Semiconductor die bonder and control method thereof

Abstract

The present application relates to the technical field of semiconductor manufacturing, in particular to a semiconductor die bonder and a control method thereof. The die bonder comprises a swing arm, which is rotationally connected to a rack through a rotating shaft. The swing arm is driven by a first driving source, which is controlled by a controller. A first substrate and a second substrate are respectively arranged below the swing arm and can move forward and backward or left and right relative to the rotating shaft in a horizontal plane. The movement of the first substrate and the second substrate is respectively controlled by a second driving source and a third driving source, which are both controlled by the controller. The present application provides a semiconductor die bonder and a control method thereof. Real-time monitoring and feedback are performed by a camera, and a control strategy is combined, so that the position and posture of the chip can be perceived in real time during the whole die bonding process, and flexible adjustment can be made according to the actual situation, thereby ensuring accurate alignment of the chip on the target substrate.

Description

Technical Field

[0001] This invention relates to the field of semiconductor manufacturing technology, and in particular to a semiconductor die bonder and its control method. Background Technology

[0002] In the semiconductor manufacturing field, die bonding is a crucial step in precisely mounting chips onto a substrate to form electronic components or integrated circuits. With the rapid development of semiconductor technology, chip sizes are constantly shrinking and integration levels are continuously increasing, posing unprecedented challenges to the precision and efficiency of die bonding. Traditional die bonding processes often rely on manual operation or simple mechanical devices, methods that are inadequate to meet the high precision and high efficiency demands of modern semiconductor manufacturing.

[0003] Manual operation in the die bonding process has several drawbacks. First, due to human physiological limitations, prolonged operation can easily lead to fatigue, resulting in decreased positioning accuracy. Second, manual operation is greatly affected by subjective factors; differences in the skill levels of different operators can lead to inconsistencies in product quality. Furthermore, manual operation is inefficient and cannot meet the demands of large-scale production.

[0004] While simple mechanical devices improve production efficiency to some extent, they still lack flexibility when dealing with microchips, high-density integration, and complex substrate structures. These devices often lack real-time monitoring and feedback mechanisms, making it impossible to accurately sense changes in the chip's position and orientation during the die bonding process, thus making it difficult to ensure accurate alignment of the chip on the target substrate.

[0005] For example, patent publication number CN111370350B, entitled "Die Bonding Machine," includes a stand; a dispensing device; a dispensing displacement mechanism; a feeding mechanism; a die bonding arm device; a die supply platform; a die bonding displacement mechanism; and a receiving mechanism. The die bonding arm device includes a rotating frame, multiple die bonding arms, a lifter, and a die bonding motor, with each die bonding arm equipped with a nozzle. The dispensing device includes multiple dispensing modules. The disadvantage is that although this type of die bonder improves die bonding efficiency to some extent, it lacks real-time monitoring and feedback devices such as cameras, making it unable to perceive the chip's position and orientation in real time. Therefore, it is insufficient in ensuring accurate alignment of the chip on the target substrate.

[0006] Existing die bonders also have limitations in their control strategies. They often employ a single, fixed control strategy, unable to flexibly adjust according to the characteristics of the chip and substrate, as well as changes in the actual production process. This makes it difficult to guarantee the precision and stability of the bonding process when encountering chips and substrates of different sizes, shapes, or materials. Summary of the Invention

[0007] To address the shortcomings of existing die bonders in ensuring accurate chip alignment on the target substrate, this invention provides a semiconductor die bonder and its control method. By using a camera for real-time monitoring and feedback, combined with a control strategy, the entire die bonder process can sense the chip's position and orientation in real time and make flexible adjustments according to the actual situation, thereby ensuring accurate chip alignment on the target substrate.

[0008] To achieve the above-mentioned technical objectives, the present invention provides a semiconductor die bonder, comprising:

[0009] frame;

[0010] A swing arm is rotatably connected to the frame via a rotating shaft. The swing arm is driven by a first drive source, which is controlled by a controller.

[0011] A first substrate and a second substrate, wherein the first substrate is used to support the chip to be soldered, and the second substrate is used to receive and solder the chip. The first substrate and the second substrate are respectively disposed below the swing arm and can move back and forth or left and right relative to the pivot in the horizontal plane. The movement of the first substrate and the second substrate is controlled by a second drive source and a third drive source, respectively. Both the second drive source and the third drive source are controlled by a controller.

[0012] A vacuum adsorption device, mounted on a swing arm, is used to move the chip from the first substrate to the second substrate;

[0013] A solder spraying device, mounted on a swing arm, includes a soldering head and is used to spray solder onto chips.

[0014] A heating device, mounted on the second substrate, is used for localized heating of the welding area;

[0015] A camera is positioned above and below the first substrate and the second substrate to capture image information of the chip, the first substrate and the second substrate. The camera is connected to a controller to transmit the captured image information to the controller.

[0016] The controller executes the following linkage control strategy:

[0017] During the welding process, the camera captures image information of the current welding point and acquires image information of the target area for the next welding while the previous welding was being performed.

[0018] Based on the image processing algorithm, the captured image information is analyzed to determine the optimal welding strategy for the current welding point and the optimal welding point for the next welding target area;

[0019] Based on the determined optimal welding strategy and optimal welding point, control commands are generated. The controller simultaneously controls the movement of the swing arm, the first substrate, and the second substrate, so that while the swing arm is welding the current chip to the second substrate, the first substrate moves to the position of the next chip to be welded, and the second substrate moves to the target position for the next welding.

[0020] In this technical solution, the swing arm is flexibly connected to the frame via a rotating shaft and precisely driven by a first drive source, achieving rapid and stable movement. The first and second substrates are respectively positioned below the swing arm, allowing free movement in the horizontal plane. Controlled by independent second and third drive sources, this ensures rapid switching and precise positioning of the chip between different locations. A vacuum adsorption device and a solder spraying device are cleverly integrated into the swing arm, enabling rapid chip transfer and precise soldering. A heating device is located on the second substrate, providing localized heating to the soldering area, improving soldering quality and reliability. A 360-degree camera system captures real-time images of the chip, the first substrate, and the second substrate, providing rich data support for the controller. The controller executes a precise linkage control strategy, rapidly analyzing image information through image processing algorithms to determine the optimal soldering strategy and position. During the soldering process, the controller can simultaneously control the movement of the swing arm, the first substrate, and the second substrate, achieving synchronization and high efficiency in the soldering process. While the swing arm is soldering the current chip to the second substrate, the first substrate has already moved to the next chip to be soldered, and the second substrate has also moved to the next target position for soldering, significantly shortening the soldering cycle and improving production efficiency.

[0021] The present invention is further configured such that: both the first substrate and the second substrate are provided with a hollow disk, and the chip is arranged on the hollow disk.

