Fault isolation during semiconductor manufacturing using automated discovery from wafer tracking databases

Describes the use of automated discovery from databases for diagnosing the causes of misprocessing during semiconductor manufacturing. The database contains the history of the semiconductor wafers as they undergo various processing steps. A generate-and-test approach is taken for using such a database for automated diagnosis. Based on prior manual use of such databases, classes of queries to the database useful for fault isolation are identified. Patterns in the responses to these queries that are useful for fault isolation are also identified. Automated diagnosis is accomplished by automating query generation and the detection of potentially useful patterns. A prototype system was implemented and tested on a database from a wafer grinding and polishing facility. In addition to identifying previously known faults, the system also identified previously unknown faults