基于SSA-GRNN的铜CMP抛光液抛光速率预测

Prediction of Copper CMP Polishing Rate Based on SSA-GRNN

铜化学机械抛光(CMP)是集成电路制造的关键步骤之一,其中铜抛光速率是衡量抛光液性能的关键指标。在CMP过程中,由于铜抛光液中各组分与铜之间的化学反应复杂,需要大量的数据实验来实现可调的抛光速率。为提高铜CMP抛光速率预测的准确性,利用麻雀搜索算法对广义回归神经网络的平滑因子进行优化,提出了一种基于麻雀搜索算法的广义回归神经网络(SSA-GRNN)铜CMP抛光液抛光速率预测模型。首先,在MATLAB中建立SSA-GRNN网络模型,然后输入抛光液各组分数据,预测在不同组分下抛光液的抛光速率,最后将SSA-GRNN模型的预测结果与BP神经网络模型(BP-NCABC)的预测结果对比。结果表明,SSA-GRNN模型在训练集上的平均绝对百分比误差(MAPE)比BP-NCABC模型降低4.82百分点,在测试集上的MAPE比BP-NCABC模型降低1.78百分点;SSA-GRNN模型在训练集上的最优预测精度比BP-NCABC模型提高0.09百分点,在测试集上的最优预测精度比BP-NCABC模型提高0.32百分点。上述研究结果表明,在CMP抛光速率的预测上SSA-GRNN模型比BP-NCABC模型的准确性更高,这为指导CMP实验、提升实验效率、降低研发成本和优化抛光液组分提供了一种可选的模型。

Copper chemical mechanical polishing(CMP)is one of the key steps in the manufacture of integrated circuits,and the copper polishing rate is the key index to measure the performance of the polishing slurry.It requires vast amounts of experimental data to achieve adjustable polishing rates due to the complex chemical reactions between various components of the copper polishing slurry and copper during the CMP process.To enhance the accuracy of copper CMP polishing rate prediction,sparrow search algorithm is used to optimize the smoothing operator of generalized regression neural network,and a prediction model of copper CMP polishing rate based on SSA-GRNN is proposed.Firstly,an SSA-GRNN network model is established in MATLAB.Subsequently,the component data of the polishing slurry is inputted to predict the polishing rate of the slurry under different compositions.Finally,a comparison is made between the prediction results of the SSA-GRNN model and the BP-NCABC model.The comparative results indicate that the SSA-GRNN model exhibited a 4.82 percentage point reduction in mean absolute percentage error(MAPE)on the training set and a 1.78 percentage point reduction in MAPE on the testing set.The SSA-GRNN model demonstrated a 0.09 percentage point improvement in optimal prediction accuracy on the training set and a 0.32 percentage point improvement on the testing set compared to the BP-NCABC model.These findings indicate the SSA-GRNN model exhibits higher accuracy than the BP-NCABC model in predicting CMP polishing rates,which provides an alternative model for guiding CMP experiments to enhance experimental efficiency and reduce research and development costs and optimizing the components of the polishing slurry.