Electronic information intelligent processing recommendation system and method based on multi-modal data fusion
By combining RGB-D visual data and 3D point cloud data in a multimodal fusion manner, semiconductor packaging defects can be identified and process parameters can be adjusted. This solves the problem of insufficient accuracy and precision in identifying packaging defects in traditional methods, thereby improving the quality and performance of semiconductor packaging.
Patent Information
- Application Number
- CN202511471332.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-15
- Publication Date
- 2026-01-16
AI Technical Summary
Traditional electronic information intelligent processing recommendation systems and methods rely solely on single data points to determine whether semiconductor packaging has defects. This results in insufficient precision and accuracy in identifying packaging defects, failing to guarantee the performance of the packaged semiconductor.
An electronic information intelligent processing recommendation system based on multimodal data fusion is adopted. Combining RGB-D visual data and 3D point cloud data, the data analysis module obtains the type and cause of packaging defects, and the intelligent recommendation module adjusts process parameters to improve packaging quality.
It improves the effectiveness and accuracy of packaging defect identification, ensures semiconductor performance, and enables accurate identification of packaging defect types and effective adjustment of process parameters.
Smart Images
Figure CN121350348A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of data fusion technology, specifically to an intelligent electronic information processing recommendation system and method based on multimodal data fusion. Background Technology
[0002] Semiconductor packaging, as a key step in the semiconductor manufacturing process, plays a vital role in ensuring chip performance, improving system reliability, and achieving product miniaturization. With the continuous development of technology, electronic products are constantly evolving towards high performance, miniaturization, and multifunctionality, which places more stringent demands on semiconductor packaging technology.
[0003] Traditional intelligent processing recommendation systems and methods for electronic information rely solely on RGB-D visual data or 3D point cloud data of the packaged semiconductor to determine if packaging defects exist. If defects are found, various process parameters are collected, abnormal process parameters are identified, and these abnormal process parameters are adjusted to their standard ranges. Clearly, such intelligent processing recommendation systems and methods for electronic information have the following shortcomings: 1. Traditional intelligent processing recommendation systems and methods for electronic information rely solely on single data points to determine if semiconductor packaging defects exist. Single data detection has limited accuracy and cannot accurately identify complex packaging defect types, thus failing to guarantee the effectiveness of packaging defect identification.
[0004] Traditional electronic information intelligent processing recommendation systems and methods collect various process parameters after defect identification, and obtain abnormal process parameters, defect types and defect causes based on the values of each process parameter. However, some packaging defect types cause the same abnormal parameters, so it is impossible to determine the type of packaging defect based solely on the abnormal parameters. Therefore, the accuracy of packaging defect type identification cannot be guaranteed.
[0005] Traditional electronic information intelligent processing recommendation systems and methods only adjust abnormal process parameters to their standard value ranges when adjusting them, without analyzing the impact of abnormal process parameter values on the performance of the packaged semiconductor. Therefore, the performance of the semiconductor cannot be guaranteed after adjusting the abnormal process parameters. Summary of the Invention
[0006] To address the aforementioned technical shortcomings, the present invention aims to provide an electronic information intelligent processing recommendation system and method based on multimodal data fusion.
[0007] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: In a first aspect, the present invention provides an electronic information intelligent processing recommendation system based on multimodal data fusion, comprising the following modules: a data acquisition module, a data analysis module, a defect identification module, an intelligent recommendation module, and a database.
[0008] The data acquisition module is used to acquire RGB-D visual data and three-dimensional point cloud data of the semiconductor after semiconductor packaging, and to obtain process parameter text from the database.
[0009] The data analysis module is used to obtain RGB-D data, 3D point cloud data, process parameter text and packaging inspection results of semiconductors during each historical packaging process from the database, and to analyze the data feature set of each packaging defect type under different defect causes and the abnormal process parameters of each packaging defect type under each defect cause, as well as the data feature set when there are no packaging defects in the semiconductor packaging.
[0010] The defect identification module is used to acquire the RGB-D visual data, three-dimensional point cloud data and process parameter text of the semiconductor during this packaging process, and to determine whether there are defects in this semiconductor packaging. If there are defects, the module acquires the type of defect and the reason for the defect.
