A Computational Study of Conduction in Dielectrics, with Implications for Neuromorphic Computing

Dielectrics are vital for a wide range of important devices, from MOSFETs and FinFETs to emerging nonvolatile memory devices such as Resistive Random Access Memories (RRAMs), and metal-insulator-semiconductor (MIS) contacts for photovoltaics and nanoelectronics. Developing an understanding and intuition for the underlying physical mechanisms that result in conduction in dielectrics is a challenging task to undertake experimentally, since the electron transport mechanisms are stochastic and unpredictable. In this work, we use Ginestra, a commercial tool developed by MDLSoft (acquired by Applied Materials in March 2019) to simulate and provide explanations for anomalies that were observed in experimental data. This work has three focus areas: (1) Using Ginestra to explain endurance issues observed in a Hf-based RRAM device, (2) Using Ginestra to verify Fermi Level pinning factor for ZnO, that had been extracted experimentally, and (3) studying the use of nonfilamentary RRAMs for Neuromorphic Computing applications.

We find that Ginestra is well-suited for studying metal-insulator-metal (MIM) device structures, providing key insights into how the endurance and reliability of RRAM cells can be optimized. However, Ginestra’s simulations suggested a trend opposite to the one observed experimentally for our ZnO pinning factor experiments, suggesting that its model set may need to be enhanced to account for effects specific to MIS structures. Our analysis of nonfilamentary RRAMs for Neuromorphic Computing suggests that a hybrid pulsing scheme, comprised of a stepped pulsing scheme for potentiation (LTP), and a standard uniform pulsing scheme for depression (LTD) produces a synaptic signature that has the linearity and symmetry that is desired for neuromorphic computing applications.