Welding spot simulation life prediction method and system based on random forest

Abstract

The invention discloses a welding spot simulation life prediction method and system based on a random forest, and belongs to the technical field of welding spot simulation prediction, and the method comprises the steps: S1, carrying out the simulation of the thermal fatigue life of welding spots of various types of BGA chip products through CAE simulation software, collecting simulation parameters and simulation results, and constructing a CAE historical simulation database; s2, for each BGA chip category, collecting important structure parameters, material parameters and welding spot simulation life values from a CAE historical simulation database to form a data set; s3, based on the collected data set, using a random forest algorithm to construct a welding spot life prediction model; and S4, performing welding spot simulation life prediction by using the constructed welding spot life prediction model. According to the scheme, traditional CAE software simulation and machine learning are combined, so that the prediction efficiency is greatly improved while the accuracy of the prediction result of the simulation life of the welding spot is ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of solder joint simulation prediction, and in particular to a solder joint simulation life prediction method and system based on random forest. Background Art

[0002] In the current electronic product development and manufacturing process, BGA (Ball Grid Array) chips on circuit boards are mechanically and electrically interconnected through solder joints. In a system consisting of BGA chips, PCB (Printed Circuit Board) and solder joints, the thermal fatigue life of solder joints deserves attention. Compared with actual reliability tests, CAE (Computer-Aided Engineering) simulation is a faster way to obtain the thermal fatigue life of solder joints.

[0003] Finite element analysis is a commonly used CAE simulation technology. By establishing a finite element model, the manufacturing process of electronic devices and various temperature / humidity working environments can be simulated and analyzed. In the field of microelectronic packaging, the finite element software ANSYS is most widely used in thermal structure coupling analysis. Through finite element analysis, reliability data of new products or new designs can be obtained, avoiding the trouble caused by experiments, saving development costs, shortening product development cycles, and not only gaining a deep understanding of the failure mechanism, but also optimizing the entire process through parameter analysis.

[0004] However, with rapid product iterations, the structural parameters and material parameters of electronic products sometimes only change partially. Since CAE simulation involves a complex modeling process, including structural, thermal and other aspects, it takes a lot of time to re-model or modify the model every time and then perform solder joint thermal fatigue life simulation. Summary of the invention

[0005] The technical problem to be solved by the present invention is: in view of the low efficiency of CAE solder joint thermal fatigue life simulation, a random forest-based solder joint simulation life prediction method and system are provided to quickly predict the solder joint simulation life based on CAE historical simulation data.

[0006] To achieve the above object, the present invention provides a method for predicting solder joint life simulation based on random forest, comprising the following steps:

[0007] S1, use CAE simulation software to simulate the thermal fatigue life of solder joints of various types of BGA chip products, collect simulation parameters and simulation results to build a CAE historical simulation database;

[0008] S2, for each BGA chip category, collect important structural parameters, material parameters and solder joint simulation life from the CAE historical simulation database to form a data set, the important structural parameters and material parameters include: chip Die length, chip Die width, device length, device width, device CTE, substrate CTE, PCB CTE;

[0009] S3, based on the collected data set, a solder joint life prediction model is constructed using the random forest algorithm;

[0010] S4, using the constructed solder joint life prediction model, perform solder joint simulation life prediction.

[0011] The present invention also provides a solder joint simulation life prediction system based on random forest, comprising:

[0012] CAE historical simulation database construction unit: used to collect, query and display CAE historical simulation data and build a CAE historical simulation database;

[0013] Data acquisition unit: used to collect important structural parameters, material parameters and solder joint simulation life data sets from the historical simulation database, wherein the important structural parameters and material parameters include: chip Die length, chip Die width, device length, device width, device CTE, substrate CTE, PCB CTE;

[0014] Model training unit: used to train the random forest prediction model using the random forest algorithm based on the data set collected by the data collection unit;

[0015] Solder joint life prediction unit: used to predict the solder joint simulation life based on the input structural parameters, material parameters of the BGA chip product and the solder joint life prediction model trained by the model training unit.

[0016] The technical solution of the present invention is adopted, and CAE simulation software is used to simulate the thermal fatigue life of solder joints of various BGA chip products, and a CAE historical simulation database is constructed; for different product types, the key parameters are selected from the structural parameters and material parameters of the solder joints according to the experimental experience of engineers, and the corresponding CAE software historical simulation data are collected to form a data set; a random forest prediction model for solder joint life is constructed based on the collected data set; and the solder joint life prediction model is used to realize rapid simulation life prediction. The present invention combines traditional CAE software simulation with machine learning, so that the prediction of the simulated life of the solder joint can ensure the accuracy of the prediction results while greatly improving the prediction efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0018] Figure 1 A prediction method flow chart is provided for an embodiment of a random forest-based solder joint simulation life prediction method and system of the present invention.

