Embedded substrate welding spot bending-torsion coupling stress optimization method based on CS-BPNN
By combining the Cuckoo optimization algorithm with the BP neural network, the problem of insufficient accuracy of embedded active devices under bending and torsional coupling loads was solved, the structural parameters of the solder joint were optimized, and the reliability and durability of the solder joint were improved.
Patent Information
- Application Number
- CN202510867295.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-26
- Publication Date
- 2025-10-28
AI Technical Summary
Existing technologies cannot accurately reflect actual working conditions when studying bending and torsional coupled loads on embedded active devices. Furthermore, traditional optimization methods lack accuracy in multi-physics coupled scenarios, and particle swarm optimization algorithms are prone to getting trapped in local optima, leading to accelerated fatigue failure of solder joints.
By combining the Cuckoo Optimization Algorithm with the BP Neural Network, and through the establishment of a finite element model, orthogonal experiments, grey relational analysis, and optimization of the neural network, the structural parameters of the weld point are optimized, thereby achieving accurate prediction and minimization of bending-torsional coupling stress.
It effectively optimized the bending and torsional coupling stress of the embedded substrate solder joints, improved the reliability and durability of the solder joints, reduced fatigue failure, and enhanced the comparison and verification effect between simulation and experimental results.
Smart Images
Figure CN120850653A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of electronic packaging reliability technology, and in particular to a stress optimization method for embedded substrate CSP solder joints under bending and torsional coupling loads based on the fusion of Cuckoo Search (CS) algorithm and BP neural network. Background Art
[0002] Embedded active devices are subjected to both bending and torsion during service, leading to complex stress concentrations at the solder joints. Traditional single-load studies (such as three-point bending) cannot reflect actual operating conditions, while bending-torsional coupling significantly accelerates solder joint fatigue failure. Existing optimization methods (such as response surface methodology) lack accuracy in multi-physics coupled scenarios, and particle swarm optimization is prone to getting trapped in local optima. The Cuckoo Search (CS) algorithm enhances global search capabilities through the Lévy flight mechanism and, combined with a backpropagation neural network, can accurately capture the nonlinear mapping relationship between solder joint geometric parameters and bending-torsional coupling stress. Summary of the Invention
[0003] This invention provides a method for optimizing the bending-torsional coupling stress of solder joints in embedded substrates based on CS-BPNN, characterized by comprising:
[0004] 1): A three-dimensional finite element model of the embedded substrate solder joints was established based on ANSYS, including the embedded chip, PCB substrate, solder joint array and resin filler.
[0005] 2): The orthogonal experimental design method was used to design the parameter combination of solder joint diameter, solder pad diameter and solder joint height, and the maximum bending-torsional coupling stress of the solder joint under each combination was obtained by finite element simulation.
[0006] 3) Fabricate an experimental prototype identical to the finite element model. First, mill a groove slightly larger than the embedded chip on the PCB substrate. Fabricate pads in the groove, solder the embedded chip onto the pads, attach right-angle strain rosettes to the embedded chip, and finally fill with resin filler. Conduct bending-torsional coupling strain measurement experiments and compare the results with simulation results to verify the effectiveness.
[0007] 4): The influence of each structural parameter on stress was ranked by magnitude through grey relational analysis;
[0008] 5): Construct a BP neural network optimized by the cuckoo search algorithm. The parameters of the cuckoo algorithm are set as follows: population size 30, discovery probability 0.25, maximum number of iterations 50, Levy flight step size a = 0.01, input layer is the weld joint structure parameters, and output layer is the predicted stress value.
[0009] 6): With the goal of minimizing weld stress, the optimal combination of structural parameters is solved using an optimized neural network;
[0010] 7): Verify the optimization effect through finite element simulation.
[0011] Optionally, the solder joint diameter is 0.26 mm, the solder joint height is 0.16 mm, the pad diameter is 0.21 mm, and the solder joint material is SAC305.
[0012] Optionally, when comparing and verifying the experimental results with the simulation results, two indicators, the first principal strain and the third principal strain, can be used. Attached Figure Description
[0013] Figure 1 A finite element model of the embedded substrate CSP solder joint is provided as an example of an embodiment of the present invention.
[0014] Figure 2 A bending-torsional coupled strain measurement platform provided as an example of an embodiment of the present invention.
[0015] Figure 3 A cloud map showing the stress distribution of the weld joint under bending and torsional coupling, provided as an example of an embodiment of the present invention.
[0016] Figure 4 The CS-BPNN regression analysis results are shown in the figure provided as an example of an implementation of the present invention. Detailed Implementation
[0017] The following describes in detail the implementation method of the present invention for optimizing the bending-torsional coupling stress of embedded substrate solder joints based on CS-BPNN, with reference to the accompanying drawings.
[0018] The APDL module of ANSYS finite element analysis software was used to perform stress-strain finite element simulation analysis on the CSP solder joint of an embedded substrate under coupled bending and torsional loads. The established finite element model of the embedded substrate solder joint is shown below. Figure 1 As shown, the model consists of four parts: nine WLCSP chips, a PCB substrate, solder joints, and resin filler.
[0019] To fabricate an experimental prototype identical to the finite element model, a groove slightly larger than the embedded chip was first milled into the PCB substrate. Pads were then fabricated within the groove, and the embedded chip was soldered onto these pads. Right-angle strain rosettes were then attached to the embedded chip, and finally, resin filler was used to fill the groove. A bending-torsional coupled strain measurement platform was then constructed, such as… Figure 2 As shown, the bending-torsional coupling strain measurement platform consists of four parts: a computer, a bending-torsional coupling fixture, an experimental sample, and a dynamic strain measuring instrument.