[0022] In this technical solution, firstly, both the first and second substrates are provided with perforated disks, allowing the chips to be arranged more orderly on the disk surface, facilitating subsequent transfer and soldering operations. The perforated structure not only reduces the weight of the substrates but also improves the ventilation and heat dissipation of the chips, helping to maintain the stability and reliability of the chips during the die bonding process. Secondly, a heating device is provided on the second substrate, providing more precise temperature control for the chip soldering process.

[0023] Another technical solution provided by the present invention is a control method for a semiconductor die bonder, comprising the following steps:

[0024] S1, the chip on the first substrate is adsorbed by the vacuum adsorption device on the swing arm, and the chip is visually inspected by a camera to determine the optimal soldering point of the chip.

[0025] S2, control the swing arm to position the optimal solder point of the chip downwards, capture the initial position image of the solder point on the second substrate through the camera, and perform initial alignment;

[0026] S3, based on the image processing algorithm, analyze the image data captured by the camera, calculate the deviation between the current position of the chip soldering point and the target soldering position of the second substrate, and generate motion commands to drive the first substrate and / or the second substrate to move until the soldering point is aligned.

[0027] S4, activate the heating device on the second substrate to preheat the second substrate, and after confirming that the solder joints are aligned, control the solder spraying device to spray solder onto the solder joints.

[0028] S5. After the solder coating is completed, the welding result is detected in real time by a camera to obtain the detection result; based on the detection result, the welding quality is predicted by a pre-trained machine learning model to obtain the prediction result information.

[0029] S6. Based on the prediction results, retrieve the corresponding action instruction set from the pre-built database. The action instruction set includes swing arm instruction, first substrate action instruction, and second substrate action instruction, which are used to adjust the position or state of the swing arm, the first substrate, and the second substrate to optimize the subsequent welding process or correct the deviation in the current welding.

[0030] In this technical solution, firstly, the chip is adsorbed by a vacuum adsorption device on a swing arm, and a camera is used for visual inspection to ensure the integrity and accuracy of the chip before transfer. Simultaneously, the optimal solder joint of the chip is determined, laying the foundation for subsequent alignment and soldering operations. Next, the swing arm is controlled to position the optimal solder joint of the chip downwards, and a preliminary position image of the solder joint on the second substrate is captured by the camera for initial alignment. The high-precision positioning capability of the camera ensures the initial alignment accuracy between the chip and the substrate. Then, the image data captured by the camera is analyzed using an image processing algorithm to calculate the deviation between the current position of the chip solder joint and the target soldering position on the second substrate. Motion commands are generated to drive the first and / or second substrates for fine-tuning, achieving precise control of the soldering process and ensuring accurate alignment of the solder joint. During the soldering preparation stage, the heating device on the second substrate is activated to preheat the substrate, providing suitable temperature conditions for subsequent solder spraying. After confirming the solder joint alignment, solder is sprayed onto the solder joint using a solder spraying device, completing the chip soldering process. After welding is completed, the welding result is monitored in real time using a camera. Based on the monitoring results, necessary adjustments are made to the swing arm, the first substrate, and / or the second substrate to ensure welding quality. This demonstrates the feedback mechanism of the control method, ensuring the stability and reliability of the welding result through real-time monitoring and adjustment. Finally, based on machine learning algorithms, real-time and historical data during the welding process are analyzed to predict and optimize the operating parameters of the solder spraying device, showcasing the intelligence level of the control method. Through data analysis and algorithm optimization, welding efficiency and quality are continuously improved.

[0031] The present invention is further configured such that: the step of using a camera to perform visual inspection of the chip and determine the optimal solder joint of the chip includes:

[0032] The electrodes and pads on the chip are identified using image recognition methods. The flatness and cleanliness of the chip surface are evaluated, and areas with no surface defects and clear electrodes are selected as potential soldering points.

[0033] Collect chip image data, train a solder joint model based on machine learning algorithm, automatically identify the best soldering area on the chip, input real-time captured chip images into the trained solder joint model, and the solder joint model predicts the position of the optimal solder joint by comparing the features in the input image with those in the training data.

[0034] A laser ranging method is used to perform three-dimensional topography detection on the chip surface, construct and analyze a three-dimensional model of the chip surface, and determine the flattest and defect-free area on the chip surface as the soldering point by analyzing the surface undulation and defect information in the three-dimensional model.

[0035] The temperature distribution of the chip after preheating is detected by thermal imaging. The temperature distribution in the thermal imaging image is analyzed by temperature analysis algorithm, and the corresponding temperature distribution uniformity value is calculated. The area with a temperature distribution uniformity value less than a preset threshold is selected as the welding point.

[0036] In this technical solution, firstly, key structures such as electrodes and pads on the chip are identified using image recognition methods. Simultaneously, the physical characteristics of the chip surface, such as flatness and cleanliness, are evaluated. Advanced image processing technology is utilized to quickly and accurately acquire basic information about the chip surface, providing crucial information for subsequent solder point selection. Next, chip image data is collected, and a machine learning algorithm is used to train a model. Leveraging the advantages of big data and artificial intelligence, the model automatically identifies the optimal soldering area on the chip through learning from a large number of samples. During real-time operation, simply inputting the captured chip image into the trained model allows for rapid prediction of the optimal solder point location, significantly improving detection efficiency and accuracy. Furthermore, laser ranging is employed to perform three-dimensional topography detection on the chip surface. Constructing and analyzing a three-dimensional model of the chip surface provides a more intuitive view of its topographic features, helping to determine the flattest, defect-free area as the solder point. Through three-dimensional topography detection, the position and shape of the solder points can be controlled more precisely, improving soldering quality. Finally, thermal imaging was used to detect the temperature distribution of the chip after preheating. Taking into account the thermal characteristics of the chip during the preheating process, by selecting areas with uniform temperature distribution and good heat dissipation performance as soldering points, the thermal stress during the soldering process can be further reduced, thereby improving the reliability and lifespan of the chip.

[0037] The present invention is further configured such that: in step S2, capturing a preliminary position image of the solder joint on the second substrate using a camera and performing preliminary alignment includes:

[0038] Based on the edge detection method, the edge of the welding point is identified, and the feature points of the welding point are extracted and matched with the preset standard feature points;

[0039] Illuminate the welding point with a light source of a specific wavelength, and adjust the brightness and angle of the light source according to actual needs;

[0040] The captured images are processed in real time to identify the location of the welding point and compare it with the preset location;

[0041] Based on the identification results, the positions of the first substrate and / or the second substrate are automatically adjusted.

