隨著半導體製程邁入ULSI (超大尺寸整合) 階段與產業隨競爭者的增加,並在幾乎所有製造廠皆擁有類似的製造技術與設備下,使得半導體產業成為高度標準化的產業。大量的生產以降低成本,造成供過於求的情況,也使得價格也嚴重下跌,因此,想要在低價的情況下獲利,除了降低生產成本外,嚴格的製程控制已成為重要目標。 目前大多數半導體廠其製程配方建立與調整的依據大多來自前人累積的經驗以及試誤法來調整,這種方法需要投入大量的人力以及時間,調整後亦需要測機以及產品試產,以現行半導體廠幾乎接近自動化的趨勢來看,投入大量人力無疑導致成本增加,更是降低在同業間的競爭力。 化學機械研磨在半導體製程中運用了大量的耗材,並有著製程不穩定性與性能偏移之特性,為能有效控制這種特性,避免突發異狀所帶來之製造成本上的損失以及人力的耗費,本論文運用徑向基底函數網路﹙Radial Basis Function Network,簡稱RBFN﹚以及田口基因演算法﹙Taguchi Genetic Algorithm,簡稱TGA﹚,針對化學機械研磨製程中所特有的製程特性進行分析,並經由調整配方成分、控制,來修正各種發生的干擾與誤差,讓量測延遲對產品造成的損失減到最低,在不增加設備與人力成本的狀況下,使設備、製程工程師們可以經由產品研磨後的變化,及時發現非預期性的問題,大幅減少產品報廢的發生,對於降低人力的負擔以及提升產品的良率有很大的幫助。
As the semiconductor manufacturing strides forward into the ULSI (Ultra-large-scale integration) fabrication and the competitors increase with similar technology and facilities, the semiconductor industry has become a highly standardized industry. However, the prices of semiconductor products dramatically decline because of outstripped supply from mass production. Therefore, the manufacturers must reduce their production cost in order to gain a profit. Stringent process control is the hinge. In most semiconductor companies, the development and adjustment of process recipes are based on either experience or method of trial and error, being a labor-intensive and time-consuming procedure. After the adjustment of process recipe, the trial productions are also required. For the automatic production in semiconductor industry, the requirement of manpower for process development and adjustment reduces manufacturer’s competitiveness. Chemical mechanical planarization (CMP) is a semiconductor manufacturing process using a lot of consumptive materials, such as polishing pad and slurry. This process is unstable due to drift characteristics. To effectively control this process and to avoid unexpected costs arising from defect products, Radial Basis Function Network (RBFN) and Taguchi genetic algorithm (TGA) was used in this study to analyze the process characteristics of CMP. By means of adjusting the process recipe, this proposed method can control the CMP process subject to disturbance and metrology delay. The process engineers can discover any abnormalities immediately from the polished products, without increasing equipment and manpower. Substantially, the scraps can be reduced. Moreover, it is also helpful for decreasing the burden of labors and promoting the product yield.