In manufacturing, data sets tend to be high-dimensional, with a low number of labels and, features show spurious correlations with respect to a target key performance indicator. As a consequence, costly manual feature engineering by domain experts is required prior to prediction. To improve this process, we propose an interactive feature engineering scheme based on dimensionality reduction. Low-dimensional embeddings of selected features are visualized and guide the domain experts towards effective feature engineering. We show that by engineering features we obtain higher predictive capabilities and we improve the interpretability of the model.

Controlling the operations and resolving product performance issues in today's high-tech production systems, such as semiconductor fabs, becomes a cumbersome task, even for experienced field engineers. To address the pressing need for assisted diagnostics approaches, in this paper we propose a model-based step-wise methodology, based on domain-specific languages and Bayesian networks, to capture domain knowledge and allow automated and guided reasoning in complex end-to-end diagnostics flow. We illustrate the methodology components and show its applied strength in a real industrial setting of semiconductor production chains.

In semiconductor manufacturing, continuous on-line monitoring prevents production stop and yield loss. The challenges towards this accomplishment are: 1) the complexity of lithography machines which are composed of hundreds of mechanical and optical components, 2) the high rate and volume data acquisition from different lithography and metrology machines, and 3) the scarcity of performance measurements due to their cost. This paper addresses these challenges by 1) visualizing and ranking the most relevant factors to a performance metric, 2) organizing efficiently Big Data from different sources and 3) predicting the performance with machine learning when measurements are lacking. Even though this project targets semiconductor manufacturing, its methodology is applicable to any case of monitoring complex systems, with many potentially interesting features, and imbalanced datasets.