[0042] In this technical solution, firstly, the edges of the welding points are identified using edge detection methods, accurately outlining their contours and providing a foundation for subsequent feature point extraction and matching. Next, feature points of the welding points, such as shape, size, or color, are extracted and matched with preset standard feature points. Utilizing the accuracy of feature point matching, precise alignment of the welding points is achieved, improving welding precision. The welding points are illuminated using a specific wavelength light source, with the brightness and illumination angle adjusted according to actual needs. The impact of lighting conditions on image capture and feature point recognition is fully considered. By optimizing the light source parameters, clear and accurate images of the welding points are captured, improving alignment reliability. Regarding image capture and processing, the captured images are processed in real time to identify the position of the welding points and compare them with preset positions. This enables dynamic monitoring and real-time adjustment of the welding points, ensuring the stability and accuracy of the welding process. Finally, based on the identification results, the positions of the first and / or second substrates are automatically adjusted to achieve precise alignment of the welding points, demonstrating the automation and intelligence of the control method. Real-time feedback and adjustment ensure smooth welding processes and high welding quality standards.

[0043] The present invention is further configured such that: in step S3, the step of analyzing the image data captured by the camera according to the image processing algorithm, calculating the deviation between the current position of the chip soldering point and the target soldering position of the second substrate, and generating motion commands to drive the first substrate and / or the second substrate to move includes:

[0044] Based on the image processing algorithm, feature points of the welding point and the target welding position are extracted from the image, and the extracted feature points of the welding point are matched with the feature points of the target welding position.

[0045] Based on the matched feature points, the precise location of the welding point in the current image is calculated using a geometric transformation method;

[0046] The calculated welding point position is compared with the target welding position, and the deviation between the two is calculated.

[0047] Based on the calculated deviation, motion commands are generated to drive the first substrate and / or the second substrate to move.

[0048] Based on deep learning algorithms, feature extraction and matching are performed on images captured from multiple angles, and overall welding point information is obtained through image fusion technology;

[0049] During the movement, the positional changes of the welding point are analyzed in real time, and the movement commands are dynamically adjusted based on the analysis results.

[0050] In this technical solution, firstly, feature points of the welding point and the target welding position are extracted from the image using image processing algorithms and precisely matched to ensure the accuracy of feature point extraction and the efficiency of matching, laying a solid foundation for subsequent position calculation and deviation analysis. Next, the precise position of the welding point in the current image is calculated using geometric transformation methods and compared with the target welding position to determine the deviation between the two. Through precise mathematical calculations, a quantitative analysis of the welding point position is achieved, providing a scientific basis for the generation of motion commands. When generating motion commands, parameters such as the movement range, speed, and acceleration of the first and second substrates are fully considered to ensure the smoothness and accuracy of the motion, reflecting the refinement and intelligence of the control method. This allows for flexible adjustment of motion parameters according to actual conditions to adapt to different welding requirements. Furthermore, the description also mentions using deep learning algorithms to extract and match features from images captured from multiple angles, and obtaining overall welding point information through image fusion technology. This fully utilizes the powerful capabilities of deep learning in image processing, achieving comprehensive, multi-angle analysis of the welding point and improving welding accuracy and reliability. During the process, the positional changes of the welding point are analyzed in real time, and the motion commands are dynamically adjusted based on the analysis results. This demonstrates the real-time and dynamic nature of the control method, which can adjust the motion strategy in a timely manner according to the actual situation in the welding process to ensure the smooth progress of the welding process.

[0051] The present invention is further configured such that: in step S5, the real-time detection of the welding result by a camera and the adjustment of the swing arm, the first substrate, and / or the second substrate based on the detection result include:

[0052] Based on the size, shape deviation, voids and cracks of the weld joints, the welding results are automatically detected to identify potential welding defects.

[0053] Based on the image analysis results, the deviation between the actual position and the ideal position of the welding point, as well as the deviation between the welding quality index and the standard value, are calculated.

[0054] Based on the deviation calculation results, the position and orientation of the swing arm, the first substrate, and / or the second substrate are dynamically adjusted.

[0055] In this technical solution, the real-time monitoring of welding results via camera is crucial. It not only ensures immediate feedback on welding quality but also provides accurate data for subsequent adjustments. By automatically detecting potential defects such as weld joint size, shape deviations, voids, and cracks, the system can quickly identify problems during the welding process, thus preventing defective products. Based on image analysis, the system calculates the deviation between the actual and ideal positions of the weld joints, as well as the deviation between welding quality indicators and standard values. This deviation calculation provides data support for subsequent dynamic adjustments, making the adjustment process more scientific and accurate. Based on the deviation calculation results, the system can dynamically adjust the position and orientation of the swing arm, the first substrate, and / or the second substrate to correct deviations in the welding process in real time, ensuring optimal welding quality. This not only improves welding accuracy and stability but also significantly enhances the production efficiency and product qualification rate of the semiconductor die bonder.

[0056] The present invention further specifies that the calculation formula for the welding quality index is:

[0057] ;

[0058] in, and It is a weighting coefficient, and + =1, Here, is the actual area of ​​the solder joint, Aideal is the ideal area of ​​the solder joint, and h is the height of the solder joint, with the ideal height ranging from [value missing]. Q represents the welding quality index.

[0059] The present invention is further configured such that: in step S6, based on the prediction result information, a corresponding action instruction set is retrieved from a pre-built database. The action instruction set includes a swing arm instruction, a first substrate action instruction, and a second substrate action instruction, used to adjust the position or state of the swing arm, the first substrate, and the second substrate to optimize the subsequent welding process or correct deviations in the current welding process, including:

[0060] The welding process uses cameras, temperature sensors, and pressure sensors to collect images, temperature, and pressure data of the welding points in real time.

[0061] Integrate historical data from past welding processes, including welding results under different operating parameters, chip type, and substrate material data;

[0062] The collected real-time and historical data are cleaned, normalized, and preprocessed with feature extraction. The preprocessed historical data is then used to train the deep neural network model.

[0063] During the welding process, real-time collected images, temperature, and pressure data are input into a trained deep neural network model;

[0064] Based on the prediction results output by the model, it is determined whether the current welding process meets the requirements. If the prediction results show that the welding quality is not good, or there is room for optimization of the working parameters, the parameters are adjusted. The working parameters of the solder spraying device are dynamically adjusted through the optimization algorithm. The optimization algorithm calculates new parameter values ​​based on the prediction results and the current working parameters so that the welding quality is expected to reach the optimal level.

[0065] This technical solution achieves comprehensive monitoring and intelligent analysis of the welding process by comprehensively applying sensor technology and machine learning algorithms. It not only collects multi-dimensional data (images, temperature, and pressure) of the welding points in real time, but also integrates historical welding data, including key information such as welding results under different operating parameters, chip type, and substrate material. Preprocessing provides a high-quality data foundation for subsequent machine learning model training. The trained deep neural network model can receive real-time data during the welding process and quickly predict the current welding result and the effect of operating parameters. This allows the system to promptly identify potential welding problems and dynamically adjust the operating parameters of the solder spraying device through optimization algorithms, thereby ensuring the stability and consistency of welding quality.

