Yield Management: Present and Future
Fourmun Lee Motorola, Chandler, Arizona -- Semiconductor International, 3/1/2000
|
At a Glance |
|
|
|
|
|
||
|
|||
Yield management has emerged as an area of increasing importance, playing a central role in new technology development, technology transfer into manufacturing, and continuous improvement. Product cycles are getting shorter with each new generation of chip technology. With the selling price of a new product declining more rapidly after introduction, the goal is to send product to market as early as possible and with the highest yield possible to maximize revenue. An effective yield management program is essential to meeting this goal.
Yield management encompasses many different fab activities, including process and device development, process and device integration, process tool management, process control, line monitoring, and real-time problem solving. A successful yield improvement program requires personnel in all areas to support yield-related activities as a top priority. Decision-making in all these areas relies on collecting and extracting information from the immense volume of process and electrical test data generated during device fabrication. Rapid data collection and analysis minimizes response time to yield and process problems, and reduces the time needed for problem identification and resolution, ultimately reducing die loss.
The volume of data generated in a modern fab is growing rapidly as metrology instrumentation improves in speed and capability, and factory throughput increases. Fab personnel spend considerable time and effort collecting the data needed for yield improvement activities. Efficient and effective data collection requires having the appropriate measurement tools as well as the software and infrastructure to store and retrieve measurement results. Once the desired data is collected, it has to be formatted for analysis.
In many cases, much of this work still requires collecting data from many separate, incompatible databases and tabulating data manually. Extracting the data from the various sources has required knowledge of multiple software languages and applications running under different operating environments, and correlating the data has required literal overlay of printed wafer maps. These approaches are cumbersome and time consuming. Networking measurement tools, storing measurement results in shared databases, and implementing a well-designed yield management system (YMS) to allow access to relevant data through a common interface can effectively address a fab's yield management needs (Fig. 1).
|
| 1. Data essential to the support of the manufacturing operation, including yield management efforts, should be stored in central databases accessible throughout the factory. |
The essentials
Essential elements in a yield management program are strategy, staffing and tools.1 First and foremost is the establishment of consistent and coherent organizational goals and priorities. The goals must be well defined and supported by top management, and departments within the organization must establish strategies to achieve them. Departmental goals must be aligned with organizational goals, with support of yield management established as a priority. The individuals in each department must then support yield management activities as one of their top priorities. The continuous improvement mentality, once established, provides the foundation for success.
With the goals and strategy in place, adequate staffing and tools are needed. Ideally, an autonomous group should be established, with the charter of monitoring performance of the manufacturing line, identifying excursions and improvement opportunities, and leading and helping with problem-solving activities. This group also should perform specialized tasks such as initiating problem-solving activities, leading and evaluating process improvements, assessing yield and reliability impact, and supporting tool qualification and process development.
A variety of inspection tools are available2-4, and the tools selected for a particular fab depend on the application and sensitivity needs as well as the tool budget. Design of the sampling plan is based on historical data and confidence in each process step, and sampling requirements can vary significantly from factory to factory. Since there is a cost associated with this activity, sampling plans often represent a balance between inspection tool capacity, cycle time and staffing, among many other considerations.
Data storage and retrieval are very important considerations in a yield management strategy. Using automated metrology tools and electrical test equipment, data can be generated at an astonishing rate. But that data has no value if it cannot be stored, easily retrieved and analyzed. An infrastructure must be established to access, analyze and correlate data from multiple hardware and software systems. Connecting metrology and test equipment via a fab-wide or company-wide network, and storing the data generated in one or more dedicated databases for later retrieval, will allow rapid retrieval and visualization of the data.
What it takes now
|
| 2. Staff, metrology, analysis tools, databases and the software tools to tie them together are key elements of a successful yield management system. |
At a typical high-volume fab, effective yield management requires skilled people, metrology tools, analytical tools and software tools. All these elements must be integrated to form the yield management system (Fig. 2). Inadequate resources in any area can seriously impact yield improvement efforts.
