品質異常與對策之知識管理：以TFT-LCD製程為例

Abstract

TFT-LCD產業的技術變革很快，新產品和新製程不斷引進。再者，由於人力的快速流動，使得經驗累積與交接的時間愈來愈短，企業想要從量產之初就有高良率高產出，就必須從過去的產品製造中吸取經驗及教訓，亦即將資深人員的經驗知識，有系統的轉化成便於檢索與存取的資訊系統。本研究擬以一TFT-LCD個案企業為例，探討如何對品質異常的資訊建立一知識管理系統。 本研究所討論的個案企業原已設有一套品質異常改善流程，此流程係針對過去TFT-LCD製程所發生的品質異常進行解析，找到根本原因，進而提出矯正及預防對策，最後寫成書面報告，以藉此紀錄品質管理的知識。然而由於新製程不斷引進，以書面報告方式保存品質管理知識，最後將造成資料量相當龐大，新進人員不容易檢索出相關知識，以即時解決問題。 本研究提出一套知識管理方法，期能將品質異常與矯正的知識，有系統的轉化成便於存取的資訊格式。本研究是應用階層式分類(Hierarchical Classification)，先將品質異常分類，亦即每一種異常可視為一個向量；然後利用類似程度檢索(Similarity Match for Retrieval)的機制，比對向量的類似性，以快速檢索出相關的資訊，期能使有效利用過去的製程經驗，進而使新產品的良率能快速提昇。

The manufacture technology of TFT-LCD frequently changes due to continuous introduction of new product and new process. Knowledge of experienced engineers is thus very important asset for tuning the process yield, in particular in cases when experienced engineers resign and leave the company. This research attempts to develop a case-based reasoning (CBR) system for help junior engineers to acquire the knowledge of senior engineers in order to quickly solve the quality issue problems. The basic idea is by encoding the quality symptom and the responsive action described in each prior quality-resolving report. That is, each prior quality-resolving report is represented by a pair of vectors: (1) symptom vector, and (2) responsive vector. And all such vector pairs are stored in a database. Then, while a quality symptom of a new process appears, junior engineers could encode the symptom and retrieve most relevant quality reports from the database, as a guide for solving the quality problem. Each responsive vector includes the identification of the process where the quality symptom appears. Therefore, the other application of such a database is to help junior engineers know all the possible quality symptoms that might appear in the process he is in charge of.