[0066] The present invention is further configured such that after step S6, the method further includes: after confirming that the welding quality meets the requirements, controlling the swing arm to release the chip that has been welded from the second substrate and preparing to perform the welding operation of the next chip.

[0067] This technical solution not only emphasizes the system's automation and intelligence but also highlights its high efficiency and continuity. The control method seamlessly connects each welding step, ensuring a smooth welding process while guaranteeing welding quality and production efficiency. The coherent workflow design is of great significance for improving the overall performance and reliability of the semiconductor die bonder.

[0068] The beneficial effects of the present invention are as follows: (1) Real-time monitoring and feedback are performed by a camera, combined with a control strategy, so that the position and orientation of the chip can be sensed in real time during the entire die bonding process, and can be flexibly adjusted according to the actual situation, thereby ensuring the accurate alignment of the chip on the target substrate; (2) The chip is adsorbed by the vacuum adsorption device on the swing arm, and the appearance is inspected by the camera, ensuring the integrity and accuracy of the chip before transfer, and determining the optimal solder point of the chip, laying the foundation for subsequent alignment and soldering operations. The swing arm is controlled to position the optimal solder point of the chip downward, and the initial position image of the solder point on the second substrate is captured by the camera for initial alignment. The high-precision positioning capability of the camera is used to ensure the initial alignment accuracy between the chip and the substrate. The image data captured by the camera is analyzed according to the image processing algorithm, the deviation between the current position of the chip solder point and the target soldering position of the second substrate is calculated, and motion commands are generated to drive the first substrate and / or the second substrate to make fine adjustments, thereby realizing the precise control of the soldering process and ensuring the accurate alignment of the solder points. Attached Figure Description

[0069] Figure 1 This is a schematic diagram of the structure of a semiconductor die bonder according to the present invention;

[0070] Figure 2 This is a flowchart of a control method for a semiconductor die bonder according to the present invention.

[0071] In the diagram: 1. Frame; 2. Swing arm; 3. Rotating shaft; 4. First substrate; 5. Second substrate; 6. Camera; 7. Hollow disk. Detailed Implementation

[0072] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only one preferred embodiment of this invention and are only used to explain this invention. They do not limit the scope of protection of this invention. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0073] like Figure 1 As shown, as an embodiment of the present invention, a semiconductor die bonder includes:

[0074] Rack 1;

[0075] The swing arm 2 is rotatably connected to the frame 1 via a rotating shaft 3. The swing arm 2 is driven by a first drive source, which is controlled by a controller.

[0076] The first substrate 4 and the second substrate 5 are respectively arranged below the swing arm 2 and can move back and forth or left and right relative to the rotating shaft 3 in the horizontal plane. The movement of the first substrate 4 and the second substrate 5 is controlled by the second driving source and the third driving source, respectively. The second driving source and the third driving source are both controlled by the controller.

[0077] A vacuum adsorption device is mounted on the swing arm 2 to move the chip from the first substrate 4 to the second substrate 5.

[0078] A solder spraying device, mounted on the swing arm 2, includes a soldering head and is used to spray solder onto the chip.

[0079] A heating device is disposed on the second substrate 5 for local heating of the welding area;

[0080] Camera 6 is disposed above and below the first substrate 4 and the second substrate 5, and is used to capture image information of the chip, the first substrate 4 and the second substrate 5. Camera 6 is connected to the controller to transmit the captured image information to the controller.

[0081] The controller executes the following linkage control strategy:

[0082] During the welding process, camera 6 captures image information of the current welding point and acquires image information of the target area for the next welding while simultaneously performing the previous welding operation.

[0083] Based on the image processing algorithm, the captured image information is analyzed to determine the optimal welding strategy for the current welding point and the optimal welding point for the next welding target area;

[0084] Based on the determined optimal welding strategy and optimal welding point, control commands are generated. The controller simultaneously controls the movement of the swing arm 2, the first substrate 4, and the second substrate 5, so that while the swing arm 2 is welding the current chip to the second substrate 5, the first substrate 4 moves to the position of the next chip to be welded, and the second substrate 5 moves to the target position for the next welding.

[0085] In this embodiment, the swing arm 2 is flexibly rotatably connected to the frame 1 through a rotating shaft 3 and is precisely driven by a first driving source to achieve fast and stable movement. The first substrate 4 and the second substrate 5 are respectively arranged below the swing arm 2 and can move freely in the horizontal plane and are controlled by independent second and third driving sources, ensuring fast switching and precise positioning of the chip between different positions. The vacuum adsorption device and the solder spraying device are ingeniously arranged on the swing arm 2 to achieve fast transfer and precise welding of the chip. The heating device is arranged on the second substrate 5 to locally heat the welding area, improving the welding quality and reliability. The camera 6 is arranged omnidirectionally and can capture the image information of the chip, the first substrate 4 and the second substrate 5 in real time, providing rich data support for the controller. The controller executes a precise linkage control strategy, quickly analyzes the image information through an image processing algorithm, and determines the optimal welding strategy and welding position. During the welding process, the controller can simultaneously control the movement of the swing arm 2, the first substrate 4 and the second substrate 5, realizing the synchronization and high efficiency of the welding process. While the swing arm 2 welds the current chip to the second substrate 5, the first substrate 4 has moved to the position of the next chip to be welded, and the second substrate 5 has also moved to the target position for the next welding, greatly shortening the welding cycle and improving the production efficiency.

[0086] It can be understood that the controller executes the following linkage control strategy to achieve the optimal control strategy:

[0087] During the welding process, the image information of the current welding point is captured by the camera, and the image information of the target welding area for the next welding is obtained while the previous welding is in progress;

[0088] According to the image processing algorithm, analyze the captured image information, calculate the deviation between the current position of the chip welding point and the target welding position, and determine the optimal alignment strategy for the current welding point;

[0089] At the same time, based on the machine learning algorithm, analyze the real-time data and historical data during the welding process, predict and optimize the working parameters of the solder spraying device, such as the solder volume, spraying speed, heating temperature, to form an optimal welding parameter strategy;

[0090] According to the determined optimal alignment strategy and optimal welding parameter strategy, generate control instructions, and simultaneously control the movement of the swing arm, the first substrate, and the second substrate through the controller, so that while the swing arm welds the current chip to the second substrate, the first substrate moves to the position of the next chip to be welded, and the second substrate moves to the target position for the next welding, to achieve the synchronization, high efficiency and high quality of the welding process.

