Optimized system and method for solder ball flattening in printed circuit board manufacturing process

Abstract

The optimized system and method for solder ball flattening in printed circuit board manufacturing processes disclosed herein can generate corresponding flattening condition parameters based on machine characteristics and material properties. These parameters, recommended by the optimized model, can be used in subsequent processes to reduce poor flattening yield. This disclosure not only reduces the need for small-scale testing during pilot production but also improves flattening yield, thus reducing rework.

Description

Technical Field

[0001] This invention relates to a process optimization system and method, and more particularly to an optimization system and method for flattening solder balls in a printed circuit board manufacturing process. Prior Technology

[0002] The back-end processes in printed circuit board (PCB) manufacturing mainly include solder resist application and solder ball or solder paste bumping, followed by flattening of the solder balls or solder paste. Before shipment, the finished products must be inspected to confirm yield. This yield is affected by variations in the size and thickness of the solder resist openings in each production batch. Therefore, before flattening the solder balls (or solder paste), a small sample must be taken for trial pressing and measurement. Flattening conditions are then determined empirically, occasionally resulting in poor flattening yields and even requiring rework for a second flattening cycle. Therefore, determining the optimal flattening production conditions by selecting key parameters before solder ball flattening is an important issue. Summary of the Invention

[0003] In view of this, this disclosure provides an optimization system and method for solder ball flattening in printed circuit board manufacturing processes, which uses data from past processes to model and provide flattening production conditions for solder ball flattening.

[0004] This disclosure provides an optimization system for solder ball flattening in a printed circuit board manufacturing process, including a memory and a processor. The memory stores multiple modules. The processor is coupled to the memory and configured to: acquire multiple data sets; preprocess the multiple data sets, wherein the preprocessing includes classifying the multiple data sets according to process equipment characteristics and process material characteristics; execute a parameter selection module among the multiple modules to acquire multiple parameters based on the multiple data sets; execute a machine learning module among the multiple modules to build an optimization model based on the multiple preprocessed data sets and the multiple parameters; and execute a prediction module among the multiple modules to perform reverse prediction testing on the optimization model.

[0005] This disclosure also provides an optimization method for flattening solder balls in a printed circuit board manufacturing process, comprising: acquiring multiple data; preprocessing the multiple data, wherein the preprocessing includes classifying the multiple data according to the characteristics of the process equipment and the characteristics of the process materials; acquiring multiple parameters based on the multiple data; establishing an optimization model based on the multiple preprocessed data and the multiple parameters; and performing reverse prediction testing on the optimization model.

[0006] Based on the above, the optimized system and method for solder ball flattening in printed circuit board manufacturing provided in this disclosure can generate corresponding flattening condition parameters according to machine characteristics and material properties. In subsequent processes, the flattening condition parameters recommended by the optimized model can be used to reduce the problem of poor flattening yield. Through the content of this disclosure, not only can the process of using small samples for trial pressing be reduced during pilot production, but the flattening yield can also be improved to reduce rework. Simple Explanation of the Diagram

[0007] Figure 1 is a schematic diagram of the optimized system for flattening solder balls in the printed circuit board manufacturing process disclosed herein. Figure 2 is a flowchart of the optimized system for flattening solder balls in the printed circuit board manufacturing process disclosed herein. Figure 3 is a schematic diagram of the printed circuit board manufacturing process. Figure 4 is a detailed process diagram of the optimized system for flattening solder balls in the printed circuit board manufacturing process disclosed herein. Implementation

[0008] The exemplary embodiments disclosed herein will now be described in detail with reference to the accompanying drawings. The terms "first," "second," etc., used throughout this specification (including the claims) are used to name elements or distinguish different embodiments or scopes, and are not intended to limit the upper or lower limit of the number of elements, nor to limit the order of elements. Furthermore, wherever possible, elements / components using the same reference numerals in the drawings and embodiments represent the same or similar parts.

[0009] Figure 1 is a schematic diagram of the optimized system for solder ball flattening in the printed circuit board manufacturing process disclosed herein. The optimized system 100 for solder ball flattening in the printed circuit board manufacturing process may include a processor 110 and a memory 120.

[0010] In the embodiments disclosed herein, the processor 110 is, for example, a central processing unit (CPU), or other programmable general-purpose or special-purpose microcontrollers (MCUs), microprocessors, digital signal processors (DSPs), programmable controllers, application-specific integrated circuits (ASICs), graphics processing units (GPUs), image signal processors (ISPs), image processing units (IPUs), arithmetic logic units (ALUs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), or other similar elements or combinations thereof. In the process parameter optimization system 200, the processor 110 may be coupled to memory 120, and the processor 110 may execute modules stored in memory 120.

