| Summary: | 碩士 === 元智大學 === 工業工程與管理學系 === 96 === It has highly correlation between TEG(Test Element Group) electric measure value on Array’s process and Array NG Rank what is abnormal defect on Cell’s process.
The panel be judged as Array NG Rank 4 will be scrapped.It can reduce the cost waste resulted from unnecessary material be input in scrapped panel if we can predict what’s Rank in Array NG precisely.Therefore,the article’ topical subject research in construct prediction model of the abnormal defect by analyzing electric character on Array’s process.It will be divide into two parts to proceed.
In first phase,by different method:Multinomial Logit model and a decision tree algorithm C5.0 to find out the better predictive rate in predicting Array NG’s Rank and induce the key factor from the inside of TEG(Test Element Group) electric measure value: Ion,Ioff,Vth, Ufe,Idl,Gm,Rgl,Rs1,Rc1,Rge1,Rge2, Rsd1 and Rsd2. In second phase,i will construct prediction model of the abnormal defect by analyzing electric character on Array’s process to understand that TEG electric measure value influence the degree of Array NG Rank’s judgement.
According to analysis result in article,in first phase, the models were evaluated based on the predictive accuracy rate for test sets.The Multinomial Logit model had better predictive rate(96.17%) than the predictive rate(96.10%) of C5.0 model and the key factors included Vth,Rs1,Rge2,Rsd1 and Rsd2.In second phase,the prediction model be constructed by analyzed the correlation between TEG electric measure value and Array NG Rank can have higher ability to predict Array NG Rank judged as 4.The ability can reach to 96.15% and it can save the cost includes NT 840,600 dollars and 45 hours in work per month.
|
|---|
