Method, system and application of designing high glass transition temperature die attach adhesive based on machine learning

Abstract

The application belongs to the technical field of machine learning and chip manufacturing, and discloses a method, system and application for designing a high glass transition temperature chip underfill adhesive based on machine learning, which takes the high glass transition temperature Tg of an epoxy resin-curing agent system as the main target parameter, and comprises the following steps: S1, constructing polymer molecular structure parameters and process parameters; S2, screening a key feature set; S3, machine learning prediction; and S4, obtaining an overall optimization scheme. Experimental verification shows that the measured Tg of the preferred formula (such as a three-functional epoxy / diphenylolpropane epoxy / naphthalene anhydride compound system) obtained by the application is 205-215 DEG C, which is more than 20% higher than that of a traditional formula, and the model prediction error is less than ± 6 DEG C. The application solves the bottleneck of the trial-and-error method for designing the Tg of an epoxy resin matrix in an underfill adhesive, can shorten the development cycle by more than 60%, is suitable for high-reliability packaging scenarios such as 5G chips and power devices, and has a wide range of applications.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of machine learning and chip manufacturing technology, in particular to a method, system and application of designing a high glass transition temperature chip underfill adhesive based on machine learning. BACKGROUND

[0002] In the field of high-density packaging of integrated circuits, the chip underfill adhesive, as a key interfacial material, needs to maintain dimensional stability and mechanical strength during high-temperature reflow soldering (peak temperature 260-280℃) and long-term service, and its glass transition temperature (Tg) directly determines the thermal fatigue resistance of the packaging structure. Epoxy resin has become the mainstream matrix material for underfill adhesive due to its excellent adhesion, chemical resistance and controllable curing network. In traditional technology, the Tg of epoxy resin is improved in two ways. On the one hand, the molecular structure of monomers is designed, such as introducing multifunctional epoxy monomers (such as tetrafunctional TGDDM, hexafunctional HP-7200) to increase crosslinking density, or using rigid aromatic ring structure monomers (such as biphenyl epoxy, naphthalene ring modified epoxy) to limit molecular chain movement. For example, the Tg of biphenyl epoxy (BPA type) can be increased by about 30-50℃ compared with standard bisphenol A epoxy (DGEBA), but high functionality monomers can easily lead to a sharp increase in system viscosity, affecting filler dispersion and flow filling. On the other hand, by selecting appropriate curing agents or modified curing agents, such as using high reactivity aromatic amines (such as 4,4'-diaminodiphenyl sulfone DDS) or rigid anhydrides (such as methyl nadic anhydride MNA) as curing agents, the Tg is increased by enhancing the interaction between molecular chains. However, the matching degree of the molecular weight distribution, steric hindrance effect (such as ortho-substituted groups) of the curing agent and the epoxy monomer needs to be verified repeatedly by experiments, and high rigidity curing agents often accompany with increased curing stress, leading to the risk of interfacial debonding. Finally, by using a multi-stage temperature program (such as 120℃ preheating-170℃ main curing-200℃ post-curing) to promote the uniformization of the crosslinked network and reduce the residual unreacted groups. However, the process parameters (temperature gradient, holding time, pressure) and material composition (monomer / curing agent ratio, catalyst content) have a strong coupling effect, and the traditional trial-and-error method is difficult to analyze the synergistic mechanism.

[0003] For example, CN201711069729.4 discloses a high-efficiency high-temperature-resistant heat-conducting underfill adhesive and a preparation method thereof, which is a single-component epoxy resin adhesive. The glass transition temperature of the adhesive is usually lower than 130℃, and the glass transition temperature fluctuates greatly between different embodiments, which is unstable, and thus it is difficult to balance the flowability, viscosity and high-temperature resistance of the material.

[0004] Therefore, the prior art has many bottlenecks, for example: there is a lack of analytical model for the quantitative influence of molecular parameters such as the number of functional groups and the density of aromatic rings on Tg, design relies on empirical rules (such as "rigid group ratio > 40%"), it is difficult to accurately guide molecular modification; the research and development efficiency is low, formula screening requires hundreds of experiments, the development cycle is as long as 6-12 months, and it is difficult to balance Tg and process properties (such as viscosity and curing rate); in addition, the calculation cost is very high, although molecular dynamics (MD) or first principle simulation can predict the crosslinking network formation process, the calculation resource consumption of complex multi-component systems is great, and high-throughput design cannot be realized. SUMMARY

