Research On Material Removal Analysis And Prediction Methods In Chemical Mechanical Polishing

Semiconductor chip industry is an advanced industry that drives the country’s industrial development and national economy,and its manufacturing technology is one of the core technologies in the electronic information industry.In the semiconductor chip manufacturing process,chemical mechanical polishing is currently the only surface precision processing technology that can achieve the global planarization.Establishing an accurate removal function model is a basic prerequisite for improving the certainty of chemical mechanical polishing as well as the surface quality and surface accuracy of the processed workpiece.At the same time,material removal rate is an important index for regulating process parameters and evaluating processing effect in chemical mechanical polishing process,and its accurate prediction is a key prerequisite for obtaining highly precise and efficient intelligent polishing process for semiconductor wafers.The main contents are as follows:Firstly,based on chemical mechanical polishing technology,this paper aims to optimize the removal function of planetary motion polishing.The means such as theoretical modeling and simulation analysis are used for a detailed study of the characteristics of the removal function as well as the variation trends.The main components include: establishing the material removal function model based on the Preston equation,and discussing the effects of eccentricity and rotation speed ratio on the removal function characteristics through simulation.Using the tending factor as the evaluation index,and deriving the optimal removal function model through simulation.The 3D model of rough texture surface close to the actual processing is generated by the algorithm.On this basis,two methods of dwell time solution,pulse iterative method and convolution iterative method,are researched by using the idea of discretization,and the residual error of the workpiece surface obtained by simulation are compared.The convolution iteration method is used as the dwell time solving algorithm,which can verify the error correction ability of the optimal removal function,explore the influence of the feed step on the residual error of the workpiece surface,and give its influence law.Secondly,from a data-based perspective,BP neural network,BP neural network optimized by genetic algorithm and deep belief network optimized by genetic algorithm are used for constructing the polishing material removal rate prediction model.At first,the experimental data collected by the detection sensors during the polishing process are extracted and pre-processed.Moreover,the random forest feature selection algorithm is used for selecting the characteristic variables that have a significant impact on the material removal rate,and then it is used as the input of the prediction model.Next,the prediction models are trained separately by using the above methods.Ultimately,the trained prediction models are used for polished material removal rate prediction,and the effectiveness of the prediction models are measured using evaluation metrics such as root mean square error.Finally,the experimental data from relevant literature are used for verifying the material removal rate prediction model established in this paper,and the simulation results show that the deep belief network optimized by genetic algorithm has a better prediction effect.Meanwhile,comparing with the existing literature reviewed,it is known that the prediction model of material removal rate based on deep belief network optimized by genetic algorithm proposed in this paper has the best effect.