Packaging method and system for semiconductor chips

Abstract

The application relates to the technical field of semiconductor packaging, and discloses a packaging method and system for a semiconductor chip. The method comprises the following steps: acquiring temperature, pressure and lead interconnection data in a production environment in real time, and extracting parameter fluctuation characteristics through a time sequence analysis method; matching real-time characteristic data with parameter combination records in a historical database to determine a backtracking combination candidate; iteratively optimizing the backtracking combination by using an optimization algorithm to generate an optimized adjustment scheme; verifying the optimized scheme by using a simulation module, updating the historical database and recording a dynamic mechanism; and finally reclassifying stable state groups according to the updated historical database, calibrating production parameters and optimizing packaging quality. The application can effectively reduce a packaging defect rate, improve packaging quality and production stability, and can adapt to changes in different production environments in real time by fusing dynamic monitoring data and historical analysis.

Description

Technical Field

[0001] This application relates to the field of semiconductor packaging technology, and in particular to a packaging method and system for semiconductor chips. Background Technology

[0002] Semiconductor chip packaging is a crucial step in the integrated circuit manufacturing process, directly affecting chip performance, reliability, and production costs. With the continuous development of semiconductor technology, the complexity and precision requirements of packaging processes are gradually increasing, especially in high-density interconnects, large-scale integrated circuits, and high-frequency, high-speed applications, where the demands on packaging quality are becoming increasingly stringent.

[0003] Currently, many unpredictable defects exist in the packaging process, such as uneven stress in packaging materials and unstable lead interconnects, all of which can lead to chip failure. Traditional packaging quality control methods mainly rely on experience and preset standards, making it difficult to respond promptly to dynamic changes in the production process, and also unable to achieve automated and precise adjustment of production parameters. This not only leads to low production efficiency but also increases the risk of defects.

[0004] In recent years, with the continuous development of sensor technology and data analysis methods, real-time monitoring and big data analysis have become important means for the semiconductor packaging industry to improve quality and stability. However, existing technologies are mostly focused on data acquisition and single process optimization, lacking comprehensive analysis of multiple parameters such as temperature, pressure, and lead interconnection, and are insufficiently adaptable to dynamic changes in the production environment, making it difficult to achieve efficient and accurate packaging quality control.

[0005] Therefore, this invention aims to propose an optimization and adjustment method based on a combination of real-time data acquisition and historical data analysis. Through simulation verification and time series similarity calculation, production parameters are accurately calibrated, thereby achieving dynamic adjustment and quality control in the packaging process, effectively reducing packaging defects and improving production stability. Summary of the Invention

[0006] This application provides a packaging method and system for semiconductor chips, which aims to improve packaging quality, reduce defect rate, and achieve automation and intelligence in the production process by comprehensively analyzing dynamic changes in the production environment and accurately adjusting packaging parameters.

[0007] In a first aspect, this application provides a packaging method for a semiconductor chip, the method comprising: Step 1: Acquire temperature, pressure, and lead interconnection data in the production environment in real time, and process the acquired data using time series analysis methods to obtain parameter fluctuation characteristics; Step 2: Compare the parameter fluctuation characteristics with the preset fluctuation amplitude threshold, determine whether there is a potential defect risk, and extract real-time feature data based on the determination result; Step 3: Obtain parameter combination records from the historical database, classify the parameter combination records using a classification method, and obtain the stable state group; Step 4: By calculating temporal similarity, the stable state group is matched with the real-time feature data to determine the backtracking combination candidates; Step 5: Use an optimization algorithm to iteratively optimize the backtracking combination candidates, and integrate the current temperature data, pressure data and lead interconnection data to obtain an optimized adjustment scheme; Step 6: Verify the optimization and adjustment scheme through the simulation module, and update the historical database based on the verification results to record the dynamic mechanism; Step 7: Reclassify the stable state groups based on the updated historical database, determine the periodic pattern reference, and calibrate the production parameters.

[0008] Secondly, this application provides a packaging system for semiconductor chips, the system comprising: The data acquisition module is used to acquire temperature data, pressure data, and lead interconnection data in the production environment in real time, and to process the acquired data using time series analysis methods to obtain parameter fluctuation characteristics. The feature extraction module is used to compare the parameter fluctuation characteristics with the preset fluctuation amplitude threshold, determine whether there is a potential defect risk, and extract real-time feature data based on the judgment result. The data grouping module is used to retrieve parameter combination records from the historical database, classify the parameter combination records using a classification method, and obtain stable state groups. The candidate determination module is used to match the stable state group with real-time feature data through temporal similarity calculation to determine the backtracking combination candidates; The scheme acquisition module is used to iteratively optimize the backtracking combination candidates using optimization algorithms, and integrates the current temperature data, pressure data and lead interconnection data to obtain the optimized adjustment scheme. The verification and update module is used to verify the optimization and adjustment scheme through the simulation module, and update the historical database based on the verification results to record the dynamic mechanism. The parameter calibration module is used to reclassify the steady state group based on the updated historical database, determine the periodic pattern reference, and calibrate production parameters.

[0009] Compared with the prior art, the beneficial effects of the technical solution of this application are at least as follows: 1. By monitoring parameters such as temperature, pressure, and lead interconnection in the production environment in real time, factors that may lead to defects during the packaging process can be identified and adjusted promptly. Precise parameter calibration effectively controls packaging quality, reduces defect rates, and improves production stability.

