Semiconductor chip packaging quality inspection device and method

Abstract

The present invention discloses a semiconductor chip packaging quality inspection device and method, which relates to the field of computer science and technology. The method includes: a quality inspection index determination module, which is used to determine multiple preset quality inspection indexes of the semiconductor to be inspected; a distribution probability analysis module, which performs defect distribution probability analysis on each quality inspection index and generates defect probability; a sampling inspection module, which performs sampling inspection according to defect probability and the number of products in the batch to be inspected, and generates multiple unqualified rates; a compensation fusion module, which inputs the unqualified rate into a parallel fusion network for compensation fusion and generates a fusion unqualified rate; a quality inspection result generation module, which generates the final quality inspection result according to the fusion unqualified rate. The technical problems of low comprehensive inspection efficiency and waste of resources in the existing semiconductor chip quality inspection are solved, and the technical effect of improving the quality inspection efficiency of semiconductor chip packaging and ensuring the accuracy of quality inspection is achieved.

Description

Technical Field

[0001] The present application relates to the field of computer science and technology, and in particular to a semiconductor chip packaging quality inspection device and method. Background Art

[0002] With the rapid development of the semiconductor industry, especially in the production of highly integrated and miniaturized chips, traditional comprehensive quality inspection methods face problems such as low efficiency and waste of resources. Traditional quality inspection methods usually require comprehensive testing of each quality inspection indicator of each product (such as surface defects, electrical performance, electrostatic discharge, etc.). Although this method can ensure quality, it is time-consuming and difficult to adapt to the needs of high-capacity production lines. In addition, quality control often relies on manual judgment and historical experience, has errors, and is difficult to dynamically optimize during the production process. In order to improve production efficiency and meet the needs of high-precision quality control, there is an urgent need for a new quality inspection solution that can reduce unnecessary comprehensive testing and optimize the quality inspection process.

[0003] At the current stage, relevant technologies exist in the technical problems of low efficiency and waste of resources in comprehensive inspection of semiconductor chips. Summary of the invention

[0004] The present application provides a semiconductor chip packaging quality inspection and testing device and method, which adopts a quality inspection index determination module to determine multiple preset quality inspection indexes of the semiconductor to be inspected; a distribution probability analysis module to perform defect distribution probability analysis on each quality inspection index and generate a defect probability; a sampling inspection module to perform sampling inspection according to the defect probability and the number of products in the batch to be inspected and generate multiple unqualified rates; a compensation fusion module to input the unqualified rate into a parallel fusion network for compensation fusion and generate a fused unqualified rate; a quality inspection result generation module to generate a final quality inspection result according to the fused unqualified rate, thereby achieving the technical effect of improving the quality inspection efficiency of semiconductor chip packaging and ensuring the accuracy of quality inspection.

[0005] The present application provides a semiconductor chip packaging quality inspection device, comprising:

[0006] A quality inspection index determination module, the quality inspection index determination module is used to determine multiple preset quality inspection indexes of the semiconductor to be inspected; a distribution probability analysis module, the distribution probability analysis module is used to perform defect distribution probability analysis on the multiple preset quality inspection indexes respectively, and generate multiple defect probabilities; a sampling detection module, the sampling detection module is used to perform sampling detection of multiple detection indicators based on the multiple defect probabilities and the number of batches of products to be inspected of the semiconductor to be inspected, and generate multiple indicator failure rates; a compensation fusion module, the compensation fusion module is used to input the multiple indicator failure rates into a pre-trained parallel fusion network for parallel existence compensation fusion, and generate a fusion failure rate; a quality inspection result generation module, the quality inspection result generation module is used to generate the quality inspection result of the semiconductor to be inspected according to the fusion failure rate.

[0007] The present application provides a semiconductor chip packaging quality inspection method, comprising:

[0008] Determine multiple preset quality inspection indicators of the semiconductor to be inspected; perform defect distribution probability analysis on the multiple preset quality inspection indicators respectively to generate multiple defect probabilities; perform sampling inspection of multiple inspection indicators based on the multiple defect probabilities and the number of batches of products to be inspected of the semiconductor to be inspected to generate multiple indicator failure rates; input the multiple indicator failure rates into a pre-trained parallel fusion network for parallel existence compensation fusion to generate a fusion failure rate; generate a quality inspection result of the semiconductor to be inspected based on the fusion failure rate.

[0009] A semiconductor chip packaging quality inspection and testing device and method proposed in this application include: first, a quality inspection index determination module, which is used to determine multiple preset quality inspection indexes of the semiconductor to be inspected; a distribution probability analysis module, which performs defect distribution probability analysis on each quality inspection index and generates a defect probability; a sampling inspection module, which performs sampling inspection according to the defect probability and the number of products in the batch to be inspected, and generates multiple unqualified rates; a compensation fusion module, which inputs the unqualified rate into a parallel fusion network for compensation fusion, and generates a fused unqualified rate; a quality inspection result generation module, which generates a final quality inspection result according to the fused unqualified rate, thereby achieving the technical effect of improving the quality inspection efficiency of semiconductor chip packaging and ensuring the accuracy of quality inspection. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solution of the embodiment of the present invention, the accompanying drawings of the embodiment of the present invention will be briefly introduced below. A flow chart is used in the present application to illustrate the operations performed by the device according to the embodiment of the present application. It should be understood that the previous or following operations are not necessarily performed precisely in order. On the contrary, various steps can be processed in reverse order or simultaneously as needed. At the same time, other operations can also be added to these processes, or one or more operations can be removed from these processes.

[0011] Figure 1 A schematic diagram of the structure of a semiconductor chip packaging quality inspection device provided in an embodiment of the present application;

[0012] Figure 2 A schematic diagram of a process flow of a semiconductor chip packaging quality inspection method provided in an embodiment of the present application.

[0013] Explanation of the reference numerals: quality inspection index determination module 10 , distribution probability analysis module 20 , sampling detection module 30 , compensation fusion module 40 , quality inspection result generation module 50 . DETAILED DESCRIPTION

[0014] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below.

[0015] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings. The described embodiments should not be regarded as limiting the present application. All other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of this application.

[0016] In the following description, reference is made to "some embodiments", which describe a subset of all possible embodiments, but it is understood that "some embodiments" may be the same subset or different subsets of all possible embodiments, and may be combined with each other without conflict, and the terms "first\second" involved are merely to distinguish similar objects and do not represent a specific ordering of objects. The terms "including" and "having" and any variations are intended to cover non-exclusive inclusions, for example, a process, method, device, product, or server that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or modules that are not clearly listed or inherent to these processes, methods, products, or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those generally understood by technicians in the technical field of this application. The terms used herein are for the purpose of describing the embodiments of the present application only.

