A wire bonding process defect detection system and a wire bonding process defect detection method

Abstract

The present application relates to a kind of bonding wire process defect detection system and process defect detection method, the process defect detection system includes: two-dimensional image acquisition module, three-dimensional data acquisition module, motion and control module and computer processing module;Two-dimensional image acquisition module is used to obtain the image information of the bonding wire to be measured;Three-dimensional data acquisition module is used to obtain the three-dimensional data of the bonding wire to be measured;Motion and control module is used to according to the specified command of computer processing module, cooperate two-dimensional image acquisition module and / or three-dimensional data acquisition module, accurately move the hybrid integrated circuit board to be measured, complete two-dimensional and / or three-dimensional image scanning;Computer processing module is used to process the image information and / or three-dimensional data of the bonding wire to be measured, unified analysis processing the preliminary detection result and accurate detection result of bonding wire defect, will defect detection result output.The present application improves the quality and production efficiency of hybrid integrated circuit, provides certain degree of guidance value for practical engineering application.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of process defect detection, in particular to a bonding wire process defect detection system and a process defect detection method. BACKGROUND

[0002] With the development of electronic equipment towards miniaturization and multi-function, the demand for multi-chip assemblies is increasing, and the assembly density is also increasing. Wire bonding is the most common process method for interconnection between chips, substrates and IO. There are hundreds or even thousands of bonding wires in a multi-chip assembly, and the interconnection quality directly determines the performance and reliability of the assembly, so the quality inspection of the bonding wire is very important.

[0003] At present, manual visual inspection is a common method for detecting bonding wire process defects, but as the application of multi-chip assembly technology becomes more and more widespread, its production and module assembly density are also increasing dramatically. The traditional manual visual inspection method relies heavily on the experience and working state of the inspector, and the work intensity is high, which is prone to false rejection and missed detection, so the detection efficiency and accuracy of manual visual inspection cannot meet the actual production needs.

[0004] In recent years, with the rapid development of automatic optical detection technology, the detection accuracy and efficiency have been continuously improved, and automatic optical detection technology has been widely applied in the field of industrial defect detection. Conventional automatic optical detection technology is mainly used for detection with standard images, such as standard packaging components, substrates and even chips, etc. However, in the field of multi-chip assemblies, there are many types of components, high assembly density, large size variation of detection objects and lack of standard patterns, especially in the field of micro-defect detection such as bonding wires, there are still many problems.

[0005] The difficulty of bonding wire defect detection mainly has the following two reasons: first, the bonding wire is very thin, with a diameter of microns, and the bonding span is large. The reflection angle of light is different due to the bending of the bonding wire, which affects the detection image of the bonding wire. The curvature changes are determined by the arch shape and height, so the pixels within the lead span change significantly, and the detection difficulty of effective pixels increases. Second, the background image of the detected bonding wire changes greatly. A lead often crosses different pads, substrates and chips, etc. The change of the background makes the processing of the bonding wire detection image very complex. When using traditional automatic optical detection technology for detection, the requirements for light source types and lighting methods are high, and the pixel requirements for the camera lens are also high, which makes it difficult to ensure high defect detection accuracy and efficiency. In addition, the current defect detection for bonding wires is mainly in the two-dimensional image field, and there is a lack of detection methods for three-dimensional defects of bonding wires. SUMMARY

[0006] The purpose of the present application is to solve the above-mentioned deficiencies of the prior art, propose a bonding wire process defect detection system and process defect detection method, so as to comprehensively, quickly and accurately detect the process defects of the bonding wire, reduce the labor intensity of the detection personnel, and improve the production quality and production efficiency.

[0007] In order to solve the above technical problems, the technical scheme of the present application is as follows:

[0008] A bonding wire process defect detection system comprises a two-dimensional image acquisition module, a three-dimensional data acquisition module, a motion and control module, and a computer processing module.

[0009] The two-dimensional image acquisition module is used to acquire image information of the bonding wire to be detected.

[0010] The three-dimensional data acquisition module is used to acquire three-dimensional data of the bonding wire to be detected.

[0011] The motion and control module is used to accurately move the hybrid integrated circuit board to be detected according to the specified command of the computer processing module, cooperate with the two-dimensional image acquisition module and / or the three-dimensional data acquisition module, and complete two-dimensional and / or three-dimensional image scanning.

