PI: Bilge Yildiz, Departments of Nuclear Science and Engineering and Department of Materials Science and Engineering, MIT
The goal of this proposal is to design an all solid inorganic device that serves for resistive switching based on proton intercalation, while advancing the scientific understanding of mechanisms and kinetics governing the resistive switching in this mode. The outcomes of this project are important for obtaining energy efficient and fast neuromorphic computing devices to implement artificial intelligence algorithms. Resistive switching based on conducting filament forming or phase change materials currently have difficulties to demonstrate energy efficiency and accuracy in neural architectures. We believe that alternative materials that switch by a fundamentally different mechanism are worthwhile and exciting to pursue. A mechanism that does not rely on conducting filaments (which require large electric fields to form) or on phase change (which require large currents to heat) is desirable. Cation intercalation is one such attractive mechanism that has the potential to have energy consumption per synaptic event that is comparable to or even lower than that in the brain. We propose that a 3-terminal device based on H+ cation (proton) intercalation should be the most promising compared to other cations. Hydrogen has the smallest ionic radius and the highest cation mobility that should improve the device energy efficiency (lower voltages needed to shuffle protons between electrodes). We are inspired by electrochromic materials and devices which exhibit large changes in electronic structure and conductance upon insertion and removal of small cations like H+, with accompanying valence change of the other element(s). The proposed devices will have a proton intercalating oxide layer as the resistive switching medium, a solid tunable hydrogen source and a solid electrolyte. There are two design concepts described here: i) all solid inorganic devices with solid electrolytes, and ii) hybrid devices with a liquid electrolyte contained in a solid polymer trap. The work combines electrochemical, electronic and spectroscopic measurements, to assess both the device performance and the mechanisms of intercalation and rate limiting steps. By doing so, we envision to predict solid electrolytes that may surpass the proton conductivity of currently available materials, and interface chemistries that accelerate charge transfer between the electrolyte and the channel materials. As a result, this proposal presents original solutions for reliable materials in resistive processing units for neuromorphic computing.