[0010] 2. Through automated parameter adjustment and optimization scheme verification, the production efficiency of the packaging process can be significantly improved. Compared with traditional experience-based adjustment methods, data-driven approaches can accurately formulate adjustment schemes, avoiding errors and uncertainties associated with manual adjustments, and achieving higher production efficiency and lower production costs.

[0011] 3. Real-time acquisition and processing of temperature, pressure, and lead interconnection data, and automatic adjustment of production parameters based on dynamic changes, making it particularly suitable for high-frequency, complex production environments. When faced with constantly changing production conditions, the dynamic optimization mechanism provided by this invention can adapt to environmental changes in real time, ensuring that packaging quality is not affected by external factors.

[0012] 4. Through time series analysis and similarity calculation, potential defect risks can be accurately identified, especially in highly variable environments, where parameters can be automatically adjusted to reduce the occurrence of packaging defects. Furthermore, the verification through simulation modules further enhances the practicality and reliability of the solution, avoiding the risks associated with traditional methods that rely excessively on experience or static data.

[0013] 5. This invention provides a comprehensive closed loop for data acquisition, analysis, optimization, and feedback. Leveraging modern sensor technology, machine learning algorithms, and simulation verification methods, it promotes the intelligentization and automation of the packaging process. Through a data-driven decision support system, production process optimization can achieve low-intervention and high-efficiency operation, further advancing the development of intelligent manufacturing in the semiconductor packaging field. Attached Figure Description

[0014] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 This is a flowchart of the packaging method for semiconductor chips according to this application; Figure 2 This is a schematic diagram of the packaging system for semiconductor chips according to this application. Detailed Implementation

[0016] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms “comprising” or “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0017] For ease of understanding, the specific process of the embodiments of this application is described below. Figure 1 The diagram shows a flowchart of a semiconductor chip packaging method provided by the present invention, which specifically includes the following steps: Step 1: Acquire temperature, pressure, and lead interconnection data in the production environment in real time, and process the acquired data using time series analysis methods to obtain parameter fluctuation characteristics.

[0018] In one specific embodiment, the process of performing step 1 may specifically include the following steps: Temperature data, pressure data, and lead interconnect data are collected in real time through a sensor network. The lead interconnect data includes lead tension and interconnect resistance. Time series analysis was used to process temperature data, pressure data, and lead interconnection data to extract fluctuation amplitude and peak offset. The fluctuation amplitude is compared with the preset benchmark amplitude to generate the fluctuation amplitude comparison result; The peak offset is compared with a preset offset threshold to identify abnormal points and generate abnormal point information. By integrating the fluctuation amplitude comparison results and anomaly location information, parameter fluctuation characteristics that characterize the dynamic changes in temperature, pressure, and lead interconnection are generated.

[0019] Specifically, multiple sensor nodes are deployed in the production environment to collect temperature, pressure, and lead interconnect data in real time. Lead interconnect data, including lead tension and interconnect resistance, is crucial for detecting abnormal changes during the packaging process. The appropriate sensor placement ensures comprehensive and real-time data acquisition. For example, on a semiconductor packaging line, sensors can monitor temperature fluctuations and pressure changes in real time, and collect lead tension and interconnect resistance values. Lead tension refers to the tensile force on the lead wire during connection, and interconnect resistance refers to the resistance between connection points. The temperature, pressure, lead tension, and interconnect resistance data are collected once per second. This high-frequency acquisition ensures the timeliness and accuracy of the data, enabling rapid response to abnormal changes occurring during production.

[0020] The collected raw data were initially filtered using a moving average filter to eliminate noise interference. Subsequently, time series analysis was employed to process the filtered temperature, pressure, and lead interconnect sequences. Specifically, an autoregressive moving average model was applied to analyze the dynamic changes of the sequences. Stationarity was achieved by calculating sequence differences, and model parameters were estimated to extract fluctuation patterns. The fluctuation amplitude, i.e., the difference between the maximum and minimum values ​​within a time window, was calculated from the processed sequences. Peak offset, i.e., the deviation between the actual peak time and the historical average peak time, was also identified. Fluctuation amplitude and peak offset provide quantitative indicators of the dynamic changes in temperature, pressure, and lead interconnect during the packaging process, including pressure fluctuation amplitude, temperature fluctuation amplitude, and lead interconnect fluctuation amplitude, as well as pressure peak offset, temperature peak offset, and lead interconnect peak offset.

[0021] The calculated fluctuation amplitude is compared point-by-point with a preset benchmark amplitude. If it exceeds the benchmark, it is marked as high fluctuation; otherwise, it is marked as normal, generating a fluctuation amplitude comparison result sequence. The benchmark amplitude is usually calculated based on historical stable data; for example, 1.5 times the average fluctuation amplitude is taken as the upper limit of the fluctuation amplitude. Simultaneously, the peak offset is compared with a preset offset threshold. If the offset exceeds the threshold, that time point is marked as an anomaly, generating anomaly information. The peak offset identification threshold can be adjusted for different parameters; for example, the threshold for pressure data is 3 units. Finally, the fluctuation amplitude comparison results and anomaly information are fused to generate parameter fluctuation characteristics characterizing the dynamic changes in temperature, pressure, and lead interconnection. These characteristics include multi-dimensional data such as the average fluctuation amplitude, peak offset standard deviation, anomaly density, current fluctuation amplitude, current peak offset value, current temperature data, pressure data, and lead interconnection data.

