A design method and design device of an underfill adhesive for packaging, and an electronic device

Abstract

This invention provides a design method, design apparatus, and electronic device for underfill adhesives used in encapsulation. The design method includes: acquiring a database containing underfill adhesive formulations and properties; establishing and training a property prediction model for the underfill adhesive based on the database, using the underfill adhesive formulation as input parameters and the underfill adhesive properties as output parameters; obtaining a preset formulation based on target properties; inputting the preset formulation into the property prediction model for property prediction to obtain predicted property values; and obtaining the target formulation for the underfill adhesive used in encapsulation based on the degree of matching between the predicted and target properties. This invention, by combining machine learning methods to establish a prediction model, can quickly analyze and predict various properties of underfill adhesive formulations, and screen formulations of suitable underfill adhesive materials based on the prediction results, rapidly identifying the direction for formulation optimization, thereby reducing experimental costs and significantly improving material development efficiency.

Description

Technical Field

[0001] This invention belongs to the field of encapsulation adhesive technology, specifically relating to a design method, design device, and electronic device for an encapsulation underfill adhesive. Background Technology

[0002] In the chip packaging process, the underfill adhesive penetrates between the chip and the substrate through a dispensing process, and then cures to form a stable adhesive layer. This reduces the stress on the solder joints due to the mismatch in the coefficients of thermal expansion between the chip and the substrate, alleviates the stress concentration on the chip and solder joints under thermal cycling, avoids interface delamination and solder joint cracking, and protects the solder joints from the influence of the surrounding environment such as moisture and ionic contaminants. This greatly improves the reliability of the packaging structure and the service life of the product.

[0003] Underfill adhesives typically use epoxy resin as the matrix, combined with fillers such as spherical silica and various additives to form a blend system. For example, the underfill adhesive disclosed in CN117801745A includes the following components by weight percentage: 24-33% epoxy resin, 58-69% silica, 8-19% curing agent, 0.1-0.3% coloring agent, and 0.2-0.4% adhesion promoter; wherein the epoxy resin is a compound of trifunctional epoxy resin, polyether-modified epoxy resin, and naphthalene-type epoxy resin. The underfill adhesive for packaging chips disclosed in CN116063968A includes the following components: 10-25 wt% organosilicon hybrid epoxy resin, 10-25 wt% curing agent, 50-70 wt% inorganic filler, 0.1-2 wt% curing accelerator, and 0.1-2 wt% coupling agent. The bottom filler composition disclosed in CN114989761A includes: 60-70 parts of pretreated spherical silica powder, 3-4 parts of amino epoxy resin, 9-10 parts of bisphenol F epoxy resin, 6-7 parts of naphthalene ring epoxy resin, 15-20 parts of acid anhydride curing agent, 0.3-0.5 parts of silane coupling agent, 3-5 parts of accelerator, 0.3-0.5 parts of surface treatment agent, and 0.05-0.4 parts of color paste.

[0004] Current underfiller formulations typically involve blending over a dozen compounds in appropriate proportions to achieve optimal property combinations. Traditional formulation development methods are based on one or more classic epoxy resin monomers (such as bisphenol A, bisphenol F, etc.), combining them with curing agents, fillers, and various additives. Design of Experiment (DOE) is used to identify key components and optimize their proportions to obtain the best formulation. However, DOE-based underfiller formulation design suffers from several problems: firstly, the vast formulation space makes component selection difficult, hindering the search for a global optimum; secondly, underfiller materials involve multiple process requirements and various bulk and interfacial properties after curing, making it a multi-objective optimization problem. Formulation designers must utilize their extensive knowledge of basic organic chemistry, polymer physical chemistry, and materials science to design experiments based on specific optimization objectives, resulting in high design barriers and significant human and time costs. Therefore, improving the R&D efficiency of underfillers is a pressing issue in this field. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the present invention aims to provide a design method, design device, and electronic device for underfill adhesives used in encapsulation. The design method, by combining artificial intelligence and machine learning methods, enables rapid design and formulation optimization of underfill adhesive materials, thereby significantly accelerating material development efficiency and reducing experimental costs.

[0006] To achieve this objective, the present invention adopts the following technical solution:

[0007] In a first aspect, the present invention provides a design method for an encapsulating underfill adhesive, the design method comprising:

[0008] Obtain a database containing underfill formulations and underfill properties;

[0009] Based on the database, a bottom filler property prediction model is established and trained, using the bottom filler formulation as the input parameter and the bottom filler properties as the output parameter.

[0010] A preset formula is obtained based on the target properties. The preset formula is then input into the filler property prediction model to predict the properties and obtain the predicted property data values.

[0011] Based on the matching degree between the predicted values ​​of the property data and the target properties, the target formulation of the underfill adhesive for encapsulation is obtained.

[0012] This invention provides a design method for underfill adhesives for packaging based on artificial intelligence. By combining machine learning methods, it can quickly predict various properties of underfill adhesive formulations and screen formulations of underfill adhesive materials that meet the requirements based on the prediction results. This greatly reduces experimental costs, significantly improves material development efficiency, and solves the technical problems of complex formulations, long material development cycles, difficulty in selecting components, and low DOE experimental efficiency of underfill adhesives for packaging.

[0013] The following are preferred technical solutions of the present invention, but are not intended to limit the technical solutions provided by the present invention. The purpose and beneficial effects of the present invention can be better achieved and realized through the following preferred technical solutions.

[0014] As a preferred embodiment of the present invention, the data in the database is derived from existing DOE experimental results and supplementary results based on machine learning.

[0015] As a preferred embodiment of the present invention, the underfill adhesive formulation includes a combination of epoxy resin, curing agent, filler and optional additives.

[0016] As a preferred embodiment of the present invention, the total mass of all components in the underfill adhesive formulation is 100%, and the total mass of the epoxy resin and curing agent is 10-60%.

[0017] As a preferred embodiment of the present invention, the filler comprises 40-90% of the total mass of all components in the bottom filler formulation, which is 100% of the total mass.

[0018] As a preferred embodiment of the present invention, the mass of the additive is 0-15% based on the total mass of all components in the bottom filler formulation being 100%.

[0019] To ensure a stable three-dimensional network structure after the underfill adhesive cures, as a preferred embodiment of the present invention, the characteristic parameters of the epoxy resin and curing agent in the underfill adhesive formulation are denoted as r, where 0.6 < r < 1.5, and the calculation formula for r is shown in Equation I:

[0020] Formula I.

[0021] In formula I, m A,i The mass of each component of the epoxy resin is represented by m. B,i c represents the mass of each component of the curing agent. A,i c represents the epoxy equivalent of each component of the epoxy resin. B,i This indicates the active hydrogen equivalent (N) of each component in the curing agent. A This indicates the component number of the epoxy resin, N. B This indicates the number of components in the curing agent.

