An integrated IC substrate appearance inspection and AI re-inspection device based on multiple images
By integrating multi-image and AI re-inspection equipment, the problems of single inspection dimension, high labor cost and misjudgment in IC substrate appearance inspection have been solved, realizing efficient and low-damage defect detection and sorting, and improving inspection coverage and judgment consistency.
Patent Information
- Application Number
- CN202611623231.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-10-20
- Publication Date
- 2026-11-17
AI Technical Summary
Existing IC substrate appearance inspection technologies suffer from problems such as limited inspection dimensions, easy to miss defects, high labor costs, inconsistent judgment standards, easy to misjudge due to fatigue, and high risk of secondary damage.
An integrated IC carrier appearance inspection and AI re-inspection device based on multiple images is adopted. Combining multi-source imaging, AI high-magnification imaging and manual final judgment, it realizes double-sided inspection of IC carriers, AI preliminary judgment and manual confirmation, reducing manual contact and completing inspection, judgment and sorting in the same device.
It significantly improves the imaging coverage of minute defects, reduces the risk of missed detection, reduces the workload of manual review, reduces manpower and costs, avoids product damage, and improves production efficiency and the consistency of defect judgment.
Smart Images

Figure CN123111964A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of semiconductor packaging inspection technology, and in particular to an integrated device for IC substrate appearance inspection and AI re-inspection based on multiple images. Background Technology
[0002] Currently, IC substrate visual inspection (AVI) mainly relies on a single-channel optical imaging system with a resolution of 2.5μm or 5μm for initial scanning. After detecting suspected defects, secondary confirmation is required by manual inspection at a review station (VRS) using a high-powered microscope.
[0003] However, the aforementioned prior art has the following drawbacks: Firstly, the detection dimension is limited, and a single-channel image cannot fully capture the defect features of different materials, heights, or reflective properties, which easily leads to omissions. Secondly, labor costs are high, as all defect images need to be manually judged, resulting in a huge manpower burden during the mass production stage; Third, inconsistent judgment standards can easily lead to fatigue-induced misjudgments during long-term operation, especially misjudging real defects as false defects, i.e., missed judgments or false negatives (i.e., misjudging real defects as no defects), which can lead to quality complaints. Fourth, there is a risk of secondary damage. Manual re-inspection involves direct contact between the product and the re-inspection platform, which can easily cause scratches or contamination to the delicate IC substrate.
[0004] To address this, a device integrating IC substrate appearance inspection and AI re-inspection based on multiple images is proposed. Summary of the Invention
[0005] The purpose of this invention is to provide an integrated IC substrate appearance inspection and AI re-inspection device based on multiple images, thereby solving or at least alleviating one or more of the above-mentioned problems and other problems existing in the prior art.
[0006] To achieve the above objectives, the main technical solution adopted by the present invention includes: an integrated IC carrier board appearance inspection and AI re-inspection device based on multiple images, including a frame, on which a feeding module, an appearance inspection module, an AI re-inspection positioning and high-magnification imaging module, a temporary storage module, a sorting execution module, a manual judgment module and a main controller are arranged; The loading module is used to transfer the IC substrate to be inspected to the appearance inspection module; The appearance inspection module is used to acquire images of side A and side B of the IC carrier board, obtain multiple image data, and output the location of suspected defect marks on the IC carrier board based on the multiple image data. The AI re-inspection positioning and high-magnification imaging module includes a high-magnification acquisition imaging module and an AI processing module. The high-magnification acquisition imaging module is used to acquire high-magnification defect images of the corresponding surfaces of the IC carrier board according to the suspected defect mark positions. The AI processing module is used to determine whether the high-magnification defect images are true or false defects and output high-confidence false defects, high-confidence true defects, and low-confidence defects. The temporary storage module is used to receive and temporarily store all IC carrier boards that have been detected by the AI re-inspection and positioning and high-magnification imaging module, and to maintain the temporary storage state before all defects of the IC carrier board are judged by AI and low-confidence defects are judged manually. The manual judgment module is used to receive low-confidence defects output by the AI processing module, retrieve the high-magnification defect image corresponding to the low-confidence defect, and output the final manual judgment result. The sorting execution module is used to sort the IC carrier boards in the temporary storage module to different bins after all defects on the IC carrier board have been judged, based on the high-confidence true defects and high-confidence false defects output by the AI processing module and the final judgment result output by the manual judgment module. The feeding module, the appearance inspection module, the AI re-inspection positioning and high-magnification imaging module, the temporary storage module, the manual judgment module, and the sorting execution module are all communicatively connected to the main controller.
[0007] Furthermore, the AI processing module includes a defect region enhancement submodule, a multi-scale feature extraction submodule, a contrastive learning feature encoding submodule, and a genuine / fake defect determination submodule; The defect region enhancement submodule is data-connected to the high-magnification acquisition imaging module and is used to receive the high-magnification defect image and perform local contrast enhancement, edge response enhancement and high-frequency texture enhancement on the candidate defect region in the image. The multi-scale feature extraction submodule is data-connected to the defect region enhancement submodule and is used to receive the enhanced defect image. The multi-scale feature extraction submodule includes multiple feature branches, each of which has a different receptive field. The multiple feature branches include at least a first feature branch, a second feature branch, and a third feature branch. The receptive field of the first feature branch is smaller than that of the second feature branch, and the receptive field of the second feature branch is smaller than that of the third feature branch. The contrastive learning feature encoding submodule is data-connected to the multi-scale feature extraction submodule and is used to encode the features output by the multi-scale feature extraction submodule. During training, it uses defect samples of the same type as positive sample pairs and defect samples of different types or with different true and false attributes as negative sample pairs, and trains through feature distance constraints. The true / false defect determination submodule is data-connected to the contrastive learning feature encoding submodule, and is used to output the high-confidence fake defect, the high-confidence true defect, and the low-confidence defect based on the features output by the contrastive learning feature encoding submodule.
[0008] Furthermore, the appearance inspection module includes an A-side inspection component, a B-side inspection component, and a transfer and flipping component. The A-side inspection component is used to inspect the A-side of the IC carrier, and the B-side inspection component is used to inspect the B-side of the IC carrier. The A-side inspection component and the B-side inspection component are respectively disposed on both sides of the transfer and flipping component. The A-side inspection component and the B-side inspection component have the same structure, both including a support frame and a first Y-axis linear module. The support frame and the first Y-axis linear module are both fixedly installed on the frame. A first Z-axis linear module and a combined light source are fixedly installed on the support frame. A line scan camera is fixedly installed on the slide of the first Z-axis linear module. The combined light source is located below the lens of the line scan camera. A first X-axis linear module is fixedly installed on the slide of the first Y-axis linear module. A first negative pressure suction cup is fixedly installed on the slide of the first X-axis linear module. The combined light source includes a coaxial light source, a uniform light source, a multi-angle side light source, and an infrared light source.
[0009] Furthermore, the transplanting and turning assembly includes a transplanting unit and a turning unit. The transplanting unit includes an X-axis two-moving linear module. The X-axis two-moving linear module is fixedly installed on the frame by a first column. A second gripper is installed on one of the moving parts of the X-axis two-moving linear module, and a third gripper is installed on the other moving part. The flipping unit is disposed between the second gripper and the third gripper. The flipping unit includes a support base, which is fixedly mounted on the frame. A flipping motor and a third negative pressure suction cup are fixedly mounted on the support base. A flipping shaft is fixedly connected to the rotating shaft of the flipping motor via a coupling. A second negative pressure suction cup is fixedly mounted on the flipping shaft. The second gripper can transfer the IC carrier board on the A-side detection assembly to the second negative pressure suction cup. When the flipping shaft rotates, flipping the IC carrier board on the second negative pressure suction cup to the third negative pressure suction cup, the third gripper can transfer the IC carrier board on the third negative pressure suction cup to the first negative pressure suction cup in the B-side detection assembly.
[0010] Furthermore, the high-magnification acquisition imaging module includes an A-side re-inspection component and a B-side re-inspection component; The A-side re-inspection assembly includes a second X-axis linear module and a second Y-axis linear module. The second X-axis linear module is fixedly mounted on the frame via a second column. A second Z-axis linear module is fixedly mounted on the slide of the second X-axis linear module. A first connecting seat is fixedly mounted on the slide of the second Z-axis linear module. A fourth negative pressure suction cup is fixedly mounted at the bottom of the first connecting seat. The fourth negative pressure suction cup is used to suction the B-side of the IC carrier board from above, so that the A-side is facing down. The second Y-axis linear module is fixedly mounted on the frame and located at the work position below the fourth negative pressure suction cup. A first high-definition camera is mounted on the slide of the second Y-axis linear module via a second connecting seat. The lens of the first high-definition camera faces upward and is used to photograph the A-side of the IC carrier board facing down. The second X-axis linear module moves the product it is holding, and the second Y-axis linear module moves the first high-definition camera to jointly complete the high-magnification imaging of the A-side.
[0011] The B-side re-inspection assembly includes a third X-axis linear module and a third Y-axis linear module. The third X-axis linear module is fixedly mounted on the frame via a third column. A third Z-axis linear module is fixedly mounted on the slide of the third X-axis linear module. A second camera bracket is fixedly mounted on the slide of the third Z-axis linear module, and a second high-definition camera is fixedly mounted on the second camera bracket, with the lens of the second high-definition camera facing downwards. The third Y-axis linear module is fixedly mounted on the frame, and a third camera bracket is fixedly mounted on the slide of the third Y-axis linear module. A B-side re-inspection stage is provided, on which a positioning suction cup is provided for fixing the IC carrier board with the B-side facing upwards. The lens of the second high-definition camera is aimed at the B-side of the IC carrier board on the B-side re-inspection stage. The fourth negative pressure suction cup can pick up the IC carrier board on the B-side inspection component and make its B-side be adsorbed and its A-side face downwards, so that the first high-definition camera can capture a high-magnification defect image of the A-side. After the A-side image is captured, the fourth negative pressure suction cup places the IC carrier board on the B-side re-inspection stage, keeping the B-side facing upwards, so that the second high-definition camera can capture a high-magnification defect image of the B-side.