[0011] The intelligent recommendation module is used to recommend adjustment values for each process parameter based on the type and cause of the packaging defect when a packaging defect exists in the semiconductor packaging process, and then provide feedback to the staff.
[0012] The database is used to store RGB-D data, 3D point cloud data, process parameter text, packaging test results, performance test results and environment of semiconductors during each packaging process in history, as well as the packaging defect type, defect cause and the adjusted values of each abnormal historical package.
[0013] Secondly, the present invention provides an electronic information intelligent processing recommendation method based on multimodal data fusion, including the following steps: S1, data acquisition: after semiconductor packaging, the RGB-D visual data and three-dimensional point cloud data of the semiconductor are acquired, and the process parameter text is obtained from the database.
[0014] S2. Data Analysis: Obtain RGB-D data, 3D point cloud data, process parameter text, and packaging inspection results of semiconductors from the database for each historical packaging inspection. Analyze the RGB-D visual data, 3D point cloud data, and process parameter characteristics of semiconductors under different defect causes for each type of packaging defect, as well as the RGB-D visual data, 3D point cloud data, and process parameter characteristics of semiconductors when there are no packaging defects.
[0015] S3 Defect Identification: During this packaging inspection, the RGB-D visual data, 3D point cloud data, and process parameter text of the semiconductor are obtained, and it is determined whether there are defects in the semiconductor packaging. If defects are found, the type of defect and the cause of the defect are obtained.
[0016] S4 Intelligent Recommendation: When a semiconductor packaging defect occurs, it recommends adjusted values for each process parameter based on the defect type and cause, and provides feedback to the staff.
[0017] The beneficial effects of this invention are as follows: 1. This invention provides an electronic information intelligent processing recommendation system and method based on multimodal data fusion, which acquires RGB-D visual data, three-dimensional point cloud data and process parameter text of semiconductors, and retrieves various information of semiconductors from the database for each historical packaging process. It analyzes the data feature set of each packaging defect type under each defect cause, each abnormal process parameter and the standard data feature set of the semiconductor, and determines whether there is a packaging defect in this packaging. If there is, it acquires the packaging defect type, defect cause and each abnormal process parameter, analyzes the adjustment value of each abnormal process parameter and feeds it back to the staff, thus ensuring the performance of the semiconductor and the effectiveness and accuracy of packaging defect identification.
[0018] 2. This invention retrieves RGB-D data, 3D point cloud data, process parameter text, and packaging inspection results of semiconductors from a database for each historical packaging process. Historical packaging processes with abnormal results are designated as "abnormal packaging," while those with normal results are designated as "normal historical packaging." These abnormal historical packaging processes are further divided into historical packaging groups. Within each historical packaging group, data feature sets for each packaging defect type under various defect causes, as well as abnormal process parameters for each packaging defect type under various defect causes, are obtained. Simultaneously, a standard set of semiconductor data features is acquired, along with the RGB-D visual data features, 3D point cloud data features, and process parameter features of the semiconductor during the current packaging process. The system analyzes whether defects exist in this packaging process. If defects are found, the packaging defect type, defect cause, and abnormal process parameters are obtained, ensuring the effectiveness of packaging defect identification and the accuracy of packaging defect type identification.
[0019] 3. When a defect exists in the current packaging process, this invention acquires the environment of the current semiconductor packaging and the environments of previous semiconductor packaging processes, determines the first adjustment value of each abnormal process parameter, acquires the current value of each abnormal process parameter, and determines the second adjustment value of each abnormal process parameter based on the current value of each abnormal process parameter. Finally, based on the performance of each semiconductor under the first adjustment value of each abnormal process parameter and the performance of each semiconductor under the second adjustment value of each abnormal process parameter, the adjustment value of each abnormal process parameter is obtained, thus ensuring the performance of the semiconductor after adjusting each abnormal process parameter. Attached Figure Description
[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0021] Figure 1 This is a schematic diagram of the system structure connection of the present invention.