[0005] The purpose of the present application is to provide a method, system and application for designing high glass transition temperature chip underfill adhesive based on machine learning, which combines machine learning with components, proportions, processes and other specific technical fields of chip manufacturing underfill adhesive, uses machine learning algorithm to analyze the influence of epoxy resin monomer and curing agent molecular structure characteristics on glass transition temperature (Tg), and applies it to design high Tg temperature chip underfill adhesive, to obtain an efficient design method and system that combines multi-scale features and has prediction accuracy and physical interpretation ability, which breaks through the technical difficulties of Tg optimization of epoxy resin matrix, realizes efficient optimization of complex formula and process under multiple constraints, and is easy to realize high-throughput design, to solve the above technical problems.

[0006] In order to achieve the above purpose, the technical scheme provided by the present application is:

[0007] The method for designing high glass transition temperature chip underfill adhesive based on machine learning takes high glass transition temperature Tg of epoxy resin-curing agent system as the main target parameter, which includes the following steps:

[0008] S1: Constructing polymer molecular structure parameters and process parameters

[0009] Using a public polymer database, the molecular structure parameters of epoxy resin monomers and curing agents and the corresponding Tg measured values are obtained, the molecular structure parameters include the number of functional groups of epoxy monomers (nEpoxy), the aromatic ring density (Ar-density), the molecular weight (MW), and the amine hydrogen equivalent (AHE) and the steric hindrance parameter (Steric-index) of the curing agent, to obtain the corresponding molecular parameter descriptor set and the process parameter descriptor set;

[0010] S2: Screening key feature set

[0011] Recursive feature elimination (RFE) combined with XGBoost algorithm is used to screen the key feature set from the molecular parameter descriptor set and the process parameter descriptor set, wherein the process parameters include curing ladder temperature and curing time;

[0012] S3: Machine learning-based prediction

[0013] Based on the screened feature set, a Gaussian process regression (GPR) model is constructed to predict the Tg of the epoxy resin-curing agent system, and a candidate scheme including components, ratios, and process parameters is obtained; then the SHAP method is used to analyze the contribution of each feature to Tg, and the candidate components, ratios, and process parameters are optimized.

[0014] S4: Obtain the overall optimization scheme

[0015] With Tg≥180℃ and viscosity≤5000cP as the constraint conditions, a Bayesian optimization algorithm is used to search for the Pareto optimal solution in the epoxy monomer-curing agent ratio space, and an optimization scheme of the epoxy resin-curing agent system with high glass transition temperature as the target parameter is obtained, including the overall optimization scheme of the components, ratios, and process parameters.

[0016] A system for designing a high glass transition temperature chip underfill adhesive based on machine learning, which is used to implement the aforementioned method, comprising:

[0017] 1) Data input module: used for importing the molecular structure of epoxy monomer and curing agent and process parameters;

[0018] 2) Feature engineering module: perform RFE-XGBoost feature screening to generate a key feature set;

[0019] 3) Model calculation module: run the GPR model to predict Tg, and output the feature contribution degree through the SHAP interpreter;

[0020] 4) Optimization output module: generate a combination of formulations and process parameters that meet the Tg and viscosity constraints based on Bayesian optimization.

[0021] The optimization output module is connected to the interface of the molecular dynamics simulation software to verify the crosslinking network morphology of the predicted formulation, and outputs a candidate system with a crosslinking degree≥85%.

[0022] A high glass transition temperature chip underfill adhesive, characterized in that it is designed by the aforementioned method and is made of components in the following proportions: bisphenol A type epoxy (DGEBA) 40-55wt%, tetrafunctional epoxy (TGDDM) 15-35wt%, naphthalene ring modified epoxy (NE) 5-10wt%, and 20-30wt% boron nitride (BN) as a thermal conductive filler, the glass transition temperature Tg≥195℃, the viscosity≤4500cP (25℃); its curing process is three-stage gradient heating: 120℃ / 1h-170℃ / 3h-200℃ / 1h, the heating rate is 2 ℃ / min.

[0023] Compared with the prior art, the present application at least includes the following beneficial effects:

[0024] 1. The present application provides a machine learning-based design method and system for epoxy resin-curing agent system chip underfill adhesive with high glass transition temperature (Tg) as the core target parameter. The method mainly quantifies the influence of molecular structure and process parameters on Tg through machine learning, realizes the generation of candidate schemes and subsequent efficient optimization. The present application quickly predicts and optimizes the Tg of epoxy resin matrix through machine learning model, avoids a large number of repeated experiments of traditional trial-and-error method, significantly shortens the research and development cycle, and improves the research and development efficiency.