[0012] The computer processing module is used to process the image information of the bonding wire to be detected acquired by the two-dimensional image acquisition module, acquire bonding wire positioning information and a preliminary detection result of bonding wire defects, process the three-dimensional data of the bonding wire to be detected acquired by the three-dimensional data acquisition module, perform bonding wire three-dimensional model reconstruction, and obtain an accurate detection result of bonding wire defects after analysis and processing; unify the preliminary detection result and the accurate detection result of the bonding wire defects, and output the final defect detection result.

[0013] In the above technical scheme, the two-dimensional image acquisition module comprises a CCD camera and an integrating sphere light source.

[0014] The CCD camera is used to take pictures of the bonding wire to be detected under illumination conditions, acquire corresponding image information, and upload the image information to the computer processing module for processing.

[0015] The integrating sphere light source is used to provide a multi-angle light source for illuminating the center position of the measurement surface.

[0016] In the above technical scheme, the three-dimensional data acquisition module comprises at least two high-speed CMOS camera lenses and a laser emitter.

[0017] The at least two high-speed CMOS camera lenses are used to acquire three-dimensional information of the bonding wire at multiple angles.

[0018] The laser emitter is used to project laser to the bonding wire area of the hybrid integrated circuit to be tested, and cooperates with the motion and control module to perform accurate movement scanning to obtain a complete three-dimensional scanning image of the bonding wire to be tested.

[0019] In the technical solution, the motion and control module comprises a conveyor belt, a precision motion control platform, a controller and a good and bad product sorter.

[0020] The conveyor belt is controlled by the controller, and after the hybrid integrated circuit to be tested is placed in a designated position, the controller is used to control the conveyor belt to move forward according to the designated command of the computer processing module, so that the circuit board to be tested is first sent to the position directly below the two-dimensional image acquisition module, and then the precision motion control platform is used to accurately adjust the bonding wire area to be photographed to a designated photographing area for image acquisition; after the two-dimensional image information acquisition is completed, the two-dimensional image information is transmitted to the computer processing module for processing, and the obtained bonding wire positioning information is transmitted to the controller.

[0021] The controller is also used to control the conveyor belt to move forward to the three-dimensional data acquisition area to wait for scanning by the three-dimensional line laser scanner; the controller controls the three-dimensional line laser scanner and the precision motion control platform to accurately move according to the bonding wire positioning information, so that the laser emitter is directly opposite the bonding wire area to be tested, and then the precision motion control platform is used to step by step move to obtain complete three-dimensional scanning data of the bonding wire; finally, the acquired three-dimensional data is transmitted to the computer processing module for analysis and processing; after the computer processing module completes the analysis and processing of the bonding wire defect information, the detection result is displayed, and the controller controls the good and bad product sorter to sort the circuit board to be tested according to the detection result, so that the qualified product is transmitted to the next process, and the unqualified product is transmitted to the bad product recycling center for processing.

[0022] In the technical solution, the computer processing module comprises a computer system.

[0023] The computer system is used to process the image information of the bonding wire to be tested acquired by the two-dimensional image acquisition module, to obtain bonding wire positioning information and preliminary detection results of bonding wire defects; according to the bonding wire positioning information, accurate motion instructions are transmitted to the controller; the three-dimensional data of the bonding wire to be tested acquired by the three-dimensional data acquisition module is processed, a three-dimensional model of the bonding wire is reconstructed, and accurate detection results of the bonding wire defects are obtained after analysis and processing; the preliminary detection results and the accurate detection results of the bonding wire defects are uniformly analyzed and processed, and the final defect detection results are output to the good and bad product sorter for processing.

[0024] A bonding wire process defect detection method suitable for the bonding wire process defect detection system, comprising the following steps:

[0025] Step 1, bonding wire positioning and preliminary defect detection:

[0026] The two-dimensional image acquisition module uses a target detection algorithm to complete the positioning detection of all bonding wires on the mixed integrated circuit to be measured, and obtains image information of the bonding wire to be measured; then the image information of the bonding wire to be measured is input to the computer processing module;

[0027] Step 2, three-dimensional model reconstruction of bonding wire:

[0028] The computer processing module processes the image information of the bonding wire to be measured, controls the precise movement of the three-dimensional line laser scanner and the precision motion platform, realizes complete scanning measurement of the three-dimensional data of the bonding wire, and obtains three-dimensional data of the bonding wire to be measured;

[0029] Step 3, precise defect detection:

[0030] The bonding wire defect is determined by comparing and analyzing the template information; the image information of the bonding wire to be measured and the three-dimensional data of the bonding wire to be measured are input to the computer system for unified analysis and processing, and the final defect detection result is output.

