Reactive Sampling for Efficient Defect Source Identification

Defectivity control in semiconductor manufacturing is crucial to improve the product quality and to reduce the production cost. When defects are detected, the objective is to identify the tool that generates them. Tool commonality analysis (TCA) is believed to be an efficient method for defect source identification. The critical element of successful TCA is that multiple entry lots within the same process flow are available for the analysis. Hence, since deployment of dynamic sampling strategies, lots selected for inspection are taken from different process flows at different manufacturing levels. In these cases, the TCA cannot identify the defect source but a set of potentially faulty tools. This paper consists in increasing the number of inspected lots that can be used for the TCA. The approach is a combination of two methods. The first one is the comparable lots identification processed by a sequence alignment algorithm. The second method is the reactive sampling algorithm based on the set covering model, this algorithm consists in sampling the minimal number of lots that cover the maximal number of potentially faulty tools. The industrial experiments of the proposed algorithms show a significant increase of number of available lots for analysis.