Advantages and implementation of integrated after-develop inspection
05/01/2003
Overview
In an industry moving rapidly to full automation, integrated metrology has been somewhat elusive. The combination of risks has kept its adoption at 2% or less. However, working from a clear set of requirements necessary to make integrated metrology successful, not the least of which is tight matching of integrated and stand-alone tools, this application provides fast, reliable, integrated after-develop lithography inspection in a 300mm fab.
The potential benefits of integrated metrology are widely recognized. Moving the metrology directly onto process equipment can provide virtually instantaneous detection of process excursions and drift. This can prevent additional wafers from being misprocessed and allows problems to be diagnosed and quickly corrected.
The benefit of early detection and correction on fab yield increases with the value of chips and the size of the wafer. In a 300mm fab, the value of a finished wafer can be tens of thousands of dollars, so an integrated metrology system that improves yield by a single percent can save millions of dollars per year.
null
Nevertheless, the adoption of integrated metrology has been slow for several reasons. Integrating a metrology module into process equipment can increase its price and often its footprint, and can lower throughput or reduce capacity. In addition, most process equipment is inherently stable, so these extra costs, coupled with lower performance, cannot justify a relatively small potential yield increase.
Equally important is the concern that the metrology module could halt the production of wafers on an otherwise properly functioning process tool. This could happen for a variety of reasons, including reliability, time required for metrology recipe creation and optimization, or even metrology results that falsely indicate a problem or are ambiguous.
Finally, integrated metrology can be difficult to implement because it requires close and ongoing relationships between the metrology supplier, the equipment manufacturer, and users, all of whom must work together to successfully install, use, and maintain the equipment.
Thus far, these negatives appear to have, in most cases, outweighed the positives. While estimates vary, integrated metrology appears to have captured no more than 1–2% of the overall metrology market. However, when the right integrated technology is installed in a process where defects can be quickly and economically reworked, and a strategy is in place to allow integrated modules throughout the fab to be matched and to share recipes and data, real yield improvement can be achieved.
After-develop inspection
With the various process steps in lithography, the chance of misprocessing during any one step is quite small, but the cumulative probability of a yield-robbing event becomes significant. The rare opportunity with lithography is that potentially yield-robbing defects can be economically reworked if they are detected before an irreversible process step such as etch or implant.
Lithography defects that arise from coat, expose, and develop steps vary widely in size, shape, color, and overall appearance. Not all lithography defects should be reworked; some will have little or no impact on overall wafer yield. Data on all defects is important for process control, but it is only when a defect's impact on yield begins to equal or exceed the cost of rework that the wafer should be stripped and reprocessed.
An example is shown in the figure. The poorly developed wafer would be reworked as nearly half the die on the wafer are affected. However, unless the number of particles exceeded a certain user-defined limit, the wafer with the particle defect would generally not be reprocessed as the defect affects only a portion of one of the wafer's die.
Accordingly, after-develop inspection (ADI) can be a critical enabler for high-yield manufacturing, but it must be sophisticated. Conventionally, human operators were required to detect the wide range of defects, to learn new defects as a process changed, and to make decisions based on fab-specific criteria.
However, manual inspection of wafers is a slow process requiring a limited lot-to-lot or within-lot sampling strategy. As a result, a significant number of wafers can be misprocessed before a problem is identified. Manual inspection is also inconsistent, as detection, classification, and dispositioning decisions tend to vary from person to person. Manual inspection faces another significant challenge in 300mm fully automated fabs, as wafers ideally should never leave their isolated FOUP environments.
At Rudolph Technologies, we have developed an automated ADI metrology tool, called WaferView, that can outperform human inspectors, providing accurate and consistent defect detection and classification 24 hours a day and at much higher speeds. Programmable decision-making criteria can automatically disposition wafers for rework or flag yield engineers when process excursions occur. These advances now allow 100% surface inspection of every wafer at stepper speeds and allow the implementation of integrated macrodefect detection and classification.