[0022] This technology can monitor key parameters in the packaging process in real time and accurately extract dynamic change features through time-series analysis. It not only effectively improves the monitoring accuracy of packaging quality, but also enables early detection of potential packaging defects, reduces scrap rate, and improves production efficiency, thereby solving the problem of lack of effective dynamic control of parameters in the packaging process.

[0023] Step 2: Compare the parameter fluctuation characteristics with the preset fluctuation amplitude threshold, determine whether there is a potential defect risk, and extract real-time feature data based on the judgment result.

[0024] In one specific embodiment, the process of performing step 2 may specifically include the following steps: Determine whether the fluctuation amplitude in the parameter fluctuation characteristics is greater than the fluctuation amplitude threshold. If so, trigger the quality monitoring module. The quality monitoring module uses a trained machine learning classifier to analyze parameter fluctuation characteristics in order to determine whether there are potential packaging defects. If a potential packaging defect risk is identified, real-time feature data is extracted from the parameter fluctuation characteristics. The real-time feature data includes at least the current fluctuation amplitude and peak offset value.

[0025] Specifically, if the fluctuation amplitude exceeds the corresponding fluctuation amplitude threshold, it indicates that there has been a significant fluctuation in temperature, pressure, or lead interconnection status in the current production environment, which may lead to potential defect risks. Therefore, the system immediately triggers the quality monitoring module to initiate deep analysis. In the quality monitoring module, the parameter fluctuation feature vector is analyzed using a trained machine learning classifier. This classifier is trained using historical normal and abnormal production data, and can map the input features to a high-dimensional space and distinguish states through a separating hyperplane; the classifier outputs a defect risk probability value, and when this probability value exceeds a set threshold, it is determined that there is a potential packaging defect risk. For example, when the fluctuation amplitude of temperature data (12%) exceeds the corresponding set threshold (10%), the system will automatically activate the quality monitoring module to further evaluate the current parameter fluctuation characteristics and determine whether potential packaging defects have occurred, such as thermal stress caused by uneven temperature or packaging deformation caused by unstable pressure. To improve the accuracy of the judgment, the quality monitoring module will also combine data from historical stable state groups and use classification techniques such as temporal similarity calculation and support vector machines to further verify whether the current data is in an abnormal range.

[0026] Once a potential packaging defect risk is identified, the system extracts real-time feature data from the parameter fluctuation characteristics, including the current fluctuation amplitude value and the current peak offset value, as well as temperature data, pressure data, lead interconnection data, etc., to form a concise feature subset. This feature subset is stored in a temporary database, and a data synchronization mechanism is used to ensure compatibility with the historical database, providing a data foundation for subsequent matching operations with the stable state group.

[0027] This technical solution enables real-time monitoring and defect prediction during semiconductor chip packaging production, improving the automation level of the production line and the stability of product quality, while reducing the scrap rate. Simultaneously, by combining machine learning classifiers with real-time feature data extraction, it can provide early warnings of minute changes in the production process, preventing the accumulation of potential problems and effectively improving overall production efficiency and packaging quality.

[0028] Step 3: Obtain parameter combination records from the historical database, classify the parameter combination records using a classification method, and obtain the stable state group.

[0029] In one specific embodiment, the process of performing step 3 may specifically include the following steps: Extract parameter combination records containing temperature data, pressure data, and lead interconnection data from the historical database; A support vector machine classifier is used to train and classify the parameter combination records, resulting in several initial classification groups; For any initial classification group, time series analysis is used to extract its periodic pattern features; The periodic pattern characteristics are compared with the preset stability conditions to select the parameter combinations that meet the preset stability conditions. The selected parameter combinations are stored as a steady-state group, which includes stable parameter combinations for temperature distribution, pressure gradient, and lead interconnection.

[0030] Specifically, parameter combination records containing temperature, pressure, and lead interconnect data are extracted from historical databases. These records are typically collected by sensors in the production environment and stored in the database to reflect real-time data and status at different production stages. This data provides a comprehensive understanding of the changes and fluctuations in various indicators during the production process, particularly regarding key factors such as temperature, pressure, and lead interconnects.

[0031] After acquiring this data, a Support Vector Machine (SVM) classifier is used to train and classify the records, generating initial classification groups. For the extracted parameter combination records, the SVM uses historical labeled data (e.g., stable and unstable labels) to train the model. In this process, temperature, pressure, and lead interconnection data are converted into multidimensional feature vectors. Radial basis function kernels are used to map the features to a high-dimensional space. By finding the maximum margin hyperplane, the parameter combinations are divided into stable and unstable classes, resulting in several initial classification groups.

[0032] For any initial classification group, time-series analysis is used to extract its periodic pattern features. These periodic pattern features are identified by using time-series analysis to pinpoint recurring patterns from the steady-state group. The analysis process includes: first, performing a Fourier transform on the parameter sequences within the group to detect frequency components and identify the dominant period, such as the daily temperature fluctuation cycle; then, calculating the autocorrelation function to confirm pattern repeatability; for example, if the autocorrelation peak appears at a fixed lag, periodicity is confirmed; finally, features such as cycle length and fluctuation amplitude are extracted for subsequent comparison. This extraction process helps capture the dynamic patterns of the production environment and improves the accuracy of defect prediction.