[0011] Storage 120 may be any type of fixed or removable random access memory (RAM), read-only memory (ROM), flash memory, hard disk drive (HDD), solid-state drive (SSD), or similar elements or combinations thereof, for storing multiple models or various applications executable by processor 110. In this embodiment, storage 120 may store at least parameter selection module 121, machine learning module 122, and prediction module 123.

[0012] Please refer to Figure 2, which is a flowchart illustrating the optimization system for flattening solder balls in a printed circuit board manufacturing process disclosed herein. In process S210, processor 110 acquires multiple data points. In process S220, processor 110 preprocesses the multiple data points, including classifying them based on process equipment characteristics and process material characteristics. In process S230, processor 110 executes a parameter selection module among multiple modules to acquire multiple parameters based on the multiple data points. In process S240, processor 110 executes a machine learning module among multiple modules to build an optimization model based on the multiple preprocessed data points and the multiple parameters. In process S250, processor 110 executes a prediction module among multiple modules to perform reverse prediction testing on the optimization model.

[0013] Figure 3 is a schematic diagram of the printed circuit board (PCB) manufacturing process. This is a conventional PCB manufacturing process, including the PCB front-end process S310, the PCB back-end process (green paint station S320, solder balling station S330, flattening station S340, and finished product inspection S350). In the embodiments disclosed herein, process optimization is mainly performed on processes S320 and S330. In the embodiments disclosed herein, the processor 110 can first obtain multiple past process data for the aforementioned processes S320 and S330 for subsequent use in building an optimization model.

[0014] Please refer to Figure 4, which is a detailed flow diagram of the optimized system for flattening solder balls in the printed circuit board manufacturing process disclosed herein. In process S410, processor 110 can collect production data, such as obtaining multiple past process data from processes S320 and S330. In process S420, processor 110 can collect specification factor data of the production batch. Specifically, specification factor data, i.e., parameters, may include production and measurement parameters of the green paint process, production and measurement parameters of the solder ball (solder paste) process station, and specification parameters of different types of film materials used in the production batch.

[0015] Please refer to Figure 4. In process S430, processor 110 can remove outliers and misprinted data, and perform data preprocessing. Preprocessing may include at least classifying process data based on the characteristics of the manufacturing equipment and the characteristics of the manufacturing materials. For example, in the process of printed circuit board manufacturing, different manufacturing equipment may have different characteristics, and using different manufacturing equipment may result in different parameters that need to be emphasized in the process. Furthermore, the manufacturing equipment may have different parameters to emphasize during the process due to differences in the parts and the years of use. Alternatively, the different proportions and styles of various materials used in the process may also result in different parameters that need to be emphasized in the process. Therefore, processor 110 can classify the data based on different characteristics of manufacturing equipment and manufacturing materials, so that when using similar manufacturing equipment and similar manufacturing materials, the model established by the corresponding data can be used to select the parameters to be emphasized.

[0016] In the embodiments disclosed herein, key parameters affecting the process may include the open-loop size of the green paint, the flattening temperature, the flattening pressure, and the solder ball size. In other embodiments disclosed herein, key parameters affecting the process may be one or a combination of the aforementioned parameters. Referring again to Figure 4, in process S440, the processor 110 can identify the aforementioned key parameters affecting the process by using feature engineering to filter parameters.

[0017] In the embodiments disclosed herein, the processor 110 can perform cluster analysis on the process data. That is, the processor 110 can find a group of classes from the process data, and the processor 110 can determine the distance between each piece of process data and the group of classes. Furthermore, the processor 110 can remove outliers from the process data to avoid outliers affecting subsequent parameter selection and model building.

[0018] In other embodiments disclosed herein, the processor 110 can also directly examine the process data and determine whether to remove obviously misplaced data. For example, when the values ​​of the first parameter in all process data fall between 0 and 1, and the processor 110 can determine, according to conventional techniques in the art, that the first parameter will not be greater than 1, then the processor 110 can delete unreasonable process data where the first parameter is greater than 1, in order to avoid outliers affecting subsequent parameter selection and model building.

[0019] In other embodiments of this disclosure, the processor 110 may also use random forest feature importance or step-wise regression to obtain key parameters from the preprocessed process data. The selection parameters of process S440 in this disclosure are not limited to any feature engineering method; any algorithm that can obtain key parameters can be used as the selection parameter process of process S440 in this disclosure.

