Multivariate process modeling: the ´preprocessing´ challenge

Semiconductor manufacturing of integrated circuits produces gigabytes of data detailing the manufacturing processes of the many steps involved in the production process. While this appears to be a large quantity of data, analysis of this data differs from data mining techniques developed for business sales information, mark-et-basket analysis, and spatial data because of the large number of variables and comparably small number of observations. A device involving 22 layers of semiconductor can involve 500 processing steps over 3 months with over 10⁵ process variables. The measurements are commonly collected on different aggregations of parts at the chip, wafer, lot, and batch levels. Since the measurements for a particular chip are spread out over time, collected at different aggregation levels and are many with respect to the yield data, current data mining and analysis techniques do not readily enable modeling of semiconductor manufacturing environments. The processing of the original data into a minable form is a difficult part of the process. The target of this work is at the system operational level and is to extract the full benefits of the data in even the most sophisticated processes to improve operations-that is to improve productivity, decrease ramp-up time, and identify and validate quality control parameters. This talk summarizes our work with a memory chip fabricator to build a ´golden signature´ of the production process-a multivariate system-level model