[0022] Figure 2 This is a schematic diagram of the implementation steps of the method of the present invention. Detailed Implementation
[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0024] Please see Figure 1 As shown, the present invention provides an intelligent electronic information processing and recommendation system based on multimodal data fusion, including: a data acquisition module, a data analysis module, a defect identification module, an intelligent recommendation module, and a database.
[0025] The data acquisition module is connected to the data analysis module, the data analysis module is connected to the defect identification module, the defect identification module is connected to the intelligent recommendation module, and the database is connected to the data analysis module and the intelligent recommendation module.
[0026] The data acquisition module is used to acquire RGB-D visual data and three-dimensional point cloud data of the semiconductor after semiconductor packaging, and to obtain process parameter text from the database.
[0027] It should be noted that RGB-D visual data includes color data and depth data, which are collected using an RGB camera and a depth sensor, and three-dimensional point cloud data are obtained by scanning the semiconductor surface with a laser beam.
[0028] Color data refers to the RGB values of the semiconductor surface, which can provide information such as the color, texture shape, and edges of the semiconductor. Depth data refers to the distance from each point on the semiconductor to the camera, which provides information such as the height, thickness, and relative position of each point on the semiconductor.
[0029] It should also be noted that the process parameter text includes the values of each process parameter at each packaging moment. Each process parameter includes temperature, pressure, ultrasonic power, and curing time, etc.
[0030] The data analysis module is used to obtain RGB-D data, 3D point cloud data, process parameter text and packaging inspection results of semiconductors during each historical packaging process from the database, and to analyze the data feature set of each packaging defect type under different defect causes and the abnormal process parameters of each packaging defect type under each defect cause, as well as the data feature set when there are no packaging defects in the semiconductor packaging.
[0031] It should be noted that semiconductor packaging defects include chip bonding defects, bonding defects, flip-chip bonding defects, and molding defects.
[0032] It should also be noted that the causes of defects in different packaging types are also different. For example, the causes of chip bonding defects include uneven application of adhesive, chip surface contamination, and substrate surface contamination. The causes of bonding defects include improper bonding parameters and excessive mechanical stress.
[0033] In a specific embodiment, the data analysis module performs the following process: it retrieves the RGB-D data, three-dimensional point cloud data, process parameter text, and packaging detection results of the semiconductor from the database for each historical packaging process, and refers to each historical packaging process with a packaging anomaly as an abnormal packaging process, and refers to each historical packaging process with a packaging normal process as a normal packaging process.
[0034] The database is used to retrieve the packaging defect types and causes of semiconductor packaging defects during each abnormal historical packaging, and these are compared. Abnormal historical packages with the same packaging defect type and the same defect cause are divided into a historical package group. This method is used to obtain each historical package group. In each historical package group, based on the RGB-D data, 3D point cloud data and process parameter text of the semiconductor during each abnormal historical packaging, the data feature set of each packaging defect type under each defect cause, as well as the abnormal process parameters of each packaging defect type under each defect cause, are obtained.
[0035] The defect types and causes of semiconductor packaging defects during normal historical packaging are obtained from the database. Following the method of obtaining the data feature set of each packaging defect type under each defect cause, the data feature set when there are no packaging defects in the semiconductor package is obtained, and this is called the standard data feature set.
[0036] The specific process for obtaining the data feature set of each packaging defect type under each defect cause is as follows: In each historical packaging group, the RGB-D data and 3D point cloud data of the semiconductor at each abnormal historical packaging time are obtained, and the values of each process parameter at each packaging time are obtained from the process parameter text. At the same time, the RGB-D data features, 3D point cloud data features, and process parameter features of the semiconductor at each abnormal historical packaging time are obtained. In each abnormal historical package, the RGB-D data features, 3D point cloud data features, and process parameter features of the semiconductor are fused to obtain a feature element. The feature elements of the historical packaging group are obtained in this way, and the feature elements are integrated to obtain the data feature set of the packaging defect type under the defect cause in the historical packaging group. The data feature set of each packaging defect type under each defect cause is obtained in this way.
[0037] It should be noted that deep learning technology is used to obtain the RGB-D data features, 3D point cloud data features, and various process parameter features of semiconductors.
[0038] It should also be noted that the feature elements include RGB-D data features, 3D point cloud data features, and features of various process parameters.
