论文部分内容阅读
用作压电材料的氮化铝薄膜要求有100或002晶体取向。薄膜的晶体取向与薄膜制备时的溅射气压、靶基距、靶功率有关。本工作应用支持向量机算法总结实验数据的数学模型,并用留一法检验模型的预报能力。计算表明:支持向量机算法的预报能力较传统的模式识别算法如Fisher法或KNN法为佳。有希望成为实验设计的有用工具。
Aluminum nitride films used as piezoelectric materials require a 100 or 002 crystal orientation. The crystal orientation of the film is related to the sputtering pressure, target distance and target power when the film is prepared. This work applies the support vector machine algorithm to summarize the mathematical model of the experimental data, and uses the left-over method to test the forecasting ability of the model. The results show that the prediction ability of SVM is better than that of traditional pattern recognition algorithms such as Fisher and KNN. Hope to be a useful tool for experimental design.