数据挖掘在IGBT模型参数提取中的应用研究  

Application Study on Data Mining in Parameter Extraction of IGBT Model

作　　者：严利人[1] 刘道广[2] 刘志弘[1] YAN Liren;LIU Daoguang;LIU Zhihong(Institute of Microelectronics,Tsinghua University,Beijing 100084,P.R.China;Institute of Nuclear and New Energy Technology,Tsinghua University,Beijing 100084,P.R.China)

机构地区：[1]清华大学微电子学研究所,北京100084 [2]清华大学核能技术研究院,北京100084

出　　处：《微电子学》2020年第6期848-852,共5页Microelectronics

基　　金：广东省重点领域研发计划项目(2019B010143002)。

摘　　要：针对半导体器件的SPICE模型参数提取,提出了一种正向处理技术。对于选定的器件和模型,大量运行不同模型参数组合下的SPICE仿真,获得各种不同的电特性曲线,形成超大规模的数据集。若通过测试得到了确实的测试数据,则通过数据挖掘和人工智能中的数据处理算法得到数据集中、最匹配的曲线项,直接给出模型参数的估计值。针对IGBT模型,通过批量仿真获得约15 k个数据,使用kNN算法和多元回归法对测试曲线构成的测试集进行了参数提取。结果表明,该方法能快速获取器件的模型参数,具有稳健性的优点。该方法为研究者对器件模型特性提供了有益的认识。A forward processing method for extracting SPICE model parameters of semiconductor devices was proposed. SPICE simulations under different model parameters were performed on the selected devices and models to obtain various electrical characteristics curves and formed a very large data set. If the real test data was obtained through the test, the most matching curve item in the data set was obtained through data mining and data processing algorithm in artificial intelligence, and the estimation values of model parameters were directly given. In this paper, about 15 k data were obtained through batch simulation for IGBT model, and the parameters of test set composed of test curve were extracted by kNN algorithm and multivariate regression method. The results showed that this method could get the model parameters of the device quickly and had the advantages of robustness. This method provided researchers with a useful understanding of device model characteristics.

关 键 词：批量仿真 数据挖掘 KNN算法 多元回归 

分 类 号：TN307[电子电信—物理电子学]