[0025] 2. The machine learning of the multi-parameter and multi-constraint system adopted by the present application reduces the number of experiments and material consumption, reduces the research and development cost, avoids the waste of resources caused by improper experimental conditions, can fully explore the complex formula and process parameter space, and screen out the optimal formula and process parameter combination, significantly improve the glass transition temperature of the chip underfill adhesive, enhance its thermal stability and reliability, and meet the demand of chip packaging for high-performance underfill adhesive. The method and system provided by the present application collect and analyze a large amount of experimental data, build a machine learning model, provide a method for predicting the glass transition temperature of epoxy resin, can accelerate the discovery of new high glass transition temperature epoxy resin, and significantly improve the glass transition temperature of the chip underfill adhesive.

[0026] 3. The present application is verified by actual experiments and shows that the measured Tg of the preferred formula (such as trifunctional epoxy / diphenyl A epoxy / naphthalene anhydride complex system) obtained by the present application is 205-215℃, which is more than 20% higher than that of the traditional formula, and the model prediction error is <±6℃.

[0027] 4. The present application adopts a new technical idea, uses RFE nested XGBoost to screen features, uses XGBoost algorithm to build a model, uses SHAP method to explain the relationship between structure and performance, and predicts high Tg value epoxy resin; the method and system have low running cost, are simple and efficient, have complete and accurate data, and the whole process is environmentally friendly and pollution-free, solve the bottleneck of epoxy resin matrix Tg design relying on trial-and-error method, can shorten the research and development cycle by more than 60%, and are suitable for 5G chip, power device and other high reliability packaging scenes.

[0028] 5. The whole design process of the present application does not involve complex chemical experiments and high-energy consumption equipment, conforms to the development concept of green environmental protection, and is easy to realize large-flux design and industrialized preparation. BRIEF DESCRIPTION OF DRAWINGS

[0029] Figure 1 The system composition and method flowchart for predicting and optimizing the glass transition temperature of epoxy resin of the present application embodiment. DETAILED DESCRIPTION

[0030] The application will be further described in details below with reference to the accompanying drawings and examples, so that those skilled in the art can implement the application according to the description.

[0031] It should be understood that the terms such as "have", "contain" and "include" used herein do not exclude the presence or addition of one or more other elements or combinations thereof.

[0032] Unless otherwise specifically stated, the various raw materials, reagents, instruments and equipment used in the application can be purchased on the market or can be prepared by existing methods.

[0033] Basic embodiment

[0034] Referring to the accompanying drawings, Figure 1 The method for designing a high glass transition temperature chip underfill adhesive based on machine learning provided by the embodiment of the application takes the high glass transition temperature Tg of an epoxy resin-curing agent system as the main target parameter, and includes the following steps:

[0035] S1: Constructing polymer molecular structure parameters and process parameters

[0036] Using a public polymer database, the molecular structure parameters of epoxy resin monomers and curing agents and the corresponding Tg measured values are obtained, the molecular structure parameters include the number of functional groups of epoxy monomers (nEpoxy), the aromatic ring density (Ar-density), the molecular weight (MW), and the amine hydrogen equivalent (AHE) and the steric hindrance parameter (Steric-index) of the curing agent, to obtain the corresponding molecular parameter descriptor set and the process parameter descriptor set;

[0037] S2: Screening a key feature set

[0038] Recursive feature elimination (RFE) combined with XGBoost algorithm is used to screen a key feature set from the molecular parameter descriptor set and the process parameter descriptor set, wherein the process parameters include curing ladder temperature and curing time;

[0039] S3: Machine learning-based prediction

[0040] Based on the screened feature set, a Gaussian process regression (GPR) model is constructed to predict the Tg of the epoxy resin-curing agent system, to obtain candidate schemes including components, ratios and process parameters; and SHAP method is used to analyze the contribution of each feature to the Tg, to optimize the candidate components, ratios and process parameters;

[0041] S4: Obtaining an overall optimization scheme

[0042] With Tg≥180℃ and viscosity≤5000cP as constraint conditions, the Bayesian optimization algorithm is used to search for Pareto optimal solution in the space of epoxy monomer-curing agent ratio, and the optimization scheme of the epoxy resin-curing agent system with high glass transition temperature as the target parameter is obtained, including the overall optimization scheme of components, ratio and process parameters.