[0031] In the above technical solution, the step 1 specifically comprises the following steps:

[0032] An open source tool labelimg is used to make a data set required for deep learning training, and all normal bonding wires and various bonding wire defects on the mixed integrated circuit board are labeled;

[0033] A traditional data augmentation method is used to realize data enhancement;

[0034] The deep learning network is trained using the bonding wire image data set to be measured;

[0035] The computer processing system transmits the image captured by the two-dimensional image acquisition system to the trained deep learning network for detection, and obtains the positioning information and preliminary defect detection result of all bonding wires.

[0036] In the above technical solution, the step 2 specifically comprises the following steps:

[0037] Scan control: obtain the motion parameters of the high-speed precision electric control translation stage, and calculate the coordinate value of the measurement point in the horizontal direction;

[0038] Three-dimensional data processing: the three-dimensional image of the bonding wire is collected by the camera, the image is preprocessed, the light strip center pixel coordinates are extracted and substituted into the sensor measurement model, and the three-dimensional coordinates of the measurement point are calculated in combination with the motion parameters of the precision motion platform;

[0039] Sensor calibration: a large number of known coordinate target feature points are used to solve the sensor mathematical model parameters, and the three-dimensional model reconstruction of the bonding wire is completed.

[0040] The present application has the following beneficial effects:

[0041] The present application takes the defect detection of the bonding wire process of the hybrid integrated circuit as the target, combines deep learning and defect detection technology, and innovatively designs a detection system capable of detecting two-dimensional and three-dimensional defects of the bonding wire in view of the special structural characteristics of the bonding wire. The advantages of deep learning in target detection and identification are fully utilized, the accuracy, timeliness and efficiency of the defect detection of the hybrid integrated circuit are effectively improved, the quality and production efficiency of the hybrid integrated circuit are improved, and a certain degree of guidance value is provided for actual engineering application. BRIEF DESCRIPTION OF DRAWINGS

[0042] The present application will be further described in detail below in combination with the drawings and specific embodiments.

[0043] Figure 1 It is a schematic diagram of the defect type of the bonding wire.

[0044] Figure 2 It is a schematic diagram of the overall design structure of the system.

[0045] Figure 3 It is a schematic diagram of the structure of the integrating sphere light source.

[0046] Figure 4 It is a schematic diagram of the double-lens three-dimensional line laser scanner.

[0047] Figure 5 It is a schematic diagram of the overall detection process.

[0048] Figure 6 It is a schematic diagram of the YOLOV5 network structure.

[0049] Figure 7 It is a schematic diagram of the positioning and preliminary detection result of the bonding wire.

[0050] Figure 8 It is a schematic diagram of the basic process of three-dimensional structure detection of the bonding wire.

[0051] Figure 9 It is a schematic diagram of the structure of the bonding wire to be detected and the three-dimensional scanning model.

[0052] The reference signs in the drawings represent:

[0053] 1-CCD camera; 2-integrating sphere light source; 3-high-speed CMOS camera lens; 4-laser emitter; 5-good and bad product sorter; 6-conveyer belt; 7-circuit board to be detected; 8-precision motion control platform; 9-controller; 10-computer system. DETAILED DESCRIPTION

[0054] The application will be described in detail below with reference to the accompanying drawings.

