Low-temperature Anode Bonding Multi-objective Process Modeling And Parameter Intelligent Optimization

Anode bonding has the advantages of simple process and good bonding quality,and is widely used in packaging processes such as MEMS devices and integrated circuits.The conventional anodic bonding process needs to be completed under high temperature conditions,causing thermal stress matching problems due to different thermal expansion coefficients of materials,affecting the performance of bonded components,especially high-performance components.Therefore,low-temperature anodic bonding has been the most recent One of the research hotspots.Plasma-based surface modification is one of the important process approaches for low-temperature anodic bonding,and has great development prospects.However,due to the problems of high-dimensional and strong dispersion of the process parameters of the low-temperature anode bonding experiment,it is difficult to obtain perfect process optimization results.Therefore,the process modeling and parameter optimization of low temperature anode bonding have become a difficult problem to be solved.Firstly,for low-temperature anodic bonding process,the characteristics of the process parameters are analyzed,and it is found that the low-temperature anodic bonding process parameters have high dimensionality,wide probability distribution range,and complicated coupling.Less amount;analysis of the applicability of conventional process modeling methods,preferably three models of back propagation neural network,support vector machine and extreme learning machine suitable for low temperature anode bonding process modeling;optimization and verification of process models Through experimental analysis,it was found that the conventional single process modeling method solves the shortcomings of the low-temperature anode bonding process modeling process,and it is verified that the multi-model combination modeling can better solve the problem of modeling low-temperature anode bonding at different process parameter intervals.Then build and verify the intelligent decision-making strategy for low temperature anode bonding multi-model combination process modeling.In order to overcome the influence of the evaluation measure on the evaluation due to measurement errors andinterference saturation,the four sets of evaluation criteria were parameterized and the model evaluation measure was constructed;the Bayesian multi-arm slot machine algorithm was used to train the parameters of the evaluation criteria based on Low-temperature anodic bonding data was used for Monte Carlo Bayesian posterior mining,and the expected values ??of the highest posterior probability distribution of the parameters were: 0.60,0.69,0.75,and 0.86;the method of multi-model switching function in Bayesian probability space was analyzed,To build an intelligent decision algorithm based on Bayesian probability,experimental analysis found that the built intelligent strategy has a better model selection decision at each sample point of low temperature anode bonding test sequence,which verified the feasibility of multi-model adaptive optimization.Then build a low-temperature anode bonding process modeling system.Constructed a framework and platform for mathematical statistical modeling of low-temperature anode bonding process and intelligent parameter optimization system;based on the characteristics of low-temperature anode bonding process parameters,introduced three sets of nonlinear consensus equations with the same characteristics in different fields(including: physical field: gravitational equation,Mass-energy formula;biological field: Mie equation)to train multi-model parameters;and experimental verification using the non-linear test function Ma F15,the results show that within 95% confidence interval,the built multi-model is high-dimensional,coupled,Normal distribution and other nonlinear problems have high prediction accuracy and fast intelligent decision-making,which proves the reliability of multiple models.Finally,the low-temperature anode bonding process optimization and experimental verification were carried out.Design low-temperature anodic bonding process experiment,determine the sample set collection plan,establish the sample set,and use the Bayesian average model,Gaussian process regression model and multi-model method to compare the low-temperature anodic bonding process experiment,from the confidence and performance evaluation indicators And reliability analysis,the results show that: in the surface treatment stage,the performance of the multi-model is 55.56% and 66.67% higher than the Bayesian average model and the Gaussian process regression model;in the anodic bonding stage,the performance of the multi-model They are respectively 37.5% and 87.5%higher than the Bayesian average model and Gaussian process regression model.And the multi-model method has the best reliability,which improves the stability oflow-temperature anodic bonding process modeling.After that,the optimal process parameters of low-temperature anodic bonding are obtained by the step size method,and verified with actual experiments,which proves the feasibility of the multi-model modeling method.