电子测量与仪器学报

郑江云,詹文法,蔡雪原.芯片多参数一致性的筛选方法[J].电子测量与仪器学报,2025,39(7):81-87

芯片多参数一致性的筛选方法

Screening method for multi parameter consistency of chips

DOI：

英文关键词:chip screening testing parameters normal distribution K-means method

基金项目:安徽省高校自然科学研究项目(2023AH050500)、国家自然科学基金(62474002)、宜城精英项目(202371)、集成与微组装技术国家地方联合工程实验室开放课题（KFJJ20230101）项目资助

作者	单位
郑江云	安庆师范大学电子工程与智能制造学院安庆246133
詹文法	安庆师范大学电子工程与智能制造学院安庆246133
蔡雪原	安庆师范大学电子工程与智能制造学院安庆246133

Author	Institution
Zheng Jiangyun	School of Electronic Engineering and Intelligent Manufacturing, Anqing Normal University, Anqing 246133,China
Zhan Wenfa	School of Electronic Engineering and Intelligent Manufacturing, Anqing Normal University, Anqing 246133,China
Cai Xueyuan	School of Electronic Engineering and Intelligent Manufacturing, Anqing Normal University, Anqing 246133,China

摘要点击次数: 786

全文下载次数: 562

中文摘要:

由于制造工艺的偏差，芯片有些参数值离散性较大，为了提高电子系统的稳定性和可靠性，设计一种提高芯片一致性的多参数低相关聚类优选（multi parameter low correlation clustering selection，MLCS）算法。该算法首先计算参数间的斯皮尔曼秩相关系数，选择相关性小的测试参数进行筛选，提高选择的效率，然后对多个参数分别用1维K-means方法进行3级聚类，再综合它们的分类结果，筛选出集中于聚类中心的芯片。实验结果表明，该算法能够实现测试芯片的自动筛选，位于中间聚类中心的芯片参数值都在均值附近，上下不偏离1个方差，且分类界限清晰、聚类效果不受筛选参数个数的限制；894个样本按照2个参数筛选，散点图显示的效果明显优于常规的二维模糊聚类和二维K-means算法；所用时间约0.04 s，而模糊聚类算法耗时超过12 s。该算法具有良好的适应性，能够有效选出多种参数值都接近均值的芯片。

英文摘要:

Due to manufacturing process deviations, some parameter values of chips have large discreteness. In order to improve the stability and reliability of electronic systems, this paper designs a multi parameter low correlation clustering selection (MLCS) algorithm to enhance chip consistency. The algorithm first calculates the Spearman rank correlation coefficient between parameters, selects test parameters with low correlation for screening, improves the efficiency of selection, and then uses 1D K-means method to perform 3-level clustering on multiple parameters. Based on their classification results, chips concentrated in the cluster center are selected. The experimental results show that the algorithm can achieve automatic screening of test chips. The parameter values of chips located in the middle cluster center are all around the mean, with no deviation of one variance, and the classification boundary is clear. The clustering effect is not limited by the number of screening parameters; 894 samples were screened according to 2 parameters, and the scatter plot showed significantly better results than conventional 2D fuzzy clustering and 2D K-means algorithm; The time taken is about 0.04 seconds, while the fuzzy clustering algorithm takes over 12 seconds. This algorithm has good adaptability and can effectively select chips with multiple parameter values that are close to the mean.

查看全文查看/发表评论下载PDF阅读器