A method for predicting reflow process quality based on data mining

Abstract

The present invention discloses a reflow soldering process quality prediction method based on data mining, which belongs to the field of microelectronic parts manufacturing technology, and includes the following steps: S1: preliminarily determining the influencing factors and quality indicators of the reflow soldering process quality and their value ranges; S2: screening the main influencing factors and quality indicators of the reflow soldering process quality and performing data collection; S3: establishing a reflow soldering process quality prediction model; S4: optimizing the reflow soldering process quality prediction model. The present invention establishes a simulation model of the reflow soldering process through multiple computer simulation software, and obtains a data set of the relationship between the main quality influencing factors and quality indicators through experimental design plus simulation. Compared with physical experiments, the experimental cost is greatly reduced and the research cycle of the process is shortened; the mainstream machine learning algorithm and the commonly used proxy model algorithm are compared, and the more advanced Boosting‑MKELM algorithm with more accurate calculation results is adopted, which can well meet the requirements of quality prediction and analysis.

Description

Technical Field

[0001] The invention relates to the technical field of microelectronic parts manufacturing, and in particular to a reflow soldering process quality prediction method based on data mining. Background Art

[0002] The integrated circuit industry has become the key to the development of the national economy, and integrated circuit design, integrated circuit manufacturing, integrated circuit packaging and integrated circuit testing are the four pillar industries of the development of the integrated circuit industry. They are both independent and inseparable, and not only affect the development of the electronic information industry and even the national economy, but are also closely related to the lives of every family. With the miniaturization and high performance of microelectronic systems, the impact of electronic packaging on the system has become as important as chips.

[0003] Chip production often faces a production model of multiple varieties, small batches, and short delivery cycles. It has become increasingly difficult to rely on experience and a large number of process tests to ensure product quality. The process that plays a vital role in the quality of the final product in the chip manufacturing process is mainly the reflow soldering process. At present, in the product design process, designers still manually preset parameters according to the properties of consumables and past experience based on the different products and consumables, and then obtain the final results through simulation. However, due to the non-unique quality indicators, it is sometimes necessary to establish multiple simulation models, and the simulation solution time is long, which undoubtedly increases the time cost in the design and production process. The above problems need to be solved urgently. To this end, a reflow soldering process quality prediction method based on data mining is proposed. Summary of the invention

[0004] The technical problem to be solved by the present invention is: how to complete the full-process information management of prior notification, in-process attention and post-event evaluation through the associated application and analysis of the early warning prediction information of the caring object based on the collection of information of each element in the grid, and provide a reflow soldering process quality prediction method based on data mining.

[0005] The present invention solves the above technical problems through the following technical solutions, and the present invention comprises the following steps:

[0006] S1: Preliminary determination of the factors affecting the quality of the reflow soldering process, quality indicators and their value ranges

[0007] Through literature and process research, the factors and quality indicators that affect the quality indicators of the reflow soldering process are preliminarily analyzed to obtain the factors that affect the quality indicators and the upper and lower limits of the quality indicators;

[0008] S2: Screening the main influencing factors and quality indicators of reflow process quality and data collection

[0009] Select fractional factorial test design to obtain sample points, use computer simulation method to obtain the response value of sample points, use Spearman correlation coefficient to analyze the correlation between quality indicators and factors affecting quality indicators, retain the combination of quality indicators and factors affecting quality indicators with strong correlation, determine the final factors and quality indicators that mainly affect quality indicators, and collect data through MES system based on the factors and quality indicators that mainly affect quality indicators;

[0010] S3: Establish a reflow process quality prediction model

[0011] The sample space is constructed by Latin hypercube sampling to obtain data and use it as modeling data. The non-continuous variables of reflow soldering are fixed, and multiple training models based on multiple algorithms are established and trained to obtain multiple algorithm models. The accuracy of the prediction results of multiple algorithm models is compared, and the algorithm model with the highest prediction accuracy is selected as the final reflow soldering process quality prediction model.

[0012] S4: Optimizing the reflow process quality prediction model

[0013] By using the existing actual production data and historical prediction data, the reflow soldering process quality prediction model is continuously trained to achieve the optimization of the reflow soldering process quality prediction model.

[0014] Furthermore, in step S2, the correlation coefficient between the quality indicator with strong correlation and the combination of factors affecting the quality indicator is greater than 0.5.

[0015] Furthermore, in step S3, the discontinuous variables of reflow soldering include solder joint types and solder paste material types.

[0016] Furthermore, in step S3, the algorithm includes a proxy model algorithm and a Boosting-MKELM algorithm.

[0017] Furthermore, in step S4, the actual production data are the factors and quality indicators that mainly affect the quality indicators and the equipment status parameter data collected by the MES system.