[0017] The present application embodiment provides a semiconductor chip packaging quality inspection device, such as Figure 1 As shown, the device comprises:

[0018] The quality inspection index determination module 10 is used to determine a plurality of preset quality inspection indexes of the semiconductor to be inspected. Specifically, in the quality inspection process of semiconductor chip packaging, it is very important for the quality inspection index determination module 10 to determine a plurality of preset quality inspection indexes of the semiconductor to be inspected. It will screen the indexes according to the product characteristics and relevant quality inspection requirements of the semiconductor to be inspected, taking into account the performance requirements of the semiconductor in different application scenarios and the quality inspection standards generally recognized by the industry, covering multiple aspects from appearance to internal performance. These include surface defect indicators, which will clearly specify the defect types, inspection standards and methods to be inspected, and use specific equipment to check whether there are scratches and other defects and their severity; electrical performance indicators, which determine the measurement range and accuracy requirements of relevant parameters such as resistance, and use professional electrical testing equipment to test according to the prescribed process to ensure that the parameters are within the qualified range; electrostatic discharge (ESD) tolerance test indicators, which clearly specify specific test methods and standards, and use a special simulator to perform simulation tests according to specific parameter settings to evaluate whether its tolerance meets the requirements, so as to accurately determine the preset quality inspection indicators including surface defects, electrical performance, electrostatic discharge (ESD) tolerance tests, etc., and lay the foundation for subsequent quality inspection work.

[0019] The distribution probability analysis module 20 is used to perform defect distribution probability analysis on the multiple preset quality inspection indicators respectively to generate multiple defect probabilities. Specifically, when the distribution probability analysis module 20 performs defect distribution probability analysis on multiple preset quality inspection indicators (such as surface defects, electrical performance, electrostatic discharge (ESD) tolerance test, etc.) to generate multiple defect probabilities, it is necessary to first obtain the preset quality inspection indicator information and a large amount of historical quality inspection data. Then, the analysis is carried out for each indicator. Taking the surface defect indicator as an example, the relevant records are screened from the historical quality inspection data, the number of surface defect records is counted, and the defect probability is obtained by calculation; for the electrical performance indicator, the relevant records are also extracted from the historical quality inspection data, and the number of electrical performance defect records is counted to calculate the corresponding defect probability; for the electrostatic discharge (ESD) tolerance test indicator, the relevant records are also found, the number of defect records is counted, and the defect probability is obtained. Finally, through the analysis and calculation of each preset quality inspection indicator, the corresponding defect probability is obtained respectively. These defect probabilities presented in specific values ​​can reflect the possibility of defects in each indicator in previous quality inspections, and serve as an important basis for subsequent sampling inspections and other operations, so as to reasonably arrange inspection resources and improve quality inspection efficiency and accuracy.

[0020] In a possible implementation, the distribution probability analysis module 20 further includes: a data set acquisition unit, which is used to collect historical quality inspection data sets with the multiple preset quality inspection indicators and the model of the semiconductor to be inspected as constraints. Specifically, the data set acquisition unit is responsible for collecting historical quality inspection data sets, and the acquisition process uses multiple preset quality inspection indicators and the model of the semiconductor to be inspected as constraints. The multiple preset quality inspection indicators cover aspects such as surface defects, electrical performance, and electrostatic discharge (ESD) tolerance testing. Different models of semiconductors to be inspected have different quality inspection conditions due to differences in production processes, design features, etc. The data set acquisition unit selects historical quality inspection records that meet specific preset quality inspection indicators and are for the corresponding semiconductor models to be inspected from various possible data sources, such as previous quality inspection record databases, data stored in the quality monitoring system on the production line, etc., and integrates them to form a complete historical quality inspection data set to provide a data basis for subsequent analysis.

[0021] A quality inspection data extraction unit, the quality inspection data extraction unit is used to extract the first historical quality inspection data corresponding to the first preset quality inspection indicator based on the historical quality inspection data set. Specifically, after obtaining the historical quality inspection data set, the quality inspection data extraction unit extracts the first historical quality inspection data corresponding to the first preset quality inspection indicator based on the data set. If the first preset quality inspection indicator is a surface defect indicator, the quality inspection data extraction unit uses a specific screening and extraction method in the entire historical quality inspection data set to find all historical quality inspection record contents related to the surface defect indicator, including detailed information such as specific test results, test methods, and test time when surface defects are detected for different batches of semiconductor products, so as to form the first historical quality inspection data for subsequent in-depth analysis and processing of the indicator.

[0022] The first defect probability generating unit is used to count the proportion of data in the first historical quality inspection data that are marked as unqualified by the first preset quality inspection indicator, generate a first defect probability, and add it to the multiple defect probabilities. Specifically, after the quality inspection data extraction unit extracts the first historical quality inspection data, the first defect probability generating unit is responsible for counting the proportion of data in the first historical quality inspection data that are marked as unqualified by the first preset quality inspection indicator, and then generating a first defect probability and adding it to the multiple defect probabilities. Assuming that the first preset quality inspection indicator is a surface defect indicator, the first defect probability generating unit carefully checks each record of surface defect detection in the extracted first historical quality inspection data, and counts the records that are judged to be unqualified. Assuming that there are a total of Y records in the first historical quality inspection data, of which X records are marked as unqualified surface defects, the first defect probability value corresponding to the surface defect indicator is obtained by dividing X by Y. For example, when X is 10 and Y is 100, the first defect probability is 0.1 (i.e., 10%). After the first defect probability is generated, it is added to multiple defect probability sets calculated for different preset quality inspection indicators, so that the entire defect probability set is complete, which comprehensively reflects the possibility of unqualified conditions in each preset quality inspection indicator during the historical quality inspection process, and provides a basis for subsequent sampling inspection and other operations.

[0023] The sampling detection module 30 is used to perform sampling detection of multiple detection indicators based on the multiple defect probabilities and the number of batches of products to be inspected of the semiconductor to be inspected, and generate multiple indicator failure rates. Specifically, the sampling detection module 30 works based on the multiple defect probabilities corresponding to different preset quality inspection indicators (such as surface defects, electrical properties, electrostatic discharge (ESD) tolerance tests, etc.) generated by the previous steps and the number of batches of products to be inspected of the semiconductor to be inspected. First, the defect probability data is analyzed to understand the risk level of defects in each indicator, and then the sampling detection ratio of each preset quality inspection indicator in the sampling detection is determined according to each defect probability ratio. Generally, the sampling detection ratio of indicators with high defect probabilities is increased accordingly. Then, the number of sampling samples for each indicator is calculated in combination with the number of batches of products to be inspected, and samples are selected by random sampling method. Afterwards, for each preset quality inspection indicator, the corresponding professional testing equipment and methods are used to strictly test the sampling samples. Finally, the ratio of the number of products judged as unqualified for each preset quality inspection indicator in the sampling samples to the number of sampling samples is counted, that is, multiple indicator unqualified rates are generated. These indicator unqualified rates can reflect the quality of each indicator in this sampling test and provide a basis for subsequent analysis and processing.

