Defect Detection Method for Batch Chips from Multiple Angles, Control Device and AOI Device

Abstract

The present application relates to optical measurement technology, and discloses a method for defect detection of a batch of chips from multiple angles, a control device, and an AOI device, including: placing a chip array after wire bonding into a detection area of the AOI device; controlling a camera array to capture panoramic images of the chip array from multiple orientations; performing image segmentation on the panoramic images from multiple orientations to obtain multi-angle images corresponding to each chip; performing defect detection on chips without unqualified images; intelligently planning an image re-capture path according to the chip positions and orientations to which the unqualified images belong; controlling a moving camera to re-capture the unqualified images along the image re-capture path; performing defect detection on the chips after image re-capture; and generating defect detection results of the chips. The present application also discloses a computer-readable storage medium. The present application aims to improve the efficiency of defect detection of a batch of chips while ensuring the quality of defect detection of the batch of chips.

Description

Technical Field

[0001] The present application relates to the technical field of optical measurement, and particularly to a method for detecting defects of a batch of chips at multiple angles, a control device, an AOI device, and a computer-readable storage medium. Background Art

[0002] In modern semiconductor manufacturing, the quality inspection of chips is a key link to ensure the reliability and performance of products. With the continuous improvement of chip integration and complexity, the requirements for the efficiency and accuracy of chip defect detection are also getting higher and higher. Automated Optical Inspection (AOI), as an efficient detection means, is widely used in the chip manufacturing process to ensure chip quality and production efficiency.

[0003] Currently, the method of using an AOI device to perform automatic defect analysis on chips is generally to take corresponding chip images one by one for each chip and each orientation. Among them, each chip and each orientation need to be independently photographed. This process requires the camera to continuously turn and pause between different orientations to ensure comprehensive and detailed image acquisition for each orientation of each chip.

[0004] This method of taking pictures one by one for each chip and each orientation, although ensuring the comprehensiveness and accuracy of the images to a certain extent, also brings some problems: First, since each chip and each orientation need to be independently photographed, the overall detection time is too long. Especially in the case of detecting a large number of chips, this method of taking pictures one by one greatly extends the detection cycle and affects production efficiency. Second, during the shooting process, the frequent turning and pausing of the camera may lead to unstable movement, which in turn affects the clarity and consistency of the images. And the frequent turning and pausing also increase the mechanical wear of the camera holder, reducing the stability and service life of the device.

[0005] The above content is only used to assist in understanding the technical solution of the present application, and does not represent an admission that the above content is prior art. Summary of the Invention

[0006] The main purpose of the present application is to provide a method for detecting defects of a batch of chips at multiple angles, a control device, an AOI device, and a computer-readable storage medium, aiming to improve the efficiency of defect detection for a batch of chips while ensuring the quality of defect detection for a batch of chips.

[0007] To achieve the above object, the present application provides a method for detecting defects of a batch of chips at multiple angles, including the following steps:

[0008] Place the chip array after wire bonding into the detection area of the AOI device; wherein, a camera array and a moving camera are arranged in the detection area;

[0009] Control the camera array to operate at multiple shooting angles to capture panoramic images of multiple orientations of the chip array; among them, during the shooting process of the camera array, each camera focuses independently;

[0010] Perform image segmentation on the panoramic images of multiple orientations to obtain multi-angle images corresponding to each chip;

[0011] Perform image quality inspection on the multi-angle images and screen out unqualified images;

[0012] For the chips without unqualified images, splice the multi-angle images to obtain the overall chip image, and perform image analysis on the overall chip image based on the AOI technology to detect defects in the chips; and, according to the chip positions and orientations to which the unqualified images belong, intelligently plan the image reshooting path; among them, calculate the path in the same orientation according to the shortest distance of the connection line of the chip positions to which all unqualified images in the same orientation belong; connect the paths in different orientations according to the shortest switching distance between the paths in different orientations to form the image reshooting path;

[0013] Control the moving camera to reshoot the unqualified images along the image reshooting path; among them, the interval path of the moving camera between two adjacent reshooting points successively includes an acceleration section, a deceleration section, a constant-speed preparation section, and a constant-speed capture section, and the length of the constant-speed preparation section is positively correlated with the deceleration of the deceleration section; the end point of the constant-speed capture section of the previous interval path is the start point of the acceleration section of the next interval path;

[0014] Replace the unqualified images with the reshot images;

[0015] For the chips after image reshooting, splice the multi-angle images to obtain the overall chip image, and perform image analysis on the overall chip image based on the AOI technology to detect defects in the chips;

[0016] Mark the chips without defects as qualified products, mark the chips with defects as unqualified products, mark the chips that fail to complete defect detection as products to be confirmed, and generate the defect detection results of the chips.

[0017] To achieve the above object, the present application also provides a control device, the control device includes: a memory, a processor, and a computer program stored on the memory and executable on the processor, and when the computer program is executed by the processor, the steps of the above-mentioned method for defect detection of multiple angles of a batch of chips are implemented.

[0018] To achieve the above object, the present application also provides an AOI device, the AOI device includes a camera array, a moving camera, and a control device; among them, the camera array and the moving camera are both communicatively connected to the control device and are controlled by the control device;

[0019] The control device is the control device as described above.

[0020] To achieve the above object, the present application further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the defect detection method for multi-angle of batch chips as described above are implemented.

[0021] The defect detection method for multi-angle of batch chips, control device, AOI device and computer-readable storage medium provided by the present application place the chip array after wire bonding into the detection area of the AOI device, and simultaneously capture multiple angles of multiple chips through the camera array, significantly reducing the independent shooting time of each chip and improving the overall detection efficiency. While detecting chip defects in qualified images, a motion camera can also be used to intelligently retake some images with insufficient quality captured by the camera array, ensuring the detection quality and efficiency. And during the retaking process, by intelligently planning the path and optimizing the motion control of the camera, unnecessary movements and pauses are reduced, not only taking into account the retaking efficiency and the quality of the retaken images, but also reducing mechanical wear and extending the service life of the device. In this way, it helps to improve the overall defect detection efficiency of batch chips, thereby improving the production efficiency of batch chips and ensuring the chip quality. Description of the Drawings

[0022] Figure 1 It is a schematic diagram of the steps of the defect detection method for multi-angle of batch chips in an embodiment of the present application;

[0023] Figure 2 It is a schematic diagram of the speeds of each section of the interval path in an embodiment of the present application;

[0024] Figure 3 It is a schematic diagram of the internal architecture of the control device in an embodiment of the present application.

[0025] The realization, functional features and advantages of the object of the present application will be further described in conjunction with the embodiments with reference to the drawings. Detailed Embodiments

[0026] The embodiments of the present application are described in detail below. The examples of the embodiments are shown in the drawings, where the same or similar reference numerals denote the same or similar elements or elements with the same or similar functions from beginning to end. The embodiments described below with reference to the drawings are exemplary and are intended to explain the present application and should not be construed as a limitation to the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present application.

[0027] In addition, if the description in this application involves "first", "second", etc., it is only for descriptive purposes (such as for distinguishing the same or similar features), and cannot be understood as indicating or implying its relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include at least one such feature. In addition, the technical solutions between various embodiments may be combined with each other, but it must be based on the ability of those of ordinary skill in the art to implement. When the combination of technical solutions results in contradictions or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection required by this application.