[0055] The technical solutions of the application mainly include two parts. The first part is to design a special detection system according to the characteristics of the bonding wire process defect detection, which specifically includes a special two-dimensional image and three-dimensional data acquisition device, an illumination system, a controller, a precision motion control platform, and a sorter. The second part is the detection algorithm part. In view of the problem that the traditional algorithm has poor detection accuracy and efficiency, the application innovatively uses a deep learning algorithm to analyze and detect the bonding wire process defect information. In view of the problem that it is too difficult to directly scan the three-dimensional information of the bonding wire and reconstruct the three-dimensional model of the hybrid integrated circuit board, and it is difficult to achieve, the application innovatively proposes a method of two-dimensional positioning first and three-dimensional detection second. In view of the problem that it is difficult to use a single camera to collect three-dimensional information of the bonding wire due to the small diameter of the bonding wire, large bonding span, complex appearance characteristics and complex background, and the precision is poor, the application innovatively uses a three-dimensional line laser scanning method of double cameras to obtain the three-dimensional information of the bonding wire. In view of the problem that the defect detection using two-dimensional images only has a high omission rate and limited defect types, the application innovatively uses a combination of two-dimensional defect preliminary detection of the bonding wire and three-dimensional defect accurate detection of the bonding wire. The detection method not only can realize the detection of two-dimensional defects such as broken wire, missing wire, flying wire, misconnection and offset short connection of the bonding wire, but also can realize the three-dimensional parameter measurement and the detection of the bonding wire sag defect. The bonding wire defect type diagram is shown in Figure 1 .

[0056] The first part: the bonding wire detection system design.

[0057] The bonding wire process defect detection system of the application mainly includes a two-dimensional image acquisition module, a three-dimensional data acquisition module, a motion and control module, and a computer processing module. The overall structure diagram of the bonding wire detection system is shown in Figure 2 .

[0058] The two-dimensional image acquisition module mainly includes a CCD camera 1 and an integrating sphere light source 2. The CCD camera 1 is used to take pictures of the bonding wire part of the hybrid integrated circuit under the illumination condition, obtain the corresponding image information, and upload the image information to the computer processing module for processing. The illumination part adopts the vertical downward illumination of the integrating sphere light source 2. The integrating sphere light source 2 is a LED light source which is reflected multiple times by the spherical inner wall with good integrating effect, and has a 360-degree reflected light superposition effect. Therefore, the integrating sphere light source 2 can ensure the uniformity of the whole image illumination in the illumination detection. In addition, the integrating sphere light source 2 can provide light sources at different angles, which can concentrate the light source at the center of the measurement surface and eliminate the interference of the shadow, so as to highlight the appearance characteristics of the bonding wire. The structure diagram of the integrating sphere light source 2 is shown in Figure 3as shown.

[0059] The three-dimensional data acquisition module is mainly composed of a three-dimensional line laser scanner. The three-dimensional line laser scanner adopts a direct reflection type and a double-lens placement mode. The light is concentrated, and the scanning effect is good. Two high-speed CMOS camera lenses 3 can obtain three-dimensional information of the bonding wire at multiple angles, which can greatly reduce the reflection error caused by the deformation of the bonding wire to a certain extent, and help to construct a more accurate three-dimensional model of the bonding wire. The three-dimensional line laser scanner is shown in Fig. 3. Figure 4 The three-dimensional data acquisition module transmits the collected data to the computer processing module for processing.

[0060] The motion and control module mainly includes a conveying belt 6, a precise motion control platform 8, a controller 9, and a good-bad product sorter 5. The conveying belt 6 is controlled by the controller 9. After the mixed integrated circuit to be tested is placed in the designated position, the controller 9 controls the conveying belt 6 to move forward according to the designated command of the computer processing module, so that the mixed integrated circuit to be tested is first sent to the position directly below the two-dimensional image acquisition module, and then the precise motion control platform 8 is used to adjust the position accurately, so that the bonding wire area to be photographed and detected is adjusted to the designated photographing area for image acquisition. After the two-dimensional image information acquisition is completed, the two-dimensional image information is transmitted to the computer processing module for further processing, and the bonding wire positioning information obtained is transmitted to the controller 9. The controller 9 controls the conveying belt 6 to continue moving forward and to the three-dimensional data acquisition area to wait for scanning by the three-dimensional line laser scanner. The controller 9 controls the three-dimensional line laser scanner and the precise motion control platform 8 to move accurately according to the bonding wire positioning information, so that the laser emitter 4 is directly opposite the bonding wire area to be tested. Then, the precise motion control platform is moved step by step to obtain complete three-dimensional scanning data of the bonding wire. Finally, the collected three-dimensional data is transmitted to the computer processing module for analysis and processing. The good-bad product sorter 5 is used to distinguish qualified circuit boards from defective circuit boards. After the computer processing module completes the analysis and processing of the bonding wire defect information, the detection result is displayed, and the controller 9 controls the good-bad product sorter 5 to sort the tested circuit board 7 according to the product category, so that the qualified products are conveyed to the next process, and the unqualified products are conveyed to the defective product recycling center for processing.