[0024] In a possible implementation, the sampling inspection module 30 further includes: a detection ratio configuration unit, which is used to configure the sampling inspection ratios of the multiple preset quality inspection indicators based on the ratios of the multiple defect probabilities. Specifically, the detection ratio configuration unit configures the sampling detection ratios of multiple preset quality inspection indicators based on the ratio of multiple defect probabilities. It first receives multiple defect probability data corresponding to different preset quality inspection indicators (such as surface defects, electrical performance, electrostatic discharge (ESD) tolerance test, etc.) obtained by the analysis of the previous module. Each defect probability reflects the possibility of defects in the corresponding indicator in the previous quality inspection. Then, the defect probability is analyzed and processed, and the defect probability ratio of each preset quality inspection indicator is calculated by a specific algorithm. For example, the defect probability of the surface defect indicator is 0.2, and the defect probability of the electrical performance indicator is 0.1, and the defect probability ratio of the two is 2:1. The sampling detection ratio of each preset quality inspection indicator is configured according to the calculated defect probability ratio, following the rule that the higher the defect probability, the higher the sampling detection ratio. For example, according to the established rules, the overall sampling ratio is determined to be 10%. According to the above 2:1 ratio, the surface defect indicator can be configured with a sampling detection ratio of approximately 6.67%, and the electrical performance indicator can be configured with a sampling detection ratio of approximately 3.33%, so as to determine a reasonable sampling detection ratio for subsequent random sampling.

[0025] A random sampling unit is used to perform random sampling according to the sampling detection ratio and the number of the batch of products to be tested, and generate multiple sampling results. Specifically, the random sampling unit generates multiple sampling results by random sampling based on the sampling inspection ratio determined by the inspection ratio configuration unit and the number of product batches to be inspected. The sampling inspection ratio information of each preset quality inspection indicator and the number of product batches to be inspected are obtained first. The number of product batches to be inspected clarifies the overall scale of this quality inspection product. For each preset quality inspection indicator, the number of sampling samples is calculated according to its sampling inspection ratio and the number of product batches to be inspected. For example, if the number of product batches to be inspected is 1,000 pieces, the sampling inspection ratio of a preset quality inspection indicator (such as surface defect indicator) is 6.67%. After calculation, the number of sampling samples is 66.7 pieces. The actual operation is rounded (such as 67 pieces). According to the calculated number of sampling samples for each indicator, the corresponding number of samples are selected from the product batches to be inspected using a random sampling method to ensure the randomness and representativeness of the samples. A group of corresponding sampling samples are selected for each preset quality inspection indicator to form multiple sampling results for subsequent result detection unit detection.

[0026] The result detection unit is used to detect the multiple sampling results through the quality inspection modules corresponding to the multiple preset quality inspection indicators, and to count the unqualified rates to generate the unqualified rates of the multiple indicators. Specifically, the result detection unit detects the multiple sampling results through the quality inspection modules corresponding to the multiple preset quality inspection indicators and counts the unqualified rates to generate the unqualified rates of the multiple indicators. First, multiple sampling results and quality inspection modules corresponding to each preset quality inspection indicator are obtained. Each quality inspection module has the ability to accurately detect specific indicators. For example, the surface defect indicator quality inspection module uses a high-resolution microscope to inspect the surface according to specific standards, and the electrical performance indicator quality inspection module uses professional electrical testing equipment to detect according to standard parameters. Then, the sampling results of each preset quality inspection indicator are tested one by one with the corresponding quality inspection module. Each quality inspection module strictly follows the detection process and standards set by itself. Accurate operation ensures that the test results are accurate and reliable. For example, when testing the sampling results of surface defect indicators, use a high-resolution microscope to check whether there are scratches and other defects on the surface under specified conditions; when testing the sampling results of electrical performance indicators, use relevant equipment to accurately measure various electrical performance parameters and compare them with standard parameters to determine whether they meet the requirements. After completing the sampling results of all preset quality inspection indicators, the number of products judged as unqualified in the sampling results of each preset quality inspection indicator is counted. If the number of sampling results of a preset quality inspection indicator (such as surface defect indicator) is N, and the number of products judged as unqualified is M, then the unqualified rate of this indicator is M / N. In this way, the indicator unqualified rate is calculated for each preset quality inspection indicator to form multiple indicator unqualified rates, reflecting the quality of the sampling results of each indicator in this sampling test, and providing a basis for subsequent analysis and processing.

[0027] In a possible implementation, the result detection unit further includes: a detection parameter set determination subunit, and the detection parameter set determination subunit is used to determine a plurality of detection parameter sets corresponding to the plurality of preset quality inspection indicators. Specifically, the detection parameter set determination subunit is responsible for determining multiple detection parameter sets corresponding to multiple preset quality inspection indicators. In the semiconductor chip packaging quality inspection process, different preset quality inspection indicators require specific detection parameter sets to achieve accurate detection. For preset quality inspection indicators such as surface defects, electrical performance, and electrostatic discharge (ESD) tolerance test, the corresponding detection parameter sets are clearly defined according to the characteristics of each indicator and the detection requirements. Taking specific detection parameters as an example, multiple detection parameter sets cover aspects such as images, ultrasound, X-rays, and pin currents. For surface defect indicators, the image-related detection parameter set can obtain a clear image of the chip packaging surface through a high-resolution image acquisition device, and set image parameters such as resolution, contrast, and brightness to analyze whether there are scratches, cracks, and other defects on the surface; for electrical performance indicators, related detection parameter sets such as pin current are involved, and parameters such as current range and current fluctuation allowable value are determined to accurately measure electrical performance; for electrostatic discharge (ESD) tolerance test indicators, the detection parameter set is composed of specific electrical parameters and test environment parameters to ensure accurate simulation of electrostatic discharge conditions and detect chip tolerance. In this way, the corresponding detection parameter set is determined for each preset quality inspection indicator, laying the foundation for subsequent detection work.

[0028] A parameter statistics subunit, the parameter statistics subunit is used to perform parameter statistics from macro defects to micro defects for the multiple detection parameter sets, and generate data that meet the common accuracy requirements of macro defect detection and micro defect detection. Specifically, after the detection parameter set determination subunit determines multiple detection parameter sets, the parameter statistics subunit performs parameter statistics from macro defects to micro defects for these detection parameter sets, aiming to generate data that meet the common accuracy requirements of macro defect detection and micro defect detection. For each detection parameter set, such as an image detection parameter set, statistics are first analyzed and counted from the perspective of macro defects. In image detection, macro defects appear as large scratches and obvious stains on the surface of the chip package. Parameters related to macro defect detection, such as the length range of large scratches and the proportion of stain coverage area, are counted. After analyzing a large amount of sample data, parameter statistics that can effectively identify macro defects are determined. Then, statistics are performed from the perspective of micro defects. At the micro level of image detection, attention is paid to tiny cracks on the surface of the chip package, particle adhesion, etc. The parameters such as the width of tiny cracks, particle size and distribution density are counted, and the parameter statistical values ​​that can accurately identify micro defects are determined through analysis of a large amount of sample data. For other detection parameter sets such as ultrasound, X-ray, pin current, and so on, the parameter statistics of the relevant macro defects and micro defects are carried out in order from macro to micro. For example, in ultrasonic testing, macro defects involve large voids inside the chip, and micro defects are tiny stratification phenomena. The ultrasonic reflection parameters, propagation time and other parameters corresponding to these defects are counted respectively. By comprehensively and systematically performing parameter statistics from macro defects to micro defects on multiple detection parameter sets, the statistical results from all aspects are integrated to generate data that meets the common accuracy requirements, so that the subsequent quality inspection modules can carry out effective detection operations based on this.