[0019] Figure 2 A prediction system block diagram is provided for an embodiment of a random forest-based solder joint simulation life prediction method and system of the present invention.

[0020] Figure 3 A processing flow chart of a solder joint life prediction unit provided in an embodiment of a solder joint simulation life prediction method and system based on random forest of the present invention. DETAILED DESCRIPTION

[0021] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0022] The general idea of ​​the present invention is: use CAE simulation software to simulate the thermal fatigue life of solder joints of various categories of BGA chip products, and build a CAE historical simulation database; for different product types, select key parameters from the structural parameters and material parameters of the solder joints, and collect the corresponding CAE software historical simulation data to form a data set; build a random forest prediction model for solder joint life based on the collected data set; use the solder joint life prediction model to achieve rapid simulation life prediction.

[0023] The embodiments of the present invention are further described in detail below in conjunction with the accompanying drawings of the specification. It should be understood that the embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0024] The present invention is applicable to the simulation life prediction of BGA chip solder joints in electronic products.

[0025] Figure 1 FIG. 1 is a flow chart of a method for predicting solder joint life simulation based on random forests provided by an embodiment of the present invention. Figure 1 As shown, the method comprises the following steps:

[0026] S1A, classify BGA chip products based on BGA package type and number of BGA chip stacking layers.

[0027] BGA packages are divided into PBGA (Plastic BGA), CBGA (Ceramic BGA), TBGA (Tape BGA), and other types. The number of stacked grains in different BGA chips may also be different. The BGA package type and the number of stacked grains have an important impact on the simulated life of the solder joints. Therefore, in an embodiment of the present invention, the BGA chips are first classified according to the BGA type and the number of BGA chip stacks, and then a machine learning training prediction model is used separately for each BGA chip product type. In practical applications, classification can be performed as needed. If the product is relatively simple, this step can be omitted. If there are many types of products, it is necessary to increase the classification dimension.

[0028] S1, use CAE simulation software to simulate the thermal fatigue life of solder joints of various types of BGA chip products, collect simulation parameters and simulation results to build a CAE historical simulation database.

[0029] In the embodiment of the present invention, ANSYS simulation software is used to simulate the thermal fatigue life of the solder joints of the BGA chip product. The simulation is performed on a device basis. For the same device, the structure, quantity, and layout of the solder joints are relatively fixed. The specific simulation process is as follows:

[0030] (1) Model the system consisting of chips, solder joints, and PCB in ANSYS.

[0031] (2) Measure the structural parameters and material parameters of each component of the system, including grain size, chip size, and CTE (Coefficient of Thermal Expansion) of substrates, devices, and PCB boards.

[0032] (3) Perform finite element simulation based on the modeling results and various structural parameters and material parameters, and output the simulation results, which include the simulated life of the solder joints.

[0033] (4) After the simulation is completed, the structural parameters, material parameters and simulation results are entered into the CAE historical simulation database.

[0034] Using ANSYS to simulate the thermal fatigue life of BGA chip product solder joints belongs to the prior art, and the specific details are not repeated here. In order to ensure the accuracy of the prediction, each type of BGA chip product needs to have sufficient CAE simulation data samples. During the implementation of the embodiment of the present invention, it was found that for the same type of BGA chip products, more than 1,000 data are used as the base to start model training.

[0035] S2, for each BGA chip category, important structural parameters, material parameters and solder joint simulation life values ​​are collected from the CAE historical simulation database to form a data set.

[0036] The simulated life of a solder joint refers to the number of cycles of a solder joint under the temperature and humidity life acceleration test, measured in times. According to the experimental experience of engineers, for the same product type, the key structural parameters and material parameters screened include: chip Die length, chip Die width, device length, device width, device CTE, substrate CTE, and PCB CTE. Among them, chip Die refers to the chip bare die, also known as bare chip, bare die, and crystal grain. It is a small piece of integrated circuit body made of semiconductor materials and not packaged. The device refers to the IC device formed after the chip Die and the substrate are packaged.

