基于BP神经网络的刻蚀偏差预测模型

The Etch Bias Prediction Model Based on BP Neural Network

刻蚀是将设计版图转化为到晶圆图形的重要步骤,刻蚀质量的好坏直接关系到芯片或集成电路的性能,而影响刻蚀质量(蚀刻偏差)的孔径效应和微负载效应在很大程度上取决于版图密度、图案间距等布局特征。为了探究在固定刻蚀工艺参数下的版图特征对刻蚀偏差影响,本文提出了基于BP神经网络的刻蚀偏差预测模型。首先,对一维版图的特征提取并建立用于训练模型和测试模型的训练集与测试集,然后,训练和优化该BP神经网络模型。实验结果表明,该模型预测值的绝对误差可达±2 nm以下,而相对于真实刻蚀偏差的相对误差可达10%以下。因此在较大技术节点下,这种基于BP神经网络模型的预测精度是可以接受的。

Etching is an important step in the transfer of design layout to wafer results,its quality is directly related to the performance of chip or integrated circuit. The etch aperture effect and micro-loading effect that affect the etch quality( etch bias) are considerably dependent on layout features such as pattern density,spacing between patterns,etc. Under the fixed etching process parameters,a predicted model of etch bias was proposed. First,the feature set of the one-dimensional layout was extracted and a training set and a test set for training the model and the test model were established,and then the BP neural network model was trained and optimized. The experimental results show that the absolute error of the predicted value of the model can reach within ±2 nm,and the relative error can reach less than 10%. Therefore,under the larger technology node,the prediction accuracy based on the BP neural network model was acceptable.