- 學位論文
應用關聯法則於半導體表面黏著技術異常診斷之研究
Applying Association rule technique for SMT Failure analysis
摘要
半導體於現今社會是非常競爭的產業，若是產品品質無法達到客戶的要求，生產者除必須負擔退貨的有形成本外，還必須承受商譽損失以及產品形象等無形成本。因此，生產者必須針對客戶回饋的不良產品進行分析，找出廠內異常原因並改善之，才能夠持續改善品質，以保有產品的競爭力。本研究主要是利用關聯法則（Association Rule）的技術,針對既有的半導體中客戶回饋的不良訊息、失效模式以及異常原因三者間進行關聯分析，並藉以產生關聯性規則，此外，當發生新案例無法比對出結果的狀況時，研究中利用相似係數統計的手法，以找出近似的預測結果。若發生新案例比對出兩種以上結果的情形時，則統計規則子集合的各結果及其信賴度作為選定預測結果的原則。當客戶回饋新的不良訊息時，系統能自動判別，並立即找出可能的失效模式及異常原因，讓廠內人員減少找尋問題的時間，並立即針對製程作改善，減少不良品產出。
並列摘要
Semi-conductor is more competitive than any industrial field in nowadays. Producers have to listen the VOC (voice of customer) and provide after-sales service to improve customer satisfaction, otherwise they have to suffer the loss of customer returned goods and get worse image in product and enterprise. Thus, producers shall assign much human resource to find the failure root cause of failed goods and trace back to correct the issue in production line to keep advantage in this field. In order to save the human resource and analysis time, this research applies association rules technique to find the relationship between VOC, failure mode and failure root cause. Based on this we can generate associational rule to predict the root cause from VOC immediately instead of analyzing every failed goods. Sometimes new case can’t match any rule, we use Jaccard similarity coefficient to find the approximate rule and set the consequent for prediction. And if confliction rule is happened we use sub-itemsets calculation to choose the highly confidence consequent to be prediction result.