[0043] The epoxy resin monomer in the step S1 is at least one of bisphenol A type epoxy (DGEBA), tetrafunctional epoxy (TGDDM), and naphthalene ring modified epoxy, and the compounding ratio meets: 40-70wt% of difunctional epoxy and 30-60wt% of multifunctional epoxy.

[0044] The curing agent in the step S1 is an aromatic amine or an anhydride compound, and the steric hindrance parameter Steric-index is obtained by calculating the van der Waals radius superposition volume through molecular dynamics simulation, and the value range is 0.6-0.9.

[0045] The curing agent is selected from at least one of 4,4'-diamino diphenyl sulfone (DDS), methyl nadic anhydride (MNA), and benzophenonetetracarboxylic dianhydride (BTDA), and the amine hydrogen equivalent (AHE) is 80-120g / eq.

[0046] The optimization of process parameters in the step S3 includes: dividing the step curing program into 3-5 stages, the initial curing temperature is 80-120℃, the final curing temperature is 160-200℃, and the total curing time is 2-8h.

[0047] In the SHAP analysis process of the step S3, the total contribution weight of the number of epoxy monomer functional groups (nEpoxy) and the aromatic ring density (Ar-density) is≥60%, and the absolute value of the negative weight of the contribution degree of the steric hindrance parameter (Steric-index) of the curing agent is≤15%.

[0048] A system for designing a high glass transition temperature chip underfill adhesive based on machine learning, which is used to implement the aforementioned method, comprising:

[0049] 1) Data input module: for importing the molecular structure of epoxy monomer, curing agent and process parameters;

[0050] 2) Feature engineering module: perform RFE-XGBoost feature selection to generate a key feature set;

[0051] 3) Model calculation module: run GPR model to predict Tg, and output feature contribution degree through SHAP interpreter;

[0052] 4) Optimization output module: generate the combination of formula and process parameters that meet the Tg and viscosity constraints based on Bayesian optimization.

[0053] The optimization output module is connected with a molecular dynamics simulation software, and crosslinking network morphology verification is performed on the predicted formula, and a candidate system with a crosslinking degree of ≥85% is output.

[0054] A high glass transition temperature chip underfill adhesive is designed by the method, and is made of components in the following proportions: bisphenol A type epoxy (DGEBA) 40-55wt%, tetrafunctional epoxy (TGDDM) 15-35wt%, naphthalene ring modified epoxy (NE) 5-10wt%, and 20-30wt% boron nitride (BN) is added as a thermal conductive filler, the glass transition temperature Tg≥195℃, the viscosity ≤4500cP (25℃); the curing process is three-stage gradient heating: 120℃ / 1h-170℃ / 3h-200℃ / 1h, the heating rate is 2℃ / min.

[0055] The above technical solutions are further described in combination with specific example embodiments, and the preferred embodiments of the application are described in detail as follows:

[0056] Example 1

[0057] The method, system and application for designing a high glass transition temperature chip underfill adhesive based on machine learning provided by the embodiments of the application are specific to the foregoing basic embodiments, and differ in that:

[0058] The method for designing a high glass transition temperature chip underfill adhesive based on machine learning is to take the high glass transition temperature Tg of the epoxy resin-curing agent system as the main target parameter, collect and preprocess the data of factors affecting the predicted glass transition temperature of the epoxy resin, and then perform prediction, analysis and optimization, and the step S1 specifically includes the following steps:

[0059] (1) 200 groups of epoxy resin-curing agent system data are obtained from PolyInfo and experiments, covering bisphenol A type epoxy (DGEBA), tetrafunctional epoxy (TGDDM), naphthalene ring modified epoxy (NE), etc. monomers, and curing agents DDS, MNA, BTDA, etc. Contain Tg (120-250℃, DMA test, tanδ peak value method), record viscosity (25℃) as process constraint at the same time;

[0060] (2) Use ChemAxon to optimize the molecular structure, and extract descriptors for the epoxy monomer and curing agent. For the epoxy monomer, the number of functional groups of DGEBA, TGDDM and NE is set to 2, 4 and 3 respectively; the aromatic ring density is determined by the number of aromatic ring atoms per molecule, for example, DGEBA = 0.032 atoms / g / mol; the polar surface area (TPSA) is calculated by Dragon software (e.g. DGEBA = 45.2 Ų); the amine hydrogen equivalent (AHE) of curing agents DDS and MNA is 107 g / eq and 178 g / eq respectively; the steric hindrance parameter is based on the calculation of atomic packing volume (DDS = 0.72, MNA = 0.65) by molecular dynamics simulation;

[0061] (3) Collect process parameters, such as: the number of step-up temperature segments is 2-5 segments, the final segment temperature is set to 160-220℃, and the curing time is set to 1-6h;

[0062] (4) Collect the filler information, mainly boron nitride filler, content is 20-40wt%, particle size is 1-10μm.