[0091] In one embodiment of the present invention, both the first substrate 4 and the second substrate 5 are provided with a cutout disk 7, and the chip is arranged on the cutout disk 7. Firstly, the cutout disk 7 on both the first substrate 4 and the second substrate 5 allows the chip to be arranged more orderly on the disk surface, facilitating subsequent transfer and soldering operations. The cutout structure not only reduces the weight of the substrate but also improves the ventilation and heat dissipation of the chip, helping to maintain the stability and reliability of the chip during the die bonding process.

[0092] Understandably, the motor drives the rotating shaft 3, thereby causing the swing arm 2 to rotate between the first substrate 4 and the second substrate 5.

[0093] like Figure 2 As shown in Embodiment 2 of the present invention, a control method for a semiconductor die bonder includes the following steps:

[0094] S1, the chip on the first substrate 4 is adsorbed by the vacuum adsorption device on the swing arm 2, and the chip is visually inspected by the camera 6 to determine the optimal soldering point of the chip.

[0095] S2, control the swing arm 2 to position the optimal solder point of the chip downwards, capture the initial position image of the solder point on the second substrate 5 through the camera 6, and perform initial alignment;

[0096] S3, based on the image processing algorithm, analyze the image data captured by the camera 6, calculate the deviation between the current position of the chip soldering point and the target soldering position of the second substrate 5, and generate motion commands to drive the first substrate 4 and / or the second substrate 5 to move until the soldering point is aligned.

[0097] S4, activate the heating device on the second substrate 5 to preheat the second substrate 5, and after confirming that the solder joints are aligned, control the solder spraying device to spray solder onto the solder joints.

[0098] S5. After the solder coating is completed, the welding result is detected in real time by camera 6 to obtain the detection result; based on the detection result, the welding quality is predicted by a pre-trained machine learning model to obtain the prediction result information.

[0099] S6. Based on the prediction results, retrieve the corresponding action instruction set from the pre-built database. The action instruction set includes the swing arm 2 instruction, the first substrate 4 action instruction, and the second substrate 5 action instruction, which are used to adjust the position or state of the swing arm 2, the first substrate 4, and the second substrate 5 to optimize the subsequent welding process or correct the deviation in the current welding.

[0100] In this embodiment, firstly, the chip is adsorbed by the vacuum adsorption device on the swing arm 2, and visual inspection is performed using the camera 6 to ensure the integrity and accuracy of the chip before transfer. Simultaneously, the optimal solder joint of the chip is determined, laying the foundation for subsequent alignment and soldering operations. Next, the swing arm 2 is controlled to position the optimal solder joint of the chip downwards, and the camera 6 captures an initial position image of the solder joint on the second substrate 5 for preliminary alignment. The high-precision positioning capability of the camera 6 ensures the initial alignment accuracy between the chip and the substrate. Then, the image data captured by the camera 6 is analyzed using an image processing algorithm to calculate the deviation between the current position of the chip solder joint and the target soldering position on the second substrate 5, and motion commands are generated to drive the first substrate 4 and / or the second substrate 5 for fine-tuning. This achieves precise control of the soldering process and ensures accurate alignment of the solder joint. During the soldering preparation stage, the heating device on the second substrate 5 is activated to preheat the substrate, providing suitable temperature conditions for subsequent solder spraying. After confirming the solder joint alignment, solder is sprayed onto the solder joint using a solder spraying device, completing the chip soldering process. After welding is completed, the welding result is monitored in real time by camera 6. Based on the monitoring results, necessary adjustments are made to the swing arm 2, the first substrate 4, and / or the second substrate 5 to ensure welding quality. This demonstrates the feedback mechanism of the control method, ensuring the stability and reliability of the welding result through real-time monitoring and adjustment. Finally, based on machine learning algorithms, real-time and historical data during the welding process are analyzed to predict and optimize the operating parameters of the solder spraying device, showcasing the intelligence level of the control method. Through data analysis and algorithm optimization, welding efficiency and quality are continuously improved.

[0101] Understandably, the generation of the pre-trained machine learning model includes: collecting a large amount of welding process data, including images of welding points and welding quality indicators; performing preprocessing operations such as data cleaning, labeling, and normalization; learning and extracting key features from the training data according to a neural network algorithm to predict welding quality; adjusting model parameters through optimization algorithms to minimize the loss function so that the model can better fit the training data; evaluating the model's performance on independent validation and test sets to ensure its good generalization ability; deploying the trained machine learning model into the control system of the semiconductor die bonder; and during actual production, the welding result data acquired by camera 6 will be input into the model, and the model will output prediction result information to guide subsequent adjustments and optimization operations.

[0102] Understandably, the database contains action instruction sets corresponding to various welding conditions. These action instruction sets are designed based on a large amount of historical welding data and experimental results to ensure the accuracy and stability of welding.

[0103] In one embodiment of the present invention, step S1, which involves using camera 6 to perform visual inspection on the chip and determine the optimal solder joint of the chip, includes:

[0104] The electrodes and pads on the chip are identified using image recognition methods. The flatness and cleanliness of the chip surface are evaluated, and areas with no surface defects and clear electrodes are selected as potential soldering points.

[0105] Collect chip image data, train a solder joint model based on machine learning algorithm, automatically identify the best soldering area on the chip, input real-time captured chip images into the trained solder joint model, and the solder joint model predicts the position of the optimal solder joint by comparing the features in the input image with those in the training data.

[0106] A laser ranging method is used to perform three-dimensional topography detection on the chip surface, construct and analyze a three-dimensional model of the chip surface, and determine the flattest and defect-free area on the chip surface as the soldering point by analyzing the surface undulation and defect information in the three-dimensional model.

[0107] The temperature distribution of the chip after preheating is detected by thermal imaging. The temperature distribution in the thermal imaging image is analyzed by temperature analysis algorithm, and the corresponding temperature distribution uniformity value is calculated. The area with a temperature distribution uniformity value less than a preset threshold is selected as the welding point.

[0108] In this technical solution, firstly, key structures such as electrodes and pads on the chip are identified using image recognition methods. Simultaneously, the physical characteristics of the chip surface, such as flatness and cleanliness, are evaluated. Advanced image processing technology is utilized to quickly and accurately acquire basic information about the chip surface, providing crucial information for subsequent solder point selection. Next, chip image data is collected, and a machine learning algorithm is used to train a model. Leveraging the advantages of big data and artificial intelligence, the model automatically identifies the optimal soldering area on the chip through learning from a large number of samples. During real-time operation, simply inputting the captured chip image into the trained model allows for rapid prediction of the optimal solder point location, significantly improving detection efficiency and accuracy. Furthermore, laser ranging is employed to perform three-dimensional topography detection on the chip surface. Constructing and analyzing a three-dimensional model of the chip surface provides a more intuitive view of its topographic features, helping to determine the flattest, defect-free area as the solder point. Through three-dimensional topography detection, the position and shape of the solder points can be controlled more precisely, improving soldering quality. Finally, thermal imaging was used to detect the temperature distribution of the chip after preheating. Taking into account the thermal characteristics of the chip during the preheating process, by selecting areas with uniform temperature distribution and good heat dissipation performance as soldering points, the thermal stress during the soldering process can be further reduced, thereby improving the reliability and lifespan of the chip.

