, while the light intensity of the ring of pixels mainly reflects the energy of photons that have deviated from the center due to scattering or diffraction.

[0067] Tracking the intensity gradient trend is a quantification of the dynamic evolution of the optical effects of defects. The photon space phase jump curve is a key intermediate product of this invention. It does not refer to the actual phase of the electromagnetic wave, but rather uses the term phase to vividly describe the reversal or jump phenomenon of the relative strength relationship between the central light intensity and the ring light intensity. This jump is directly related to the physical refraction and diffraction processes that occur at the interface of the medium.

[0068] For example, following the previous embodiment, the processor receives a sequence of 100 frames of deformation-resistant infrared transmission images acquired at coordinates (36.10165, 63.20165) mm. The processor first constructs a three-dimensional voxel space and labels each frame of image according to its depth. Insert. Then, lock the voxel column centered at physical coordinates (36.10165, 63.20165) that runs through the 3D space. For each depth layer... ,For example The processor reads the pixel value at that coordinate point as the transmitted light intensity of the center pixel, assuming its grayscale value is... Simultaneously, the grayscale values ​​of the surrounding eight pixels are read, assuming they are {1480, 1495, 1510, 1520, 1515, 1505, 1490, 1485}, and their average value is calculated to obtain the diffraction intensity of the surrounding adjacent pixels. The processor repeats this calculation for all 100 depth layers. Assuming at depth... Nearby, the calculated light intensity value is ,and ; in a deeper At that location, the light intensity value becomes ,and Calculate all depth layers and Connecting these values ​​forms two curves that vary with depth. One curve represents the change in light intensity at the center, and the other represents the change in diffracted light intensity in the annular region; together, they constitute the... Domain phase reversal feature. This feature is stored in the form of a graph or data pair for further analysis of the physical phenomenon.

[0069] In one embodiment of the present invention, step S6 includes the following steps: Along the depth coordinates of the Z-domain phase reversal feature, divergent phase nodes are identified where the central pixel exhibits a local minimum and a ring of adjacent pixels derives a bright diffraction ring exhibiting a local maximum. By comparing the Z-domain phase reversal feature in deeper dielectric layers, refocusing phase nodes are detected where the central pixel's light intensity replays and exhibits a local maximum, and the bright diffraction ring's energy is extracted, collapses, and darkens, exhibiting a local minimum. When divergent and refocusing phase nodes appear sequentially in a continuous longitudinal depth to form a light intensity crossover point, the current region is classified as a first intrinsic defect. First intrinsic defects include bubbles and pores hidden inside the wafer.

[0070] The nonlinear photoelectric abrupt change process within the Z-domain phase reversal feature is analyzed. In response to the capture of the intensity crossover point, it is classified as a first intrinsic defect exhibiting physical refraction. This step is performed by a rule-based pattern recognition engine configured to analyze the two intensity curves contained within the Z-domain phase reversal feature. This engine operates along the depth coordinates... ,from arrive The evolution direction, scanning the transmitted light intensity curve of the center pixel point by point. Diffraction intensity curves of adjacent pixels .

[0071] During this process, the engine will recognize a specific pattern: the light intensity of the center pixel. A local minimum occurs, and at that minimum point... Nearby, the light intensity value shows a decreasing trend; at the same time, the diffracted light intensity of adjacent pixels on the periphery... At this point The area near this point reaches a distinct local maximum, forming a bright diffraction ring. When the central darkening and outer ring brightening occur simultaneously, this depth point... These are the nodes that are marked as candidates for diverging phase.

[0072] Subsequently, the pattern recognition engine detects the existence of complementary patterns within a preset depth window following the diverging phase node, moving along the direction of increasing depth. The feature configuration of this complementary pattern is: the light intensity of the center pixel. After the previous descent, at another depth point The rebound at the point forms a local maximum; at the same time, the diffracted light intensity of the surrounding adjacent pixels... At this point A localized drop in light intensity occurs nearby. This complementary phenomenon is satisfied at this depth point. The nodes are marked as candidate reconverging phase nodes. Finally, when a diverging phase node... With a refocusing phase node Sequential association and matching along a continuous vertical depth, i.e. At that time, the feature formed by this pair of nodes is defined as the light intensity crossover point.

[0073] The system performs classification based on preset optical judgment rules: the aforementioned optical behavior of first diverging and then refocusing corresponds to the physical phenomenon caused by local refraction effects when probe light passes through low-refractive-index inclusions (such as bubble pores) in a high-refractive-index medium. Therefore, when the feature sequence contains a complete light intensity crossover point, the current analysis region is classified as a primary intrinsic defect hidden inside the wafer.