[0063] (5) The first specific raw material component and process of the underfill adhesive are: bisphenol A type epoxy (DGEBA) 50wt%, tetrafunctional epoxy (TGDDM) 22wt%, naphthalene ring modified epoxy (NE) 8wt%, and 20wt% boron nitride (BN) is added as a thermal conductive filler, the measured Tg≥ 195℃, the viscosity ≤ 4500cP (25℃); its curing process is three-stage gradient heating: 120℃ / 1h-170℃ / 3h-200℃ / 1h, the heating rate is 2 ℃ / min;

[0064] The epoxy resin monomer in the embodiment is bisphenol A type epoxy (DGEBA), and the compounding ratio meets: the proportion of bifunctional epoxy is 40wt%, and the proportion of multifunctional epoxy is 60wt%; the compounding ratio of other embodiments can also meet: the proportion of bifunctional epoxy is 50wt%, and the proportion of multifunctional epoxy is 50wt%; or the proportion of bifunctional epoxy is 70wt%, and the proportion of multifunctional epoxy is 30wt%.

[0065] Embodiment 2

[0066] The method, system and application for designing high glass transition temperature chip underfill adhesive based on machine learning provided by the embodiment of the application are specific to the foregoing basic embodiment, which are basically the same as embodiment 1, and the difference is that the step S2 of screening the key feature set further includes the following steps:

[0067] The model training and verification for predicting the glass transition temperature of the epoxy resin include the following steps:

[0068] (1) 85 features including epoxy monomers, curing agents and process parameters;

[0069] (2) Top 8 key features are: functional group number, aromatic ring density, polar surface area, amine hydrogen equivalent, steric hindrance, final temperature, curing time and boron nitride content, which are screened by recursive feature elimination (RFE) combined with XGBoost algorithm;

[0070] (3) Gaussian process regression (GPR) is used, and the kernel function is Matern 5 / 2;

[0071] (4) 160 groups (80%) of training set and 40 groups (20%) of test set are divided, and 5-fold cross-validation is used;

[0072] (5) The R² of the model training set is 0.93, and the MAE is 5.8℃; the R² of the test set is 0.89, and the MAE is 7.1℃;

[0073] (6) SHAP value analysis is used for explainability analysis, wherein the positive contributions include functional group number (+42%), aromatic ring density (+28%) and final temperature (+15%), and the negative contributions include steric hindrance parameter (-10%, which leads to crosslinking hindrance), and the Tg is increased by 32±5℃ (confidence 95%) on average for each increase of 1 of the functional group of the epoxy monomer.

[0074] (7) The second specific raw material component and process of the underfill adhesive are obtained: bisphenol A type epoxy (DGEBA) 40wt%, tetrafunctional epoxy (TGDDM) 35wt%, naphthalene ring modified epoxy (NE) 5wt%, and 20wt% boron nitride (BN) is added as a heat conducting filler, the measured Tg is ≥195℃, and the viscosity is ≤4500cP (25℃); the curing process is three-stage gradient heating: 120℃ / 1h-170℃ / 3h-200℃ / 1h, and the heating rate is 2℃ / min.

[0075] Example 3

[0076] The method, system and application for designing high glass transition temperature chip underfill adhesive based on machine learning provided by the embodiments of the application are specific to the foregoing basic embodiments, which are basically the same as embodiments 1 and 2, and the difference lies in that the formula optimization and experimental verification of predicting the glass transition temperature of the epoxy resin are further included in step S3 based on machine learning prediction, and the step S3 based on machine learning prediction further includes the following steps:

[0077] (1) The target parameter is optimized to maximize Tg (≥180℃) with the constraint of viscosity ≤5000cP (25℃);

[0078] The filler boron nitride content is fixed at 35wt% (particle size 5μm);

[0079] (2) The optimal formulation using Bayesian optimization is that the epoxy monomer component is 40% TGDDM + 30% NE + 30% DGEBA, the curing agent component is DDS (amine hydrogen equivalent weight is 107 g / eq, and the steric hindrance parameter is 0.72), and the process parameter is three-stage curing (120 DEG C / 1 h→ 170 DEG C / 3 h→ 200 DEG C / 1 h);

[0080] (3) According to the above prediction, the predicted Tg is 218 DEG C ± 6 DEG C, and the viscosity is 4800 cP ± 300 cP.