[0012] Furthermore, the temporary storage module includes a fourth gripper and a temporary storage compartment. The fourth gripper is capable of moving the IC carrier board on the B-side re-inspection platform into the temporary storage compartment, which is then pulled out and installed on the rack.
[0013] Furthermore, the sorting execution module includes a fifth gripper, a cleaning chamber, a qualified chamber, and a scrap chamber. The cleaning chamber, the qualified chamber, and the scrap chamber are all pulled out and installed on the rack. The fifth gripper can move the IC carrier board into the cleaning chamber, the qualified chamber, or the scrap chamber according to the final judgment result.
[0014] Furthermore, the feeding module includes a feeding bin and a first gripper. The feeding bin is pulled out and installed on the frame. The first gripper can transfer the IC carrier board of the feeding bin to the first negative pressure suction cup in the A-side detection component.
[0015] Furthermore, an X-axis three-moving linear module is fixedly installed on the frame via a fourth column, and the first gripper, the fourth gripper, and the fifth gripper are respectively fixed on the corresponding moving parts in the X-axis three-moving linear module.
[0016] Furthermore, the main controller is configured to perform the following steps: S1. Loading materials: The main controller controls the first gripper to take out the IC carrier board to be tested from the loading bin and transfer the IC carrier board to the first negative pressure suction cup in the A-side detection component; the first negative pressure suction cup adsorbs and fixes the IC carrier board so that the A-side of the IC carrier board faces upward. S2, A-side multiple image scanning: The main controller controls the first Y-axis linear module and the first X-axis linear module to move the first negative pressure suction cup and the IC carrier board on it to below the line scan camera; the main controller controls the first Z-axis linear module to adjust the focal length of the line scan camera; the main controller controls the combined light source to flash sequentially according to a preset timing sequence, and synchronously triggers the line scan camera to scan and image the A-side of the IC carrier board, obtaining multiple image data of the A-side; the appearance inspection module outputs the suspected defect marking positions of the A-side of the IC carrier board based on the multiple image data of the A-side; S3, Flip: The main controller controls the second gripper to transfer the IC carrier board on the A-side detection assembly from the first negative pressure suction cup to the second negative pressure suction cup in the flipping unit; the main controller controls the flipping motor to drive the flipping shaft to rotate 180°, so that the IC carrier board on the second negative pressure suction cup is flipped onto the third negative pressure suction cup; the main controller controls the third gripper to transfer the IC carrier board on the third negative pressure suction cup to the first negative pressure suction cup in the B-side detection assembly, so that the B-side of the IC carrier board is facing upwards; S4, B-side multiple image scanning: The main controller controls the operation of the first Y-axis linear module and the first X-axis linear module in the B-side inspection assembly to move the first negative pressure suction cup and the IC carrier board on it in the B-side inspection assembly to below the line scan camera in the B-side inspection assembly; the main controller controls the first Z-axis linear module in the B-side inspection assembly to adjust its focus; the main controller controls the combined light source in the B-side inspection assembly to flash sequentially according to a preset timing sequence, and synchronously triggers the line scan camera in the B-side inspection assembly to scan and image the B-side of the IC carrier board, obtaining multiple image data of the B-side; the appearance inspection module outputs the suspected defect marking positions on the B-side of the IC carrier board based on the multiple image data of the B-side; S5, AI-powered re-examination positioning and high-magnification imaging: S51, High-magnification acquisition on side A: The main controller controls the second X-axis linear module and the second Z-axis linear module to move the fourth negative pressure suction cup above the first negative pressure suction cup of the B-side detection component; the main controller controls the negative pressure source to make the fourth negative pressure suction cup pick up the B-side of the IC carrier board from the B-side initial inspection station, so that the A-side is facing down; the main controller controls the second X-axis linear module to move the product it is holding according to the suspected defect mark position on the A-side, and at the same time controls the second Y-axis linear module to move the first high-definition camera facing upward, so that the first high-definition camera is aligned with each suspected defect mark position on the A-side of the IC carrier board in sequence, and acquires high-magnification defect images of the A-side; the second Z-axis linear module is used for product picking and height adjustment.
[0017] S52, B-side high-magnification acquisition: After the A-side image is acquired, the main controller controls the fourth negative pressure suction cup to place the IC carrier board onto the B-side re-inspection stage on the third Y-axis linear module slide and fix it with the positioning suction cup, keeping the B-side facing upwards. According to the position of the suspected defect mark on the B-side, the main controller controls the third X-axis linear module and the third Z-axis linear module to drive the downward-facing second high-definition camera to move. At the same time, the third Y-axis linear module drives the B-side re-inspection stage to move, so that the second high-definition camera is aligned with each suspected defect mark position on the B-side of the IC carrier board in sequence, and high-magnification defect images of the B-side are acquired.
[0018] S6, AI authenticity defect determination: The AI processing module receives the high-magnification defect image of side B and the high-magnification defect image of side A; the defect region enhancement submodule first performs local contrast enhancement, edge response enhancement, and high-frequency texture enhancement on the candidate defect regions in the high-magnification defect image; the multi-scale feature extraction submodule extracts first-scale features, second-scale features, and third-scale features from the enhanced image and fuses them; the contrastive learning feature encoding submodule encodes the fused features; and the true / false defect determination submodule outputs high-confidence fake defects, high-confidence true defects, and low-confidence defects based on the encoded features. S7, Temporary Storage: The main controller controls the fourth gripper to move the IC carrier board on the B-side re-inspection platform to the temporary storage compartment; the temporary storage compartment receives and temporarily stores all IC carrier boards that have been detected by the AI re-inspection positioning and high-magnification imaging module, and maintains the temporary storage state until all defects of the IC carrier board are judged by AI and low-confidence defects are judged manually. S8. Sorting: The main controller determines whether all defects on the same IC carrier board in the temporary storage compartment have been identified. After identification, the main controller controls the fifth gripper to determine the final confirmed true defect data of the IC carrier board based on the high-confidence true defects and high-confidence false defects output by the AI processing module and the final judgment result output by the manual judgment module, and performs the following sorting actions: If the IC carrier board is full of high-confidence defects, the high-confidence defects are directly filtered out and not included in the sorting defect statistics, and the fifth gripper is controlled to transfer the IC carrier board to the qualified warehouse. If at least one high-confidence true defect and / or low-confidence defect exists on the IC carrier, the following sorting action shall be performed on the IC carrier: Specifically, high-confidence true defects on the IC carrier are directly included in the final confirmed true defect data of the IC carrier; low-confidence defects on the IC carrier are subject to manual final judgment by the manual judgment module. If the manual final judgment confirms that the defect is a true defect, it is included in the final confirmed true defect data of the IC carrier; if the manual final judgment confirms that the defect is a false defect, it is directly filtered out and not included in the sorting defect statistics; high-confidence false defects on the IC carrier are directly filtered out and not included in the sorting defect statistics. The defects in the final confirmed true defect data of the IC carrier board are divided into fatal defects and repairable defects. The fatal defects include at least one of board damage, short circuit, open circuit, and missing parts. The repairable defects include at least one of foreign matter and dirt. If the fatal defect is found in the final confirmed true defect data of the IC carrier, the fifth gripper is controlled to transfer the IC carrier to the scrap bin. If the final confirmed true defect data of the IC carrier board only contains the repairable defect, then control the fifth gripper to transfer the IC carrier board to the cleaning chamber; If the final confirmed true defect data of the IC carrier is empty, the fifth gripper is controlled to transfer the IC carrier to the qualified warehouse.
[0019] The present invention has at least the following beneficial effects: The combined light source in the appearance inspection module includes a coaxial light source, a uniform light source, a multi-angle side light source, and an infrared light source. During the scanning process, the light source sequentially flashes and synchronously triggers the line scan camera to image. Multiple image data can be obtained in a single scan. Compared with the existing 2.5μm or 5μm single-channel imaging system, this application can simultaneously capture defect features of different materials, different heights, and different reflectivity, significantly improving the imaging coverage and detection stability of micro-defects and reducing the risk of missed detection.
[0020] The AI-based re-inspection and positioning module acquires high-magnification defect images of the corresponding surfaces, and the AI processing module determines whether the defects are genuine or fake, outputting high-confidence fake defects, high-confidence true defects, and low-confidence defects. Among them, high-confidence fake defects are directly filtered out, high-confidence true defects are directly adopted, and only low-confidence defects enter the manual judgment module for remote final judgment. Compared with the existing manual full-scale re-judgment method, this application significantly reduces the workload of manual re-judgment and significantly reduces the manpower load and labor costs in the mass production stage.
[0021] The AI processing module utilizes a defect region enhancement submodule, a multi-scale feature extraction submodule, a contrastive learning feature encoding submodule, and a genuine / fake defect determination submodule to determine the authenticity of suspected defects using a unified standard. Specifically, the defect region enhancement submodule suppresses interference from copper foil textures, lighting fluctuations, resin patterns, and normal process textures; the multi-scale feature extraction submodule simultaneously handles defects of varying sizes, including micron-level cracks, small foreign objects, broken wires, large-area contamination, and edge chipping; and the contrastive learning feature encoding submodule improves the separability of genuine and false defects. Compared to manual judgment, this application reduces judgment fluctuations caused by fatigue, lowers the risk of misclassifying genuine defects as false defects (False Negative), and enhances the consistency and reliability of defect data.
