多色光激發光電流量測系統的實現與神經網絡在材料特性分析的應用

Abstract

在許多材料特性的量測領域裡，光學激發技術是十分重要的，如定型化半導體中的激發子與晶格震動。然而在非晶半導體中有著更加複雜的能階與光致激發途徑。為了深入探討這些複雜的機制，新的光學激發技術與數據分析模型是不可或缺的。 在此篇論文裡，針對新的光學激發技術，我們在實驗室中開發並架設多色光激發光電流量測系統。此套系統是基於現有的雙波長雷射激發光電流量測系統的拓展與延伸。此套系統的優勢為此系統可同時以十台不通波長的雷射來激發樣品；除此之外，雷射的參數調控皆可由電腦完成，因此此系統亦可執行自動化多參數光電流量測，並建構龐大的數據資料庫。 在此篇論文裡，針對新的數據分析模型，我們在多年未演進的光電流相位分析技術的基礎上引入了卷積神經網絡來近似出以往無解析解的鎖相光電流。此數據分析模型在未來有潛力在無人協助下，針對實驗數據，提出可能的能階分布。 我們提出的多色光激發光電流量測系統在非晶矽的鎖相光電流實驗中發現了未曾被觀測過的共振現象，並自證其重要性；我們提出的數據分析模型成功的解釋了以往模型無法解釋的現象，並自證其重要性。 此論文在複雜能階的材料檢測上開闢了新的實驗方法與數據分析模型。

Optical spectroscopy is a powerful technique for the characterization of collective phenomena, such as excitons and vibrations, in crystalline semiconductors. More complex excitation processes and the analysis of amorphous materials, however, are more challenging for optical spectroscopy, due to the importance of interactions between extended electronic states. We here realize a new spectroscopy technique that utilizes large numbers of simultaneous excitations to investigate the transition between extended defect states in amorphous silicon. Our approach is an extension of existing dual-beam photocurrent spectroscopy which generates large datasets. Our second advance is the automated analysis of the photocurrent spectra created by the simultaneous utilization of 10 laser sources with a variable phase difference and amplitude ratio. A neural network is built by the essence of the evolution operator and the concept of weight sharing in the convolution neural network. This neural network can deliver possible LASER-induced band transitions without any hypothesis from researchers. This function is proven useful by solving the resonance effect of modulated photocurrent in a-Si which has not been previously achieved. Our results open up a route toward materials characterization beyond the simple semiconductor picture.