晶圆缺陷检测与分类的卷积神经网络
晶圆缺陷检测与分类的卷积神经网络；针对晶圆检验时扫描电镜图像的缺陷检测和缺陷分类两问题，采用了“ ZFNet”的卷积神经网络来分类晶圆缺陷，并基于该分类器实现了一种“基于块的卷积神经网络”缺陷检测算法。为了提高准确率和加快速度，又改动“更快的区域卷积神经网络”实现了另一种检测算法。第卷第期邡鑫,史峥:晶圆缺陌检测与分类的卷积神经网络ZENet classifierDarker ImIn.ril” HumpBitel检测算法示意图在训练检测器时,数据集是检测器原始尺寸的图像,且包含标记好的缺陷区域和类型。我们结构通过·系列数据扩张操作,得到组数据,随机选取相比于检测算法主作为训练集,作为测试集。要从以下三方面进行了针对性的改进算法中需要优化的参数有滑动窗口尺寸滑()针对重复计算卷积的缺点,采用先动步幅、概率阙值、面积阙值,由于无法求出统一计算特征图,再按)进行映射各参数与检测结果的明确关系式,所以采用遍历法优化参截取的办法。如图,先通过卷积网络(数。因为检测到的缺陷尽量正确和尽量检测到所有缺陷是)对输入图像计算得到其特征图,因为在输入图像矛盾的,故以精确率和召回率的调和平均值作为优上的都能映射到特征图上,所以从输入图像上按化目标,也可根据实际需要调整两者权重满足不同侧重。割取图像进行卷积运算可以替代为直接从特征图上按测试结果映射后的范围割取,从而避免多次重复计算卷积。由于用训练好的检测模型对测试集检的大小形状不·,而全连接层的神经元连接数是固定的,测,计算模式下每张图大概耗时如果检测到的缺所以对割取得到的子特征图,通过层次采样到统陷与标准答案的且类型相同,则判为正确,否尺寸以连接到全连接层。则判为错误。得到结果如表,计算得:laut Image精确率Feature Map召回率ROI其屮正确缺陷的平均表检测器测试结果数量正确错误network有缺陷(正类)图映射示意图从检测结果来看该算法基本实现∫对图像上晶圆()针对滑动窗口尺寸单·的缺点,增缺陷的检测和分类,但是值较低,缺陷检测位置不加了滑动窗口的尺寸类型,并且增加由一个全卷积网络准确,检测耗时较长,分析其原囚如下)组成的()检测出错的数据中,缺陷较大的类型易判断错,)来预判断是否有缺陷。本文采用面积缺陷较小的容易被漏掉,说明只使用一种尺寸的滑动框很分别为,长宽比分别为、共难适应尺寸变化范围较大的缺陷种尺寸的滑动窗口,依次计算其中有缺陷的概率,再从中)滑动框步幅减小则算法耗时平方倍增加,而步幅筛选出一定数量最有可能有缺陷的区域,进行非极大值抑过长造成缺陷概率分布图分辨率较差,从而检测到缺陷位制(),最后得到一定数置准确度较差量的候选区域。()相邻滑动框都有大量重叠,所以每个区域都被多()针对缺陷检测位置准确度差的缺点,次重复送入计算卷积,导致算法耗吋较长。在全连接层后连接一个边界回归层在与上述检测算法相似的图像目标检测领域,近来出用来修正缺陷位置,该回归层与分类层并列。现的很好的克服了以上缺点并取得了很好的针对本文的缺陷检测问题,直接套用标准效果,所以下面介绍如何通过改动实现品圆并不能解决问题。因为判断晶圆的缺陷类型通常需缺陷的检测与分类。要结合缺陷区域周围的图形信息,而在预判断是否有C1994-2017ChinaAcademicJournalElcctronicPublishinghOusc.Allrightsrescrved.http://www.cnki.nct计算机工程年月日缺陷吋还进行了边界回归。虽然更加准确的给出缺陷的位()将原尺寸为的图像调整为置,但送入检测网络的特征儿乎不包含缺陷周围图肜信息,使得滑动窗口尺寸能够适应缺陷大小的变化范围,也可以导致缺陷分类不准。故木文对标准徹了一些根据实际情况来具体调整。改动:得到缺陷检测算法如图,卷积网络(()将改为只判断滑动窗口内是否有缺陷,而,)将输入图僚转换成多种特征图;根据不进行边界回归,也就是只计算所有滑动窗口有缺陷的概特征图从滑动窗口中选出最有可能存在缺陷的率,选取可能性最大的个,做非极大值抑制,再选出层根据特征图中抽取出对应特征组成特可能性最大的个进行检测。征向量;检测网络()根据特征向()将卷积层的尺寸加大为,加大感受野量判断缺陷类型,并进行边界回归;最后通过和概率),从而在判断滑动框內是否有缺陷吋能参阈值对候选缺陷进行过滤即可得到最终缺陷。考更多的周围信息。Detection NetworkonFolutionnl actorSoftmaxRuI Puling liver,e Prop卟 edMS+PrubilitessionInput Image 1024*1024Fully 10 dyercrectCcrvchrionalLaver size 747图检测算法示意图模型训练和平均值作为优化目标,并且使用相同的训练集和图中的检测算法也是基于架构实现,因为卷测试集积网络提取的特征类型对相似普遍有效,故其卷积网络的测试结果参数是直接迁移第章分类器的卷积层参数。但是用训练好的检测模型对测试集检测,和的参数则需要通过方法进行训练,标准计算模式下每张图大概耗时,采用相同判定标准,提供了分开和联合两种训练方式。为了节约得到检测结果如表(其中负类总数与表中总数不同是因时间,本文采用联合训练方式,并结合缺陷检测问题的实为同一张图屮可能检测到多个缺陷),计算得际情况调整超参数精确率在训练时,对每张输入图像,要计算的滑动窗口召回率数量庞大(种尺寸的滑动窗口,滑动步幅)。所以从中随机抽取个作为训练集,其中正例其中正确缺陷的平均负例,且正例占比不超过。分类器采用表检测器测试结果损失函数数量正确错误在训练时,设置提供个,从中随有缺陷(正类)机选取个作为训练集,其屮正例无缺陷(负类)负例,且正例占比不超过。另外设置学从结果来看该算法各方面都优于检测算习率分类器采用损失函数,而边界回法和值更高说明检测检测缺陷类型正确归采用函数。