[0022] By remotely retrieving high-magnification defect images corresponding to low-confidence defects through a manual judgment module for final judgment, the manual personnel no longer need to directly contact the IC substrate or search for defects under a microscope. Compared with the existing method of manual secondary confirmation using a high-magnification microscope at the re-judgment station, this application eliminates the contact-based re-judgment steps such as manual handling and microscope observation, effectively avoiding potential secondary physical damage to the product during the circulation process. At the same time, it allows the manual personnel to make decisions only by looking at the images, improving the judgment speed of individual defects.
[0023] This application integrates the feeding module, appearance inspection module, AI re-inspection positioning and high-magnification imaging module, temporary storage module, sorting execution module, manual judgment module and main controller into the same rack, and completes double-sided appearance inspection, AI high-magnification re-inspection, manual remote final judgment, temporary storage and sorting in the same equipment. This reduces the repetitive process of repeated loading and unloading of products, reduces the waste of factory resources, avoids the damage to products caused by repeated loading and unloading, and improves factory production efficiency. Attached Figure Description
[0024] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a schematic diagram of the structure of the present invention; Figure 2 This is one of the partial structural schematic diagrams of the present invention; Figure 3 This is a schematic diagram of the structure of the qualified warehouse of the present invention; Figure 4 This is a second partial structural schematic diagram of the present invention; Figure 5 This is the third partial structural schematic diagram of the present invention; Figure 6 This is a schematic diagram of the structure of the A-side detection component of the present invention; Figure 7 This is a schematic diagram of the structure of the B-side detection component of the present invention; Figure 8 This is a schematic diagram of the transplanting and flipping component of the present invention; Figure 9 This is a schematic diagram of the transplanting unit of the present invention; Figure 10 This is a schematic diagram of the structure of the flipping unit of the present invention; Figure 11 This is a schematic diagram of the structure of the A-side re-inspection component of the present invention; Figure 12 This is a schematic diagram of the structure of the B-side re-inspection component of the present invention; Figure 13 This is a schematic diagram of the X-axis three-moving-sub linear module of the present invention; Figure 14 This is a schematic diagram of the architecture of the AI processing module of the present invention.
[0025] Explanation of icon numbers: 1. Rack; 2. Feeding module; 201. Feeding bin; 202. First gripper; 3. Appearance Inspection Module; 301. A-side Inspection Component; 302. B-side Inspection Component; 303. Transplanting and Tilting Component; 3031. Transplanting Unit; 3032. Tilting Unit; 304. Support Frame; 305. First Z-axis Linear Module; 306. Linear Scan Camera; 307. Combined Light Source; 308. First Y-axis Linear Module; 309. First X-axis Linear Module; 310. First Negative Pressure Suction Cup; 311. X-axis Two-Wheel Linear Module; 312. Second Gripper; 313. Third Gripper; 314. Support Base; 315. Tilting Motor; 316. Tilting Axis; 317. Second Negative Pressure Suction Cup; 318. Third Negative Pressure Suction Cup; 4. AI Re-inspection Positioning and High-Magnification Imaging Module; 401. A-side Re-inspection Component; 402. B-side Re-inspection Component; 403. Second X-axis Linear Module; 404. Second Z-axis Linear Module; 405. First Connecting Seat; 406. Fourth Negative Pressure Suction Cup; 407. Second Y-axis Linear Module; 408. Second Connecting Seat; 409. First High-Definition Camera; 410. Third X-axis Linear Module; 411. Third Z-axis Linear Module; 412. Second High-Definition Camera; 413. Third Y-axis Linear Module; 414. Fifth Negative Pressure Suction Cup; 5. Temporary storage module; 501. Fourth gripper; 502. Temporary storage bin; 6. Sorting execution module; 601. Fifth gripper; 602. Cleaning bin; 603. Qualified bin; 604. Scrap bin; 7. X-axis three-movement sub-linear module; 8. AI processing module; 801. Multi-scale feature extraction sub-module; 802. Defect region enhancement sub-module; 803. Contrastive learning feature encoding sub-module; 804. Authentic and fake defect determination sub-module; 9. Manual judgment module. Detailed Implementation
[0026] The following will describe in detail the implementation of this application with reference to the accompanying drawings and embodiments, so that the implementation process of how this application uses technical means to solve technical problems and achieve technical effects can be fully understood and implemented accordingly.
[0027] Example 1 Please refer to Figures 1 to 14 As shown, an embodiment of the present invention provides an integrated IC carrier board appearance inspection and AI re-inspection device based on multiple images, including a frame 1, on which a feeding module 2, an appearance inspection module 3, an AI re-inspection positioning and high-magnification imaging module 4, a temporary storage module 5, a sorting execution module 6, a manual judgment module 9, and a main controller are arranged. The loading module 2 is used to transfer the IC carrier board to be inspected to the appearance inspection module 3; The appearance inspection module 3 is used to acquire images of the A and B sides of the IC substrate, obtain multiple image data, and output the location of suspected defects on the IC substrate based on the multiple image data. The AI re-inspection positioning and high-magnification imaging module 4 includes a high-magnification acquisition imaging module and an AI processing module 8. The high-magnification acquisition imaging module is used to acquire high-magnification defect images of the corresponding surfaces of the IC carrier board based on the suspected defect marking positions. The AI processing module 8 is used to determine whether the high-magnification defect images are true or false defects and output high-confidence false defects, high-confidence true defects, and low-confidence defects. The temporary storage module 5 is used to receive and temporarily store all IC carrier boards that have been inspected by the AI re-inspection and positioning and high-magnification imaging module 4, and maintains the temporary storage state before all defects of the IC carrier board are judged by AI and low-confidence defects are judged manually. The manual judgment module 9 is used to receive the low-confidence defects output by the AI processing module 8, retrieve the high-magnification defect image corresponding to the low-confidence defect, and output the final manual judgment result. After all defects on the IC carrier board have been judged, the sorting execution module 6 sorts the IC carrier boards in the temporary storage module 5 to different bins based on the high-confidence true defects and high-confidence false defects output by the AI processing module 8 and the final judgment result output by the manual judgment module 9. The feeding module 2, appearance inspection module 3, AI re-inspection positioning and high-magnification imaging module 4, temporary storage module 5, manual judgment module 9, and sorting execution module 6 are all connected to the main controller.
[0028] In this embodiment, the AI processing module 8 includes a defect region enhancement submodule 802, a multi-scale feature extraction submodule 801, a contrastive learning feature encoding submodule 803, and a true / false defect determination submodule 804. The defect region enhancement submodule 802 is connected to the high-magnification acquisition imaging module for receiving high-magnification defect images and performing local contrast enhancement, edge response enhancement, and high-frequency texture enhancement on candidate defect regions in the images. The multi-scale feature extraction submodule 801 is data-connected to the defect region enhancement submodule 802 and is used to receive the enhanced defect image. The multi-scale feature extraction submodule 801 includes multiple feature branches, each of which has a different receptive field. The multiple feature branches include at least a first feature branch, a second feature branch, and a third feature branch. The receptive field of the first feature branch is smaller than that of the second feature branch, and the receptive field of the second feature branch is smaller than that of the third feature branch. The contrastive learning feature encoding submodule 803 is connected to the multi-scale feature extraction submodule 801 to encode the features output by the multi-scale feature extraction submodule 801. During training, it uses defect samples of the same type as positive sample pairs and defect samples of different types or with different true and false attributes as negative sample pairs, and trains through feature distance constraints. The true / false defect determination submodule 804 is data-connected to the contrastive learning feature encoding submodule 803, and is used to output high-confidence fake defects, high-confidence true defects, and low-confidence defects based on the features output by the contrastive learning feature encoding submodule 803.
[0029] By adopting the above technical solution, the defect region enhancement submodule 802 first performs pixel-level preprocessing on the original high-magnification defect image to suppress the interference of copper foil texture, illumination fluctuation, resin texture and normal process texture, and highlight the contrast and edge features of the defect region; the multi-scale feature extraction submodule 801 uses multiple feature branches with different receptive fields to extract features of different scales respectively. The first feature branch has a receptive field of 8×8, which corresponds to capturing micro-defects such as micron-level cracks and small foreign objects; the second feature branch has a receptive field of 32×32, which corresponds to capturing medium-sized defects such as broken wires and local contamination; the third feature branch has a receptive field of 128×128, which corresponds to capturing large-sized defects such as large-area contamination and edge chipping. Multi-scale feature fusion is achieved through feature stitching, while retaining the key feature information of defects of different sizes. The contrastive learning feature encoding submodule 803 encodes the fused features. During the training phase, a triplet loss function is used, with samples of the same type and attribute of defects as positive sample pairs and samples of different types or different true and false attributes as negative sample pairs. Through feature distance constraints, features of the same type of defects are closer in the feature space, and features of different types of defects are farther apart, thereby improving the separability of true and false defects. The true / false defect determination submodule 804 outputs defect classification results and confidence scores based on the encoded features. The determination submodule outputs background / no defects and probability maps of each defect category. If a candidate ROI meets the high confidence condition for the background, it is judged as false; otherwise, the category probability map is segmented, size parameters are extracted, and compared with the customer's preset specifications. Defects that exceed the specifications and meet the determination conditions are indeed true defects. Defects that cannot be confirmed as true and cannot be judged as false with high confidence are marked as Unknown. A confidence threshold is set: a score ≥ 0.92 is a high confidence true defect, a score ≤ 0.08 is a high confidence false defect, and 0.08 < score < 0.92 is a low confidence defect, thereby achieving three-level diversion and reducing the amount of manual review.