[0039] It needs to be explained that in each historical encapsulation group, each feature element is compared, and the same feature elements are divided into a feature element pile. Then, a feature element is randomly selected from the feature element pile as a marker feature element. This method is used to obtain each marker feature element. The marker feature elements are combined to obtain the data feature set of the historical encapsulation group. This method is used to obtain the data feature set of each historical encapsulation combination.
[0040] The above describes the process of obtaining abnormal process parameters for each type of packaging defect under each defect cause. The specific process is as follows: retrieve the process parameter text of the semiconductor during each normal historical packaging from the database, and at the same time, retrieve the values of each process parameter at each packaging moment during each normal historical packaging, and compare them to obtain the standard value range of each process parameter.
[0041] In each abnormal historical package of each historical package group, the process parameter text of the semiconductor is obtained, and the value of each process parameter at each packaging time is obtained from the process parameter text. The value of each process parameter at each packaging time is compared with the standard value range of each process parameter. If a certain process parameter has a value that is not within the range of the marked value of the process parameter, it means that the process parameter is an abnormal process parameter of the abnormal historical package. The abnormal process parameters of each abnormal historical package are obtained in this way.
[0042] The abnormal process parameters of each abnormal historical package in each historical package group are integrated to obtain the abnormal process parameter set of each historical package group. Each abnormal process parameter in the abnormal process parameter set of each historical package group is used as the abnormal process parameter of each package defect type under each defect cause in each historical package group. In this way, the abnormal process parameters of each package defect type under each defect cause are obtained.
[0043] It should be noted that the method for obtaining the set of abnormal process parameters is the same as the method for obtaining the set of data features.
[0044] The defect identification module is used to acquire the RGB-D visual data, three-dimensional point cloud data and process parameter text of the semiconductor during this packaging process, and to determine whether there are defects in this semiconductor packaging. If there are defects, the module acquires the type of defect and the reason for the defect.
[0045] In a specific embodiment, the defect identification module performs the following process: acquiring the RGB-D visual data, 3D point cloud data, and process parameter text of the semiconductor during this packaging process, and acquiring the RGB-D visual data features, 3D point cloud data features, and various process parameter features, and simultaneously fusing them to obtain the feature elements of this semiconductor packaging.
[0046] The characteristic elements of this semiconductor package are compared with the characteristic elements of the standard data characteristic set. If the characteristic elements of this semiconductor package are the same as one of the characteristic elements in the standard data characteristic set, it means that there is no packaging defect in this semiconductor package. If the characteristic elements of this semiconductor package are different from all the characteristic elements in the standard data characteristic set, it means that there is a packaging defect in this semiconductor package. At this time, the type and cause of the packaging defect of this semiconductor package are analyzed.
[0047] The above-mentioned analysis of the packaging defect type and defect cause of this semiconductor packaging is carried out in the following specific process: When there is a packaging defect in this semiconductor packaging, the feature elements of this semiconductor packaging are compared with the feature elements in the data feature set of each packaging defect type under each defect cause. If the feature elements of this semiconductor packaging are the same as a feature element in the data feature set of a certain packaging defect type under a certain defect cause, then it means that the packaging defect type and the defect cause are the packaging defect type and defect cause of this semiconductor packaging.
[0048] The intelligent recommendation module is used to recommend adjustment values for each process parameter based on the type and cause of the packaging defect when a packaging defect exists in the semiconductor packaging process, and then provide feedback to the staff.
[0049] In a specific embodiment, the intelligent recommendation module performs the following process: when there is a packaging defect in the semiconductor packaging, the packaging defect type and defect cause are obtained, and the abnormal process parameters of the semiconductor packaging are obtained according to the packaging defect type and defect cause, while the environment of the semiconductor packaging is also obtained.
[0050] The environment of each historical semiconductor packaging process is retrieved from the database and compared with the environment of the current semiconductor packaging process. Historical semiconductor packaging processes with the same environment as the current semiconductor packaging process are called marked historical packages. At the same time, the packaging inspection results of each marked historical package are obtained. Marked historical packages with normal packaging inspection results are called secondary marked historical packages. The values of each abnormal process parameter in each secondary marked historical package are obtained, and the average value of each abnormal process parameter in each secondary marked historical package is calculated and used as the first adjustment value of each abnormal parameter.