[0020] A 1mm bending-torsional coupled displacement load was applied to the four corners of the PCB substrate, and a 2mm bending displacement load was applied to the center. Finite element simulation analysis was then performed, and the stress cloud diagram of the solder joints was read. Figure 3 As shown.
[0021] Using three structural parameters—solder joint diameter, solder pad diameter, and solder joint height—as factors, and the solder joint bending-torsional coupling stress as the dependent variable, each factor is assigned four levels, and L is used. 16 (4 3 The structural parameter level combination design was carried out using an orthogonal array, resulting in 16 factor level combinations. Based on these 16 weld point structural parameter levels in the orthogonal array, 16 corresponding finite element simulation models were established and bending-torsional coupled loads were applied. The maximum weld point stress results were read, and the stress results are shown in Table 1.
[0022] Table 4 Results of the orthogonal experiment
[0023]
[0024] With a confidence level of 95%, an analysis of variance was performed based on the results of the bending-torsional coupling stress of the weld joints in Table 1.
[0025] The 16 factor level combinations in Table 1 were used for mean value processing to eliminate the influence of different dimensions and orders of magnitude of each factor. Subsequently, the gray relation values between the bending and torsional coupling stress of the solder joint and the solder joint diameter, solder pad diameter, and solder joint height were solved, and the gray relation degree values were solved. The results are shown in Table 2.
[0026] Table 2. Association Results
[0027]
[0028] The significance analysis in Table 1 shows that the solder joint diameter, pad diameter, and solder joint height all have a significant impact on the bending-torsional coupling stress of the solder joints in the embedded substrate. Therefore, these three structural parameters were selected as the input to the BP neural network, and the bending-torsional coupling stress of the solder joints was used as the output. Sixteen orthogonal experimental groups from Table 1 were selected as prediction samples, and the sample size was expanded to 50 groups. The first 40 groups were selected as the training group, and the last 10 groups as the test group. The torsional stress of the solder joints obtained after finite element simulation of these 50 groups of solder joints with different structural parameter levels is shown in Table 3.
[0029] Table 3 CS-BP Neural Network Prediction Samples
[0030]
[0031]
[0032] A backpropagation neural network (BP neural network) optimized for the cuckoo search algorithm was constructed. The cuckoo algorithm parameters were set as follows: population size 30, discovery probability 0.25, maximum number of iterations 50, Levy flight step size a = 0.01, maximum number of iterations for the BP neural network was set to 1000, and error threshold was set to 10. -6The learning rate was 0.01. After training the neural network, the regression results were analyzed, and the regression analysis results are as follows: Figure 4 As shown.
[0033] With the goal of minimizing the bending-torsional coupling stress of the solder joints on the embedded substrate, the solder joint diameter was set to a range of 0.24mm-0.27mm, the pad diameter to a range of 0.19mm-0.22mm, and the solder joint height to a range of 0.14mm-0.17mm. The net file of the neural network and the normalized file were imported. After minimizing the bending-torsional coupling stress, the solder joint structural parameters were read, resulting in a solder joint diameter of 0.2400mm, a pad diameter of 0.2199mm, and a solder joint height of 0.1699mm. The results were approximated and rounded, ultimately yielding optimized solder joint structural parameters of a solder joint diameter of 0.24mm, a pad diameter of 0.22mm, and a solder joint height of 0.17mm. Finally, simulations were performed to verify the optimization and calculate the optimization rate.
[0034] The above description is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the protection scope of the present invention.
Claims
1. A method for optimizing the bending-torsional coupling stress of solder joints in embedded substrates based on CS-BPNN, characterized in that, include: 1): A three-dimensional finite element model of the embedded substrate solder joints was established based on ANSYS, including the embedded chip, PCB substrate, solder joint array and resin filler. 2): The orthogonal experimental design method was used to design the parameter combination of solder joint diameter, solder pad diameter and solder joint height, and the maximum bending-torsional coupling stress of the solder joint under each combination was obtained by finite element simulation. 3) Fabricate experimental specimens identical to the finite element model, conduct bending-torsional coupled strain measurement experiments, and compare the results with simulation results to verify the effectiveness; 4): The influence of each structural parameter on stress was ranked by magnitude through grey relational analysis; 5): Construct a BP neural network optimized by the cuckoo search algorithm, with the input layer being the weld joint structural parameters and the output layer being the predicted stress value; 6): With the goal of minimizing weld stress, the optimal combination of structural parameters is solved using an optimized neural network; 7): Verify the optimization effect through finite element simulation.
2. The method according to claim 1, characterized in that: The loading method for bending-torsional coupling in step 2) is as follows: apply a torsional displacement of 1 mm in different directions at the four corners of the PCB substrate, and at the same time apply a bending displacement of 2 mm at the center of the substrate.
3. The method according to claim 1, characterized in that: The sample fabrication method in step 3) is as follows: mill a groove slightly larger than the embedded chip on the PCB substrate, make pads in the groove, solder the embedded chip to the pads in the groove, attach right-angle strain roses to the embedded chip, and finally fill with resin filler.
4. The method according to claim 1, characterized in that: In step 4), the parameters of the cuckoo algorithm are set as follows: population size 30, discovery probability 0.25, maximum number of iterations 50, and Levi's flight step size a = 0.01.