[0029] The quality inspection module generation subunit is used to configure the detection module and the detection analysis network according to the common accuracy requirement, and generate the quality inspection modules corresponding to the multiple preset quality inspection indicators. Specifically, the quality inspection module generation subunit configures the detection module and the detection analysis network according to the common accuracy requirements generated by the parameter statistics subunit, and then generates quality inspection modules corresponding to multiple preset quality inspection indicators, obtains data that meets the common accuracy requirements of macro defect detection and micro defect detection determined by the parameter statistics subunit, clarifies the various parameter standards required for accurate detection of macro defects and micro defects under different detection parameter sets, and configures the detection module according to the common accuracy requirements for the detection parameter sets corresponding to each preset quality inspection indicator. Taking the surface defect indicator as an example, if its corresponding detection parameter set mainly involves image detection and the common accuracy requirements determine that a high-resolution image acquisition device and a specific image analysis algorithm are required to accurately detect surface defects, then a suitable high-resolution image acquisition device is selected and the corresponding image analysis algorithm is configured as the detection module. For electrical performance indicators, if its detection parameter set involves parameters such as pin current and the common accuracy requirements stipulate that a high-precision current measurement device and a specific current analysis method are required, then the corresponding high-precision current measurement device is selected and the relevant current analysis method is configured as the detection module. For other preset quality inspection indicators, suitable detection modules are configured according to their corresponding detection parameter sets and common accuracy requirements. After the detection modules are configured, the detection analysis network is configured. The detection and analysis network is used to further analyze and process the data collected by the detection module. According to the common accuracy requirements and the characteristics of different preset quality inspection indicators, appropriate data analysis algorithms and models are selected to construct the detection and analysis network. For example, for image detection of surface defect indicators, an image recognition model based on deep learning can be selected as part of the detection and analysis network to more accurately analyze image data and identify surface defects; for pin current detection of electrical performance indicators, a data analysis network based on electrical principles can be used to analyze current data and determine whether the electrical performance meets the requirements. By carefully configuring the detection module and the detection and analysis network according to the common accuracy requirements, corresponding quality inspection modules are generated for each preset quality inspection indicator, so that it has the ability to accurately detect specific preset quality inspection indicators and meet the needs of subsequent quality inspection work.

[0030] The unqualified rate statistical subunit is used to collect detection data of the multiple sampling results by the detection module in the quality inspection module and input them into the detection analysis network, and to calculate the unqualified rate according to the detection analysis results to generate the unqualified rates of the multiple indicators. Specifically, the unqualified rate statistics subunit uses the detection module in the quality inspection module to collect detection data on multiple sampling results and then inputs them into the detection analysis network, and calculates the unqualified rate according to the detection and analysis results, and then generates multiple indicator unqualified rates, and obtains the multiple sampling results generated in the previous steps and the quality inspection modules corresponding to multiple preset quality inspection indicators. The sampling results are obtained through the previous random sampling and other operations, and each corresponds to a preset quality inspection indicator. The quality inspection module is tailored for each indicator and has accurate detection capabilities. For each preset quality inspection indicator, the unqualified rate statistics subunit uses the detection module in its corresponding quality inspection module to collect detection data on the sampling results of the indicator. Taking the surface defect indicator as an example, if the detection module in its quality inspection module includes a high-resolution image acquisition device, the device is used to acquire images of the sampling results of the surface defect indicator to obtain detailed image data as part of the detection data; for electrical performance indicators, if the detection module in its quality inspection module includes a high-resolution image acquisition device, the device is used to acquire images of the sampling results of the surface defect indicator to obtain detailed image data as part of the detection data; for electrical performance indicators, if the detection module in its quality inspection module The detection module includes a high-precision current measurement device, and the device is used to measure the current of the sampling results of the electrical performance indicators to obtain current data as the detection data. After the detection data collection is completed, the collected detection data is input into the detection analysis network in the quality inspection module. The detection analysis network analyzes and processes the input detection data according to pre-configured algorithms, models, etc. For example, for image data of surface defect indicators, the detection analysis network can analyze based on the deep learning image recognition model to determine whether there are surface defects such as scratches and cracks in the image; for current data of electrical performance indicators, the detection analysis network can analyze based on the data analysis network of electrical principles to determine whether the current parameters meet the requirements, and the unqualified rate is calculated based on the detection and analysis results. Suppose the number of sampling results of a preset quality inspection indicator (such as surface defect indicator) is N, and the number of products judged as unqualified is M, then the unqualified rate of this indicator is M / N. In this way, for each preset quality inspection indicator, the unqualified rate statistics subunit can calculate its indicator unqualified rate. These indicator unqualified rates constitute multiple indicator unqualified rates, which accurately reflect the quality of the sampling results of each indicator in this sampling test, and provide an important basis for subsequent further analysis and processing.

[0031] The compensation fusion module 40 is used to input the multiple indicator failure rates into the pre-trained parallel fusion network for parallel existence compensation fusion to generate a fusion failure rate. Specifically, the compensation fusion module 40 first obtains the multiple indicator failure rates generated by the previous quality inspection process link. These indicator failure rates correspond to preset quality inspection indicators such as surface defects, electrical performance, and electrostatic discharge (ESD) tolerance testing, and carry the quality information of each indicator in the sampled test sample. Then, the multiple indicator failure rates are input into the pre-trained parallel fusion network. The network has been pre-trained and optimized with a large amount of historical data, and has a strong data analysis and processing capability. When receiving data, it will perform preliminary processing according to the preset format and rules. Then the network performs parallel existence compensation fusion operations on the input indicator failure rates, and will perform parallel processing based on the importance of each indicator corresponding to the preset quality inspection indicator, the correlation between indicators, and the weight of each indicator in the historical data on the overall quality. For example, a higher weight is given to the key indicator, the correlation between related indicators is considered, and the processing method is adjusted accordingly. Compensation operations will also be performed for the deviation or incompleteness of the indicator failure rate that may be caused by sampling detection. Finally, after the above complex process, the network generates a fused failure rate, which is a failure rate for the batch of semiconductor products to be inspected. It integrates the quality conditions of multiple preset quality inspection indicators and can more comprehensively and accurately reflect the overall quality status of the batch of products to be inspected, providing an important basis for subsequent judgment of the quality inspection results of the semiconductors to be inspected.

[0032] In a possible implementation, the compensation fusion module 40 further includes: a parallel combination result generating unit, the parallel combination result generating unit is used to perform parallel combination on the multiple preset quality inspection indicators to generate P parallel combination results, wherein the number of indicators in each parallel combination result is at least 2, P is an integer greater than 0, and Q is the total number of the plurality of preset quality inspection indicators. Specifically, the parallel combination result generation unit performs a parallel combination operation on the plurality of preset quality inspection indicators in accordance with the regulations to generate P parallel combination results. The plurality of preset quality inspection indicators involved cover aspects such as surface defects, electrical performance, and electrostatic discharge (ESD) tolerance testing, reflecting the quality characteristics of the semiconductor to be inspected from different angles. The combination rule specifies that the number of indicators in each parallel combination result is at least 2, and P is based on the formula To calculate (where Q is the total number of multiple preset quality inspection indicators, P is an integer greater than 0), that is, to generate results according to multiple combinations of two indicators, three indicators, four indicators, etc. For example, assuming that there are 5 preset quality inspection indicators (i.e., Q=5), according to the formula It is necessary to generate 10 parallel combination results. For example, surface defects and electrical performance indicators can be combined into one parallel combination result, or surface defects, electrical performance and electrostatic discharge (ESD) tolerance test indicators can be combined into another parallel combination result, etc. By exhausting all kinds of combinations that contain at least two indicators, P parallel combination results can be generated to provide a data basis for the subsequent construction of a parallel fusion network.

