Authors:
Andrea Schirru
1
;
Simone Pampuri
2
;
Cristina De Luca
3
and
Giuseppe De Nicolao
1
Affiliations:
1
University of Pavia, Italy
;
2
University of Pavia and Infineon Technologies Austria, Italy
;
3
Infineon Technologies Austria, Austria
Keyword(s):
Semiconductors, Machine learning, Entropy, Kernel methods.
Related
Ontology
Subjects/Areas/Topics:
Industrial Automation and Robotics
;
Industrial Engineering
;
Informatics in Control, Automation and Robotics
;
Systems Modeling and Simulation
Abstract:
In this paper, a novel learning methodology is presented and discussed with reference to the application of virtual sensors in the semiconductor manufacturing environment. Density estimation techniques are used jointly with Renyi’s entropy to define a loss function for the learning problem (relying on Information Theoretic Learning concepts). Furthermore, Reproducing Kernel Hilbert Spaces (RKHS) theory is employed to handle nonlinearities and include regularization capabilities in the model. The proposed algorithm allows to estimate the structure of the predictive model, as well as the associated probabilistic uncertainty, in a nonparametric fashion. The methodology is then validated using simulation studies and process data from the semiconductor manufacturing industry. The proposed approach proves to be especially effective in strongly nongaussian environments and presents notable outlier filtering capabilities.