The yield enhancement team is responsible for real-time, in-line inspection of product in process; monitoring and characterizing process anomalies (not limited to defects); identifying and investigating issues; communicating findings; and participating in problem solving. The team can perform a key role in troubleshooting as well as evaluating the process output (i.e., what happens on the wafer when a process change is made). Relying solely on electrical test results, collected at the completion of processing, is inefficient.
Table 1 lists the primary responsibilities of yield enhancement personnel. For a typical production fab, adequate staffing must be available to provide continuous yield enhancement coverage (24 hours/day, 7 days/week). It is assumed that other operational functions are staffed correspondingly and are available to respond.
| Table 1. Yield Enhancement Staffing and Responsibilities | |||
| Responsibility | Operator | Technician | Engineer |
| Run inspection tools | X | X | X |
| Review and classify defects | X | X | X |
| Basic data analysis | X | X | X |
| Communicate results | X | X | X |
| Expedite material | X | X | X |
| Project team support | X | X | X |
| Generate inspection recipes | X | X | |
| Support inspection operations | X | X | |
| Disposition lots | X | X | |
| Investigate new issues | X | X | |
| Lead project teams | X | ||
| Yield analysis | X | ||
| Inspection strategy | X | ||
| Tool development | X | ||
The type and quantity of metrology tools used in fabs varies greatly, depending on the age of the fab, wafer size, product mix and technology level of the parts manufactured. Measurements are tailored for individual unit processes and are used to monitor the well-being of the process and perform statistical process control (SPC). In addition, particle and process-induced defects must be monitored on every process tool. Most of this testing has been performed on test wafers, but on-product measurements are now used whenever possible to reduce test wafer cost and consumption of valuable tool capacity for non-value-added processing. For throughput, sensitivity, capacity and cost considerations, a combination of different wafer inspection tools is used to service specific applications5-6.
Wafer inspection tools employ two basic detection techniques: brightfield and darkfield. Both types can be used effectively for general line monitoring applications. Brightfield tools are generally slower but more sensitive, while darkfield tools are generally faster but less sensitive.
Darkfield tools are preferred for most unpatterned wafer applications, and are commonly used for on-product monitoring for processes such as film deposition and chemical mechanical polishing (CMP). Brightfield tools are the tools of choice where there is a need to detect all defects generated within a process module.
It is essential to review defects after inspection to classify them and save the classification and defect image information in a database. A combination of optical- and SEM-based review tools are used to classify and characterize defects. Recently introduced automatic defect classification (ADC) technology improves throughput and accuracy of defect inspection and review, but it is still relatively immature. Work is in progress at several equipment suppliers to improve ADC speed, accuracy, and integration to hardware and software tools.
Data storage and retrieval presents unique challenges due to the diversity of fab data. Organization of the data tends to be site specific and is highly dependent on the database, network infrastructure and software tools available at each site. To facilitate data collection and retrieval, a common network backbone should connect tools that generate data and all databases used to store data (Fig. 1). This setup allows all non-image data to be accessible from remote terminals or workstations located in the fab or office area, though not necessarily using a common interface or analysis application. The image database can be accessed through dedicated terminals in analog format or through remote terminals in digital format.
It generally is undesirable to store the different types of fab data in a single database due to database size, access time, and differences in the nature and usage of the data. Table 2 lists some commonly encountered data types and their physical associations. Defect data typically is stored in a defect database system, which also acts as the focal point for wafer defect inspection and review activities. The data must be readily retrievable in the fab for use in line monitoring, process qualification and tool qualification activities.
| Table 2. Common data types and their physical associations | ||
| Data Description | Physical Associations | Data Type |
| Yield | die, wafer | continuous |
| Bin | die, wafer | discrete |
| Bitmap | die, wafer | discrete |
| Process | die, wafer, lot | continuous |
| Parametric | die, wafer, lot | continuous |
| Defect | event, die, wafer | discrete |
| Image | event | analog or digital |
| WIP | lot | discrete |
Third-party analysis applications are useful for common analyses like the ones shown in Figure 3. In this example, a common signature (center stripe) can be seen in the in-line defect map, composite bitmap and bin map. Defect, bin, bitmap and parametric data can contain unique signatures7,8. The on-wafer location and orientation of a signature can serve as a fingerprint for a process or process integration problem and help identify the defect origin.