[0109] Understandably, the solder joint model can automatically identify the optimal soldering area on the chip by learning features (such as electrode shape, pad position, surface defects, etc.) from a large amount of chip image data.

[0110] It is understood that the image recognition method is a feature-based image recognition method, which performs recognition by extracting key features (such as edges, corners, and textures) in the image, and uses an edge detection method to identify the edges of the welding points.

[0111] Furthermore, in step S2, the camera 6 captures a preliminary position image of the solder joint on the second substrate 5, and preliminary alignment is performed, including:

[0112] Based on the edge detection method, the edge of the welding point is identified, and the feature points of the welding point are extracted and matched with the preset standard feature points;

[0113] Illuminate the welding point with a light source of a specific wavelength, and adjust the brightness and angle of the light source according to actual needs;

[0114] The captured images are processed in real time to identify the location of the welding point and compare it with the preset location;

[0115] Based on the recognition results, the positions of the first substrate 4 and / or the second substrate 5 are automatically adjusted.

[0116] In this technical solution, firstly, the edges of the welding points are identified using edge detection methods, accurately outlining their contours and providing a foundation for subsequent feature point extraction and matching. Next, feature points of the welding points, such as shape, size, or color, are extracted and matched with preset standard feature points. Utilizing the accuracy of feature point matching, precise alignment of the welding points is achieved, improving welding precision. The welding points are illuminated using a specific wavelength light source, with the brightness and illumination angle adjusted according to actual needs. The impact of lighting conditions on image capture and feature point recognition is fully considered. By optimizing the light source parameters, clear and accurate images of the welding points are captured, improving alignment reliability. Regarding image capture and processing, the captured images are processed in real-time to identify the position of the welding points and compare them with preset positions. This enables dynamic monitoring and real-time adjustment of the welding points, ensuring the stability and accuracy of the welding process. Finally, based on the identification results, the positions of the first substrate 4 and / or the second substrate 5 are automatically adjusted to achieve precise alignment of the welding points. This demonstrates the automation and intelligence of the control method. Through real-time feedback and adjustment, the smooth progress of the welding process and high welding quality standards are ensured.

[0117] Understandably, the characteristic features of a weld joint are its shape, size, or color.

[0118] Understandably, a light source of a specific wavelength refers to a laser source in the near-infrared band (such as 800-1100nm), because lasers of these wavelengths can be well absorbed by metal materials, improving welding efficiency and quality. Furthermore, tin-based solder has good absorption capacity in the near-infrared band, and lasers of these wavelengths can effectively heat and melt tin-based solder, achieving good welding results.

[0119] In one embodiment of the present invention, step S3, which involves analyzing the image data captured by the camera 6 according to the image processing algorithm, calculating the deviation between the current position of the chip soldering point and the target soldering position of the second substrate 5, and generating motion commands to drive the first substrate 4 and / or the second substrate 5 to move, includes:

[0120] Based on the image processing algorithm, feature points of the welding point and the target welding position are extracted from the image, and the extracted feature points of the welding point are matched with the feature points of the target welding position.

[0121] Based on the matched feature points, the precise location of the welding point in the current image is calculated using a geometric transformation method;

[0122] The calculated welding point position is compared with the target welding position, and the deviation between the two is calculated.

[0123] Based on the calculated deviation, motion commands are generated to drive the first substrate 4 and / or the second substrate 5 to move.

[0124] Based on deep learning algorithms, feature extraction and matching are performed on images captured from multiple angles, and overall welding point information is obtained through image fusion technology;

[0125] During the movement, the positional changes of the welding point are analyzed in real time, and the movement commands are dynamically adjusted based on the analysis results.

[0126] In this technical solution, firstly, feature points of the welding point and the target welding position are extracted from the image using image processing algorithms and precisely matched to ensure the accuracy of feature point extraction and the efficiency of matching, laying a solid foundation for subsequent position calculation and deviation analysis. Next, the precise position of the welding point in the current image is calculated using geometric transformation methods and compared with the target welding position to determine the deviation between the two. Through precise mathematical calculations, a quantitative analysis of the welding point position is achieved, providing a scientific basis for the generation of motion commands. When generating motion commands, parameters such as the movement range, speed, and acceleration of the first substrate 4 and the second substrate 5 are fully considered to ensure the smoothness and accuracy of the motion, reflecting the refinement and intelligence of the control method. This allows for flexible adjustment of motion parameters according to actual conditions to adapt to different welding requirements. Furthermore, the description also mentions using deep learning algorithms to extract and match features from images captured from multiple angles, and obtaining overall welding point information through image fusion technology. This fully utilizes the powerful capabilities of deep learning in image processing, achieving comprehensive, multi-angle analysis of the welding point and improving welding accuracy and reliability. During the process, the positional changes of the welding point are analyzed in real time, and the motion commands are dynamically adjusted based on the analysis results. This demonstrates the real-time and dynamic nature of the control method, which can adjust the motion strategy in a timely manner according to the actual situation in the welding process to ensure the smooth progress of the welding process.

[0127] Understandably, the image processing algorithms mentioned are SIFT, SURF, and ORB.

[0128] Understandably, the deviation can be expressed as the coordinate difference in a two-dimensional plane.

[0129] Understandably, the motion command takes into account the range of movement, speed, and acceleration parameters of the first substrate 4 and the second substrate 5.

[0130] Understandably, in image processing, feature points refer to pixels in an image that have significant features (such as corner points, edge points, texture change points, etc.). These feature points are unique and stable in the image and can be used for image matching and localization. Extracting these significant feature points from an image is usually achieved using specific algorithms (such as SIFT, SURF, ORB, etc.).

[0131] Understandably, in image processing, geometric transformation refers to the process of transforming an image from one coordinate system to another. Commonly used geometric transformation methods include translation, rotation, scaling, affine transformation, and perspective transformation. In the control methods of semiconductor die bonders, geometric transformation methods are used to calculate the precise position of the welding point in the current image based on the matched feature points. For example, if the coordinates of the target welding position in the image are known, as well as the relative positional relationship between the welding point feature points and the target feature points (such as described by affine transformation), the precise coordinates of the welding point in the current image can be calculated through geometric transformation.