[0033] A parallel fusion network construction unit, the parallel fusion network construction unit is used to combine the P parallel combination results and the historical quality inspection data set to construct the parallel fusion network, wherein the parallel fusion network includes M parallel fusion integration models, wherein M≤P. Specifically, the parallel fusion network construction unit uses the P parallel combination results generated by the parallel combination result generation unit and the historical quality inspection data set to construct a parallel fusion network, the network includes M parallel fusion integration models, and M≤P, to obtain P parallel combination results and a rich historical quality inspection data set, the historical quality inspection data set contains a large number of detailed records of quality inspections for similar semiconductor products in the past, the records are of great reference value for analyzing the relationship between different quality inspection indicators and the overall quality status, by analyzing the correlation between the P parallel combination results and the data points in the historical quality inspection data set, the rules and patterns contained therein are mined, for example, observing the relationship between the P parallel combination results and the data points in the historical quality inspection data set in different historical quality inspection records. The corresponding relationship between the unqualified indicators involved and the overall product quality is constructed based on the above analysis and mining results, with the help of appropriate algorithms and technical means to build the architecture of the parallel fusion network, and determine the M parallel fusion integration models contained therein. Each model is responsible for the function of performing specific analysis and processing on the input data. The number of models M is determined based on the actual situation and must satisfy the condition of M≤P. For example, assuming that 10 parallel combination results have been generated (i.e., P=10), M=8 parallel fusion integration models may be determined. After reasonable configuration of these models and the connection relationship between them, the construction of the parallel fusion network is completed, so that the network has the ability to effectively fuse and process subsequent input data.

[0034] A weighted calculation unit is used to combine the unqualified rates of the multiple indicators and then input them into the M parallel fusion integration models for analysis and fusion, and to perform weighted calculation on the output M parallel fusion results to generate the fusion unqualified rate. Specifically, the weighted calculation unit performs relevant processing on the unqualified rates of multiple indicators to generate the fusion unqualified rate, obtains the unqualified rates of multiple indicators generated by the previous quality inspection process link, and the unqualified rates of indicators correspond to the preset quality inspection indicators respectively, which can reflect the proportion of unqualified situations of each indicator in the relevant processes such as sampling detection. After the unqualified rates of multiple indicators are combined and processed, they are input into the M parallel fusion integration models in the constructed parallel fusion network to carry out analysis and fusion operations. Each parallel fusion integration model conducts in-depth analysis on the input combination data based on its own algorithm and internal structure, and mines the hidden information and the mutual relationship between the unqualified rates of each indicator. For example, it will analyze the coordinated changes of the unqualified rates of indicators involved in the specific parallel combination results, and the impact of such changes on the overall quality assessment. After the analysis and fusion processing of the M parallel fusion integration models, After that, M parallel fusion results will be output. At this time, the weighted calculation unit will perform weighted calculation operations on these M parallel fusion results based on the importance evaluation of each parallel fusion result and their relative weights in the overall quality evaluation. For example, based on previous experience data and research results on the relationship between different parallel fusion results and the overall quality, if a certain parallel fusion result is determined to play a more critical role in reflecting the overall quality failure, then it will be given a relatively high weight; and for relatively minor parallel fusion results, they will be given a relatively low weight. Through weighted calculation operations, the fusion failure rate is finally generated. The fusion failure rate comprehensively considers the failure rates of multiple indicators and the relevant information of each parallel fusion result, and can more comprehensively and accurately reflect the overall quality status of the batch of products to be inspected of the semiconductor to be inspected, thereby providing an important basis for the subsequent judgment of the quality inspection results of the semiconductor to be inspected.

[0035] In a possible implementation, the parallel fusion network construction unit further includes: a sample data set generation subunit, the sample data set generation subunit is used to extract data from the historical quality inspection data set in combination with the P parallel combination results to generate P sample data sets, wherein any sample data set includes a quality inspection indicator failure rate sample and a corresponding full inspection failure rate sample corresponding to the parallel combination result. Specifically, the sample data set generation subunit performs data extraction on the historical quality inspection data set based on the P parallel combination results to generate P sample data sets, the P parallel combination results are formed by combining multiple preset quality inspection indicators according to specific rules, and each parallel combination result contains at least two preset quality inspection indicators, reflecting the combination of quality inspection indicators related to the semiconductor to be inspected from different levels. When extracting data from the historical quality inspection data set, for each parallel combination result, the corresponding quality inspection indicator failure rate sample and full inspection failure rate sample are searched in the historical quality inspection data set. Among them, the quality inspection index unqualified rate sample is the probability data related to the specific quality inspection index contained in the parallel combination result that failed in the past quality inspection; the full inspection unqualified rate sample is not the probability that all quality inspection indicators are unqualified, but the probability of unqualified products obtained after comprehensive testing, that is, the probability of unqualified products determined when only one indicator is unqualified or multiple indicators are unqualified. After matching and extracting each parallel combination result with the corresponding data in the historical quality inspection data set, P sample data sets are finally generated. Each sample data set completely covers the quality inspection index unqualified rate samples and full inspection unqualified rate samples corresponding to the specific parallel combination results, providing a targeted data basis for subsequent analysis and processing.

[0036] The feature correlation analysis subunit is used to perform the feature correlation analysis of the index correlation and the input and output for the P parallel combination results, and perform dimensionality reduction according to the analysis results to generate M target parallel combination results. Specifically, the feature correlation analysis subunit carries out the feature correlation analysis of the index correlation and the input and output for the P parallel combination results, performs dimensionality reduction according to the analysis results, and then generates M target parallel combination results. For the P parallel combination results, the index correlation analysis is first carried out. The degree of correlation between different quality inspection indicators within each parallel combination result is explored, and by analyzing historical data and related quality inspection records, it is clarified whether there are relationships such as mutual influence and coordinated changes between different indicators, such as unqualified conditions of certain quality inspection indicators or specific changes with other indicators, and such relationships are quantified to determine the index correlation, and then the feature correlation analysis of input and output is carried out. From the perspective of data processing flow, when these parallel combination results are used as input, the feature correlation between them and the final output results (such as product quality determination results, etc.) is analyzed. Analyze the changing characteristics of the output results and the strength of the correlation between the two under different input parallel combination results to determine the characteristic correlation between the input and output. After completing the above two correlation analysis, perform dimensionality reduction processing based on the obtained analysis results. In view of the fact that the P parallel combination results may have high data dimensions and information redundancy, through dimensionality reduction operations, select some parallel combination results that have a more critical impact on the final result and a higher correlation, and finally generate M target parallel combination results, so that subsequent processing can focus more on key information and improve processing efficiency and accuracy.