[0081] (4) The epoxy monomer and DDS are mixed according to the ratio, the BN filler is added, vacuum degassing is performed for 30 min, the bottom filling adhesive sample is prepared according to the optimized process ladder curing, the actual measured Tg is 215 DEG C (DMA, tan delta peak value), the viscosity is 4650 cP (25 DEG C, Brookfield DV2T), and the prediction error is Tg deviation -1.4% (<5% threshold, the model is effective);

[0082] (5) Compared with the traditional sample DGEBA 100% + MNA (Tg=165 DEG C), the optimized formulation is improved by 30.3%.

[0083] (6) A third specific raw material component and process of the bottom filling adhesive are obtained: bisphenol A type epoxy (DGEBA) 40 wt%, tetrafunctional epoxy (TGDDM) 20 wt%, naphthalene ring modified epoxy (NE) 10 wt%, and 30 wt% boron nitride (BN) is added as a heat conducting filler, the actual measured Tg is greater than or equal to 195 DEG C, and the viscosity is less than or equal to 4500 cP (25 DEG C); and the curing process is three-stage gradient heating: 120 DEG C / 1 h-170 DEG C / 3 h-200 DEG C / 1 h, and the heating rate is 2 DEG C / min.

[0084] Example 4

[0085] The method, system and application for designing a high glass transition temperature chip bottom filling adhesive based on machine learning provided by the embodiment of the application are specific to the foregoing basic embodiment, which are basically the same as examples 1-3, and the difference lies in that the model robustness verification step for predicting the glass transition temperature of the epoxy resin comprises the following contents:

[0086] (1) Input a new type of epoxy monomer (pentafunctional PX-100, nEpoxy=5, Ar_density=0.045), predict Tg=242 DEG C; actual measured Tg=230 DEG C, error 4.9% (verify model generalization ability).

[0087] (2) Curing end temperature fluctuation ±10°C (190-210°C): predicted Tg variation ±8°C, measured ±7-9°C; model accurately captures process sensitivity, guiding production line temperature control precision (required ±5°C).

[0088] (3) Fourth specific raw material component and process for obtaining the underfill adhesive: bisphenol A type epoxy (DGEBA) 55wt%, tetrafunctional epoxy (TGDDM) 15wt%, naphthalene ring modified epoxy (NE) 7.5wt%, and 22.5wt% boron nitride (BN) added as a thermal conductive filler, measured Tg≥ 195°C, viscosity ≤ 4500 cP (25°C); its curing process is three-stage gradient heating: 120°C / 1h-170°C / 3h-200°C / 1h, heating rate 2°C / min.

[0089] Example 5

[0090] In this embodiment, on the basis of Examples 1-4, further feedback iteration steps for predicting the glass transition temperature of epoxy resins are provided, including the following contents:

[0091] (1) Input (50%) NE + (Steric_index = 0.65) MNA, predicted Tg = 188°C; measured Tg = 162°C, error 14.3%, triggering feature re-screening;

[0092] (2) According to the above analysis, the reason may be that the compatibility (polarity difference) of MNA and NE is not considered, a new feature "epoxy-curing agent polarity matching degree" is added; after subsequent iteration, the model R² is improved to 0.91, MAE = 6.2°C.

[0093] In summary, the method provided by the above embodiments for predicting and optimizing, verifying the glass transition temperature of epoxy resins includes the following steps:

[0094] (1) Using the public polymer database, the molecular structure parameters of epoxy resin monomers and curing agents and the corresponding measured Tg values are obtained, the molecular structure parameters include the number of functional groups of epoxy monomers (nEpoxy), the aromatic ring density (Ar-density), the molecular weight (MW), and the amine hydrogen equivalent (AHE) of the curing agent, the steric hindrance parameter (Steric-index);

[0095] (2) Recursive feature elimination (RFE) combined with XGBoost algorithm is used to screen the key feature set from the molecular structure parameters and process parameters, the process parameters include curing ladder temperature, curing time;

