| dc.description.abstract | The purpose of this investigation is to demonstrate value for Auto Defect
Classification (ADC) for patterned wafer inspection systems within a high volume
manufacturing fabrication in the Process Limited Yield (PLY) defect area. Process
excursions in all functional Unit Process (UP) areas, examples are of etch, litho,
diffusion, are monitored by PLY.
Troubleshooting of process excursions using added defect density count with a
small percentage (random or largest 50 examples) of and inline Scanning Electron
Microscope (SEM) data classification review does not give a clear indication of the full
wafer data. Statistical Process Control (SPC) trigging on total counts or defect density is
not as powerful as making excursion decisions on classified data from ADC (Fisher,
2002).
The ADC data gives classification of the entire wafer rather than a smaller sample
making signature analysis to be an additional troubleshooting tool. The inline ADC data
does not have near the resolution of the SEM but can be used to help make important
decisions to what is occurring in the manufacturing line. The interest is to gain a full
understanding of the current capabilities and limitation of ADC and to apply the learning
to enable faster reaction and visibility into process and tool excursions within a high
volume manufacturing fabrication.
The Technical Learning Vehicle (TLV), high running product layer at the leading
design rule, there were approximately 10,000 wafers a week with 1000 wafer die (chips)
per wafer. A sustained improvement in yield of 1% across the entire manufacturing line
would equate to almost 1 million dollars a month of saving. With the ability to tightly
control multiple etch process tools, the resulting yield improvement was 3% across 15%
of the line. With the baseline yield improvement along with ability to react quickly to
process excursions, the combined improvement resulted in excessive of 5 million dollar a
year of reoccurring savings. | |