[0028] Referring to Figure 1 , in one embodiment, a method for defect detection of a batch of chips at multiple angles includes:

[0029] Step S10: Place the chip array after wire bonding into the detection area of the AOI device; wherein, a camera array and a moving camera are provided in the detection area;

[0030] Step S20: Control the camera array to operate at multiple shooting angles to capture panoramic images of multiple orientations of the chip array; wherein, during the shooting process of the camera array, each camera focuses independently;

[0031] Step S30: Perform image segmentation on the panoramic images of multiple orientations to obtain multi-angle images corresponding to each chip;

[0032] Step S40: Check the image quality of the multi-angle images and screen out unqualified images;

[0033] Step S50: For the chips without unqualified images, splice the multi-angle images to obtain an overall image of the chip, and perform image analysis on the overall image of the chip based on the AOI technology to detect defects of the chip; and, according to the chip positions and orientations to which the unqualified images belong, intelligently plan an image re-shooting path; wherein, calculate the path in the same orientation according to the shortest distance of the connection line of the chip positions to which all unqualified images in the same orientation belong; connect the paths in different orientations according to the shortest switching distance between the paths in different orientations to form an image re-shooting path;

[0034] Step S60: Control the moving camera to re-shoot the unqualified images along the image re-shooting path; wherein, the interval path of the moving camera between two consecutive re-shooting points sequentially includes an acceleration section, a deceleration section, a constant-speed preparation section, and a constant-speed capture section, and the length of the constant-speed preparation section is positively correlated with the deceleration of the deceleration section; the end point of the constant-speed capture section of the previous interval path is the start point of the acceleration section of the next interval path;

[0035] Step S70: Replace the unqualified images with the re-shot images;

[0036] Step S80: For the chip after image retaking, perform stitching of multi-angle images to obtain an overall chip image, and perform image analysis on the overall chip image based on AOI technology to detect defects of the chip;

[0037] Step S90: Mark the defect-free chips as qualified products, mark the defective chips as unqualified products, mark the chips for which defect detection cannot be completed as products to be confirmed, and generate the defect detection results of the chips.

[0038] In this embodiment, the execution terminal of the embodiment may be the control device of the AOI device.

[0039] As described in step S10, before entering this step, the chip has been assembled through the gold wire bonding process. Gold wire bonding generally connects thin metal wires to a substrate or carrier to achieve electrical connection. This process is called bonding wire connection, and usually uses thermocompression bonding or ultrasonic bonding technology to weld the metal wire between the pads on the chip and the pins on the substrate.

[0040] The chips after gold wire bonding are placed on a dedicated tray or carrier in an array form (usually in the format of rows and columns). The position of each chip in the array needs to be accurately identified for positioning and detection in subsequent steps.

[0041] Optionally, the operator or the automated system of the AOI device (such as the gold wire bonding AOI sorter) smoothly moves the carrier with the chip array into the detection area of the AOI device. Then the carrier can be transported to the shooting position through a conveyor belt, a robotic arm or other automated devices. In the detection area of the AOI device, a camera array and a moving camera are provided.

[0042] Optionally, the camera array can be composed of multiple high-resolution cameras, and the specific number depends on the detection requirements and the chip size (it is not necessary that the number of cameras in the camera array is the same as the number of chips, and the shooting range of one camera can cover multiple chips). For example, it can be 4, 8 or more cameras to cover different angles of the chip.

[0043] Optionally, the moving camera is installed on a movable motion control platform, which is composed of a precision servo motor and a high-precision guide rail system and can move freely in the detection area. The moving camera can be a high-resolution single camera or a multi-camera system, and the specific configuration depends on the detection requirements. The moving camera has a sufficient moving range to cover all areas of the chip array.

[0044] Of course, the camera array also has a certain degree of movement ability (the range of movement and flexibility do not necessarily need to reach the level of action cameras), so that multiple shooting angles can be operated during shooting. Among them, the movement of the camera can be controlled by an electric motor or a stepper motor to achieve precise position control; some cameras can also have the function of height adjustment, allowing them to move up and down within a certain range to adapt to chips or detection requirements at different heights.

[0045] As described in step S20, the initial position of the camera array can be located directly above the camera array, and a top-down panoramic view of the camera array is collected first.

[0046] Optionally, ensure that each camera in the camera array is in the initial position, usually the top-down position. Each camera needs to preset an initial focus point to ensure quick focusing during shooting. Each camera performs the focusing operation independently to ensure that the captured images are clear. Then, the images captured by each camera are stitched together to obtain a panoramic image of the chip array from the top-down perspective.

[0047] Then, according to the required shooting angles, plan the movement path of the camera array to switch to the next shooting angle, such as the bird's-eye view, side view angle, etc. Among them, the camera array can be controlled by an electric motor or a stepper motor to adjust the position and angle of the camera array to ensure that each camera can accurately reach the predetermined shooting angle. At each new angle, each camera performs the focusing operation independently again to ensure that the images are clear at the new shooting angle.

[0048] Taking the bird's-eye view shooting as an example, it can be divided into bird's-eye view angles of different orientations in front of, behind, left, and right of the chip. After adjusting the camera array to the bird's-eye view shooting angle of the corresponding orientation, each camera performs the focusing operation independently to ensure that the captured images are clear. Then, the chip array images captured by each camera at the bird's-eye view angle of the corresponding orientation are stitched together to obtain a panoramic image at the bird's-eye view angle of the corresponding orientation. These images will be used for subsequent image processing and analysis.

[0049] For example, each camera captures local images of the chip array at the bird's-eye view angle in front of the chip array, and then stitches the captured images together to obtain a panoramic image at the front-view bird's-eye view angle; then, by controlling the camera array with an electric motor or a stepper motor, after adjusting to the bird's-eye view angle on the left side of the chip array, each camera captures local images of the chip array at the bird's-eye view angle on the left side of the chip array, and then stitches the captured images together to obtain a panoramic image at the left-view bird's-eye view angle.

[0050] Optionally, if images at the side view angle (which can be the left, right, front, or back side) are needed, the position and angle of the camera array can be further adjusted. At each new side view angle, each camera performs the focusing operation independently again to ensure that the images are clear at the new shooting angle, and a panoramic image of the chip array is captured at the side view angle.

[0051] In this way, a multi-directional panoramic image of the chip array can be effectively obtained, providing high-quality basic data support for subsequent image processing and defect detection.

[0052] As described in step S30, after the camera array captures panoramic images of the chip array from multiple angles (the panoramic image is a large image containing multiple chips, and each chip occupies a specific position in the image), for panoramic images at different angles, taking each chip as a unit, image segmentation can be performed on the panoramic images at different angles.

[0053] Optionally, before performing image segmentation, some preprocessing operations can be performed on the panoramic image to improve the accuracy and efficiency of segmentation. The preprocessing steps can include: removing noise in the image to reduce interference during the segmentation process; enhancing the contrast of the image to make the edges of the chips clearer and facilitate subsequent segmentation operations.

[0054] The purpose of image segmentation is to separate each chip in the panoramic image to obtain an independent image of each chip. Optional image segmentation algorithms include: using the Canny edge detection algorithm or other edge detection methods to identify the edges of the chips; or, by setting appropriate thresholds, separating the chip regions in the image from the background regions.

[0055] In the panoramic image, the position of each chip is known (it can have been identified in step S10). Based on this position information, the segmentation operation can be performed more precisely. The coordinate range of each chip in the panoramic image can be determined according to the pre-identified chip positions, and then using an image processing tool (such as OpenCV), according to the coordinate range of the chip, the image of each chip can be extracted from the panoramic image.

[0056] Since the panoramic image is captured from multiple angles, each chip has corresponding images at different angles. During the segmentation process, it is necessary to ensure that the images of each chip at each angle are correctly extracted. Therefore, the panoramic images can be classified according to the shooting angles (such as top view, bird's-eye view, side view, etc.), and then the panoramic images at each angle are segmented to extract the images of each chip at that angle.

