在許多理論與實驗的文獻中,反射式散射儀已經被證實在量測奈米尺度週期性光柵結構的外貌上具有很高的準確性,但吾人認為在量測方法與技術上仍有改善的空間。而且,對於量測製作於透明基材上的光柵結構,反射式散射儀的準確性仍有待考驗。因此本論文將探討增進反射式散射儀的量測解析度的一種方式;另一方面,並採用類神經網路結合穿透式散射儀,以彌補反射式散射儀在量測透明材質的不準度。 為了改善傳統反射式散射儀的量測解析度,我們提出了結合波長及角度資訊的3D特徵圖譜,並以嚴格耦合波理論進行分析。實驗上,我們也建立一台反射式散射儀,並與商用機台比對後,發現在量測3D特徵圖譜上有很高的一致性 此外,由於嚴格耦合波理論在模擬穿透頻譜上有其不足之處,我們建立穿透式的散射儀,取得穿透頻譜,建立資料庫並利用類神經網路加以訓練而後輸入待測樣品推測其參數值(線寬、周期、高度),並與電子顯微鏡比對,發現可以得到很好的準確度。
In many theoretical and experimental studies, a reflection type scatterometer has been proved to have high accuracy when used to measure the profiles of periodic grating structures of nanometric scale. However, we think there is still room for improvement in terms of methodology and techniques. Besides, the accuracy yet remains to be validated when a reflection type scatterometer is used to characterize grating structures on transparent substrates. In this thesis, we will first verify a modified method to enhance the measurement resolution of a reflection type scatterometer. Secondly, we will adopt artificial neural network along with a transmission type scatterometer to remedy the inaccuracy issue when a reflection type scatterometer is used to measure transmissive grating samples. To enhance measurement resolution of a conventional reflection type scatterometer, a modified method called 3D signature is proposed, which includes both scanned wavelengths and angles derived from rigorous coupled wave analysis (RCWA). We also build up a reflection type scatterometer for such measurments. The experimental results thus obtrained are compared with to a commercial equipment, VASE, indicating a fairly good agreement is achieved. Because RCWA has not developed well for the analysis of transmission spectra, we build up a transmission type scattermeter to obtain transmission spectra as database, and adopt artificial neural network (ANN) for training. Afterwards, transmission spectra of test samples whose parameters such as CD, period and height are to be identified are input to the model to predict those parameters. When compared with SEM results, it is found that the parameters obtained through such method are in good agreement with SEM ones.