[0051] Simultaneously, the values of each abnormal process parameter are acquired, and based on these values, a second adjustment value for each abnormal process parameter is obtained. The adjustment value for each abnormal process parameter is then determined, and the adjustment value for each abnormal process parameter is fed back to the staff.
[0052] The above-mentioned process for obtaining the second adjustment value of each abnormal process parameter is as follows: each marked historical package whose packaging detection result is abnormal and has the same defect type and defect cause as the current packaging is referred to as each abnormal marked historical package, and each abnormal process parameter of each abnormal marked historical package is obtained.
[0053] The abnormal process parameters of each historical package marked with an anomaly are compared with the abnormal process parameters of the current semiconductor package. The historical packages marked with anomalies that have the same abnormal process parameters as the current semiconductor package are called the historical packages marked with anomalies for secondary marking. The similarity coefficient between the abnormal process parameters of each historical package marked with anomalies for secondary marking and the abnormal process parameters of the current semiconductor package is determined and compared. The historical package marked with anomalies for secondary marking with the largest similarity coefficient is selected. The adjusted values of each abnormal parameter in the historical package marked with anomalies for secondary marking are obtained from the database and used as the second adjustment values of each abnormal process parameter.
[0054] It should be noted that after normalizing the values of each abnormal process parameter, then... In the formula Representing the The first abnormal secondary marking history encapsulation An abnormal process parameter value, Representing the first semiconductor packaging An abnormal process parameter value, Representing the The similarity coefficient between the historical packaging anomaly process parameters and the current semiconductor packaging anomaly process parameters is determined by secondary marking of the anomaly. This represents the number of the encapsulation of each exception's secondary marking history. The numbers representing the abnormal process parameters, Represents the total number of abnormal process parameters. , and All are positive integers.
[0055] The specific process for determining the adjustment values of each abnormal process parameter is as follows: when each abnormal process parameter is at the first adjustment value of each abnormal process parameter, the performance test results of each semiconductor after packaging are obtained, and each semiconductor with a good performance test result is referred to as the semiconductor with good performance at the first adjustment value.
[0056] It should be noted that the historical values of each abnormal process parameter during each packaging process are obtained from the database and compared with the first adjustment value of each abnormal process parameter. The historical packaging processes that are the same as the first adjustment value of each abnormal process parameter and whose other process parameters are all within the standard value range are selected and referred to as the first value historical packaging processes. The performance test results of the semiconductors of the first value historical packaging processes are obtained from the database and used as the performance test results of each semiconductor after packaging when each abnormal process parameter is at the first adjustment value of each abnormal process parameter.
[0057] When each abnormal process parameter is at the second adjustment value of each abnormal process parameter, the performance test results of each semiconductor after packaging are obtained, and each semiconductor with a good performance test result is called the semiconductor with good performance at the second adjustment value.
[0058] It should be noted that the method for obtaining the performance test results of each semiconductor after packaging when each abnormal process parameter is at the second adjustment value of each abnormal process parameter is the same as the method for obtaining the performance test results of each semiconductor after packaging when each abnormal process parameter is at the first adjustment value of each abnormal process parameter.
[0059] Calculate and compare the proportion of high-performance semiconductors with the first adjustment value and the proportion of high-performance semiconductors with the second adjustment value. If the proportion of high-performance semiconductors with the first adjustment value is greater than the proportion of high-performance semiconductors with the second adjustment value, then the first adjustment value of each abnormal parameter is used as the adjustment value of each abnormal parameter. If the proportion of high-performance semiconductors with the second adjustment value is greater than the proportion of high-performance semiconductors with the first adjustment value, then the second adjustment value of each abnormal parameter is used as the adjustment value of each abnormal parameter.
[0060] The database is used to store RGB-D data, 3D point cloud data, process parameter text, packaging test results, performance test results and environment of semiconductors during each packaging process in history, as well as the packaging defect type, defect cause and the adjusted values of each abnormal historical package.
