本論文提出一新式晶圓定位方法，應用機器視覺及影像處理技術，僅需利用一張影像即可計算晶圓在機械手臂之相對位置。將雷射光經由柱狀透鏡展開成一線段光線並投射於散射面上，架設CCD (Charge-Coupled Device)照相機於此散射面之上方。當特殊設計之機械手臂傳送晶圓通過該線段光線上方時，將部分遮斷線段光線。使用CCD 照相機擷取散射面上被遮斷之光線影像，再利用影像處理與所發展之演算法，即可得到晶圓與機械手臂之相對位置。 在取像參數設定方面，實際拍攝擴束光線影像，以影像灰階值不飽和為原則，選用適當的光圈及曝光時間，再由MTF (Modulation Transfer Function)曲線決定最佳之鏡頭工作距離，並針對鏡頭畸變現象進行校正。在定位實驗部分，先以較小的四吋晶圓進行靜態定位實驗，決定系統誤差之校正方法與實驗流程。之後以八吋晶圓進行靜態定位實驗，決定八吋晶圓定位實驗之校正參數，最後再進行八吋晶圓動態定位實驗，修正曝光時間後，成功擷取移動中之晶圓與手臂影像並進行晶圓定位計算，所得之平均定位誤差約0.37mm，誤差之標準差為0.18mm。最後亦說明實驗誤差來源以及改善系統準確度之建議方法。 This thesis proposes a novel method for detecting wafer position on a robot blade on the basis of machine vision. The wafer positioning system comprises two line-shaped beams spreading toward a scattering surface on the way of wafer transfer. The images taken from a CCD camera mounted above the scattering surface are analyzed, and the position of the wafer center on the robot blade can be determined after image processing. The lens parameters are predetermined, in order to obtain a better precision of positioning results. The aperture and the exposure time are chosen on the grounds that the gray-level values from images of spread beams are unsaturated. The working distance is determined based on the MTF curves. Besides, the lens distortion is also corrected. The procedure of the positioning experiment and the calibrating approaches of the system errors are determined in accordance with the positioning experiments of 4" wafers. Then, the calibrating parameters for 8" wafers were calculated through static positioning experiments of 8" wafers. Finally, real-time positioning experiments are successfully performed. The positions of wafer and robot can be estimated from the images captured when the wafer and robot are in motion. The experimental results reveal that the average error is about 0.37mm and the standard deviation of error is about 0.18mm.