[0030] In some embodiments, the multi-scale feature extraction submodule 801 may employ a lightweight convolutional neural network MobileNetV2 as the backbone network to reduce the number of model parameters and improve online inference speed.
[0031] The specific implementation of the defect region enhancement submodule 802 is as follows: local contrast enhancement is performed using Limit Contrast Adaptive Histogram Equalization (CLAHE), with the clipLimit parameter set to 0.02 and the grid size set to 8×8; edge response enhancement is performed using the Sobel operator to extract the edge gradients in the horizontal and vertical directions; and high-frequency texture enhancement is performed using Gaussian difference filtering, with σ1 set to 1.0 and σ2 set to 2.0. The difference operation highlights the high-frequency defect texture and suppresses the low-frequency process texture in the background.
[0032] In this embodiment, the appearance inspection module 3 includes an A-side inspection component 301, a B-side inspection component 302, and a transfer and flipping component 303. The A-side inspection component 301 is used to inspect the A-side of the IC carrier, and the B-side inspection component 302 is used to inspect the B-side of the IC carrier. The A-side inspection component 301 and the B-side inspection component 302 are respectively disposed on both sides of the transfer and flipping component 303. The A-side inspection component 301 and the B-side inspection component 302 have the same structure, both including a support frame 304 and a first Y-axis linear module 30. 8. The support frame 304 and the first Y-axis linear module 308 are both fixedly installed on the frame 1. The first Z-axis linear module 305 and the combined light source 307 are fixedly installed on the support frame 304. The line scan camera 306 is fixedly installed on the slide of the first Z-axis linear module 305. The combined light source 307 is located below the lens of the line scan camera 306. The first X-axis linear module 309 is fixedly installed on the slide of the first Y-axis linear module 308. The first negative pressure suction cup 310 is fixedly installed on the slide of the first X-axis linear module 309. The combined light source 307 includes a coaxial light source, a uniform light source, a multi-angle side light source, and an infrared light source.
[0033] By adopting the above technical solution, the A-side detection component 301 scans and images the A-side of the IC carrier, the B-side detection component 302 scans and images the B-side of the IC carrier, and the transfer and flipping component 303 is set between the A-side detection component 301 and the B-side detection component 302. It is used to flip the IC carrier 180 degrees and send it to the B-side detection component 302 after the A-side detection is completed. During the scanning and imaging, the first Z-axis linear module 305 adjusts the focal length of the line scan camera 306, the combined light source 307 flashes sequentially according to the preset timing sequence, the line scan camera 306 scans synchronously, and the first Y-axis linear module 308 and the first X-axis linear module 309 drive the first negative pressure suction cup 310 and the IC carrier on it to move, thereby acquiring images under various lighting conditions such as coaxial light, uniform light, multi-angle side light and infrared light in a single scan.
[0034] Among them, coaxial light source is used to highlight metal reflections, resin or translucent material surface contamination, uniform light source is used to highlight discoloration and foreign objects on metal surfaces, multi-angle side light source is used to highlight protrusions, depressions and scratches, and infrared light source is used to improve the visibility of features under resin or translucent materials.
[0035] The synchronization mechanism between multi-source strobe and line scan imaging is as follows: the line scan camera's line frequency is set to 40kHz, and the carrier plate's moving speed is matched with the camera's line frequency to ensure that the physical size of each pixel is 2.5μm; the four light sources of the combined light source are strobe sequentially in the order of "coaxial-uniform-side-infrared", and the ratio of the strobe frequency to the camera's line frequency is 1:4, that is, a round of switching of the four light sources is completed every 4 lines of images; after the entire image is scanned, it is split and reconstructed into four complete images corresponding to different light sources according to the line order, and the four images are aligned at the pixel level, so as to obtain multiple image data in a single scan.
[0036] In this embodiment, the transplanting and flipping assembly 303 includes a transplanting unit 3031 and a flipping unit 3032. The transplanting unit 3031 includes an X-axis two-moving linear module 311. The X-axis two-moving linear module 311 is fixedly installed on the frame 1 by a first column. A second gripper 312 is installed on one of the moving parts of the X-axis two-moving linear module 311, and a third gripper 313 is installed on the other moving part. The flipping unit 3032 is disposed between the second gripper 312 and the third gripper 313. The flipping unit 3032 includes a support base 314, which is fixedly mounted on the frame 1. A flipping motor 315 and a third negative pressure suction cup 318 are fixedly mounted on the support base 314. A flipping shaft 316 is fixedly connected to the rotating shaft of the flipping motor 315 through a coupling. A second negative pressure suction cup 317 is fixedly mounted on the flipping shaft 316. The second gripper 312 can transfer the IC carrier board on the A-side detection assembly 301 to the second negative pressure suction cup 317. When the flipping shaft 316 rotates, flipping the IC carrier board on the second negative pressure suction cup 317 onto the third negative pressure suction cup 318, the third gripper 313 can transfer the IC carrier board on the third negative pressure suction cup 318 to the first negative pressure suction cup 310 in the B-side detection assembly 302.
[0037] By adopting the above technical solution, after the A-side inspection is completed, the second gripper 312 transfers the IC carrier board from the first negative pressure suction cup 310 to the second negative pressure suction cup 317. The flipping motor 315 drives the second negative pressure suction cup 317 to rotate 180 degrees through the flipping shaft 316, so that the IC carrier board is flipped onto the third negative pressure suction cup 318. The third gripper 313 then transfers the IC carrier board from the third negative pressure suction cup 318 to the first negative pressure suction cup 310 in the B-side inspection assembly 302, so that the B-side of the IC carrier board faces upward.
[0038] The flipping unit 3032 is also equipped with a vacuum pressure detection sensor and an anti-drop protection mechanism: vacuum pressure sensors are connected in series in the air paths of the second negative pressure suction cup 317 and the third negative pressure suction cup 318 to detect the negative pressure value inside the suction cup in real time. When the gauge pressure is higher than the -60kPa safety threshold and the vacuum is insufficient, the main controller triggers a shutdown alarm and stops the flipping; a mechanical limit stop is provided at the end of the flipping shaft 316 to limit the flipping angle to ±185° to prevent the carrier plate from falling off due to over-rotation.
[0039] In this embodiment, the high-magnification acquisition imaging module includes an A-side re-inspection component 401 and a B-side re-inspection component 402; The A-side re-inspection assembly 401 includes a second X-axis linear module 403 and a second Y-axis linear module 407. The second X-axis linear module 403 is fixedly mounted on the frame 1 via a second column. A second Z-axis linear module 404 is fixedly mounted on the slide of the second X-axis linear module 403. A first connecting seat 405 is fixedly mounted on the slide of the second Z-axis linear module 404. A fourth negative pressure suction cup 406 is fixedly mounted on the bottom of the first connecting seat 405. The fourth negative pressure suction cup 406 suctions the B-side of the IC carrier board from above. The first high-definition camera 409 is mounted on the slide of the second Y-axis linear module 407 with the lens facing upwards, and is used to capture the A-side of the IC carrier board facing downwards. The second X-axis linear module 403 moves the product it holds, and the second Y-axis linear module 407 moves the first high-definition camera 409 to complete the high-magnification image capture of the A-side.
[0040] The B-side re-inspection assembly 402 includes a third X-axis linear module 410 and a third Y-axis linear module 413. The third X-axis linear module 410 is fixedly mounted on the frame 1 via a third column. A third Z-axis linear module 411 is fixedly mounted on the slide of the third X-axis linear module 410. A second camera bracket is fixedly mounted on the slide of the third Z-axis linear module 411, and a second high-definition camera 412 is fixedly mounted on the second camera bracket, with the lens of the second high-definition camera 412 facing downwards. The third Y-axis linear module 413 is fixedly mounted on the frame 1, and a third Y-axis linear module 413 is fixedly mounted on the slide. A B-side re-inspection stage is installed, and a fifth negative pressure suction cup 414 is set on the B-side re-inspection stage to fix the IC carrier board with the B-side facing upward. The lens of the second high-definition camera 412 is aimed at the B-side of the IC carrier board on the B-side re-inspection stage. The fourth negative pressure suction cup 406 can pick up the IC carrier board on the B-side detection component 302 and make its B-side be attracted and its A-side face downward. The first high-definition camera 409 captures a high-magnification defect image of the A-side. After the A-side image is captured, the fourth negative pressure suction cup 406 places the IC carrier board on the B-side re-inspection stage, keeping the B-side facing upward, and the second high-definition camera 412 captures a high-magnification defect image of the B-side.
[0041] During the A-side re-inspection, the fourth negative pressure suction cup 406 picks up the IC carrier board from the first negative pressure suction cup 310 of the B-side detection component 302, so that its B-side is adsorbed and its A-side is facing down; the second X-axis linear module 403 and the second Z-axis linear module 404 drive the fourth negative pressure suction cup 406 to move along the X and Z directions, and the second Y-axis linear module 407 drives the first high-definition camera 409, which is facing up, to move along the Y direction. The two work together to make the first high-definition camera 409 sequentially align with each suspected defect mark position on the A-side of the IC carrier board and collect high-magnification defect images of the A-side; During the B-side re-inspection, after the A-side image is acquired, the fourth negative pressure suction cup 406 places the IC carrier board onto the B-side re-inspection platform on the slide of the third Y-axis linear module 413 and fixes it with the fifth negative pressure suction cup 414, keeping the B-side facing upwards; the third X-axis linear module 410 and the third Z-axis linear module 411 drive the second high-definition camera 412 to move along the X and Z directions, and the third Y-axis linear module 413 drives the B-side re-inspection platform to move along the Y direction. The two work together to make the second high-definition camera 412 sequentially align with the suspected defect mark positions on the B-side of the IC carrier board and acquire high-magnification defect images of the B-side.
