ANALYSIS AND METROLOGY—YIELD MANAGEMENT

Implementing a yield management program to improve die yield

Avraham I. Ron, Tower Semiconductor

Because it reflects the amount of product available for sale, die yield is critical to the success of a semiconductor fabrication plant. Yield is also the most important factor in overall processing cost. This article describes a yield improvement project undertaken by Tower Semiconductor, a foundry in Migdal Haemek, Israel, that produces several dozen different products, each with the relatively low volume of a few hundred wafers. Five different technologies (1.0, 0.8, 0.6, and 0.5 µm, and EPROM) are fabricated with variations of dual and triple metal. The same facility also is used to develop such new technologies as analog, 0.35 µm, and 0.5- and 0.35-µm Flash. With this complex product mix, the company needs a yield management program that operates effectively to ensure that high yield levels are consistently maintained.

The ambitious goals of Tower's yield improvement effort included not only decreasing the defect densities of existing technologies within 6 months and bringing the foundry's die yield to a worldwide level based on an entitlement database, but also setting a yield improvement infrastructure in place for all future technologies. The following sections outline the systematic approach that was taken to meet these goals and describe how a yield management methodology was successfully modified to fit the fab's needs. This article also describes how the company's goals were successfully met.

Seven Steps to Yield Enhancement

During the first half of 1996, Tower invited the KLA-Tencor yield management group to collaborate on its yield improvement project. The group analyzed the status of the fab's existing technologies and, through detailed reports, proposed a new methodology for handling yield enhancement, which evolved into the following seven-step plan: (1) gaining and broadcasting management commitment and support, (2) establishing yield expectations, (3) setting the yield learning rate, (4) establishing baseline process defect levels and goals, (5) establishing equipment defect levels and goals, (6) supplementing defect density reduction efforts with end-of-line testing, and (7) making yield improvement a part of the company culture. A flowchart of selected elements of this plan is shown in Figure 1.

Figure 1: Flowchart of the systematic die yield and defect density improvement scheme.

Management Commitment and Support. Obtaining the commitment and support of fab management are essential prerequisites for achieving and maintaining high yield rates. There are several actions management should take to broadcast the importance of yield enhancement activities to the success of the fab. For example, monthly die yield review meetings attended by all the top fab managers should be scheduled. Management should also encourage the individuals and teams who contribute to die yield improvement by posting their names on the company bulletin board and by offering financial incentives in the form of bonuses for goals met. Another essential ingredient for successful yield improvement is cooperation between yield enhancement, process engineering, and other involved engineering and production groups.

Top-level management must also recognize the importance of yield by providing the financial resources needed to implement and maintain a yield management infrastructure. Human resources allocations may need to be increased based on yield expectations and the company's overall business environment, and funds should be set aside for the purchase of capital equipment.

Yield Expectations. In Tower's case, management was already committed to obtaining yield improvements when its project began. Thus, the first active step in its effort was setting yield expectations, or, more simply, establishing a specific goal for each technology targeted for improvement. Following an analysis of the existing tool set, the KLA-Tencor process entitlement defect density database (Murphy model), containing information from more than 80 fabrication plants worldwide, was used to set the plant entitlement. This graph is shown in Figure 2. Entitlement in this application is defined as the best sustainable yield for a given technology. Once the plant entitlement was set, it was easy to determine each technology's goal. The real challenge was how to achieve better defect densities than the entitlement.

Figure 2: Process entitlement defect density graph (Murphy model) for four equipment sets: 0.65, 0.50, 0.35, and 0.25 µm. Design rules are on-reticle active poly line or space widths.

Learning Rate. When the fab had established defect density goals for each technology, the next step was to determine the pace at which the goals would be met. A monthly learning rate was set based on the defect density database and on prior experience that suggested that in order to achieve a 1% yield learning rate per month, the equivalent of three engineers would need to be deployed (recognizing that, in some cases, a highly experienced technician is equal to an engineer). This model was later verified to be realistic. Additional elements also were taken into account, including the fab's business environment (mainly customer needs and competitor performance) and available human resources, and the determinations made by the yield enhancement department manager in conjunction with process engineering, equipment engineering, and production managers regarding the resources required to achieve the yield improvement goals. The original plan called for a learning rate of 8% per month.

Baseline Process Defect Levels and Goals. Setting defect density goals will not lead to actual improvements unless engineers and technicians understand what these high-level goals mean in terms of day-to-day activities. Translating the technology defect density goals to lower-level goals was done in the following manner. Each technology was divided into two major sections, front-end and back-end, which in turn were subdivided into zones containing several process steps. For example, each metal layer was defined as a zone. Front- and back-end teams were established and given the task of defining specific defect density goals and monthly improvement rates for each zone. A zone's defect density improvement rate was based on the engineers' experience that it should be higher than the monthly learning rate for that technology.

Figure 3: Typical zone defect density tracking sheet used to define process goals.

After choosing monitoring points for each zone under its responsibility, each team tracked the defect density levels in the respective zones using patterned wafer inspection tools. All of the monitoring points were sampled on a daily basis. Figure 3 shows a typical zone process capability (Cp) tracking sheet. Figure 4 presents graphs of average defect density from several zones. Based on such data, each team defined its zone defect density goals and identified a list of projects for improvement.

Figure 4: Average zone defect density for monitoring points in four process zones: (a) poly, (b) implant, (c) BPSG, and (d) metal-2 sputtering.