[0022] As a preferred technical solution of the present invention, the properties of the bottom filler include any one or a combination of at least two of the following: glass transition temperature, modulus, coefficient of thermal expansion, viscosity, and interfacial adhesion strength.

[0023] As a preferred technical solution of the present invention, the establishment and training of the filler adhesive property prediction model includes:

[0024] Data preprocessing;

[0025] Feature engineering;

[0026] The dataset is divided into training and validation sets; and

[0027] Machine learning model training.

[0028] As a preferred embodiment of the present invention, the data preprocessing includes special data processing and data standardization.

[0029] As a preferred embodiment of the present invention, the special data processing includes any one or a combination of at least two of missing data processing, duplicate data processing, and abnormal data processing.

[0030] As a preferred technical solution of the present invention, the data standardization method includes any one or a combination of at least two of the following: standard deviation standardization, range standardization, robust standardization, and interval scaling.

[0031] As a preferred technical solution of the present invention, the algorithm used in the feature engineering includes any one or a combination of at least two of the following: Principal Component Analysis (PCA), Linear Discriminant Analysis (LDA), Singular Value Decomposition (SVD), Locally Linear Embedding (LLE), Stochastic Neighbor Embedding (SNE), t-Distributed Neighborhood Embedding (t-NSE), Autoencoders, LASSO (Least Absolute Value Convergence and Selection) method, and Feature Recursive Elimination (RFE).

[0032] As a preferred embodiment of the present invention, with the total number of samples in the dataset being 100%, the sample size of the training set is 70-95%.

[0033] As a preferred technical solution of the present invention, the machine learning model training includes Linear Regression (LR), Polynomial Regression, K-Nearest Neighbors Regressor, Adaboost, Ridge Regression, Support Vector Machine Regression (SVM), Gaussian Process Regression (GPR), Decision Tree Regressor (DT), Random Forest (RF), Gradient Boost Decision Tree (GBDT), Extreme Gradient Boost (XGBoost), Light Gradient Boosting Machine (LightGBM), Categorical Boosting (CatBoost), Artificial Neural Network (ANN), Recursive Neural Network (RNN), Convolutional Neural Network (CNN), and Graph Convolutional Neural Network (CNN). Any one or at least two combinations of gradient boosting trees and / or extreme gradient boosting networks (GCNs), preferably gradient boosting trees and / or extreme gradient boosting.

[0034] To ensure the accuracy of model predictions, as a preferred technical solution of the present invention, the trained prediction model for the physical properties of the filler glue satisfies: the correlation coefficient R of the validation set. 2 >88%.

[0035] As a preferred technical solution of the present invention, the trained prediction model for the physical properties of the filler glue satisfies the following condition: the ratio of the root mean square error of the verification set to the mean of the target property is <10%.

[0036] As a preferred embodiment of the present invention, the preset formulation includes a combination of epoxy resin, curing agent, filler and additives.

[0037] As a preferred embodiment of the present invention, based on the total mass of all components in the preset formula being 100%, the total mass of the epoxy resin and curing agent is 10-60%, the mass of the filler is 40-90%, and the mass of the additives is ≤15%.

[0038] To ensure a stable three-dimensional network structure after the bottom filler has cured, as a preferred embodiment of the present invention, the characteristic parameters of the epoxy resin and curing agent in the preset formula are denoted as r, where 0.6 < r < 1.5, and the calculation formula for r is shown in Equation I:

[0039] Formula I.

[0040] In formula I, m A,i The mass of each component of the epoxy resin is represented by m. B,i c represents the mass of each component of the curing agent. A,i c represents the epoxy equivalent of each component of the epoxy resin. B,i This indicates the active hydrogen equivalent (N) of each component in the curing agent. A This indicates the component number of the epoxy resin, N. B This indicates the number of components in the curing agent.

[0041] As a preferred technical solution of the present invention, the method for judging the matching degree includes: calculating the formula matching index S, wherein the smaller the formula matching index S, the better the matching degree, that is, the closer the properties corresponding to the preset formula are to the target.

[0042] The formula for calculating the formula matching index S is shown in Equation II:

[0043] Formula II;

[0044] In Equation II, N X X represents the property number. i Let X represent the target value of the i-th property. i pre Let f represent the predicted value of the i-th property, g represent the standardization function used to scale the value to 0-1, and f represent the predicted value of the i-th property. i This represents the weight coefficient of the i-th property, which is usually a positive real number.

[0045] As a preferred embodiment of the present invention, the design method further includes:

[0046] Several preset formulas are obtained based on the target properties. These preset formulas are then input into the filler property prediction model to predict the properties, and the property data prediction values ​​for each preset formula are obtained.

[0047] Substituting the predicted value of the property data into Equation II, we obtain the formula matching index S for each preset formula. Based on the formula matching index S, we sort several preset formulas and obtain at least one formula with good matching degree as the target formula for the bottom filler adhesive for encapsulation.

[0048] As a preferred technical solution of the present invention, the design method includes the following steps:

[0049] S1. Obtain the database, which includes the underfiller formulation and underfiller properties;

[0050] S2. Based on the database, using the underfiller formulation as the input parameter and the underfiller properties as the output parameter, establish and train a predictive model for the physical properties of the underfiller.

[0051] The establishment and training of the filler property prediction model includes:

[0052] Data preprocessing includes special data processing and data standardization;

[0053] Feature engineering;

[0054] The dataset is divided into training and validation sets; and

[0055] Machine learning model training;

[0056] The trained prediction model for the physical properties of the filler glue satisfies the following: the correlation coefficient R on the validation set. 2 >88%, and / or, the ratio of the root mean square error of the verification set to the mean of the target property is <10%;

[0057] S3. Obtain several preset formulas according to the target properties, input the several preset formulas into the filler property prediction model to predict the properties, and obtain the property data prediction value of each preset formula respectively;

[0058] S4. Calculate the formula matching index S based on the predicted values ​​of the property data and the target properties. The formula matching index S is calculated using the formula II shown in Equation II:

[0059] Formula II;

[0060] In Equation II, N X X represents the property number. i Let X represent the target value of the i-th property. i pre Let f represent the predicted value of the i-th property, g represent the standardization function used to scale the value to the range of 0-1; i This represents the weight coefficient of the i-th property;

[0061] Based on the formula matching index S, several preset formulas are sorted to obtain at least one formula with a small formula matching index S and a good matching degree as the target formula for the bottom filler adhesive for encapsulation.

[0062] As a preferred technical solution of the present invention, the design method further includes: obtaining the measured values ​​of the property data of the preset formula, and determining whether to perform iteration based on the deviation between the measured values ​​of the property data and the predicted values ​​of the property data.

[0063] The iterative method includes:

[0064] The measured values ​​of the aforementioned properties (and the corresponding formulations) are incorporated into the database for data updates, and the filler property prediction model is retrained based on the updated database.