[0042] Among them, the first high-definition camera 409 and the second high-definition camera 412 have a resolution of 0.8 micrometers and use metallographic objectives with a 20x long working distance to achieve clear imaging of micrometer-level defects.
[0043] The coordinate calibration method for the initial inspection and re-inspection systems is as follows: a nine-point calibration method is adopted, in which nine marker points are set in a 3×3 array on the standard calibration plate, with a physical spacing of 10mm between the marker points. First, the line scan camera of the appearance inspection module acquires the image of the calibration plate and obtains the image pixel coordinates of the nine marker points. Then, the calibration plate is placed on the B-side re-inspection stage, and the A-side is facing down to be photographed by the first high-definition camera, simulating the suction state of the fourth negative pressure suction cup. The high-definition camera acquires the image of each marker point and records the encoder coordinates of the corresponding motion axis. The least squares method is used to fit and establish the mapping matrix between the pixel coordinate system of the initial inspection image and the physical coordinate system of the re-inspection motion, and the backlash error of the linear module is compensated with a compensation accuracy of ±1μm to ensure that the high-magnification camera can accurately align with the defect position of the initial inspection mark.
[0044] In this embodiment, the temporary storage module 5 includes a fourth gripper 501 and a temporary storage compartment 502. The fourth gripper 501 can move the IC carrier board on the B-side re-inspection platform into the temporary storage compartment 502. The temporary storage compartment 502 is pulled out and installed on the rack 1.
[0045] By adopting the above technical solutions, the high-speed AI re-inspection process is decoupled from the relatively slow manual final judgment and sorting process, avoiding equipment downtime or product backlog caused by delays in manual judgment.
[0046] The temporary storage compartment 502 adopts a 10×4 matrix multi-station layout with a total of 40 storage stations. Each station is equipped with a unique number and a position sensor. Each IC carrier board has a unique QR code printed on its corner. The loading station is equipped with a fixed barcode scanner to read the carrier board ID and bind it with the defect data. When the fourth gripper 501 puts the carrier board into the temporary storage station, the main controller automatically binds and stores the carrier board ID with the station number. After the manual final judgment is completed, the main controller quickly indexes the corresponding station according to the carrier board ID and controls the fifth gripper 601 to accurately pick up the material, realizing orderly temporary storage and rapid addressing.
[0047] In this embodiment, the sorting execution module 6 includes a fifth gripper 601, a cleaning chamber 602, a qualified chamber 603, and a scrap chamber 604. The cleaning chamber 602, the qualified chamber 603, and the scrap chamber 604 are all pulled out and installed on the rack 1. The fifth gripper 601 can transfer the IC carrier board to the cleaning chamber 602, the qualified chamber 603, or the scrap chamber 604 according to the final judgment result.
[0048] By adopting the above technical solution, defects can be automatically sorted, eliminating the need for manual sorting and improving the efficiency of subsequent processing.
[0049] The IC carrier board in the cleaning chamber 602 can be put back into the loading chamber 201 for a second inspection after plasma cleaning or wet cleaning. After the defects are confirmed to be removed, it is put into the qualified chamber, forming a complete process closed loop. The defect judgment and sorting rules are configured according to product requirements and customer preset specifications.
[0050] In this embodiment, the feeding module 2 includes a feeding bin 201 and a first gripper 202. The feeding bin 201 is pulled out and installed on the frame 1. The first gripper 202 can transfer the IC carrier board of the feeding bin 201 to the first negative pressure suction cup 310 in the A-side detection component 301.
[0051] By adopting the above technical solution, the feeding bin 201 is installed on the frame 1 in a pull-out manner. The first gripper 202 transfers the IC carrier board of the feeding bin 201 to the first negative pressure suction cup 310 in the A-side detection component 301, thereby automatically transferring the IC carrier board to be tested to the A-side detection component 301, reducing manual feeding operations, and improving feeding efficiency and positioning consistency.
[0052] In this embodiment, an X-axis three-moving linear module 7 is fixedly installed on the frame 1 by a fourth column, and the first gripper 202, the fourth gripper 501 and the fifth gripper 601 are respectively fixed on the corresponding moving parts in the X-axis three-moving linear module 7.
[0053] By adopting the above technical solution, an X-axis three-moving linear module 7 is fixedly installed on the frame 1 via a fourth column. The first gripper 202, the fourth gripper 501, and the fifth gripper 601 are respectively fixed on the corresponding moving parts in the X-axis three-moving linear module 7, thereby simplifying the equipment structure, reducing the space occupied, and improving the motion coordination between the grippers.
[0054] The Y-axis alignment of each gripper with each workstation is achieved through the Y-axis adjustment mechanism at the workstation end: the loading workstation, temporary storage workstation, and sorting workstation are all equipped with Y-axis fine-tuning platforms. During the debugging phase, the fine-tuning platforms are used to align the Y-axis center of each workstation with the motion path of the X-axis three-movement linear module 7, ensuring that the gripper can accurately pick up and place the carrier plate.
[0055] It should be noted that the first gripper 202, the second gripper 312, the third gripper 313, the fourth gripper 501, and the fifth gripper 601 all include a fourth Z-axis linear module. A suction cup holder is fixedly mounted on the slide of the fourth Z-axis linear module, and multiple suction cups are fixedly mounted on the suction cup holder. The fourth Z-axis linear module is used to drive the suction cup holder and multiple suction cups to move vertically up and down, so that the suction cups approach or move away from the IC carrier. The multiple suction cups are used to adsorb the IC carrier under the action of a negative pressure source, thereby realizing the picking, placing, and transferring of the IC carrier.
[0056] In this embodiment, the main controller is configured to perform the following steps: S1. Loading materials: The main controller controls the first gripper 202 to take out the IC carrier board to be tested from the loading bin 201 and transfer the IC carrier board to the first negative pressure suction cup 310 in the A-side detection component 301; the first negative pressure suction cup 310 adsorbs and fixes the IC carrier board so that the A-side of the IC carrier board faces upward. S2, A-side multiple image scanning: The main controller controls the first Y-axis linear module 308 and the first X-axis linear module 309 to move the first negative pressure suction cup 310 and the IC carrier board on it to below the line scan camera 306; the main controller controls the first Z-axis linear module 305 to adjust the focal length of the line scan camera 306; the main controller controls the combined light source 307 to flash sequentially according to a preset timing sequence, and synchronously triggers the line scan camera 306 to scan and image the A-side of the IC carrier board, obtaining multiple image data of the A-side; the appearance inspection module 3 outputs the suspected defect marking position of the A-side of the IC carrier board based on the multiple image data of the A-side; S3, Flip: The main controller controls the second gripper 312 to transfer the IC carrier board on the A-side detection assembly 301 from the first negative pressure suction cup 310 to the second negative pressure suction cup 317 in the flipping unit 3032; the main controller controls the flipping motor 315 to drive the flipping shaft 316 to rotate 180°, so that the IC carrier board on the second negative pressure suction cup 317 is flipped to the third negative pressure suction cup 318; the main controller controls the third gripper 313 to transfer the IC carrier board on the third negative pressure suction cup 318 to the first negative pressure suction cup 310 in the B-side detection assembly 302, so that the B-side of the IC carrier board is facing upward; S4, B-side multiple image scanning: The main controller controls the first Y-axis linear module 308 and the first X-axis linear module 309 in the B-side inspection assembly 302 to move the first negative pressure suction cup 310 and the IC carrier board on it in the B-side inspection assembly 302 to below the line scan camera 306 in the B-side inspection assembly 302; the main controller controls the first Z-axis linear module 305 in the B-side inspection assembly 302 to adjust the focal length; the main controller controls the combined light source 307 in the B-side inspection assembly 302 to flash sequentially according to a preset timing sequence, and synchronously triggers the line scan camera 306 in the B-side inspection assembly 302 to scan and image the B-side of the IC carrier board, obtaining multiple image data of the B-side; the appearance inspection module 3 outputs the suspected defect marking position of the B-side of the IC carrier board based on the multiple image data of the B-side; S5, AI-powered re-examination positioning and high-magnification imaging: S51, High-magnification acquisition on side A: The main controller controls the second X-axis linear module 403 and the second Z-axis linear module 404 to move the fourth negative pressure suction cup 406 above the first negative pressure suction cup 310 of the B-side detection component 302. The main controller controls the negative pressure source to make the fourth negative pressure suction cup 406 pick up the B-side of the IC carrier board from the B-side initial inspection station, so that the A-side is facing down. According to the position of the suspected defect mark on the A-side, the main controller controls the second X-axis linear module 403 to move the product it is holding, and at the same time controls the second Y-axis linear module 407 to move the first high-definition camera 409 facing up, so that the first high-definition camera 409 is aligned with each suspected defect mark position on the A-side of the IC carrier board in sequence to collect high-magnification defect images of the A-side. The second Z-axis linear module 404 is used for product picking and height adjustment.