[0061] Please see Figure 2 As shown, the present invention provides an electronic information intelligent processing recommendation method based on multimodal data fusion, including: S1, data acquisition: after semiconductor packaging, RGB-D visual data and three-dimensional point cloud data of the semiconductor are acquired, and process parameter text is obtained from the database.
[0062] S2. Data Analysis: Obtain RGB-D data, 3D point cloud data, process parameter text, and packaging inspection results of semiconductors from the database for each historical packaging process. Analyze the data feature sets of each packaging defect type under different defect causes, the abnormal process parameters of each packaging defect type under each defect cause, and the data feature set when there are no packaging defects in the semiconductor packaging.
[0063] S3. Defect Identification: Obtain the RGB-D visual data, 3D point cloud data, and process parameter text of the semiconductor during this packaging process, and determine whether there are defects in this semiconductor packaging. If defects exist, obtain the type of defect and the reason for the defect.
[0064] S4. Intelligent Recommendation: When a packaging defect exists in the semiconductor packaging process, the system recommends adjustment values for each process parameter based on the type and cause of the defect and provides feedback to the staff.
[0065] This invention acquires RGB-D visual data, 3D point cloud data, and process parameter text of semiconductors, and retrieves information about semiconductors from historical packaging processes from a database. It analyzes the data feature sets of each packaging defect type under each defect cause, as well as the abnormal process parameters and the standard data feature set of the semiconductor, and determines whether a packaging defect exists in the current packaging. If a defect exists, it acquires the packaging defect type, defect cause, and abnormal process parameters, analyzes the adjustment values of each abnormal process parameter, and provides feedback to the staff. This ensures the performance of the semiconductor and guarantees the effectiveness and accuracy of packaging defect identification.
[0066] The above description is merely an example and illustration of the concept of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described or use similar methods to replace them, as long as they do not deviate from the concept of the invention or exceed the scope defined in this specification, they should all fall within the protection scope of the present invention.
Claims
1. An electronic information intelligent processing recommendation system based on multi-modal data fusion, characterized in that, The method comprises the following modules: The data acquisition module is used for acquiring RGB-D vision data and three-dimensional point cloud data of the semiconductor after semiconductor packaging, and acquiring process parameter texts from a database; The data analysis module is used for acquiring historical RGB-D data, three-dimensional point cloud data, process parameter texts and packaging detection results of the semiconductor during each packaging from the database, and analyzing data feature sets of each packaging defect type under different defect causes, each abnormal process parameter of each packaging defect type under each defect cause, and the data feature set when the semiconductor packaging does not have packaging defects; The defect identification module is used for acquiring the RGB-D vision data, three-dimensional point cloud data and process parameter texts of the semiconductor during this packaging, and judging whether the semiconductor packaging has defects, and if so, acquiring the type of the defects and the cause of the defects; The intelligent recommendation module is used for recommending the adjustment values of each process parameter according to the packaging defect type and the defect cause when the semiconductor packaging has packaging defects, and feeding back to the staff; The database is used for storing the historical RGB-D data, three-dimensional point cloud data, process parameter texts, packaging detection results, performance detection results and environment of the semiconductor during each packaging, and the packaging defect type, defect cause and values of each abnormal process parameter after adjustment of each abnormal historical packaging. 2.The electronic information intelligent processing recommendation system based on multi-modal data fusion according to claim 1, characterized in that, The data analysis module specifically comprises the following steps: Acquiring the historical RGB-D data, three-dimensional point cloud data, process parameter texts and packaging detection results of the semiconductor during each packaging from the database, and regarding the historical packaging with the packaging detection results as abnormal as each abnormal historical packaging, and regarding the historical packaging with the packaging detection results as normal as each normal historical packaging; Acquiring the packaging defect type and the defect cause of the semiconductor packaging defects during each abnormal historical packaging from the database, and comparing the packaging defect type and the defect cause, dividing each abnormal historical packaging with the same packaging defect type and the same defect cause into a historical packaging group, obtaining each historical packaging group in this way, and obtaining the data feature set of each packaging defect type under each defect cause and each abnormal process parameter of each packaging defect type under each defect cause according to the RGB-D data, three-dimensional point cloud data and process parameter texts of the semiconductor during each abnormal historical packaging; Acquiring the defect type and the defect