Integrated auto-ADI
Integrating auto-ADI metrology directly onto a lithography track will provide immediate notification of process excursions, improved statistical process control (SPC) data, and better root-cause analysis. However, to be successful, an integrated auto-ADI module must provide:
- high speed that enables defect detection and classification for the entire surface of each wafer processed;
- high defect capture rate so the cost/defect missed becomes the key driver of CoO and ROI;
- highly consistent defect classification that results in correct rework decisions, better statistical information on defect types, and maximum yield improvement;
- off-line recipe creation, optimization, and defect review to prevent track throughput reduction;
- high-reliability integrated modules that are significantly more reliable than the track equipment and have backup metrology available to reduce the risk of track throughput reduction;
- matching between integrated modules and with stand-alone auto-ADI metrology tools for consistent detection, classification, and rework decision making throughout the fab; and
- recipe management, defect management, and data storage that are all effectively managed from a central location to ensure all integrated modules use the same recipes, generate the same alarms, and follow the same dispositioning rules so yield engineers can access all data for a wafer or wafer lot.
Technology requirements
To provide the required speed, defect capture, and classification rates, the WaferView's auto-ADI metrology captures full-color, whole-wafer images using simultaneous dark- and bright-field illumination. Color bright-field mimics the response of the human eye and allows detection of defects that are invisible to monochrome or gray-scale systems. Using patented algorithms, the resulting image is compared to that of an ideal or "golden" wafer having no defects. When a defect is detected, its image is broken down into multidimensional vectors that are compared with a library of known classified defects stored in a central database. Using vectors for comparison provides fast and repeatable image classification.
 |
null
The metrology system comes with a library of standard defects developed through years of fab experience and can learn an unlimited number of new defect classes. The defects are cataloged by Rudolph's Knowledge Base software. This allows each fab to tailor the metrology to its own internal defect classes, which allows for better decision making and more valuable SPC data. In addition, user-defined defects extend the metrology to new processes and materials having unexpected properties or a unique appearance.
Our auto-ADI inspection has a throughput of up to 120 200mm wafers/hr, and up to 110 300mm wafers/hr, which exceeds stepper throughputs by a comfortable margin. The high throughput comes from single-pass, bright-dark field detection and rapid vector-based classification. Full-color detection is the key to the high capture rate of up to 95%, because it can identify changes in film thickness, film properties, and other color variation defects that negatively impact yield, but cannot be detected by monochrome technology. Vector comparison, the wide range of available defect types, and the ability to generate more defect types based on fab-specific criteria enable correct classification up to 85% of the time.
Defect capture results
Recently, we installed eight integrated auto-ADI modules and one stand-alone auto-ADI tool in a 300mm US fab. To test the ability of the nine systems to repeatably catch the same defects with identical sensitivity, we used a defect standard wafer (DSW) containing three classes of defects in various sizes:
- defocus defects from +0.3 to -0.4µm in 0.1µm increments;
- silicon defects (mounds of oxide) from 100–40µm in 20µm increments; and
- silicon dioxide defects (holes in an oxide layer) from 100–40µm in 20µm increments.
The criteria were that the eight integrated modules and the stand-alone tool had to detect and map all the DSW defects correctly. The gathered data showed that the nine auto-ADI tools met the criteria and detected DSW defects 100% of the time, with the exception of the 40µm silicon defects, which were still captured at a relatively high rate (Table 1). The 40µm silicon and silicon dioxide defects are below the specification threshold of the inspection technology that is tuned to capture defects ≥50µm. The camera pixel size is 18µm, however, which allows capture of smaller defects. Boosting the sensitivity can improve the capture rate of defects below 50µm, but it may increase false positive defect detection.