[0033] The extracted periodic pattern features are compared with preset stability conditions to screen parameter combinations that meet these conditions. These conditions include an allowable range for cycle length and a fluctuation amplitude threshold. For example, if the cycle length is within the preset range and the amplitude is below the threshold, it is considered compliant. Stability conditions can be set based on production experience and historical data; for example, the amplitude of the periodic pattern in temperature distribution should not exceed 5%, and the fluctuation amplitude of the pressure gradient should be less than a certain threshold. This comparison effectively eliminates unstable parameter combinations caused by abnormal fluctuations, retaining only those records with good stability. This screening process typically uses similarity calculation techniques, such as Dynamic Time Warping (DTW), to measure the similarity between the current pattern and historical stable patterns, thereby improving the accuracy of the screening. The screened parameter combinations must meet all conditions, such as stable resistance measurements of lead interconnects within the cycle. The screened parameter combinations are stored as a stable state group, which contains stable parameter combinations of temperature distribution patterns, pressure gradient features, and lead interconnects. These combinations represent the ideal parameter range under normal production conditions.

[0034] The technical solution of this invention can effectively extract stable production patterns from historical data and identify and predict defect risks based on these patterns. Compared with traditional experience-based judgment methods, the use of support vector machine classification and time series analysis can more objectively and accurately screen out stable parameter combinations, improving the system's automation level and predictive capabilities. Especially in the field of semiconductor chip packaging, minute fluctuations in temperature, pressure, and lead interconnection during the production process can directly affect packaging quality. Therefore, extracting stable state groups based on historical data and performing periodic pattern analysis has significant technical advantages for improving production efficiency and packaging quality.

[0035] Step 4: By calculating temporal similarity, the stable state group is matched with the real-time feature data to determine the backtracking combination candidates.

[0036] In one specific embodiment, the process of performing step 4 may specifically include the following steps: A temporal similarity calculation method is used to calculate the similarity score between the steady-state group and the real-time feature data; If the similarity score is lower than the preset score threshold, then anomaly points are marked on the time series corresponding to the real-time feature data. Based on the marked outlier locations, select the group from the stable state group that is closest to the marked outlier locations as the backtracking candidate group. Obtain the temperature distribution mapping and pressure gradient data of the current production environment, and fuse them with the backtracking candidate group to generate a backtracking combined candidate.

[0037] Specifically, a temporal similarity calculation method is used to match the steady-state group with the real-time feature data and calculate their similarity score. The steady-state group is extracted from a historical database and contains stable data from different production cycles, while the real-time feature data comes from data such as temperature, pressure, and lead interconnection collected in real time by the sensor network.

[0038] Real-time feature data is compared with the parameter combinations recorded in each steady-state group to calculate a similarity score. A set of similarity scores is obtained using a time-series similarity calculation method (e.g., Dynamic Time Warping (DTW)) to measure the degree of matching between each steady-state record and the real-time feature data. Choosing the Dynamic Time Warping (DTW) algorithm to calculate similarity effectively solves problems such as nonlinear deformation matching of time-series data along the time axis during semiconductor manufacturing. If the maximum similarity score is lower than a preset threshold (e.g., 0.7), it indicates a significant difference between the current production state and historical steady-state states. In this case, it is necessary to mark outlier points on the time-series sequence of the real-time feature data, indicating data points with large fluctuations. The Local Anomaly Factor (LOF) method is used to identify outliers. LOF is a density-based anomaly detection algorithm that calculates the ratio of the local density of each point to the average density of its neighborhood. If the ratio is lower than a set threshold, it is determined to be an outlier.

[0039] Preferably, if any similarity value in the similarity score array is lower than a preset threshold, it can be determined as an anomaly and the anomaly point needs to be marked; or, if multiple values ​​in the similarity score array are lower than the threshold, it is determined as an anomaly and the anomaly point needs to be marked.

[0040] After identifying outliers, the group closest to the identified outliers from the historical stable state groups is selected as the backtracking candidate group. This selection process is based on the similarity of time-series data and can find the closest stable state group by calculating Euclidean distance, among other methods. The selected backtracking candidate group can reflect historically successful production patterns and parameter combinations under similar production conditions, providing a basis for adjusting the current production process.

[0041] The system acquires the current production environment's temperature distribution map and pressure gradient data, and then fuses them with the backtracking candidate group. The temperature distribution map is generated as a two-dimensional temperature field by spatial interpolation of sensor point data using the Kriging interpolation method; the pressure gradient data is obtained by calculating the pressure change rate between adjacent sensor points using the finite difference method. The parameters of the backtracking candidate group are weighted and fused with the current temperature distribution map, and then integrated with the pressure gradient vector to finally generate a combined backtracking candidate group containing spatial temperature distribution and pressure gradient information. For example, the candidate's temperature parameter multiplied by a weight of 0.6 is added to the map value multiplied by a weight of 0.4, and then integrated with the pressure gradient vector by a dot product to form fused data. Through this fusion process, multiple environmental factors in the production process can be considered, such as the impact of temperature changes on physical stress and the distribution of pressure gradients at different locations, ensuring that the backtracking candidates are not merely data matches, but a comprehensive consideration of the interaction of multiple factors in the production process.