且位置准确,而且速度也大大提高(检测一张图像耗时从为了与检测算法对比,在最后通过遍历法缩小到)。如图为检测缺陷示例,共中标注了缺陷优化和概率阈值时,同样以精确率和召回率的调位置、类型和对应概率C1994-2017ChinaAcademicJournalElcctronicPublishinghOusc.Allrightsrescrved.http://www.cnki.nct邡鑫,史峥:晶圆缺陷检测与分类的卷积神经网络I I图检测结果示例图结束语而对图像上的缺陷检测和缺陷分类这两个问题,本文提出的改动后的检测算法能够精准、快速地从图像中检测出缺陷并同吋进行分类。得益于卷积神经网络良好的特征学习能力,该检测算法能够根据标记好缺陷位置和类型的数据自动学习特征,从而尽量避免人工千预,使算法具有较强的适应能力。参考文献徐姗姗刘应安徐昇基于卷积神经网络的木材缺陷识别山东大学学报工学版刘云杨建滨王传旭基于卷积神经网络的苹果缺陷检测算法电子测量技术江帆刘辉王彬等基于模型的图像识别计算机工程C1994-2017ChinaAcademicJournalElcctronicPublishinghOusc.Allrightsrescrved.http://www.cnki.nct
- 2021-05-06下载
- 积分：1
CMW500仪器编程手册
CMW500的资料不多，这是非常不错的参考资料R&s CMW 500Contents overviewContents overview1 Preparing the Instrument for Use2 Getting Started3 System Overview4 Basic Instrument functions5 Remote Control6 System Command Reference7 General Purpose RF Applications8 GSM Applications9 WCDMA Applications10 WiMAX Applications11 AnnexesNote about Faceless InstrumentsChapter 1 of this manual gives an overview of the front panel controls and connectorsof the R&s CMW 500 Wideband Radio Communication Testers with display and givesall information that is necessary to put the instrument into operation and connectexternal devices. The application examples in Chapter 2 and the following chapters arealso based on a r&S CMW 500 with displayThe measurement functionality of the two instrument types is identical. You can test allmeasurement examples reported in this manual using an r&S CMW 500 withoutdisplay that is controlled from the Graphical User Interface displayed on an externalmonitor or pcFor specific information concerning faceless instruments refer to your quick start guide.Operating Manual 1202. 3986.32-03R&s CMW 500ContentsContents1 Preparing for Use…日日画1.1 Front Panel Tour1.1.1 Utility Keys…1.1.2 Status LEDs and Standby Key1.1.3 Display…1.1.4 Softkeys and Hotkeys1.1.5 Setup Keys1.1.6 Data Entry Keys223334561.1.7 Rotary Knob and Navigation Keys1.1.8 Front panel connectors1.1.8. 1 RF Connectors1.1.8.2 LAN Connector1.183 SENSOR Connector1.184 USB Connectors1.1.8.5 AF Connectors888881.2 Rear panel tour:::::B:1.3 Putting the Instrument into Operation1.3.1 Unpacking the instrument and checking the shipment1.3.2 Instrument Setup...............001.3.3 Bench Top Operation1.3.4 Mounting in a 19 Rack121.3.5 EMI Protective measures131.3.6 Connecting the Instrument to the AC Supply131.3.7 Power on and off…131.3.8 Replacing Fuses141.3.9 Standby and ready state141.4 Maintenance15Operating manual 1202.3986.32-03R&s CMW 500Contents1.4.1 Storing and Packing151.5 Connecting External Accessories…………,…,…,…,…,…,…,………,……….151.5.1 Connecting a mouse161.5.2 Connecting a Keyboard161.5.3 Connecting a Printer1.5. 