Configuration and matching
A key requirement for the auto-ADI system was that it did not reduce track and stepper throughput. Therefore, reliable operation is essential. In the 300mm fab trial, the scheduled downtime for the integrated modules was <2% and, six months after acceptance, unscheduled downtime is also <2%. However, integrated module downtime is not the only potential source for track throughput reduction. Valuable track and stepper process equipment could sit idle if recipe creation, recipe optimization, and data review were performed on the integrated modules. Stand-alone auto-ADI setup tools are required to perform these tasks. Stand-alone tools can also provide auto-ADI metrology for track systems used for processing levels where features are larger and the process is well understood and controlled, or for older track systems that cannot be retrofitted to integrated auto-ADI.
 |
null
 |
null
The ideal ratio of integrated modules to stand-alone tools is fab specific. But it is critical to any fab-wide auto-ADI installation that all the metrology tools match. A defect that is detected and classified on one tool must be detected and classified the same way on all other tools in the fab, or else rework decisions and SPC data will not be consistent. Matching also enables recipes to be shared and the workflow to be modified without extensive reprogramming of the metrology system.
Matching results
In this installation, the eight integrated modules and one stand-alone system all matched in detecting the same defects, with the one exception noted for 40µm silicon defects that are smaller than specified. The ability of the tools to "color match" was also tested. The stand-alone tool was considered the reference tool and integrated modules were calibrated to match the RGB values of the reference tool using the DSW wafer. This method enables the various metrology tools to be insensitive to variations in the intensity of their light sources. After a lamp change, the same calibration procedure, which typically takes 15 min, enables the stand-alone tool (or any integrated module) to be precisely recalibrated so recipes do not have to be reoptimized. Table 2 shows the color confidence match for two product wafers with different devices; the color calibration matched to >97% on all the systems.
The nine systems were also tested for their ability to repeatably detect defects on product wafers. Two different patterned device wafers were run through the stand-alone tool and defects were detected. The die with defects were classified as defective die. The fab required that 85% of the defective die had to be detected by the integrated modules. The stand-alone tool detected 31 defective die on device wafer No. 1 and three defective die on device wafer No. 2. These defective die were accurately identified by the integrated modules more than 87% of the time (Table 3).
Fabwide auto-ADI management
Advanced technology, a well-designed configuration, and tool matching are critical enablers for integrated metrology. However, effective ADI also requires sophisticated centralized software to tie together stand-alone and integrated metrology tools and all the data they output.
Centralized software can track an entire wafer lot through various lithography steps. This allows fab engineers to correlate final yield with defects and to work back to find yield-robbing problems, even though different auto-ADI tools collected data. Tracking lots is also important as defects tend to propagate up through several layers. A defect detected on one level may be visible, and therefore redetected, for the next several levels of processing. By providing previous-level subtraction filters, the metrology can only consider defects caused by the current lithography step. Without this feature, rework decisions might be based on defects that cannot be corrected by reprocessing.
Efficient recipe management is another main function of centralized software. The auto-ADI metrology can build an initial recipe from a wafer's shot map. After the first recipe is created, higher-level recipes can be created automatically and downloaded to all ADI tools. Also, when a recipe is optimized or changed, the software can automatically issue the new recipe to all tools, instantaneously.
Centralized software also allows engineers to review data from any stand-alone or integrated tool from their desks. The same network allows all auto-ADI systems to communicate with a fab's advanced process control (APC) system so alarms can be passed to the appropriate engineer and sophisticated pass-fail criteria can be programmed into APC for distribution to all the auto-ADI tools.
Bill Welch, Hemant Amin, Rudolph Technologies Yield Management Group, Richardson, Texas
Jana Clerico, Rudolph Technologies, Flanders, New Jersey
Acknowledgments
The authors extend a special thanks to Stephen Lickteig for collecting data that made this article possible, and to David Pham and Bill Schymik for technical editing. WaferView and Knowledge Base are trademarks of RudolphTechnologies Inc.
Bill Welch received his BS in electrical engineering from the University of Oklahoma. He is manager of applications and quality control at Rudolph Technologies Yield Management Group, 1100 W. Campbell Rd., Richardson, TX 75080; ph 469/624-4636, fax 973/691-5480, e-mail [email protected].
Hemant Amin received his BS in computer science from the University of Texas, Dallas. He is an applications and quality control engineer at Rudolph Technologies Yield Management Group.
Jana Clerico received her BS in electrical engineering from Stevens Institute of Technology, and her MBA from Fairleigh Dickinson University. She is manager of marketing communications at Rudolph Technologies.