[0057] S52, B-side high-magnification acquisition: After the A-side image is acquired, the main controller controls the fourth negative pressure suction cup 406 to place the IC carrier board onto the B-side re-inspection stage on the slide of the third Y-axis linear module 413 and fix it with the fifth negative pressure suction cup 414, keeping the B-side facing upwards. According to the position of the suspected defect mark on the B-side, the main controller controls the third X-axis linear module 410 and the third Z-axis linear module 411 to drive the downward-facing second high-definition camera 412 to move. At the same time, the third Y-axis linear module 413 drives the B-side re-inspection stage to move, so that the second high-definition camera 412 is aligned with each suspected defect mark position on the B-side of the IC carrier board in sequence, and high-magnification defect images of the B-side are acquired.
[0058] S6, AI authenticity defect determination: AI processing module 8 receives high-magnification defect images from side B and side A; defect region enhancement submodule 802 first performs local contrast enhancement, edge response enhancement, and high-frequency texture enhancement on candidate defect regions in the high-magnification defect images; multi-scale feature extraction submodule 801 extracts first-scale features, second-scale features, and third-scale features from the enhanced images and fuses them; contrastive learning feature encoding submodule 803 encodes the fused features; and true / false defect determination submodule 804 outputs high-confidence fake defects, high-confidence true defects, and low-confidence defects based on the encoded features. S7, Temporary Storage: The main controller controls the fourth gripper 501 to move the IC carrier board on the B-side re-inspection platform to the temporary storage compartment 502; the temporary storage compartment 502 receives and temporarily stores all IC carrier boards that have been inspected by the AI re-inspection positioning and high-magnification imaging module 4, and maintains the temporary storage state until all defects of the IC carrier board are judged by AI and low-confidence defects are judged manually. S8. Sorting: The main controller checks whether all defects on the same IC carrier board in the temporary storage compartment 502 have been determined. After the determination is completed, the main controller controls the fifth gripper 601 to determine the final confirmed true defect data of the IC carrier board based on the high-confidence true defects and high-confidence false defects output by the AI processing module 8 and the final judgment result output by the manual judgment module 9, and performs the following sorting actions: If the IC carrier is full of high-confidence defects, the high-confidence defects are directly filtered out and not included in the sorting defect statistics. The fifth gripper 601 is then controlled to transfer the IC carrier to the qualified warehouse 603. If at least one high-confidence true defect and / or low-confidence defect exists on the IC carrier, the following sorting action shall be performed on the IC carrier: Specifically, high-confidence true defects on the IC carrier are directly included in the final confirmed true defect data of the IC carrier; low-confidence defects on the IC carrier are subject to final manual judgment by the manual judgment module 9. If the final manual judgment confirms that the defect is a true defect, it is included in the final confirmed true defect data of the IC carrier; if the final manual judgment confirms that the defect is a false defect, it is directly filtered out and not included in the sorting defect statistics; high-confidence false defects on the IC carrier are directly filtered out and not included in the sorting defect statistics. The defects in the final confirmed true defect data of the IC substrate are divided into fatal defects and repairable defects. Fatal defects include at least one of board damage, short circuit, open circuit, and missing parts, while repairable defects include at least one of foreign matter and dirt. If a fatal defect is found in the final confirmed true defect data of the IC carrier, the fifth gripper 601 is controlled to transfer the IC carrier to the scrap bin 604. If the final confirmed true defect data of the IC carrier board only contains repairable defects, then control the fifth gripper 601 to transfer the IC carrier board to the cleaning chamber 602. If the final confirmed true defect data of the IC carrier is empty, the fifth gripper 601 will transfer the IC carrier to the qualified warehouse 603.
[0059] The interaction and synchronization logic of the manual judgment module 9 is as follows: The manual judgment terminal adopts a web-based interactive interface, which automatically pushes a list of low-confidence defects. Each defect displays a high-magnification image, defect location, defect type, and confidence score. Operators can complete the judgment of a single defect with one click by using the "True Defect" or "False Defect" buttons. When all low-confidence defects on the same IC carrier board have been judged, the terminal automatically sends a "Judgment Completed" signal to the main controller, triggering the subsequent sorting process. The terminal supports batch operations and historical record backtracking, and the average judgment time for a single defect is ≤2 seconds.
[0060] In this embodiment, the main controller can also associate the confirmed genuine defect data with the product ID and upload it to the MES manufacturing execution system. The uploaded data includes product identifier, face, defect number, defect physical coordinates, defect category, defect size parameters, judgment source and judgment time. If the upload fails, the corresponding product identifier and defect data are added to the retransmission queue and uploaded again after communication is restored, so as to realize the data closure and quality traceability of the whole process.
[0061] Example 2: AI Online Incremental Learning Mechanism To further enhance the detection stability and intelligence of this equipment in mass production environments, this embodiment introduces an online incremental learning mechanism in the AI processing module 8. When the manual judgment module 9 completes a final judgment on a low-confidence defect and confirms it as a true or false defect, the final judgment result and its corresponding high-magnification defect image are automatically stored as newly labeled samples in the sample cache of the AI processing module 8. When the cumulative number of newly labeled samples of the same type of defect reaches a preset threshold (e.g., 50), the AI processing module 8 automatically triggers an incremental training process. The newly labeled samples and existing training samples are used to fine-tune the parameters of the comparative learning feature encoding submodule 803 and the true / false defect judgment submodule 804. During the fine-tuning process, a low learning rate and experience playback strategy are adopted to prevent the model's ability to recognize historical defect types from degrading. Through this online incremental learning mechanism, the equipment can continuously absorb manual final judgment experience during continuous production, gradually improving the accuracy of identifying defect morphologies specific to certain production lines and reducing the frequency of subsequent manual re-judgments of similar defects.
[0062] Example 3: Attention-Enhanced Multi-Scale Feature Extraction Architecture In this embodiment, the multi-scale feature extraction submodule 801 adopts an architecture combining a lightweight convolutional neural network and an attention mechanism. Specifically, after the first, second, and third feature branches output feature maps of different scales, the channel weights of each scale feature map are adaptively recalibrated through a channel attention module, and the spatial position of the candidate defect region is weighted and enhanced through a spatial attention module. Finally, the weighted feature maps of the three scales are fused element-wise. This architecture, while maintaining a low number of parameters (no more than 5M), can effectively enhance the feature saliency of micron-level fine defects in complex background textures and improve the ability of the multi-scale feature extraction submodule 801 to suppress interference such as copper foil texture and resin texture on the IC substrate surface.
[0063] Example 4: Defect Data Closed-Loop Traceability and Process Feedback Mechanism In this embodiment, the main controller of the equipment is also equipped with a defect data closed-loop traceability and process feedback module. The main controller associates the final confirmed true defect data of each IC carrier board with the product identifier and uploads it to an external data system. The uploaded data includes the product identifier, face value, defect number, defect physical coordinates, defect category, defect size parameters, judgment source (AI high-confidence or manual final judgment), and judgment time. The external data system performs statistical analysis on the uploaded defect data. When the frequency of a certain type of defect within a preset time window exceeds a preset threshold, or the defect clustering degree within a certain physical coordinate area exceeds a preset threshold, it automatically generates process anomaly warning information and feeds it back to the production line management system. This closed-loop traceability mechanism allows the detection data not only to be used for sorting individual products but also to support the optimization and adjustment of upstream process parameters, forming a complete data closed loop of detection-sorting-process improvement.
[0064] Example 5: RFID slot positioning structure for temporary storage to prevent mixing of materials In this embodiment, the temporary storage compartment 502 of the temporary storage module 5 is equipped with multiple independent partitions. Each partition has a positioning slot that matches the shape of the IC carrier board, and an RFID reading antenna is embedded in the positioning slot. When each IC carrier board is placed in the temporary storage compartment 502, its product identification is automatically read by the RFID reading antenna and bound to the slot number of the positioning slot, which is stored in the main controller. After the manual judgment module 9 completes the final judgment of low-confidence defects, the main controller queries the slot number of the corresponding IC carrier board in the temporary storage compartment 502 according to the product identification, and controls the fifth gripper 601 to accurately remove the IC carrier board from the slot and perform the sorting action, avoiding material picking errors caused by mixing multiple IC carrier boards. This slot binding mechanism is particularly necessary when the capacity of the temporary storage compartment 502 is large, ensuring that product misalignment does not occur when the high-speed AI re-inspection process and the low-speed manual final judgment process are decoupled.
[0065] Example 6: Adsorption Posture Pressure Detection Mechanism In this embodiment, pressure-sensing films are evenly distributed on the suction cup surfaces of the fourth negative pressure suction cup 406 and the positioning suction cup of the B-side re-inspection stage. After the IC carrier is adsorbed and fixed, the pressure-sensing films detect the pressure distribution in each area of the suction cup and transmit the pressure distribution signal to the main controller. The main controller determines whether the IC carrier is warped or misaligned based on the pressure distribution: if the pressure distribution unevenness exceeds a preset threshold (e.g., the difference between the maximum and minimum pressure exceeds 30% of the average pressure), it determines that the adsorption posture of the IC carrier is abnormal, controls the negative pressure source to adjust the adsorption force, or controls the corresponding linear module to re-pick up and place the IC carrier. This pressure distribution detection mechanism can effectively avoid high-magnification image defocusing or defect mark position shifting caused by poor adsorption posture, improving the imaging quality of the high-magnification acquisition imaging module 4 and the accuracy of AI genuine and fake defect determination.
