應用先進製程控制虛擬量測技術於LED磊晶製程波長預測

Abstract

磊晶製程(Epitaxy)為LED的首要製程，影響產品良率、產能與市佔率。磊晶波長的量測是判定磊晶製程成功與否的關鍵，一但偏離目標波長太多，則表示磊晶機台對於製程控制的能力需要調校，或須排除人為因素以提昇磊晶製程穩定性。因此，波長量測在LED製造流程有其重要性。 然而波長量測常因人為失誤或量測機台當機導致產能利用率的下降，因此，本研究參考半導體廠先進製程控制(APC)的經驗，提供虛擬量測(VM)的解決方案於LED產業，使用決策樹、類神經網路以及最近鄰居法等三種方法進行研究評估有效的波長虛擬量測方法，以期縮短實體波長量測的等待時間、即時回饋機台所需之量測資訊、提高產能利用率與降低製程所需成本。

The Epitaxy is the primary process of LED, in which the stability of Epitaxy may affect the quality of end production, capacity of factory and market share. The result of wavelength measurement is the criteria to make sure the Epitaxy process is successful or not. Once the measuring result is too far away from the target wavelength, it implies that process controlling ability of the Epitaxy tool needs to be tuned, or some human errors need to be corrected for increasing the stability of Epitaxy tool. Consequently, the wavelength measurement is important over the LED manufacturing process. Wavelength measurement is often interrupted because of human errors and the crashes of measurement tool, which leads to the decline of the tool utilization ratio. This research proposes to apply virtual metrology to predict wavelength of LED Epitaxy Process based the Advanced Process Control (APC) in semiconductor industry. This research compares three virtual metrology methods, including Decision Tree, ANN and kNN methods, and evaluates their effectiveness in predicting the wavelength of LED Epitaxy process. The proposed virtual metrology can shorten the waiting time of the batches for wavelength measurement, respond feedback information rapidly, raise utilization ratio of production and reduce manufacturing cost.