本征型聚酰亚胺介电常数的定量构效关系模型构建与研究  

Construction and Study of Quantitative Structure-Property Relationship Model for Intrinsic Polyimide Dielectric Constant

作　　者：范振国[1,2] 刘四委 池振国[1] 张艺[1] 许家瑞[1] Zhen-guo Fan;Si-wei Liu;Zhen-guo Chi;Yi Zhang;Jia-rui Xu(Key Laboratory for Polymeric Composite and Functional Materials of Ministry of Education,Guangdong Laboratory of High-Performance Polymer Composites,Guangdong Engineering Technology Research Center for High-Performance Organic and Polymer Photoelectric Functional Films,School of Chemistry,Sun Yat-sen University,Guangzhou 510275;Guangdong Utrust Investment Holdings Co.,Ltd,Guangzhou 511045)

机构地区：[1]中山大学聚合物复合材料及功能材料教育部重点实验室,高性能树脂基复合材料广东省重点实验室广东省高性能有机聚合物光电功能薄膜工程技术研究中心,化学学院,广州510275 [2]广东粤财投资控股有限公司,广州511045

出　　处：《高分子学报》2021年第7期750-761,I0004,共13页Acta Polymerica Sinica

基　　金：广东省科技计划项目(项目号2020B010182001,2019B040401002);国家自然科学基金(基金号U20A20255,51873239);国家重点基础研究发展计划(973计划,项目号2014CB643605);广东省“特支计划”科技创新领军人才项目(项目号2016TX03C295)资助.

摘　　要：综合运用量子化学方法及基团贡献法,采集了78种不同化学结构聚酰亚胺的结构参数.采用通径分析法从16种参数中筛选出了对介电常数具有显著影响的8种结构参数,在此基础上,构建了2种针对聚酰亚胺介电常数的定量构效关系模型,平均相对误差均在10%以内.研究认为影响聚酰亚胺薄膜介电常数的最重要因素为分子体积,从宏观角度来看即体系的自由体积尺寸.评价了2种模型的适用性及稳定性,对比发现人工神经网络模型具备更高的精度,相对误差均在5%以内,多元线性回归模型的精度在10%以内,但具备更好的物理意义.设计了18种低介电常数的聚酰亚胺结构单元,并预测其介电常数.研究发现增加体系的含氟量可以降低其介电常数,最佳含氟量为0.25~0.37之间.侧基占比S/M最佳比例为0.5~0.6左右,侧基占比过大将会导致介电常数上升.基于上述研究,设计了3种超低介电常数的聚酰亚胺结构,其中最低预测介电常数为1.22.The structure parameters of 78 polyimides with different chemical structures were collected by using quantum chemistry method and group contribution method.Eight structural parameters with significant influence on dielectric constant were selected from 16 kinds of parameters by path analysis.On this basis,two kinds of quantitative structure-property relationship models for polyimide were constructed,and the average relative errors were within 10%.It is considered that the most important factor affecting the permittivity of polyimide film is molecular volume,which is the free volume size of the molecular from the macroscopic point of view.The applicability and stability of the two models are evaluated.It is found that artificial neural network has higher accuracy,the relative error is less than 5%,and the accuracy of multiple linear regression model is less than 10%,but it has better physical significance.Eighteen polyimide structural units with low dielectric constant were designed and their dielectric constants were predicted.It is shown that the dielectric constant can be reduced by increasing the fluorine content of the molecular,and the optimum fluorine content is between 0.25 and 0.37.The optimal ratio of side chain to main chain is about 0.5-0.6,which will lead to the increase of dielectric constant.Based on the above research,three polyimide structures with ultra-low dielectric constant were designed,and the lowest predicted dielectric constant was 1.22.

关 键 词：聚酰亚胺 介电常数 定量构效关系 多元线性回归 人工神经网络 

分 类 号：TQ323.7[化学工程—合成树脂塑料工业]