[0132] Understandably, in the control methods of semiconductor die bonders, deep learning algorithms can be used to extract features of the bonding points from images captured from multiple angles. Information from multiple images is fused to obtain more comprehensive and accurate information. In the control methods of semiconductor die bonders, image fusion technology can be used to obtain overall bonding point information, improving the accuracy and reliability of the bonding process.

[0133] In step S5, the step of using camera 6 to detect the welding result in real time and adjusting the swing arm 2, the first substrate 4, and / or the second substrate 5 based on the detection result includes:

[0134] Based on the size, shape deviation, voids and cracks of the weld joints, the welding results are automatically detected to identify potential welding defects.

[0135] Based on the image analysis results, the deviation between the actual position and the ideal position of the welding point, as well as the deviation between the welding quality index and the standard value, are calculated.

[0136] Based on the deviation calculation results, the position and orientation of the swing arm 2, the first substrate 4 and / or the second substrate 5 are dynamically adjusted.

[0137] In this technical solution, the real-time monitoring of the welding results via camera 6 is crucial. It not only ensures immediate feedback on welding quality but also provides accurate data for subsequent adjustments. By automatically detecting potential defects such as weld joint size, shape deviations, voids, and cracks, the system can quickly identify problems during the welding process, thus preventing defective products. Based on image analysis, the system calculates the deviation between the actual and ideal positions of the weld joints, as well as the deviation between welding quality indicators and standard values. This deviation calculation provides data support for subsequent dynamic adjustments, making the adjustment process more scientific and accurate. Based on the deviation calculation results, the system can dynamically adjust the position and orientation of the swing arm 2, the first substrate 4, and / or the second substrate 5 to correct deviations in the welding process in real time, ensuring optimal welding quality. This not only improves welding accuracy and stability but also significantly enhances the production efficiency and product qualification rate of the semiconductor die bonder.

[0138] Understandably, the welding quality indicators include weld area, height, and uniformity.

[0139] Understandably, the formula for calculating the welding quality index is as follows:

[0140] ;

[0141] in, and It is a weighting coefficient, and + =1, Here, is the actual area of ​​the solder joint, Aideal is the ideal area of ​​the solder joint, and h is the height of the solder joint, with the ideal height ranging from [value missing]. Q represents the welding quality index.

[0142] In this formula, the welding quality index Q takes into account both the area and height of the weld joint and is used to evaluate the welding quality.

[0143] In step S6, based on the prediction results, the corresponding action instruction set is retrieved from a pre-built database. This action instruction set includes swing arm instructions, first substrate action instructions, and second substrate action instructions, used to adjust the position or state of the swing arm, first substrate, and second substrate to optimize the subsequent welding process or correct deviations in the current welding process.

[0144] The camera 6, temperature sensor, and pressure sensor collect images, temperature, and pressure data of the welding point in real time during the welding process.

[0145] Integrate historical data from past welding processes, including welding results under different operating parameters, chip type, and substrate material data;

[0146] The collected real-time and historical data are cleaned, normalized, and preprocessed with feature extraction. The preprocessed historical data is then used to train the deep neural network model.

[0147] During the welding process, real-time collected images, temperature, and pressure data are input into a trained deep neural network model;

[0148] Based on the prediction results output by the model, it is determined whether the current welding process meets the requirements. If the prediction results show that the welding quality is not good, or there is room for optimization of the working parameters, the parameters are adjusted. The working parameters of the solder spraying device are dynamically adjusted through the optimization algorithm. The optimization algorithm calculates new parameter values ​​based on the prediction results and the current working parameters so that the welding quality is expected to reach the optimal level.

[0149] In this embodiment, by comprehensively applying sensor technology and machine learning algorithms, comprehensive monitoring and intelligent analysis of the welding process are achieved. It can not only collect multi-dimensional data (images, temperature, and pressure) of the welding points in real time, but also integrate historical welding data, including key information such as welding results under different operating parameters, chip type, and substrate material. Preprocessing provides a high-quality data foundation for subsequent machine learning model training. The trained deep neural network model can receive real-time data during the welding process and quickly predict the current welding result and the effect of operating parameters. This allows the system to promptly identify potential welding problems and dynamically adjust the operating parameters of the solder spraying device through optimization algorithms, thereby ensuring the stability and consistency of welding quality.

[0150] Understandably, the relationship between the prediction results and the parameters is established through the mapping relationship learned by the model during training. The optimization process will continuously iterate until the best combination of parameters is found.

[0151] Understandably, the optimization algorithms mentioned are gradient descent and genetic algorithms.

[0152] Following step S6, the process further includes: after confirming that the welding quality meets the requirements, controlling the swing arm 2 to release the welded chip from the second substrate 5 and preparing for the next chip welding operation. This not only emphasizes the system's automation and intelligence but also highlights its high efficiency and continuity. The control method seamlessly connects each welding step, ensuring a smooth welding process while guaranteeing welding quality and production efficiency. A coherent workflow design is crucial for improving the overall performance and reliability of semiconductor die bonders.

[0153] The above embodiments, which describe the specific features of the present invention, are only used to further illustrate the present invention and should not be construed as limiting the scope of protection of the present invention. Any non-essential improvements and adjustments made to the present invention by those skilled in the art based on the above description of the invention shall fall within the scope of protection of the present invention.

Claims

1. A semiconductor die bonder, characterized in that, include: frame; A swing arm is rotatably connected to the frame via a rotating shaft. The swing arm is driven by a first drive source, which is controlled by a controller. A first substrate and a second substrate, wherein the first substrate is used to support the chip to be soldered, and the second substrate is used to receive and solder the chip. The first substrate and the second substrate are respectively disposed below the swing arm and can move back and forth or left and right relative to the pivot in the horizontal plane. The movement of the first substrate and the second substrate is controlled by a second drive source and a third drive source, respectively. Both the second drive source and the third drive source are controlled by a controller. A vacuum adsorption device, mounted on a swing arm, is used to move the chip from the first substrate to the second substrate; A solder spraying device, mounted on a swing arm, includes a soldering head and is used to spray solder onto chips. A heating device, mounted on the second substrate, is used to locally heat the welding area; A camera is positioned above and below the first substrate and the second substrate to capture image information of the chip, the first substrate and the second substrate. The camera is connected to a controller to transmit the captured image information to the controller. The controller executes the following linkage control strategy: During the welding process, the camera captures image information of the current welding point and acquires image information of the target area for the next welding while the previous welding was being performed. Based on the image processing algorithm, the captured image information is analyzed to determine the optimal welding strategy for the current welding point and the optimal welding position for the next welding target area. Based on the determined optimal welding strategy and optimal welding position, control commands are generated. The controller simultaneously controls the movement of the swing arm, the first substrate, and the second substrate, so that while the swing arm is welding the current chip to the second substrate, the first substrate moves to the position of the next chip to be welded, and the second substrate moves to the target position for the next welding.

