阻絲與非阻絲型態電阻式記憶體之數值模擬研究

Abstract

由於元件微縮，固態記憶體與硬碟兩者之間存取效能差異越來越顯著，為了彌補差異，儲存記憶體（Storage class memory）被提出廣泛討論，其中尤以電阻式記憶體（Resistive switching random access memory; RRAM）最具優勢。今年七月，英特爾與美光提出Xpoint Array，雖未表明記憶體元件為何，但猜測為RRAM，這驗證其未來的關鍵角色。 然而RRAM的廣泛研究中，皆渴望能有一物理模型可解釋阻值切換與傳導機制行為。本論文利用數值模擬方式，建立了阻絲與非阻絲型態的RRAM物理模型；在TiN/HfOx/Pt元件下，藉由滲透理論與動態蒙地卡羅離子移動，描述氧離子與氧缺間的交互影響，解釋了阻絲生成行為，且利用氧缺間的分布解釋電性記憶窗口的展現。另一模型是基於Ta/TaOx/TiO¬2/Ti 雙層RRAM結構，藉由缺陷分布於TiO2區域，電子被捕捉與釋放而改變了TiO2的介電常數，使得跨壓分布在兩氧化物中有明顯的差異，解釋此非阻絲RRAM元件的電性特性。 在這兩個模型中，其電性行為與實驗結果皆有很好的一致性，我們藉此模型，了解高低阻態的切換機制，阻絲、非阻絲模型的型態也能合理描述，對於RRAM未來發展，相信這是很有用的研究與分析。

The discrepancy of access time between memories and hard disk drives becomes significant because of the device scaling. To bridge the access time gap, the storage class memory (SCM) was proposed. Among various candidates, the resistive switching random access memory (RRAM) is one of the most promising. On July 2015, Intel and Micron launched the Xpoint memory technology speculatively based on RRAM, highlighting its strong potential in the future. However, the complete model of RRAM to elucidate device properties such as resistive switching and conduction mechanism is still under active research. In this thesis, we built two numerical models for both filamentary and non-filamentary RRAMs. The filamentary RRAM model based on kinetic Monte Carlo (KMC) ions migration and percolation theory describes the interaction between oxygen ions and vacancies in the TiN/HfOx/Pt device. This model can interpret the formation and rupture of the conduction path in HfOx and thus the I-V characteristics. On the other hand, the non-filamentary RRAM model described electron (carrier) trapping/detrapping in the Ta/TaOx/TiO2/Ti device. Furthermore, the permittivity modulation in TiO2 depending on the carrier concentration leads to different voltage drops across the TaOx and TiO2 layers, and thus different high and low resistance states. Both models show good agreements with the measurement results. The characteristics of filamentary and non-filamentary RRAMs can be reasonably depicted. We believe that these studies would benefit further investigations and applications of RRAM.