cause of the semiconductor packaging defects during each normal historical packaging from the database, obtaining the data feature set when the semiconductor packaging does not have packaging defects according to the method of obtaining the data feature set of each packaging defect type under each defect cause, and regarding the data feature set as a standard data feature set. 3.The electronic information intelligent processing recommendation system based on multi-modal data fusion according to claim 2, characterized in that, The method of obtaining the data feature set of each packaging defect type under each defect cause specifically comprises the following steps: In each historical packaging group, the RGB-D data and three-dimensional point cloud data of the semiconductor at each abnormal historical packaging time are acquired, the values of each process parameter at each packaging time are acquired from the process parameter text, and the RGB-D data features, three-dimensional point cloud data features and process parameter features of the semiconductor at each abnormal historical packaging time are acquired. In each abnormal historical packaging, the RGB-D data features, three-dimensional point cloud data features and process parameter features of the semiconductor are fused to obtain a feature element. In this way, the feature elements of the historical packaging group are acquired, and the feature elements are integrated to obtain a data feature set of the packaging defect type under the defect reason in the historical packaging group. In this way, the data feature sets of each packaging defect type under each defect reason are acquired. 4.The electronic information intelligent processing recommendation system based on multi-modal data fusion according to claim 2, characterized in that, Each abnormal process parameter of each packaging defect type under each defect reason is acquired, and the specific process is as follows: The process parameter text of the semiconductor at each normal historical packaging time is acquired from the database, and the values of each process parameter at each packaging time are acquired at each normal historical packaging time, and the standard value range of each process parameter is obtained by comparison; In each abnormal historical packaging of each historical packaging group, the process parameter text of the semiconductor is acquired, and the values of each process parameter at each packaging time are acquired from the process parameter text. The values of each process parameter at each packaging time are compared with the standard value range of each process parameter. If the value of a certain process parameter is not within the standard value range of the process parameter, it represents that the process parameter is an abnormal process parameter of the abnormal historical packaging. In this way, the abnormal process parameters of each abnormal historical packaging are acquired; The abnormal process parameters of each abnormal historical packaging in each historical packaging group are integrated to obtain an abnormal process parameter set of each historical packaging group. Each abnormal process parameter in the abnormal process parameter set of each historical packaging group is taken as each abnormal process parameter of the packaging defect type under the defect reason in each historical packaging group. In this way, each abnormal process parameter of each packaging defect type under each defect reason is acquired. 5.The electronic information intelligent processing recommendation system based on multi-modal data fusion according to claim 1, characterized in that, The specific process of the defect identification module is as follows: The RGB-D vision data, three-dimensional point cloud data and process parameter text of the semiconductor at this packaging time are acquired, and the RGB-D vision data features, three-dimensional point cloud data features and process parameter features are acquired. The features are fused to obtain a feature element of the semiconductor packaging at this time. The feature element of the semiconductor packaging at this time is compared with each feature element of the standard data feature set. If the feature element of the semiconductor packaging at this time is the same as a certain feature element in the standard data feature set, it represents that the semiconductor packaging at this time does not have a packaging defect. If the feature element of the semiconductor packaging at this time is different from each feature element in the standard data feature set, it represents that the semiconductor packaging at this time has a packaging defect. At this time, the packaging defect type and defect reason of the semiconductor packaging at this time are analyzed. 6.The electronic information intelligent processing recommendation system based on multi-modal data fusion according to claim 5, characterized in that, The specific process of analyzing the packaging defect type and defect reason of the semiconductor packaging at this time is as follows: When the semiconductor package has a packaging defect, the feature elements of the semiconductor package are compared with the feature elements in the data feature set of each packaging defect type under each defect cause, and if the feature elements of the semiconductor package are the same as the feature elements in the data feature set of a packaging defect type under a defect cause, it means that the packaging defect type and the defect cause are the packaging defect type and the defect cause of the semiconductor package. 