[0096] (3) Based on the screened feature set, a Gaussian process regression (GPR) model is constructed to predict the Tg of the epoxy resin-curing agent system, and the SHAP method is used to analyze the contribution of each feature to the Tg;

[0097] (4) Taking Tg≥180℃ and viscosity≤5000cP as constraint conditions, a Bayesian optimization algorithm is used to search for Pareto optimal solution in the epoxy monomer-curing agent ratio space; the epoxy resin monomer is at least one of bisphenol A type epoxy (DGEBA), tetrafunctional epoxy (TGDDM), and naphthalene ring modified epoxy, and the compounding ratio meets: the proportion of difunctional epoxy is 40-70wt%, and the proportion of multifunctional epoxy is 30-60wt%;

[0098] The curing agent is an aromatic amine or an anhydride compound, and the steric index is obtained by calculating the van der Waals radius superposition volume through molecular dynamics simulation, and the value range is 0.6-0.9; the curing agent is selected from at least one of 4,4'-diaminodiphenyl sulfone (DDS), methyl nadic anhydride (MNA), and benzophenonetetracarboxylic dianhydride (BTDA), and the amine hydrogen equivalent (AHE) is 80-120g / eq; the process parameter optimization includes: the step curing procedure is divided into 3-5 stages, the initial curing temperature is 80-120℃, the final curing temperature is 160-200℃, and the total curing time is 2-8h;

[0099] (5) In the SHAP analysis, the total contribution weight of the number of epoxy monomer functional groups (nEpoxy) and the aromatic ring density (Ar-density) is≥60%, and the absolute value of the negative weight of the steric index of the curing agent (Steric-index) is≤15%.

[0100] The design method provided by each of the above embodiments avoids repeated experiments and the process of trial and error, uses Chem3D to optimize molecules, and uses ChemAxon software to generate structure descriptors, which are screened and modeled through RFE nested XGBoost method, so that the Tg value of the epoxy resin can be predicted in advance, the operation process is simple, the cost is low, and there is no pollution, and only one person is needed to complete the operation; The features are explained by SHAP and reference literature in the above embodiment method, and a QSPR model is constructed, which better explains the relationship between molecular structure and performance, and tries to solve the shortcomings of most black box methods; The above embodiment method uses XGBoost to establish a model for quickly predicting the Tg value of the epoxy resin, which can greatly shorten the research and development time of new epoxy resins and reduce the research and development cost.

[0101] The above embodiments of the present application focus on: first, constructing a set of molecular descriptors of epoxy monomers (number of functional groups, density of aromatic rings) and curing agents (amine hydrogen equivalent, steric hindrance parameter); then using recursive feature elimination (RFE) combined with XGBoost algorithm to screen key features, establishing a Gaussian process regression (GPR) model to predict Tg, and analyzing the structure contribution degree by SHAP method; then combining Bayesian optimization to screen candidate systems with high Tg (> 180℃) and low viscosity (< 5000cP) in the epoxy-curing agent ratio and ladder curing process coordination space; finally, the scheme is optimized and verified.

[0102] Experimental verification shows that the measured Tg of the preferred formula (such as a three-functional epoxy / diphenyl A epoxy / naphthalene anhydride anhydride compound system) reaches 205-215℃, which is more than 20% higher than that of the traditional formula, and the model prediction error is < ± 6℃.

[0103] The above embodiments of the present application respectively successfully design an epoxy resin-based chip underfill adhesive with high glass transition temperature through machine learning technology, significantly improve the thermal stability and reliability of the underfill adhesive, provide a new idea and method for the research and development of high-performance chip packaging materials, solve the bottleneck of epoxy resin matrix Tg design relying on trial and error method, and can shorten the research and development period by more than 60%, suitable for high-reliability packaging scenarios such as 5G chips and power devices. The present application not only improves the research and development efficiency and reduces the research and development cost, but also meets the development concept of green environmental protection, and has wide application prospect.

[0104] It should be particularly pointed out that other technical solutions obtained by specific selection within the scope of the components, ratios and process parameters described in the present application can achieve the technical effects of the present application, so they will not be listed one by one. Other technical solutions obtained by using components, ratios, preparation methods and applications equivalent to those described in the present application are included in the protection scope of the present application.