[0037] The convergence accuracy acquisition subunit is used to obtain the M sample data sets corresponding to the M target parallel combination results, train the M parallel fusion integration models based on the machine learning model, and obtain the convergence accuracy. Specifically, the convergence accuracy acquisition subunit performs two operations: obtaining the M sample data sets corresponding to the M target parallel combination results, and training the M parallel fusion integration models based on the machine learning model and obtaining the convergence accuracy. First, the M target parallel combination results generated by the feature correlation analysis subunit and their corresponding M sample data sets are obtained. After the sample data set is extracted by the previous sample data set generation subunit and processed by the feature correlation analysis subunit, it accurately matches the target parallel combination results, contains rich quality inspection index unqualified rate samples and full inspection unqualified rate samples and other related information, and provides sufficient data support for the machine learning model training. Then, the M parallel fusion integration models are trained based on the machine learning model, and the M target parallel combination results and their corresponding M sample data sets are input into the machine learning model according to the appropriate format and rules, so that the model can learn the data patterns and laws therein. During the training process, the model continuously adjusts its own parameters to adapt to the data. As the training progresses, the model output results gradually become stable. At this time, the convergence accuracy is determined by monitoring the changes in the model output results. The convergence accuracy reflects the degree of stability reached during the model training process and is an important indicator for measuring the model training effect. By obtaining the convergence accuracy, it is possible to evaluate whether the M parallel fusion integration models have achieved the expected training effect, providing a basis for subsequent network construction.

[0038] A parallel fusion network generation subunit, the parallel fusion network generation subunit is used to configure the result weighting layer of M parallel fusion integration models according to the convergence accuracy, and integrate and connect with the M parallel fusion integration models to generate the parallel fusion network. Specifically, the parallel fusion network generation subunit configures the result weighting layer of M parallel fusion integration models according to the convergence accuracy, and integrates and connects with the M parallel fusion integration models to generate a parallel fusion network, first obtains the convergence accuracy information determined by the convergence accuracy acquisition subunit, and the convergence accuracy is used as a key indicator to measure the training effect of the M parallel fusion integration models. Its value reflects the degree of stable state reached during the model training process and the degree of fitting to the data, and then configures the result weighting layer of M parallel fusion integration models according to the convergence accuracy. The function of the result weighting layer is to weight the output results of each parallel fusion integration model to make the final fusion result more reasonable and accurate. According to different values ​​of convergence accuracy, the weight distribution method of different parallel fusion integration models in the final fusion result is determined. If the convergence accuracy of a parallel fusion integration model is high, it indicates that it has a good fitting effect on the data, and it may be given a relatively high weight in the result weighting layer; conversely, if the convergence accuracy is low, it may be given a relatively low weight. Finally, the M parallel fusion integration models with the result weighting layer configured are integrated and connected. Establish a suitable connection method and data transmission channel, combine each parallel fusion integration model into an organic whole, and form a parallel fusion network. This parallel fusion network has the ability to effectively fuse the input data. After multiple related data are input, they can be processed by each parallel fusion integration model and weighted by the result weighting layer, and output the comprehensive fusion result, providing strong support for subsequent quality inspection related analysis and processing.

[0039] In a possible implementation, the feature correlation analysis subunit further includes: an indicator correlation generation microunit, and the indicator correlation generation microunit is used to perform unqualified correlation analysis on the indicators in the P parallel combination results to generate P indicator correlations. Specifically, the indicator correlation generation micro-unit performs unqualified correlation analysis on the indicators in P parallel combination results, and then generates P indicator correlations. The P parallel combination results are composed of multiple preset quality inspection indicators according to specific rules. Each parallel combination result contains at least two preset quality inspection indicators, which reflect the combination of quality inspection indicators of the semiconductor to be inspected from different levels. When performing unqualified correlation analysis on the indicators in each parallel combination result, based on historical quality inspection data and related records, the possibility and correlation degree of other indicators being unqualified when one of the indicators is unqualified are analyzed. For example, when a parallel combination result contains surface defect indicators and electrical performance indicators, the probability of the electrical performance indicators being unqualified (such as parameters such as resistance and capacitance exceeding the normal range) when the surface defect indicator is unqualified (such as obvious scratches or cracks on the surface) and the degree of correlation between the two are analyzed. After analyzing each indicator in the P parallel combination results in this way, P indicator correlations are generated according to the correlation relationship and degree of correlation between the unqualified situations of each indicator. The correlation degree of each indicator accurately reflects the unqualified correlation characteristics of each indicator in the corresponding parallel combination result, providing a data basis for subsequent analysis and processing.

[0040] An input-output correlation generation micro-unit is used to perform feature correlation analysis between input and output for P sample data sets to generate P input-output correlations. Specifically, the input-output correlation generation micro-unit performs feature correlation analysis between input and output for P sample data sets to generate P input-output correlations. First, P sample data sets are obtained, which are generated by the previous related operations. Each sample data set contains information such as quality inspection indicator failure rate samples and full inspection failure rate samples corresponding to specific parallel combination results, reflecting data characteristics under different quality inspection situations from different aspects. Feature correlation analysis between input and output is performed for each sample data set. From the perspective of data processing flow, the input part of the sample data set (input data such as quality inspection indicator failure rate samples corresponding to the parallel combination results) and the output part (full inspection failure rate samples) are used as the input part. The output results (such as whether the final product is qualified or not) are taken as the analysis object, and the changing characteristics of the output results and the strength of the correlation between the input data and the output results are analyzed under different input data conditions. For example, if the sample value of the unqualified rate of a certain quality inspection indicator is changed, the changes in the qualified situation of the products reflected by the full inspection unqualified rate samples, as well as the degree of change and the law are observed. Based on this, the correlation degree between the input and output characteristics is determined. After performing the above-mentioned input and output feature correlation analysis on P sample data sets one by one, P input-output correlation degrees are generated according to the analysis results. Each input-output correlation degree accurately reflects the correlation characteristics of the input and output features of the corresponding sample data set, providing a data basis for subsequent comprehensive analysis and processing.

[0041] The target parallel combination result generation micro-unit is used to delete the parallel combination results whose indicator correlation or input-output correlation is less than the preset correlation according to the P indicator correlations and the P input-output correlations, and generate M target parallel combination results. Specifically, the target parallel combination result generation micro-unit generates M target parallel combination results based on the P indicator correlations and the P input-export correlations. First, the P indicator correlations and the P input-export correlations generated by the first two micro-units are obtained. These correlation data reflect the correlation characteristics between indicators in the parallel combination results and the input and output correlation characteristics of the sample data set from different angles. As a basis for subsequent operations, each parallel combination result is screened according to the preset correlation standard. For each parallel combination result, its corresponding indicator correlation and input-export correlation are compared with the preset correlation. If the indicator correlation or input-export correlation of a parallel combination result is less than the preset correlation, it indicates that If the parallel combination result does not meet the preset requirements in terms of the closeness of indicator correlation or input-output correlation, it will be deleted. For example, if the preset correlation is 0.5, and the indicator correlation of a parallel combination result is 0.3 and the input-output correlation is 0.4, the parallel combination result will be deleted because the indicator correlation is less than the preset correlation. After screening all parallel combination results in this way, the parallel combination results whose indicator correlation or input-export correlation is less than the preset correlation are deleted, and the parallel combination results that meet the requirements are retained, and the number of them is M, that is, M target parallel combination results, which meet the preset requirements in terms of indicator correlation and input-output correlation, and provide an accurate and effective data basis for subsequent related operations.