[0074] In this embodiment, the system identifies the existence of light intensity crossover points by determining whether the following preset conditions are met simultaneously: Condition 1 (Depth Timing Condition): Within the preset scanning depth range, divergent phase nodes and convergent phase nodes are detected sequentially, and the depth coordinate value of the convergent phase node is greater than the depth coordinate value of the divergent phase node. Condition 2 (Divergence Feature Condition): Within the local depth window where the diverging phase node is located, the transmitted light intensity of the central pixel exhibits a local minimum, and the diffracted light intensity of the corresponding peripheral adjacent pixels exhibits a local maximum. Condition 3 (Re-convergence feature condition): Within the local depth window where the re-convergence phase node is located, the transmitted light intensity of the central pixel exhibits a local maximum, and the diffracted light intensity of the corresponding peripheral adjacent pixels exhibits a local minimum.

[0075] A valid light intensity crossover point is determined to have been extracted if and only if all three of the above conditions are met simultaneously.

[0076] It should be noted that the requirement for the diverging phase node and the converging phase node to be sequentially associated in the longitudinal depth eliminates misjudgments caused by random noise or other types of isolated light intensity fluctuations.

[0077] For example, the pattern recognition engine begins analyzing the generated Z-domain phase inversion features. The engine scans two light intensity curves and finds that in the depth... At the center of the light intensity It is the minimum value in its neighborhood, while the ring light intensity , is the maximum value in its nearest neighborhood. The system will match the depth of the feature of this diverging phase node. Marking. The engine continues scanning downwards, at depth... At this location, the central light intensity was detected. It exhibits a peak value, while the ring light intensity The system will match the depth of the re-aggregated phase node features, exhibiting a valley value. Marking is performed. At this point, the system detects divergent and re-aggregating node pairs, and Based on this judgment logic, the system determines that there are bubble-like physical obstructions between 55.2μm and 64.8μm depth inside the wafer at physical coordinates (36.10165, 63.20165) mm, and classifies them as first-intrinsic defects.

[0078] In one embodiment of the present invention, step S7 includes the following steps: Determine whether the representative light intensity of the central pixel in the Z-domain phase reversal feature conforms to the monotonic photon occlusion absorption property as the absolute depth label increases; if the entire jump curve does not capture the preset light intensity crossover point and confirms that only monotonic linear decay is present, it is determined that the refractive index rearrangement has not occurred at the probe medium interface, and it is forcibly classified and graded as the second intriguing defect. The second intriguing defect includes surface particles or surface protrusions; summarize the information of the first intrinsic defect and the second intriguing defect, and output the wafer defect physical grading closed-loop data.

[0079] If the Z-domain phase inversion feature exhibits a monotonically decreasing decay without abrupt changes, it is classified as a second intriguing defect caused by pure physical occlusion, and the closed-loop data for wafer defect physical classification is output. First, the classification engine performs a second-mode matching on the Z-domain phase inversion feature generated by S5. The system analyzes the transmitted light intensity of the center pixel as a function of the absolute depth label. The increased evolutionary process, i.e., the curve This is to determine whether it conforms to the exponential decay model as a whole.

[0080] Specifically, the system attempts to use a preset decay function. The transmitted light intensity data of the center pixel is fitted, where It is the initial light intensity. It is the attenuation coefficient. If the coefficient of determination obtained from the fitting... If the score exceeds a preset goodness threshold, such as 0.95, the curve segment is determined to exhibit monotonic photon occlusion absorption properties. Based on this, the system performs a final confirmation check: on the corresponding Z-domain phase reversal characteristic transition curve, it confirms that no intensity crossover point satisfying the definition in step S6 has been captured. When the attenuation curve exhibits monotonic attenuation and no corresponding phase reversal node exists, the system determines that the attenuation caused by the probe beam in the propagation path is primarily due to physical occlusion or material absorption, rather than a hole-like refraction effect. The system then classifies the target area as a second intrinsic defect, which includes, but is not limited to, opaque particles or surface protrusions located on the surface. All analysis results are then summarized to generate a structured report.

[0081] If the target is identified as a primary intrinsic defect, the report will include its three-dimensional depth coordinates, as well as the equivalent astigmatic volume estimated by combining the nodal depth difference and the change in light intensity; if it is identified as a secondary intriguing defect, the report will record its two-dimensional surface coordinates. This dataset constitutes a closed-loop dataset for physical classification of wafer defects and can be configured to be output externally via the system interface.