[0037] In the embodiment of the present invention, the simulation life of the solder joints of the memory chip is predicted. The structure, quantity and layout of the solder joints of the memory chip are relatively fixed. The structural parameters and material parameters that change frequently and have an important impact on the life of the solder joints include: chip Die length, chip Die width, device length, device width, device CTE, substrate CTE, PCB CTE. Only relying on these 7 parameters can make a more accurate prediction of the simulation life of the solder joints. In actual implementation, the collected parameters can be adjusted in combination with the division dimension of the product type in step S1 and the actual situation of the BGA chip product.

[0038] Assume that the number of collected data is k, each data contains m parameters, the parameter data set is x, and the solder joint simulation life data set is y, then:

[0039]

[0040]

[0041] In the embodiment of the present invention, part of the collected data is shown in Table 1 below, wherein the first 7 columns correspond to the 7 parameters of chip Die length, chip Die width, device length, device width, device CTE, substrate CTE, and PCB CTE, respectively, and the last column corresponds to the simulated life obtained by CAE simulation, and each row of data corresponds to a simulation process of a BGA packaged device.

[0042] Table 1 Part of the historical simulation data collected by the embodiment of the present invention

[0043]

[0044]

[0045] S3, based on the collected data set, a random forest algorithm is used to build a solder joint life prediction model.

[0046] The random forest algorithm is an integrated learning method based on self-help aggregation. In the random forest, each decision tree is an estimator. Part of the data set is randomly selected from the input data set to form the input of each decision tree. Multiple decision trees are operated in parallel, and the outputs of all decision trees are combined to obtain the prediction results. Therefore, the random forest algorithm has the advantages of high accuracy, fast operation speed, and good support for high-dimensional data. It is very suitable for the scenario of the present invention that requires the combination of multiple dimensions of parameters for fast and accurate prediction.

[0047] In the embodiment of the present invention, the platform used to build the prediction model is Anaconda, the programming language used is Python, and the Python libraries involved mainly include sklearn, pandas, and numpy. The specific steps of step S3 are as follows:

[0048] S3.1, deduplicate the data in the dataset and divide the dataset into a training set and a test set.

[0049] In the embodiment of the present invention, the train_test_split method of the sklearn library is used to split the data set into a training set and a test set, and the specific split ratio is determined according to the actual situation. The calling method of the train_test_split method is as follows:

[0050] x_train,x_test,y_train,y_test=train_test_split(x,y,test_size=0.2,random_state=42)

[0051] The parameter data set x is divided into a training set x_train and a test set x_test, and the simulation life data set y is divided into a training set y_train and a test set y_test.

[0052] S3.2, train the prediction model using the random forest algorithm based on the training set.

[0053] In the embodiment of the present invention, the prediction model is trained using the random forest regression class RandomForestRegressor of the sklearn library, and the training set feature data x_train and the simulated life training set y_train are respectively passed into the random forest algorithm as variables and target values ​​to train the prediction model. The key parameters used include: the number of decision trees is 30, the maximum depth is 10, and the minimum number of leaf node samples is 10.

[0054] S3.3, pass the test set feature x_test into the prediction model, get the prediction result y_test, and then calculate the mean absolute error MAE to get the accuracy of the model. The MAE calculation formula is as follows:

[0055]

[0056] where y i Represents the simulated life of the solder joint, y′ i It represents the predicted life of solder joints by simulation, and n represents the number of solder joints.

[0057] In the embodiment of the present invention, a total of 1676 CAE historical simulation data were collected, of which 48 data were used as test sets and the rest were used as training sets. The results and prediction errors of the test set predicted by the model trained with the training set are shown in Table 2 below, and the average prediction error is 7.6%. For CAE simulation, an error within 30% is basically acceptable. Therefore, the error of the embodiment of the present invention fully meets the error range requirements.

[0058] In Table 2, the first 7 columns correspond to the 7 parameters of chip die length, chip die width, device length, device width, device CTE, substrate CTE, and PCB CTE, respectively. The 8th column corresponds to the simulated life obtained by CAE simulation, the 9th column is the predicted life predicted by the training model, and the 10th column is the error value of the predicted life relative to the CAE simulated life. The average error is calculated using the MAE calculation formula in step S3.3. The test set data in Table 2 is the data sample provided in Table 1.

[0059] Table 2 Comparison table of test set data and test results of the embodiment of the present invention

[0060]

[0061]

[0062] If the accuracy of the trained model does not meet the requirements, it is necessary to analyze it based on the actual situation and consider improving it from the following aspects:

[0063] (1) Improve the accuracy of CAE simulation results: Review the CAE simulation process to confirm whether there are any errors in the modeling, parameter measurement, and simulation process, and improve the accuracy of the simulation results.