[0057] As described in step S40, although multiple angles of multiple chips can be captured at once by the camera array, which can significantly reduce the independent shooting time for each chip and improve the overall shooting efficiency. However, due to the arrangement of the camera array, for chips at certain positions or angles, it is inevitable that the captured image quality is poor due to insufficient field of view (for example, chips at certain edge positions or specific angles may not be fully covered or blocked, or the captured images at certain positions may be affected by optical distortion, resulting in edge distortion or blurring of the image). Therefore, it is necessary to screen out the images with poor quality from them.

[0058] That is, the purpose of image quality inspection is to screen out those images that do not meet the quality standards due to poor shooting conditions, equipment failures or other reasons. These unqualified images may affect the subsequent defect detection results, so they need to be screened before further processing.

[0059] When performing image quality inspection, multiple evaluation metrics can be used to judge the quality of the image. Commonly used evaluation metrics include:

[0060] (1) Sharpness: The sharpness of an image is one of the important metrics for measuring image quality. A sharp image can provide more detailed information, facilitating subsequent defect detection;

[0061] (2) Contrast: The contrast of an image reflects the brightness difference between different regions in the image. An image with high contrast is easier to distinguish different regions and features;

[0062] (3) Noise level: Noise in the image will interfere with the detailed information in the image and reduce the image quality;

[0063] (4) Exposure: The exposure of an image reflects the brightness level of the image. Appropriate exposure can ensure that the detailed information in the image is not masked by overexposure or underexposure;

[0064] (5) Distortion: Image distortion may be caused by improper camera lenses or shooting angles. Distortion will affect the geometric shape of the image, thus affecting subsequent analysis and detection.

[0065] Among them, for sharpness evaluation, an edge detection algorithm can be used to calculate the sharpness of the image. A sharp image usually has a higher edge intensity.

[0066] Among them, for contrast evaluation, the contrast of the image can be calculated by computing the gray-level histogram of the image. An image with high contrast will have obvious peaks and valleys in the histogram.

[0067] Among them, for noise evaluation, the signal-to-noise ratio of the image can be calculated to evaluate the noise level.

[0068] Among them, the exposure evaluation can be carried out by calculating the average brightness value of the image to determine whether the image is overexposed or underexposed.

[0069] Among them, the distortion evaluation can be carried out by calculating the geometric shape change of the image to determine whether the image is distorted. Optional methods include Harris corner detection and image registration.

[0070] According to the results of the image quality evaluation, unqualified images are screened out and marked.

[0071] Optionally, according to the preset image quality threshold, those images that do not meet the requirements are screened out. For example, images with clarity lower than a certain threshold, insufficient contrast, high noise level, improper exposure or distortion.

[0072] The unqualified images screened out are marked, and the chip positions and shooting angles to which they belong are recorded for subsequent image reshooting and processing.

[0073] As described in step S50, when the quality of each angular image corresponding to the chip is qualified, the chip is marked as having no unqualified images; when any one of the angular images of the chip is unqualified, the chip is marked as having unqualified images.

[0074] Optionally, for chips with no unqualified images, feature detection algorithms (such as SIFT, SURF or ORB) are used to extract feature points from each angular image and perform matching to find the corresponding relationships between each angular image.

[0075] Based on the feature matching results, image stitching algorithms (such as stitching, fusion, etc.) are used to merge the multi-angular images into a complete overall chip image. In this process, geometric transformation and color correction can be carried out when necessary to ensure the naturalness and accuracy of the stitching effect.

[0076] The overall chip image obtained by stitching is analyzed, and AOI image processing technology is used to identify potential defects (vision recognition based on machine learning can be added when necessary). This can include but is not limited to detecting whether there are multiple or misaligned chip leads, and whether the chip body is complete or missing, etc.

[0077] While detecting defects in chips with no unqualified images, for unqualified images, the position information and orientation information of all unqualified images are collected (each unqualified image has a clear chip position and shooting angle (orientation)).

[0078] For each specific orientation, calculate the shortest path from the current position to all the positions of non-conforming chips within that orientation. This can be solved using a variant algorithm of the Traveling Salesman Problem (TSP) to ensure that the camera can visit each chip position that needs to be reshot in the shortest possible time. At the same time, establish an adjacency matrix in which the distances between each pair of chip positions are calculated and stored, enabling the moving camera to visit each chip position in sequence at the shortest distance.

[0079] In the planned reshooting path, an orientation corresponding to a chip position is a reshooting point.

[0080] After completing the path planning for each orientation, it is necessary to consider how to switch from the current orientation to the next orientation that needs to be reshot. This involves calculating the shortest switching path from the last chip position of the current orientation to the first chip position of the next orientation. Among them, the shortest path from the end point of one orientation to the starting point of another orientation can be calculated to ensure that the moving camera can switch orientations quickly and efficiently.

[0081] Optionally, connect the shortest path within the same orientation with the shortest switching paths between different orientations to form a complete path. This path ensures that the moving camera can visit all the positions of non-conforming images that need to be reshot, regardless of which orientation they are in.

[0082] During the process of connecting the paths of each orientation, the total path can be further optimized to ensure that the moving camera can complete all reshooting tasks in the shortest possible time with the minimum number of turns.

[0083] As described in step S60, before the reshooting starts, confirm that the moving camera is in a ready state and all necessary parameters (such as exposure time, focal length, etc.) have been preset in advance to adapt to the reshooting environment. Load the previously planned image reshooting path into the control system of the moving camera.

[0084] In the image reshooting path, the path between two adjacent reshooting points before and after is defined as an "interval path". Refer to Figure 2 , each interval path sequentially includes an acceleration section, a deceleration section, a constant-speed preparation section, and a constant-speed capture section.

[0085] Optionally, in the acceleration section, the moving camera starts from a stationary or lower speed (when shooting the first image in each orientation of the reshooting, the initial speed is 0; when shooting the second image onwards, the initial speed is the speed at the end of the constant-speed capture section), and gradually accelerates to the preset target speed v 1 . This acceleration process is smooth to reduce mechanical shock and image distortion. Among them, the value of the acceleration is determined by the mechanical characteristics of the motion control platform that controls the movement of the moving camera and can take the maximum value within the range allowed by the mechanical characteristics.

[0086] The deceleration stage is the stage when the action camera starts to decelerate as it approaches the next chip position, ensuring a stable and controllable speed when reaching the retake point. The magnitude of the deceleration is adjusted as needed to ensure the camera decelerates smoothly within a predetermined time. Among them, although the value of the deceleration can also be determined by the mechanical characteristics of the motion control platform that controls the movement of the action camera and can take the maximum value within the range allowed by the mechanical characteristics, since the acceleration stage does not need to consider motion jitter too much, while the deceleration stage needs to avoid severe jitter of the motion control platform or the camera, it is therefore preferred to set the value of the deceleration to be less than the acceleration of the acceleration stage.

[0087] The constant-speed preparation stage is used to relieve or eliminate the camera jitter that may be brought about by the deceleration stage. When the action camera decelerates in the deceleration stage, due to the action of mechanical inertia, the camera may generate a certain amount of jitter or vibration. This kind of jitter may affect the clarity and quality of the image; during the deceleration process, the speed fluctuation may also cause the camera to be unable to stabilize immediately when approaching the retake point, thus affecting the capture effect.

[0088] The constant-speed preparation stage provides a stable environment for the camera by maintaining a constant speed, enabling it to eliminate the jitter and vibration brought about by the deceleration stage before capture. Among them, the length of the constant-speed preparation stage is positively correlated with the deceleration of the deceleration stage, which means that if the deceleration of the deceleration stage is large, the length of the constant-speed preparation stage will also increase accordingly. This design provides the camera with more time to stabilize the speed, thereby reducing jitter.

