论文部分内容阅读
采用响应曲面法(RSM)和人工神经网络(ANN)分别对化学机械抛光(CMP)碱性铜抛光液的主要成分(Si O2磨料、FA/O型螯合剂、H2O2氧化剂)进行优化研究。采用RSM优化,当抛光液中磨料、氧化剂和FA/O型螯合剂的体积分数分别为10.57%,1.52%和2.196%时,Cu的抛光速率的预测值和实测值分别为924.29和908.96 nm/min;采用ANN结合人工蜂群算法(ABC)优化,当抛光液中磨料、氧化剂和FA/O型螯合剂的体积分数分别为11.58%,1.467%和2.313%时,Cu的抛光速率的预测值和实测值分别为947.58和943.67 nm/min,其拟合度为99.36%,高于RSM的94.63%,且均方根误差较低为0.199 3。结果表明,在抛光液配比优化方面,RSM和ANN都是可行的,但后者比前者具有更好的拟合度和预测准确度,为更加高效科学地优化抛光液配比提供了一种新的思路和方法。
The main components (Si O2 abrasive, FA / O chelating agent, H2O2 oxidant) of chemical mechanical polishing (CMP) alkaline copper polishing solution were optimized by RSM and ANN respectively. Using RSM optimization, when the volume fraction of abrasive, oxidant and FA / O chelating agent in the polishing solution were 10.57%, 1.52% and 2.196% respectively, the predicted and measured values of Cu polishing rate were 924.29 and 908.96 nm / min. When the volume fraction of abrasive, oxidant and FA / O chelating agent in the polishing solution were 11.58%, 1.467% and 2.313%, respectively, the predictive value of the polishing rate of Cu was optimized by ANN combined with artificial bee colony algorithm (ABC) And the measured values were 947.58 and 943.67 nm / min respectively, the fitting degree was 99.36%, higher than that of RSM 94.63%, and the lower root mean square error was 0.199 3. The results show that both RSM and ANN are feasible in polishing solution optimization, but the latter has a better fitting degree and prediction accuracy than the former, which provides a more efficient and scientifically optimized polishing solution ratio New ideas and methods.