[0061] The computer processing module includes a computer system 10. The functions of the computer processing module include:

[0062] 1. Processing the image information collected by the two-dimensional image acquisition module to obtain bonding wire positioning information and preliminary detection results of bonding wire defects;

[0063] 2. According to the bonding wire position information, accurate motion instructions are transmitted to the controller;

[0064] 3. The three-dimensional data of the bonding wire collected by the three-dimensional data collection module is processed, the three-dimensional model of the bonding wire is reconstructed, and the accurate detection result of the bonding wire defect is obtained after analysis and processing;

[0065] 4. The preliminary detection result and the accurate detection result of the bonding wire defect are uniformly analyzed and processed, and the final defect detection result is output to the good / bad product sorter for further processing.

[0066] Second part: bonding wire detection algorithm design.

[0067] The bonding wire detection algorithm is described in two parts: the first part is bonding wire positioning and defect preliminary detection, and the second part is bonding wire three-dimensional model reconstruction and defect accurate detection. The two parts cooperate with each other to realize accurate detection of all defects of the bonding wire. The overall detection process is as shown in Figure 5 .

[0068] Because the types of bonding wire pictures are various, the background is complex, and it is too difficult to directly detect the bonding wire defects on the whole hybrid integrated circuit board, so the method of positioning first and then detecting is adopted for bonding wire defect detection. First, the target detection algorithm is used to complete the positioning and detection of all bonding wires on the hybrid integrated circuit to be tested, and the preliminary detection of the two-dimensional defects of the bonding wire is realized. Then, the bonding wire positioning result is input to the computer system, the computer system processes the related positioning information and transmits instructions to the controller, the controller controls the accurate movement of the three-dimensional line laser scanner and the precision motion platform, and realizes the complete scanning measurement of the three-dimensional data of the bonding wire. After the three-dimensional data of the bonding wire is collected, the three-dimensional model of the bonding wire is reconstructed through related processing algorithm processing and model parameter calibration, and the judgment of the bonding wire defect is completed by comparing and analyzing the template information. Finally, the preliminary detection result and the accurate detection result of the bonding wire are input to the computer system for unified analysis and processing, and the final defect detection result is output.

[0069] 1. Bonding wire positioning and defect preliminary detection

[0070] First, the open source tool labelimg is used to generate the data set required for deep learning training, and all normal bonding wires and various bonding wire defects on the hybrid integrated circuit board are labeled. The data format is yolo format.

[0071] Second step: the deep learning algorithm has strong dependence on the data set, and the number of mixed integrated circuit boards containing bonding wire defects in actual industrial production is insufficient. The data enhancement is realized by using traditional data augmentation method. Through random cropping, translation transformation, brightness change, noise addition, rotation angle and mirror image of the data set image, the number of data set is greatly expanded, which meets the basic requirements of training.

[0072] Third step, using the image data set of the bonding wire to be tested to train the deep learning network, the target detection model designed by the application is based on YOLOV5 network, and the YOLOV5 network structure diagram is as shown in Figure 6 YOLOV5 network is composed of four parts of input end (Input), backbone network (Backbone), neck layer and output layer (Prediction). The defect detection principle is mainly that the adaptive anchor box extracts image features through Backbone, the Neck part is responsible for fusing image feature values, then the image features are predicted, finally the prediction category is generated, and the bounding box coordinates of the target position are returned. YOLOV5 uses CSP as the backbone network on the basis of the cross-stage local network idea, mainly including focus structure, convolution module, bottleneck layer (C3) and spatial pyramid pooling (SPP). The focus structure is to reconstruct the pixel points in the high-resolution image to the low-resolution image, and this module is mainly used to speed up the calculation speed. The SPP module adopts 5 / 9 / 13 maximum pooling respectively, and the contact fusion is carried out to improve the receptive field. The structure of FPN+PAN is used in the neck network to strengthen the feature fusion ability. Finally, the loss function GIoU_LOSS is used in the output end, and the GIoU function contains the area of the minimum box of the predicted box and the real box. The GIoU calculation process is as shown in the following formula, and the function can enhance the detection ability.

