- 學位論文
微影製程對位精度之即時調整系統
Real-time Adjustment System for Lithography Alignment Accuracy
摘要
微影製程中，曝光機在執行基板粗對位(coarse alignment)時的誤差精度準確與否，將影響生產的產量及順暢性。粗對位誤差越小越好，不佳的誤差將導致對位時間過久，增加生產的時間成本。粗對位誤差過大時，甚至將無法對位，需要以人工方式確認處理，增加品質風險。若處理人員判斷錯誤，更有可能造成曝錯，造成產品圖形偏移。機台內機構的精度誤差，會因為長時間而累積。目前業界尚無可以針對曝光機粗對位誤差進行長期監控的系統，亦無法立即發現異常來調整補正。當發現問題時，需要針對各個單元一一澄清，若無法即時修正並改善，長期下來將損失很多產能。 本論文參考Nikon4.5代曝光機的對位系統。首先收集機台生產資料，分析對位誤差與對位時間的關係；並收集歷史露光中心補正調整與對位誤差的相關資料，以倒傳遞類神經網路演算模式，訓練並建立出調整補正量與粗對位誤差間的關係，改善現行靠人力判斷的方式，減少補值後的誤差，增加調整後的對位精度，維持一個良好的狀況，增加機台生產的順暢性。 在製程監控方面，套入多變量統計分析的概念，模擬出五項對位誤差數據之間的影響關係，訓練出較佳的管制規格，以此規格監控實際生產時的粗對位誤差，提早發現異常點並即時處理。 本文研究粗對位誤差產生的原因，建立出精確可靠的誤差估測方式，以解決目前業者無法監控，免於大量產品對位時間過長的情況發生。
並列摘要
In lithography, the accuracy of the coarse substrate alignment (coarse alignment) will affect the production yield and smoothness. It is better to have the error of the coarse alignment as small as possible. Otherwise, more time will be wasted and the time cost of the production will be larger. The alignment may even fail if the error of the coarse alignment is too big. In this case, the production needs to be judged by human, increasing the risk on quality. For example, if a handler makes a mistake in his judgment, wrong exposure may occur, resulting in graphics offset. Overall, the precision error of the machine will accumulate as time goes on. However, no companies now have a system to monitor the error of the coarse alignment of the exposure for a long period of time. Besides, the error could not be adjusted immediately. In other words, as a problem is detected, it needs to be clarified according to each unit. The lateness of the correction will result in a loss of capacities. We used Nikon 4.5 Optical Alignment System of exposure machine for this study. At first, we collected data of production machine to analyze the correlation between alignment inaccuracy and the time of alignment. Furthermore, we gathered the information about the adjustment of alignment error from the exposure center. We then used the back-propagation neural network to model the relationship between the adjustment and coarse alignment error. In this way we could not only improve the disadvantage of human judgment, but also reduce the error after correction and increase the alignment precision after adjustment. Therefore, the machine could be kept in good situation and could preserve the smoothness during production. In the process monitoring, we used multivariate statistical analysis to simulate the relationships among the five items from the errors, and to develop a better control specifications, by which we could monitor the coarse alignment during production. Thus, we are able to detect the error in the first place and to solve it promptly. This research aimed to find the reason for the inaccuracy of coarse alignment, and to build an accurate estimation of the error, so that companies could prevent their products from spending too much time in alignment.