[0089] Optionally, the length of the constant-speed preparation stage is determined according to the deceleration of the deceleration stage. The following formula can be used for calculation:

[0090] L = ;

[0091] where, v 1 is the target speed before deceleration, v 2 is the constant speed after deceleration, a is the deceleration, and k is an empirical coefficient. The value of the empirical coefficient k is usually determined according to the actual application scenario and test data, such as set between 2 and 10.

[0092] During the constant-speed preparation stage, the speed of the action camera is precisely controlled at a constant value. The control system will calculate the appropriate length of the constant-speed preparation stage according to the deceleration of the deceleration stage and the mechanical characteristics of the camera to ensure that the camera reaches a stable state before capture. Even though the distances between the retake points may be different, it is still necessary to ensure that the length of the constant-speed capture section of each interval path can meet the requirement that the action camera has enough time to capture the chip image. Therefore, the length of the constant-speed capture section needs to be determined jointly according to the actual constant speed of the constant-speed capture section and the shooting parameters of the action camera.

[0093] Optionally, during the constant-speed preparation stage, the system can also finely adjust the position of the camera to ensure that the position of the camera is accurate during the capture.

[0094] Through the design of the constant-speed preparation stage, the camera can reach a stable state before the capture, thus significantly improving the clarity and quality of the image. The design of the constant-speed preparation stage not only improves the image quality but also ensures that the camera can efficiently move and capture between multiple reshooting points, improving the overall operation efficiency.

[0095] Optionally, in actual operation, the length and speed control strategy of the constant-speed preparation stage can also be dynamically adjusted according to the real-time monitored camera state and environmental conditions to meet different reshooting requirements.

[0096] Optionally, the acceleration sensor installed on the moving camera is used to monitor the acceleration and vibration of the camera in real time. According to the real-time monitored acceleration data, the acceleration and deceleration of the camera are dynamically adjusted. For example, by optimizing the acceleration curve (such as an S-shaped curve), the acceleration change becomes smoother, reducing mechanical shock and shaking; if it is detected that the camera shakes significantly during the deceleration stage, the deceleration can be appropriately reduced or the length of the constant-speed preparation stage can be increased.

[0097] The constant-speed capture stage is to capture the image when the speed of the moving camera is stable and the position is accurate. This stage needs to ensure that the camera quickly and accurately captures the image of the chip.

[0098] Once the capture of the current chip is completed, the end point of the constant-speed capture stage immediately becomes the start point of the speed-up section of the next interval path. This design ensures that the camera can continuously and efficiently move and reshoot between multiple reshooting points.

[0099] The above process is repeatedly executed on each interval path until all reshooting points are visited and the images are successfully reshot. During the entire reshooting process, the system will monitor the state of the camera, the accuracy of the path, and the quality of the reshot images in real time. If any problems or deviations are found, the system will make timely adjustments to ensure the smooth completion of the reshooting task.

[0100] It should be noted that the design of the speed-up section is to make the moving camera approach the next reshooting point as fast as possible; the design of the deceleration section is to make the moving camera have a stable and controllable speed when reaching the reshooting point (that is, to ensure that the camera has an appropriate speed during the capture to avoid blurry images caused by too fast speed); the design of the constant-speed preparation stage is to make the moving camera enter an absolutely stable and uniform translation state before the official capture (that is, to relieve or eliminate the camera shake that may be brought about by the deceleration section, and by maintaining a constant speed, the camera reaches a stable state before the capture); the design of the constant-speed capture stage is to capture a clear and high-quality chip image when the moving camera stably and uniformly translates through the reshooting point.

[0101] In this way, there is no need to pause the action camera during the capture process, which can not only improve the reshooting efficiency but also ensure high-quality chip images are captured.

[0102] Since the appearance of unqualified images is mostly caused by the field of view of the camera array, and the occurrence of this reason is affected by factors such as the position and angle of the camera array relative to the chip array, environmental light, and focusing parameters, resulting in uneven distribution of unqualified images, which also makes the distribution of the positions of reshooting points uneven. By designing that the interval path between the front and rear reshooting points successively includes an acceleration section, a deceleration section, a constant-speed preparation section, and a constant-speed capture section, the action camera can move and capture efficiently between the unevenly distributed reshooting points while ensuring the quality of the images.

[0103] As described in step S70, the action camera reshoots the specified chip position and orientation according to the planned path to generate new image data.

[0104] According to the metadata of the reshot image or through image processing technology, determine the chip position and shooting orientation corresponding to the reshot image. Register the reshot image with the original image set to ensure that the new image can accurately replace the original unqualified image.

[0105] Optionally, in the image database, find the record of the unqualified image corresponding to the chip position and orientation, and replace it with the reshot image. Update the metadata information of the image, including shooting time, camera parameters, position information, etc., to reflect this replacement operation.

[0106] As described in step S80, for the chip after image reshooting, use the feature detection algorithm to extract feature points from the images at each angle and perform matching to find the corresponding relationship between the images at each angle.

[0107] Based on the feature matching results, use the image stitching algorithm to merge the multi-angle images into a complete overall chip image. During this process, geometric transformation and color correction can be performed if necessary to ensure the naturalness and accuracy of the stitching effect.

[0108] Analyze the overall chip image obtained by stitching, and use AOI image processing technology to identify potential defects (vision recognition based on machine learning can be added if necessary). This can include but is not limited to detecting whether there are multiple lines or misaligned lines in the chip leads, as well as whether the chip body is complete or missing.

[0109] As described in step S90, for those chips that have passed the AOI technology detection and no defects are found, mark them as "qualified products". Qualified products mean that the chips have passed all quality inspection standards and can enter the next production link or be packaged and shipped out.

[0110] For chips detected with defects, they are marked as "non-conforming products" according to information such as the type and severity of the defects. These chips need to undergo additional processing, such as repair, scrapping, or re-inspection.

[0111] For those chips that fail to complete defect detection due to certain reasons (such as insufficient light, blurred images, not meeting the recognizable standards of the defect detection algorithm, etc.), they are marked as "products to be confirmed". These chips need to be re-inspected or manually confirmed to determine their final quality status.

[0112] Optionally, update the detection results of each chip to the database, recording information such as the chip number, detection time, detection results (qualified, unqualified, to be confirmed), and defect type. Ensure the traceability and integrity of all data.

[0113] Optionally, associate the detection results with the corresponding chip images to form a complete detection report. The detection report can include the original image, the stitched panoramic image, the defect marked image, etc., providing detailed quality information.

[0114] Optionally, generate a detailed quality report. The report format can include the basic information of the chips (such as batch number, production date, etc.), statistical information on the detection results (quantity of qualified products, quantity of unqualified products, quantity of products to be confirmed), defect type and distribution charts, defect severity analysis, etc.

[0115] Optionally, use a combination of charts, images, and text to generate an intuitive and easy-to-understand quality report. The report should include a defect distribution chart, a defect type proportion chart, a defect severity analysis chart, etc., to help managers quickly understand the quality status of the chips.

[0116] Optionally, feedback the generated quality report to the production line or the quality control department to guide subsequent production adjustments and quality improvements. For unqualified chips, it is recommended to perform repair or scrapping; for chips to be confirmed, it is recommended to re-inspect or manually confirm.

[0117] Optionally, conduct data analysis on the defect detection results to identify the distribution pattern and possible causes of the defects. Through data analysis, potential problems in the production process can be discovered, providing a basis for process improvement and equipment calibration.

[0118] Optionally, archive all the generated quality reports and related image files for future reference and auditing. Ensure the traceability and integrity of all data to meet the requirements of the quality management system.

