Ab initio Quantum Transport in Conductive Bridging Random Access Memory

Abstract

In line with the growing popularity of data-driven IT applications, the importance of data storage and memory has not stopped increasing during the last decade. However, the complementary metal-oxide-semiconductor (CMOS)-based storage and memory technology is reaching its scaling limits and has become ill-suited for low power and energy efficient operations. Therefore, new nonvolatile memory technologies are being developed and are slowly entering the mass market. Among those, conductive bridging random access memory (CBRAM) cells promise great advances in terms of power consumption, integration density, and cointegration with CMOS-based logic circuits. CBRAMs are metal-insulator-metal heterojunctions through which a metallic filament grows and dissolves, which creates two distinctive resistance states corresponding to logical 1 or 0. Nevertheless, several challenges remain to be addressed before they can compete with traditional technologies, particularly with respect to variability and reliability. Solving these issues is complicated by the fact that the operating mechanisms of CBRAMs are not fully understood. The aim of this thesis was to use quantum transport simulations to elucidate the switching characteristics of CBRAM cells. We used density-functional theory to compute the electronic structure of nanoscale components, from which structural, electrical, thermal and electro-thermal properties of memory cells were derived in a parameter-free manner. First, the theoretical foundation of first principles-based quantum transport is presented. Such simulation techniques induce an immense computational burden when applied to large models such as CBRAM. This challenge is assessed and addressed in the subsequent chapters. For that purpose, the so-called mode-space approximation originally derived within the effective mass context was generalized and applied to CBRAM systems. Moreover, the process of obtaining the required mode-space transformation matrix was largely automatized which drastically simplifies the usage of this approach. Its strength and reliability is demonstrated with the help of a CBRAM configuration whose electrical current was computed both with and without the mode-space approximation. We found that the discrepancy between both models can be kept below 2%. Furthermore, the reduction of the required computational resources was valuated and we observed that it can be as large as two to three orders of magnitude, depending on the targeted accuracy. Next, a model for a Cu/SiO2/Cu CBRAM cell is presented. Its ON-OFF switching behavior was investigated and an estimate of the number of atoms that contribute to the process is provided. Multiple intermediate states were simulated to present the electrical conductance as a function of the number of “dissolved” atoms. The results qualitatively agree with experimental data. Moreover, the electro-thermal properties for CBRAM cells of varying size were determined, leading to an explanation for the experimentally discovered improved performance of ultra-scaled devices. Bringing the memory cell closer to their ballistic transport limit by decreasing their thickness minimizes the interaction between electrons and atomic lattice vibrations. This in turn lowers the lattice temperature within the cell, thereby improving its thermal stability. In the last chapter of this thesis, the impact of the CBRAM material stack on the electrical current is examined. We observed that for a given atomic configuration of the filamentary structure in the ON-state, the metal of the counter electrode has little influence on the conductance. However, the current density is affected by the choice of the electrode material. This fact is likely due to the asymmetric configuration of typical CBRAM cells, which is challenging to account for in ab initio quantum transport. However, to assess the resistance state of a filament, a we found that assuming symmetric metallic electrodes is sufficient. Yet, calculating further properties of the system such as the interaction with the surrounding oxide, requires representing the full asymmetric nature of CBRAM stacks.