[0073]

[0074]

[0075] Fourth step, the computer processing system transmits the image shot by the two-dimensional image acquisition system to the trained deep learning network for detection, and obtains the positioning information and preliminary defect detection result of all bonding wires, as shown in Figure 7

[0076] 2, bonding wire three-dimensional model reconstruction and defect accurate detection

[0077] The basic flow of bonding wire three-dimensional structure detection is as shown in Figure 8 ​The system mainly includes three parts: scanning control, three-dimensional data processing and sensor calibration. The scanning control part can obtain the motion parameters of the high-speed precision electric control translation stage to calculate the coordinate values of the measuring points in the lateral direction. The three-dimensional data processing includes using the camera to collect the three-dimensional image of the bonding wire, preprocessing the image, extracting the light strip center pixel coordinates and substituting them into the sensor measurement model, combining the motion parameters of the precision motion platform to calculate the three-dimensional coordinates of the measuring points. The calibration part mainly uses a large number of target feature points with known coordinates to solve the sensor mathematical model parameters. Finally, the three-dimensional model of the bonding wire is reconstructed. The key technology design is described as follows:

[0078] In the vision detection system based on line structured light, the mathematical model of the sensor is to determine the coordinate conversion relationship of the spatial points in the related coordinate system. To realize a high-precision three-dimensional detection system, a suitable sensor mathematical model needs to be established, and a suitable calibration method is used to obtain accurate model parameters.

[0079] The system uses a sensor mathematical model based on DLT. The DLT method applies the space mapping theory to describe the relationship between the three-dimensional coordinates P(X w ,Y w ,Z w ) of the spatial coordinate points and the ideal computer image coordinates (u, v) as a direct linear transformation. The specific mathematical expression is as follows:

[0080]

[0081] In the formula, Q is a matrix composed of related parameters in the DLT model, which needs to be determined through calibration.

[0082] In actual measurement process, the lens distortion problem should also be considered to achieve high precision. Under the distortion condition, the relationship between the ideal image coordinates (u, v) and the actual image coordinates (u d ,v d ) is nonlinear, which can be represented by a polynomial. Any measured point P(X w ,Y w ,Z w ) and its actual image coordinates (u d ,v d ) have the following relationship:

[0083]

[0084] In the formula, C ij , D ij represent the model parameters, and the subscripts i, j are the orders of u d , v d , respectively. n is the highest order of the polynomial. The formula is the DLT model based on the polynomial used by the system.

[0085] This system uses line structured light to be vertically projected onto the bonding wire area of ​​the hybrid integrated circuit board under test. By substituting the image coordinates of the light stripe points obtained from image processing into the sensor's mathematical model, their corresponding two-dimensional coordinates (X, Y, F, Z) in the measurement coordinate system can be obtained. w ,Y w To obtain the complete three-dimensional coordinates (X, Y, X) of a point in the measurement coordinate system. w ,Y w Z w The system adopts a passive scanning measurement method, that is, the line structure light emitter and two CMOS cameras are placed at a fixed angle, and the sensor position is fixed; the hybrid integrated circuit board under test is placed on a precision motion control platform, and the controller controls the movement of the precision motion platform according to the positioning information of the two-dimensional positioning system to realize the complete scanning measurement of the bonding wire under test.

[0086] During measurement, the precision motion platform moves in steps, with a step distance ΔZ. k This refers to the coordinate value in the Z direction. It is different from the two-dimensional coordinate (X) calculated using the DLT model. w ,Y w By combining these, the three-dimensional coordinates (X, Y, Z) of each measurement point can be obtained. w ,Y w Z w The specific measurement model expression is as follows:

[0087]

[0088] In the formula, v t t represents the scanning speed and scanning time interval of the precision motion platform, respectively. The bonding wire to be tested is scanned to obtain three-dimensional data of the morphological characteristics of the bonding wire.

[0089] To address the limited measurement range of the three-dimensional detection system for bonding wire defects in hybrid integrated circuits, and considering the need to unify the relevant coordinate systems for real-time mutual compensation between the two sensors during measurement, this invention designs a suitable target and a unified dual-sensor calibration method. Utilizing a designed parallel-line target, images of the target at different locations are acquired, target feature points are extracted, and calibration calculations are performed. Simultaneously, the polynomial parameters C of the left and right sensors are obtained. ij and D ij , thus completing coordinate system one.