[0119] In one embodiment, the chip array after gold wire bonding is placed in the detection area of the AOI device, and multiple angles of multiple chips are simultaneously captured by the camera array, significantly reducing the independent shooting time of each chip and improving the overall detection efficiency. While detecting chip defects in the qualified images, a motion camera can be used to intelligently retake some of the images with insufficient quality captured by the camera array, ensuring the detection quality and efficiency. And during the retaking process, by intelligently planning the path and optimizing the motion control of the camera, unnecessary movements and pauses are reduced, not only taking into account the retaking efficiency and the quality of the retaken images, but also reducing mechanical wear and extending the service life of the device. In this way, it helps to improve the overall defect detection efficiency of batch chips, thereby improving the production efficiency of batch chips and ensuring the chip quality.

[0120] In one embodiment, based on the above embodiment, the step of performing image analysis on the overall chip image based on the AOI technology to detect defects in the chip includes:

[0121] Locate the gold wire image in the overall chip image based on the edge detection algorithm and analyze whether the wire connection in the gold wire image is correct;

[0122] Based on the image segmentation technology, separate the chip body image in the overall chip image and detect whether there are defects in the chip body image.

[0123] In this embodiment, when performing defect detection on chips without unqualified images or on chips after image retaking, image analysis can be performed on the overall chip image based on the AOI technology to detect defects in the chips.

[0124] Optionally, when necessary, image preprocessing such as denoising, grayscale conversion, and binarization can be performed on the overall chip image first.

[0125] Among them, defect detection includes but is not limited to detecting whether there are problems such as multiple wires or misaligned wires in the chip leads, and whether the chip body is complete or missing.

[0126] Optionally, when detecting whether there are problems such as multiple wires or misaligned wires in the chip leads, the Canny algorithm can be used to detect the edges in the image, identify the contour of the gold wire, and then detect the straight line through the Hough transform to further confirm the position and direction of the gold wire. Among them, the contour of the gold wire is extracted from the edge image, and the geometric features such as the length, angle, and position of the gold wire are analyzed to determine the specific position of the gold wire.

[0127] Compare the detected gold wire image with a pre - defined standard gold wire template, and judge whether the gold wire is correctly connected by calculating the degree of difference (such as Euclidean distance, correlation coefficient). Among them, geometric transformation is performed on the gold wire image to align it with the standard template, and then check whether there are extra gold wires, or whether the gold wire is connected to the wrong pin or position.

[0128] Optionally, when detecting whether the chip body is complete, missing, etc., based on image segmentation technology, the chip body image in the overall chip image can be separated. Among them, starting from the seed point, the area of the chip body can be gradually grown; then by setting an appropriate threshold, the chip body can be separated from the background; finally, operations such as dilation and erosion are used to further optimize the segmentation result.

[0129] Optionally, calculate the area and perimeter of the chip body, compare them with the standard values, and detect whether there are missing or damaged parts. Or analyze the roundness and shape of the chip body to detect whether there are abnormalities.

[0130] Optionally, based on the gray - level co - occurrence matrix, analyze the texture features of the chip body to detect whether there are abnormal areas.

[0131] Optionally, check whether the chip body is a single - connected region to detect whether there are broken or separated parts.

[0132] If the gold wire connection of the chip is correct and the chip body is complete and defect - free, it is a qualified product; if there are problems such as multiple wires, wrong wires, etc., and / or the chip body has problems such as missing, damaged, abnormal texture, etc., it is an unqualified product; if it is impossible to determine whether it is qualified due to image quality problems or other reasons and manual re - inspection is required, it is a product to be confirmed.

[0133] In this way, the AOI device can comprehensively and accurately analyze the overall chip image, detect problems such as multiple wires and wrong wires of the chip leads and the integrity of the chip body, thereby ensuring the quality of the chip.

[0134] In one embodiment, on the basis of the above - mentioned embodiment, after the step of marking defect - free chips as qualified products, defective chips as unqualified products, chips that have not completed defect detection as products to be confirmed, and generating the defect detection results of the chips, the following steps are further included:

[0135] Associate the defect detection results corresponding to consecutive batches of chips with the corresponding bonding parameters and array positions to generate an analysis sample; among them, the arrangement array during chip defect detection is the same as the arrangement array during gold wire bonding;

[0136] Input the analysis sample into a pre-trained artificial intelligence model for defect cause analysis to obtain a cause report for each type of defect; among them, the training samples used when training the artificial intelligence model include analysis samples of defective chips and analysis samples of non-defective chips.

[0137] In this embodiment, in order to conduct in-depth defect cause analysis, it is necessary to associate the defect detection results with relevant manufacturing parameters and location information. Among them, the bonding parameters include but are not limited to bonding pressure, temperature, time, etc. These parameters are directly related to the quality of bonding and are key factors that may cause defects; the array position is the specific position of the chip in the detection area and its position during the gold wire bonding process. Since the chips are processed in batches, different positions may be affected by the equipment uniformity. Therefore, the location information is very important for analyzing the distribution pattern of defects.

[0138] Match the defect detection results with the corresponding bonding parameters and array position information to generate analysis samples. Each analysis sample contains the following information: chip ID, defect detection result, bonding parameter, array position.

[0139] Among them, ensure that the arrangement array during chip defect detection is the same as the arrangement array during gold wire bonding, so that defects can be accurately associated with specific manufacturing conditions and avoid analysis errors caused by position mismatch.

[0140] For defect cause analysis, a trained artificial intelligence model is required. This model can learn from the analysis samples and identify which bonding parameters and location factors are associated with specific types of defects. Among them, the training samples should include two types of data: one is the analysis samples of defective chips, which contain known defects and their corresponding bonding parameters and location information, and are used to help the model learn the characteristics of defects; the other is the analysis samples of non-defective chips, which are used as references for normal chips to help the model distinguish normal and abnormal situations and improve the prediction accuracy.

[0141] Optionally, various machine learning or deep learning models can be selected for the model, such as decision trees, random forests, support vector machines, neural networks, etc. The specific selection depends on the characteristics of the data and the requirements of the analysis. Then determine which bonding parameters and location information are the most influential features on defects, which can be achieved through feature selection algorithms to improve the efficiency and accuracy of the model. The specific training process of the model is as follows:

[0142] (1) Data preparation

[0143] First, a large amount of historical data needs to be collected, including analysis samples of defective chips and analysis samples of non-defective chips. Clean the collected data to handle missing values, outliers, and inconsistent data.

[0144] Then, extract useful features from the original data. These features can include: bonding parameters and chip location information. Additionally, if combinations between certain parameters may have a greater impact on defects, interaction features can also be created.

[0145] Split the dataset into a training set, a validation set, and a test set. Optional ratios are 70%, 15%, 15% or 80%, 10%, 10%.

[0146] (2) Model Selection

[0147] Select a suitable machine learning or deep learning model. For defect cause analysis, a random forest model is preferably used. The random forest model is suitable for multi-classification problems, can handle high-dimensional data, and has a certain degree of interpretability.

[0148] (3) Model Training

[0149] Use the training set data to train the selected model. In this process, the model will learn the relationship between features and defect types. Also, use the validation set to tune the hyperparameters of the model to obtain the best performance. Methods such as grid search, random search, or Bayesian optimization can be used.

[0150] (4) Model Evaluation

[0151] Use the test set to evaluate the performance of the model. Commonly used evaluation metrics include: Accuracy, Precision, Recall, F1-score, and confusion matrix (intuitively showing the classification situation of the model).

[0152] (5) Model Deployment

[0153] Deploy the trained model to the production environment for real-time or batch defect cause analysis. As new data is collected, the model can also be updated regularly to maintain its performance. Additionally, it can be integrated with existing production systems and databases to achieve automatic data flow.