[0105] The above only describes the preferred embodiments of the present application and is not intended to limit the present application. For those skilled in the art, the present application can have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for designing high glass transition temperature die attach adhesive based on machine learning, characterized in that, The high glass transition temperature Tg of the epoxy resin-curing agent system is taken as the main target parameter, and the method comprises the following steps: S1: constructing polymer molecular structure parameters and process parameters Using a public polymer database, the molecular structure parameters of epoxy resin monomers and curing agents and the corresponding Tg measured values are obtained, the molecular structure parameters include the number of functional groups nEpoxy of epoxy monomers, the aromatic ring density Ar-density, the molecular weight MW, and the amine hydrogen equivalent AHE and the steric hindrance parameter Steric-index of the curing agent, to obtain the corresponding molecular parameter descriptor set and the process parameter descriptor set; S2: screening a key feature set Recursive feature elimination RFE combined with XGBoost algorithm is used to screen a key feature set from the molecular parameter descriptor set and the process parameter descriptor set, wherein the process parameters include curing ladder temperature and curing time; S3: machine learning-based prediction Based on the screened feature set, a Gaussian process regression GPR model is constructed to predict the Tg of the epoxy resin-curing agent system, and a candidate scheme including components, ratios and process parameters is obtained; SHAP method is used to analyze the contribution of each feature to Tg, and the candidate components, ratios and process parameters are optimized; S4: obtaining an overall optimization scheme Under the constraint conditions of Tg≥180℃ and viscosity≤5000cP, a Bayesian optimization algorithm is used to search for Pareto optimal solutions in the epoxy monomer-curing agent ratio space, and an optimization scheme of the epoxy resin-curing agent system with high glass transition temperature as the target parameter is obtained, including the overall optimization scheme of components, ratios and process parameters.

2. The method for designing a chip underfill adhesive with high glass transition temperature based on machine learning according to claim 1, wherein the epoxy resin monomer in step S1 is at least one of bisphenol A type epoxy DGEBA, tetrafunctional epoxy TGDDM, and naphthalene ring modified epoxy, and the compounding ratio satisfies that the double-functional epoxy accounts for 40-70wt%, and the multi-functional epoxy accounts for 30-60wt%.

3. The method for designing a chip underfill adhesive with high glass transition temperature based on machine learning according to claim 1, wherein the curing agent in step S1 is an aromatic amine or an acid anhydride compound, and the steric hindrance parameter Steric-index is obtained by calculating the atomic van der Waals radius superposition volume through molecular dynamics simulation, and the value range is 0.6-0.

9.

4. The method for designing a chip underfill adhesive with high glass transition temperature based on machine learning according to claim 3, wherein the curing agent is at least one of 4,4'-diamino diphenyl sulfone DDS, methyl nadic anhydride MNA, and benzophenonetetracarboxylic dianhydride BTDA, and the amine hydrogen equivalent AHE is 80-120g / eq.

5. The method for designing a chip underfill adhesive with high glass transition temperature based on machine learning according to claim 1, wherein ​ ​ ​ The optimization of the process parameters in the step S3 includes: dividing the step curing program into 3-5 stages, the initial curing temperature is 80-120 DEG C, the final curing temperature is 160-200 DEG C, and the total curing time is 2-8 h.

6. The method for designing high glass transition temperature chip underfill adhesive based on machine learning according to claim 1, wherein, In the SHAP analysis process of the step S3, the total contribution weight sum of the number of epoxy monomer functional groups nEpoxy and the aromatic ring density Ar-density is greater than or equal to 60%, and the absolute value of the negative contribution weight of the steric hindrance parameter Steric-index of the curing agent is less than or equal to 15%.

7. A system for designing high glass transition temperature chip substrate filler based on machine learning, characterized in that, It is used to implement the method of any one of claims 1-6, which comprises: 1) Data input module: for importing the molecular structure of epoxy monomer, curing agent and process parameters; 2) Feature engineering module: perform RFE-XGBoost feature selection to generate a key feature set; 3) Model calculation module: run GPR model to predict Tg, and output feature contribution through SHAP interpreter; 4) Optimization output module: generate the combination of formula and process parameters that meet the Tg and viscosity constraints based on Bayesian optimization.

8. The system for designing high glass transition temperature underfill adhesive for a chip based on machine learning of claim 7, wherein, The optimization output module is connected with the interface of molecular dynamics simulation software to verify the crosslinking network morphology of the predicted formula, and output the candidate system with crosslinking degree greater than or equal to 85%.

Images