7.The electronic information intelligent processing recommendation system based on multi-modal data fusion according to claim 1, characterized in that, The intelligent recommendation module has the following specific process: When the semiconductor package has a packaging defect, the packaging defect type and the defect cause are obtained, and the abnormal process parameters of the semiconductor package are obtained according to the packaging defect type and the defect cause, and the environment of the semiconductor package is obtained; The environment of each historical semiconductor package is obtained from the database and compared with the environment of the semiconductor package, the historical semiconductor package with the same environment as the semiconductor package is called a marked historical package, and the packaging detection result of the marked historical package is obtained, the marked historical package with a normal packaging detection result is called a second marked historical package, the values of the abnormal process parameters in the second marked historical package are obtained, and the average value of the abnormal process parameters in the second marked historical package is calculated as the first adjustment value of the abnormal parameter; The values of the abnormal process parameters at this time are obtained, and the second adjustment value of the abnormal process parameters is obtained according to the values of the abnormal process parameters at this time, and the adjustment value of the abnormal process parameters is determined, and the adjustment value of the abnormal process parameters is fed back to the worker. 8.The electronic information intelligent processing recommendation system based on multi-modal data fusion according to claim 7, characterized in that, The specific process of obtaining the second adjustment value of the abnormal process parameter is as follows: The marked historical package with a packaging defect and the same packaging defect type and defect cause is called an abnormal marked historical package, and the abnormal process parameters of the abnormal marked historical package are obtained; The abnormal process parameters of the abnormal marked historical package are compared with the abnormal process parameters of the semiconductor package, the abnormal marked historical package with the same abnormal process parameters as the semiconductor package is called an abnormal second marked historical package, the similarity coefficient of the abnormal process parameters of the abnormal second marked historical package and the abnormal process parameters of the semiconductor package is determined, and the similarity coefficient is compared, the abnormal second marked historical package with the largest similarity coefficient is selected, and the value of the abnormal parameter after adjustment in the abnormal second marked historical package is obtained from the database, which is used as the second adjustment value of the abnormal process parameter. 9.The electronic information intelligent processing recommendation system based on multi-modal data fusion of claim 7, wherein, The specific process of determining the adjustment value of the abnormal process parameter is as follows: When the abnormal process parameter is at the first adjustment value of the abnormal process parameter, the performance detection result of each semiconductor after packaging is obtained, and each semiconductor with a good performance detection result is called a first adjustment value performance good semiconductor; Obtaining performance detection results of the packaged semiconductors when each abnormal process parameter is at the second adjustment value of each abnormal process parameter, and each semiconductor with a good performance detection result is referred to as a second adjustment value performance good semiconductor; Calculating the proportion of first adjustment value performance good semiconductors and the proportion of second adjustment value performance good semiconductors, and comparing them, if the proportion of first adjustment value performance good semiconductors is greater than the proportion of second adjustment value performance good semiconductors, then the first adjustment value of each abnormal parameter is taken as the adjustment value of each abnormal parameter, if the proportion of second adjustment value performance good semiconductors is greater than the proportion of first adjustment value performance good semiconductors, then the second adjustment value of each abnormal parameter is taken as the adjustment value of each abnormal parameter.
10. An electronic information intelligent processing recommendation method of an electronic information intelligent processing recommendation system based on multi-modal data fusion according to any one of claims 1-9, characterized in that, Comprise: S1, data acquisition: after packaging the semiconductor, collect the RGB-D vision data of the semiconductor and the three-dimensional point cloud data of the semiconductor, and obtain the process parameter text from the database; S2, data analysis: obtain the RGB-D data, three-dimensional point cloud data, process parameter text and packaging detection result of the semiconductor in each packaging from the database, and analyze the data feature set of each packaging defect type under different defect causes, the abnormal process parameters of each packaging defect type under each defect cause, and the data feature set of the semiconductor packaging without packaging defects; S3, defect identification: obtaining the RGB-D vision data, three-dimensional point cloud data and process parameter text of the semiconductor at this time, and judging whether there is a defect in this semiconductor packaging, if there is a defect, obtaining the type of the defect and the reason for the existence of the defect; S4, intelligent recommendation: when there is a packaging defect in this semiconductor packaging, recommending the adjustment value of each process parameter according to the packaging defect type and the defect reason, and feeding back to the staff.