SEMICON EUROPA

camLine together with Elisa IndustrIQ will be represented by Kalle Ylä-Jarkko, Senior Data Scientist at Elisa IndustrIQ.
In his presentation, you’ll get to know how information obtained from images and time series can be significantly increased by integrating machine learning as an integral part of the current SPC framework.

Abstract:
Traditionally, the detection of Out-Of-Control (OOC) signals as well as the stability of the process flow in semiconductor manufacturing is based on control charts using aggregated data such as mean and range, or standard deviation. Information loss is already generic in extracting only parts of the raw data, especially in complex cases like images and time series. This makes the detection of critical events limited to rule-based automation. Thus, this is where cognitive automation comes in with pattern recognition. Nowadays, multi-dimensional data that carries tens and hundreds of variables have exceeded the limitations of the human observer and current root cause analysis techniques. To solve the problem, camLine aims to introduce machine learning into the SPC framework. The enhancement will combine historical data, inline/online process data, and machine learning algorithms, highlighting the most probable cause from Normal Operation Conditions (NOC). To simplify and automate the analysis of wafer maps in semiconductors, the machine learning techniques create a system that can extrapolate from any set of measurement points to create comparable wafer maps across different measurement sites. Unlike deep learning-based solutions, our system can be trained with relatively small datasets. Only a few hundred wafers from any mix of product designs will suffice. Even for big volume wafers, the system acts efficiently with a snap of response time on wafer scoring, typically within one second.