低溫多晶矽薄膜電晶體元件特性分布於用電路模擬之研究

Abstract

多晶矽薄膜電晶體(poly-Si TFTs)基於其優於非晶矽薄膜電晶體(amorphous silicon TFTs)的電流驅動能力，最近在液晶顯示器(AMLCD)及有機發光二極體(AMOLED)顯示器的週邊電路整合應用上皆備受矚目。在本文中，我們將對低溫多晶矽薄膜電晶體(low temperature poly-Si TFTs)的元件特性作一統計性的研究。對於在固定距離下兩元件間特性如臨界電壓(threshold voltage)及遷移率(mobility)之差異，會做進一步的討論。這些元件間差異行為的變異性(variation)分布將可以用我們所提出的數學模型加以描述，取代之前所廣泛採用的高斯分布。而在這些我們所提出函數對於實際量測到的分布之比較中，經過回歸分析所得之回歸變異係數(R square)皆在0.95之上。此一結果代表我們所提出的變異性的模型與實際分布情況十分吻合，也反映出該模型的適用性。更進一步的，本文所提出的模型會用於在積體電路中常用到的差動對(differentia pair)電路與電流鏡(current mirror)電路之模擬。我們將可以從模擬的結果之中，了解電路上元件間的變異性對電路性能產生之影響。而從模擬結果，我們發現電晶體變異性行為，也可以藉由定義原始資料的四分位差值來描述，與函數並無絕對相關，並且也可以得到一個與真實分布類似的分布。因此我們在一般商用模擬軟體中，仍然可以採用高斯分布來進行電路模擬，並可以得到比過去更加精準的模擬結果。

Low Temperature Polycrystalline Silicon (LTPS) thin film transistors (TFTs) have attracted much attention in the application on the integrated peripheral circuits of display electronics such as active matrix liquid crystal displays (AMLCDs) and active matrix organic light emitting diodes (AMOLEDs) due to its better current driving compared with a-Si (amorphous silicon) TFTs. In this thesis, the variation characteristics of LTPS TFTs are statistically investigated. The differences of the threshold voltage and mobility with the same device distance are further studied. The difference shows the distribution much centered than the Gaussian distribution and a proper model is proposed to describe the variation behaviors with difference device distances, for which the R squares (Coefficient of Determination) are higher than 0.95, reflecting the validity of the model. Furthermore, the proposed models are used to simulate the performance of the differential pair and current mirror circuit, which are commonly used in VLSI. Simulation results show the effects of the variation behavior on the estimation of the circuit performance. Besides, from the simulation results, it is found that the Gaussian distributions defined by the inter-quartile range of parameters difference data have a good fitness for the real data distribution compared with Gaussian distribution defined by the standard deviation. Therefore, Monte Carlo analysis with Gaussian distribution still can be used to simulate LTPS TFT circuits in simulation tools. Furthermore, the circuit simulation results will be more accurate than before.