[0042] The quality inspection result generation module 50 is used to generate the quality inspection result of the semiconductor to be inspected according to the fusion unqualified rate. Specifically, the quality inspection result generation module 50 first obtains the fusion unqualified rate generated by the previous series of complex quality inspection processes and data processing operations, which integrates multiple preset quality inspection index information to comprehensively reflect the overall quality status of the semiconductor to be inspected and the batch of products to be inspected. Then, the quality inspection result is generated based on the judgment criteria that are pre-determined based on industry specifications, internal quality requirements of the enterprise, and a large amount of past quality inspection experience. The judgment criteria may set a specific quality threshold. If the fusion unqualified rate is higher than the threshold, it means that there are major problems with the overall quality of the batch of products to be inspected and it does not meet the requirements; if it is lower than the threshold, it is preliminarily determined that the overall quality meets the requirements and can enter the next process or be delivered for use. There may also be complex rules for further detection or analysis of the fusion unqualified rate in a specific range due to different product application scenarios, special customer requirements, etc. Then, the quality inspection results are formally generated based on the set judgment standards. If the fusion unqualified rate meets the quality qualification conditions, the overall quality of the batch of products to be inspected is judged to be qualified and a detailed quality inspection report is generated to indicate information such as the product quality meets the requirements; if it does not meet the quality qualification conditions, the overall quality is judged to be unqualified and a report is generated to explain the problem and recommend measures such as rework and adjustment of production processes to improve product quality, so as to accurately generate the quality inspection results of the semiconductors to be inspected, providing important basis for subsequent links.

[0043] The embodiment of the present application adopts a quality inspection index determination module for determining multiple preset quality inspection indexes of the semiconductor to be inspected; a distribution probability analysis module for performing defect distribution probability analysis on each quality inspection index to generate a defect probability; a sampling detection module for performing sampling detection according to the defect probability and the number of products in the batch to be inspected to generate multiple unqualified rates; a compensation fusion module for inputting the unqualified rate into a parallel fusion network for compensation fusion to generate a fused unqualified rate; a quality inspection result generation module for generating a final quality inspection result according to the fused unqualified rate, thereby achieving the technical effect of improving the quality inspection efficiency of semiconductor chip packaging and ensuring the accuracy of quality inspection.

[0044] In the above, refer to Figure 1 A semiconductor chip packaging quality inspection device according to an embodiment of the present invention is described in detail. Figure 2 A semiconductor chip packaging quality inspection method according to an embodiment of the present invention is described.

[0045] A semiconductor chip packaging quality inspection method, such as Figure 2 As shown, the method includes:

[0046] Determine multiple preset quality inspection indicators of the semiconductor to be inspected; perform defect distribution probability analysis on the multiple preset quality inspection indicators respectively to generate multiple defect probabilities; perform sampling inspection of multiple inspection indicators based on the multiple defect probabilities and the number of batches of products to be inspected of the semiconductor to be inspected to generate multiple indicator failure rates; input the multiple indicator failure rates into a pre-trained parallel fusion network for parallel existence compensation fusion to generate a fusion failure rate; generate a quality inspection result of the semiconductor to be inspected based on the fusion failure rate.

[0047] In one possible implementation, the semiconductor chip packaging quality inspection method also includes: collecting a historical quality inspection data set based on the multiple preset quality inspection indicators and the model of the semiconductor to be inspected; extracting a first historical quality inspection data set corresponding to a first preset quality inspection indicator based on the historical quality inspection data set; counting the proportion of data in the first historical quality inspection data in which the first preset quality inspection indicator is marked as unqualified, generating a first defect probability, and adding it to the multiple defect probabilities.

[0048] In a possible implementation, the semiconductor chip packaging quality inspection method further includes: configuring the sampling inspection ratio of the multiple preset quality inspection indicators based on the ratio of the multiple defect probabilities; performing random sampling according to the sampling inspection ratio and the number of products in the batch to be inspected to generate multiple sampling results; inspecting the multiple sampling results through the quality inspection modules corresponding to the multiple preset quality inspection indicators, and counting the unqualified rates to generate the unqualified rates of the multiple indicators.

[0049] In one possible implementation, the semiconductor chip packaging quality inspection method also includes: determining multiple inspection parameter sets corresponding to the multiple preset quality inspection indicators; performing parameter statistics from macro defects to micro defects for the multiple inspection parameter sets to generate a common accuracy requirement that meets macro defect detection and micro defect detection; configuring the inspection module and the inspection analysis network according to the common accuracy requirement to generate a quality inspection module corresponding to the multiple preset quality inspection indicators; using the inspection module in the quality inspection module to collect inspection data for the multiple sampling results and then inputting the data into the inspection analysis network, and statistically analyzing the unqualified rate based on the inspection analysis results to generate the unqualified rates of the multiple indicators.

[0050] In a possible implementation, the semiconductor chip packaging quality inspection method further includes: combining the plurality of preset quality inspection indicators in parallel to generate P parallel combination results, wherein the number of indicators in each parallel combination result is at least 2, P is an integer greater than 0, and Q is the total number of the multiple preset quality inspection indicators; the parallel fusion network is constructed in combination with the P parallel combination results and the historical quality inspection data set, wherein the parallel fusion network includes M parallel fusion integration models, wherein M≤P; the multiple indicator failure rates are combined and input into the M parallel fusion integration models for analysis and fusion, and the output M parallel fusion results are weighted calculated to generate the fusion failure rate.

[0051] In a possible implementation, the semiconductor chip packaging quality inspection and detection method also includes: extracting data from the historical quality inspection data set in combination with the P parallel combination results to generate P sample data sets, wherein any sample data set includes a quality inspection indicator failure rate sample and a corresponding full inspection failure rate sample corresponding to the parallel combination result; performing indicator correlation and input and output feature correlation analysis on the P parallel combination results, performing dimensionality reduction based on the analysis results, and generating M target parallel combination results; obtaining M sample data sets corresponding to the M target parallel combination results, training the M parallel fusion integration models based on the machine learning model, and obtaining convergence accuracy; configuring the result weighting layer of the M parallel fusion integration models with the convergence accuracy, and integrating and connecting them with the M parallel fusion integration models to generate the parallel fusion network.

[0052] In a possible implementation, the semiconductor chip packaging quality inspection method also includes: performing unqualified correlation analysis on the indicators within the P parallel combination results to generate P indicator correlations; performing feature correlation analysis of input and output on P sample data sets to generate P input-output correlations; based on the P indicator correlations and the P input-output correlations, deleting parallel combination results whose indicator correlations or input-output correlations are less than a preset correlation, to generate M target parallel combination results.

