在本研究中，我們提交了由光學發射光譜(optical emission spectroscopy, OES)收集的復雜的及時電漿數據。在不考慮複雜因素的情況下，從一組復雜的物理參數，如氮化鋁(Aluminum Nitride, AlN)薄膜的殘餘應力獲取了一系列解決方案。AlN具有較高的應力穩定性、熱穩定性和化學穩定性，我們採用脈沖直流電濺射法在矽基板上沉積AlN。我們想要知道的一個重要答案是，沈積的薄膜的應力是壓縮的還是拉伸的。為了回答這個問題，我們可以訪問任意多的光譜數據，記錄數據生成一個庫，並利用主成分分析(Principal Component Analysis, PCA)來降低復雜數據的復雜性。經過PCA預處理後，我們試圖證明我們是否可以採用標準的人工神經網路(Artificial Neural Network, ANN)，以獲得一個足夠解析度的機器思維分類方法來區分AlN薄膜的應力類型。因此，通過這些機器學習練習，這些輔助分類可以擴展到未來其他感興趣的半導體研究。;In this study, we present complex real-time plasma data collected by optical emission spectroscopy (OES). A series of solutions were obtained from a set of complex physical parameters, such as the residual stress of Aluminum Nitride (AlN) films, without taking into account complex factors. AlN has high stress stability, thermal stability and chemical stability. AlN was deposited on silicon substrate by pulsed direct current sputtering. One of the key answers we want to know is whether the stresses on the deposited film are compressed or stretched. To answer this question, we can access as much spectral data as we want, record the data to generate a library, and use Principal Component Analysis (PCA) to reduce the complexity of complex data. After PCA pretreatment, we tried to prove whether we could use standard Artificial Neural networks (ANN) to obtain a machine-mind classification method with sufficient resolution to distinguish stress types in AlN films. Therefore, through these machine learning exercises, these auxiliary classifications can be extended to other interesting semiconductor research in the future.
