May 29, 2013

PhD, Machine Learning, LIRIS Laboratory Lyon, DL. 30.06

The Data Mining and Machine Learning group (DM2L) at LIRIS laboratory (UMR 5205 CNRS) invites applications for a PhD position in Machine Learning.


This is a European project entitled "Integrated Solutions for Agile Manufacturing in High-mix Semiconductor Fabs" (INTEGRATE) with 28 European partners in which one PhD student will be funded for 3 years. The INTEGRATE project aims to enhance European semiconductor fabs efficiency by providing methods and tools to better control the process variability, reduce the cycle time and enhance the effectiveness of the production equipment. The PhD student will develop powerful machine learning algorithms for analyzing large unbalanced data sets including sensor data streams (at varying temporal resolution), selecting and extracting features, assessing their relevance and performing both fault detection and fault classification in a supervised or semi-supervised context.

Work environment

Lyon is France's second largest city and capital of the Rhone-Alpes region. Combining an exceptional historical heritage with a natural liking for good food, Lyon is an ideal city for discovering all the charm of the French way of life. A stage for more than 2000 years of history, the city has a remarkable architectural heritage. Expanding
towards the east throughout the centuries, without destroying the existing areas, 500 hectares of its city centre became a Unesco World Heritage Site in 1998.

University Lyon 1 is one of the leading academic communities in France. Renowned for its leafy campus (443,000 m2), and state-of-the- art equipment, it enrols over 35, 000 students in approximately hundreds of study programs. University Lyon 1 employs 2630 researchers and teachers.

The Data Mining and Machine Learning group (DM2L) at LIRIS laboratory (UMR 5205 CNRS), focuses on the development of principled approaches to machine learning and data mining, and their applications to diverse areas including bioinformatics, anomaly detection, forecasting, process monitoring, medical diagnosis etc. DM2L
currently consists of 12 researchers and 5 PhD students.

Project description

State-of-the- art semiconductor processes are often pushed to the limits of the current technology, resulting in processes that have little or no margin for error. Increasingly there is a need for fast, accurate, and sensitive detection and classification of equipment and process faults to maintain high process yields and high throughput in manufacturing. Detection of process and tool faults in the shortest time possible is critical to minimize scrap wafers and improve product yields for semiconductor manufacturing. Recently there has been a move towards fault detection directly on the manufacturing tool, through monitoring of tool-state data and in-situ process-state sensor data. Many manufacturing tools are beginning to have the ability to collect a large amount of data in real-time, which can be then accessed for tool fault detection. Unfortunately, with such an abundant amount of
data available, it is often difficult to extract useful information such as when the tool is ! no longer operating properly (i.e. detecting a fault). Over the past years, a number of machine learning techniques have emerged to process large amounts of data and convert them to useful information. Several of these are starting to gain usage for tool fault detection. Once the fault is detected, the fault must be classified for some assignable cause, which ultimately must be corrected before the tool is returned to normal operation. Classification of the cause of the fault is equally critical to detecting the fault, because rapid classification and corrective action will lead to minimized tool downtime and increased throughput.

The PhD candidate will focus on automating the fault classification with powerful machine supervised and semi-supervised learning approaches (e.g. graphical models, SVMs, neural nets, ensemble methods) for automatically classifying faults based on historical data and identifying some fault signature. The overall aim is to detect and
classify faults faster and more accurately, resulting in improved process yields and higher throughput, while controlling the false alarm rate.

What we expect from you

You should meet the following requirements:
- A Master's degree (or equivalent) in Computer Science, Electrical Engineering or Statistics with a strong interest in machine learning, pattern recognition and data analysis;
- Strong programming skills in R or Matlab;
- Good knowledge in probability and statistical inference;
- Commitment and a cooperative attitude;
- Excellent proficiency in written and spoken English.

If you are interested in this position, please provide a detailed curriculum vitae, a short explanation of your interest in the proposed research topic, a list of courses (including grades) that you have successfully completed, a publication list, copy of your publication( s) in English and the names of two references, and all other information that might be relevant to your application

Please send your application by mail not later than June 30th to:

Prof. Alexandre AUSSEM,
Email : aaussem@univ- Phone: +33 (0)4 26 23 44 66.