[0064] (2) Expand the sample size in the CAE historical simulation database.

[0065] (3) Expand the parameters collected from the CAE historical simulation database.

[0066] (4) Adjust the ratio of training set and test set.

[0067] (5) Adjust the random forest algorithm parameters.

[0068] In the early stage of model training, the full amount of historical simulation data can be obtained from the CAE historical simulation database on a daily basis to retrain and obtain the latest prediction model.

[0069] S4, using the constructed solder joint life prediction model, perform solder joint simulation life prediction.

[0070] In the embodiment of the present invention, after the accuracy of the prediction model meets the requirements, the model is solidified and published using a web service. Then, a rapid prediction WEB service for the simulated life of a solder joint based on a B / S architecture is developed, and the solder joint life prediction model is called to realize rapid simulation life prediction.

[0071] Using the solder joint life prediction service of the embodiment of the present invention, the solder joint life prediction can be completed within one minute, while using ANSYS to perform solder joint life prediction requires half a day to run the simulation results. If the model building step is included, it will take several days. Therefore, the technical solution of the embodiment of the present invention can greatly improve the efficiency of solder joint life prediction.

[0072] like Figure 2 As shown, an embodiment of the present invention further proposes a solder joint simulation life prediction system based on random forest, the system comprising:

[0073] Pre-classification unit: used to classify BGA chip products and provide the classification results to the CAE historical simulation database construction unit. The classification basis includes BGA package type and BGA chip stacking number.

[0074] CAE historical simulation database construction unit: used to collect, query and display CAE historical simulation data and build CAE historical simulation database.

[0075] Data acquisition unit: used to collect important structural parameters, material parameters and solder joint simulation life data sets from the CAE historical simulation database.

[0076] Model training unit: used to train the random forest solder joint life prediction model using the random forest algorithm based on the data set collected by the data collection unit.

[0077] Solder joint life prediction unit: used to predict the solder joint simulation life based on the input structural parameters, material parameters of the BGA chip product and the solder joint life prediction model trained by the model training unit.

[0078] In an embodiment of the present invention, a WEB service based on a B / S architecture is developed for collecting, querying and displaying CAE historical simulation data. After each CAE simulation is completed, simulation parameters and simulation results are entered through a WEB page and saved in a database. The database serves as a CAE historical simulation database and can continuously accumulate simulation data in practice.

[0079] In the embodiment of the present invention, the important structural parameters and material parameters collected by the data collection unit include chip Die length, chip Die width, device length, device width, device equivalent CTE, substrate CTE, and PCB CTE.

[0080] In an embodiment of the present invention, the model training unit includes: a device for performing deduplication processing on data in a data set and dividing the data set into a training set and a test set; a device for training a prediction model using a random forest algorithm based on the training set; and a device for passing the test set features into the prediction model to obtain a prediction result, and then calculating the mean absolute error MAE to obtain the accuracy of the model, wherein the MAE calculation formula is as follows:

[0081]

[0082] where y i Represents the simulated life of the solder joint, y′ i It represents the predicted life of solder joints by simulation, and n represents the number of solder joints.

[0083] In the embodiment of the present invention, the model training unit is developed based on the sklearn library of python, and the random forest regression class RandomForestRegressor of the sklearn library is used to train the prediction model.

[0084] In the embodiment of the present invention, the solder joint life prediction unit is built using the Django framework, including a front-end simulation parameter input interface, a back-end data preprocessing and model prediction service. Figure 3 The solder joint life prediction unit process is shown below:

[0085] (1) The user logs in to the rapid welding thermal fatigue simulation prediction page, enters the key parameters of the BGA chip simulation, clicks Next, and transfers the data to the backend.

[0086] (2) The backend judges the rationality of the data. If it is not rational, it will directly return an error to the frontend and prompt the user to re-enter the data. If the data is rational, it will be passed to the latest prediction model. The model obtains the prediction results and returns them to the frontend through the interface. The frontend displays the prediction results on the page. Rationality mainly includes the basic legality and logical judgment of the parameters. For example, the length and width of the device must be greater than the length and width of the chip Die.