|
| 3. Signature analysis, trend analysis and correlation of visual defects to electrical failures are common fab data analyses. In this example, a common signature (center stripe) appears in the in-line defect map, composite bitmap and bin map. |
Other areas where yield management systems have performed well are defect partitioning and trend analysis. Using inspection data from two or more process steps, defect partitioning identifies new defects added by each process step, defects that became permanent, and defects removed by subsequent processing. An integral part of this analysis is the grouping of defects according to a pre-defined classification scheme. The physical characteristics of each defect often provide information about the defect's origin.
Using a YMS, trend analysis can be performed on all types of data with the push of a few buttons. The data can be filtered by a multitude of criteria set by the user. If a desirable trend is found, the YMS can help pinpoint the change(s) that resulted in the improvement. Alternatively, if an undesirable trend is found, the YMS can help in determining what went wrong. Other types of analyses are possible using a YMS, but this potential has not yet been fully realized.
What it will take
Current state-of-the-art fabs are processing 200 mm wafers at a rate of 6000 or more wafers per week. The transition to deep sub-micron geometries and 300 mm wafers will bring a substantial increase in die/wafer output, resulting in increased data generation and data storage needs. Effectively analyzing these "mountains" of data (data mining) will require better software tools to improve data integration and database management and reduce analysis time.
|
| 4. Overlay of in-line defect inspection data with electrical bin data correlates die failures with visual defects. Colored stars represent different types of visual defects. Die color coding indicates electrical test results (green dice are good; all other s failed). |
Several yield management systems are available commercially, but all fall short or miss the mark with respect to some of the required capabilities. In my experience each system offers some desired features, but none provides all. Key issues include system speed, difficulty with data set selection, difficulty with performing the desired analysis, limited capability and difficulty of use (too many options, not enough options). The systems available today are deficient in the areas of correlation capability, flexibility and automation.
Visual/electrical correlation is an important yield improvement and failure analysis tool.7 It is a powerful tool for reducing failure analysis time and increasing failure analysis efficiency. Analysis tools need to provide visual representations of the correlation results, such as the in-line defect to bin data overlay map shown in Figure 4. In this figure, colored stars represent different types of visual defects. Die colors indicate electrical testing results, with green indicating a good die. Locations where colored stars coincide with non-green dice indicate a correlation of die failures to visual defects.
With sufficient data, graphical analysis of correlation data can identify major failure modes. Figure 5 is an example of an analysis of bitmap data for four wafers from the same lot. Figure 5A shows the correlation of failed bits with adder defects for six inspection steps, and Figure 5B presents the same data with correlated failures grouped by failure mode. Analyses of this type allow identification of specific failure mechanisms prior to physical analysis, reducing the time and effort spent on physical analyses. The system should allow the user to output the coordinates of selected correlated failures and uncorrelated failures. This coordinate information would be used by analytical instruments to locate specific defects or electrical failures of interest and confirm the cause of device failures.
|
| 5. Correlation of bitmap failures with in-line defect data helps to identify failure modes. Figure 5A shows the percentage of failed bits correlated with adder defects for six inspection steps. Figure 5B presents the same data with correlated failures grou ped by failure mode. |
Correlation of yield and parametric data is difficult and limited in scope with current tools. If device performance is sensitive to changes in specific parametric values, yield loss can be associated with shifts or variance in device characteristics such as Leff, Vt and contact resistance. There is a need for software that can detect shifts or changes in variance appearing across a wafer, wafer-to-wafer, or lot-to-lot.
One example of the potential power of a yield management system is the investigation of low-yielding lots. A typical investigation involves a series of time-consuming steps (Fig. 6). Collecting and tabulating the data for an investigation can take several hours or more, excluding failure analysis work.
With the help of a YMS, data extraction can be performed in one hour or less, excluding failure analysis work. This reduces the response time to address issues and allows fab personnel to focus their attention and efforts on solving problems. The YMS would provide access to yield, defect, WIP and even parametric data. Fab personnel can extract and view yield and defect trend charts, defect maps, bin maps and more from a single interface. Correlation can be performed if the required data exist.
|
| 6. Low yield lot investigation involves many steps, and it can be long and tedious without an integrated data management system. |
How do we get there?
