Consultants Engineer Reduction in Test Wafer Usage
Yield management purchases no longer just pay for themselves in theory.
Laura Peters, Senior Editor -- Semiconductor International, 3/1/1998
During semiconductor industry slowdowns, fabs often intensify cost-reduction activities. Programs that can reduce test wafer usage, from ratios as high as 2:1 (test wafers to production wafers) to as low as 0.5 to 0.3, are possible. TEFEN (Foster City, Calif.), an industrial engineering firm focused on improving manufacturing efficiency in the semiconductor industry, uses a sophisticated test wafer model designed to make such reductions possible.During initial phases of productivity improvement, the company builds a comprehensive database including work-in-process (WIP) management to show the flow of test wafers and how the information gathered from the wafers is being used. Breakdowns of defectivity levels and the flow can be used to optimize test wafer usage and redesign a process flow for maximum efficiency. According to Haim Albalak, president of TEFEN, often changes in the route of the test wafers, definition of usage and changes in work procedures are needed to dramatically reduce test wafer usage.
The company's Test Wafer Management System (TWMS) determines test wafer requirements by facility, functional area and monitor. Specifically, the TWMS can help determine the quantity of test wafers used in a period; the effects of changes in qualifications on the usage of test wafers in the facility; the number of test wafers that are reclaimed, repolished or scrapped; and the number of new test wafers needed.
Usage can be reduced
from 2:1 ratios to 0.5-0.3.In order for the transition from 200 mm to 300 mm wafers to be successful, large reductions in test wafer usage from present levels must occur. TEFEN's engineers are developing fab layout methodologies and staffing analyses for 300 mm fabs. The company is working with Lockwood Greene (Spartanburg, S.C.) on engineering aspects of 300 mm fab design and with Bechtel (San Francisco, Calif.) on virtual reality modeling. Such modeling allows users to see how the 300 mm fab will operate, including material handling, staffing requirements, tooling requirements, automation, etc.
TEFEN further specializes in WIP management, overall equipment efficiency (OEE) improvement, cycle time management, yield vs. on-time delivery analyses and cost modeling. The user-friendly models allow a variety of "what-if" analyses to be performed by a variety of fab personnel in only minutes. Visit www.tefen.com for more details.
Evaluating the Benefits of Yield Improvement Investments
One of the most challenging cost analyses the fab manager faces is evaluating the cost savings associated with implementing or upgrading the investment in yield management tools. Using a modified cost-of-ownership model, engineers from KLA-Tencor (San Jose, Calif.), SEMATECH (Austin, Texas), AMD (Austin, Texas) and Stanford University (Stanford, Calif.) effectively estimate the value of using tools to improve yields. The model accounts for the time value of yield management investments. Use of the model was demonstrated at the IEEE/SEMI Advanced Semiconductor Manufacturing Conference in mid-September.
The new value of ownership model accounts for benefits caused by faster yield excursion detection and long-term increases in yield enabled through the use of the inspection tools. Discounted cash flow and net present value models can be used to compare purchases of machines that best reflect a fab's current and future capacity needs to determine the combination that provides the best value to the fab. The analysis includes an approach to solicit the probability distribution of the yield outcomes from fab yield management groups.
The example showed a six-month payback
period for an investment of $2.2 billion.An effective yield management system reduces the percentage of excursion lots through more in-line inspection steps, higher in-line sampling frequency, accurate and integrated defect classification and analysis, and fabwide yield analysis. The example modeled by the engineers in the project showed a six-month payback period for a yield management investment of $2.2 billion, with annual operating expenses of $250,000. The fab starts 5000 wafers per week (200 mm), processes 300 die per wafer, sells the die for $20 each, has line yields averaging 95%, has average in-control die yields of 75% and has average die yield at excursions of 55%. Importantly, the defect reduction team must estimate the probability of reducing the percentage of excursion lots. In this example, the expected revenue increase because of a reduction in the percentage of excursion lots is $379,000 per month. The one-year ROI is 196%.
The model goes one step further to estimate the value of investment for yield learning through baseline improvement. The model relies on fab engineers' experience to estimate performance improvements associated with the purchase of different inspection tools. In cases where investment values can be explicitly modeled, for example, with automatic defect classification systems investments, results can be readily incorporated into the model.