[0042] This technical solution effectively solves the problem of inaccurate matching between real-time and historical data during the production process, ensuring that the most representative stable state can be quickly located in complex production environments, providing reliable data support for optimization and adjustment, and significantly improving the stability of the production process and the accuracy of defect identification.

[0043] Step 5: Use an optimization algorithm to iteratively optimize the backtracking combination candidates, and integrate the current temperature data, pressure data and lead interconnection data to obtain an optimized adjustment scheme.

[0044] In one specific embodiment, the process of performing step 5 may specifically include the following steps: Using backtracking combination candidates as the initial population, a genetic algorithm is used for iterative optimization to generate optimal parameter combinations; Based on the current lead interconnect data, perform lead tension calibration and interconnect resistance measurement to obtain calibrated lead interconnect data; The optimized parameter combination, current temperature data, current pressure data, and calibrated lead interconnection data are weighted and fused to generate an initial adjustment scheme. Through a data synchronization mechanism, the parameters in the initial adjustment plan are verified and ensured to match the real-time production data, thereby obtaining an optimized adjustment plan.

[0045] Specifically, backtracking candidate combinations are used as the initial population for the optimization process. Each candidate combination represents a set of historical data records, including key parameters such as temperature, pressure, and lead interconnection. A genetic algorithm optimizes these combinations across multiple generations by simulating natural selection, crossover, and mutation mechanisms. During each generation, the algorithm evaluates the fitness of each backtracking combination based on its temporal similarity to real-time data collected in the current production environment. If the temporal similarity of a combination is higher than a preset threshold, that combination is considered to have high fitness and can better predict or match the current production state. New backtracking combinations are generated by crossover operations on combinations with high fitness, and mutation operations are introduced to avoid local optima, thereby increasing the probability of finding the global optimum. This process continues iterating until convergence, ultimately generating an optimized parameter combination.

[0046] Building upon this, further calibration operations are performed on the real-time acquired lead interconnect data, including lead tension calibration and interconnect resistance measurement. Using a tension sensor, the real-time lead tension value is compared with a standard value. If a deviation exists, appropriate adjustments are made to ensure the accuracy of the tension data. For example, if the real-time lead tension is 5 Newtons and the standard is 4 Newtons, a calibration force of 1 Newton is applied. Simultaneously, a resistance meter is used to accurately record the resistance value, and outliers are marked as needed. This calibration process ensures the reliability of the lead interconnect data and eliminates potential measurement errors, providing more accurate data support for optimizing adjustment schemes.

[0047] After the candidate backtracking combinations are optimized by a genetic algorithm, the current temperature and pressure data, along with the calibrated lead interconnection data, are integrated to obtain a comprehensive optimization and adjustment scheme. In this process, a weighted fusion method is used to combine different data sources according to predetermined weights. For example, the weight of the optimized parameter combination is 0.6, and the weight of the real-time data is 0.4, to reflect the different impacts of each parameter on the optimization scheme. Through weighted fusion, the relative importance of each parameter can be reasonably adjusted in the actual production environment, thereby obtaining an adjustment scheme that meets actual needs. For example, the optimized parameter combination is (temperature 60℃, pressure 2.3MPa, lead tension 4.9N, interconnection resistance 0.50Ω), the real-time collected data is (temperature 62℃, pressure 2.5MPa, lead tension 4.8N, interconnection resistance 0.52Ω), and the initial adjustment scheme obtained through weighted fusion is (temperature 60.8℃, pressure 2.38MPa, lead tension 4.86N, interconnection resistance 0.508Ω).

[0048] Finally, the generated initial adjustment plan needs to be verified through a data synchronization mechanism to ensure that the parameters in the plan match the real-time production data. Clock synchronization ensures that each adjustment parameter is aligned with the real-time data. If the matching degree is lower than a predetermined threshold (e.g., 0.9), data fusion needs to be repeated until the plan perfectly matches the real-time production data. This mechanism effectively avoids excessive deviations between optimized parameters and actual production, ensuring that the optimized adjustment plan has high accuracy and adaptability in practical applications.

[0049] This technical solution enables precise adjustment of production environment parameters, significantly improving the quality of semiconductor packaging, reducing the risk of packaging defects, and increasing production efficiency. Furthermore, this optimization process can adapt to variations in different production batches, ensuring effective risk control and performance optimization under different environmental conditions, thus producing significant technical benefits in practical applications.

[0050] Step 6: Verify the optimization and adjustment scheme through the simulation module, and update the historical database based on the verification results to record the dynamic mechanism.

[0051] In one specific embodiment, the process of performing step 6 may specifically include the following steps: The optimization and adjustment scheme is input into the simulation module based on finite element analysis to generate the corresponding simulated packaging quality results; Acquire real-time monitoring data and compare and analyze it with simulated packaging quality results; If the deviation between the simulated packaging quality result and the real-time monitoring data is lower than the preset deviation threshold, the optimization and adjustment scheme is confirmed to be effective. Confirmed and effective optimization and adjustment schemes are recorded as parameter combinations and stored in the historical database, generating new dynamic mechanism records. Based on the dynamic mechanism records, the parameters of the peak offset recognition model are updated to enhance the sensitivity of production anomaly detection.