[0065] Based on the target properties, the preset formula is obtained again and input into the retrained filler property prediction model to obtain new property data prediction values;

[0066] Based on the matching degree between the predicted values ​​of the new property data and the target properties, a new target formulation for the underfill adhesive for encapsulation is obtained.

[0067] Preferably, if the relative error between the measured value and the predicted value of the property data is greater than 30%, then iteration is performed.

[0068] In a second aspect, the present invention provides a design apparatus for an encapsulating underfill adhesive, the design apparatus comprising:

[0069] The database module is used to obtain and / or store underfill formulations and underfill properties;

[0070] The prediction model generation module is used to establish and train a prediction model of the physical properties of the filler using the filler formulation as input parameters and the filler properties as output parameters.

[0071] The recipe generation module is used to generate preset recipes; and,

[0072] The formulation design module is used to input the preset formulation into the filler property prediction model to obtain property data prediction values, and to obtain the target formulation of the bottom filler for encapsulation based on the matching degree between the property data prediction values ​​and the target properties.

[0073] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the design method for underfill adhesive for encapsulation as described in the first aspect.

[0074] Fourthly, the present invention provides an electronic device, the electronic device comprising:

[0075] Memory;

[0076] Processor; and,

[0077] A computer program stored in the memory and executable on a processor;

[0078] When the processor executes the computer program, it implements the design method for underfill adhesive for packaging as described in the first aspect.

[0079] Compared with the prior art, the present invention has at least the following beneficial effects:

[0080] This invention provides an artificial intelligence-based design method for underfill adhesives used in encapsulation. By combining machine learning methods to establish a predictive model of the underfill adhesive's physical properties, it can quickly analyze and predict various properties of the underfill adhesive formulation. Based on the prediction results, it can screen underfill adhesive materials that meet the requirements, quickly identify the direction for formulation optimization, thereby greatly reducing experimental costs, significantly improving material development efficiency, and shortening the development cycle to 1-2 months. Simultaneously, this design method can reduce the learning costs for relevant personnel, avoid designing ineffective formulations due to a lack of professional knowledge, and improve experimental accuracy. Attached Figure Description

[0081] Figure 1 A schematic diagram of the dispensing and curing process of an encapsulation underfill adhesive provided for one specific embodiment;

[0082] Figure 2 A schematic diagram of a database provided for one specific implementation method;

[0083] Figure 3a The prediction effect diagram of the filler adhesive property prediction model based on support vector machine regression provided for a specific implementation method;

[0084] Figure 3b A prediction result diagram of a filler adhesive property prediction model based on a fully connected neural network provided for a specific implementation;

[0085] Figure 3c The prediction effect of a Gaussian process regression-based model for predicting the physical properties of filler adhesives is shown in a specific implementation diagram.

[0086] Figure 3d The prediction effect diagram of the random forest-based filler property prediction model provided for a specific implementation method;

[0087] Figure 3e A prediction result diagram of a gradient boosting tree-based filler property prediction model provided for a specific implementation;

[0088] Figure 3f The prediction effect diagram of the filler property prediction model based on extreme gradient boosting provided for a specific implementation method;

[0089] Figure 4a A graph showing the predicted glass transition temperature for one specific implementation method;

[0090] Figure 4b A graph showing the predicted high-temperature modulus for one specific implementation method;

[0091] Figure 4c A graph showing the predicted viscosity for one specific implementation method;

[0092] Figure 4d A diagram showing the predicted effect of silicon-based adhesive force for a specific implementation method;

[0093] Figure 4e A diagram showing the predicted adhesion of silicon after damp heat aging, provided for a specific implementation method;

[0094] Figure 4f A diagram showing the predicted effect of the adhesive strength of the molding compound as provided in a specific implementation method;

[0095] Figure 5 A schematic diagram of the iterative and formulation screening process provided for one specific implementation;

[0096] Figure 6a A comparison graph of the measured glass transition temperature and the target value of the target formulation provided for a specific embodiment;

[0097] Figure 6b A comparison chart of the measured and target values ​​of the high-temperature modulus of the target formulation provided for a specific implementation method;

[0098] Figure 6c A comparison chart of measured and target values ​​of silicone-based adhesive strength for a target formulation provided in a specific embodiment;

[0099] Figure 7 A schematic flow diagram of a design apparatus for an encapsulation underfill adhesive provided in one specific embodiment. Detailed Implementation

[0100] To facilitate understanding of the present invention, specific embodiments are provided to further illustrate the technical solution of the present invention. Those skilled in the art should understand that the embodiments are merely illustrative of the present invention and should not be considered as specific limitations thereof.

[0101] The terms “comprising,” “including,” “having,” “containing,” or any other variations thereof, as used herein, are intended to cover non-exclusive inclusion. For example, a composition, step, method, article, or apparatus that includes the listed elements is not limited to those elements and may also include other elements not expressly listed or elements inherent to such composition, step, method, article, or apparatus.

[0102] "Optional" or "any one" means that the matter or event described thereafter may or may not occur, and the description includes both the possibility that the event may occur and the possibility that the event may not occur.

[0103] The indefinite articles “a” and “an” preceding an element or component of this invention do not impose any limitation on the quantity (i.e., number of occurrences) of the element or component. Therefore, “an” or “a” should be interpreted as including one or at least one, and the singular form of an element or component also includes the plural form, unless the quantity clearly refers only to the singular form.

[0104] In this invention, terms such as "first," "second," "third," and "fourth" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or quantity, nor should they be construed as implicitly indicating the importance or quantity of the indicated technical features. Moreover, terms such as "first," "second," "third," and "fourth" serve only as a non-exhaustive enumeration and should be understood not to constitute a closed limitation on quantity.

[0105] Those skilled in the art will understand that the order in which the steps are written in the various embodiments or examples does not imply a strict execution order and does not limit the implementation process in any way. The detailed execution order of each step should be determined by its function and possible internal logic. Unless otherwise specified, all steps of the present invention may be performed sequentially or randomly, but are preferably performed sequentially. For example, the method includes steps (a) and (b), indicating that the method may include steps (a) and (b) performed sequentially, or it may include steps (b) and (a) performed sequentially. Furthermore, the mention that the method may also include step (c) indicates that step (c) can be added to the method in any order.

[0106] One specific embodiment of the present invention provides a design method for an underfill adhesive for encapsulation, the design method comprising:

[0107] Obtain a database containing underfill formulations and underfill properties;

[0108] Based on the database, a bottom filler property prediction model is established and trained, using the bottom filler formulation as the input parameter and the bottom filler properties as the output parameter.

[0109] A preset formula is obtained based on the target properties. The preset formula is then input into the filler property prediction model to predict the properties and obtain the predicted property data values.

[0110] Based on the matching degree between the predicted values ​​of the property data and the target properties, the target formulation of the underfill adhesive for encapsulation is obtained.

