Semiconductor manufacturing processes are prone to process deviations or other production issues. Quality assurance of every processing step and measuring wafer test values is crucial for finding possible root causes of these problems. Automated visual inspection and recognition of patterns in wafermap data obtained during different processing steps has a potential to signifficantly improve the efficiency of finding early production issues and even help with adjustment of the production parameters to automatically resolve them. In this paper, we present a machine learning approach for unsupervised clustering of spatial patterns in wafermap measurement data. Measured test values are first pre-processed using some computer vision techniques, followed by a feature extraction based on variational autoencoders to decompose high-dimensional wafermaps into a low-dimensional latent representation. Final step is to detect the structure of this latent space and assign individual wafers into clusters. We experimentally evaluate the performance of the proposed method over a real dataset.
In Proc. of IJCNN'18, the 2018 International Joint Conference on Neural Networks, Rio, Brasil, July, 2018, IEEE.
*This work was partially supported by the NSF-Frontiers Cyber-Cardia
Award, the US-AFOSR Arrive Award, the EU-Artemis EMC2 Award, the
EU-Ecsel Semi40 Award, the EU-Ecsel Productive 4.0 Award, the
AT-FWF-NFN RiSE Award, the AT-FWF-LogicCS-DC Award, the AT-FFG
Harmonia Award, the AT-FFG Em2Apps Award, and the TUW-CPPS-DK Award.