[0082] In this embodiment, the classification and grading logic for the second intriguing defect must simultaneously satisfy independent fitting and exclusion conditions: Fitting conditions: The extracted transmitted light intensity data of the center pixel is fitted to a preset monotonic attenuation model, for example... When the fit index is greater than the preset threshold, it is determined that the light intensity attenuation process conforms to the physical occlusion characteristics. Exclusion condition: The system confirms that within the current corresponding scanning area, the paired divergent phase nodes and convergent phase nodes have not been extracted according to the logic described in step S6, that is, the effective light intensity cross-flip feature has not been extracted to exclude the inclusion refraction effect.

[0083] Based on a comprehensive assessment of the above conditions, the system classifies the anomaly point with monotonic light intensity attenuation and no abrupt change in refraction phase as a second eigendefect.

[0084] For example, suppose that in the aforementioned analysis, the pattern recognition engine did not extract the feature representation of the corresponding intensity crossover point in the Z-domain phase reversal feature. The engine continues to execute the judgment logic configured in step S7: extract the transmitted light intensity curve of the center pixel. The system calculates and determines that, throughout the entire depth scanning range (e.g., 0 to 775 μm), the light intensity decreases from an initial grayscale value of 3800 to 500 without any fluctuation or rebound characteristics. After fitting this curve data to the attenuation model, the system obtains the coefficient of determination. The value is 0.98, which is greater than the set threshold of 0.95. Simultaneously, the system verified and ruled out the existence of divergent and convergent phase nodes. Based on the above data parameters, the system determined that the optical response characteristics of this region conformed to physical occlusion behavior, and ultimately classified the defect located at the corresponding physical coordinates (36.10165, 63.20165) mm as a second intriguing defect (corresponding physical morphology as surface particles). The system further recorded the corresponding entry in the wafer defect physical classification closed-loop data report: "Defect type: Second intriguing defect; Location two-dimensional coordinates: (36.10165, 63.20165) mm". Conversely, if the previous embodiment (first intrinsic defect) holds true, the corresponding output entry is configured as: "Defect type: First intrinsic defect; Three-dimensional coordinates: (36.10165, 63.20165, 55.2) mm; Equivalent astigmatic volume". : [System calculated value]".

[0085] See appendix Figure 2 The present invention also proposes a wafer intrinsic defect classification system based on infrared penetration characteristics, comprising the following modules: The defect location module is used to drive the infrared scanning module to perform a global scan of the wafer, extract the transmission abnormal area, and generate a suspected defect location signal containing absolute physical coordinates. The warpage detection module is used to detect the wafer warpage bias at the physical location corresponding to the suspected defect location signal, and to generate an analog modulated voltage signal for underlying hardware interlocking based on the wafer warpage bias. The scanning trajectory generation module is used to dynamically adjust the exposure time of the infrared detection component and the downward stepping distance of the longitudinal stepper actuator according to the analog modulation voltage signal, so as to construct a dynamic aberration-free scanning trajectory that eliminates wavefront distortion. The tomographic image acquisition module is used to drive the infrared array acquisition module to perform a wandering tomographic scan along a dynamic aberration-free scanning trajectory and capture a deformation-resistant infrared transmission image sequence with absolute depth labels. The optical feature extraction module is used to process the deformation-resistant infrared transmission image sequence to aggregate pixel blocks with overlapping spatial coordinates and extract the Z-domain phase reversal features that evolve layer by layer with the detection depth. The intrinsic defect classification module is used to analyze the nonlinear photoelectric abrupt change process within the phase reversal feature of the Z domain. In response to the capture of the light intensity crossover reversal point, the defects in the corresponding region are classified and graded as first intrinsic defects with physical refraction effects. The defect classification and output module is used to classify and classify the defects in the corresponding region as second intriguing defects with pure physical occlusion in response to the monotonous linear decay without abrupt changes in the phase reversal characteristics of the Z-domain, and output closed-loop data of wafer defect physical classification.

[0086] Each of the modules can be implemented in whole or in part through software, hardware, or a combination thereof. It supports hardware embedded in or independent of the processor in the computer device, and also supports software stored in the memory of the computer device, so that the processor can call and execute the operations corresponding to each of the above modules.

