New Approaches to Yield management

08/01/1997

New approaches to yield management

Mary Swedberg

Yield management is changing. Simple particle monitoring is no longer sufficient for the yield management of fast complex chips where performance is limited by the electrical characteristics of the interconnect system.

Defect data management solutions, which automatically gather and analyze data from defect review tools, are more prevalent than even a year ago, but these solutions address only one part of today`s yield challenge. As designers push the limits of the chip performance envelope, manufacturers must tighten the control of the electrical characteristics of interconnect materials.

Parametric test gains importance

Before deep submicron technology, monitoring extrinsic defects using an optical microscope was sufficient, as only larger particles affected the functional operation of the chip. As feature sizes decreased, the size of killer defects decreased, but the activities of yield programs remained pretty much the same - routine maintenance of equipment could reduce defects based on a typical pareto (80/20) approach, and systematic use of test wafers was sufficient to predict process results - affording some degree of stability. But particle counting will only get us so far.

Electrical properties will begin to dominate yield issues because economic yield is no longer just a matter of whether or not the chip works, but rather - does it work at the speed (or other performance criteria) for which it was designed. For example, both 50-MHz Pentium and 300-MHz Pentium chips may work functionally, but clearly the real money is in the chip`s performance - at the same or similar cost of production.

The intrinsic characteristics of the interconnect system, which can limit a chip`s performance and parametric testing, are becoming as important as the number of particles in determining yield. The "parametric yield" of the interconnect system is based on the electrical properties of the materials that comprise it. By correlating the data from testing various parametric parameters, like sheet resistance, leakages, and threshold voltages, and monitoring that data over time, it`s possible to predict the chip`s performance when it gets to functional test.

Parametric test results correlated to bin sort and other data can be used to improve the overall performance of the chips produced. Comparing parametric data, for example, with end yield, production, and WIP tracking information - such as lot history, performance metrics of equipment, and recipes - will allow fabs to characterize their processes more easily in order to pinpoint the cause of yield variations and the sources of problems. Links to parametric results with bin sort data will allow fabs to optimize performance because adjustments in processing and electrical measurements can be compared to performance-based yield metrics.

Closed-loop solutions

However, the integration of this data adds to the industry`s information management crisis as the industry moves to 0.18-0.25 ?m and 300-mm wafer processing. Smaller device features, denser patterns, and higher performance circuits mean considerably more data is being collected at each inspection step. A typical fab with weekly wafer starts of 6000 currently generates about 2.5 Gbytes of data/week. Existing in uncorrelated islands, the amount of this data that can be analyzed manually and comprehended by humans is statistically insignificant, and corrective action for yield excursions is far too slow to be economically feasible.

Integration of data along the entire process flow - from design through process to final test - will be necessary to turn information into insights on how to successfully design, process, and test effectively and economically for deep submicron technologies.

To make it even more difficult, the industry is held hostage to individual equipment company proprietary data standards that are maintained to increase profits through companion sales of data extraction and converter systems. Yield managers need open data standards that will make information integration easier and will eventually drive equipment companies to cooperate. Closed-loop systems can collect data, relate it to end yield, and feed information back to process development and design.

You may recall the United Negro College Fund advertisements from several years ago showing a picture of a young person with a chain and padlock around his head and concluding with the tag line "A mind is a terrible thing to waste." The objective of the advertisement was to get people to make a donation in the belief that there was a great deal of untapped potential in our youth. I am not sure why, but that advertisement made an impression on me and has shaped my expectations of the people with whom I work.

When I joined Eaton`s Semiconductor Equipment Operations ten years ago, senior managers frequently speculated that employee teams, rather than isolated, functionally organized groups, would improve the quality of our finished product, ion implantation equipment. Though not in Operations at the time, I thought it obvious that this structure would also allow team members to make contributions to operational performance for which there was no avenue in the traditional organization. Being able, permitted, and, in fact, encouraged to make meaningful contributions is fundamental to job satisfaction.

A few years ago, these early ideas started to become reality. Teams, made up of members with a single skill, formed around major modules of the ion implantation machine. The teams were responsible for the quality, cycle time, and schedule attainment for the entire module they owned. Finished modules were delivered to the teams` internal customer for integration and commissioning.

In less than three years, defects found by the teams` internal customers dropped by two thirds. Team members investigate every defect, implement corrective action, and keep a database to analyze systematic or other causal factors and improve their delivered quality continuously. Cycle time dropped to a quarter of pre-implementation length and continues to drop. Team members look for ways to sequence their work to minimize total calendar time. They actively contribute to organizing and staffing multiple shifts. Schedule attainment, though recently affected by the surge in the capital equipment market, improved by a factor of 30%. Team members know the ultimate customer for whom they build their modules and take seriously their responsibility for meeting that customer`s requirements.

At the outset, we expected improvements in assembly quality, but we had no way to estimate the improvement in advance of establishing the teams. We are most pleased with the results of conversion to cell manufacturing and team management in the area of product quality. Cell teams began receiving feedback on their quality from the group commissioning the integrated modules starting immediately after team formation. A team member reviews the problem log, notes the team`s issues, researches the root cause of the problem, ensures that a corrective action plan is initiated, and has the information recorded in a database. The database is analyzed by the team for trends, systemic problems, or other issues through the use of time charts and paretos. The team determines which problems to work on, performs root cause analysis, and contacts the appropriate support group - e.g., manufacturing engineering or supplier quality engineering, or even their own training spoke - to develop solutions.

The database is also used to develop team quality metrics. A few months after the teams formed and became familiar with their quality capability, they negotiated their team goals for quality (and other) metrics with their Cell Coach. Fifty-five percent of an individual team member`s annual performance is based on the team`s record on quality, safety, cycle time, and on-time.

A cell team member`s day is significantly different from a day spent working in a functional department. In today`s environment, a cell member works with a planning spoke, another member of the team, to determine priorities for the day. A cell member participates in spoke teams (safety, planning, quality, process, or training) and takes responsibility to ensure that the rest of his or her team is following the processes established by the spoke teams. Team members may also be assigned to new product launch teams to facilitate the introduction of new assembly and test operations into the cell. Presentations to the team are standard practice and generally use metrics to keep everyone informed of progress toward team goals. to maximize factory profitability. However, the challenge of linking the data from all the process steps is one that is much too large for any one company to accomplish.

Join the team

A recent collaboration among a number of companies supplying tools to IC manufacturers is taking proactive steps to develop this integrated approach to data management. In cooperation with IC manufacturers, member companies of the Yield Management Team (YMT) have agreed to share equipment and software specifications in order to provide interoperability among their tools. These integrated tools will evolve over the next few years, but YMT members Electroglas and Keithley Instruments have already announced new tools that enable the seamless integration of data to central yield management systems.

Open to all equipment and software suppliers in the semiconductor manufacturing industry, the YMT will effectively define data integration requirements across the whole fab. Solutions will encompass software, hardware, applications, and consulting services to maximize yield - resulting in improved fab profitability.

This is our call to the industry to join the YMT in this cooperative effort. The input of every company, whether as a member or an adviser, will improve the results for our collective customers and help develop factory-wide integrated information systems. For details, call me at ph 408/988-0600.

Tom Sherby is president of Knights Technology, a wholly owned subsidiary of Electroglas. Knights Technology, 155 North Wolfe Road, Sunnyvale, CA 94086; ph 408/988-0600, fax 408/739-4438.

Mary Swedberg is plant manager at Eaton Corporation`s Semiconductor Equipment Division; ph 508/921-0750,fax 508/927-3652.