2. The semiconductor die bonder according to claim 1, characterized in that, Both the first substrate and the second substrate are provided with a cutout disk, and the chip is arranged on the cutout disk.

3. A control method for a semiconductor die bonder, characterized in that, A semiconductor die bonder according to claim 1 or 2 includes the following steps: S1, the chip on the first substrate is adsorbed by the vacuum adsorption device on the swing arm, and the chip is visually inspected by a camera to determine the optimal soldering point of the chip. S2, control the swing arm to position the optimal solder point of the chip downwards, capture the initial position image of the solder point on the second substrate through the camera, and perform initial alignment; S3, based on the image processing algorithm, analyze the image data captured by the camera, calculate the deviation between the current position of the chip soldering point and the target soldering position of the second substrate, and generate motion commands to drive the first substrate and / or the second substrate to move until the soldering point is aligned. S4, activate the heating device on the second substrate to preheat the second substrate, and after confirming that the solder joints are aligned, control the solder spraying device to spray solder onto the solder joints. S5. After the solder coating is completed, the welding result is detected in real time by a camera to obtain the detection result; based on the detection result, the welding quality is predicted by a pre-trained machine learning model to obtain the prediction result information. S6. Based on the prediction results, retrieve the corresponding action instruction set from the pre-built database. The action instruction set includes swing arm instruction, first substrate action instruction, and second substrate action instruction, which are used to adjust the position or state of the swing arm, the first substrate, and the second substrate to optimize the subsequent welding process or correct the deviation in the current welding.

4. The control method for a semiconductor die bonder according to claim 3, characterized in that, In step S1, the step of using a camera to perform visual inspection of the chip and determine the optimal solder joints of the chip includes: The electrodes and pads on the chip are identified using image recognition methods. The flatness and cleanliness of the chip surface are evaluated, and areas with no surface defects and clear electrodes are selected as potential soldering points. Collect chip image data, train a solder joint model based on machine learning algorithm, automatically identify the best soldering area on the chip, input real-time captured chip images into the trained solder joint model, and the solder joint model predicts the position of the optimal solder joint by comparing the features in the input image with those in the training data. A laser ranging method is used to perform three-dimensional topography detection on the chip surface, construct and analyze a three-dimensional model of the chip surface, and determine the flattest and defect-free area on the chip surface as the soldering point by analyzing the surface undulation and defect information in the three-dimensional model. The temperature distribution of the chip after preheating is detected by thermal imaging. The temperature distribution in the thermal imaging image is analyzed by temperature analysis algorithm, and the corresponding temperature distribution uniformity value is calculated. The area with a temperature distribution uniformity value less than a preset threshold is selected as the welding point.

5. The control method for a semiconductor die bonder according to claim 3, characterized in that, In step S2, capturing a preliminary position image of the solder joint on the second substrate using a camera and performing preliminary alignment includes: Based on the edge detection method, the edge of the welding point is identified, and the feature points of the welding point are extracted and matched with the preset standard feature points; The welding point is illuminated with a light source of a specific wavelength. The brightness and illumination angle of the light source are adjusted according to actual needs. The light source of the specific wavelength is a laser light source in the near-infrared band of 800-1100nm. The captured images are processed in real time to identify the location of the welding point and compare it with the preset location; Based on the identification results, the positions of the first substrate and / or the second substrate are automatically adjusted.

6. The control method for a semiconductor die bonder according to claim 3, characterized in that, In step S3, the step of analyzing the image data captured by the camera according to the image processing algorithm, calculating the deviation between the current position of the chip soldering point and the target soldering position of the second substrate, and generating motion commands to drive the first substrate and / or the second substrate to move includes: Based on the image processing algorithm, feature points of the welding point and the target welding position are extracted from the image, and the extracted feature points of the welding point are matched with the feature points of the target welding position. Based on the matched feature points, the precise location of the welding point in the current image is calculated using a geometric transformation method; The calculated welding point position is compared with the target welding position, and the deviation between the two is calculated. Based on the calculated deviation, motion commands are generated to drive the first substrate and / or the second substrate to move. Based on deep learning algorithms, feature extraction and matching are performed on images captured from multiple angles, and overall welding point information is obtained through image fusion technology; During the movement, the positional changes of the welding point are analyzed in real time, and the movement commands are dynamically adjusted based on the analysis results.

7. The control method for a semiconductor die bonder according to claim 3, characterized in that, In step S5, the real-time detection of the welding results using a camera, and the adjustment of the swing arm, the first substrate, and / or the second substrate based on the detection results, includes: Based on the size, shape deviation, voids and cracks of the weld joints, the welding results are automatically detected to identify potential welding defects. Based on the image analysis results, the deviation between the actual position and the ideal position of the welding point, as well as the deviation between the welding quality index and the standard value, are calculated. Based on the deviation calculation results, the position and orientation of the swing arm, the first substrate, and / or the second substrate are dynamically adjusted.

8. The control method for a semiconductor die bonder according to claim 7, characterized in that, The formula for calculating the welding quality index is as follows: ； in, and It is a weighting coefficient, and + =1, Here, is the actual area of ​​the solder joint, Aideal is the ideal area of ​​the solder joint, and h is the height of the solder joint, with the ideal height ranging from [value missing]. Q represents the welding quality index.

9. The control method for a semiconductor die bonder according to claim 3, characterized in that, In step S6, based on the prediction results, the corresponding action instruction set is retrieved from a pre-built database. This action instruction set includes swing arm instructions, first substrate action instructions, and second substrate action instructions, used to adjust the position or state of the swing arm, first substrate, and second substrate to optimize the subsequent welding process or correct deviations in the current welding process. The welding process uses cameras, temperature sensors, and pressure sensors to collect images, temperature, and pressure data of the welding points in real time. Integrate historical data from past welding processes, including welding results under different operating parameters, chip type, and substrate material data; The collected real-time and historical data are cleaned, normalized, and preprocessed with feature extraction. The preprocessed historical data is then used to train the deep neural network model. During the welding process, real-time collected images, temperature, and pressure data are input into a trained deep neural network model; Based on the prediction results output by the model, it is determined whether the current welding process meets the requirements. If the prediction results show that the welding quality is not good, or there is room for optimization of the working parameters, the parameters are adjusted. The working parameters of the solder spraying device are dynamically adjusted through the optimization algorithm. The optimization algorithm calculates new parameter values ​​based on the prediction results and the current working parameters so that the welding quality is expected to reach the optimal level.

10. A control method for a semiconductor die bonder according to any one of claims 3 to 9, characterized in that, After step S6, the method further includes: after confirming that the welding quality meets the requirements, controlling the swing arm to release the chip that has been welded from the second substrate and preparing to perform the welding operation of the next chip.

Images