[0066] Example 7: Light Source-Scan Phase-Locking Imaging Mechanism In this embodiment, a phase-locked relationship is established between the strobe timing of the combined light source 307 and the scanning line frequency of the line scan camera 306. The main controller controls the combined light source 307 to strobe sequentially in the order of "coaxial light source → uniform light source → multi-angle side light source → infrared light source". The strobe period of each light source is strictly synchronized with the time it takes for the line scan camera 306 to scan a complete line of pixels, so that the line scan camera 306 receives illumination from only a single light source within the exposure window of each line of pixels. This allows for the acquisition of multiple image data under four strictly registered lighting conditions in a single scan, avoiding image misalignment or aliasing caused by asynchrony between light source switching and scanning line frequency. This phase-locking mechanism achieves true single-scan multiple image acquisition without the need for additional beam splitters or multiple scanning processes.
[0067] Example 8: Switchable Polarizer High-Magnification Imaging Mechanism In this embodiment, a switchable polarizer module is provided in front of the lenses of both the first high-definition camera 409 and the second high-definition camera 412. The polarizer module includes a linear polarizer and a circular polarizer, which can be automatically switched according to the defect type under the control of the main controller. For reflective foreign objects on the metal surface, a linear polarizer is used to suppress specular reflection and highlight the contrast between the foreign object and the metal substrate; for defects under resin or translucent materials, a circular polarizer is used to eliminate the interference of surface reflected light and enhance the visibility of deep features. This automatic polarizer switching mechanism further improves the signal-to-noise ratio of high-magnification defect images, which helps the AI processing module 8 to accurately determine low-contrast defects.
[0068] The foregoing description illustrates and describes several preferred embodiments of the present invention. However, as previously stated, it should be understood that the present invention is not limited to the forms disclosed herein and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be altered within the scope of the inventive concept by means of the foregoing teachings or techniques or knowledge in related fields. Any modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims.
Claims
1. A multi-image integrated device for IC substrate appearance inspection and AI re-inspection, characterized in that, Includes a frame (1), on which are provided a feeding module (2), an appearance inspection module (3), an AI re-inspection positioning and high-magnification imaging module (4), a temporary storage module (5), a sorting execution module (6), a manual judgment module (9) and a main controller; The loading module (2) is used to transfer the IC carrier board to be inspected to the appearance inspection module (3). The appearance inspection module (3) is used to acquire images of the A and B sides of the IC carrier board, obtain multiple image data, and output the location of suspected defect marks on the IC carrier board based on the multiple image data. The AI re-inspection positioning and high-magnification imaging module (4) includes a high-magnification acquisition imaging module and an AI processing module (8). The high-magnification acquisition imaging module is used to acquire high-magnification defect images of the corresponding surface of the IC carrier board according to the suspected defect mark position. The AI processing module (8) is used to determine the authenticity of the high-magnification defect images and output high-confidence fake defects, high-confidence true defects, and low-confidence defects. The temporary storage module (5) is used to receive and temporarily store all IC carrier boards that have been detected by the AI re-inspection positioning and high-magnification imaging module (4), and maintain the temporary storage state before all defects of the IC carrier board are judged by AI and low-confidence defects are judged manually. The manual judgment module (9) is used to receive the low confidence defect output by the AI processing module (8), retrieve the high magnification defect image corresponding to the low confidence defect, and output the final manual judgment result. The sorting execution module (6) is used to sort the IC carrier boards in the temporary storage module (5) to different bins after all defects on the IC carrier board have been judged, based on the high confidence true defects and high confidence false defects output by the AI processing module (8) and the final judgment result output by the manual judgment module (9). The feeding module (2), the appearance inspection module (3), the AI re-inspection positioning and high-magnification imaging module (4), the temporary storage module (5), the manual judgment module (9), and the sorting execution module (6) are all connected to the main controller.
2. The IC substrate appearance inspection and AI re-inspection integrated device based on multiple images according to claim 1, characterized in that: The AI processing module (8) includes a defect region enhancement submodule (802), a multi-scale feature extraction submodule (801), a contrast learning feature encoding submodule (803), and a true / false defect determination submodule (804); the defect region enhancement submodule (802) is data-connected to the high-magnification acquisition imaging module and is used to receive the high-magnification defect image and perform local contrast enhancement, edge response enhancement, and high-frequency texture enhancement on the candidate defect regions in the image; The multi-scale feature extraction submodule (801) is data-connected to the defect region enhancement submodule (802) and is used to receive the enhanced defect image. The multi-scale feature extraction submodule (801) includes multiple feature branches, each of which has a different receptive field. The multiple feature branches include at least a first feature branch, a second feature branch, and a third feature branch. The receptive field of the first feature branch is smaller than that of the second feature branch, and the receptive field of the second feature branch is smaller than that of the third feature branch. The contrastive learning feature encoding submodule (803) is data-connected to the multi-scale feature extraction submodule (801) and is used to encode the features output by the multi-scale feature extraction submodule (801). During training, it uses defect samples of the same type as positive sample pairs and defect samples of different types or with different true and false attributes as negative sample pairs, and trains through feature distance constraints. The true / false defect determination submodule (804) is data-connected to the contrastive learning feature encoding submodule (803) and is used to output the high-confidence fake defect, the high-confidence true defect, and the low-confidence defect based on the features output by the contrastive learning feature encoding submodule (803).
3. The IC substrate appearance inspection and AI re-inspection integrated device based on multiple images according to claim 1, characterized in that: The appearance inspection module (3) includes an A-side inspection component (301), a B-side inspection component (302), and a transplanting and flipping component (303). The A-side inspection component (301) is used to inspect the A-side of the IC carrier board, and the B-side inspection component (302) is used to inspect the B-side of the IC carrier board. The A-side inspection component (301) and the B-side inspection component (302) are respectively disposed on both sides of the transplanting and flipping component (303). The A-side inspection component (301) and the B-side inspection component (302) have the same structure, both including a support frame (304) and a first Y-axis linear module (308). The support frame (304) and the first Y-axis linear module (308) are both fixedly installed on the frame (1). The support frame (304) is fixedly installed with the first Z-axis linear module (305) and the combined light source (307). The first Z-axis linear module (305) is fixedly installed on the slide table of the first Z-axis linear module (305). The combined light source (307) is located below the lens of the line scanning camera (306). The first X-axis linear module (309) is fixedly installed on the slide table of the first Y-axis linear module (308). The first negative pressure suction cup (310) is fixedly installed on the slide table of the first X-axis linear module (309). The combined light source (307) includes a coaxial light source, a uniform light source, a multi-angle side light source, and an infrared light source.
4. The IC substrate appearance inspection and AI re-inspection integrated device based on multiple images according to claim 3, characterized in that: The transplanting and turning assembly (303) includes a transplanting unit (3031) and a turning unit (3032). The transplanting unit (3031) includes an X-axis two-moving linear module (311). The X-axis two-moving linear module (311) is fixedly installed on the frame (1) by a first column. A second gripper (312) is installed on one of the moving parts of the X-axis two-moving linear module (311), and a third gripper (313) is installed on the other moving part. The flipping unit (3032) is disposed between the second gripper (312) and the third gripper (313). The flipping unit (3032) includes a support base (314), which is fixedly mounted on the frame (1). A flipping motor (315) and a third negative pressure suction cup (318) are fixedly mounted on the support base (314). A flipping shaft (316) is fixedly connected to the rotating shaft of the flipping motor (315) through a coupling. A flipping shaft (316) is fixedly mounted on the flipping shaft (316). Equipped with a second negative pressure suction cup (317), the second gripper (312) can transfer the IC carrier board on the A-side detection assembly (301) to the second negative pressure suction cup (317). When the flipping shaft (316) rotates, flipping the IC carrier board on the second negative pressure suction cup (317) onto the third negative pressure suction cup (318), the third gripper (313) can transfer the IC carrier board on the third negative pressure suction cup (318) onto the first negative pressure suction cup (310) in the B-side detection assembly (302).
5. The IC substrate appearance inspection and AI re-inspection integrated device based on multiple images according to claim 4, characterized in that: The high-magnification acquisition and imaging module includes an A-side re-examination component (401) and a B-side re-examination component (402). The A-side re-inspection component (401) includes a second X-axis linear module (403) and a second Y-axis linear module (407). The second X-axis linear module (403) is fixedly mounted on the frame (1) via a second column. A second Z-axis linear module (404) is fixedly mounted on the slide of the second X-axis linear module (403). A first connecting seat (405) is fixedly mounted on the slide of the second Z-axis linear module (404). A fourth negative pressure suction cup (406) is fixedly mounted at the bottom of the first connecting seat (405). The fourth negative pressure suction cup (406) is used to suction the IC from above. The B side of the carrier board is positioned so that the A side faces downwards; the second Y-axis linear module (407) is fixedly mounted on the frame (1) and located at the work station below the fourth negative pressure suction cup (406). A first high-definition camera (409) is mounted on the slide of the second Y-axis linear module (407) via a second connecting seat (408). The lens of the first high-definition camera (409) faces upwards and is used to photograph the A side of the IC carrier board facing downwards; the second X-axis linear module (403) moves the product it holds, and the second Y-axis linear module (407) moves the first high-definition camera (409) to jointly complete the high-magnification imaging of the A side; The B-side re-inspection assembly (402) includes a third X-axis linear module (410) and a third Y-axis linear module (413). The third X-axis linear module (410) is fixedly mounted on the frame (1) via a third column. A third Z-axis linear module (411) is fixedly mounted on the slide of the third X-axis linear module (410). A second camera bracket is fixedly mounted on the slide of the third Z-axis linear module (411). A second high-definition camera (412) is fixedly mounted on the second camera bracket, with the lens of the second high-definition camera (412) facing downwards. The third Y-axis linear module (413) is fixedly mounted on the frame (1). A B-side re-inspection stage is fixedly installed on the slide of 413). A fifth negative pressure suction cup (414) is provided on the B-side re-inspection stage to fix the IC carrier board and make the B-side face up. The lens of the second high-definition camera (412) is aimed at the B-side of the IC carrier board on the B-side re-inspection stage. The fourth negative pressure suction cup (406) can pick up the IC carrier board on the B-side detection component (302) and make its B-side be adsorbed and its A-side face down. The first high-definition camera (409) collects the high-magnification defect image of the A-side. After the A-side image is acquired, the fourth negative pressure suction cup (406) places the IC carrier board on the B-side re-inspection stage, keeping the B-side face up, and the second high-definition camera (412) collects the high-magnification defect image of the B-side.
