Home > Data Analysis

Fault Detection and Classification (FDC)

In modern semiconductor manufacturing with continously reducing device and process margins as well as high costs for wafer scrap the precise control of equipment status and performance is essential. The idea is that by well controlled internal equipment and process parameters, like mass flow controller opening or plasma density the outcome on wafers will be also well controlled. FDC will detect abnormalities of such key parameters within a short time allowing to stop the related equipment, eliminating scrap.

Means the main purpose of FDC is at first to detect an abnormal status of the equipment or the process running on it (fault detection - FD). The second step is then to classify the detected failure, a leak in the chamber or a problem in a RF power supply, to give engineers and technicians a hint where to start to search for the root cause (fault classification - FC). A prerequisite for this activity is to get the related data from the tool reliably. This data e. g. throttle valve position, chamber temperature, RF reflected power ... is checked by univariate or multivariate statistical or knowledge based methods.

On this site you can find several presentations and papers showing the implementation of Hotelling T2 charts (a multivariate fault detection possibility) and other techniques to monitor a via etching process. The mentioned multivariate methods require software capable of handling large matrices, in this case the open source software R was used successfully.

Furtheron you can find apc-gs68 a framework written in Python that enables hopefully the easy implementation and testing of FDC algorithms.

| News | Data Analysis and Integration | Plasma Etching | Publications | Tutorials | Function Library | Software |

Other important things: Legal Notice, Contact us

Copyright © 2001-2012 Spitzlsperger Technologies GmbH