[0111] In one specific embodiment, a schematic diagram of the dispensing and curing process of the underfill adhesive for encapsulation is shown below. Figure 1 As shown, the underfill adhesive (UF adhesive) is penetrated between the chip and the carrier board through a dispensing process and then cured to form an adhesive layer.

[0112] This invention conducts in-depth research on underfill adhesives for encapsulation and proposes an artificial intelligence-based design method. The property prediction model for the underfill adhesive, established through machine learning strategies, can simultaneously analyze all components in the formulation design, quickly identifying the direction for formulation optimization. This avoids multiple DOE (design of experiment) tests required in traditional methods, accelerating the development speed and shortening the development cycle from over 6 months to approximately 1-2 months. Furthermore, the AI-based design method reduces the learning costs for relevant personnel, avoids designing ineffective formulations due to a lack of professional knowledge, and improves experimental accuracy.

[0113] In this invention, the data in the database comes from existing DOE experimental results and supplementary results based on machine learning.

[0114] In one specific embodiment, the database includes two parts: underfill formulation and underfill properties, as illustrated in the diagram below. Figure 2 As shown.

[0115] In one embodiment, the underfill formulation comprises a combination of epoxy resin, curing agent, filler, and optionally additives.

[0116] In one specific embodiment, the epoxy resin includes any one or a combination of at least two of glycidyl ether epoxy resin, glycidyl ester epoxy resin, aliphatic epoxy resin, alicyclic epoxy resin, polycyclic aromatic epoxy resin, and phenolic epoxy resin.

[0117] In one specific embodiment, the epoxy resin contains 1 to 5 types of molecules, such as 2, 3, 4 or 5 types, usually not less than 1 type, and usually not more than 5 types.

[0118] In one specific embodiment, the curing agent includes any one or a combination of at least two of the following: amine curing agents, acid anhydride curing agents, phenol curing agents, isocyanate curing agents, alcohol curing agents, ester curing agents, urea curing agents, and hydrazide curing agents.

[0119] In one specific embodiment, the curing agent contains 1 to 5 types of molecules, such as 2, 3, 4 or 5 types, usually not less than 1 type, and usually not more than 5 types.

[0120] In one specific embodiment, the filler pack is an inorganic oxide microsphere at the nanometer to micrometer scale, including but not limited to: any one or a combination of at least two of silicon dioxide, aluminum oxide, aluminum nitride, and silicon nitride.

[0121] In one specific embodiment, the surface of the filler may be subjected to surface modification treatment.

[0122] In one specific embodiment, the filler contains 1 to 5 types of microspheres with different particle size distributions and / or different surface modifications, such as 2, 3, 4 or 5 types, usually not less than 1 type, and usually not more than 5 types.

[0123] The additives are other additives besides the components mentioned above that can adjust and / or improve the bulk and interfacial properties, including macromolecular additives and / or small molecule additives, with small molecule additives being preferred.

[0124] In one specific embodiment, the additives include any one or a combination of at least two of the following: curing accelerators, interface modifiers, viscosity modifiers, toughening agents, stabilizers, antioxidants, leveling agents, defoamers, ion trapping agents, and colorants.

[0125] In one specific embodiment, the adjuvant contains 1 to 25 types of molecules, such as 2, 3, 4, 5, 6, 7, 8, 9, 10, 12, 14, 15, 16, 18, 20, 22, or 24 types, as well as specific point values ​​between the above-mentioned point values. Due to space limitations and for the sake of brevity, this invention will not exhaustively list the specific point values ​​included in the range. Preferably, there are usually no fewer than 3 types, and the upper limit is usually no more than 25 types.

[0126] In one specific embodiment, with the total mass of all components in the underfill adhesive formulation being 100%, the total mass of the epoxy resin and curing agent is 10-60%, for example, it can be 15%, 20%, 25%, 30%, 35%, 40%, 45%, 50% or 55%, and specific values ​​between the above points. Due to space limitations and for the sake of brevity, the present invention will not exhaustively list the specific values ​​included in the range.

[0127] In one specific embodiment, with the total mass of all components in the underfill adhesive formulation being 100%, the mass of the filler is 40-90%, for example, it can be 45%, 50%, 55%, 60%, 65%, 70%, 75%, 80% or 85%, and specific values ​​between the above points. Due to space limitations and for the sake of brevity, the present invention will not exhaustively list the specific values ​​included in the range.

[0128] In one specific embodiment, with the total mass of all components in the underfill adhesive formulation as 100%, the mass of the additive is 0-15%, for example, it can be 0.1%, 0.5%, 0.8%, 1%, 2%, 3%, 4%, 5%, 6%, 7%, 8%, 9%, 10%, 11%, 12%, or 14%, as well as specific values ​​between the above values. Due to space limitations and for the sake of brevity, the present invention will not exhaustively list the specific values ​​included in the range.

[0129] In one specific embodiment, the characteristic parameters of epoxy resin and curing agent in the underfill adhesive formulation are denoted as r, where 0.6 < r < 1.5. For example, r can be 0.7, 0.8, 0.9, 1, 1.1, 1.2, 1.3, or 1.4, as well as specific values ​​between the above values. Due to space limitations and for the sake of brevity, this invention will not exhaustively list all the specific values ​​included in the range, thereby ensuring that a stable three-dimensional network structure is formed after the underfill adhesive is cured.

[0130] The formula for calculating r is shown in Equation I:

[0131] Formula I;

[0132] In formula I, m A,i The mass of each component of the epoxy resin is represented by m. B,i c represents the mass of each component of the curing agent. A,i c represents the epoxy equivalent of each component of the epoxy resin. B,i This indicates the active hydrogen equivalent (N) of each component in the curing agent. A This indicates the component number of the epoxy resin, N. B This indicates the number of components in the curing agent.

[0133] The characteristic parameter r represents the molar ratio of actual active hydrogen to epoxy groups in the formulation.

[0134] In one specific embodiment, the properties of the underfill adhesive include any one or a combination of at least two of the following: glass transition temperature, modulus, coefficient of thermal expansion, viscosity, and interfacial bond strength.

[0135] In one specific embodiment, the glass transition temperature (T) g The test can be performed in a variety of ways, including but not limited to: differential scanning calorimetry (DSC), thermomechanical analysis (TMA), dynamic mechanical analysis (DMA), etc.

[0136] In one specific embodiment, the modulus includes a high-temperature modulus and a low-temperature modulus, which correspond to the Young's modulus, flexural modulus, and shear modulus of the material before and after the glass transition temperature, respectively, and can be tested in various ways; for example, the Young's modulus can be measured by uniaxial tension, such as with reference to standard ASTM D638, the flexural modulus can be measured by the three-point / four-point bending method, such as with reference to standard ASTM D790, and the shear modulus can be measured by V-notch beam shear, such as with reference to standard ASTM D5379.

