機器視覺之馬可夫算程－半導體銲線製程之應用

積體電路封裝品質的良窳，銲線是一關鍵製程。目前IC封裝業的銲線機已全面自動化，同時也配備了機器視覺系統。在晶墊間接定位的視覺系統環境，銲線的生產力，大量依賴圖鑑教讀和圖樣辨識的科技及技術。然而，縱使視覺系統善於辨識，封裝製程的銲線能力，仍然與晶體數位化影像的「清晰程度」息息相關。晶體二值影相是從按由數值影像轉換而來。就技術而言，遑論整批，即使同片晶圓上各粒晶體其影像亦不盡相同，導致現有的各種最佳門檻技術都不適用於IC封裝產業。探索影像處理的機遇本質，運用馬可夫鏈之狀態和遷移概念，經由辨識編態與門檻搜尋方式的設計，本論文發展出乙套質優的馬可夫算程。對IC封裝業的銲線製程，該算程除了能輔助視覺系統提高辨識生產力外，尚可因銲墊定位準度的改善，而提昇銲線的品質能力。

關鍵字

IC銲線；銲墊定位；二值影像；門檻灰度；馬可夫算程

並列摘要

Wire-bonding is one of the critical processes that determine the quality of IC packaging. Nowadays outstanding wire-bonders are all automatic and they are equipped well with machine-vision either. With the indirect pad-positioning system, the bonding productivity depends heavily upon technology as well as techniques in pattern recognition. Even if the vision system is good at recognizing, the bonding capability is still strongly related to the ”clearness” of the binary die image. The binary die image in the wire bonding process is converted from its digital die image according to the gray threshold. Technically, for the determination of gray threshold, all optimization methods existing are not appropriate for the IC packaging industry. The reason is that the digital image of each individual die varies at all, even of the same wafer instead of the same lot. In order to overcome the aforementioned positioning problem, this paper is to explore the stochastic features of the die images, and then to develop some robust Markovian algorithms aided to pad-positioning both effectively and efficiently.