Electrical parameter control for semiconductor manufacturing
MetadataShow full item record
The semiconductor industry is highly competitive environment where modest improvements in the manufacturing process can translate to significant cost savings. An area where improvements can be realized is reducing the number of wafers that fail to meet their electrical specifications. Wafers that fail to meet electrical specifications are scrapped, which negatively impacts yield and increases manufacturing costs. Most of the existing semiconductor process control research has focused on controlling individual steps during the manufacturing process via run-to-run control, but almost no work has looked at directly controlling device electrical characteristics. Since meeting electrical specifications is so critical to reducing scrap a fab-wide electrical parameter control scheme is proposed to directly control electrical parameter values. The goal of the controller is reducing the variation in the electrical parameters. The control algorithm uses a model to predict electrical parameter values after each processing step. Based on this prediction the decision to make a control move is made. If a control move is necessary, optimal adjustments for the subsequent processing steps are determined. The process model is continually updated so that it reflects the current process. A simple implementation using a least squares model is first proposed. Simulations and an industrial case study demonstrate the potential improvements that can be achieved with the algorithm and the limitations of the simple implementation are discussed. A partial least squares modeling and control algorithm combined with missing data algorithms are proposed as enhancements to the electrical parameter control algorithm to address many of the issues faced when implementing such a control strategy in real manufacturing environments. The enhancements take the input variable correlations into account when making control moves and utilize the correlation structure to make better model predictions. Simulations are performed to determine the effectiveness of the enhancements. A cost function formulation and a Bayesian based alternative are also presented and evaluated. The cost function implementation uses a different method to determine the optimal set points for the subsequent processing steps than the other implementations use. Simulations are used to compare the cost function formulation with the other methods presented. The Bayesian implementation addresses the stochastic nature of the manufacturing process by dealing with the probabilities of events occurring. A simulation of the Bayesian algorithm is preformed and the algorithms limitations are discussed.