The parallel SNN-based manufacturing yield prediction model

In the production line of hard disk drive (HDD) manufacturing, the machine parameters directly affect the production yield. The problems on the manufacturing line are called the root cause. By accurately identifying the root cause, we can suggest solutions for yield improvement. This research focuses on the design of an effective parallel algorithm for prediction required at the end of analysis in order to validate the suggested solutions by simulation. From previous experimental results, it can be concluded that the multiple regression method has a high error rate, which can lead to faulty predictions. Also, HGA yield prediction is proved to be non-linear. Therefore, we employ Stochastic Neural Networks (SNNs) for yield prediction in this problem domain. Genetic algorithm is used as the learning algorithm instead of backpropagation in order to handle the non-linear and stochastic relationships between input parameters. Our SNNs-based prediction model gives favorable prediction results with very low error rates. However, our version of SNNs is highly compute-intensive. In order to improve the performance, parallel algorithms are applied to all procedures in our variation of SNNs-based prediction model. The parallel algorithms and performance are described in this paper.