4 Connecting a monitor1.5.5 Connecting a LAN Cable788916 Starting the R&scMW500 and Shutting D。wn…,.....,.,…,,,191.7 Remote Operation in a LAN...........-.201.7.1 Assigning an IP Address201.7.2 Remote Desktop Connection221.8 Windows xP国国国面1.9 Firmware Update…,,…,,,,,,,,,",…,…222 Getting Started,…,…,…252.1 Basic tasks.…252.1.1 Accessing Dialogs252.1.2 Using Keyboard Shortcuts272.1.3 Data entr272.14 Using Front Panel Keys.……282.1.5 Using an External Keyboard2.1.6 Task bar302.2 Sample Session…312.2.1 Generating an rf signal312.2.1.1 GPRF Generator2.2.2 Measuring an RF Signal332.2.2.1 GPRE Power333 System Overview…363.1 Generators363.1.1 Generator Control363.1.2 RF Path Settings(Generators)37Operating manual 1202.3986.32-03R&s CMW 500Contents3.2 Measurements383.2.1 Measurement control383.2.2 Connection Control(Measurements393.2.3 Statistical Settings3. 2. 4 Statistical Results3.2.4.1 Statistics Type423.2.4.2 Detectors433.2.4.3Peak∨ alues.433244 Averaging…443.245 Standard deviation143.2.5 Trigger Settings453.2.6 TX Measurements453.2.6.1 Power results463.2.6.2 Modulation accuracy3.2.6.3 Adjacent Channel Power(Spectrum)493.2.6.4 Spectrum Emission Mask493.2. 6.5 Code domain power503.2.6.6 Multi-Evaluation measurements4 Basic Instrument Functions534.1.1 Startup Dialog534.2 Utility Dialogs544.2.1 Reset Dialog..4.2.2 Print Dialog554.2.3 Save/Recall Dialog564.3 Setup Dialog…,,,,…574.3.1 Activating Options584.3.2 Selftests…594.3.2.1 General test features604.3.2.2 Board Tests614.3.23 System Tests…62Operating Manual 1202.3986.32-03R&s CMW 500Contents4.3.2.4 Performing Selftests24.3.25 Selftest Parameters634.3.3 Reference Frequency654.3.3.1 Reference Frequency Settings654.3.4 Measurement Controller Dialog664.3.5 Generator Controller Dialog665 Remote Control685.1 Remote Control Operation685.1.1 Establishing and Testing a LAN Connection705.1.2 Switchover to remote control5.1.3 Return to Manual Operation715.2 Messages国国国面…725.2.1 VXI-11 Interface Messages725.2.2 GPIB Bus Interface Messages..725.2.3 Device Messages(Commands and Device Responses)735.2. 