[0020] Please refer to Figure 4. In process S450, processor 110 can establish a machine learning algorithm model. Specifically, processor 110 can use ridge regression, least absolute shrinkage and selection operator (LASSO), or linear regression and derived algorithms to establish a linear machine learning algorithm optimization model. Alternatively, processor 110 can use decision tree, random forest, extreme gradient boosting (XGBoost), neural network, or gradient boosting regression to establish a nonlinear machine learning algorithm optimization model. The method of establishing the optimization model in process S450 of this disclosure is not limited to using any particular machine learning algorithm model; any algorithm that can be used to establish an optimization model can be used as the optimization model establishment process in process S450 of this disclosure.

[0021] Following the previous section, the evaluation metrics (e.g., coefficient of determination R²) of the optimized model obtained by processor 110 can influence the flow of the process. Specifically, if the coefficient of determination R² of the optimized model obtained by processor 110 is greater than a first evaluation threshold (e.g., greater than 0.7), processor 110 can continue to process S460, which involves testing the model's reverse prediction. If processor 110 determines that the coefficient of determination R² is between 0.4 and 0.7 (0.4 ≤ R² ≤ 0.7), processor 110 can determine that the selected key parameters are insufficient or need adjustment. Therefore, processor 110 can return to process S440 and use feature engineering different from that used previously to obtain key parameters. In this way, it can be understood that different key parameter compositions can train different optimized models, thus processor 110 can obtain new evaluation metrics for the optimized model.

[0022] Continuing from the previous step, if the processor 110 determines that the determination coefficient R² is less than the second evaluation threshold (e.g., less than 0.4), the processor 110 can return to process S410 to re-collect data or increase the amount of data. In addition, the processor 110 can re-evaluate whether the collected data conforms to the conventional standards of the process, such as whether the collected data includes at least the consistency of process equipment characteristics or material properties. This improves the evaluation metrics for subsequently building the optimization model.

[0023] Please refer to Figure 4. In process S460, processor 110 can perform reverse prediction testing on the model. Specifically, after the aforementioned processes S410 to S450, processor 110 needs to further confirm whether the established optimized model is suitable for subsequent practical application in the production process. In the embodiments disclosed herein, processor 110 can, for example, input process data not used to establish the optimized model into the obtained optimized model to determine the effectiveness of the optimized model. In process S470, assuming the aforementioned optimized model's effectiveness meets expectations, processor 110 can put the optimized model online for actual use on the production line. The aforementioned indicator of meeting expectations (prediction threshold) is, for example, an accuracy rate greater than or equal to 70%.

[0024] Following the previous section, if the processor 110 determines that the accuracy rate is less than 70%, the processor 110 will return the process to process S410 to collect data again, increase the amount of data, or re-evaluate whether the collected data meets the normal standards of the process.

[0025] Table 1 below lists case results using at least one of the processes shown in Figure 4 above in this disclosure. Table 1 Case process Case 1 Case 2 Case 3 Case 4 Comparative Example 1 Comparative Example 2 S410-S420 P P P P P P S430 P P P P P P S440 x x P P P P S450 linear Nonlinear Nonlinear Nonlinear Nonlinear Nonlinear Does the S450 use...? Hyperparameter optimization x x P P x P S460 x x x P x P Model metrics R 2 0.15-0.40 0.60-0.65 0.70-0.85 0.70-0.85 0.65-0.75 0.70-0.85 Error MAPE (%) 8-10 6-8 5-6 5-6 5.5-7.0 5-6 Reverse prediction Accuracy (%) Note - - - 85-90% - 60-85%

[0026] In Case 1, processor 110 collects production data and production specification factors for the green paint station and solder ball station, totaling approximately 20 parameters. These parameters include familiar green paint process production and measurement parameters, solder ball (solder paste) process station production and measurement parameters, and different specification parameters for the production batch (e.g., the type of film material used). The data consists of 5,000 records. After simple logical deletion of erroneous data, processor 110 does not use feature engineering to filter parameters in process S440. Furthermore, processor 110 uses a linear machine learning model in process S450 without using transcendental parameter optimization. Since the optimized model output from process S450 needs to be evaluated using evaluation metrics before entering process S460, and the R² in Case 1 is clearly not greater than 0.4, it cannot enter process S460.