[0154] Optionally, the generated analysis samples are input into a trained artificial intelligence model, which analyzes the causes of defects based on the learned patterns and relationships and generates a cause report for each type of defect. The report will indicate which bonding parameters or location factors each defect type may be related to, and how the interaction between these factors affects the occurrence of defects. To enable engineers to understand the analysis results of the model, visual charts or explanatory texts can be provided to clearly show the relationship between defects and parameters.

[0155] In one embodiment, based on the defect cause analysis, the bonding process parameters can be adjusted specifically to optimize the production process and reduce the defect rate; by analyzing the distribution and trend of defects, potential problems of the equipment can be predicted for preventive maintenance to reduce the downtime due to failures. This can reduce the proportion of defective chips, directly reduce the production cost, and improve the production efficiency.

[0156] In this way, by using big data and artificial intelligence technologies, data-driven decision-making is realized, and the intelligent level of the entire manufacturing process is improved. It can not only improve the efficiency and accuracy of defect detection, but also promote the continuous improvement and optimization of the production process through in-depth analysis of the causes of defects.

[0157] In one embodiment, on the basis of the above embodiment, after the step of inputting the analysis samples into a pre-trained artificial intelligence model for defect cause analysis to obtain a cause report for each type of defect, it further includes:

[0158] Based on the cause report, the artificial intelligence model is used to further analyze the optimal bonding parameters corresponding to each array position;

[0159] Based on the optimal bonding parameters, the bonding parameters selected by the chip bonding equipment associated with the AOI equipment are updated.

[0160] In this embodiment, in addition to pre-training the defect cause analysis model, the artificial intelligence model can also synchronously train the optimal bonding parameter analysis model.

[0161] On the basis of training the defect cause analysis model, by defining an optimization objective (such as minimizing the defect rate or maximizing the bonding quality), and selecting a suitable optimization model (such as genetic algorithm, particle swarm optimization, Bayesian optimization, etc.), the optimal combination of bonding parameters can be searched.

[0162] Optionally, an optimization algorithm is used to search for the optimal bonding parameters for each array position. For each position, the model will try different parameter combinations to find the optimal solution that can minimize the defect rate.

[0163] After obtaining the cause reports of various types of defects, the artificial intelligence model can further analyze the optimal bonding parameters corresponding to different array positions. The purpose of this step is to identify which bonding parameter settings can minimize the occurrence of defects at different positions.

[0164] Optionally, analyze the relationship between different array positions and bonding parameters to find out which parameter combinations at specific positions lead to defects. For each type of defect, determine which bonding parameters are the main influencing factors.

[0165] Optionally, use optimization algorithms (such as genetic algorithms, particle swarm optimization, etc.) to search for the optimal bonding parameter combinations for each array position. When necessary, physical models or simulation tools can also be used to predict the bonding quality under different parameter settings, so as to assist in determining the optimal parameters.

[0166] Through the above analysis, a set of optimal bonding parameters is determined for each array position. These parameters are designed to minimize the defect rate and improve the bonding quality.

[0167] Generate a mapping table that details the optimal bonding parameters corresponding to each array position. Provide specific parameter adjustment suggestions, such as the set values of bonding pressure, temperature, time, etc.

[0168] Optionally, select some array positions in actual production, perform bonding according to the new parameter settings, and monitor whether the defect rate has improved. Verify the effectiveness of the new parameter settings. Once the optimal bonding parameters are determined and their effectiveness is verified through experiments, these parameters can be updated to the chip bonding equipment associated with the AOI device.

[0169] Optionally, through system integration, achieve automatic parameter synchronization between the artificial intelligence model and the bonding equipment. Continuously monitor the bonding quality after updating the parameters to ensure that the defect rate is effectively controlled. Regularly re-evaluate and update the optimal parameters to adapt to changes in the production process.

[0170] It should be noted that during the operation of the chip bonding equipment, due to mechanical wear, material deformation, or other physical factors, the actual effects of parameters such as bonding pressure, temperature, and time at different positions may vary. For example, the bonding heads at some positions may be worn more severely, or there may be looseness in the mechanical structure of the equipment in some areas. And during the bonding process, the temperature distribution in the heating area may be uneven. Even if an average temperature is set, the chips at different positions may receive different temperatures, thus affecting the bonding effect.

[0171] Through data analysis and optimization algorithms, the optimal bonding parameters can be customized for each array position to minimize the occurrence of defects, thereby reducing rework and scrap caused by defects and lowering production costs. The entire process is based on data analysis and artificial intelligence technologies to achieve intelligent management of the production process.

[0172] In this way, by applying the analysis results of the artificial intelligence model to the optimization of bonding parameters and updating them in real time to the production equipment, a closed-loop production optimization system can be formed to continuously improve the quality and efficiency of chip manufacturing.

[0173] In one embodiment, based on the above embodiment, after the step of updating the bonding parameters selected by the chip bonding equipment associated with the AOI equipment based on the optimal bonding parameters, the following steps are further included:

[0174] After the bonding parameters of the chip bonding equipment are updated, if the defect detection results of the latest batch of chips are improved, the model parameters of the artificial intelligence model are updated using the analysis samples corresponding to the latest batch of chips;

[0175] Based on the updated artificial intelligence model, continue to optimize the bonding parameters.

[0176] In this embodiment, after the bonding parameters of the chip bonding equipment are updated, the latest batch of chips are subjected to defect detection, and the detection results are recorded. The defect detection results of the new batch and the old batch of chips are compared to evaluate the effect after the bonding parameters are updated. If the defect detection results of the latest batch of chips are improved (for example, the number of unqualified chips is reduced), then continue to the next step.

[0177] Collect the bonding parameters, array position information, and defect detection results of the latest batch of chips. After cleaning and annotating the new sample data, use the new sample data to retrain or incrementally train the existing artificial intelligence model to update the model parameters.

[0178] Use the updated artificial intelligence model to generate a defect cause report. Based on the latest cause report, continue to use optimization algorithms (such as genetic algorithms, particle swarm optimization, etc.) to search for the optimal combination of bonding parameters for each array position. List in detail the latest optimal bonding parameters corresponding to each array position, and update the new optimal bonding parameters to the bonding equipment.

[0179] In actual production tests, some array positions can be selected to perform bonding according to the new parameter settings, and monitor whether the defect rate is further improved. Regularly re-evaluate and update the optimal parameters to adapt to changes in the production process.

[0180] Through the above steps, a closed-loop system for continuous improvement can be achieved. After each update of the bonding parameters, the latest defect detection results are evaluated to determine whether to further update the artificial intelligence model. In this way, the model can continuously learn new data, optimize the bonding parameters, and further improve the quality and efficiency of chip manufacturing. The entire process is based on data analysis and artificial intelligence technology to achieve intelligent management of the production process.

[0181] In one embodiment, based on the above embodiment, after the step of marking the defect-free chips as qualified products, the defective chips as unqualified products, the chips that fail to complete the defect detection as products to be confirmed, and generating the defect detection results of the chips, the following steps are further included:

[0182] Sort the qualified products, unqualified products, and products to be confirmed, and convey the qualified products to the downstream equipment of the AOI equipment or store them in a cassette, convey the unqualified products to the recycling area, and convey the products to be confirmed to the manual inspection area.

[0183] In this embodiment, during the defect detection process, record the specific position of each chip in the array (for example, represented using a coordinate system). Associate the defect detection result of each chip with its position information in the array to form a mapping table containing position information and detection results.