[0137] In one specific embodiment, the coefficient of thermal expansion (CTE) includes high-temperature CTE and low-temperature CTE, which correspond to the coefficient of thermal expansion of the material as the temperature increases before and after the glass transition temperature, respectively. The measurement methods include, but are not limited to, thermomechanical analysis (TMA) methods.

[0138] In one specific embodiment, the viscosity test temperature range is 10-250℃, specifically 20℃, 25℃, 30℃, 40℃, 50℃, 60℃, 70℃, 80℃, 90℃, 100℃, 120℃, 150℃, 180℃, 200℃, 220℃, 240℃, etc.; viscosity can be measured in various ways, including but not limited to using a viscometer, rheometer, etc.

[0139] In one specific embodiment, the interfacial adhesive strength includes the properties between the underfill adhesive and various material interfaces. These interfacial materials include various metals, non-metals, and organic materials, exemplarily including but not limited to silicon, silica, copper, aluminum, iron, and epoxy molding compounds. Furthermore, the adhesive strength includes both conventional and damp heat aging strengths. The damp heat aging process can be conducted in various ways, including but not limited to high accelerated temperature and humidity pressure (HAST) testing. The adhesive strength can be measured in various ways, including but not limited to preparing a pudding mold and testing it using a push-pull shear force tester (e.g., referring to the disclosure in prior art CN116296653A).

[0140] In one specific implementation, a formula-property relationship model is established by using the underfiller formulation information in the database as input and the underfiller property information as output.

[0141] In one specific implementation, the establishment and training of the filler adhesive property prediction model includes:

[0142] Data preprocessing;

[0143] Feature engineering;

[0144] The dataset is divided into training and validation sets; and

[0145] Machine learning model training.

[0146] In one specific implementation, the data preprocessing includes special data processing and data standardization.

[0147] In one specific implementation, the special data processing includes any one or a combination of at least two of missing data processing, duplicate data processing, and abnormal data processing.

[0148] In one specific implementation, some features may be missing due to the inability to sample or the lack of observations, i.e., there is missing data (null values); the methods for handling missing data (filling null values) are exemplarily including but not limited to: fixed value method, mean method, mode method, upper and lower data method, and various machine learning algorithms.

[0149] In one specific implementation, the processing of duplicate data (data duplicate values) includes: determining whether it should be deleted directly based on actual needs.

[0150] In one specific implementation, outlier data (outlier values) processing includes: removing outliers using various algorithms, including but not limited to: statistical outlier detection algorithms (using range, interquartile range, mean, standard deviation, etc. for judgment), distance-based outlier detection algorithms (using absolute distance, Euclidean distance, Mahalanobis distance, etc. for judgment), density-based outlier detection algorithms, etc.

[0151] Data standardization is a key step in improving model accuracy and accelerating convergence. In one specific implementation, the data standardization method includes any one or a combination of at least two of the following: standard deviation standardization, range standardization, robust standardization, and interval scaling.

[0152] Because the recipe space has a huge dimensionality, feature engineering is needed to reduce it. In one specific implementation, the algorithm used for feature engineering includes any one or a combination of at least two of the following: Principal Component Analysis (PCA), Linear Discriminant Analysis (LDA), Singular Value Decomposition (SVD), Locally Linear Embedding (LLE), Stochastic Nearest Neighbor Embedding (SNE), t-Distributed Neighborhood Embedding (t-NSE), Autoencoders, LASSO (Least Absolute Value Convergence and Selection) method, and Recursive Feature Eradication (RFE).

[0153] The feature engineering helps to reduce the dimension of the feature space to a reasonable range and eliminate the influence of components that are less correlated with specific properties.

[0154] In one specific implementation, with the total number of samples in the dataset being 100%, the sample size of the training set is 70-95%, for example, it can be 72%, 75%, 78%, 80%, 82%, 85%, 88%, 90%, 92%, or 94%, as well as specific point values ​​between the above point values. Due to space limitations and for the sake of brevity, this invention will not exhaustively list the specific point values ​​included in the range.

[0155] In one specific implementation, the number of samples required for training the machine learning model is ≥100.

[0156] In one specific implementation, the machine learning model training includes any one or a combination of at least two of the following: Linear Regression (LR), Polynomial Regression, K-Nearest Neighbors Regressor, Adaboost, Ridge Regression, Support Vector Machine Regression (SVM), Gaussian Process Regression (GPR), Decision Tree Regressor (DT), Random Forest (RF), Gradient Boost Decision Tree (GBDT), Extreme Gradient Boost (XGBoost), Light Gradient Boosting Machine (LightGBM), Categorical Boosting (CatBoost), Artificial Neural Network (ANN), Recursive Neural Network (RNN), Convolutional Neural Network (CNN), and Graph Convolutional Neural Network (GCN).

[0157] In one specific implementation, the machine learning model training includes Gradient Boosting Tree (GBDT) and / or Extreme Gradient Boosting (XGBoost), which outperform other models on both the training and validation sets.

[0158] In one specific implementation, the property prediction performance of different machine learning models is shown in the following figure. Figure 3a , Figure 3b , Figure 3c , Figure 3d , Figure 3e and Figure 3f As shown, different machine learning models were used to determine the glass transition temperature T. g The prediction, specifically, Figure 3a This is a graph showing the prediction results of a prediction model based on support vector machine regression. Figure 3b The image shows the prediction results of a prediction model based on a fully connected neural network (ANN, with 3 layers). Figure 3c This is a graph showing the prediction results of a prediction model based on Gaussian process regression. Figure 3d The image shows the prediction results of a random forest-based prediction model. Figure 3e This is a prediction result diagram of the gradient boosting tree-based prediction model. Figure 3f The graphs show the prediction results of the prediction model based on extreme gradient boosting. The horizontal axis T of each graph is... g -exp represents the glass transition temperature T. g The experimental values, ordinate T g -pre indicates T g The predicted values ​​are given by R. The training set is the training set, and the validation set is the validation set. The training RMSE represents the root mean square error of the training set, and the validation RMSE represents the root mean square error of the validation set. 2 This represents the correlation coefficient; comparing the above figures, it can be seen that GBDT and XGBoost have higher correlation coefficients R on the training and validation sets. 2 It has higher accuracy and better performance.

[0159] To ensure the accuracy of model predictions, in one specific implementation, the trained prediction model for the physical properties of the filler adhesive satisfies the following: the correlation coefficient R on the validation set... 2 >88%, for example, it can be 89%, 90%, 91%, 92%, 93%, 94%, 95%, 96%, 97%, 98%, 99%, 99.5%, 99.9%, 99.95%, 99.99%, 99.992%, 99.995%, 99.998%, 99.999%, 99.9999%, etc., with R being preferred. 2 ≥89%.

