A Neural Network Based Approach for Surveillance and Diagnosis of Statistical Parameters in the IC Manufacturing Process
MetadataShow full item record
Despite advances in integrated circuits (IC) equipment and fabrication techniques, there still exist random fluctuations or statistical disturbances in any IC manufacturing facility, which can adversely affect the production yield. Actually devices and circuits are being designed with increasingly tighter parameter and performance margins. As a result, chip performance becomes even more sensitive to the statistical variations, and this may result in low production yield. Based a statistical process simulator, a methodology of tracking and diagnosing statistical variations of a real manufacturing process in a bid to implement real time statistical quality control of IC manufacturing process is presented in this thesis.<P>The main contributions of this thesis include the following. A neural network based approach for IC process diagnosis is proposed and has been realized. This approach needs a very short time in diagnosing significant variations of an IC process, hence is practical to be used in real-time monitoring and diagnosing of the process disturbances. Another contributive feature of this approach is that process diagnosis is a high dimension problem, and in our approach all variables are handled simultaneously, instead of eliminating of some variables that may have small but important contributions as in previous approaches. Other contributions include an algorithm to evaluate the fault observability and disturbance diagnosability. In addition, thresholding and coding methods are developed for pattern generation of the neural networks. A special sampling distribution is employed for simulation of samples, in conjunction with latin hypercube sampling techniques. Finally the approach is applied to a general example to show its efficiency with some experimental results.