Equipment Defect Levels and Goals. Improving a process's defect density level requires reducing the defect contributions of the process equipment. Part of the KLA-Tencor team's contribution to the Tower project was to perform an equipment defectivity analysis. This activity involved implementing optimized defect-monitoring procedures for all processes and tools and then collecting data and comparing them to the benchmarks in the process entitlement database. In this way, the processes and tools with the most need for improvement were identified and placed in the equipment defect density improvement plan. An important feature of the analysis was that the monitor wafers were run under actual processing conditions so that the monitoring data are representative of the environment that the product wafers encounter. Figure 5 shows the equipment defect improvement bar graph that compares the equipment defect counts taken during April 1996 with those taken in March 1997. The graph also includes the respective benchmarks from the database.

Figure 5: Equipment defect improvement graph.

The same approach taken to setting zone defect density goals and identifying improvement projects was adopted for equipment defect density improvement, except that the equipment goals were set based on the benchmark data. In order to meet the defined goals, the back-end and front-end teams identified, executed, and tracked the results of a set of agreed-on projects. The teams also continued to collect monitor data on a daily basis and ensured that all out-of-control and defect excursion lots would be analyzed and the proper correction implemented.

End-of-Line Testing. Increasing die yield is a complex activity. Although much can be done to reduce the number of killer defects in the fabrication process, product die can also fail because of a combination of parametric and design sensitivity factors. The most important element behind the well-known concept of probing both a technology's electrical parameters and die performance at the end of the line is obtaining the fastest possible feedback about product failures. Thus, an important part of the ongoing yield enhancement program is monitoring and tracking the end-of-line yield loss events and ensuring that each of these events is analyzed until the root cause is found and corrective action is taken. Performing such tracking from the early days of production has resulted in good learning rates and an increase in die yield.

Yield Improvement Culture. In order to maintain die yields, the concept of yield improvement must become part of the fab's culture. In other words, defect density reduction has to become everyone's business. Managers must make sure that all yield and defect density tracking data can be easily accessed by engineers and technicians. The most important information should be placed on company bulletin boards. In addition, at Tower the die yield performance data are presented on a regular basis to different forums of employees. One example is a weekly presentation focusing on the past week's main die yield issues along with specific containment efforts and root cause solutions. Monthly die yield steering committee meetings are also held, after which plant managers review the yield data with their staffs. All of these actions have helped create an excellent yield improvement culture at the foundry.

Figure 6: Defect density for Product A before and after implementation of the yield improvement program.

Figure 7: Monthly learning rates for Product A before and after implementation of the yield improvement program. The cumulative rate increased from <1% to 12.7%.

Yield Enhancement Project Results

Figures 6—9 show some of the results of Tower's yield enhancement project. Figures 6 and 7, which depict a defect density graph and a learning rate bar graph, respectively, highlight the remarkable improvement in yield performance for the fab's Product A that took place within a period of 7 months after the implementation of the yield enhancement project. The original goal of bringing the fab to the worldwide level (based on the process entitlement database) within 6 months was met, and the cumulative learning rate exceeded the targeted 8%, reaching a level of 12%. Moreover, adopting the systematic approach to yield enhancement also proved successful in reducing defect density for the 1.0-, 0.8-, 0.6-, and 0.5-µm technologies as the data shown in Figures 8 and 9 illustrate.

Figure 8: Defect density trends by month for three process flows following implementation of the yield improvement program.

Figure 9: Defect density trends by month for the 0.5-µm technology following implementation of the yield improvement program.

Acknowledgments

The author would like to thank all of his colleagues at Tower Semiconductor and the members of the KLA-Tencor yield management group, who made the project's success possible. Special thanks are due to Ilan Rabinovitch and Jeffry Levy from Tower Semiconductor and to Rich Martin from KLA-Tencor.

Bibliography

Castrucci P, "Utilizing an Integrated Yield Management System to Improve Return on Investment in IC Manufacturing," presented at IEEE/SEMI International Semiconductor Manufacturing Science Symposium, Burlingame, CA, 1991.

Laughlin DH, "Equipment Defect Reduction Methodology," presented at Tencor Yield Management Seminar, Geneva, Switzerland, March 1996.

Maly W, Trifilo B, Hughes RA, and Miller A, "Yield Diagnosis through Interpretation of Tester Data," in Proceedings of the International Test Conference 1987: Integration of Test with Design and Manufacturing, Washington, DC, IEEE Computer Society Press, pp xxxi, 1151:10—20, 1987.

Martin RJ, "Measuring Defect Reduction Programs," presented at Tencor Yield Management Seminar, Geneva, Switzerland, March 1996.

Nurani RK, Strojwas AJ, Maly WP, et al., "In-Line Yield Prediction Methodologies Using Patterned Wafer Inspection Information," IEEE Transactions on Semiconductor Manufacturing, 11(1):40—47, 1998.

Wong S, "Defect Isolation and Reduction Techniques," presented at Tencor Yield Management Seminar, Austin, TX, October 1995.

Avraham I. Ron is manager of the yield and electrical test department at Tower Semiconductor Israel, a position he has held since July 1996. Before joining Tower, he worked at the Intel Development Center in Haifa, Israel. His most recent position at Intel was back-end engineering manager responsible for quality and reliability, test engineering, and product engineering groups. Ron began his career as system reliability engineer at Elbit Computers in 1980. He received his BSEE from the Technion in Haifa, Israel, in 1980 and MSc in industrial management from the Technion in 1985. (Ron can be reached via fax at +972 6 6547788.)

MicroHome | Search | Current Issue | MicroArchives
Buyers Guide | Media Kit