[0160] In one specific implementation, the trained prediction model for the physical properties of the filler glue satisfies the following condition: the ratio of the root mean square error of the validation set to the mean of the target property is <10%, for example, it can be 0.1%, 0.5%, 0.8%, 1%, 2%, 3%, 4%, 5%, 6%, 7%, 8%, 9%, or 9.5%, etc.

[0161] In one specific implementation, the prediction results of some models are shown in the following figure. Figure 4a , Figure 4b , Figure 4c , Figure 4d , Figure 4e and Figure 4f shown, specifically, Figure 4a This is a graph showing the predicted glass transition temperature. Figure 4b This is a graph showing the predicted results for the high-temperature modulus. Figure 4c This is a graph showing the predicted viscosity. Figure 4d This is a diagram showing the predicted adhesion strength of the silicon-based substrate. Figure 4e This is a diagram showing the predicted adhesion strength of silicone after damp heat aging. Figure 4f The graphs show the predicted adhesion strength of the molding compound. The horizontal axis (exp) represents the experimental value of this property, and the vertical axis (pre) represents the predicted value. The training set and validation set are shown. Training RMSE represents the root mean square error of the training set, and Valid RMSE represents the root mean square error of the validation set. R0 2 Represents the correlation coefficient; the correlation coefficient R of the validation set in each figure. 2 ≥89%, reaching 89.2-94.4%, training set correlation coefficient R 2 ≥99.8%, which can be 99.8-100%; at the same time, the ratio of the root mean square error of the validation set (Valid RMSE) to the mean of the target property is <10%, indicating that the prediction model of the filler property after training in this invention has high accuracy.

[0162] In one specific embodiment, the preset formulation includes a combination of epoxy resin, curing agent, filler and additives.

[0163] In one specific embodiment, the epoxy resin includes any one or a combination of at least two of glycidyl ether epoxy resin, glycidyl ester epoxy resin, aliphatic epoxy resin, alicyclic epoxy resin, polycyclic aromatic epoxy resin, and phenolic epoxy resin.

[0164] In one specific embodiment, the epoxy resin contains 1 to 5 types of molecules, such as 2, 3, 4 or 5 types, usually not less than 1 type, and usually not more than 5 types.

[0165] In one specific embodiment, the curing agent includes any one or a combination of at least two of the following: amine curing agents, acid anhydride curing agents, phenol curing agents, isocyanate curing agents, alcohol curing agents, ester curing agents, urea curing agents, and hydrazide curing agents.

[0166] In one specific embodiment, the curing agent contains 1 to 5 types of molecules, such as 2, 3, 4 or 5 types, usually not less than 1 type, and usually not more than 5 types.

[0167] In one specific embodiment, the filler pack is an inorganic oxide microsphere at the nanometer to micrometer scale, preferably any one or a combination of at least two of silicon dioxide, aluminum oxide, aluminum nitride, and silicon nitride.

[0168] In one specific embodiment, the surface of the filler may be subjected to surface modification treatment.

[0169] In one specific embodiment, the filler contains 1 to 5 types of microspheres with different particle size distributions and / or different surface modifications, such as 2, 3, 4 or 5 types, usually not less than 1 type, and usually not more than 5 types.

[0170] The additives are other additives besides the components mentioned above that can adjust and / or improve the bulk and interfacial properties, including macromolecular additives and / or small molecule additives, with small molecule additives being preferred.

[0171] In one specific embodiment, the additives include any one or a combination of at least two of the following: curing accelerators, interface modifiers, viscosity modifiers, toughening agents, stabilizers, antioxidants, leveling agents, defoamers, ion trapping agents, and colorants.

[0172] In one specific embodiment, the adjuvant contains 1 to 25 types of molecules, such as 2, 3, 4, 5, 6, 7, 8, 9, 10, 12, 14, 15, 16, 18, 20, 22, or 24 types, as well as specific point values ​​between the above-mentioned point values. Due to space limitations and for the sake of brevity, this invention will not exhaustively list the specific point values ​​included in the range. Preferably, there are usually no fewer than 3 types, and the upper limit is usually no more than 25 types.

[0173] In one specific embodiment, the formulation includes a combination of the aforementioned epoxy resin, curing agent, filler, and additives, and the formulation as a whole contains 6-40 components (e.g., 7, 8, 10, 12, 15, 18, 20, 22, 25, 28, 30, 32, 35, 38, etc.), and the sum of the mass contents of all components is 100%.

[0174] In one specific embodiment, taking the total mass of all components in the preset formula as 100%, the total mass of the epoxy resin and curing agent is 10-60%, for example, it can be 15%, 20%, 25%, 30%, 35%, 40%, 45%, 50%, or 55%, etc.; the mass of the filler is 40-90%, for example, it can be 45%, 50%, 55%, 60%, 65%, 70%, 75%, 80%, or 85%, etc.; the mass of the additives is ≤15%, for example, it can be 0.1%, 0.5%, 0.8%, 1%, 2%, 3%, 4%, 5%, 6%, 7%, 8%, 9%, 10%, 11%, 12%, or 14%.

[0175] In one specific embodiment, to ensure that a stable three-dimensional network structure is formed after the bottom filler adhesive cures, the characteristic parameters of epoxy resin and curing agent in the preset formula are denoted as r, where 0.6 < r < 1.5. For example, r can be 0.7, 0.8, 0.9, 1, 1.1, 1.2, 1.3 or 1.4, etc.

[0176] The formula for calculating r is shown in Equation I:

[0177] Formula I.

[0178] In formula I, m A,i The mass of each component of the epoxy resin is represented by m. B,i c represents the mass of each component of the curing agent. A,i c represents the epoxy equivalent of each component of the epoxy resin. B,i This indicates the active hydrogen equivalent (N) of each component in the curing agent. A This indicates the component number of the epoxy resin, N. B This indicates the number of components in the curing agent.

[0179] The characteristic parameter r represents the molar ratio of actual active hydrogen to epoxy groups in the formulation. In this invention, the obtained preset formulation must satisfy all the above conditions to ensure that the preset formulation conforms to the underfill adhesive design specifications.

[0180] In this invention, given a target combination of properties and formulation boundary conditions (mass content range of each component), a preset formulation is obtained, the preset formulation is input into the filler property prediction model, and a formulation that satisfies the combination of properties is found based on a strategy of matching degree and formulation matching index.

[0181] In one specific implementation, the method for determining the matching degree includes: calculating the formula matching index S, wherein the smaller the formula matching index S, the better the matching degree, that is, the closer the properties corresponding to the preset formula are to the target.

[0182] The formula for calculating the formula matching index S is shown in Equation II:

[0183] Formula II;

[0184] In Equation II, N X X represents the property number. i Let X represent the target value of the i-th property. i pre Let f represent the predicted value of the i-th property, g represent the standardization function used to scale the value to 0-1, and f represent the predicted value of the i-th property. i This represents the weight coefficient of the i-th property, usually a positive real number, used to balance the emphasis on different properties.

