面對電腦及通訊產品功能的快速提昇,以及其多元化與輕巧化的品質需求,大部份只針對單一品質特性製程最佳化做探討,然而工程人員在實際的製程中碰到的往往是兩個或兩個以上的多重品質特性問題,一般處理多重品質特性問題只能依靠工程人員的經驗來選擇因子水準,而由於工程人員的認定標準不一,容易產生不確定性與模糊性,且品質特性數目愈多其品質之相關性愈強,同時帶給工程人員解決品質特性衝突上之難度。 本文針對陣列錫球封裝製程中,主要探討BGA金線銲接製程多重品質特性最佳化之製程技術,先計算個別品質特性損失,其次將各品質損失標準化,利用統計軟體SAS進行主成份分析,並選取特徵值大於1的主成份,來代表原有品質特性,此主成份亦為最佳化參數組合。在結合羅吉斯迴歸、倒傳遞神經網路模式及支援向量機,三種分類方法進行資料的分類以選取最佳模型,進而改善金線焊接製程能力。
Facing the fast promotion of the computer and the communication product function, and its multiplication and the dexterous quality demand, the major part only aims at the single quality characteristic optimization to make the discussion, however the engineers often encountered two or more than two multiple quality characteristic question, generally the multiple quality characteristic questions can only depend upon the experience of engineers to choose the factors, but because the engineers identified different standards, have the uncertainty and the fuzziness, and the number of quality characteristics are more relevance of its quality to be stronger, simultaneously takes to the engineers to solve in the difficulty of the quality characteristic conflict. In this paper, array solder ball packaging process, investigate the multiple quality characteristics optimization of BGA gold wire bonding process, first calculating the loss of individual quality characteristics, and second, the standardization of the quality loss. Using statistical software SAS to calculate principal component analysis, and select the eigenvalues of the main components is greater than 1, to represent the original quality characteristics, the principal component is also optimized parameters. In combination with logistic regression, back propagation neural network models and support vector machines, three kinds of classification methods for data classification to select the best model, then improve the ability of gold wire bonding process.