[0053] A semiconductor chip packaging quality inspection and testing device provided in an embodiment of the present invention can execute a semiconductor chip packaging quality inspection and testing method provided in any embodiment of the present invention, and has corresponding functional modules and beneficial effects of the execution method.

[0054] Although the present application makes various references to certain modules in the device according to the embodiments of the present application, any number of different modules may be used and run on the user terminal and / or server, and the various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of the present invention.

[0055] The above specific implementation manner does not constitute a limitation to the protection scope of the present application. It should be understood by those skilled in the art that various modifications, combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present application should be included in the protection scope of the present application. In some cases, the actions or steps recorded in the present application can be performed in an order different from that in the embodiment and can still achieve the desired results. In addition, the process depicted in the accompanying drawings does not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

Claims

1. A semiconductor chip packaging quality inspection device, characterized in that: include: A quality inspection index determination module, the quality inspection index determination module is used to determine a plurality of preset quality inspection indexes of the semiconductor to be inspected; A distribution probability analysis module, the distribution probability analysis module is used to perform defect distribution probability analysis on the multiple preset quality inspection indicators respectively to generate multiple defect probabilities; A sampling inspection module, the sampling inspection module is used to perform sampling inspections of multiple inspection indicators based on the multiple defect probabilities and the number of products in the batch to be inspected of the semiconductor to be inspected, and generate multiple indicator failure rates; A compensation fusion module, wherein the compensation fusion module is used to input the multiple indicator failure rates into a pre-trained parallel fusion network for parallel existence compensation fusion to generate a fusion failure rate; A quality inspection result generating module, the quality inspection result generating module is used to generate a quality inspection result of the semiconductor to be inspected according to the fusion failure rate; The compensation fusion module includes: A parallel combination result generating unit, the parallel combination result generating unit is used to perform parallel combination on the plurality of preset quality inspection indicators to generate P parallel combination results, wherein the number of indicators in each parallel combination result is at least 2, , P is an integer greater than 0, and Q is the total number of the plurality of preset quality inspection indicators; A parallel fusion network construction unit, the parallel fusion network construction unit is used to combine the P parallel combination results and the historical quality inspection data set to construct the parallel fusion network, wherein the parallel fusion network includes M parallel fusion integration models, wherein M≤P; A weighted calculation unit, the weighted calculation unit is used to combine the multiple indicator failure rates and input them into the M parallel fusion integration models for analysis and fusion, and perform weighted calculation on the output M parallel fusion results to generate the fusion failure rate; The parallel fusion network construction unit includes: A sample data set generation subunit, the sample data set generation subunit is used to extract data from the historical quality inspection data set in combination with the P parallel combination results to generate P sample data sets, wherein any sample data set includes a quality inspection indicator failure rate sample corresponding to the parallel combination result and a corresponding full inspection failure rate sample; A feature correlation analysis subunit, the feature correlation analysis subunit is used to perform an indicator correlation analysis and an input and output feature correlation analysis on the P parallel combination results, perform dimensionality reduction according to the analysis results, and generate M target parallel combination results; A convergence accuracy acquisition subunit, the convergence accuracy acquisition subunit is used to obtain M sample data sets corresponding to the M target parallel combination results, train the M parallel fusion integration models based on a machine learning model, and obtain convergence accuracy; A parallel fusion network generation subunit is used to configure the result weighted layer of M parallel fusion integration models with the convergence accuracy, and integrate and connect with the M parallel fusion integration models to generate the parallel fusion network.

2. A semiconductor chip packaging quality inspection device as claimed in claim 1, characterized in that: The distribution probability analysis module comprises: A data set collection unit, the data set collection unit is used to collect historical quality inspection data sets based on the plurality of preset quality inspection indicators and the model of the semiconductor to be inspected; A quality inspection data extraction unit, the quality inspection data extraction unit is used to extract first historical quality inspection data corresponding to a first preset quality inspection indicator based on the historical quality inspection data set; A first defect probability generating unit, wherein the first defect probability generating unit is used to count the proportion of data in the first historical quality inspection data that is marked as unqualified by the first preset quality inspection indicator, generate a first defect probability, and add it to the multiple defect probabilities.

3. A semiconductor chip packaging quality inspection device as claimed in claim 1, characterized in that: The sampling detection module comprises: A detection ratio configuration unit, the detection ratio configuration unit is used to configure the sampling detection ratios of the plurality of preset quality inspection indicators based on the ratios of the plurality of defect probabilities; A random sampling unit, the random sampling unit is used to perform random sampling according to the sampling detection ratio and the number of the batch of products to be tested, and generate multiple sampling results; A result detection unit is used to detect the multiple sampling results through the quality inspection modules corresponding to the multiple preset quality inspection indicators, and to count the unqualified rates to generate the unqualified rates of the multiple indicators.

4. A semiconductor chip packaging quality inspection device as claimed in claim 3, characterized in that: The result detection unit comprises: A detection parameter set determination subunit, the detection parameter set determination subunit is used to determine a plurality of detection parameter sets corresponding to the plurality of preset quality inspection indicators; A parameter statistics subunit, the parameter statistics subunit is used to perform parameter statistics from macro defects to micro defects for the multiple detection parameter sets, and generate parameters that meet the common accuracy requirements of macro defect detection and micro defect detection; A quality inspection module generation subunit, the quality inspection module generation subunit is used to configure the detection module and the detection analysis network according to the common accuracy requirement, and generate quality inspection modules corresponding to the multiple preset quality inspection indicators; The unqualified rate statistical subunit is used to collect detection data of the multiple sampling results by the detection module in the quality inspection module and input them into the detection analysis network, and to calculate the unqualified rate according to the detection analysis results to generate the unqualified rates of the multiple indicators.

5. A semiconductor chip packaging quality inspection device as claimed in claim 1, characterized in that: The feature correlation analysis subunit comprises: An indicator correlation generation micro-unit, wherein the indicator correlation generation micro-unit is used to perform unqualified correlation analysis on the indicators in the P parallel combination results to generate P indicator correlations; An input-output correlation generation micro-unit, wherein the input-output correlation generation micro-unit is used to perform feature correlation analysis between input and output for P sample data sets to generate P input-output correlations; The target parallel combination result generation micro unit is used to delete the parallel combination results whose indicator correlation or input-output correlation is less than the preset correlation according to the P indicator correlations and the P input-output correlations, and generate M target parallel combination results.

6. A semiconductor chip packaging quality inspection method, characterized in that: The method is applied to a semiconductor chip packaging quality inspection device according to any one of claims 1 to 5, and the method comprises: Determine multiple preset quality inspection indicators of the semiconductor to be inspected; Perform defect distribution probability analysis on the multiple preset quality inspection indicators respectively to generate multiple defect probabilities; Performing sampling inspection on multiple inspection indicators based on the multiple defect probabilities and the number of inspection batches of the semiconductor to be inspected, and generating multiple indicator failure rates; Inputting the multiple indicator failure rates into a pre-trained parallel fusion network for parallel existence compensation fusion to generate a fusion failure rate; A quality inspection result of the semiconductor to be inspected is generated according to the fusion failure rate.

Images