[0185] In one specific embodiment, the design method further includes:

[0186] Several preset formulas are obtained based on the target properties. These preset formulas are then input into the filler property prediction model to predict the properties, and the property data prediction values ​​for each preset formula are obtained.

[0187] Substituting the predicted value of the property data into Equation II, we obtain the formula matching index S for each preset formula. Based on the formula matching index S, we sort several preset formulas and obtain at least one formula with good matching degree as the target formula for the bottom filler adhesive for encapsulation.

[0188] In one specific implementation, several preset formulas are sorted according to the formula matching index S, and the formulas corresponding to the 10 smallest S values ​​are selected for recommendation.

[0189] In one specific embodiment, the design method includes the following steps:

[0190] S1. Obtain the database, which includes the underfiller formulation and underfiller properties;

[0191] S2. Based on the database, using the underfiller formulation as the input parameter and the underfiller properties as the output parameter, establish and train a predictive model for the physical properties of the underfiller.

[0192] The establishment and training of the filler property prediction model includes:

[0193] Data preprocessing includes special data processing and data standardization;

[0194] Feature engineering;

[0195] The dataset is divided into training and validation sets; and

[0196] Machine learning model training;

[0197] The trained prediction model for the physical properties of the filler glue satisfies the following: the correlation coefficient R on the validation set. 2 >88%, and / or, the ratio of the root mean square error of the verification set to the mean of the target property is <10%;

[0198] S3. Obtain several preset formulas according to the target properties, input the several preset formulas into the filler property prediction model to predict the properties, and obtain the property data prediction value of each preset formula respectively;

[0199] S4. Calculate the formula matching index S based on the predicted values ​​of the property data and the target properties. The formula matching index S is calculated using the formula II shown in Equation II:

[0200] Formula II;

[0201] In Equation II, N X X represents the property number. i Let X represent the target value of the i-th property. i pre Let f represent the predicted value of the i-th property, g represent the standardization function used to scale the value to the range of 0-1; i This represents the weight coefficient of the i-th property;

[0202] Based on the formula matching index S, several preset formulas are sorted to obtain at least one formula with a small formula matching index S and a good matching degree as the target formula for the bottom filler adhesive for encapsulation.

[0203] In one specific embodiment, the design method further includes: obtaining the measured values ​​of the property data of the preset formula, and determining whether to perform iteration based on the deviation between the measured values ​​of the property data and the predicted values ​​of the property data.

[0204] The iterative method includes:

[0205] The measured values ​​of the aforementioned properties (and the corresponding formulations) are incorporated into the database for data updates, and the filler property prediction model is retrained based on the updated database.

[0206] Based on the target properties, the preset formula is obtained again and input into the retrained filler property prediction model to obtain new property data prediction values;

[0207] Based on the matching degree between the predicted values ​​of the new property data and the target properties, a new target formulation for the underfill adhesive for encapsulation is obtained.

[0208] In one specific implementation, if the relative error between the measured value and the predicted value of the property data is greater than 30%, then iteration is performed.

[0209] Figure 5 This diagram illustrates an iterative process for formula selection in one specific implementation. If the measured value of a certain property of the recommended formula (preset formula) deviates significantly from the target property combination (e.g., relative error > 30%), further iterations can be used to obtain the next recommended formula. The specific method includes: updating the measured data in a database; further training the model using the updated database; generating a new preset formula near the formula space where the measured property data in the current formula is closest to the target property combination; performing a new round of predictions using the model to obtain new predicted property data values; and then recommending a new formula based on the matching degree between the predicted property data values ​​and the target property. This iterative process can continue until a target formula that meets the requirements of the target property combination is found.

[0210] In one specific implementation, a comparison is made between the measured values ​​and target values ​​of some recommended formulations (target formulations). Figure 6a , Figure 6b , Figure 6c shown, specifically, Figure 6a This is a comparison chart of the measured glass transition temperature and the target value of the recommended formulation that can be used as the target formulation. Figure 6b This is a comparison chart of the measured and target values ​​of the high-temperature modulus of the target formulation. Figure 6c The graph shows a comparison between the measured values ​​and target values ​​of the silicone-based adhesive strength of the target formulation. The measured values ​​of the properties of the recommended formulations in each graph are very close to the target values, indicating that the filler property prediction model of the present invention has high accuracy. The design method of the present invention can accurately and quickly obtain the target formulation that meets the requirements.

[0211] Another specific embodiment of the present invention provides a design apparatus for an encapsulation underfill adhesive, the design apparatus comprising:

[0212] The database module is used to obtain and / or store underfill formulations and underfill properties;

[0213] The prediction model generation module is used to establish and train a prediction model of the physical properties of the filler using the filler formulation as input parameters and the filler properties as output parameters.

[0214] The recipe generation module is used to generate preset recipes; and,

[0215] The formulation design module is used to input the preset formulation into the filler property prediction model to obtain property data prediction values, and to obtain the target formulation of the bottom filler for encapsulation based on the matching degree between the property data prediction values ​​and the target properties.

[0216] In one specific embodiment, the design apparatus for the underfill adhesive for encapsulation adopts the design method for the underfill adhesive for encapsulation proposed in this invention, the flowchart of which is shown below. Figure 7 As shown, it includes the following four modules: a formulation generation module, a database module, a prediction model generation module (machine learning model training module), and a formulation design module. The formulation generation module searches the space based on the addition ratio of each component to form sample formulations to be screened, obtains preset formulations, and establishes a sample formulation database. The database module stores the formulation of each bottom filler sample and its corresponding macroscopic physical property data. The prediction model generation module (machine learning model training module) preprocesses the data in the database and selects various machine learning models for training to obtain and save the bottom filler property prediction model. The formulation design module uses the formulation generation module to search the component space to form preset formulations (candidate formulations), then uses the bottom filler property prediction model to obtain the corresponding physical properties of all candidate formulations, and recommends the optimal target formulation for the bottom filler.

[0217] Another embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the design method for underfill adhesive for encapsulation as described in the foregoing embodiments of the present invention.

[0218] Another specific embodiment of the present invention provides an electronic device, the electronic device comprising:

[0219] Memory;

[0220] Processor; and,

[0221] A computer program stored in the memory and executable on a processor;

[0222] When the processor executes the computer program, it implements the design method for underfill adhesive for packaging as described in the foregoing specific embodiments of the present invention.

[0223] The present invention has been illustrated through the above embodiments, but the present invention is not limited to the above process steps, that is, it does not mean that the present invention must rely on the above process steps to be implemented. Those skilled in the art should understand that any improvements to the present invention, equivalent substitutions of the raw materials used in the present invention, additions of auxiliary components, and selection of specific methods, etc., all fall within the protection scope and disclosure scope of the present invention.