[0027] In Case 2, processor 110 uses the same processes S410 to S430 as in Case 1, but processor 110 does not use the feature engineering screening parameters of process S440. In process S450, processor 110 uses a non-linear machine learning model and does not use transcendental parameter optimization. Since the optimized model produced by process S450 needs to be evaluated by evaluation metrics before entering process S460, and R² in Case 2 is obviously not greater than 0.7, it cannot enter process S460.

[0028] In Case 3, processor 110 uses the same processes S410 to S430 as in Case 1. Processor 110 uses feature engineering to select parameters in process S440, and employs a non-linear machine learning model with transcendental parameter optimization in process S450. As shown in Table 1, the R² of Case 3 is greater than 0.7. Although Case 3 did not enter the model reverse prediction test in process S460, at least the evaluation metric of the optimized model is greater than the threshold. Therefore, this optimized model can at least be used for the process deployment disclosed in this paper.

[0029] In Case 4, the only difference from Case 3 is that the processor 110 further executes the model reverse prediction test of process S460, and the processor 110 also sets upper and lower limits for the parameter range based on the characteristics of the process. As shown in Table 1, the accuracy of the model reverse prediction test in Case 4 is between 85% and 90%, which is greater than the expected 70%. Therefore, this optimized model can be used for the process deployment disclosed in this paper.

[0030] In Comparative Example 1, the process of Case 3 can be referenced. The difference between Comparative Example 1 and Case 3 is that Comparative Example 1 does not use hyperparameter optimization in process S450. As shown in Table 1, the R² of Comparative Example 1 is between 0.65 and 0.75, meaning that the optimized model produced by processor 110 according to the process of Comparative Example 1 is between usable and unusable. This also highlights the importance of using hyperparameter optimization in process S450.

[0031] In Comparative Example 2, the process of Case 4 can be referenced. The difference is that processor 110 does not set upper and lower limits for the reverse prediction flattening conditions in Comparative Example 2. As shown in Table 1, the accuracy of the reverse prediction test of the model in Comparative Example 2 is between 60% and 85%, which is between being usable online and not usable online. This also highlights the importance of setting upper and lower limits for the reverse prediction flattening conditions in process S460.

[0032] In the various embodiments disclosed herein, the definition of a correct reverse prediction test is: when the flattening yield is better than the flattening yield using the flattening conditions of the optimized model, the prediction is correct.

[0033] This disclosure also provides an optimized method for flattening solder balls in a printed circuit board manufacturing process, which can be implemented by the processor 110 of a printed circuit board process solder ball flattening optimization system 100. The various processes and detailed technical features of the optimization method have been described in the preceding paragraphs and will not be repeated here.

[0034] The optimization model generated by the optimization system and method for solder ball flattening in the printed circuit board manufacturing process disclosed herein can be applied to the flattening process in the printed circuit board manufacturing process, thereby improving the yield of solder ball or solder paste flattening.

[0035] In summary, the optimized system and method for solder ball flattening in printed circuit board manufacturing provided in this disclosure can generate corresponding flattening condition parameters based on machine characteristics and material properties. These flattening condition parameters, recommended by the optimized model, can be used in subsequent processes to reduce the problem of poor flattening yield. Through this disclosure, not only can the process of using small samples for trial flattening be reduced during pilot production, but the flattening yield can also be improved, thereby reducing rework.

[0036] 100: Optimization system for solder ball flattening in printed circuit board manufacturing process 110: Processor 120: Storage 121: Parameter Selection Module 122: Machine Learning Module 123: Prediction Module S210, S220, S230, S240, S250, S310, S320, S330, S340, S350, S410, S420, S430, S440, S450, S460, S470: Process

Claims

1. An optimized system for flattening solder balls in a printed circuit board manufacturing process, comprising: Storage, storing multiple modules; and a processor, coupled to the storage and configured to: acquire multiple data; The data is preprocessed, which includes classifying the data according to the characteristics of the process equipment and the process materials; executing a parameter selection module in the modules to obtain multiple parameters based on the data; executing a machine learning module in the modules to build an optimization model based on the preprocessed data and the parameters; and executing a prediction module in the modules to perform reverse prediction testing on the optimization model.

2. The optimized system for flattening solder balls in a printed circuit board manufacturing process as described in claim 1, wherein the pretreatment further includes: The processor is used to perform cluster analysis on this data; And remove multiple outliers from the cluster analysis data.

3. An optimization system for flattening solder balls in a printed circuit board manufacturing process as described in claim 1, wherein the processor is further configured to: use random forest feature importance or step-wise regression to obtain the parameters based on the data.