[0052] Specifically, the optimization and adjustment scheme includes key parameters of the production process, such as temperature, pressure, lead interconnection data, temperature distribution mapping, and pressure gradient data. After this data is imported into the simulation module, the module simulates the packaging process using finite element analysis (FEM) to generate corresponding simulated packaging quality results. These results typically include potential defect risk values ​​during the packaging process, such as the probability of defect occurrence. For example, the optimization and adjustment scheme is input into a FEM-based simulation module. This module meshes the temperature distribution mapping in the scheme to generate a simulated temperature field. Based on this simulated temperature field and combined with pressure gradient data, it calculates the stress distribution of the packaging material and outputs simulated packaging quality results containing defect risk values. For instance, in semiconductor chip packaging, assuming the optimization and adjustment scheme sets the temperature to 80 degrees Celsius and the pressure to 2 MPa, the simulation module might generate a defect risk value of 5%, providing fundamental data for further verification of the scheme's effectiveness.

[0053] Based on this, real-time monitoring data is collected in real time via a sensor network, including temperature, pressure, and lead interconnection data in the current production environment. The acquired real-time data is compared with the corresponding quality indicators in the simulation results, and the deviation value is calculated. If the deviation between the simulated packaging quality results and the real-time monitoring data is less than a preset threshold (e.g., 2%), the effectiveness of the optimization and adjustment scheme in the actual production environment can be confirmed. For example, if the real-time data shows a temperature of 82 degrees Celsius and a defect risk value of 6%, while the simulation results show a risk value of 5%, the deviation between the two is 1%, and the scheme is considered effective.

[0054] Once the optimization and adjustment scheme is confirmed to be effective, the relevant scheme parameters (such as temperature distribution mapping, pressure gradient, temperature data, pressure data, lead tension data, etc.) are stored in the historical database, forming new dynamic mechanism records. Each record not only saves the current timestamp but also includes a quantified matching score between the scheme and historical data. These data and records support subsequent real-time adaptation and anomaly detection. Through dynamic mechanism recording, key parameters in the production process can be continuously optimized and adjusted, enabling the system to iterate and evolve continuously in different production cycles.

[0055] Furthermore, based on the dynamic mechanism records, the peak offset identification model is updated accordingly. This model uses algorithms such as support vector machines to classify extracted fluctuation features (such as fluctuation amplitude and peak offset) to generate an updated stable state group. Through time series analysis, this model can not only identify potential anomalies in the production process but also adjust the labeling of anomaly points according to data trends and fluctuations, improving sensitivity to peak offset. When a new data cycle arrives, the updated model will be applied to real-time monitoring. By combining it with data such as temperature distribution mapping, the calibration of lead tension will be further optimized, improving the stability of packaging quality.

[0056] The optimization and adjustment scheme is verified through simulation, and real-time feedback and updates are provided based on the verification results to ensure that key parameters in the production process are continuously optimized. The continuous updates recorded by the dynamic mechanism and the adjustment of the peak offset recognition model not only improve the stability of the production process but also significantly enhance the sensitivity of anomaly detection, reduce the occurrence of defects in production, and significantly optimize the quality control of semiconductor chip packaging.

[0057] Step 7: Reclassify the stable state groups based on the updated historical database, determine the periodic pattern reference, and calibrate the production parameters.

[0058] In one specific embodiment, the process of performing step 7 may specifically include the following steps: Retrieve new parameter combination records from the updated historical database; A support vector machine classifier is used to reclassify the records with new parameter combinations, generating new stable state groups; For the new set of stable states, extract its periodic pattern features; By using a time-series similarity calculation method and integrating the abnormal point annotation information from real-time monitoring, a periodic pattern comparison benchmark for the next cycle is determined. Based on the periodic pattern reference and the temperature distribution mapping of the current production environment, the lead tension is calculated and calibrated to optimize the production environment.

[0059] Specifically, the latest parameter combination records are retrieved from the updated historical database. These records contain all relevant data from previous optimization and adjustment schemes, ensuring the timeliness and accuracy of the database. Each record includes production parameters such as temperature, pressure, lead tension, and interconnect resistance. Support Vector Machines (SVMs) are used to reclassify these parameter combination records, generating new stable state groups. As a powerful classification algorithm, SVMs can construct a classification hyperplane based on feature information from historical data, dividing parameter combination records into multiple categories, each representing a specific stable production state. This classification method effectively identifies which production parameter combinations are stable and which fluctuate significantly in different production cycles, thus helping to quickly identify potential production problems and improve the accuracy and controllability of production. For these new stable state groups, time series analysis methods are used to extract their periodic pattern features, including calculating the recurring fluctuation period of parameters and identifying the dominant frequency components. This process, by analyzing the fluctuation periods of parameters such as temperature and pressure in each data set, reveals periodic patterns in the production process. For example, certain temperature fluctuations may be related to specific packaging modes, and periodic temperature changes can indicate potential problems or optimization opportunities in the packaging process.

