Skip to Main content Skip to Navigation

Statistical Learning on Circular Domains For Advanced Process Control in Microelectronics

Abstract : Driven by industrial needs in microelectronics, this thesis is focused on probabilistic models for spatial data and Statistical Process Control. The spatial problem has the specificity of being defined on circular domains. It is addressed through a Kriging model where the deterministic part is made of orthogonal polynomials and the stochastic term represented by a Gaussian process. Defined with the Euclidean distance and the uniform measure over the disk, traditional Kriging models do not exploit knowledge on manufacturing processes. To take rotations or diffusions from the center into account, we introduce polar Gaussian processes over the disk. They embed radial and angular correlations in Kriging predictions, leading to significant improvements in the considered situations. Polar Gaussian processes are then interpreted via Sobol decomposition and generalized in higher dimensions. Different designs of experiments are developed for the proposed models. Among them, Latin cylinders reproduce in the space of polar coordinates the properties of Latin hypercubes. To model spatial and temporal data, Statistical Process Control is addressed by monitoring Kriging parameters, based on standard control charts. Furthermore, the monitored time – series contain outliers and structural changes, which cause bias in prediction and false alarms in risk management. These issues are simultaneously tackled with a robust and adaptive smoothing.
Document type :
Complete list of metadata

Cited literature [116 references]  Display  Hide  Download
Contributor : Abes Star :  Contact
Submitted on : Friday, December 15, 2017 - 3:48:23 PM
Last modification on : Friday, October 23, 2020 - 5:03:27 PM


Files produced by the author(s)


  • HAL Id : tel-01438684, version 2


Esperan Padonou. Statistical Learning on Circular Domains For Advanced Process Control in Microelectronics. Other. Université de Lyon, 2016. English. ⟨NNT : 2016LYSEM009⟩. ⟨tel-01438684v2⟩



Record views


Files downloads