关键词: 半导体生产线, 成品率预测, 多智能体, 模糊聚合, 支持向量回归

Abstract: Aiming at problem that the existing methods needed to consider a lot of defect data and rarely estimate prediction range in traditional yield prediction model,based on Multi-Agent System (MAS),an approach which combined fuzzy rules intersection method and Support Vector Regression (SVR) was proposed to predict the semiconductor yield.Through reducing the prediction rang gradually and adjusting parameters such as learning rate coordinately by agents,multiple fuzzy yield learning models were constructed.The fuzzy forecasts of multiple fuzzy learning models was aggregated with fuzzy rules,and the predicted accuracy was improved.SVR was adopted to defuzzify the fuzzy yield prediction.Theoretic analysis and experiments showed that the proposed method was available,the forecasting process was relatively simple as well as the prediction range was more accurate than the existing methods.

Key words: semiconductor wafer fabrication, yield prediction, multi-agent, fuzzy intersection, support vector regression