Claims

1. A design method for an underfill adhesive for encapsulation, characterized in that, The design method includes: Obtain a database containing underfill formulations and underfill properties; Based on the database, a bottom filler property prediction model is established and trained, using the bottom filler formulation as the input parameter and the bottom filler properties as the output parameter. A preset formula is obtained based on the target properties. The preset formula is then input into the filler property prediction model to predict the properties and obtain the predicted property data values. Based on the matching degree between the predicted values ​​of the property data and the target properties, the target formulation of the underfill adhesive for encapsulation is obtained.

2. The design method according to claim 1, characterized in that, The underfill formulation comprises a combination of epoxy resin, curing agent, filler, and optional additives; Preferably, based on the total mass of all components in the underfill adhesive formulation as 100%, the total mass of the epoxy resin and curing agent is 10-60%, the mass of the filler is 40-90%, and the mass of the additives is 0-15%. Preferably, the characteristic parameters of the epoxy resin and curing agent in the bottom filler formulation are denoted as r, where 0.6 < r < 1.5, and the calculation formula for r is shown in Equation I: Equation I; In formula I, m A,i The mass of each component of the epoxy resin is represented by m. B,i c represents the mass of each component of the curing agent. A,i c represents the epoxy equivalent of each component of the epoxy resin. B,i This indicates the active hydrogen equivalent (N) of each component in the curing agent. A This indicates the component number of the epoxy resin, N. B Indicates the number of components in the curing agent; Preferably, the properties of the bottom filler include any one or a combination of at least two of the following: glass transition temperature, modulus, coefficient of thermal expansion, viscosity, and interfacial bond strength.

3. The design method according to claim 1 or 2, characterized in that, The establishment and training of the filler property prediction model includes: Data preprocessing; Feature engineering; The dataset is divided into training and validation sets; and Machine learning model training; Preferably, the data preprocessing includes special data processing and data standardization; Preferably, the special data processing includes any one or a combination of at least two of the following: missing data processing, duplicate data processing, and abnormal data processing. Preferably, the data standardization method includes any one or a combination of at least two of the following: standard deviation standardization, range standardization, robust standardization, and interval scaling. Preferably, the algorithm used in the feature engineering includes any one or a combination of at least two of the following: principal component analysis, linear discriminant analysis, singular value decomposition, local linear embedding, random nearest neighbor embedding, t-distributed neighborhood embedding algorithm, autoencoder, minimum absolute value convergence and selection operator method, and feature recursive elimination method. Preferably, with the total number of samples in the dataset being 100%, the sample size of the training set is 70-95%. Preferably, the machine learning model training includes any one or at least two of the following: linear regression, multinomial regression, K-nearest neighbor regression, adaptive boosting regression, ridge regression, support vector machine regression, Gaussian process regression, decision tree regression, random forest, gradient boosting tree, extreme gradient boosting, lightweight gradient boosting machine, categorical feature boosting fully connected neural network, recurrent neural network, convolutional neural network, and graph convolutional neural network, with gradient boosting tree and / or extreme gradient boosting being more preferred.

4. The design method according to any one of claims 1-3, characterized in that, The method for judging the matching degree includes: calculating the formula matching index S, wherein the smaller the formula matching index S, the better the matching degree; The formula for calculating the formula matching index S is shown in Equation II: Formula II; In Equation II, N X X represents the property number. i Let X represent the target value of the i-th property. i pre Let f represent the predicted value of the i-th property, g represent the standardization function, and f represent the predicted value of the i-th property. i This represents the weight coefficient of the i-th property.

5. The design method according to claim 4, characterized in that, The design method further includes: obtaining several preset formulas based on the target properties, inputting the several preset formulas into the filler property prediction model for property prediction, and obtaining the property data prediction value of each preset formula; Substituting the predicted value of the property data into Equation II, we obtain the formula matching index S for each preset formula. Based on the formula matching index S, we sort several preset formulas and obtain at least one formula with good matching degree as the target formula for the bottom filler adhesive for encapsulation.

6. The design method according to any one of claims 1-5, characterized in that, The design method includes the following steps: S1. Obtain the database, which includes the underfiller formulation and underfiller properties; S2. Based on the database, using the underfiller formulation as the input parameter and the underfiller properties as the output parameter, establish and train a predictive model for the physical properties of the underfiller. The establishment and training of the filler property prediction model includes: Data preprocessing includes special data processing and data standardization; Feature engineering; The dataset is divided into training and validation sets; and Machine learning model training; The trained prediction model for the physical properties of the filler glue satisfies the following: the correlation coefficient R on the validation set. 2 >88%, and / or, the ratio of the root mean square error of the verification set to the mean of the target property is <10%; S3. Obtain several preset formulas according to the target properties, input the several preset formulas into the filler property prediction model to predict the properties, and obtain the property data prediction value of each preset formula respectively; S4. Calculate the formula matching index S based on the predicted values ​​of the property data and the target properties. The formula matching index S is calculated using the formula II shown in Equation II: Formula II; In Equation II, N X X represents the property number. i Let X represent the target value of the i-th property. i pre Let f represent the predicted value of the i-th property, g represent the standardization function used to scale the value to the range of 0-1; i This represents the weight coefficient of the i-th property; Based on the formula matching index S, several preset formulas are sorted to obtain at least one formula with a small formula matching index S and a good matching degree as the target formula for the bottom filler adhesive for encapsulation.

7. The design method according to any one of claims 1-6, characterized in that, The design method further includes: obtaining the measured values ​​of the property data of the preset formula, and determining whether to perform iteration based on the deviation between the measured values ​​of the property data and the predicted values ​​of the property data; The iterative method includes: The measured values ​​of the aforementioned properties are incorporated into the database for data updates, and the filler property prediction model is retrained based on the updated database. Based on the target properties, the preset formula is obtained again and input into the retrained filler property prediction model to obtain new property data prediction values; Based on the matching degree between the predicted values ​​of the new property data and the target properties, a new target formulation for the underfill adhesive for encapsulation is obtained; Preferably, if the relative error between the measured value and the predicted value of the property data is greater than 30%, then iteration is performed.

8. A design device for an encapsulating underfill adhesive, characterized in that, The design device includes: The database module is used to obtain and / or store underfill formulations and underfill properties; The prediction model generation module is used to establish and train a prediction model of the physical properties of the filler using the filler formulation as input parameters and the filler properties as output parameters. The recipe generation module is used to generate preset recipes; and, The formulation design module is used to input the preset formulation into the filler property prediction model to obtain property data prediction values, and to obtain the target formulation of the bottom filler for encapsulation based on the matching degree between the property data prediction values ​​and the target properties.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the design method of the underfill adhesive for encapsulation as described in any one of claims 1-7.

10. An electronic device, characterized in that, The electronic device includes: Memory; Processor; and, A computer program stored in the memory and executable on a processor; When the processor executes the computer program, it implements the design method for underfill adhesive for encapsulation as described in any one of claims 1-7.

Images