[0087] 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 the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A wafer intrinsic defect classification method based on infrared penetration characteristics, characterized in that, Includes the following steps: S1. Drive the infrared scanning module to perform a global scan of the wafer, extract the transmission abnormal area and generate a suspected defect location signal containing absolute physical coordinates; S2. Detect the wafer warpage bias at the physical location corresponding to the suspected defect location signal, and generate an analog modulated voltage signal for underlying hardware interlocking based on the wafer warpage bias. S3. Based on the analog modulation voltage signal, dynamically adjust the exposure time of the infrared detection component and the downward stepping distance of the longitudinal stepper to construct a dynamic aberration-free scanning trajectory that eliminates wavefront distortion. S4. Drive the infrared array acquisition module to perform a wandering tomography along a dynamic aberration-free scanning trajectory to capture a deformation-resistant infrared transmission image sequence with absolute depth labels. S5. Process the deformation-resistant infrared transmission image sequence to aggregate pixel blocks with overlapping spatial coordinates and extract the Z-domain phase reversal features that evolve layer by layer with the detection depth. S6. Analyze the nonlinear photoelectric abrupt change process within the phase reversal feature of the Z domain. In response to the capture of the light intensity crossover reversal point, classify and grade the defects in the corresponding region as first intrinsic defects with physical refraction effects. S7. In response to the monotonically linear decay without abrupt changes in the phase reversal characteristics of the Z-domain, the defects in the corresponding region are classified and graded as second intriguing defects with pure physical occlusion, and the closed-loop data of physical grading of wafer defects is output.

2. The wafer intrinsic defect classification method based on infrared penetration characteristics according to claim 1, characterized in that, Driving the infrared scanning module to perform a global scan of the wafer includes the following steps: The initial two-dimensional infrared grayscale image penetrating the wafer is captured using a large field-of-view time-delay integral infrared scanning module. Thresholding segmentation is performed on the initial infrared two-dimensional grayscale image to extract continuous dark spot pixel regions; The geometric center of the continuous dark spot pixel region is mapped to the physical position in the wafer absolute coordinate system through the coordinate calibration matrix, and the suspected defect location signal containing horizontal and vertical two-dimensional spatial coordinates is output.

3. The wafer intrinsic defect classification method based on infrared penetration characteristics according to claim 1, characterized in that, Detecting wafer warpage bias at the physical location corresponding to suspected defect location signals includes the following steps: The suspected defect location signal is used as a hardwired trigger switch to drive the surface manifold coherent encoder to release coherent laser to the horizontal and vertical two-dimensional spatial coordinates. Receive the interference echo reflected back from the physical surface of the wafer by coaxial coherent laser, and analyze the radius of curvature of the current tiny region in the horizontal and vertical two-dimensional spatial coordinates and the optical astigmatism bias vector as the wafer warping bias; The radius of curvature and the optical astigmatism bias vector are transformed into a continuous topological waveform that characterizes the local thickness distortion of the wafer, and the continuous topological waveform is amplified by an operational amplifier and output as an analog modulated voltage signal.

4. The wafer intrinsic defect classification method based on infrared penetration characteristics according to claim 1, characterized in that, Dynamically adjusting the exposure time of the infrared detection component and the downward stepping distance of the longitudinal stepper actuator includes the following steps: The analog modulation voltage signal is loaded into the exposure timing generator of the infrared phase gate array driver, which is an infrared detection component, and the photoelectric integration exposure time of a single local infrared pixel in the infrared phase gate array driver is dynamically adjusted by using the voltage amplitude of the analog modulation voltage signal. The inverted component of the analog modulated voltage signal is used to reshape the longitudinal micron-level downward stepping distance of the piezoelectric stepper, which is used as a longitudinal stepper, in real time when it cuts into the wafer thickness. By substituting the photoelectric integration exposure time and the longitudinal micron-level downward step spacing into the displacement compensation, a dynamic aberration-free scanning trajectory that eliminates wavefront distortion is generated.

5. The wafer intrinsic defect classification method based on infrared penetration characteristics according to claim 1, characterized in that, Driving the infrared array acquisition module to perform a walking tomographic scan along a dynamic aberration-free scanning trajectory includes the following steps: Using the high-light reflection point established by the pre-set coaxial coherent laser on the physical surface of the wafer as the absolute zero longitudinal starting point, the light-emitting and pixel photosensitive elements of the short-wave infrared array acquisition module, which serves as the infrared array acquisition module, are activated. The constrained shortwave infrared array acquisition module performs a non-equidistant downward cutting action on the vertical focusing plane along the dynamic aberration-free scanning trajectory, according to the timing sequence of the preset longitudinal micron-level downward stepping interval and the preset photoelectric integration exposure time. The transmitted photon stream that completely penetrates the corresponding regions of the horizontal and vertical two-dimensional spatial coordinates is continuously captured, and the aggregated output is a sequence of deformation-resistant infrared transmission images with sequentially increasing depth labels.