6. The IC substrate appearance inspection and AI re-inspection integrated device based on multiple images according to claim 5, characterized in that: The temporary storage module (5) includes a fourth gripper (501) and a temporary storage compartment (502). The fourth gripper (501) can move the IC carrier board on the B-side re-inspection platform into the temporary storage compartment (502). The temporary storage compartment (502) is installed on the rack (1) in a pull-out manner.
7. The IC substrate appearance inspection and AI re-inspection integrated device based on multiple images according to claim 6, characterized in that: The sorting execution module (6) includes a fifth gripper (601), a cleaning chamber (602), a qualified chamber (603), and a scrap chamber (604). The cleaning chamber (602), the qualified chamber (603), and the scrap chamber (604) are all installed in a pull-out manner on the rack (1). The fifth gripper (601) can transfer the IC carrier board to the cleaning chamber (602), the qualified chamber (603), or the scrap chamber (604) according to the final judgment result.
8. The IC substrate appearance inspection and AI re-inspection integrated device based on multiple images according to claim 7, characterized in that: The feeding module (2) includes a feeding bin (201) and a first gripper (202). The feeding bin (201) is installed on the frame (1) in a pull-out manner. The first gripper (202) can transfer the IC carrier board of the feeding bin (201) to the first negative pressure suction cup (310) in the A-side detection component (301).
9. The IC substrate appearance inspection and AI re-inspection integrated device based on multiple images according to claim 8, characterized in that: The frame (1) is fixedly mounted with an X-axis three-moving linear module (7) via a fourth column. The first gripper (202), the fourth gripper (501) and the fifth gripper (601) are respectively fixed on the corresponding moving parts in the X-axis three-moving linear module (7).
10. A multi-image integrated IC substrate appearance inspection and AI re-inspection device according to any one of claims 1 to 9, characterized in that: The main controller is configured to perform the following steps: S1. Loading materials: The main controller controls the first gripper (202) to take out the IC carrier board to be tested from the loading bin (201) and transfer the IC carrier board to the first negative pressure suction cup (310) in the A-side detection component (301); the first negative pressure suction cup (310) adsorbs and fixes the IC carrier board so that the A-side of the IC carrier board faces upward; S2, A-side multiple image scanning: The main controller controls the first Y-axis linear module (308) and the first X-axis linear module (309) to move the first negative pressure suction cup (310) and the IC carrier board on it to the underside of the line scan camera (306); the main controller controls the first Z-axis linear module (305) to adjust the focal length of the line scan camera (306); the main controller controls the combined light source (307) to flash sequentially according to the preset timing sequence, and synchronously triggers the line scan camera (306) to scan and image the A-side of the IC carrier board to obtain multiple image data of the A-side; the appearance inspection module (3) outputs the suspected defect marking position of the A-side of the IC carrier board based on the multiple image data of the A-side; S3, Flip: The main controller controls the second gripper (312) to transfer the IC carrier board on the A-side detection assembly (301) from the first negative pressure suction cup (310) to the second negative pressure suction cup (317) in the flipping unit (3032); the main controller controls the flipping motor (315) to drive the flipping shaft (316) to rotate 180°, so that the IC carrier board on the second negative pressure suction cup (317) is flipped to the third negative pressure suction cup (318); the main controller controls the third gripper (313) to transfer the IC carrier board on the third negative pressure suction cup (318) to the first negative pressure suction cup (310) in the B-side detection assembly (302), so that the B-side of the IC carrier board is facing upward; S4, B-side multiple image scanning: The main controller controls the first Y-axis linear module (308) and the first X-axis linear module (309) in the B-side inspection component (302) to move the first negative pressure suction cup (310) and the IC carrier board on it in the B-side inspection component (302) to below the line scan camera (306) in the B-side inspection component (302); the main controller controls the first Z-axis linear module (305) in the B-side inspection component (302) to adjust the focal length; the main controller controls the combined light source (307) in the B-side inspection component (302) to flash sequentially according to a preset timing sequence, and synchronously triggers the line scan camera (306) in the B-side inspection component (302) to scan and image the B-side of the IC carrier board to obtain multiple image data of the B-side; the appearance inspection module (3) outputs the suspected defect marking position of the B-side of the IC carrier board based on the multiple image data of the B-side; S5, AI-powered re-examination positioning and high-magnification imaging: S51, High-magnification acquisition on side A: The main controller controls the second X-axis linear module (403) and the second Z-axis linear module (404) to move the fourth negative pressure suction cup (406) above the first negative pressure suction cup (310) of the B-side inspection component (302); the main controller controls the negative pressure source to make the fourth negative pressure suction cup (406) pick up the B-side of the IC carrier board from the B-side initial inspection station, so that the A-side is facing down; according to the suspected defect mark position on the A-side, the main controller controls the second X-axis linear module (403) to move the product being picked up, and at the same time controls the second Y-axis linear module (407) to move the first high-definition camera (409) facing up, so that the first high-definition camera (409) is aligned with each suspected defect mark position on the A-side of the IC carrier board in sequence, and high-magnification defect images of the A-side are collected; the second Z-axis linear module (404) is used for product picking up and height adjustment; S52, B-side high-magnification acquisition: After the A-side image is captured, the main controller controls the fourth negative pressure suction cup (406) to place the IC carrier board onto the B-side re-inspection stage on the slide of the third Y-axis linear module (413) and fixes it with the fifth negative pressure suction cup (414) to keep the B-side facing upward. According to the suspected defect mark position on the B-side, the main controller controls the third X-axis linear module (410) and the third Z-axis linear module (411) to drive the downward-facing second high-definition camera (412) to move. At the same time, the third Y-axis linear module (413) drives the B-side re-inspection stage to move, so that the second high-definition camera (412) is aligned with each suspected defect mark position on the B-side of the IC carrier board in sequence to acquire high-magnification defect images of the B-side. S6, AI authenticity defect determination: The AI processing module (8) receives the high-magnification defect image of side B and the high-magnification defect image of side A; the defect region enhancement submodule (802) first performs local contrast enhancement, edge response enhancement and high-frequency texture enhancement on the candidate defect region in the high-magnification defect image; the multi-scale feature extraction submodule (801) extracts the first-scale feature, the second-scale feature and the third-scale feature of the enhanced image respectively and fuses them; the contrastive learning feature encoding submodule (803) encodes the fused features; The true / false defect determination submodule (804) outputs high-confidence fake defects, high-confidence true defects, and low-confidence defects based on the encoded features; S7, Temporary Storage: The main controller controls the fourth gripper (501) to move the IC carrier board on the B-side re-inspection platform to the temporary storage bin (502); the temporary storage bin (502) receives and temporarily stores all IC carrier boards that have been inspected by the AI re-inspection positioning and high-magnification imaging module (4), and maintains the temporary storage state until all defects of the IC carrier board are judged by AI and low-confidence defects are judged manually. S8. Sorting: The main controller checks whether all defects on the same IC carrier board in the temporary storage compartment (502) have been determined. After the determination is completed, the main controller controls the fifth gripper (601) to determine the final confirmed true defect data of the IC carrier board based on the high-confidence true defects and high-confidence false defects output by the AI processing module (8) and the final judgment result output by the manual judgment module (9), and performs the following sorting actions: If the IC carrier is full of high-confidence defects, the high-confidence defects are directly filtered out and not included in the sorting defect statistics. The fifth gripper (601) is controlled to transfer the IC carrier to the qualified warehouse (603). If at least one high-confidence true defect and / or low-confidence defect exists on the IC carrier, the following sorting action shall be performed on the IC carrier: Among them, high-confidence true defects on the IC carrier are directly included in the final confirmed true defect data of the IC carrier; low-confidence defects on the IC carrier are subject to manual final judgment by the manual judgment module (9). If the manual final judgment confirms that the defect is true, it is included in the final confirmed true defect data of the IC carrier. If the manual final judgment confirms that the defect is false, it is directly filtered and not included in the sorting defect statistics; high-confidence false defects on the IC carrier are directly filtered and not included in the sorting defect statistics. The defects in the final confirmed true defect data of the IC substrate are divided into fatal defects and repairable defects. Fatal defects include at least one of board damage, short circuit, open circuit, and missing parts, while repairable defects include at least one of foreign matter and dirt. If a fatal defect is found in the final confirmed true defect data of the IC carrier, the fifth gripper (601) is controlled to transfer the IC carrier to the scrap bin (604). If the final confirmed true defect data of the IC carrier board only contains repairable defects, then control the fifth gripper (601) to transfer the IC carrier board to the cleaning chamber (602). If the final confirmed true defect data of the IC carrier is empty, the fifth gripper (601) is controlled to transfer the IC carrier to the qualified warehouse (603).