4. An optimization system for flattening solder balls in a printed circuit board manufacturing process as described in claim 1, wherein the processor is further configured to: use ridge regression, least absolute shrinkage and selection operator (LASSO), or linear regression to establish a linear optimization model.

5. The optimization system for flattening solder balls in a printed circuit board manufacturing process as described in claim 1, wherein the processor is further configured to: use a decision tree, random forest, extreme gradient boosting (XGBoost), neural network, or gradient boosting regression to build a nonlinear optimization model.

6. The optimization system for solder ball flattening in a printed circuit board manufacturing process as claimed in claim 1, wherein the processor is further configured to: execute an evaluation module among the modules to evaluate the optimization model and obtain an evaluation index; execute the reverse prediction test in response to the evaluation index being greater than a first evaluation threshold; execute a second parameter selection module among the modules to obtain a plurality of second parameters in response to the evaluation index being not greater than the first evaluation threshold and the evaluation index being not less than a second evaluation threshold; and increase the amount of data in response to the evaluation index being less than the second evaluation threshold.

7. The optimized system for flattening solder balls in a printed circuit board manufacturing process as described in claim 6, wherein the processor is further configured to: in response to obtaining the second parameters, execute a machine learning module in the modules to build a second optimization model based on the preprocessed data and the second parameters.

8. The optimized system for flattening solder balls in a printed circuit board manufacturing process as described in claim 6, wherein the processor is further configured to: perform the preprocessing on the incremental data in response to an increase in the amount of data.

9. An optimization system for flattening solder balls in a printed circuit board manufacturing process as described in claim 1, wherein the processor is further configured to: increase the amount of data in response to the accuracy of the reverse prediction test being less than a prediction threshold.

10. An optimization system for flattening solder balls in a printed circuit board manufacturing process as described in claim 1, wherein the parameters include: The size of the green paint ring, the flattening temperature, the flattening pressure, and the size of the solder ball.

11. An optimized method for flattening solder balls in a printed circuit board manufacturing process, comprising: Acquire multiple data sets; perform preprocessing on these data sets, including classifying the data sets according to process equipment characteristics and process material characteristics; obtain multiple parameters based on these data sets; establish an optimization model based on the preprocessed data sets and the parameters; and perform reverse prediction testing on the optimization model.

12. The optimized method for flattening solder balls in a printed circuit board manufacturing process as described in claim 11, wherein the pretreatment further includes: Perform cluster analysis on this data; And remove multiple outliers from the cluster analysis data.

13. The optimized method for flattening solder balls in a printed circuit board manufacturing process as described in claim 11 further includes: Use random forest feature importance or step-wise regression to obtain these parameters from the data.

14. The optimized method for flattening solder balls in a printed circuit board manufacturing process as described in claim 11 further includes: Use ridge regression, least absolute shrinkage and selection operator (LASSO), or linear regression to build a linear optimization model.

15. The optimized method for flattening solder balls in a printed circuit board manufacturing process as described in claim 11 further includes: Use decision trees, random forests, extreme gradient boosting (XGBoost), neural networks, or gradient boosting regression to build this nonlinear optimization model.

16. The optimized method for flattening solder balls in a printed circuit board manufacturing process as described in claim 11 further includes: The optimization model was evaluated, and an evaluation index was obtained. In response to the evaluation index being greater than the first evaluation threshold, the reverse prediction test is performed; in response to the evaluation index being no greater than the first evaluation threshold and no less than the second evaluation threshold, multiple second parameters are obtained based on these data; and in response to the evaluation index being less than the second evaluation threshold, the amount of these data is increased.

17. The optimized method for flattening solder balls in a printed circuit board manufacturing process as described in claim 16 further includes: In response to obtaining these second parameters, a second optimization model is established based on the preprocessed data and these second parameters.

18. The optimized method for flattening solder balls in a printed circuit board manufacturing process as described in claim 16 further includes: In response to the increase in the amount of data, the preprocessing is performed on the incremental data.

19. The optimized method for flattening solder balls in a printed circuit board manufacturing process as described in claim 11 further includes: In response to the fact that the accuracy of the reverse prediction test is less than the prediction threshold, the amount of such data is increased.

20. An optimized method for flattening solder balls in a printed circuit board manufacturing process as described in claim 11, wherein the parameters include: The size of the green paint ring, the flattening temperature, the flattening pressure, and the size of the solder ball.

Images