[0184] The conveyor belt inside the AOI equipment starts to work, moving the chips that have completed the detection from the detection platform to the sorting area. The loading and unloading mechanism is ready to receive the chips on the conveyor belt and transfer them to the working range of the robotic arm. The robotic arm reads the position information of the chip in the array and, based on the position information and the results in the mapping table, can move the chips marked as "qualified products" to the conveyor belt of the downstream equipment to continue the subsequent processes (such as packaging, testing, etc.), or store the qualified product chips in a preset cassette for future use.

[0185] The robotic arm moves the chips marked as "unqualified products" to the conveyor belt in the recycling area, and these chips will be sent to the recycling or scrapping area.

[0186] The robotic arm moves the chips marked as "products to be confirmed" to the conveyor belt in the manual inspection area, and these chips will be further inspected and confirmed by the staff.

[0187] The AOI equipment records the sorting results of each chip, including the sorting time, sorting position, and sorting target (downstream equipment, cassette, recycling area, manual inspection area).

[0188] Through the above steps, the AOI equipment can achieve automatic classification and sorting of chips. Such an automated sorting system can ensure the efficient and accurate classification of chips, improve production efficiency, reduce manual intervention and errors, and ensure more precise and efficient quality management of chips.

[0189] In one embodiment, based on the above embodiment, the method for defect detection of a batch of chips from multiple angles further includes:

[0190] When capturing panoramic images of each orientation of the chip array, obtain the relative position relationship between the camera array and the chip array based on visual detection;

[0191] According to the relative position relationship between the camera array and the chip array, uniformly set the initial focal lengths of the cameras in the camera array;

[0192] According to the array positions of the cameras and the photosensitive data detected by each camera based on the initial focal length, determine the shooting areas responsible for each camera; wherein, the set of shooting areas responsible for each camera covers the area where the chip array is located;

[0193] Based on the relative position relationship between the camera array and the chip array, obtain the relative position relationship between each camera and the responsible shooting area based on visual detection;

[0194] Based on the relative position relationship between each camera and the responsible shooting area, and the photosensitive data detected by each camera, adjust the focal lengths of the cameras based on the initial focal length.

[0195] In this embodiment, before capturing the panoramic image of the chip array, the camera array obtains the relative position relationship between the camera array and the chip array through visual detection technology.

[0196] The parameter description of the relative position relationship may include the distance, angle, offset, etc. between the camera array and the chip array. Among them, the distance is the spatial distance between the camera array and the chip array, the angle is the angular deviation between the camera array and the chip array (such as the horizontal angle and the vertical angle), and the offset is the geometric center offset of the camera array relative to the chip array.

[0197] Since there are multiple cameras in the camera array, the distance and angle between the camera array and the shooting area can be calculated through stereo vision technology based on multi-view visual detection, and the spatial relationship between the two can be calculated by combining image processing algorithms. The system will generate an accurate relative position data for subsequent focal length setting and shooting area division.

[0198] After obtaining the relative position relationship between the camera array and the chip array, calculate a suitable focal length value according to the distance between the camera array and the chip array, and uniformly set the initial focal lengths of the cameras in the camera array. Ensure that when all cameras start shooting, the focal length setting is reasonable, and avoid image blurring or out-of-focus problems caused by focal length differences.

[0199] For example, if the distance between the camera array and the chip array is 10 cm, a suitable focal length value (such as 50 mm) is calculated and uniformly set for all cameras.

[0200] The camera array is composed of multiple cameras arranged in a certain layout (such as a linear array or a matrix array). The array position of each camera determines its position and angle in the entire detection system. Based on the array positions of the cameras and combined with their photosensitive data (such as brightness, contrast, etc.) at the initial focal length, the specific shooting area responsible for each camera is determined.

[0201] Among them, the photosensitive data can help determine whether each camera can clearly capture its responsible area at the initial focal length.

[0202] Area coverage requirement: The set of shooting areas of all cameras must cover the area where the entire chip array is located to ensure that each chip can be captured by at least one camera.

[0203] For example, if the camera array is linearly arranged, camera 1 is responsible for the left area of the chip array, camera 2 is responsible for the middle area, and camera 3 is responsible for the right area; in a matrix array, camera 1 is responsible for the upper left area, camera 2 is responsible for the upper right area, and so on.

[0204] The previously obtained relative position relationship between the camera array and the chip array can provide a global reference for the subsequent relative position detection between each camera and its shooting area. Including:

[0205] The camera array is usually composed of multiple cameras arranged in a certain layout (such as a linear array, a matrix array or a circular array). The array position of each camera determines its position and angle in the overall system.

[0206] Each camera needs to obtain the specific relative position relationship between itself and its responsible shooting area through visual detection to ensure that it can accurately focus and clearly shoot its responsible area. Among them, image matching techniques (such as SIFT, ORB, etc.) can be used to match the image captured by the camera with the pre-stored chip array template to calculate the relative position.

[0207] Each camera transmits the detected relative position data to the system, and the system integrates the relative position data of all cameras to generate a global position relationship map to ensure that each camera clearly knows its relative position to the shooting area.

[0208] Example scenario:

[0209] Example 1: The distance between camera 1 and its responsible shooting area is 12 cm, the horizontal angle deviation is 2 degrees, the vertical angle deviation is 1 degree, and the geometric center offset is (0.5 mm, 0 mm);

[0210] Example 2: The distance between camera 2 and its responsible shooting area is 10 cm, the horizontal angle deviation is 0 degrees, the vertical angle deviation is 0 degrees, and the geometric center offset is (0 mm, 0 mm).

[0211] Example 3: The distance between camera 3 and its responsible shooting area is 11 cm, the horizontal angle deviation is -1 degree, the vertical angle deviation is 0.5 degrees, and the geometric center offset is (0 mm, -0.3 mm).

[0212] Optionally, based on the precise relative position relationship between each camera and its shooting area, and the detected photosensitive data, fine-tuning is performed on the basis of the initial focal length to ensure that each camera can clearly capture its responsible shooting area.

[0213] Optionally, based on the distance and angle deviation between each camera and its shooting area, the focal length value to be adjusted is calculated. Combining the photosensitive data (such as the clarity and contrast of the image), the effect of the focal length adjustment is judged, and the focal length setting is further optimized.

[0214] For example, if the distance between a certain camera and its shooting area is far, the focal length is appropriately increased to ensure clear images; if the distance between a certain camera and its shooting area is close, the focal length is appropriately decreased to avoid image distortion. Through cyclic adjustment and detection, ensure that the focal length of each camera is suitable for its shooting area.

[0215] In this way, it can be ensured that each part of the chip array is clearly photographed. At the same time, through focal length adjustment, image blurring or distortion is avoided, so as to provide high-quality image data for subsequent defect detection and improve the accuracy and efficiency of detection.

[0216] In addition, an embodiment of the present application also provides a control device, and the internal architecture of the control device can be as Figure 3 shown, including a processor, a memory, a communication interface, and an input interface connected through a system bus. Among them, the processor is used to provide computing and control capabilities. The memory includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database is used to store data called by the computer program. The communication interface is used to communicate with an external terminal for data. The input interface is used to receive signals input by an external device. When the computer program is executed by the processor, it realizes a method for defect detection of a batch of chips at multiple angles as described in the above embodiments.

[0217] Those skilled in the art can understand, Figure 3The structure shown is only a block diagram of some of the structures related to the solution of this application, and does not constitute a limitation on the control device to which the solution of this application is applied. For example, in some alternative embodiments, the control device may further include an output interface (not shown in the figure), and the output interface is also connected to the system bus and is used to output corresponding signals to external devices.

[0218] In addition, this application also provides an AOI device, which includes a camera array, a moving camera, and a control device; wherein, the camera array and the moving camera are both communicatively connected to the control device and are controlled by the control device;

[0219] The control device is the control device as described above.