[0060] By utilizing temporal similarity calculation methods and integrating anomaly point annotation information from real-time monitoring data, a periodic pattern reference for the next cycle can be further determined. Temporal similarity calculation methods, such as Dynamic Time Warping (DTW), can measure the similarity between the fluctuations of various parameters in the current production cycle and historical cycles. Integrating anomaly point annotations marks peak offset points exceeding a threshold as anomalies and adjusts the similarity scores. Based on the adjusted similarity, a periodic pattern reference for the next cycle is determined, such as selecting the group with the highest similarity as the reference. This integration improves the robustness of pattern determination and reduces the risk of defects in production. For example, the similarity score can be adjusted based on the marked outliers. The adjustment can be achieved using the following strategies: 1) Increase the penalty on the similarity score. When an outlier exceeds a threshold in the current production cycle, the similarity score can be penalized. For example, each outlier can have a negative value added to its similarity score, or a penalty factor can be added. If there are too many outliers or the deviation is too large in the current cycle, its similarity score will decrease. 2) Reduce the contribution of outliers to the similarity score. By weighting outliers, the contribution of normal data points is made greater than that of outliers, thus making the system pay more attention to the fluctuation patterns of normal data when calculating the overall similarity. 3) Threshold adjustment. If the fluctuation range of outliers in certain cycles is particularly large, and these points occur frequently, further classification of these points can be considered to determine whether they belong to a special production state. For example, the threshold can be modified to further refine the classification of outliers, so that outliers are not considered ordinary production fluctuations, but are marked as new, independent patterns, thereby affecting the calculation of the similarity score.

[0061] Based on a defined periodic pattern reference and the temperature distribution mapping of the current production environment, a lead tension adjustment factor is calculated, and the lead tension parameters are calibrated accordingly to match the tension value with the temperature distribution gradient. Specifically, a temperature distribution mapping is generated, representing the temperature gradient distribution in the production environment using a gridding method; the periodic pattern reference is superimposed on the temperature distribution mapping to calculate the lead tension adjustment factor; the adjustment factor is applied to calibrate the lead tension parameters, for example, updating the tension value using a formula to match the gradient in the mapping, ensuring optimized production parameters. This calibration process can accurately calibrate the lead tension parameters, ensuring lead stability and resistance values ​​are within a suitable range, thereby optimizing package quality and reducing defects.

[0062] This technical solution, through support vector machine classification, temporal similarity calculation, and temperature distribution mapping fusion calibration, enables precise parameter optimization in complex production environments, significantly improving production stability and reducing defect risks caused by environmental changes. These closely integrated technical features constitute a significantly advanced packaging quality optimization technology, solving problems such as difficulty in quickly adapting to environmental changes and high defect rates. Furthermore, through continuous accumulation and analysis of historical data, production parameters and processes can be continuously optimized, improving the robustness and anti-interference capabilities of the entire production system.

[0063] The packaging method for semiconductor chips in the embodiments of this application has been described above. The packaging system for semiconductor chips in the embodiments of this application is described below. Please refer to [link / reference]. Figure 2 The present application provides a schematic diagram of a packaging system for semiconductor chips, the system comprising: The data acquisition module 10 is used to acquire temperature data, pressure data and lead interconnection data in the production environment in real time, and to process the acquired data using time series analysis methods to obtain parameter fluctuation characteristics.

[0064] The feature extraction module 20 is used to compare the parameter fluctuation characteristics with the preset fluctuation amplitude threshold, determine whether there is a potential defect risk, and extract real-time feature data based on the judgment result.

[0065] The data grouping module 30 is used to retrieve parameter combination records from the historical database, classify the parameter combination records using a classification method, and obtain a stable state group.

[0066] The candidate determination module 40 is used to match the stable state group with the real-time feature data through time series similarity calculation to determine the backtracking combination candidates.

[0067] The scheme acquisition module 50 is used to iteratively optimize the backtracking combination candidates using optimization algorithms, and integrate the current temperature data, pressure data and lead interconnection data to obtain the optimized adjustment scheme.

[0068] The verification and update module 60 is used to verify the optimization and adjustment scheme through the simulation module, and update the historical database based on the verification results to record the dynamic mechanism.

[0069] The parameter calibration module 70 is used to reclassify the steady state group based on the updated historical database, determine the periodic pattern reference, and calibrate the production parameters.