[0090] The structure of the bonding wire to be tested and its 3D scanning model are as follows: Figure 9 As shown, the 3D scan image clearly distinguishes solder joints, bonding wires, chips, and circuit boards. By comparing the morphological features of the bonding wires with the template information, it can be determined whether the bonding wires contain defects and the types of defects.

[0091] The application takes the detection of the bonding wire process defects of hybrid integrated circuits as the target, combines deep learning and defect detection technology, and innovatively designs a detection system that can detect two-dimensional and three-dimensional defects of the bonding wire in view of the special structural characteristics of the bonding wire. The advantages of deep learning in target detection and identification are fully utilized, the accuracy, timeliness and efficiency of the defect detection of the hybrid integrated circuits are effectively improved, the quality and production efficiency of the hybrid integrated circuits are improved, and a certain degree of guidance value is provided for actual engineering application.

[0092] Obviously, the above embodiments are only examples for clearly illustrating, but not limitation to the embodiments. For ordinary skilled in the art, other different forms of changes or variations can be made on the basis of the above description. Here, all the embodiments need not and cannot be exhausted. The obvious changes or variations derived therefrom are still within the protection scope of the present application.

Claims

1. A bonded wire process defect detection system characterized by, The application relates to a keying wire defect detection system, which comprises a two-dimensional image acquisition module, a three-dimensional data acquisition module, a motion and control module and a computer processing module. The two-dimensional image acquisition module is used for acquiring image information of a keying wire to be detected. The three-dimensional data acquisition module is used for acquiring three-dimensional data of the keying wire to be detected. The motion and control module is used for accurately moving a mixed integrated circuit board to be detected according to a specified command of the computer processing module, cooperating with the two-dimensional image acquisition module and / or the three-dimensional data acquisition module, and completing two-dimensional and / or three-dimensional image scanning. The computer processing module is used for processing image information of the keying wire to be detected collected by the two-dimensional image acquisition module, acquiring keying wire positioning information and a preliminary detection result of a keying wire defect, processing three-dimensional data of the keying wire to be detected collected by the three-dimensional data acquisition module, performing keying wire three-dimensional model reconstruction, and obtaining an accurate detection result of the keying wire defect after analysis and processing. The preliminary detection result and the accurate detection result of the keying wire defect are uniformly analyzed and processed, and a final defect detection result is output. The two-dimensional image acquisition module comprises a CCD camera (1) and an integrating sphere light source (2). The CCD camera (1) is used for photographing the keying wire to be detected under illumination conditions, acquiring corresponding image information, and uploading the image information to the computer processing module for processing. The integrating sphere light source (2) is used for providing multi-angle light sources for irradiating a center position of a measurement surface. The three-dimensional data acquisition module mainly comprises a three-dimensional line laser scanner, and comprises at least two high-speed CMOS camera lenses (3) and a laser emitter (4). The at least two high-speed CMOS camera lenses (3) are used for acquiring three-dimensional information of the keying wire in multiple angles. The laser emitter (4) is used for projecting laser to a mixed integrated circuit keying wire area to be detected, and cooperates with the motion and control module to accurately move and scan, so that a complete three-dimensional scanning graph of the keying wire to be detected is obtained. The motion and control module comprises a conveying belt (6), a precision motion control platform (8) and a controller (9). The circuit board to be detected (7) is first sent to the lower side of the two-dimensional image acquisition module, the keying wire area needing to be detected is adjusted to a specified photographing area for image acquisition; after the two-dimensional image information acquisition is completed, the two-dimensional image information is transmitted to the computer processing module for processing, and keying wire positioning information is obtained and transmitted to the controller (9). The controller (9) controls the three-dimensional line laser scanner and the precision motion control platform (8) to accurately move according to the keying wire positioning information, so that the laser emitter (4) is opposite to the keying wire area to be detected, and then the precision motion control platform (8) is stepwise moved to obtain complete three-dimensional scanning data of the keying wire. The motion and control module further comprises a good-bad product sorter (5).

