Quantum Point Contact Parameter Extraction of Carbon-based Resistive Memory using Hybrid Genetic Algorithm

TL;DR

This paper introduces a hybrid genetic algorithm approach to extract quantum point contact parameters from experimental data of carbon-based resistive memories, enabling better understanding of their conduction mechanisms.

Contribution

It proposes a novel hybrid genetic algorithm method for parameter extraction of the QPC model from macroscopic I-V data in carbon resistive memories.

Findings

MAPE of 3.7% in model fitting

Parameters within reasonable physical ranges

QPC model effectively describes conduction mechanisms

Abstract

Resistive switching phenomenon in carbon film is associated with formation and annihilation of low resistance sp2 nanochannels within the amorphous sp3 matrix. The thinnest point of these graphitic nanochannels behaves like quantum wire (QW) and limits current flow. Transport mechanism at these bottlenecks can be described within the framework of quantum point contact (QPC) model. The model applies mesoscopic Landauer formalism and correlates device resistance state with the density of the nanochannels as well as lateral area of its constriction. However, QPC model parameter extraction from macroscopic I-V characteristic is not feasible due to multiple nonlinear and closely coupled parameters, e.g. barrier height ({\Phi}), barrier curvature ({\alpha}) and voltage drop ratio (\b{eta}). In this work, a hybrid genetic algorithm (GA) based parameter extraction flow is proposed and is applied to extract parameters from experimental result of carbon-based resistive memories. In this proposed flow, the number of single subband channels for low resistance state (NLRS) and barrier curvature parameter at high resistance state ({\alpha}HRS) were first derived by exploring boundary conditions at low voltage. The four remaining parameters ({\Phi}HRS, {\Phi}LRS, {\alpha}LRS, and \b{eta}) were then extracted from experimental I-V characteristic by leveraging hybrid GA. Mean absolute percentage error (MAPE) of fitted QPC models is only 3.7% and the extracted parameters are within reasonable range. Hereby, we regard the QPC model as a viable model to describe conduction mechanism of graphitic nanochannels in carbon-based resistive memories.