[0087] The technical solution of the present invention is adopted, and CAE simulation software is used to simulate the thermal fatigue life of solder joints of various BGA chip products, and a CAE historical simulation database is constructed; for different product types, the key parameters are selected from the structural parameters and material parameters of the solder joints according to the experimental experience of engineers, and the corresponding CAE software historical simulation data are collected to form a data set; a random forest prediction model for solder joint life is constructed based on the collected data set; and the solder joint life prediction model is used to realize rapid simulation life prediction. The present invention combines traditional CAE software simulation with machine learning, so that the prediction of the simulated life of the solder joint can ensure the accuracy of the prediction results while greatly improving the prediction efficiency.

[0088] The above are only specific implementations of the present invention, which cannot be used to limit the scope of the present invention. Equivalent changes made by ordinary technicians in this technical field based on this creation, as well as changes known to technicians in this field, should still fall within the scope of the present invention.

Claims

1. A solder joint simulation life prediction method based on random forest, characterized in that: The following steps are involved: S1, use CAE simulation software to simulate the thermal fatigue life of solder joints of various types of BGA chip products, collect simulation parameters and simulation results to build a CAE historical simulation database; S2, for each BGA chip category, collect important structural parameters, material parameters and solder joint simulation life values ​​from the CAE historical simulation database to form a data set, wherein the important structural parameters and material parameters include: chip Die length, chip Die width, device length, device width, device CTE, substrate CTE, and PCB CTE; S3, based on the collected data set, a random forest algorithm is used to build a solder joint life prediction model; S4, using the constructed solder joint life prediction model, perform solder joint simulation life prediction.

2. The method for predicting solder joint life simulation based on random forest according to claim 1, characterized in that: Before step S1, the following steps are also included: S1A, classify BGA chip products based on BGA package type and number of BGA chip stacking layers.

3. The method for predicting solder joint life simulation based on random forest according to claim 1, characterized in that: The step S3 specifically includes: S3.1, deduplicate the data in the dataset and divide the dataset into a training set and a test set; S3.2, train the prediction model using the random forest algorithm based on the training set; S3.3, pass the test set features into the prediction model to obtain the prediction results, and then calculate the mean absolute error MAE to obtain the accuracy of the model. The MAE calculation formula is as follows: where y i Represents the simulated life of the solder joint, y′ i It represents the predicted life of solder joints by simulation, and n represents the number of solder joints.

4. The method for predicting solder joint life simulation based on random forest according to claim 3 is characterized in that: In step S3.2, the key parameters used by the random forest algorithm include: number of decision trees 30, maximum depth 10, and minimum number of leaf node samples 10.

5. The method for predicting solder joint life simulation based on random forest according to claim 3, characterized in that: In step S3.3, if the accuracy of the model does not meet expectations, improve it from the following aspects: (1) Improve the accuracy of CAE simulation results; (2) Expand the sample size in the CAE historical simulation database; (3) Expand the parameters collected from the CAE historical simulation database; (4) Adjust the ratio of training set and test set; (5) Adjust the random forest algorithm parameters.

6. A solder joint simulation life prediction system based on random forest, characterized in that: include: CAE historical simulation database construction unit: used to collect, query and display CAE historical simulation data and build a CAE historical simulation database; Data acquisition unit: used to collect important structural parameters, material parameters and solder joint simulation life data sets from the CAE historical simulation database, wherein the important structural parameters and material parameters include: chip Die length, chip Die width, device length, device width, device CTE, substrate CTE, PCB CTE; Model training unit: used to train a random forest solder joint life prediction model using a random forest algorithm based on the data set collected by the data collection unit; Solder joint life prediction unit: used to predict the solder joint simulation life based on the input structural parameters, material parameters of the BGA chip product and the solder joint life prediction model trained by the model training unit.

7. The solder joint simulation life prediction system according to claim 6, characterized in that: It also includes a pre-classification unit for classifying BGA chip products and providing classification results to the CAE historical simulation database construction unit, and the classification basis includes BGA package type and BGA chip stacking number.

8. The solder joint simulation life prediction system according to claim 6, characterized in that: The model training unit includes: a device for performing deduplication processing on the data in the data set and dividing the data set into a training set and a test set; a device for training the prediction model using a random forest algorithm based on the training set; and a device for passing the test set features into the prediction model to obtain the prediction results, and then calculating the mean absolute error (MAE) to obtain the accuracy of the model, wherein the MAE calculation formula is as follows: where y i Represents the simulated life of the solder joint, y′ i It represents the predicted life of solder joints by simulation, and n represents the number of solder joints.

Images