SPIE 1995 #2439-01 Monitoring and Diagnostic Techniques for

Semiconductor lithography manufacturing presents a major challenge for the application of classical Statistical Process Control (SPC) methodologies due to the complex nature of this process. For example, difficulties can occur due to inadequate data sampli

SPIE 1995 #2439-01

Monitoring and Diagnostic Techniques for

Control of Overlay in Steppers

Gary E. Flores, Warren W. Flack, Susan Avlakeotes

Ultratech Stepper, Inc.

San Jose, CA 95134

Mark Merrill

KLA Instruments Corporation

San Jose, CA 95161

1.0 Abstract

Semiconductor lithography manufacturing presents a major challenge for the application ofclassical Statistical Process Control (SPC) methodologies due to the complex nature of thisprocess. For example, dif culties can occur due to inadequate data sampling, nonnormal errordistributions, equipment or process instability and nonstationary random errors. Incorrect use ofclassical SPC techniques can result in the incorrect interpretation of process stability which canhave a drastic impact on productivity. Photolithography provides additional SPC challenges dueto the inherent multivariable nature of the output variables that are being controlled. This paperexamines appropriate SPC and monitoring techniques for stepper control of overlay performanceusing in-process measurement and analysis equipment to address these issues.

The average run length of three charting techniques is compared to quantify the ability of eachtechnique to detect process mean shifts. Shewart, Exponentially Weighted Moving-Average(EWMA) and Cumulative-Sum (CUSUM) charts are analyzed for a baseline process and meanshifts of 0.42, 0.85 and 1.25 standard deviations. These results illustrate the superior performanceof a CUSUM chart over Shewart and EWMA charts. In addition, the Shewart chart with WesternElectric rules produced false mean shift alarms for the baseline case. The EWMA is also observedto be sensitive to the selection of weighting factors. The effectiveness of plotting inpidual wafersis compared with plotting lot means. The plotting of inpidual wafers outperforms lot means inthe determination of baseline shifts because of the larger population size of the inpidual charts.

Semiconductor lithography manufacturing presents a major challenge for the application of classical Statistical Process Control (SPC) methodologies due to the complex nature of this process. For example, difficulties can occur due to inadequate data sampli

SPIE 1995 #2439-01

2.0 INTRODUCTION

As the semiconductor industry has matured there has been a dramatic increase in the amount ofdata collected to address manufacturing issues [1]. In the area of photolithography, these data aresupplied by in-process systems as well as independent post-process measurement systems.Overlay and critical dimensions (CD) are the predominant lithography processes measured.Proper characterization and analysis of these factors are essential for continuously improvingproduct quality, productivity and device yield.

A major goal in lithography processes is the continuous improvement of overlay control.Matching overlay between stepper manufacturers’ and different generations of steppers can be asigni cant challenge [2]. Automated overlay metrology systems provide an opportunity to collectextensive data sets on lithography tools and processes. However, as a result of this large amount ofdata there is a need to improve and evolve monitoring and diagnostic techniques to maximize thedata value. This requires the appropriate use of statistical analysis and statistical process control(SPC) techniques.

Figure 1 illustrates the typical relationship of process, stepper, metrology, analysis and SPC for alithography process. The wafers at the stepper are processed and a subset is sampled using in-process measurement systems for monitoring overlay and critical dimensions. Next, themeasurement data is analyzed using SPC techniques with appropriate overlay models anddiagnostic routines. From this analysis, essential information on the stepper and lithographyprocesses are derived. Typical information consists of possible process alarms, optimal steppersettings of focus, exposure and overlay parameters. The output of the analysis can be used tocontrol the stepper and processes using standard control techniques.

Classical SPC principles such as Shewart Control Charts used in conjunction with WesternElectric rules [3] are applied for optimal process control in many manufacturing environments.However, the process complexity of lithography manufacturing presents a major challenge for theapplication these techniques. For example, the lithographic alignment process has been observedto exhibit systematic errors rather than random errors [4]. An implicit assumption of SPCprinciples is that the processes and tools being monitored are Identically, Independent andNormally Distributed (IIDN) [5]. This implies that monitored processes and tools are from anormal data set and all measurements are uncorrelated over time.

Violation of the IIDN requirements can occur during overlay monitoring for a variety of reasons.For example, insuf cient overlay data sampling may result for infrequently processed lithographylevels. Insuf cient data can be exhibited as a nonnormal overlay error distribution. A secondscenario is a nonnormal distribution for processes and equipment that are unstable due topremature implementation in manufacturing. A third possibility is that inpidual measurementsare not statistically independent and thus are correlated over time. An example would be a processor piece of equipment that has an inherent cyclic nature. In this case, sampled overlay

Semiconductor lithography manufacturing presents a major challenge for the application of classical Statistical Process Control (SPC) methodologies due to the complex nature o …… 此处隐藏：28983字，全部文档内容请下载后查看。喜欢就下载吧 ……