國立臺灣大學應用力學研究所楊照彥2006-07-262018-06-292006-07-262018-06-292002-07-31http://ntur.lib.ntu.edu.tw//handle/246246/21647本文第一部份主要目的為以分子動 力模擬來探討影響金屬薄膜成長及大馬 士革製程中通道填充問題之相關參數， 在模擬過程中以二維沉積模式配合 Lennard–Jones Potential 來代表Cu 原子 間的交互作用力，並採用Gear 預測修正 法來求解原子運動軌跡。另外假設基板 溫度為恆溫，也就是採用速度修正的方 式來模擬入射粒子所釋放出的能量與被 基板吸收之間的關係。不同的物理參數 如溫度、入射動能、沉積率、及高溫迴 流過程等對薄膜形貌的影響亦有系統的 探討。從模擬結果得知，溫度及入射動 能效應扮演著相同角色，當高溫或高入 射動能皆有助於降低孔隙佔有率及均方 根粗糙度，沉積率方面，需要是當地搭 配入射動能才會得到理想的薄膜，也就 是之間存在一最佳分佈。而高溫迴流唯 一種對通道快速升溫及降溫的回火程 序，目的是來增加沉積原子的遷徙能力 因而舒緩薄膜中的孔隙產生。第二部分 則為提昇分子動力模擬之計算效益，採 用訊息傳遞介面為基礎的分散式計算來 處理模擬過程中最費時的分子間交互作 用部分，由結果得知，分散式計算可使 運算時間由原本正比於粒子數平方O(N²) 大幅度地降低至O(N²/p)及達到線性加速 的滿意結果，其中p 為處理器的數目。The purpose of this work are two fold﹒First﹐the molecular dynamics (MD) simulation is used to investigate the influence of process parameters on metal thin film morphology and trench – filling morphology in the damascene process﹒The MD simulation adopts a two dimensional deposition model and uses Lennard–Jones Potential to represent the intermolecular force acting between Cu atoms﹒The Gear’s predictor – corrector algorithm is used to calculate the trajectories of atoms ﹒ A constant substrate temperature is assumed﹒ The rescaling technique is utilized to model the energy absorbed by the substrate due to the impact of incident atoms ﹒ The morphologies of films for different physical parameters﹐such as temperature﹐incident energy ﹐ deposition rate and high temperature reflow process are discussed﹒ It is found that temperature and incident energy effect play a similar role and with increasing temperature or incident energy﹐ the void percentage and root mean aquare roughness are decreasing ﹒ Meanwhile ﹐ there exists an optimal region of deposition rate for an incident energy﹒Refiow process is achieved by annealing procedure ﹐ the heating up and then cooling down the trench rapidly﹐to drive the copper diffusion and remedy the void defect﹒Secondary is to improve the computational efficiency of MD simulation﹒We treat the calculation of interactions which is the most time – consuming part by using distributed computing based on the Message Passing Interface (MPI)﹒In our result﹐distributed computing is able to reduce the complexity from O(N²) to O(N²/p) on a p–processor system for N particles and achieve a linear speedup result﹒application/pdf1318486 bytesapplication/pdfzh-TW國立臺灣大學應用力學研究所分子動力模擬 濺鍍沉積 分散式計算 訊息傳遞介面Molecular Dynamics Simulation Sputter Deposition Distributed Computation Message Passing Interface (MPI)reporthttp://ntur.lib.ntu.edu.tw/bitstream/246246/21647/1/902212E002203.pdf