Improving statistical validity with Macro CD-SEM imaging
01/01/2006
A new image processing technology being added to current-generation CD-SEMs allows CD measurements to be made from multiple features from the same SEM image. Process control and characterization achieves better statistical validity, enabled by an order-of-magnitude increase in the data collection rate.
Benjamin D. Bunday, International Sematech Manufacturing Initiative (ISMI), Austin, Texas; John Allgair, ISMI, Freescale assignee, Austin, Texas; Ofer Adan, Aviram Tam, PDC Business Group, Applied Materials, Santa Clara, California
Process control continues to be a major challenge as device sizes shrink. Because critical dimension scanning electron microscopy (CD-SEM) is the most widely used image-based CD metrology method, it is continuously being improved. One such effort is the “Macro CD” capability to find the average of multiple measurements from the same image in order to reduce random variation.
Conventional CD-SEM measurement schemes typically involve single test features on multiple die on a wafer for process control. These schemes assume that a single CD measurement, which is only a single sampling of a broader population, is an accurate representation of the population. In fact, within periodic patterns that are nominally identical (e.g., a grating) there is a CD distribution with a local CD variation. This distribution may or may not be normal (Gaussian), but for simplicity, it may be considered so. With a larger sampling, the distribution can be characterized by an average value and a variance. With line features, this variance (or standard deviation) is linewidth variation, and with contact holes, it is contact hole variation.
Other metrology techniques, such as scatterometry-based optical CD, effectively measure the average CD of a regular grating structure but cannot currently characterize the linewidth variation within a grating target, beyond possibly flagging excursions from a goodness-of-fit metric. Measurement of variance quantities, obtained from measurements of many individual features, requires an image-based tool such as a CD-SEM.
There are different metrics for variation: linewidth roughness, a measure of the variation occurring along a single feature; local CD variation, a feature-to-feature variation between features in close proximity to each other; and across-chip CD variation, the variation between different features at different positions within an exposure field. Across wafer, wafer-to-wafer, and lot-to-lot variations also exist.
Local CD variation is a large component of across-chip CD variation. Each source of variation can have different physical causes; thus, a proper sampling plan must characterize most, if not all of these sources, to enable confident process control. Source-specific variations also can be used as fingerprints for specific process excursions. Resist local CD variation can have causes rooted in scanner focus, scanner slit nonuniformities, lens heating, other lens nonuniformities, post-exposure bake (PEB) temperature excursions, resist thickness nonuniformity, edge or proximity effects, or topography. Resist across-die variation can be caused by scanner-stage tilt, lens coma, other lens nonuniformities, PEB nonuniformity, edge or proximity effects, or topography. Etch local-CD variation is caused by etch loading or bad profile variation carried over from litho to etch. Etch across-die nonuniformity is caused by etch loading or effects carried over from lithography.
To control these different types of variation, the CD metrology tool must be capable of the precision [1] required by the International Technology Roadmap for Semiconductors [2], while maintaining throughput for an adequate sampling plan. Traditional sampling plans rely on single CD measurements, and focus on capturing lot-to-lot and across-wafer CD variation, occasionally sampling across-chip CD variation. Local CD variation is usually not measured due to excessive individual measurement times. However, Macro CD technology now makes it practical to use CD-SEM tools to sample local CD variation. Macro CD technology improves the CD-SEM’s available image size, processing power, and speed so multiple measurements can be obtained from a single image without a significant change in throughput. This application can be used with linewidths, spacewidths, contact holes, contact-hole edge roughness, linewidth roughness, or line-edge roughness. Since multiple discrete features are measured, the true average of the CDs at a location is determined, along with other descriptive statistics, such as the standard deviation and the max/min of the sampled features, to more precisely characterize the CD population.
Precision improvement
Improvement in precision is a direct result of the improved repeatability of the average CD, i.e., leveraging single-feature precision from use of multiple features within the same SEM image. For example, if precision of a single measured contact is σSEM = 1nm (3σ), N contacts can be measured and the average of all contact CDs calculated. The precision of the average CD, in case the CD measurements are uncorrelated, is 1nm/√N and has been confirmed experimentally [3] and through simulation. Samples used were large contact-hole arrays of various pitches and CDs in focus-exposure matrices (FEM) at the lithography stage (holes in ArF resist).
When using the new technique, imaging and measurement parameters are not very different from those used in conventional, single-feature CD measurements. Electron dose (charge per unit irradiated area) may be allowed to be slightly lower; this will have positive implications to noise, ArF resist shrinkage, charging and contamination, and sampling statistics. A tradeoff can be considered, allowing more noise from a lower dose that is balanced by the statistics of sampling multiple features.
Precision for the new method was ~34% of the precision obtained by measuring individual CDs (a 66% precision improvement), and the precision of the 3σ metric was 14% of the size of the population variation.
Sampling statistics
Population metrics (mean, standard deviation, max/min, etc.) available when using the Macro CD technique are more representative and informative than the measurement of arbitrarily chosen single features [3]. Consider a population of contact holes having a normal distribution of CDs, that is, CD ≈ N (µ, σPOP2). A measurement of a single feature from this population gives little information about the population mean, and no information about the population variance. Increasing the number of features measured (increasing sample size N) results in a better estimate of the population mean and variance, along with other characteristics of the population. As shown in more detail elsewhere [3], the estimation of the mean of such a population improves as 1/√N. The estimation of the standard deviation of the population also improves as a strong, monotonically decreasing function of N [3]. The solutions for the improvement factor of these metrics with increasing N are shown in Fig. 1.
null
Process characterization
Once the measurement process was optimized, a measurement scheme was executed to demonstrate the new method in analyzing the process variation within a FEM. This included a process window of five focus values by six exposure values. The measurement scheme was performed like a typical evaluation for across-chip CD variation, with five measurements per die: one in each die corner and one in the center. Each measurement included 16 contact holes (Fig. 2).