[0018] Compared with the prior art, the present invention has the following advantages: a simulation model of the reflow soldering process is established through multiple computer simulation software, and a data set of the relationship between the main quality influencing factors and quality indicators is obtained through experimental design plus simulation. Compared with physical experiments, the experimental cost is greatly reduced and the research cycle of the process is shortened; the mainstream machine learning algorithm and the commonly used proxy model algorithm are compared, and the more advanced Boosting-MKELM algorithm with more accurate calculation results is adopted, which can well meet the requirements of quality prediction and analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is an implementation flow chart of a reflow soldering process quality prediction method based on data mining according to Embodiment 1 of the present invention;

[0020] Figure 2 It is a flow chart of the technical route for developing a reflow soldering process quality prediction method based on data mining in the first embodiment of the present invention;

[0021] Figure 3 This is a prediction framework diagram based on the Boosting-MKELM algorithm in Example 1 of the present invention. DETAILED DESCRIPTION

[0022] The following is a detailed description of an embodiment of the present invention. This embodiment is implemented on the premise of the technical solution of the present invention, and a detailed implementation method and a specific operation process are given, but the protection scope of the present invention is not limited to the following embodiment.

[0023] Embodiment 1

[0024] like Figure 1 , 2 As shown, this embodiment provides a technical solution: a reflow soldering process quality prediction method based on data mining, comprising the following steps:

[0025] Step 1: Preliminary determination of influencing factors, quality indicators and their value ranges

[0026] Through the research of literature and process, the factors affecting the quality indexes of the reflow soldering process and the quality indexes are preliminarily analyzed, and the correlation between the parameters is preliminarily studied to realize the identification and simplification of key data, reduce the difficulty of model construction, and maintain the accuracy of the model. In order to avoid the interference of non-critical quality influencing factors and quality indicators on quality defect analysis, the data structure optimization analysis of key processes is carried out, and on this basis, the factors affecting the quality indexes and the upper and lower limits of the quality indexes are obtained.

[0027] Step 2: Screening of main influencing factors and quality indicators and data collection

[0028] In order to obtain the main influencing factors and quality indicators of the reflow soldering process quality, the factors affecting the quality indicators are used as input, and the experimental design method is adopted to obtain relevant sample points. After comparing and studying a variety of different experimental design methods, the fractional factorial experimental design is selected; compared with other experimental designs, this experimental design requires the least number of times and can effectively shorten the time required for computer simulation. After obtaining the sample points, the computer simulation method is used to obtain the response value of the sample points. The Spearman correlation coefficient is used to analyze the correlation between the quality indicators and the factors affecting the quality indicators. The combination of quality indicators and factors affecting the quality indicators with a correlation coefficient greater than 0.5 is retained to determine the final factors and quality indicators that mainly affect the quality indicators, further reducing the difficulty of model construction, while maintaining the accuracy of the model. Finally, based on the factors and quality indicators that mainly affect the quality indicators, the MES data collection is studied. The factors and quality indicators that mainly affect the quality indicators and equipment status parameters are collected through the MES system to realize the monitoring of process and equipment data.

[0029] Step 3: Establishment of reflow process quality prediction model

[0030] After obtaining the factors that mainly affect the quality index and the quality index, the Latin hypercube sampling method with good spatial projection is used to construct the sample space, and the obtained data is used as the modeling data. In order to compare the advantages and disadvantages of the common surrogate model algorithm and the Boosting-MKELM algorithm (the prediction framework diagram is shown in Figure 3 ), the prediction accuracy of these algorithms was compared, and the Boosting-MKELM algorithm with the highest prediction accuracy was selected to build the final reflow soldering process quality prediction model.

[0031] Step 4: Optimization of reflow process quality prediction model

[0032] Based on the above research on the reflow soldering process quality prediction model, the existing actual production data and historical prediction data are used to continuously train the reflow soldering process quality prediction model, which can ultimately help process personnel to select materials and quickly design process parameters, and continuously improve product quality.

[0033] Embodiment 2

[0034] This embodiment provides a technical solution: a reflow soldering process quality prediction method based on data mining, and the specific implementation points are as follows:

[0035] First point: simplify the research object of reflow process

[0036] There are many factors that affect the reflow process, which can be divided into two categories: qualitative factors and quantitative factors. The quality indicators are the size parameters and mechanical properties parameters of the solder paste after reflow. There are many products that use the reflow process, and the types of solder joints on the PCB of different products are also different. We can select several of the most representative types of solder joints and conduct research on them with different types of solder paste.

