Large scale data analytics with application to wafer defect detection

Abstract

According to Semiconductor Industry Association (SIA), the global sale rates of semiconductor products in the years 2012–2013 indicated an increase of hundreds of millions of dollars, and a further growth was witnessed in 2014. Such sale records put a huge amount of pressure on the manufacturing companies in order to produce a 100% defect free, functional wafer lots in order to be able to satisfy the huge industrial demand, and minimize the yield losses as much as possible, since even a small percent of manufacturing failures could cost the companies massive losses. Machine learning techniques were proposed to encounter such problems, as they provide an automated solution to identify and classify the root causes of manufacturing failures, giving the companies the chance of fixing the faults occurring during any of the production phases. Hence, this thesis adds to the existing literature by targeting this field of research. We propose novel algorithms capable of identifying various patterns of wafer defects, and accurately classifying them in order to trace back the point of manufacturing failure within the production line. The proposed algorithms namely randomized general regression (RGR) and simplified subspacing regression (SSR) networks outperformed all other stateof- the-art Machine Learning-based models, in terms of model accuracy and ability to fit the model, achieving 99.82% and 99.44%, and 99.88% and 99.91% for RGRN and SSRN, respectively.

Date of Award	2015
Original language	American English
Supervisor	Paul Yoo (Supervisor)