Abstract:
The analysis of data from simulations and experiments in the development phase and measurements during mass production plays a crucial role in modern manufacturing:
Experiments and simulations are performed during the development phase to ensure the design's fitness for mass production. During production, a large number of measurements in the automated production line controls a stable quality.
As the number of measurements grows, the conventional, largely manual data analysis approaches its limits, and alternative methods are needed. This thesis studies the value of machine learning methods for typical problems faced in data analysis from engineering to mass production. In a case-study, the production of integrated circuits and micro electro-mechanical systems in silicon technology is discussed in detail. A number of approaches to salient problems in industrial application have been developed in the presented work, addressing the yield as the central figure of batch processes in silicon manufacturing:
The parametric yield is governed by a design's robustness against process tolerances. This work develops a framework for doing statistical sensitivity analysis, and robust optimization which accounts for process tolerances.
Using nonparametric Gaussian process regression, the sensitivity analysis can be performed efficiently. For computationally demanding simulations a robust optimization is eventually only made feasible through the presented approach.
Being probabilistic models, Gaussian processes allow for an optimal experimental design, thus significantly reducing the number of required simulation runs. A novel approach to active learning for Gaussian process regression is proposed in this thesis, and validated experimentally.
Besides random failures, as captured by the parametric yield, systematic errors in the production can lead to additional losses.
It is hard to localize the root cause for previously unseen losses, as physical interrelations can hardly be reconstructed in complex manufacturing facilities, and as there is usually a large number of potential sources for the error.
This work shows that, using feature selection, data from quality
checks can be combined with data from manufacturing to construct an
automated localization mechanism.