[0070] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A packaging method for semiconductor chips, characterized in that, The method includes: Step 1: Acquire temperature, pressure, and lead interconnection data in the production environment in real time, and process the acquired data using time series analysis methods to obtain parameter fluctuation characteristics; Step 2: Compare the parameter fluctuation characteristics with the preset fluctuation amplitude threshold, determine whether there is a potential defect risk, and extract real-time feature data based on the determination result; Step 3: Obtain parameter combination records from the historical database, and classify the parameter combination records using a classification method to obtain stable state groups; Step 4: By calculating temporal similarity, the stable state group is matched with the real-time feature data to determine the backtracking combination candidates; Step 5: Iteratively optimize the backtracking combination candidates using an optimization algorithm, and integrate the current temperature data, pressure data, and lead interconnection data to obtain an optimized adjustment scheme; Step 6: Verify the optimization and adjustment scheme through the simulation module, and update the historical database based on the verification results to record the dynamic mechanism; Step 7: Reclassify the steady state group based on the updated historical database, determine the periodic pattern reference, and calibrate the production parameters; Step 1 includes: real-time acquisition of temperature data, pressure data, and lead interconnection data through a sensor network, wherein the lead interconnection data includes lead tension and interconnection resistance; A time-series analysis method is used to process temperature data, pressure data, and lead interconnection data to extract fluctuation amplitude and peak offset. The fluctuation amplitude is compared with a preset reference amplitude to generate a fluctuation amplitude comparison result. The peak offset is compared with a preset offset threshold to identify abnormal points and generate abnormal point information. The fluctuation amplitude comparison result and the abnormal point information are fused to generate parameter fluctuation characteristics that characterize the dynamic changes of temperature, pressure, and lead interconnection. Step 5 includes: using the backtracking candidate combinations as the initial population, performing iterative optimization using a genetic algorithm to generate optimized parameter combinations; based on the current lead interconnection data, performing lead tension calibration and interconnection resistance measurement to obtain calibrated lead interconnection data; weighting and fusing the optimized parameter combinations, current temperature data, current pressure data, and the calibrated lead interconnection data to generate an initial adjustment scheme; and verifying and ensuring that the parameters in the initial adjustment scheme match the real-time production data through a data synchronization mechanism, thereby obtaining the optimized adjustment scheme.

2. The method according to claim 1, characterized in that, Step 2 includes: Determine whether the fluctuation amplitude in the parameter fluctuation characteristics is greater than the fluctuation amplitude threshold. If so, trigger the quality monitoring module. The quality monitoring module uses a trained machine learning classifier to analyze the parameter fluctuation characteristics in order to determine whether there is a potential packaging defect risk. If the potential packaging defect risk is determined to exist, the real-time feature data is extracted from the parameter fluctuation characteristics. The real-time feature data includes at least the current fluctuation amplitude and peak offset value.

3. The method according to claim 1, characterized in that, Step 3 includes: Extract the parameter combination record containing temperature data, pressure data, and lead interconnection data from the historical database; The parameter combination records are trained and classified using a support vector machine classifier to obtain several initial classification groups. For any initial classification group, time series analysis is used to extract its periodic pattern features; The periodic pattern characteristics are compared with preset stability conditions to select parameter combinations that meet the preset stability conditions. The selected parameter combinations are stored as the stable state group, which includes stable parameter combinations of temperature distribution, pressure gradient, and lead interconnection.

4. The method according to claim 1, characterized in that, Step 4 includes: A temporal similarity calculation method is used to calculate the similarity score between the stable state group and the real-time feature data; If the similarity score is lower than a preset score threshold, then an anomaly point is marked on the time sequence corresponding to the real-time feature data; Based on the marked outlier locations, select the group from the stable state group that is closest to the marked outlier locations as the backtracking candidate group. The temperature distribution map and pressure gradient data of the current production environment are obtained and fused with the backtracking candidate group to generate the backtracking combination candidate.

5. The method according to claim 1, characterized in that, Step 6 includes: The optimization and adjustment scheme is input into the simulation module based on finite element analysis to generate the corresponding simulated packaging quality results; Acquire real-time monitoring data and compare and analyze it with the simulated packaging quality results; If the deviation between the simulated packaging quality result and the real-time monitoring data is lower than the preset deviation threshold, then the optimization and adjustment scheme is confirmed to be effective. The confirmed and effective optimization and adjustment schemes are recorded as parameter combinations and stored in the historical database to generate new dynamic mechanism records. Based on the dynamic mechanism recorded, the parameters of the peak offset recognition model are updated to enhance the sensitivity of production anomaly detection.

6. The method according to claim 1, characterized in that, Step 7 includes: Retrieve new parameter combination records from the updated historical database; The new parameter combination records are reclassified using a support vector machine classifier to generate a new stable state group; For the new set of stable states, extract its periodic pattern features; By using a time-series similarity calculation method and integrating the abnormal point annotation information from real-time monitoring, the periodic pattern reference for the next cycle is determined. Based on the periodic pattern reference and the temperature distribution mapping of the current production environment, the lead tension is calculated and calibrated to optimize the production environment.

7. A packaging system for semiconductor chips, used to implement the method as described in any one of claims 1 to 6, characterized in that, The system includes: The data acquisition module is used to acquire temperature data, pressure data, and lead interconnection data in the production environment in real time, and to process the acquired data using time series analysis methods to obtain parameter fluctuation characteristics. The feature extraction module is used to compare the parameter fluctuation characteristics with a preset fluctuation amplitude threshold, determine whether there is a potential defect risk, and extract real-time feature data based on the determination result. The data grouping module is used to retrieve parameter combination records from the historical database, classify the parameter combination records using a classification method, and obtain a stable state group. The candidate determination module is used to match the stable state group with the real-time feature data through temporal similarity calculation to determine backtracking combination candidates; The scheme acquisition module is used to iteratively optimize the backtracking combination candidates using an optimization algorithm, and to integrate the current temperature data, pressure data and lead interconnection data to obtain an optimized adjustment scheme; The verification and update module is used to verify the optimization and adjustment scheme through the simulation module, and update the historical database based on the verification results to record the dynamic mechanism. The parameter calibration module is used to reclassify the steady state group based on the updated historical database, determine the periodic pattern reference, and calibrate production parameters.

Images