In-line failure detection and diagnosis of excimer laser-based microvia fabrication using computational intelligence

In this article, two computational intelligent methodologies for run-to-run, in-line excimer laser ablation process failure detection and diagnosis are compared. The techniques used are: (1) a combination of feed-forward neural networks and Dempster-Shafer theory; and (2) adaptive neurofuzzy networks. Both methodologies employ response data originating directly from the laser equipment and characterization of microvias formed by the ablation process, which serves as evidence of equipment malfunctions affecting process parameters. Both neural networks and neurofuzzy models are trained and validated based on this data. Successful failure detection is achieved in 100% of 19 possible failure scenarios using the first technique. Moreover, successful failure diagnosis is also achieved, with only a single false alarm occurring in the 19 failure scenarios. Using adaptive neurofuzzy logic, results indicate a single false alarm in 19 possible failure detection scenarios. For failure diagnosis, a single false alarm and a single missed alarm occur. Neural networks and neuro-fuzzy networks thus achieve approximately 95% and 90% success in diagnosis, respectively.