Spatial change detectors for semiconductor manufacturing

In this paper, we address the problem of detecting process changes by monitoring spatially distributed data in semiconductor manufacturing processes. A specific question treated in this paper is how to extract information from the spatial data for change detection when we do not know what information contained in the data is useful. We adopt the idea investigated by Zamel and Hinton of using neural networks to extract information which is not known a priori. The result shows that this approach works much better than the simple mean for process changes detection.