The most effective approach for addressing analysis needs is through partnership of suppliers with knowledgeable users. The users would participate in development of the product by providing the supplier with specific information about their analysis requirements, and providing the data and environment to test new tools as they are developed. Detailed knowledge of the user's analysis methodologies early in the product development cycle would aid the supplier in designing tools that extend and enhance the user's thought process rather than forcing the user to alter the thought process to fit the tool. Such an approach ensures that the end product provides the interface, features and capability desired by the users. At the same time, it guarantees a ready market for the finished product.
Although the major suppliers of yield management solutions have taken a modular approach to software design, modules developed by any given supplier are seldom compatible with those from another supplier. This would not be a concern if the modules provided by one supplier performed all the functions desired by a particular user or fab community. In practice, a software suite from one supplier seldom can service all of a user's needs. This presents the end user with the difficult task of exporting data from application to application, which may be cumbersome at best and impossible at worst.
In the personal computer market, standardized interfaces such as the PCI bus or USB port have benefited both users and suppliers, encouraging technology adoption and market growth. Adopting standardized module interfaces for software tools can provide similar benefits to software developers and users in the yield analysis market. With this concept, each supplier can still design its own database formats and analysis tools, while allowing users the freedom and flexibility to choose the database and software tools most suited to their needs.
A yield management system is essentially a suite of proprietary software applications linked to one or more databases. Most Internet sites today consist of Web-based applications attached to one or more databases. A natural extension of the standardized interface concept is the transformation of proprietary YMS applications into Web-based visualization tools. Users finally would be able to access all required data from a single user interface, and suppliers would have increased market and demand for their products. •
Fourmun Lee, Ph.D., is a principal staff scientist at Motorola. He has B.S. degrees in chemical engineering and materials science from the University of California, Berkeley, and received his M.S. and Ph.D. from the University of Illinois, Champaign-Urbana. His current work focuses on process integration and yield enhancement in silicon device manufacturing. He has 12 years experience in semiconductor processing development and yield enhancement, has published 29 technical papers and holds three patents.
Motorola Inc., MOS12 Die Manufacturing, 1300 North Alma School Road, Chandler, Arizona 85224
email: Fourmun_Lee@email.sps.mot.com
REFERENCES
-
P. Wang, F. Lee, K.M. Chan, R. Goodner and R. Ceton, "Part II: Yield Enhancement in a High-Volume 8-inch Wafer Fab", Semiconductor International, July 1996, pp217-22.
-
K. Seshan, Handbook of Thin Film Deposition Processes and Technology: Principles, Methods, Equipment, and Applications, Noyes Publications (2nd Ed., to be published).
-
S. Stokowski and M. Vaez-Iravani, "Wafer Inspection Technology Challenges for ULSI Manufacturing", International Conference on Characterization and Metrology for ULSI Technology, NIST, (March 23-27, 1998).
-
F. Lee, "Defect Metrology for the 21st Century," Future Fab, December 1998.
-
P. Wang, F. Lee, K.M. Chan, R. Goodner and R. Ceton, "Part I: Yield Enhancement in a High-Volume 8-inch Wafer Fab," Semiconductor International, June 1996, pp 221-6.
-
R. Ceton, R. Goodner, F. Lee and P. Wang, "Comparison of Patterned Wafer Defect Detection Tools for General In-line Monitors," Proceedings of the IEEE/SEMI Advanced Semiconductor Manufacturing Conference, pp 92-99 (1996).
-
F. Lee, P. Wang and R. Goodner, "Yield Improvement through Signature Analysis and Visual/Electrical Correlation," Proceedings of IEEE/SEMI Advanced Semiconductor Manufacturing Conference, pp 272-5 (1995).
-
F. L.ee, A. Chatterjee and D. Croley, "Computer-Based Spatial Pattern Analysis," Proceedings of IEEE/SEMI Advanced Semiconductor Manufacturing Conference,pp 409-15 (1996).