4 SCPl Command structure and syntax735.2.4.1 Common commands745.2.4.2 Instrument-Control Commands5.2.4,3 Structure of a command line765.2.4.4 Responses to Queries5.2.45 SCPI Parameters,775.2.4.6 Use of SCPl Subsystems95.3 R&s CMW Software and command structure5.3.1 General command structure5.3.2 Firmware applications815.3.3 Measurement Contexts and views5.4 Control of the instrument825.4.1 Measurement Control825.4.1.1 Measurement states and measurement control commands835.4.1.2 INITiate: : MEASurement84Operating manual 1202.3986.32-03R&s CMW 500Contents5.4.1.3 ABORt: MEASurement 8554.14sTOP:< Application> MEASurement>.,……855.4.1.5 Measurement substates855.4.2 Statistical Settings865.4.3 Retrieving Measurement Results885.4,3. FEtCh.? Command885.4.3.2 READ.? Command∴8954.33 Retrieving Single∨ alues and traces.…5.4.4 Reliability Indicator5.4.4.1 Common Reliability Indicator……5.4.5 Multi-Evaluation Measurements5.4.6 Generator control925. 4.7 RF Path Settings945.4.8 Resource and path Management∴945.4.8. 1 Basic RPM Principles5.4.8.2 Queuing of Measurements∴9654.83 Causes for task Conflicts5.4.8.4 Monitoring Measurement and Generator States995.5 Command Processing...:::::B:995.5.1 Input Unit1005.5.2 Command Recognition1005.5.3 Data base and instrument hardware1015.5.4 Status Reporting System1015.5.5 Output Un1025.6 Status Reporting System1025.6.1 Overview of status Registers1035.6.2 Structure of an SCPl Status Register…1035.6.2.1 Description of the five status register parts1045.6.3 Contents of the Status Registers1055.6.3.1 STB and sre.105Operating manual 1202.3986.32-03R&s CMW 500Contents5.6.3.2 IST Flag and pPe.1065.6. 3.3 EsR and ese∴1075.6.3 4 STATus: OPERation1085.6.3.5 STATus QUEStionable1085.6.4 Application of the status reporting s ystem1085.6.4.1 Service Request1085.6.4.2 Serial poll1095.6.4.3 Parallel poll1095.6.4.4 Query of an Instrument Status1105.64.5 Error queue.….115.6.5 Reset Values of the Status Reporting System1116 Command reference∴∴11361 Special Terms and Notation…,…………,…,…,………,…,…,,…,………………1136.2 Common commands1156.3 Instrument-Control commands.117631 MMEMory Commands…..,,,…,,……1176.3.2 Ref Frequency Commands1206.3.3 STATus Commands1216.3.4 SYSTem Commands1256.3.5 LAN Services1256.3.6 Miscellaneous Instrument Settings1286. 4 Alphabetical List of Commands System)1317 GPRF Applications1337.1 GPRF Measurements and generators1337.1.1 General Purpose RF Generato.1337.1.1.1 GPRF Generator(Constant Frequency)1337.1.1.2 Arbitrary RF Generator(Option R&S CMW-B110 A)1347.1.13 List Mode∴1357.1.2 Power measurement1357.1.2.1 Test Setup…135Operating manual 1202.3986.32-03
- 2020-12-05下载
- 积分：1