2. The bonded wire process defect detection system of claim 1, wherein, ​ The conveyor belt (6) is controlled by the controller (9), after the mixed integrated circuit to be tested is placed in the designated position, the controller (9) is used to control the conveyor belt (6) to move forward according to the designated command of the computer processing module, so that the circuit board (7) to be tested is first sent to the position directly below the two-dimensional image acquisition module, then the precise position adjustment is performed by the precise motion control platform (8), the bonding wire area to be photographed and detected is adjusted to the designated photographing area for image acquisition; after the two-dimensional image information acquisition is completed, the two-dimensional image information is transmitted to the computer processing module for processing, and the obtained bonding wire positioning information is transmitted to the controller (9); The controller (9) is also used to control the conveyor belt (6) to move forward and move to the three-dimensional data acquisition area to wait for scanning by the three-dimensional line laser scanner; the controller (9) controls the three-dimensional line laser scanner and the precise motion control platform (8) to move accurately according to the bonding wire positioning information, so that the laser emitter (4) is directly opposite the bonding wire area to be tested, then the precise motion control platform (8) is stepwise moved to obtain complete three-dimensional scanning data of the bonding wire; finally, the collected three-dimensional data is transmitted to the computer processing module for analysis and processing; when the computer processing module completes the analysis and processing of the bonding wire defect information, the detection result is displayed, and the controller (9) controls the good and defective product sorter (5) to sort the types of the circuit board (7) to be tested according to the detection result, so that the qualified products are conveyed to the next process and the unqualified products are conveyed to the defective product recycling center for processing.

3. The bonded wire process defect detection system of claim 2, wherein, The computer processing module comprises a computer system (10); The computer system (10) is used to process the image information of the bonding wire to be tested collected by the two-dimensional image acquisition module, obtain the bonding wire positioning information and the preliminary detection result of the bonding wire defect, transmit the precise motion instruction to the controller (9) according to the bonding wire position information, process the three-dimensional data of the bonding wire to be tested collected by the three-dimensional data acquisition module, reconstruct the three-dimensional model of the bonding wire, and obtain the precise detection result of the bonding wire defect after analysis and processing; the preliminary detection result and the precise detection result of the bonding wire defect are uniformly analyzed and processed, and the final defect detection result is output to the good and defective product sorter (5) for processing.

4. A method for detecting defects in a wire bonding process using the system for detecting defects in a wire bonding process according to claim 1, characterized by, The method comprises the following steps: Step 1, bonding wire positioning and defect preliminary detection: The two-dimensional image acquisition module adopts a target detection algorithm to complete the positioning detection of all bonding wires on the mixed integrated circuit to be tested, and obtains the image information of the bonding wire to be tested; then the image information of the bonding wire to be tested is input to the computer processing module; Step 2, three-dimensional model reconstruction of the bonding wire: The computer processing module processes the image information of the bonding wire to be tested, controls the precise movement of the three-dimensional line laser scanner and the precise motion platform, realizes the complete scanning measurement of the three-dimensional data of the bonding wire, and obtains the three-dimensional data of the bonding wire to be tested; Step 3, precise detection of defects: The determination of the bonding wire defect is completed by comparison analysis with the template information; the image information of the bonding wire to be measured and the three-dimensional data of the bonding wire to be measured are input into a computer system for unified analysis and processing, and the final defect detection result is output.

5. The bonded wire process defect detection method of claim 4, wherein, The step 1 specifically comprises the following steps: An open source tool labelimg is used to generate a data set required for deep learning training, and all normal bonding wires and various bonding wire defects on a hybrid integrated circuit board are labeled; A traditional data augmentation method is used to realize data enhancement; A deep learning network is trained using the bonding wire image data set to be measured; The computer processing system transmits the image captured by the two-dimensional image acquisition system to the trained deep learning network for detection, and obtains the positioning information and the preliminary detection result of all the bonding wires.

6. The bonded wire process defect detection method of claim 4, wherein, The step 2 specifically comprises the following steps: Scan control: obtain the motion parameters of the high-speed precision electric control translation table, and calculate the coordinate values in the transverse direction of the measurement point; Three-dimensional data processing: the three-dimensional image of the bonding wire is collected by the camera, the image is preprocessed, the light strip center pixel coordinates are extracted and substituted into the sensor measurement model, and the three-dimensional coordinates of the measurement point are calculated in combination with the motion parameters of the precision motion platform; Sensor calibration: a large number of target feature points with known coordinates are used to solve the sensor mathematical model parameters, and the three-dimensional model reconstruction of the bonding wire is completed.

Images