Tool sensitivity analysis using neural net technique for yield improvement

In this paper we introduce the methodology of using neural networks (nonlinear regressions) as a yield ramp technique for new product introduction. Within a wafer fabrication facility, a large set of manufacturing equipment is used in sequential mode for IC manufacturing. As the technology scales, it is becoming more difficult to detect influential factors associated with specific equipment during a yield ramp. Using multiple iterations of neural networks on subsets of the yield ramp data, we have successfully isolated equipment sets that are most likely to influence yield very early in the yield ramp phase. This allows for a significantly shorter yield learning time.