 Figure 2. Images showing a) top-down view of ArF resist contact holes at best focus and exposure and b) at low exposure and poor focus, with increased local CD variation. |
From the resulting data, Bossung curves - the family of focus curves with varying exposure - are plotted (Fig. 3). Individual hole measurements (Fig. 2) lie between the error bands. The Macro CD technology resulted in a higher confidence level in the mean CD population at a given site, and gives information on the spread of the distribution. The distribution itself also changes near the process window limits, yielding another metric to monitor process centering. Considering that data such as these are used to choose the litho process window, results using the new technique are more appropriate and less prone to risk than using individual CD measurements.
With the Macro CD method, each CD measurement consists of an average value and a 3σ metric that describes the distribution of CDs at the measurement site (local CD variation). Since five measurements were taken across the die, the distributions can be summed to infer the distribution of CDs across the entire die [3]. The resulting 3σ is the quadrature sum of local CD variation plus across-chip CD variation. The rms average of the 3σ values at the five sites gives a measure of the average local CD variation, so across-chip CD variation can be calculated as a separate variation component. Such a calculation can increase confidence in detecting lithography tool stage-tilt issues, or other such systematic causes of across-die CD variation.
Throughput
Move-acquire-measure (MAM) time (average time for a single measurement) is a key component of throughput, and this study found that measurements using the Macro CD technique have only slightly higher MAM time than single CD measurements. For the following calculation, an average value found in recent CD-SEM evaluations [4] for contact-hole measurements of 8 sec is used (the tool used in this study was somewhat faster). The process window characterization in Fig. 3 consists of 5×6 die with five measurements per die, for 150 measurements, with 16 contact holes evaluated per measurement. If each measurement took 8 sec, this recipe would run in 1200 sec, or 20 min. Executing this sampling plan with single-feature measurements would require 150×16 = 2400 measurements, and would take 19,200 sec or 5 hr 20 min.
 |
For a typical nine-site measurement plan and assuming the 8 sec MAM time, the new technique would thus take ~72 sec/wafer, so pipeline throughput would still be ~50wph (neglecting exchange time between wafers), for the gain of more data (more than 100 measurements) than available with conventional, single-feature measurements.
Conclusion
A new technique called Macro CD technology - the measurement of several or many features from a single image - characterizes local CD variation with improved precision and throughput over conventional single-feature measurements. The technique allows a CD-SEM tool to use all CD information from a single image with an increased data collection rate, enhancing its capability as an instrument for process characterization and control. Benefits include improved CD precision; improved confidence in finding average CD; characterization of 3σ local CD variation and other quantities to characterize CD distribution; improved reproducibility of shrinkage trend; an order-of-magnitude more data per unit time (or more); improvement in process control statistics; options for improved process control schemes with confident detection of different variabilities [3]; through leveraging multiple measurements, options for effective sampling with fewer measurement sites [3]; and faster detection of process excursions [3]. Local process variation can then be compared to specifications along with the improved value for average CD. Process excursions can be found much faster and with greater confidence.
Acknowledgments
The authors would like to thank Applied Materials PDC and Sematech, ISMI, and ATDF personnel, as well as the ISMI Advanced Metrology Advisory Group and Project Advisory Group. Sematech, the Sematech logo, International Sematech Manufacturing Initiative (ISMI), and the ISMI logo are registered servicemarks of Sematech Inc. All other servicemarks and trademarks are the property of their respective owners.
References
- B. Bunday, B. Singh, C. Archie, W. Banke, C. Hartig, G. Cao, et al., “Unified Advanced CD-SEM Specification for Sub-90 nm Technology,” 2004 Version, Sematech Technology Transfer document # 04114595 (nonconfidential document). The specification can be viewed at http://www.sematech.org.
- International Technology Roadmap for Semiconductors, 2004 Edition, http://member.itrs.net. (See Table 117 of the Metrology section.)
- B. Bunday, D. Michelson, J. Allgair, A. Tam, D. Chase-Colin, A. Dajczman, et al., “CD-SEM Metrology: Macro CD Metrology - Beyond the Average,” Proc. SPIE 5752, print pending, 2005.
- B. Bunday, M. Bishop, J. Allgair, “Results of Benchmarking of Advanced CD-SEMs at the 90nm CMOS Technology Node,” Proc. SPIE 5375, pp 151-172, 2004.
Benjamin D. Bunday performed graduate studies in STM, AFM, and SEM of high-temperature annealed silicon surfaces at Cornell U. He is project manager of CD metrology at Sematech/International Sematech Manufacturing Initiative (ISMI), 2706 Montopolis Dr., Austin, TX, 78741; e-mail [email protected].
John Allgair received his PhD in electrical engineering from Arizona State U., and is a Freescale Semiconductor assignee responsible for coordinating litho metrology programs at ISMI.
Ofer Adan received his BSc and MSc in electronic materials engineering from Ben Gurion U. of the Negev in Israel, and is a project manager and system engineer in the metrology SEM product division at PDC Business Group, Applied Materials, 9 Oppenheimer, Rehovot 76705, Israel; e-mail [email protected].
Aviram Tam received his BSc in electronics engineering and his MSc in operational research from the Tel Aviv U. in Israel, and is an R&D manager in the metrology SEM product division at PDC Business Group, Applied Materials.