6. The wafer intrinsic defect classification method based on infrared penetration characteristics according to claim 1, characterized in that, Extracting the Z-domain phase reversal features that evolve layer by layer with the detection depth includes the following steps: By stacking the anti-deformation infrared transmission image sequence according to the sequentially increasing depth labels, a virtual three-dimensional photon transmission energy field characterizing the transmission attenuation probability inside the wafer is constructed. Lock the pixel block formed by the central pixel that vertically penetrates the horizontal and vertical two-dimensional spatial coordinates and the surrounding ring of adjacent pixels, and track the light intensity gradient trend of the pixel block from shallow to deep in the virtual three-dimensional photon transmission energy field. The photon spatial phase transition curves generated by the transmitted light intensity of the central pixel and the diffracted light intensity of the surrounding ring of pixels when crossing medium planes of different depths are extracted to generate Z-domain phase reversal features.

7. The wafer intrinsic defect classification method based on infrared penetration characteristics according to claim 6, characterized in that, In response to the capture of an intensity crossover point, the defect is classified as a first intrinsic defect with a physical refraction effect, including the following steps: Along the depth coordinates of the Z-domain phase reversal feature, we identify divergent phase nodes where the central pixel exhibits a local minimum and a ring of adjacent pixels derives a bright diffraction ring and exhibits a local maximum. By comparing the Z-domain phase reversal features under deeper dielectric layers, the refocusing phase nodes that detect the light intensity playback of the central pixel and show local maxima, and the energy extraction, collapse and darkening of the bright diffraction ring and show local minima are detected. When diverging phase nodes and reconverging phase nodes appear sequentially in a continuous longitudinal depth to form an intensity crossover point, the current region is classified as a first intrinsic defect. First intrinsic defects include bubbles and pores hidden inside the wafer.

8. The wafer intrinsic defect classification method based on infrared penetration characteristics according to claim 6, characterized in that, If the phase reversal feature in the Z-domain exhibits a monotonically linear decay without abrupt changes, then the classification and grading as a second eigenvalue defect of pure physical occlusion includes the following steps: Determine whether the representative light intensity of the center pixel in the Z-domain phase reversal feature conforms to the monotonic photon occlusion absorption property as the absolute depth label increases; If the entire jump curve fails to capture the preset light intensity crossover point and confirms that it only exhibits monotonous linear decay, it is determined that the refractive index rearrangement has not occurred at the interface of the probe medium, and it is forcibly classified and graded as a second intriguing defect. The second intriguing defect includes surface particles or surface protrusions. Summarize the information of the first intrinsic defect and the second non-intrinsic defect, and output the closed-loop data of wafer defect physical classification.

9. The wafer intrinsic defect classification method based on infrared penetration characteristics according to claim 8, characterized in that, Summarize the information of the first intrinsic defect and the second intriguing defect, and output the closed-loop data for wafer defect physical classification, including the following steps: Obtain the three-dimensional depth coordinates and equivalent astigmatic volume of the first intrinsic defect; Extract the two-dimensional coordinates of the surface of the second intrinsic defect; The three-dimensional depth coordinates, equivalent astigmatic volume, and surface two-dimensional coordinates are structured and encapsulated to generate and output closed-loop data for wafer defect physical classification based on pure physical underlying optical features.

10. A wafer intrinsic defect classification system based on infrared penetration characteristics, characterized in that, Includes the following modules: The defect location module is used to drive the infrared scanning module to perform a global scan of the wafer, extract the transmission abnormal area, and generate a suspected defect location signal containing absolute physical coordinates. The warpage detection module is used to detect the wafer warpage bias at the physical location corresponding to the suspected defect location signal, and to generate an analog modulated voltage signal for underlying hardware interlocking based on the wafer warpage bias. The scanning trajectory generation module is used to dynamically adjust the exposure time of the infrared detection component and the downward stepping distance of the longitudinal stepper actuator according to the analog modulation voltage signal, so as to construct a dynamic aberration-free scanning trajectory that eliminates wavefront distortion. The tomographic image acquisition module is used to drive the infrared array acquisition module to perform a wandering tomographic scan along a dynamic aberration-free scanning trajectory and capture a deformation-resistant infrared transmission image sequence with absolute depth labels. The optical feature extraction module is used to process the deformation-resistant infrared transmission image sequence to aggregate pixel blocks with overlapping spatial coordinates and extract the Z-domain phase reversal features that evolve layer by layer with the detection depth. The intrinsic defect classification module is used to analyze the nonlinear photoelectric abrupt change process within the phase reversal feature of the Z domain. In response to the capture of the light intensity crossover reversal point, the defects in the corresponding region are classified and graded as first intrinsic defects with physical refraction effects. The defect classification and output module is used to classify and classify the defects in the corresponding region as second intriguing defects with pure physical occlusion in response to the monotonous linear decay without abrupt changes in the phase reversal characteristics of the Z-domain, and output closed-loop data of wafer defect physical classification.

Images