[0037] Second point: Selection of main influencing factors and quality index screening methods

[0038] In addition to adopting factorial experimental design to obtain sample points, you can also use uniform experimental design, orthogonal experimental design and other methods that require fewer experiments. After obtaining the response value through simulation, it is necessary to analyze the data set of influencing factors and quality indicators. Due to the small amount of data in the data set, it is not possible to use machine learning methods for feature screening. Correlation coefficient, information entropy, variance analysis, etc. are preferred.

[0039] The third point: Construction of reflow process quality prediction model

[0040] The data source of the reflow process quality prediction model is Latin hypercube sampling. The advantage of Latin hypercube is that the selected sample points have good dispersion in space, but it cannot handle qualitative factors (types of solder joints, types of solder paste materials), so it is necessary to combine qualitative factors, use Latin hypercube sampling to obtain sample points of quantitative factors, and then obtain response values ​​through simulation, establish multiple quality prediction models, and finally integrate them to establish the reflow process quality prediction model. In order to compare the advantages and disadvantages of common proxy model algorithms and Boosting-MKELM algorithms, it is only necessary to verify under a combination of qualitative factors, and select the Boosting-MKELM algorithm with the highest prediction accuracy as the best algorithm for the entire reflow process.

[0041] Fourth point: Continuous optimization of reflow process quality prediction model

[0042] The reflow soldering process quality prediction model has a very important guiding significance for actual production. In order to better improve and optimize the quality prediction model, the actual data in production and the historical prediction data of the corresponding quality prediction model are used as supplements to the model data source. The actual data in production comes from the data collected by the MES system, which needs to be preprocessed. With the continuous use of the prediction model, the amount of data is also increasing, and its prediction accuracy will continue to improve, eventually forming a continuous optimization of the reflow soldering process quality prediction model.

[0043] In summary, the reflow soldering process quality prediction method based on data mining in the above-mentioned embodiment preliminarily determines the quality indicators of the process and the factors affecting the quality indicators through qualitative analysis of the process and research on relevant literature, preliminarily obtains sample data through experimental design, analyzes and mines the sample data to obtain the correlation between the factors and product quality, and obtains the quality indicators and the main factors affecting the quality indicators; creates a sample space of main factors through experimental design, uses computer simulation technology to obtain the quality indicator response value of the sample space of main factors, and then compares the commonly used proxy model algorithm and the mainstream machine learning algorithm to establish a model, and adopts the more advanced Boosting-MKELM algorithm with more accurate calculation results to complete the prediction of the quality indicators of the reflow soldering process.

[0044] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.

Claims

1. A reflow soldering process quality prediction method based on data mining, characterized in that: The following steps are involved: S1: Preliminary determination of the factors affecting the quality of the reflow soldering process, quality indicators and their value ranges Through literature and process research, the factors and quality indicators that affect the quality indicators of the reflow soldering process are preliminarily analyzed to obtain the factors that affect the quality indicators and the upper and lower limits of the quality indicators; S2: Screening the main influencing factors and quality indicators of reflow process quality and data collection Select fractional factorial test design to obtain sample points, use computer simulation method to obtain the response value of sample points, use Spearman correlation coefficient to analyze the correlation between quality indicators and factors affecting quality indicators, retain the combination of quality indicators and factors affecting quality indicators with strong correlation, determine the final factors and quality indicators that mainly affect quality indicators, and collect data through MES system based on the factors and quality indicators that mainly affect quality indicators; In step S2, the correlation coefficient between the quality indicator with strong correlation and the combination of factors affecting the quality indicator is greater than 0.5; S3: Establish a reflow process quality prediction model The sample space is constructed by Latin hypercube sampling to obtain data and use it as modeling data. The non-continuous variables of reflow soldering are fixed, and multiple training models based on multiple algorithms are established and trained to obtain multiple algorithm models. The accuracy of the prediction results of multiple algorithm models is compared, and the algorithm model with the highest prediction accuracy is selected as the final reflow soldering process quality prediction model. S4: Optimizing the reflow process quality prediction model By using the existing actual production data and historical prediction data, the reflow soldering process quality prediction model is continuously trained to achieve the optimization of the reflow soldering process quality prediction model.

2. The method for predicting reflow process quality based on data mining according to claim 1, characterized in that: In step S3, the discontinuous variables of the reflow soldering include the type of solder joint and the type of solder paste material.

3. The method for predicting reflow soldering process quality based on data mining according to claim 1, characterized in that: In step S3, the algorithm includes a proxy model algorithm and a Boosting-MKELM algorithm.

4. The method for predicting reflow soldering process quality based on data mining according to claim 1, characterized in that: In step S4, the actual production data are the factors and quality indicators that mainly affect the quality indicators and the equipment status parameter data collected by the MES system.

Images