Application Research On PSO-SVM Algorithm Model For Component Proportioning Optimization In CMP Polishing Fluid

At present,integrated circuit develops rapidly in accordance with Moore law,the integration of transistors reaches the stage of ultra large scale,the wiring layers exceeds ten layers.Higher requirements and challenges are put forward for the precision of ultra precision planarization technology.CMP technology is the only high precision processing technology that can realize the local and global planarization of the material surface,polishing process is affected by many factors,as an important factor affecting the flatting effect,the polishing fluid affects the chemical action and mechanical action in the process of CMP,to a large extent,determines the polishing efficiency and the surface quality of the polishing material.Therefore,how to optimize the polishing liquid distribution ratio,improve the material removal rate,and obtain high-quality and efficient polishing surface is a hot research topic at the present stage.In this thesis,the research status of CMP technology and the optimization of proportioning of components of CMP polishing liquid and the problems in the optimization proportioning of components of CMP polishing liquid are analyzed,the particle swarm optimization algorithm?PSO algorithm?and the support vector machine?SVM?are selected to set up the PSO-SVM algorithm model.By using the results of the influence of each component of the polishing fluid on the removal rate,an orthogonal optimization experiment is established to get the experimental data set optimization of proportioning of components of CMP polishing liquid.In the optimization experiment,the data are pretreated with the normalization of decimal scaling and Min-Max normalization,with the input value of pH value,H₂O₂ oxidant concentration,FA/O I chelating agent concentration,S_iO₂ abrasive concentration and active agent concentration,and the S_iO₂ removal rate as the output value.The data set is trained by the PSO-SVM algorithm model,and optimization model of proportioning of components of CMP polishing liquid is finally obtained.The prediction of the predicted samples is made by the optimization model.The maximum absolute error is 0.0016 and the mean square error is 0.0001.By comparing with only using SVM model and verifying the optimization model,the results show that the results obtained by the former are more accurate and more accurate than that of the latter.And the prediction error of optimization model of proportioning of components of CMP polishing liquid based on PSO-SVM algorithm is within the allowable scope of the engineering.What's more,aiming at the problem that the decrease of accuracy and reliability caused by temperature drift of pressure sensor,a PSO-SVR temperature compensation model combining particle swarm optimization algorithm with support vector regression machine algorithm was proposed.The penalty coefficient and kernel function parameter of SVR were optimized by using PSO to improve the problem that the particles fell into local minima.Through the prediction of the test set,the maximum absolute error of the experiment is 0.0016 and the mean square error is 0.0008%.The compensation accuracy of PSO-SVR model is higher than that of RBF network and SVM.PSO-SVR model can satisfy the actual accuracy requirements.