This report summarizes the results of a beta site evaluation performed on KLA's Segmented Auto-Threshold technology, more commonly known as S.A.T., in the TI's DP1 development facility form MArch to August 1996. This technique was primarily designed to eliminate the effect of nuisance defects at the metal etch levels and thereby allow the KLA machine to improve its sensitivity at these levels. Two S.A.T. recipe optimization techniques, basic and customized, were compared to the standard mean/range techniques being used today in manufacturing at contact etch, metal-1 etch, and metal-4 etch. The basic S.A.T. technique is a more simplified technique which is designed to be more easily implementable in a manufacturing environment but would probably not demonstrate as good a sensitivity as customized recipes. Customized recipes offer the promise of much better sensitivity but optimization is extremely involved and not readily implementable in manufacturing at this point. Results indicated that customized S.A.T. recipes captured several times more defects than seen by the traditional mean/range recipes at the metal etch levels.Conversely, basic S.A.T. showed little improvement over mean/range recipes at these types of levels. However, for contact etch only, basic S.A.T. did show significant improvement, finding almost three times as many defects as mean/range. For the metal etch levels, the significant increase in the number of defects caught with the customized S.A.T. recipes was observed for almost every defect type, including a very important category called 'shorts', which are particles whose positions on the die indicate a high probability to cause a probe failure. In addition to comparison data for defects detected on wafers randomly sampled form production lots at each of the 3 process levels, repeatability data rom dedicated product wafers ar also presented for both customized S.A.T. and basic S.A.T. recipes. The standard deviations of the repeatability data for both the customized and the basic S.A.T. recipes were as good or better than those seen with the mean/range recipes. Non- S.A.T. data are also shown for both before and after the S.A.T. hardware/software installation which verify no loss in system sensitivity due to the changes.
|