[0220] In addition, this application also provides a computer-readable storage medium, which includes a computer program. When the computer program is executed by a processor, it implements the steps of the method for defect detection of a batch of chips from multiple angles as described in the above embodiments. It can be understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0221] In summary, for the method for defect detection of a batch of chips from multiple angles, the control device, the AOI device, and the computer-readable storage medium provided in the embodiments of this application, the chip array after wire bonding is placed in the detection area of the AOI device, and multiple angles of multiple chips are simultaneously photographed by the camera array, significantly reducing the independent shooting time of each chip and improving the overall detection efficiency. While performing chip defect detection on qualified images, the moving camera can be used to intelligently reshoot some of the images with insufficient quality captured by the camera array, ensuring the detection quality and efficiency. And during the reshooting process, by intelligently planning the path and optimizing the motion control of the camera, unnecessary movements and pauses are reduced, not only taking into account the reshooting efficiency and the quality of the reshot images, but also reducing mechanical wear and extending the service life of the device. In this way, it helps to improve the overall defect detection efficiency of the batch of chips, thereby improving the production efficiency of the batch of chips and ensuring the chip quality.

[0222] Those of ordinary skill in the art can understand that all or part of the processes in the above-described method embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the above-described method embodiments. Among them, any reference to a memory, storage, database, or other medium provided in this application and used in the embodiments can include non-volatile and / or volatile memories. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and Rambus dynamic RAM (RDRAM), etc.

[0223] It should be noted that in this document, the terms "include", "comprise", or any other variant thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that includes a series of elements includes not only those elements but also other elements not expressly listed, or elements that are inherent to such process, apparatus, article, or method. Without further limitation, an element defined by the statement "including one..." does not exclude the existence of additional identical elements in the process, apparatus, article, or method that includes such element.

[0224] The above are only the preferred embodiments of this application, and do not limit the patent scope of this application. Any equivalent structure or equivalent process transformation made using the specification and drawings of this application, or directly or indirectly applied in other related technical fields, is similarly included in the patent protection scope of this application.

Claims

1. A multi-angle defect detection method for batch chips, characterized in that: include: The chip array after gold wire bonding is placed in the detection area of ​​the AOI equipment; wherein the detection area is provided with a camera array and a motion camera; Control the camera array to operate at multiple shooting angles, capture images of the chip array in multiple directions, and perform image stitching to obtain panoramic images in multiple directions; wherein, during the camera array shooting process, each camera focuses independently; Perform image segmentation on panoramic images in multiple directions to obtain multi-angle images corresponding to each chip; Perform image quality checks on images from multiple angles and filter out unqualified images; For chips without unqualified images, multi-angle images are stitched to obtain the overall image of each chip, and the overall image of the chip is analyzed based on the AOI technology to detect defects on the chip; and the image reshoot path is intelligently planned according to the chip position and orientation to which the unqualified image belongs; the path in the same orientation is calculated according to the shortest distance of the line connecting the chip positions of all unqualified images in the same orientation; and the paths in different orientations are connected to form the image reshoot path according to the shortest switching distance between the paths in different orientations; Control the motion camera to retake the unqualified images along the image retake path; wherein the interval path of the motion camera between the two retake points includes a speed-up segment, a speed-down segment, a uniform speed preparation segment and a uniform speed capture segment in sequence, and the length of the uniform speed preparation segment is positively correlated with the deceleration of the speed-down segment; the end point of the uniform speed capture segment of the previous interval path is the start point of the speed-up segment of the next interval path; Replace unqualified images with retaken images; For the chip after the image reshoot, multi-angle images are stitched together to obtain the overall image of the chip, and the overall image of the chip is analyzed based on AOI technology to detect defects on the chip; Mark non-defective chips as qualified products, mark defective chips as unqualified products, mark chips that have failed defect detection as pending products, and generate defect detection results for the chips; The step of performing image analysis on the overall image of the chip based on the AOI technology to detect defects on the chip includes: Locate the gold wire image in the overall chip image based on the edge detection algorithm and analyze whether the wiring in the gold wire image is correct; Based on image segmentation technology, the chip body image is separated from the overall chip image, and the chip body image is detected to see if there are defects.

2. The multi-angle defect detection method for batch chips according to claim 1, characterized in that: After the step of marking the non-defective chips as qualified products, marking the defective chips as unqualified products, marking the chips that failed to complete the defect detection as pending products, and generating the defect detection results of the chips, the method further includes: Correlate the defect detection results corresponding to multiple batches of chips with the corresponding bonding parameters and array positions to generate analysis samples; wherein the array arrangement during chip defect detection is the same as the array arrangement during gold wire bonding; The analysis samples are input into a pre-trained artificial intelligence model for defect cause analysis to obtain a cause report for each type of defect; wherein, the training samples used in training the artificial intelligence model include analysis samples of defective chips and analysis samples of non-defective chips.

3. The multi-angle defect detection method for batch chips as claimed in claim 2, characterized in that: After the step of inputting the analysis sample into the pre-trained artificial intelligence model to perform defect cause analysis to obtain a cause report of each type of defect, the method further includes: Based on the cause report, the artificial intelligence model is used to further analyze the optimal bonding parameters corresponding to each array position; Based on the optimal bonding parameters, the bonding parameters selected by the chip bonding equipment associated with the AOI equipment are updated.

4. The multi-angle defect detection method for batch chips as claimed in claim 3, characterized in that: After the step of updating the bonding parameters selected by the chip bonding equipment associated with the AOI equipment based on the optimal bonding parameters, the method further includes: After the bonding parameters of the chip bonding equipment are updated, if the defect detection results of the latest batch of chips are improved, the model parameters of the artificial intelligence model are updated using the analysis samples corresponding to the latest batch of chips; Based on the updated AI model, continue to optimize bonding parameters.

5. The multi-angle defect detection method for batch chips as claimed in claim 1, characterized in that: After the step of marking the non-defective chips as qualified products, marking the defective chips as unqualified products, marking the chips that failed to complete the defect detection as pending products, and generating the defect detection results of the chips, the method further includes: Qualified products, unqualified products and products to be confirmed are sorted, and qualified products are transported to the downstream equipment of the AOI equipment or stored in the material box, unqualified products are transported to the recycling area, and products to be confirmed are transported to the manual inspection area.

6. The multi-angle defect detection method for batch chips as claimed in claim 1, characterized in that: The multi-angle defect detection method for batch chips also includes: When taking images of the chip array in all directions, the relative position relationship between the camera array and the chip array is obtained based on visual detection; According to the relative position relationship between the camera array and the chip array, the initial focal length of each camera in the camera array is uniformly set; According to the array position of each camera and the light-sensing data detected by each camera based on the initial focal length, the shooting area that each camera is responsible for is determined; wherein the set of shooting areas that each camera is responsible for covers the area where the chip array is located; Based on the relative positional relationship between the camera array and the chip array, the relative positional relationship between each camera and the shooting area it is responsible for is obtained based on visual detection; Based on the relative positional relationship between each camera and the shooting area it is responsible for, and the light-sensitive data detected by each camera, the focal length of each camera is adjusted on the basis of the initial focal length.

7. A control device, characterized in that: The control device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, the steps of the multi-angle defect detection method for batch chips as described in any one of claims 1 to 6 are implemented.

8. An AOI device, characterized in that: The AOI device includes a camera array, a motion camera and a control device; wherein the camera array and the motion camera are both connected to the control device in communication and are controlled by the control device; The control device is the control device according to claim 7.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the multi-angle defect detection method for batch chips according to any one of claims 1 to 6 are implemented.

Images