基于VisionTransformer的混合型晶圆图缺陷模式识别

打开文本图片集

中图分类号：TN305；P391.4 文献标识码：A 文章编号：2096-4706（2025）19-0026-05

Abstract：Wafer testingisanimportantpartof thechipproductionproces.The identificationandclasifcationof wafer mapdefectpatems playakeyrole inimproving thefront-endmanufacturingprocessIntheactual productionprocess，various defects mayappearat thesametime，formingamixeddefect type.The traditional Deep Learning method has alowrecognition rate for mixed wafer map defect information. Therefore，this paper proposes a defect recognition method based on Vision Transformer.This methoduses the multi-head self-atention mechanism toencode theglobal features of the wafer mapand realizestheeffcientidentificationofmixedwaferdefectmaps.Theexperimentalresultsonthemixeddefectdatasetshowthat the performance ofthis method is better than that of the existing DeepLearning model，and the average accuracy is 96.2%

Keywords： computer vision; wafer map; defect recognition; Vision Transformer

0 引言

别速度。（剩余7623字）